Issuu on Google+


Ecography 000: 001–003, 2011 doi: 10.1111/j.1600-0587.2011.07338.x © 2011 The Authors. Journal compilation © 2011 Ecography

The patterns and causes of elevational diversity gradients Nathan J. Sanders and Carsten Rahbek

A major focus of research in spatial ecology over the past 25 years has been to understand why the number of species varies geographically. The most striking, and perhaps best documented, pattern in spatial ecology is the latitudinal gradient in species diversity in which the number of species, for most taxa, declines with increasing latitude. Understanding the underlying cause(s) of the latitudinal gradient has proven challenging, perhaps because there are really only two latitudinal gradients (in the northern and southern hemispheres), and because it is often difficult to perform experiments at latitudinal scales. Elevational gradients in species diversity are nearly as ubiquitous as latitudinal gradients, and they offer many characteristics that make them perhaps more suitable for uncovering the underlying cause(s) of spatial variation in diversity. First, there are many replicates of elevational diversity gradients – essentially each mountain or mountain range is a replicate, so it is possible to test for the generality of the underlying cause(s). Second, it is possible to carry out manipulative experiments along elevational gradients. Third, field data can be collected more readily along elevational gradients than along latitudinal gradients, simply because the

spatial extent of elevational gradients is small relative to latitudinal gradients. Finally, many of the potential underlying causes that covary along latitudinal gradients (history, climate, time since glaciation, area) do not covary along elevational gradients (Körner 2007). Given the benefits of elevational gradients relative to latitudinal gradients, it seems clear that they can be useful tools to understand the underlying cause(s) of diversity gradients. And, in fact, there is a growing appreciation of the utility of elevational gradients as tools to uncover the mechanisms that shape both patterns of biodiversity and the functioning of ecosystems (Fukami and Wardle 2005, Nogues-Bravo et al. 2008). Ecography has played a major role as an outlet for many studies of elevational gradient studies, and in fact such studies are one of the strengths of the journal. Since its inception, Ecography has published more than 25 papers that have explicitly focused on elevational diversity gradients. The papers highlighted in this Virtual Issue indicate that Ecography has been, and will continue to be, an important outlet for papers at the cutting edge of documenting and explaining elevational gradients in diversity. Here, our goal is to highlight some elevational diversity gradient papers published in Ecography (bold-face in reference list) that we feel have made long-lasting contributions to the study of spatial ecology. This Virtual Issue (http://tinyurl.com/cr2lkew) is about elevational diversity gradients, though we recognize that a number of key papers have been published in Ecography on topics ranging from montane diversity at regional or continental scales (Parra et al. 2004, Ricklefs et al. 2004, Ruggiero and Kitzberger 2004, Ruggiero and Hawkins 2008), population dynamics (Ramriez et al. 2006, Gimenez-Benavides et al. 2011), interactions among species (Fuentes et al. 1992, Mazia et al. 2004), adaptation (Berner et al. 2004), and climate change (Dollery et al. 2006).

The patterns Nearly 20 years ago, one of us (Rahbek 1995) asked whether the conventional wisdom about elevational diversity gradients – that they mirrored the latitudinal gradient and declined with elevation – was supported by the data. Early View (EV): 1-EV


Examining all of the literature (at the time, 97 papers) on elevational diversity gradients showed that the answer was, for the most part, ‘no’. Most studies, when sampling effort was corrected for, showed hump-shaped diversity gradients, with diversity peaking at mid-elevations. The quantitative review of published studies by Rahbek (1995) to document the generality (or lack thereof ) of the pattern was illuminating. The studies in Rahbek’s paper were from various mountain ranges, and on various taxa. One reason that different patterns of elevational diversity might occur in different systems may be that the scale and extent of the elevational gradients varied among studies (Rahbek 2005, Nogues-Bravo et al. 2008) or because different mountain ranges are embedded in different regional climatic areas with different evolutionary histories. This is an under-appreciated fact in comparative studies of elevational diversity gradients. Another approach to examine generality of elevational diversity gradients is to focus on several replicate elevational gradients within the same region, so that species occuring along the gradient might come from the same regional species pool and share similar evolutionary histories. This was the approach of Grytnes (2003), who sampled plant diversity along seven transects in northern Norway, Wang et al. (2009) who sampled tree and herb communities along six elevational gradients in northeast China, and of Sanders (2002) who compiled regional lists of the ants of Colorado, Nevada and Utah. In those studies, the patterns differed slightly among replicate samples, but the underlying causes were similar within each gradient. These results contrast with a study on non-volant mammals in several mountain ranges in Utah by Rowe (2009). In that study, the patterns of diversity with elevation were similar, but the underlying mechanisms differed among mountain ranges. But most of the elevational diversity gradient studies that have been published in Ecography have come from investigators who have compiled empirical data for a given taxon in a particular mountain range. These studies might differ in the extent and scale at which diversity is sampled, ranging from Herzog et al.’s (2005) data on bird diversity in 250 m elevational bands in the Andes to Grytnes’s (2003) 25 m2 plots in Norway. Regardless of the differences in sampling and extent among studies, most agree with the results from Rahbek’s (1995) review of the literature: in most instances, diversity peaks at mid-elevations, with a few notable exceptions (Brehm et al. 2003, Machac et al. 2011).

The underlying causes A number of factors have been implicated as underlying causes of elevational diversity gradients. Some of the most frequently tested are climate and productivity (Rahbek 1995, Odland and Birks 1999, Grytnes 2003, Fu et al. 2006, Rowe 2009, Wang et al. 2009), sourcesink dynamics (Kessler et al. 2011), area (Rahbek 1995, Sanders 2002, Jones et al. 2003, Bachman et al. 2004, Herzog et al. 2005, Romdal and Grytnes 2007), disturbance (Escobar et al. 2007, Bunn et al. 2011), geometric constraints (Sanders 2002, Bachman et al. 2004, Herzog 2-EV

et al. 2005, Fu et al. 2006, Rowe 2009) and evolutionary history (Machac et al. 2011). The diversity of results among studies, and even within studies, suggests that no single mechanism is responsible for all elevational diversity gradients. Future studies, many of which are likely to be published in Ecography (we hope), will move the field forward, perhaps by examining the interplay between contemporary and past climate (Hortal et al. 2011), integrating ecology and evolution (Graham et al. 2009, Machac et al. 2011), employing new tools (Levanoni et al. 2011) and demonstrating the effects of climatic change on current (Forister et al. 2010) and future patterns of biodiveristy (Colwell et al. 2008). Papers published in Ecography have been some of the first to test explicitly many of these mechanisms, and their generality. As the number of studies on elevational diversity gradients continues to grow (more than 300 as of 2011), Ecography will continue to play a role in shaping the field and helping to uncover the mechanisms which shape broadscale variation in species richness, especially along elevational gradients.

References Bachman, S. et al. 2004. Elevational gradients, area and tropical island diversity: an example from the palms of New Guinea. – Ecography 27: 299–310. Berner, D. et al. 2004. Grasshopper populations across 2000 m of altitude: is there life history adaptation? – Ecography 27: 733–740. Brehm, G. et al. 2003. Unique elevational diversity patterns of geometrid moths in an Andean montane rainforest. – Ecography 26: 456–466. Bunn, W. A. et al. 2011. Change within and among forest communities: the influence of historic disturbance, environmental gradients, and community attributes. – Ecography 33: 425–434. Colwell, R. K. et al. 2008. Global warming, elevational range shifts, and lowland biotic attrition in the wet tropics. – Science 322: 258–261. Dollery, R. et al. 2006. Impact of warming and timing of snow melt on soil microarthropod assemblages associated with Dryas-dominated plant communities on Svalbard. – Ecography 29: 111–119. Escobar, F. et al. 2007. From forest to pasture: an evaluation of the influence of environment and biogeography on the structure of beetle (Scarabaeinae) assemblages along three altitudinal gradients in the Neotropical region. – Ecography 30: 193–208. Forister, M. L. et al. 2010. Compounded effects of climate change and habitat alteration shift patterns of butterfly diversity. – Proc. Natl Acad. Sci. USA 107: 2088–2092. Fu, C. Z. et al. 2006. Elevational patterns of frog species richness and endemic richness in the Hengduan Mountains, China: geometric constraints, area and climate effects. – Ecography 29: 919–927. Fuentes, M. 1992. Latitudinal and elevational variation in fruiting phenology among western-European bird-dispersed plants. – Ecography 15: 177–183. Fukami, T. and Wardle, D. A. 2005. Long-term ecological dynamics: reciprocal insights from natural and anthropogenic gradients. – Proc. R. Soc. B 272: 2105–2115. Gimenez-Benavides, L. et al. 2011. Demographic processes of upward range contraction in a long-lived Mediterranean high mountain plant. – Ecography 34: 85–93.


Graham, C. C. et al. 2009. Phylogenetic structure in tropical hummingbird communities. – Proc. Natl Acad. Sci. USA l06: 19673–19678. Grytnes, J. A. 2003. Species–richness patterns of vascular plants along seven altitudinal transects in Norway. – Ecography 26: 291–300. Herzog, S. K. et al. 2005. The elevational gradient in Andean bird species richness at the local scale: a foothill peak and a high-elevation plateau. – Ecography 28: 209–222. Hortal, J. et al. 2011. Ice age climate, evolutionary constraints and diversity patterns of European dung beetles. – Ecol. Lett. 14: 741–748. Jones, J. I. et al. 2003. Area, altitude and aquatic plant diversity. – Ecography 26: 411–420. Kessler, M. et al. 2011. The impact of sterile populations on the perception of elevational richness patterns in ferns. – Ecography 34: 123–131. Körner, C. 2007. The use of ‘altitude’ in ecological research. – Trends Ecol. Evol. 22: 569–574. Levanoni, O. et al. 2011. Can we predict butterfly diversity along an elevation gradient from space? – Ecography 34: 372–383. Machac, A. et al. 2011. Elevational gradients in phylogenetic structure of ant communities reveal the interplay of biotic and abiotic constraints on diversity. – Ecography 34: 364–371. Mazia, C. N. et al. 2004. Interannual changes in folivory and bird insectivory along a natural productivity gradient in northern Patagonian forests. – Ecography 27: 29–40. Nogues-Bravo, D. et al. 2008. Scale effects and human impact on the elevational species richness gradients. – Nature 453: 216–219. Odland, A. and Birks, H. J. B. 1999. The altitudinal gradient of vascular plant richness in Aurland, western Norway. – Ecography 22: 548–566.

Parra, J. L. et al. 2004. Evaluating alternative data sets for ecological niche models of birds in the Andes. – Ecography 27: 350–360. Rahbek, C. 1995. The elevational gradient of species richness – a uniform pattern. – Ecography 18: 200–205. Rahbek, C. 2005. The role of spatial scale and the perception of large-scale species–richness patterns. – Ecol. Lett. 224– 239. Ramirez, J. M. et al. 2006. Altitude and woody cover control recruitment of Helleborus foetidus in a Mediterranean mountain area. – Ecography 29: 375–384. Ricklefs, R. E. et al. 2004. The region effect on mesoscale plant species richness between eastern Asia and eastern North America. – Ecography 27: 129–136. Romdal, T. S. and Grytnes, J. A. 2007. An indirect area effect on elevational species richness patterns. – Ecography 30: 440–448. Rowe, R. 2009. Environmental and geometric drivers of small mammal diversity along elevational gradients in Utah. – Ecography 32: 411–422. Ruggiero, A. and Hawkins, B. A. 2008. Why do mountains support so many species of birds? – Ecography 31: 306– 315. Ruggiero, A. and Kitzberger, T. 2004. Environmental correlates of mammal species richness in South America: effects of spatial structure, taxonomy and geographic range. – Ecography 27: 401–416. Sanders, N. J. 2002. Elevational gradients in ant species richness: area, geometry, and Rapoport’s rule. – Ecography 25: 25–32. Wang, X. P. et al. 2009. Relative importance of climate vs local factors in shaping the regional patterns of forest plant richness across northeast China. – Ecography 32: 133–142.

3-EV


ECOGRAPHY 27: 299 /310, 2004

Elevational gradients, area and tropical island diversity: an example from the palms of New Guinea Steven Bachman, William J. Baker, Neil Brummitt, John Dransfield and Justin Moat

Bachman, S., Baker, W. J., Brummitt, N., Dransfield, J. and Moat, J. 2004. Elevational gradients, area and tropical island diversity: an example from the palms of New Guinea. / Ecography 27: 299 /310. The factors causing spatial variation in species richness remain poorly known. In this study, factors affecting species richness of palms (Palmae/Arecaceae) were studied along the elevational gradient of New Guinea. Interpolated elevational ranges were calculated from a database of all known collections for 145 species in 32 genera. The amount of land area at different elevations greatly affects the species richness gradient. If assessed in equal-elevation bands species richness appears to decline monotonically, but when assessed in equal-area bands species richness shows a pronounced midelevation peak, due to the large proportion of lowlands in New Guinea. By randomising species ranges within the total elevational gradient for palms and accounting for area, we found the mid-elevation peak to be consistent with a middomain effect caused by the upper and lower limits to palm distribution. Our study illustrates the importance of accounting for area in macroecological studies of richness gradients and introduces a novel yet simple method for doing this through the use of equal-area bands. Together, the effect of area and the mid-domain effect explain the majority of variation in species richness of New Guinea palms. We support calls for the multivariate assessment of the mid-domain effect on an equal footing with other potential explanations of species richness. S. Bachman, Dept of Geography and Earth Sciences, Brunel Univ., Uxbridge Middlesex, U.K. UB8 3PH. / W. J. Baker (correspondence: w.baker@rbgkew.org.uk), N. Brummitt, J. Dransfield and J. Moat, The Herbarium, Royal Botanical Gardens, Kew, Richmond, Surrey, U.K. TW9 3AB.

Along with the well-known latitudinal gradients in species richness, patterns of biotic diversity along elevational gradients have been studied for centuries (Willdenow 1805, Darwin 1859, Wallace 1876, Whittaker 1960, Brown 1971, Rahbek 1995, 1997). Many of these studies have attempted to correlate observed patterns of diversity with various environmental gradients such as precipitation, temperature, humidity and productivity. It has been generally recognised that species richness declines with elevation. However, monotonic declines in richness are less typical than are unimodal peaks or patterns where species richness plateaus before decreasing (Rahbek 1995). Monotonic declines in species richness are thought by some authors to mirror a

decrease in productivity (Rahbek 1997 and references therein, Kaspari et al. 2000), although productivity gradients have also been used to explain mid-elevation peaks in species richness (Rosenzweig 1992, 1995, Rosenzweig and Abramsky 1993). However, to date, no universal relationship between productivity and richness has been elucidated. In contrast, the relationship between area and number of species is well known (Arrhenius 1921, Williams 1943, Rosenzweig 1995). Traditionally, species richness is thought to increase with increasing area following a power-law model (Williamson 1988; but see also Connor and McCoy 1979). In regions with diverse landscapes that range from sea level to high mountains, the land

Accepted 16 December 2003 Copyright # ECOGRAPHY 2004 ISSN 0906-7590 ECOGRAPHY 27:3 (2004)

299


area in different elevational bands varies greatly. Commonly, land area decreases as elevation increases such that lowlands account for the highest proportion of the total land area of a given region (MacArthur 1972); mountainous islands such as New Guinea display this feature particularly well. It seems strange then that, when examining species richness along elevational gradients, some authors have not considered the influence of area (e.g. Patterson et al. 1998, Ohlemu¨ller and Wilson 2000). Several studies show that when area has been controlled species richness peaks at mid-elevations (e.g. Lawton et al. 1987, Rahbek 1997), although unimodal peaks have been found without accounting for area (Koleff and Gaston 2001, Sanders 2002, Grytnes and Vetaas 2002). It has also been shown that area alone could explain a large proportion of the variance in the observed richness pattern (Rahbek 1997, Koleff and Gaston 2001, Sanders 2002). The mid-elevation peak is now recognised as a common pattern (Rahbek 1995, 1997, Heaney 2001, Kessler 2001, Nor 2001, Rickart 2001, Sa´nchez-Cordero 2001), but despite the recent resurgence of interest in studies of species richness along elevational gradients, a generally-accepted explanation for this mid-elevation peak has not yet been found. Recently, the potential influence of non-biological factors on species richness patterns has been highlighted (Colwell and Hurtt 1994, Colwell and Lees 2000). When species ranges are randomly placed within a geographically defined domain, a peak in richness inevitably arises towards the centre of this domain (Colwell and Lees 2000). In real terms, this domain may be an island or continent bounded by its shores, a mountain bounded by its summit and the lowlands at its base, or an ocean bounded by the surface and its floor. A gradient in elevation is a simple, one-dimensional domain. Within the domain, species with a range mid-point at midelevations will have a greater potential elevation range, or amplitude, as they can extend further both upwards and downwards than can those species whose range midpoint is found near the top or the bottom of the domain. Since species ranges are constrained geometrically in this way (they cannot extend either above the highest mountain top or below sea level), the number of overlapping elevational ranges will be greater at mid-elevations and, thus, so will species richness. This is due simply to species of moderate to large elevational range being more likely to cross the middle of the domain than to coincide at either boundary. The phenomenon, termed the ‘‘mid-domain effect’’ (MDE), is therefore dependent upon the underlying frequency distribution of relative species range-sizes. The mid-domain effect is attributed to these simple geometric constraints of the domain or, more accurately, to geometrically constrained stochastic processes affecting the underlying range-size frequency distribution (RSFD; Colwell et al. in press). 300

Despite growing evidence and support for the existence of the MDE (Colwell and Hurtt 1994, Lyons and Willig 1997, Pineda and Caswell 1998, Willig and Lyons 1998, Colwell and Lees 2000, Jetz and Rahbek 2001, Sanders 2002, Grytnes and Vetaas 2002), the concept has yet to gain widespread acceptance as a plausible factor contributing to richness patterns (Bokma and Mo¨nkko¨nen 2000, Koleff and Gaston 2001, Diniz-Filho et al. 2002). Several studies (Lees et al. 1999, Jetz and Rahbek 2001, Koleff and Gaston 2001, Sanders 2002) have investigated the combined influence of area and MDE on species richness patterns and in each case richness was explained well by the two combined factors. To date, however, few MDE studies have focused on vascular plants (but see Grytnes and Vetaas 2002), and in addition, except for a pioneering study of butterflies in Madagascar (Lees et al. 1999), no MDE studies have focused on large tropical islands such as New Guinea. New Guinea offers a perfect template for MDE studies being rich in endemic species, relative to total species richness, mountainous, and clearly defined in terms of a domain. Furthermore, the deepening biodiversity crisis should surely direct such research to those areas, such as tropical islands, where the greatest threats to biodiversity exist. This study aims to investigate the influence of available area and MDE on species richness in palms (family Arecaceae or Palmae) along the elevational gradient of New Guinea and surrounding islands, one of the most biodiverse areas in the Asia-Pacific region. We use two methods, one published (Sanders 2002) and one novel, to ask the following questions: 1) Does available area affect palm richness at different elevations in New Guinea? 2) Do geometric constraints affect palm richness along this elevational gradient? 3) Do genera of palms also show the same patterns as species do? The methods and data used in this study will be critically examined and may help to highlight important issues for other studies aiming to investigate similar themes.

Methods Study area and empirical data New Guinea is the largest tropical island in the world with a surface area of 808 510 km2 and an elevational range extending from sea level to 5030 m, the highest peak being Puncak Jaya (also known as Mt. Carstenz). The study area (Fig. 1) comprises the Indonesian province of Papua (formerly Irian Jaya) and the independent country of Papua New Guinea, and is bounded in the east by the Louisiade Archipelago and in the west by the islands Misool and Waigeo. In addition to mainland New Guinea, other major islands included in the study area are Yapen, Biak, Manus and the Bismarck Archipelago. The island of Bougainville (a part of Papua ECOGRAPHY 27:3 (2004)


Fig. 1. New Guinea and surrounding islands; the dark area represents the study region, the grey scale shading indicates elevational change. The island is divided into two political entities, the independent country Papua New Guinea to the east and the Indonesian province of Papua (formerly Irian Jaya) to the west. The islands to the north-east form the Bismarck Archipelago. Dots indicate the distribution of georeferenced collecting localities.

New Guinea) has been excluded because its biogeographic links are closer to the Solomon Islands than to New Guinea. The study area stretches over a relatively small latitudinal range (from 08 to 11840ƒS), thus minimising any possible effect of a latitudinal gradient. The primary dataset for this study was extracted from a database of /3000 herbarium specimen records from the specified area. The data were gathered from herbaria at six institutions (Royal Botanic Gardens, Kew; the Papua New Guinea Forest Research Inst., Lae; Herbarium Bogoriense, Bogor, Indonesia; the National Herbarium of the Netherlands [Leiden branch]; the Univ. of Aarhus, Denmark and the Queensland Herbarium, Brisbane, Australia). The database has been compiled as part of a current taxonomic research project on the palm flora of New Guinea (Baker 2002). Each record in the database relates to a single herbarium collection, including all known duplicates of any given collection. As well as taxonomic data, each record includes all locality and elevation data from the data labels of each specimen; collecting localities are shown in Fig. 1. The variation in the number of collections along the elevational gradient, with amount of land area in successive equal-elevation bands, is depicted in Fig. 2. Elevation ranges were calculated from the database records using two separate methods. The first method used empirical elevation readings recorded in the field by the collector; this will be referred to as the ‘‘field data method’’. This is regarded as a primary source of elevation data because it comes directly from the collector’s field measurements. These data will most likely have been gathered using a barometric altimeter, but may also have been estimated using topographic maps. In the second method, elevation ranges were derived from a digital elevation model (DEM) produced by the United States Geological Service (Anon. 2001); this will be referred to as the ‘‘DEM method’’. This is considered to be a secondary method because the elevation data are ECOGRAPHY 27:3 (2004)

indirectly derived from the locality data of the specimens through a DEM and are highly dependent on the quality of that locality data and the DEM. The DEM method was only applied to records that were geo-referenced (i.e. spatially referenced to a point on the surface of the earth), either with a GPS unit, from a gazetteer or with a reliable map estimate. Using only those species and genera for which both georeferences and field recorded elevation data were available (145 species, 32 genera) it was possible to compare the influences of the two data types on the analysis of species richness patterns. It should be noted that the two data types (field and DEM) are not wholly statistically independent. However, we include both since field and DEM data may differ slightly for the same record, and more importantly, in some cases, individual specimen records may have either georeferences or field-recorded elevation data, but not both. This comparison of field and DEM records may therefore expose potential biases in our analysis due to these differing data collection techniques.

Fig. 2. The relationship between available area and number of collections in successive equal-elevation bands. Lowlands make up by far the largest area in New Guinea; the amount of available area decreases dramatically with increasing elevation. The number of collections is not uniform over the elevational gradient but it remains proportional to the amount of available area.

301


When conducting a MDE analysis no taxon should have a range that extends beyond the domain being studied; in the narrowest sense, the species and genera should be endemic to the domain (Colwell and Lees 2000). In this study the lower limit of the domain is sea level and the upper limit is the maximum empirical elevation of palm records (Field data /2800 m, DEM / 2803 m). Of the 145 species available for analysis, 21 were not endemic to the study region. However, the elevational ranges of these non-endemic species did not exceed the limits of the domain after the examination of additional non-New Guinea records. Therefore, all 145 species were used in the analyses. Of the 32 palm genera that occur in New Guinea, only 2 are endemic to the region as a whole. The elevational ranges of all genera within New Guinea were therefore compared with their ranges outside of the region and as a result 8 genera were excluded from the analyses because their ranges outside of the study region extended beyond the domain limits within the study region. Thus, 24 genera were used in the analyses.

Gaston 2001, Colwell et al. in press). The purely theoretical Models 1 /3 (Colwell 2000) were also not used because of the implicit biological assumptions they make regarding RSFD’s (Colwell et al. in press). The model was iterated 100 times to produce the null distribution. The observed richness values and area values for each band were log-transformed to account for the relationship between species number and area. The MDE predictions were also log-transformed so that they could be compared with the log transformed observed richness values and area values. Using simple linear regression, the independent variables of area and MDE predictions were separately tested against our empirical dataset to examine the amount of variance in the observed richness that could be conditionally explained. The combined effects of both area and MDE predictions were also tested against observed richness. Tests of significance are complicated by the fact that richness figures for the elevational bands are spatially-autocorrelated and hence not statistically independent (Colwell pers. comm.). Consequently, no p-values are presented here.

Analytical methods

Method 2 Method 2 accounts for area by measuring richness along the elevational gradient in equal-area bands rather than equal-elevation bands. This method, despite its simplicity, appears not to have been used in species richness analyses so far. The elevational gradient was classified into equal-area bands using GIS software ArcView 3.2 (Anon. 1999). The original Digital Elevation Model (DEM) data (Anon. 2001) is given as integers (whole numbers), but because of the extremely large area of lowland in New Guinea this meant that the majority of pixels were classified as either 1 or 2 m elevation only. This did not allow sufficient equal-area bands to be produced, since it was impossible to split the largest band ( B/1 m elevation) any further. To overcome this problem the original elevation data was converted to decimal numbers and a random number between /0.5 and/0.5 added to the elevation of each pixel for all elevations (in ArcView Spatial Modeller the equation would be {[DEM] /0.5/Grid.MakeRandom}). This produced a grid that could be easily classified into equal-area bands. Unfortunately, however, in ArcView it is not possible to classify a floating number grid into equal areas. To determine the band boundaries, therefore, the new DEM data was exported to a database program (only land was exported / sea was treated as null, and removed) and pixels sorted by ascending value of elevation. The band boundaries were calculated from the number of pixels divided by the number of bands desired. The value for each band boundary was then simply read off from the row number within the database, for each of 5, 10 and 15 equal-area bands.

Two methods were used to investigate the influence of area and MDE on taxon richness along the elevational gradient. In order to explore the basic relationship between elevation and richness the gradient was divided into 100 m equal-elevation bands and total richness was calculated as the sum of all species or genera occurring within each band. Method 1 then specifically tests the separate and combined influences of area and MDE on observed richness patterns within these equal-elevation bands, using simple linear regression. Method 2 examines the amount of variance in observed richness that can be explained by MDE after area has been accounted for, by dividing the gradient into equal-area bands, also using simple linear regression. Method 1 Method 1 follows the procedure used by Sanders (2002). Observed richness was calculated by summing the number of species or genera in each 100 m elevation band. It was assumed that each taxon occurred in all bands between its minimum and maximum elevational limits. The amount of area in each band was calculated using the DEM. The MDE null distribution was calculated using RangeModel software (Colwell 2000). We used Model 4 (Colwell 2000) which selects (with replacement) ranges from the empirical range-size frequency distribution (RSFD) and randomly places them in the domain (the elevation gradient). Model 5 (Colwell 2000), which uses empirical midpoints and random range-sizes, was not used because the null model is too closely constrained by the empirical data (Koleff and 302

ECOGRAPHY 27:3 (2004)


With area kept constant at each band, richness can be looked at anew. However, there is no set rule regarding the number of equal-area bands into which a gradient should be split. To explore the effect of the number of bands on richness estimates, we split the elevational gradient into 5, 10 and 15 equal-area bands. The elevational ranges for all these equal-area bands are given in Appendix 1. The equal-area richness patterns were then plotted against those predicted by the MDE null model under Model 4 of RangeModel (Colwell 2000). The model was iterated 100 times using the empirical RSFD. In using equal-area bands we have created bands with unequal elevation, since the amount of area in New Guinea declines with increasing elevation (see Fig. 2). In other words, the higher the equal-area band, the broader is its elevational range. Therefore multiple linear regression was used to compare the observed richness against expected richness derived from MDE null predictions with elevational range for each equal-area band as an additional explanatory variable. The procedure was carried out for species and genera using both types of elevation data (DEM and field data). Slope values for each relationship were also calculated, under the assumption that the closer the slope is to 1, the more closely the MDE null model predicts the number of observed taxa (see also Jetz and Rahbek 2001).

Results Basic relationships between richness and elevation Without accounting for area both species and genus richness of New Guinea palms decreases as elevation increases (Fig. 3A). Generic richness peaks in the lowest elevation band (0 /100 m) for both DEM and field data methods. Species richness also peaks in the lowest elevation band using the field data method, but peaks in the second band (201 /300 m) using DEM data. The DEM-derived richness estimates for both species and genera are consistently higher than the field data estimates along the gradient, with the exception of species for the first elevation band where the field data estimate is greater than the DEM estimate. However, the overall patterns along the gradient are broadly similar for both data types. Method 1 The relationships between New Guinea palm richness, area and MDE along the elevational gradient, presented as in Sanders (2002), are given in Fig. 4A /D. All r2 values are shown in Table 1. Area alone conditionally explained around half of the variance in observed richness pattern for both species and genera irrespective of data type used (min r2 /0.45 [genera, DEM]; max r2 /0.52 [species, Field and DEM]). MDE alone conECOGRAPHY 27:3 (2004)

Fig. 3. The relationship between richness and elevation within the study region. (A) species and generic richness within 100 m equal-elevation bands; (B) species and generic richness within equal-area bands. The thick, solid line represents species richness using field data; the dot-dash line species richness with elevations calculated from a DEM; the thin, solid line generic richness using field data and the dashed line generic richness with elevations calculated from a DEM.

ditionally explained very little of the variation in observed richness regardless of data type and taxonomic scale (min r2 /0.04 [species, Field]; max r2 /0.33 [species, DEM]). However, the combined effects of area and MDE conditionally explained a great deal of the variation (min r2 /0.77 [genera, Field]; max r2 /0.89 [species, DEM]) of the observed richness pattern, with the highest r2 value for species DEM data (Table 1). In general, the combined effects of area and MDE gave higher r2 values for DEM data (species, 0.89; genera, 0.84) than for field data (species, 0.84; genera, 0.77).

Method 2 The removal of the effect of area on richness through the use of equal-area bands yielded a mid-elevation peak in both species and genus richness patterns (Fig. 3B). This contrasts markedly with the basic relationship between richness and elevation, which is roughly monotonic (Fig. 3A). The relationships between richness, area and MDE are given in Fig. 5A /D for 15 equal-area bands. All r2 values are shown in Table 2. Although the MDE null predictions consistently underestimate observed richness, the variation in observed patterns at both taxo303


Fig. 4. Comparison of species and generic richness with amount of area and MDE null-model predictions along the elevational gradient. (A) species richness with elevations from field data; (B) generic richness with elevations from field data; (C) species richness with elevations from DEM data; (D) generic richness with elevations from DEM data. For each graph the solid line represents species or generic richness, the solid line with hollow circles shows area and the solid line with solid circles shows MDE null-model predictions.

nomic levels and with both data types are generally comparable with the null model predictions. Linear regression shows that the variance in species richness with field data in most cases is conditionally explained remarkably well by the MDE null predictions irrespective of the number of equal-area bands (min r2 / 0.90 [5 and 15 bands]; max r2 /0.98 [10 bands]; see Table 2). However, for species richness from DEM data and generic richness with both field data and DEM data the number of equal-area bands does seem to have an influence; r2 values decrease in unison with number of

equal-area bands, and markedly so for generic richness with field data (r2 /0.70, 15 bands; 0.66, 10 bands; 0.39, 5 bands). In general the variance in richness is best explained when 15 equal-area bands are used (species richness from DEM data, r2 /0.98; generic richness from DEM data, r2 /0.89; generic richness from field data, r2 /0.70), although for species richness with field data the highest r2 value is with 10 equal-area bands. As with Method 1, DEM data consistently returns higher r2 values (species, 0.93 /0.98; genera, 0.82 /0.89) than does field data (species, 0.90 /0.98; genera, 0.39 /0.70). By

Table 1. Results from the simple linear regression analysis using Method 1. The separate and combined effects of area and MDE are shown. Note that in all cases the combined effects of area and MDE explain more of the variance in observed richness than either does alone. Data type

Elevation range (m)

Taxonomic level

Parameter

Field Field Field Field Field Field DEM DEM DEM DEM DEM DEM

0 /2800 0 /2800 0 /2800 0 /2438 0 /2438 0 /2438 0 /2803 0 /2803 0 /2803 0 /2796 0 /2796 0 /2796

Species Species Species Genera Genera Genera Species Species Species Genera Genera Genera

observed observed observed observed observed observed observed observed observed observed observed observed

304

vs vs vs vs vs vs vs vs vs vs vs vs

r2 area MDE MDE/area area MDE MDE/area area MDE MDE/area area MDE MDE/area

0.52 0.04 0.84 0.50 0.07 0.77 0.52 0.33 0.89 0.45 0.10 0.84

ECOGRAPHY 27:3 (2004)


Fig. 5. Comparison of observed richness patterns and MDE null-model predictions along the elevational gradient, split into 15 equal-area bands. (A) species richness with elevations from field data; (B) generic richness with elevations from field data; (C) species richness with elevations from DEM data; (D) generic richness with elevations from DEM data. For each graph the hollow circles show observed richness and solid circles show MDE null predictions.

adding elevational-width of equal area-bands as an additional explanatory variable in a multiple linear regression high r2 values increased marginally whereas low r2 values increased appreciably (see Table 2). Slope values for the relationship between generic richness and MDE null predictions for field and DEM data for 15 equal-area bands, and for field data for 10 equal-area bands were 5/1.09/0.1 (see Table 2). For all types of data, MDE predictions generally underestimated the observed numbers of taxa (see Fig. 5). Slope values for species-level data were all considerably further from 1 than for genus-level data (see Table 2), revealing a greater under-estimation of actual species richness

patterns by MDE null model predictions (see also Fig. 5).

Discussion Data quality With the field and DEM methods we have used two different techniques for obtaining elevation range data from specimen records. As one would expect, both methods have produced broadly similar patterns, although there are slight discrepancies with the max-

Table 2. Results from multiple linear regression analysis of richness against mid-domain effect predictions and equal-area elevational band-width using Method 2. MDE null predictions are correlated with observed richness patterns once area has already been accounted for. Data type

Elevation range (m)

Taxonomic level

No. equal area bands

Slope

r2 (MDE alone)

Multiple r2

Field Field Field Field Field Field DEM DEM DEM DEM DEM DEM

0 /2800 0 /2800 0 /2800 0 /2438 0 /2438 0 /2438 0 /2803 0 /2803 0 /2803 0 /2797 0 /2797 0 /2797

Species Species Species Genera Genera Genera Species Species Species Genera Genera Genera

5 10 15 5 10 15 5 10 15 5 10 15

1.39 1.61 1.63 0.48 0.92 0.90 2.04 1.97 1.99 0.38 0.88 1.06

0.90 0.98 0.90 0.39 0.66 0.70 0.93 0.97 0.97 0.82 0.84 0.86

0.92 0.98 0.95 0.99 0.93 0.91 0.94 0.98 0.97 0.95 0.85 0.87

ECOGRAPHY 27:3 (2004)

305


imum elevations. Assessing the accuracy of either method is problematic. The field data method may provide spurious results because pre-twentieth century herbarium records are less likely to have accurate elevation data due to imprecision of maps and inadequate measuring equipment; even modern altimeters can give misleading measurements. The DEM method may be inaccurate due to imprecise georeferencing or the inexactness of the DEM itself. A strength of this study is that it builds directly on current taxonomic expertise. Data for this analysis are derived directly from individual records for each species (i.e. herbarium specimens), which are primary observations. The database includes the vast majority of available specimen records for New Guinea palms and expert taxonomists have verified the identity of each record. Thus, we feel our data are impeccably sourced and any problem detected during the analysis can be readily investigated. Most other studies have had to rely on distribution data contained in published monographs and field guides, in the form of line maps or dot maps, or written descriptions of distributions. We have therefore avoided some of the additional assumptions associated with the use of such secondary data sources, such as poor coverage of specimens inadequately reflecting true geographic ranges, the incorrect identification of individual records, or outdated taxonomic concepts. As practising taxonomists, we encourage others to use a more explicit, specimen-based approach. Sampling procedure can have a significant influence on richness estimates (Wolda 1987, Rahbek 1995). When investigating Rapoport’s rule, for example, Colwell and Hurtt found that simply following a standard procedure, such as sampling with equal effort at points along a gradient, can produce a spurious Rapoport effect (Colwell and Hurtt 1994). The relationship between area and sampling effort in each elevational band in this study is illustrated in Fig. 2. Although the number of collections is not equal along the elevational gradient, the graph shows that collection intensity nevertheless varies in proportion with land area in each elevation band, as one would expect.

Interpolation It is an assumption inherent in this study that species occur in all elevational bands between the minimum and maximum observed values. Such interpolation is typical of analyses involving species richness estimates (Whittaker et al. 2001, Grytnes and Vetaas 2002). It has been suggested that species richness estimates based on interpolation may overestimate richness towards the centre of the gradient because species are only strictly observed in bands at the extreme ends of each species range (Grytnes and Vetaas 2002). In addition to being 306

overestimated in the centre of the gradient, richness may also be underestimated at the periphery since it cannot be interpolated beyond the range limits (Grytnes and Vetaas 2002). However, interpolating species ranges is a pragmatic solution to an intractable analytical problem, as we have no evidence that species are not found where the ranges have been interpolated, which may not noticeably alter the underlying trends in richness (Lees et al. 1999). Furthermore, after plotting the residuals from the linear regression of log area and log species/ genus richness against elevation (Rahbek 1995) the midelevation peak was still evident with all four analyses (graphs not reproduced here).

The influence of area The analysis of richness patterns of palm species and genera along the elevational gradient in New Guinea yields further evidence for the significant influence of area on these patterns. The species-area relationship has been universally acknowledged, although the exact structure of the relationship is still under discussion (Connor and McCoy 1979, Plotkin et al. 2000, Crawley and Harral 2001). Method 1 shows that area alone explains a good deal of the variation in observed richness (Table 1). Method 2 shows that by factoring out area the richness pattern changes from roughly monotonic to unimodal, irrespective of taxonomic scale or number of equal-area bands. This agrees with previous studies that have found a hump-shaped species richness pattern after accounting for area (Lawton et al. 1987, Rahbek 1997). However, if richness along the elevational gradient were entirely dependent on the effect of area, the richness pattern would vary little across the gradient when examined using equal-area bands. As can be seen in Fig. 3B this is not the case and richness peaks at mid-elevations when the effect of area is removed. The influence of area is completely removed using the equalarea band method and allows the area-controlled richness pattern to be directly compared with MDE null predictions. We suggest that all subsequent MDE analyses need to take into account the influence of area, and this can easily be achieved using the equal-area band methodology presented here. Using equal-area bands in regions where amount of available area declines steadily with elevation, such as New Guinea (see Fig. 2), means that equal-area bands increase in elevational breadth as elevation increases. This might mean that species richness for equal-area bands will increase with elevation simply because as the elevational breadth of each band increases so each band will include a greater number of species, assuming the RSFD remains roughly the same with elevation, and also because beta diversity (species turnover) tends to increase with elevation (Colwell pers. comm., see also ECOGRAPHY 27:3 (2004)


Rahbek 1997). In New Guinea palms, elevational ranges increase with elevation for both species and genera (Spearman’s rank correlation between mean elevational range and elevational breadth for equal-area bands: coefficient rs /0.72 and 0.79 for species field and DEM data, respectively, n/145; 0.85 and 0.98 for generic field and DEM data, respectively, n/24). This increase in mean elevational range accompanies a reduction in the number of taxa at higher elevations, so countering the possible artefacts from increases in the elevational breadth of equal-area bands. Furthermore, the addition of elevational width of equal-area bands to the regression model in most cases accounted for only a marginal increase in the amount of explained variation in the observed richness pattern (see Table 2).

The mid-domain effect in New Guinea palms Despite growing evidence, there is still great scepticism surrounding the importance of the mid-domain effect. So far only a handful of studies have considered the influence of area and MDE on richness (Rahbek 1997, Lees et al. 1999, Jetz and Rahbek 2001, Koleff and Gaston 2001, Sanders 2002). In these studies, the combined effects of area and MDE were found to explain a large proportion of the variance in observed richness. Our study also supports these findings. MDE predictions conditionally explained up to 98% of the variance in observed richness in this study after the effect of area had been removed (species field data, 10 equalarea bands, r2 /0.98). However, as the authors were themselves careful to point out, geometric constraints are an additional influence on patterns of species richness (hence the name mid-domain ‘‘effect’’; Colwell and Lees 2000); it is not claimed to be the sole explanation. Vascular plants have rarely been considered in MDE analyses until now (but see Grytnes and Vetaas 2002), but our results and others point to the general influence of MDE across a wide range of taxa such as birds, mammals, insects, and vascular plants. This study furthermore shows that MDE is also important at different taxonomic scales. However, despite the high r2 values (Table 2), MDE null predictions consistently underestimate observed richness patterns (Fig. 5A /D). This may be a result of the way in which the RangeModel software (Colwell 2000) calculates richness along the domain (see Fig. 1 in Colwell and Lees 2000). Setting the bin range effectively determines the number of times the domain is intersected. For instance, if the domain limits are 0 and 1, and the number of bins is 5, the domain would be intersected at 0.1, 0.3, 0.5, 0.7 and 0.9. Every time a range crosses one of these points, the richness total for that bin increases by one. If a species range extended from 0.4 to 0.8, it would intersect the bins at 0.5 and 0.7, thus ECOGRAPHY 27:3 (2004)

adding to their respective totals. Richness for each bin is calculated as the total number of ranges intersected at that point. Using this intersect method it is possible that very small ranges might lie between these points which will not be picked up in the richness totals for that bin. This is more likely to occur for RSFD’s that contain a high proportion of small ranges. Taxa are also more likely to be missed when the number of bins is small, due to the increased size of the gap between the intersection points. This may explain why observed richness is explained least well (e.g. genus field data, 5 equal-area bands, r2 /0.39) by the MDE when the gradient is split into fewer bins/equal area bands. Inspecting the slope values for all relationships, only those for generic field data for 10 and 15 equal-area bands and generic DEM data for 15 equal-area bands were within 0.1 of an ideal slope of 1.0. The slope when using only 5 equal-area bands for genera for both field data and DEM data was considerably smaller than for both 10 and 15 bands with the same data type, again suggesting that 5 equal-area bands is too few to accurately represent richness patterns over an elevational gradient of this size (see Table 2). Furthermore, the slope values are furthest from 1 for species data (especially species DEM data), showing that MDE predictions under-estimated observed numbers of taxa more when ranges were smaller, and lending weight to the supposition that small species ranges are being missed between the intersections of the domain. Analyses that have considered the combined effects of area and MDE (in one dimension), including this study, have shown that these factors can explain observed richness patterns well (Jetz and Rahbek 2001, Koleff and Gaston 2001, Sanders 2002). MDE null models have attracted criticism for only considering richness patterns across one dimension when, of course, species have ranges extending in two dimensions (Bokma and Mo¨nkko¨nen 2000, Koleff and Gaston 2001). A two-dimensional approach has also been developed (Jetz and Rahbek 2001) which indicates that MDE null predictions across two dimensions can still explain a significant amount of the variance in observed richness patterns. It must be noted, however, that these two-dimensional predictions did not explain as much as the one-dimensional models. In order to fully understand the importance of the MDE it needs to be considered alongside various other factors that may contribute to the pattern of species richness. Unfortunately, data were not available in this analysis for the investigation of other variables that may have contributed to the pattern (e.g. productivity, temperature). Two multivariate analyses have shown that the MDE can still be an important explanatory variable in studies of richness gradients (Lees et al. 1999, Jetz and Rahbek 2002); we support calls for the influence 307


of MDE to be further assessed on an equal footing with other determinants of species richness patterns.

The mid-domain effect and Rapoport’s elevational rule Rapoport’s rule states that there is a simple positive correlation between species latitudinal range-size and latitude (Stevens 1989), following an original observation by Rapoport (1982). This idea has also been extended by Stevens (1992) to elevational gradients, where species elevational ranges would increase with elevation, and so decrease in extent towards the lowlands. Rapoport’s elevational rule runs counter to the expectation of the mid-domain effect, which predicts greatest species richness at middle elevations due to the geometric constraints exerted by the upper and lower boundaries of the elevational gradient. For New Guinea palms, elevational range does indeed increase with midpoint of elevation, and, as Fig. 3A shows, species richness does appear to be greater at low elevations without accounting for the differing amount of available area at different elevations. However, it should be evident from Fig. 3B that, after accounting for area, the results presented here provide support for the expectations of the mid-domain effect model, and against those of Rapoport’s elevational rule (see also Rahbek 1997). This suggests that it is simply the greater available area of lowland regions which gives rise to their apparent diversity, and that the Rapoport rescue effect, which remains largely untested in tropical lowland environments, may not be the cause of such species richness gradients.

Taxonomic scale This study is one of the first yet completed to assess the mid-domain effect at both species and genus level. Although the use of indicator taxa remains contentious for biodiversity studies, genera are now well established as reliable higher-taxonomic surrogates for species-level diversity patterns (Williams and Gaston 1994, La Ferla et al. 2002). It is clear from Fig. 5 that patterns in taxon richness across the elevational gradient, including the results of mid-domain effect null model simulations, are similar and highly correlated between both species and genera, further demonstrating the generality of the middomain effect.

islands. Given the urgent need to understand the factors dictating the distribution of tropical diversity, it is vital that new ecological models such as the mid-domain effect are explored over the broadest taxonomic spectrum and across a wide range of geographical scales and locations. Islands such as New Guinea present significant challenges to our understanding of biodiversity being extremely rich biologically and yet poorly explored. Detailed knowledge of tropical plant diversity is still limited by a shortage of good collections, especially for plant families such as palms, which are difficult to collect and often under-represented in herbaria. Although collection densities of palms in New Guinea remain low, a one-dimensional approach yields a wellsampled gradient that can be analysed effectively using null models. It seems likely that many similar datasets exist and we encourage others to undertake analyses such as ours. This study adds significantly to the growing body of evidence supporting the influence of the mid-domain effect. This effect is undoubtedly an important consideration for any analysis of richness gradients. Using a simple yet novel method, we have demonstrated that it is essential to account for the influence of area on species richness patterns when exploring concepts such as the mid-domain effect. In our study, the observed richness patterns can be explained to a strikingly large extent by the combined role of the mid-domain effect and area. In future, these factors need to be considered in a multivariate context to understand their ultimate importance relative to additional environmental variables. While findings such as ours may continue to be controversial in the short term, we believe that, in time, the role of the mid-domain effect will become increasingly accepted as one of several factors determining species richness patterns. Acknowledgements / Many people contributed to the New Guinea palm database, in particular Roy Banka, Anders Barfod, Kate Davis, Anders Kjaer, Meesha Patel and Helen Sanderson. We thank the staff of the herbaria at Aarhus, Brisbane, Bogor, Kew, Lae and Leiden for kindly providing access to collections and data. We thank Robert Colwell, David Lees and John-Arvid Grytnes for valuable discussions and numerous comments on earlier versions of the manuscript. Carsten Rahbek made many suggestions that improved the paper. This work was supported by a student internship from the Royal Botanic Gardens, Kew to SPB, a studentship from the Bernard Sunley Charitable Trust to NAB and by funding from the BAT Biodiversity Partnership to the Palms of New Guinea project.

References Conclusions Few studies of this kind have focused on vascular plants and even fewer have been based on taxa from tropical 308

Anon. 1999. ArcView GIS. / ESRI, Redlands, California. Anon. 2001. GTOPO30 / Global topographic data. / USGS, B/http://edcdaac.usgs.gov/gtopo30/gtopo30.html /, last accessed: 30 Sept 2002. Arrhenius, O. 1921. Species and area. / J. Ecol. 9: 95 /99. ECOGRAPHY 27:3 (2004)


Baker, W. J. 2002. The palms of New Guinea project. / Fl. Males. Bull. 13: 35 /37. Bokma, F. and Mo¨nkko¨nen, M. 2000. The mid-domain effect and the longitudinal dimension of continents. / Trends Ecol. Evol. 15: 288 /289. Brown, J. H. 1971. Mammals on mountaintops: nonequilibrium insular biogeography. / Am. Nat. 105: 467 /478. Colwell, R. K. 2000. RangeModel: a Monte Carlo simulation tool for assessing geometric constraints on species richness. Ver. 3. / User’s guide and application, B/http://viceroy.eeb. uconn.edu/asn/. Colwell, R. K. and Hurtt, G. C. 1994. Nonbiological gradients in species richness and a spurious Rapoport effect. / Am. Nat. 144: 570 /595. Colwell, R. K. and Lees, D. C. 2000. The mid-domain effect: geometric constraints on the geography of species richness. / Trends Ecol. Evol. 15: 70 /76. Colwell, R. K., Rahbek, C. and Gotelli, N. J. in press. The middomain effect and species richness patterns: what have we learned so far. / Am. Nat. Connor, E. F. and McCoy, E. D. 1979. The statistics and biology of the species-area relationship. / Am. Nat. 113: 791 /833. Crawley, M. J. and Harral, J. E. 2001. Scale dependence in plant biodiversity. / Science 291: 864 /868. Darwin, C. 1859. On the origin of species by means of natural selection or the preservation of favoured races in the struggle for life. / John Murray, London. Diniz-Filho, J. A. F. et al. 2002. Null models and spatial patterns of species richness in South American birds of prey. / Ecol. Lett. 5: 47 /55. Grytnes, J. A. and Vetaas, O. R. 2002. Species richness and altitude: a comparison between null models and interpolated plant species richness along the Himalayan altitudinal gradient, Nepal. / Am. Nat. 519: 294 /304. Heaney, L. R. 2001. Small mammal diversity along elevational gradients in the Philippines: an assessment of patterns and hypotheses. / Glob. Ecol. Biogeogr. 10: 15 /39. Jetz, W. and Rahbek, C. 2001. Geometric constraints explain much of the species richness pattern in African birds. / Proc. Natl. Acad. Sci. USA 98: 5661 /5666. Jetz, W. and Rahbek, C. 2002. Geographic range size and determinants of avian species richness. / Science 297: 1548 / 1551. Kaspari, M., O’Donnell, S. and Kercher, J. R. 2000. Energy, density, and constraints to species richness: studies of ant assemblages along a productivity gradient. / Am. Nat. 155: 280 /293. Kessler, M. 2001. Pteridophyte species richness in Andean forests in Bolivia. / Biodiv. Conserv. 10: 1473 /1495. Koleff, P. and Gaston, K. J. 2001. Latitudinal gradients in diversity: real patterns and random models. / Ecography 24: 341 /351. La Ferla, B. et al. 2002. Continental scale patterns of biodiversity: can higher taxa accurately predict African plant distributions? / Bot. J. Linn. Soc. 138: 225 /235. Lawton, J. H., MacGarvin, M. and Heads, P. A. 1987. Effects of altitude on the abundance and species richness of insect herbivores on bracken. / J. Anim. Ecol. 56: 147 /160. Lees, D. C., Kremen, C. and Andriamampianina, L. 1999. A null model for species richness gradients: bounded range overlap of butterflies and other rainforest endemics in Madagascar. / Biol. J. Linn. Soc. 67: 529 /584. Lyons, S. K. and Willig, M. R. 1997. Latitudinal patterns of range size: methodological concerns and empirical evaluations for New World bats and marsupials. / Oikos 79: 568 / 580. MacArthur, R. H. 1972. Geographical ecology: patterns in the distribution of species. / Harper and Rowe. Nor, S. M. 2001. Elevational diversity patterns of small mammals on Mt. Kinabalu, Sabah, Malaysia. / Glob. Ecol. Biogeogr. 10: 41 /62.

ECOGRAPHY 27:3 (2004)

Ohlemu¨ller, R. and Wilson, J. B. 2000. Vascular plant species richness along latitudinal and altitudinal gradients: a contribution from New Zealand temperate rainforests. / Ecol. Lett. 3: 262 /266. Patterson, B. D. et al. 1998. Contrasting patterns of elevational zonation for birds and mammals in the Andes of southeastern Peru. / J. Biogeogr. 25: 593 /607. Plotkin, J. B. et al. 2000. Predicting species diversity in tropical forests. / Proc. Natl. Acad. Sci. USA 97: 10850 / 10854. Pineda, J. and Caswell, H. 1998. Bathymetric species-diversity patterns and boundary constraints on vertical range distributions. / Stud. Oceanogr. 45: 83 /101. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern? / Ecography 18: 200 /205. Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in Neotropical birds. / Am. Nat. 149: 875 /902. Rapoport, E. H. 1982. Areography: geographical strategies of species. The Fundacio´n Bariloche Series Vol. 1. / Pergamon Press. Rickart, E. A. 2001. Elevational diversity gradients, biogeography and the structure of montane mammal communities in the intermountain region of North America. / Glob. Ecol. Biogeogr. 10: 77 /100. Rosenzweig, M. L. 1992. Species diversity gradients: we know more and less than we thought. / J. Mammal. 73: 715 /730. Rosenzweig, M. L. 1995. Species diversity in space and time. / Cambridge Univ. Press. Rosenzweig, M. L. and Abramsky, Z. 1993. How are diversity and productivity related? / In: Ricklefs, R. E. and Schluter, D. (eds), Species diversity in ecological communities: historical and geographical perspectives. Univ. of Chicago Press, pp. 52 /65. Sa´nchez-Cordero, V. 2001. Small mammal diversity along elevational gradients in Oaxaca, Mexico. / Glob. Ecol. Biogeogr. 10: 63 /76. Sanders, N. 2002. Elevational gradients in ant species richness: area, geometry, and Rapoport’s rule. / Ecography 25: 25 / 32. Stevens, G. C. 1989. The latitudinal gradient in geographic range: how so many species coexist in the tropics. / Am. Nat. 133: 240 /256. Stevens, G. C. 1992. The elevational gradient in altitudinal range: an extension of Rapoport’s latitudinal rule to altitude. / Am. Nat. 140: 893 /911. Wallace, A. R. 1876. The geographical distribution of animals. / Macmillan. Whittaker, R. H. 1960. Vegetation of the Siskiyou Mountains, Oregon and California. / Ecol. Monogr. 30: 279 /338. Whittaker, R. J., Willis, K. J. and Field, R. 2001. Scale and species richness: toward a general hierarchical theory of species diversity. / J. Biogeogr. 28: 453 /470. Willdenow, K. L. 1805. The principles of botany, and vegetable physiology. / Blackwood, Cadell and Davies, London. Williams, C. B. 1943. Area and number of species. / Nature 152: 264 /267. Williams, P. H. and Gaston, K. J. 1994. Measuring more of biodiversity: can higher-taxon richness predict wholesale species richness. / Biol. Conserv. 67: 211 /217. Williamson, M. 1988. Relationship of species number to area, distance and other variables. / In: Myers, A. A. and Giller, P. S. (eds), Analytical biogeography. Chapman and Hall, pp. 94 /115. Willig, M. R. and Lyons, S. K. 1998. An analytical model of latitudinal gradients of species richness with an empirical test for marsupials and bats in the New World. / Oikos 81: 93 /98. Wolda, H. 1987. Altitude, habitat and tropical insect diversity. / Biol. J. Linn. Soc. 30: 313 /323.

309


Appendix 1. Ranges of equal-area bands for each of 5, 10 and 15 bands used. Minimum and maximum elevations in metres for each equal-area band along the elevation gradient are given (as used in Method 2). Ranges are shown separately for each data type (Field and DEM). Band

5 Bands 1 2 3 4 5

Field data, Species

Field data, Genera

DEM data, Species

DEM data, Genera

Min (m)

Min (m)

Min (m)

Min (m)

Max (m)

Max (m)

Max (m)

Max (m)

0.5 14.5 122.8 304.6 923.2

14.5 122.8 304.6 923.2 2800.0

0.5 12.7 113.3 296.6 855.4

12.7 113.3 296.6 855.4 2438.0

0.5 14.5 122.8 304.6 923.7

14.5 122.8 304.6 923.7 2803.0

0.5 14.5 122.7 304.6 922.6

14.5 122.7 304.6 922.6 2797.0

10 Bands 1 0.5 2 1.2 3 14.5 4 56.7 5 122.8 6 219.8 7 304.6 8 550.3 9 923.2 10 1631.8

1.2 14.5 56.7 122.8 219.8 304.6 550.3 923.2 1631.8 2800.0

0.5 1.2 12.7 52.7 113.3 206.2 296.6 487.0 855.4 1510.8

1.2 12.7 52.7 113.3 206.2 296.6 487.0 855.4 1510.8 2438.0

0.5 1.2 14.5 56.7 122.8 219.9 304.6 550.8 923.7 1633.5

1.2 14.5 56.7 122.8 219.9 304.6 550.8 923.7 1633.5 2803.0

0.5 1.2 14.5 56.6 122.7 219.7 304.6 550.0 922.6 1629.9

1.2 14.5 56.6 122.7 219.7 304.6 550.0 922.6 1629.9 2797.0

15 Bands 1 0.5 2 1.0 3 1.5 4 14.5 5 40.8 6 74.1 7 122.8 8 185.4 9 252.7 10 304.6 11 435.7 12 630.8 13 923.2 14 1467.3 15 2017.2

1.0 1.5 14.5 40.8 74.1 122.8 185.4 252.7 304.6 435.7 630.8 923.2 1467.3 2017.2 2800.0

0.5 1.0 1.5 12.7 37.4 69.4 113.3 172.4 238.5 296.6 376.7 587.8 855.4 1262.5 1722.5

1.0 1.5 12.7 37.4 69.4 113.3 172.4 238.5 296.6 376.7 587.8 855.4 1262.4 1722.5 2438.0

0.5 1.0 1.5 14.5 40.8 74.1 122.8 185.5 252.9 304.6 436.1 631.3 923.7 1468.7 2019.7

1.0 1.5 14.5 40.8 74.1 122.8 185.5 252.9 304.6 436.1 631.3 923.7 1468.7 2019.7 2803.0

0.5 1.0 1.5 14.5 40.8 74.0 122.7 185.3 252.6 304.6 435.4 630.3 922.6 1465.9 2015.0

1.0 1.5 14.5 40.8 74.0 122.7 185.3 252.6 304.6 435.4 630.3 922.6 1465.9 2015.0 2797.0

310

ECOGRAPHY 27:3 (2004)


ECOGRAPHY 27: 733 /740, 2004

Grasshopper populations across 2000 m of altitude: is there life history adaptation? Daniel Berner, Christian Ko¨rner and Wolf U. Blanckenhorn

Berner, D., Ko¨rner, Ch. and Blanckenhorn, W. U. 2004. Grasshopper populations across 2000 m of altitude: is there life history adaptation? / Ecography 27: 733 /740. Life history differentiation along climatic gradients may have allowed a species to extend its geographic range. To explore this hypothesis, we compared eleven Omocestus viridulus (Orthoptera: Acrididae) populations along an altitudinal gradient from 410 to 2440 m in Switzerland, both in the field and laboratory. In situ temperature records indicated a striking decline in available heat sums along the gradient, and field populations at high altitudes reached egg hatching and adulthood much later in the year than at low elevation. The reproductive period at high altitude is thus severely limited by season length, especially during a cool year. However, controlled environment experiments revealed that intrinsic rates of embryonic and juvenile development increased with the populations’ altitude of origin. This countergradient variation is largely genetic and conforms to predictions of life history theory. No corresponding differentiation in the overwintering egg stage, a pivotal determinant of phenology, was found. This trait seems conserved within the gomphocerine grasshopper subfamily. Although we found evidence for altitudinal adaptation in development, the potential of O. viridulus to adapt to cool alpine climates appears restricted by a phylogenetic constraint. D. Berner (daniel.berner@fal.admin.ch), Agroscope FAL Reckenholz, Swiss Federal Research Station for Agroecology and Agriculture, Reckenholzstr. 191, CH-8046 Zu¨rich, Switzerland. / Ch. Ko¨rner, Inst. of Botany, Univ. of Basel, Scho˝nbeinstrasse 6, CH-4056 Basel, Switzerland. / W. U. Blanckenhorn, Zoological Museum, Univ. of Zu¨rich, Winterthurerstrasse 190, CH-8057 Zu¨rich, Switzerland.

The seasonal recurrence of adverse climatic conditions is a principal force shaping ectotherm life cycles in temperate regions. Growth, development, reproduction and dormancy need to be coordinated and timed in relation to the available growing season (Taylor and Karban 1986, Danks 1994). The set of adaptations, which synchronizes the life cycle with the growing season, is reflected in the organisms’ phenology (Tauber et al. 1986). The length of the growing season generally declines with increasing latitude and altitude. Thus, geographically widespread species have to cope with a variety of climatic conditions, which can basically be achieved in two / not mutually exclusive / ways. Firstly, a generalist genotype may display plastic responses in

relation to environmental conditions (Gotthard and Nylin 1995, Schlichting and Pigliucci 1998). Phenotypic plasticity in traits relevant to seasonal timing has been documented and interpreted in adaptive terms in several insect species (Tanaka and Brookes 1983, Nylin 1994, Blanckenhorn 1997, Kingsolver and Huey 1998). Secondly, spatial variation in selection pressures may give rise to genetic differentiation between populations due to natural selection. Prerequisites are heritable genetic variation and restricted gene flow between local populations (Slatkin 1987). Both responses, local adaptation and phenotypic plasticity, may allow a species to extend its distribution across a range of altitudes and latitudes. Several studies of ectotherms report genetic life history

Accepted 30 July 2004 Copyright # ECOGRAPHY 2004 ISSN 0906-7590 ECOGRAPHY 27:6 (2004)

733


differentiation in relation to systematic geographic variation in climate (Masaki 1967, Berven and Gill 1983, Dingle et al. 1990, Ayres and Scriber 1994, Blanckenhorn and Fairbairn 1995, Telfer and Hassall 1999, Merila¨ et al. 2000; but see Lamb et al. 1987). However, almost all studies focus on latitude, whereas evidence for altitudinal life history adaptation in animals is exceedingly scarce. This distinction matters indeed: altitudinal changes in climate typically occur on a particularly small spatial scale, where continuous gene flow is likely to impede genetic differentiation unless strong local selection is acting. An insect species with a remarkable altitudinal distribution is the grasshopper Omocestus viridulus (Acrididae: Gomphocerinae). In Switzerland, it occurs in grasslands from below 400 m to above 2500 m (Thorens and Nadig 1997), making it a suited system for the study of altitudinal adaptation. Over the whole range, it displays an annual life cycle, which includes egg hatch in spring, four larval instars followed by adult molt in summer, and reproduction until autumn. The species overwinters as an egg in embryonic diapause (Ingrisch and Ko¨hler 1998). This dormant phase is characterized by suppressed development, reduced metabolism, and high tolerance to harsh environmental conditions (Danks 1987, Leather et al. 1993). The developmental stage during diapause strongly influences later phenology, as it determines how many developmental steps an embryo has to pass through before hatch in spring. In addition, seasonal timing can be achieved through adjustment of development rates (Danks 1987). Higher rates of embryonic and larval development allow reaching adulthood, and thus the subsequent reproductive phase, sooner. The diapause stage, as well as embryonic and larval development rates, can be identified as chief traits determining phenology, and are therefore of high significance to geographic life history adaptation. Along the altitudinal gradient, O. viridulus faces a decline in the length of the growing season, which is likely to require phenological adjustment. Moreover, the species is sedentary (Ingrisch and Ko¨hler 1998) and most populations are separated to some extent by migration barriers, suggesting rather low levels of gene flow. For these reasons, we hypothesize that local life history adaptation, rather than phenotypic plasticity, allowed the grasshopper to extend its distribution to the wide range of altitudes. In this case, the species would represent a fine-grained patchwork of local demes, which are differentiated in traits relevant to seasonal timing. We address the hypothesis by both field and laboratory approaches. In a first step, the natural temperature regimes and their effect on field phenologies along the gradient are explored. In a second step, we compare populations with respect to developmental rates and diapause characteristics in common laboratory environ734

ments. The latter approach serves to remove environmental variation and reveal genetic differences in life histories, if they occur.

Materials and methods Study populations The present investigation includes a total of eleven Omocestus viridulus populations. The study sites were chosen to form a transect from the Swiss lowland into the Alps, covering an altitudinal gradient of 2000 m (Fig. 1). Distribution data were provided by the Swiss center of cartography of the fauna (CSCF). We considered only sites where large populations had been reported over several years. Some 110 km separate the furthermost sites. Although this spatial scale is relatively small, the study populations can be viewed as reasonably independent, as most populations of this widespread species are isolated to some degree by natural and human dispersal barriers (e.g. forests or farmland).

Field studies To estimate the length of the growing season in the field, temperatures were recorded hourly during the 2002 season at sites 1, 5, 7 and 10 (see Fig. 1) by means of data loggers (‘‘StowAway TidbiT’’, Onset Computer Corporation, Bourne, MA, USA). We were primarily interested in the conditions the embryos (eggs) experience. Since O. viridulus lays its clutches into the top soil layer or at the base of grass tussocks (Ingrisch 1983, Berner unpubl.), we positioned the loggers’ sensors at 1 cm soil depth under natural vegetation cover. Two loggers were used per site, and their measurements averaged for all calculations. Two different indices of

Fig. 1. The rectangle in the outline map shows the location of the study area in Switzerland. The sampling sites and corresponding altitudes are 1) Neerach 410 m, 2) Birchwil 540 m, 3) Scho¨nenberg 670 m, 4) Ba¨retswil 830 m, 5) Bendel 1055 m, 6) Na¨fels 1350 m, 7) Speer 1610 m, 8) Flumserberg 1850 m, 9) Gamserrugg 2060 m, 10) Pizol 2215 m, 11) Ho¨rnli 2440 m. The triangle denotes the city of Zu¨rich (47822?N, 8831?E). ECOGRAPHY 27:6 (2004)


season length were computed: one index uses the date at which 118C was exceeded for the first time. This date roughly corresponds to the initiation of postdiapause embryonic development, which is inhibited at temperatures below ca 118C (Wingerden et al. 1991). The second index uses the cumulative degree hours above 148C between the first appearance of larvae at each site and the end of the year. This approximates the season length for larvae and adults. 148C was chosen based on a study by Hilbert and Logan (1983), because postembryonic development thresholds were unavailable for the species. However, grasshoppers are known to increase body temperature by basking (Begon 1983, Chappell 1983). Hence, the latter index must be viewed as a relatively crude, but still informative, estimate of the thermally effective season length. Field phenologies were studied at the same sites and in the same year as the temperature records. Each site was visited in regular intervals of six to eleven days over the growing season. We censused by direct observation along transects, noting the stage (larval instars 1 /4, adult) of each grasshopper. (The insertion of an additional larval instar reported from other gomphocerine grasshoppers (references in Ingrisch and Ko¨hler 1998), occurred neither in the field nor laboratory.) Although males reach adulthood slightly earlier than females in this species, the sexes were pooled post hoc for simplicity. This did not influence the results substantially. As 2002 was a rather cool and cloudy year and 2003 was particularly sunny, we also checked the stage composition at site 1 in late June and site 10 in mid July 2003. These snapshots during the second year allowed a comparison of phenologies between climatically rather different seasons.

Breeding techniques and laboratory experiments To establish breeding populations, ca 14 individuals of each sex were caught at the beginning of the reproductive period at each of the eleven sites. The populations 2, 5, 7 and 10 were sampled in 2001, all others in 2002. The animals were kept in groups in cages in a greenhouse under natural photoperiods until death. Field-cut grass (largely of Dactylis glomerata and Agropyron repens ) was provided as food. Egg pods were collected twice a week as they were laid, put in plastic vials containing moist vermiculite, and incubated at 258C for 35 d which allows the embryos to reach the diapause stage (Wingerden et al. 1991). After this, the clutches were stored at 58C. Postdiapause embryonic development time was studied in a climate chamber set to a photoperiod of 14 h at 278C. Night temperature was 88C. All eggs had spent at least three month at 58C, which is enough to break diapause (Ingrisch and Ko¨hler 1998). The vials were ECOGRAPHY 27:6 (2004)

inspected twice daily for newly hatched larvae, until no further hatch occurred. Individual hatch dates were noted and converted to degree hours with 118C as threshold. We tested clutch medians in a general linear model (GLM), with study year as a fixed factor and altitude of origin (/populations) as a continuous covariate. Effective sample size varied between 18 and 79 clutches per population. To verify the robustness of our 278C results, a subset of the clutches laid in 2002 was incubated at 198C, but otherwise treated and analyzed in the same way. Here, sample size varied between 14 and 48. To investigate larval development time, hatchlings were immediately transferred to another climate chamber with a 16 h photoperiod at 328C and a night temperature of 108C. This experiment was conducted with the seven 2002 populations only. Larvae were kept clutchwise in plastic containers (19 cm high, 8 cm in diameter) in groups of six at most. Small pots with a grass mixture provided food. Adult emergence was checked twice daily, and individual degree hours for larval development were determined using 148C as threshold. We analyzed clutch medians using GLM. Altitude was entered as a covariate, and sex as a fixed factor, since the sexes differed in development time. To assess diapause stages, ten random clutches from each of six populations (1, 2, 4, 7, 8, 9) were removed from the cold. The outer layer (chorion) of every single egg in the clutch was scraped off with a fine blade under a stereomicroscope so that the embryo could be seen and assigned a developmental stage. We used the classification system of Cherrill (1987), which divides the continuous process of differentiation to the fully developed embryo into twenty discrete morphological steps. Based on individual eggs, the clutch median stage was determined and treated as one data point. Differences between populations were tested using one-way ANOVA and a distribution-free Kruskal-Wallis test. All statistics were performed with SPSS 11.1.

Results Temperature regimes Daily mean temperatures of the top soil layer display a sharp decline with altitude (Fig. 2). Over the summer months, mean temperatures at 2215 m remain ca 78C below those at 410 m. Moreover, snow cover maintains spring temperatures around zero at the high elevation sites, most dramatically at 2215 m. Indeed, the very first hourly temperature record above the estimated embryonic threshold of 118C occurs as late as on the 31 May at 2215 m (Table 1). At the low elevation sites this threshold is exceeded almost three month earlier. Season length estimated as degree hours above 148C shows a more than tenfold reduction from 410 m to 2215 m 735


Fig. 2. Daily mean temperatures during the 2002 season at altitudes of 410 m (1), 1055 m (2), 1610 m (3) and 2215 m (4).

altitude (Table 1). Hardly any hourly records above 148C were made after the end of August at the highest site. Roughly speaking, postembryonic development was possible during seven months at low altitude, whereas only three months were available at the highest site in 2002.

population reaches adulthood by late June in both years. At 2215 m, however, the phenological difference between the two seasons is much larger. Clearly, the high altitude grasshoppers are delayed in both 2002 and 2003 relative to the lowland, but the phenological delay is more pronounced in the cooler year of 2002.

Field phenologies

Laboratory experiments

The phenology curves in Fig. 3 indicate a marked delay in the emergence of first instar larvae at the high elevation sites, where the first hatchlings appeared three (1610 m) and seven (2215 m) weeks later than at the lowest location. The delay carries over to the adults: the graphically estimated dates at which each population reaches an adult frequency of 75% are 26 June (410 m), 31 June (1055 m), 23 July (1610 m) and 19 August (2215 m). Consequently, adult emergence at the highest site is delayed by almost two months compared to the lowland site. During the 2002 season lowland grasshoppers had already started reproducing when first instar larvae just started hatching at high altitude. In accordance with the temperature regimes, the difference in phenology between the sites at 410 m and 1055 m is small. The comparison of 2002 (cool year) and 2003 (warm year) reveals a small difference in the low elevation phenology (Table 2). The greatest majority of the 410 m

There is a clear relationship between embryonic development time in the laboratory at 278C and a population’s altitude of origin (Fig. 4): high altitude embryos complete development faster, resulting in earlier hatching of the first instar larvae (F1,431 /50.9, pB/0.001). However, the maximal difference between populations in development time amounts to some ten percent only. Expressed in real time, the population averages declined from 14.1 to 12.3 d. The year factor is also significant because temperature conditions differed slightly between the years (different climate chamber types; F1,431 /40.5, p B/0.001). Faster development of the high altitude embryos was also found at the lower experimental temperature of 198C. The correlation of population averages of embryonic development time at the two incubation temperatures yields coefficients of 0.85 (Pearson’s r, p/0.015) and 0.93 (Spearman’s rank, p /0.003). The duration of development through all larval instars to adults clearly gets shorter with altitudinal origin (Fig. 5; F1,223 /51.9, pB/0.001), similar to embryonic development. As in the field, males always reach adulthood earlier than females (F1,223 /16.1, pB/ 0.001), but the altitudinal response is similar in the sexes, as indicated by a nonsignificant interaction (F1,223 /1.3, p/0.26). Again, the difference between the fastest and slowest population is only about ten percent. In real time, the population averages for larval development ranged from 23.6 to 20.9 d in males and

Table 1. Indices of the 2002 season length at four altitudes, based on hourly temperature records at 1 cm soil depth. Altitude (m) 410 1055 1610 2215

Date of first record/118C 8 March 8 March 5 April 31 May

Degree hours/148C* 12536 10019 3182 904

* From the onset of larval hatch to the end of the year.

736

ECOGRAPHY 27:6 (2004)


Fig. 3. Field phenologies of O. viridulus at four altitudes during the season of 2002. The vertical line represents the first observation of hatchlings.

from 25.2 to 22.3 d in females for low and high altitude, respectively. The stage of embryonic diapause does not significantly differ between the populations, and no altitudinal trend is evident (Fig. 6; ANOVA F5,54 /1.34, p/0.26; Kruskal-Wallis x25 /8.09, p/0.15). In all O. viridulus populations studied, the vast majority of embryos diapauses at developmental stage nine, which corresponds to stage IVd of Cherrill (1987). The embryos are then arrested just before the onset of embryonic rotation. Most clutches contained some retarded eggs, but no single embryo developed further than stage nine.

Discussion Our laboratory study documents increasing rates of embryonic and juvenile development in O. viridulus with increasing altitude. As a consequence, high altitude grasshoppers attain adulthood in shorter time than their low altitude counterparts when grown in a common environment. In contrast, the diapause stage, another

key determinant of phenology, shows no difference among the populations. The field work indicates a time constraint on the life cycle of high altitude animals. The cooler high elevation temperature regimes substantially delay larval hatch and adulthood. In a cloudy year like 2002, the reproductive life span of alpine grasshoppers is thus severely truncated and reproductive success very poor. Moreover, a considerable fraction of the produced eggs may fail to reach the overwintering stage due to insufficient late season heat. This was shown to entail delayed hatching in Chorthippus brunneus (Cherrill and Begon 1991) and survival costs in Camnula pellucida (Pickford 1966). Certainly, cool and cloudy years are severe selection events at the species’ upper range margin. Only during particularly sunny seasons like 2003 is the reproductive period sometimes terminated

Table 2. Frequency (%) of O. viridulus instars in the years 2002 and 2003 at low and high elevation. Note that the two populations were censused on different dates.

2nd instar 3rd instar 4th instar Adult

410 m, late June

2215 m, mid July

2002

2003

2002

2003

/ 3 18 79

/ / 3 97

21 64 15 /

/ 9 31 60

ECOGRAPHY 27:6 (2004)

Fig. 4. Physiological time required by O. viridulus populations from different altitudes for postdiapause embryonic development. Data from 2002 (k) and 2003 (m). Degree hours were calculated using 118C as threshold.

737


Fig. 5. Physiological time required by female (k) and male (m) grasshoppers from different altitudes for larval development. A threshold of 148C was used for calculation.

by intrinsic senescence at both low and high elevation. Thus the variance in the available season length increases with altitude. Under such a seasonal time constraint, annual organisms face the problem of optimally allocating time to development and reproduction. Life history models predict that decreased season length will favor faster development and hence decreased time to maturity (Cohen 1976, Roff 1980, 2002, Rowe and Ludwig 1991, Abrams et al. 1996). In Omocestus viridulus with its wide altitudinal distribution, therefore, we expected differences in traits determining postdiapause development time. The higher rates of development exhibited under laboratory conditions by the alpine populations thus conform well to the theoretical prediction. A genetic basis to the acceleration of development is strongly suggested because, firstly, maternal influence on offspring embryonic development appears negligible in the related Chorthippus parallelus (Ko¨hler 1983). Secondly, our field records indicate that the temperatures used in the laboratory may be experienced by all populations in the field. Absorption of solar radiation may allow even

Fig. 6. Embryonic developmental stage at diapause in O. viridulus populations from six altitudes. Plotted are clutch medians (N/10 per population).

738

high altitude larvae to rise body temperature well above 328C. Furthermore, embryonic development rates were found to increase with altitude at incubation temperatures of both 198C and 278C. Strong genotype by environment interactions are thus excluded. For these reasons we suggest that the observed developmental differences are robust and reflect an adaptive strategy. Increased embryonic and larval development rates allow high altitude animals to reach maturity relatively faster, and hence prolong reproductive life span when time is short. The hypothesis of within species life history differentiation on a small spatial scale is thus confirmed. Apparently the level of gene flow between the O. viridulus populations is too low to counteract local adaptation. Omocestus viridulus agrees well with some other ectotherms in which differentiation of development along gradients in season length has been documented (Masaki 1967, Berven and Gill 1983, Dingle et al. 1990, Ayres and Scriber 1994, Telfer and Hassall 1999). The shortening of development time proves a common adaptive response to seasonal time constraints. Intraspecific differentiation in traits related to phenology may be quite frequent in annual ectotherms with relatively long development times covering wide geographic ranges. However, our field surveys make it clear that the cool climates at high elevation retard grasshopper phenologies despite higher intrinsic rates of development in those populations. Thus, the high elevation grasshoppers are only partly able to compensate the delaying environmental influence on time to adulthood. This agrees with the relatively modest level of differentiation found in the laboratory. As the genetic response along the altitudinal gradient is opposed to the phenotypic response to the environmental conditions, O. viridulus provides an example of countergradient variation (Conover and Schultz 1995). A merely phenotypic comparison of development times within the species would have failed to demonstrate altitudinal differentiation. At the proximate level, increased development rates may be associated with metabolic temperature compensation (Danks 1987). Hadley and Massion (1985) for example report increased metabolic rates in high altitude populations of the grasshopper Aeropedellus clavatus. Likewise, latitudinal differences in metabolism were found in the butterfly Papilio canadensis (Ayres and Scriber 1994). However, physiological traits of O. viridulus populations have not been compared so far. A question arising from the observed patterns is why higher intrinsic rates of development did not evolve in the lowland populations. What could be the disadvantage of a similarly rapid development as at high elevation? On the one hand, adverse climatic conditions early in the season are likely to select against precocious ECOGRAPHY 27:6 (2004)


larval emergence. Carrie`re et al. (1996), for example, demonstrate a mortality cost associated with precocious larval hatch in Gryllus pennsylvanicus, due to unfavorable temperature conditions. Furthermore, trade-offs with other fitness components could maintain developmental rates below the physiological potential exhibited by the alpine populations (Schluter et al. 1991, Stearns 1992, Roff 2002). For instance, Tatar et al. (1997) found increased senescence in Melanoplus sanguinipes grasshoppers from high elevation sites compared to the slower developing low altitude animals. Likewise, given that juvenile development time, adult size, and fecundity are often correlated positively (Roff 1980, 2002, Rowe and Ludwig 1991, Honek 1993), a shortened juvenile development will negatively affect fecundity. According to Orr (1996) this is the case in M. sanguinipes. Most probably, elevated rates of development bear fitness costs and are selected against in the absence of a seasonal time constraint on the life cycle, as is the case at low elevation. However, low altitude seasons appear still too short for two generations, as the species exhibits an annual life cycle throughout its range. Besides embryonic and larval development rates, the stage of overwintering strongly determines time to adulthood. Central European grasshoppers of the gomphocerine subfamily are believed to show an obligatory diapause during embryonic development. According to some studies, the dormant stage is inserted shortly before embryonic rotation (Ko¨hler 1991, Ingrisch and Ko¨hler 1998). This stage has been designated IVd by Cherrill (1987). However, geographic variation in diapause stage within an insect species is possible in principle (references in Tauber et al. 1986), but has not been investigated to date in any European grasshopper. In O. viridulus we found no such altitudinal differentiation: in all populations, most clutches mainly contained embryos arrested at the aforementioned stage, and no embryo developed further. Thus, the species displays an obligatory diapause and is uniform with respect to the stage of dormancy, conforming to other members of the subfamily. This finding stands in striking contrast to other orthopteran studies, which document intraspecific variation in embryonic diapause stage and/or expression along gradients in season length (Mousseau and Roff 1989, Groeters and Shaw 1992, Tanaka 1994, Dingle and Mousseau 1994, Bradford and Roff 1995). The catantopine grasshopper Melanoplus sanguinipes, for example, occurs from sea level to above 3800 m (Chappell 1983) and displays enormous variation in embryonic diapause stage within its North American range. High elevation populations overwinter almost completely developed and attain the hatching state at low heat sums. This is interpreted as an effective means to decrease postdiapause development time under short seasons (Dingle et al. 1990, Dingle and Mousseau 1994). In M. sanguinipes, adult emergence at above 2600 m and at sea level ECOGRAPHY 27:6 (2004)

happens roughly at the same time! Not surprisingly, another catantopine, M. frigidus, is the highest reaching species in the Alps (Carron 1996). This grasshopper subfamily illustrates the importance of flexibility in the overwintering stage for altitudinal adaptation. In this light, the lack of variation in the stage of diapause within O. viridulus likely represents a phylogenetic constraint (Gould 1989, Stearns 1992, Schlichting and Pigliucci 1998) to altitudinal range expansion. The stage of dormancy as a conserved trait within the gomphocerine lineage precludes tuning of development in a way expected to be optimal under seasonal time constraints. However, this has to be confirmed by investigating other, closely related species. To summarize, O. viridulus exhibits altitudinal differentiation in development as an adaptive response to selection imposed by local climates. However, the potential for altitudinal adaptation is limited by the invariant stage of overwintering diapause, probably indicating a phylogenetic constraint within the gomphocerine grasshopper lineage. As a consequence, the degree of climatic compensation displayed by field populations along the altitudinal gradient is rather low as compared to other ectotherms. Acknowledgements / We thank T. Walter for organizing a parttime employment for D. Berner at the Swiss Federal Research Station for Agroecology and Agriculture (Agroscope FAL Reckenholz), which made this research project possible. J. Samietz, G. Ko¨hler and E.-F. Kiel gave valuable advice regarding the research design and/or grasshopper rearing techniques. S. Bosshart and M. Waldburger provided technical help and materials. P. Streckeisen kindly helped operate the climate chambers and made greenhouse space available. Distribution data of the species were obtained from the Swiss center of cartography of the fauna (CSCF). To all these people and institutions we are most grateful.

References Abrams, P. A. et al. 1996. The effect of flexible growth rates on optimal sizes and development times in a seasonal environment. / Am. Nat. 147: 381 /395. Ayres, M. P. and Scriber, J. M. 1994. Local adaptation to regional climates in Papilio canadensis (Lepidoptera: Papilionidae). / Ecol. Monogr. 64: 465 /482. Begon, M. 1983. Grasshopper populations and weather: the effects of insolation on Chorthippus brunneus. / Ecol. Entomol. 8: 361 /370. Berven, K. A. and Gill, D. E. 1983. Interpreting geographic variation in life-history traits. / Am. Zool. 23: 85 /97. Blanckenhorn, W. U. 1997. Altitudinal life history variation in the dung flies Scathophaga stercoraria and Sepsis cynipsea . / Oecologia 109: 342 /352. Blanckenhorn, W. U. and Fairbairn, D. J. 1995. Life history adaptation along a latitudinal cline in the water strider Aquarius remigis (Heteroptera: Gerridae). / J. Evol. Biol. 8: 21 /41. Bradford, M. J. and Roff, D. A. 1995. Genetic and phenotypic sources of life history variation along a cline in voltinism in the cricket Allonemobius socius. / Oecologia 103: 319 /326. Carrie`re, Y., Simons, A. M. and Roff, D. A. 1996. The effect of the timing of post-diapause egg development on survival,

739


growth and body size in Gryllus pennsylvanicus. / Oikos 75: 463 /470. Carron, G. 1996. Do alpine acridids have a shortened postembryonic development? / Articulata 11: 49 /72. Chappell, M. A. 1983. Metabolism and thermoregulation in desert and montane grasshoppers. / Oecologia 56: 126 /131. Cherrill, A. and Begon, M. 1991. Oviposition date and pattern of embryogenesis in the grasshopper Chorthippus brunneus (Orthoptera: Acrididae). / Holarct. Ecol. 14: 225 /233. Cherrill, A. J. 1987. The development and survival of the eggs and early instars of the grasshopper Chorthippus brunneus (Thunberg) in north west England. / Unpubl. Ph.D. thesis, Univ. of Liverpool. Cohen, D. 1976. The optimal timing of reproduction. / Am. Nat. 110: 801 /807. Conover, D. O. and Schultz, E. T. 1995. Phenotypic similarity and the evolutionary significance of countergradient variation. / Trends Ecol. Evol. 10: 248 /252. Danks, H. V. 1987. Insect dormancy: an ecological perspective. / Biological Survey of Canada. Danks, H. V. 1994. Insect life-cycle polymorphism: theory, evolution and ecological consequences for seasonality and diapause control. / Kluwer. Dingle, H. and Mousseau, T. A. 1994. Geographic variation in embryonic development time and stage of diapause in a grasshopper. / Oecologia 97: 179 /185. Dingle, H., Mousseau, A. and Scott, S. M. 1990. Altitudinal variation in life cycle syndromes of California populations of the grasshopper, Melanoplus sanguinipes (F.). / Oecologia 84: 199 /206. Gotthard, K. and Nylin, S. 1995. Adaptive plasticity and plasticity as an adaptation: a selective review of plasticity in animal morphology and life history. / Oikos 74: 3 /17. Gould, S. J. 1989. A developmental constraint in Cerion , with comments on the definition and interpretation of constraint in evolution. / Evolution 43: 516 /539. Groeters, F. R. and Shaw, D. D. 1992. Association between latitudinal variation for embryonic development time and chromosome structure in the grasshopper Caledia captiva (Orthoptera: Acrididae). / Evolution 46: 245 /257. Hadley, N. F. and Massion, D. D. 1985. Oxygen consumption, water loss and cuticular lipids of high and low elevation populations of the grasshopper Aeropedellus clavatus (Orthoptera: Acrididae). / Comp. Biochem. Physiol. 80A: 307 /311. Hilbert, D. W. and Logan, J. A. 1983. Empirical model of nymphal development for the migratory grasshopper, Melanoplus sanguinipes (Orthoptera: Acrididae). / Environ. Entomol. 12: 1 /5. Honek, A. 1993. Intraspecific variation in body size and fecundity in insects: a general relationship. / Oikos 66: 483 /492. Ingrisch, S. 1983. Zum Einfluss der Feuchte auf die Schlupfrate und Entwicklungsdauer der Eier mitteleuropa¨ischer Feldheuschrecken (Orthoptera: Acrididae). / Dtsch. Ent. Z. N. F. 30: 1 /15. Ingrisch, S. and Ko¨hler, G. 1998. Die Heuschrecken Mitteleuropas. / Westarp. Kingsolver, J. G. and Huey, R. B. 1998. Evolutionary analyses of morphological and physiological plasticity in thermally variable environments. / Am. Zool. 38: 545 /560. Ko¨hler, G. 1983. Investigations on the hatching polymorphism and its intrapopular consequences in Chorthippus parallelus (Zetterstedt) (Orthoptera: Acrididae). / Zool. Jb. Syst. 110: 31 /44. Ko¨hler, G. 1991. Investigations on diapause and non-diapause in central European grasshopper eggs (Acrididae: Gomphocerinae). / Zool. Jb. Syst. 118: 323 /344.

740

Lamb, R. J., MacKay, P. A. and Gerber, G. H. 1987. Are development and growth of pea aphids, Acyrtosiphon pisum , in North America adapted to local temperatures? / Oecologia 72: 170 /177. Leather, S. R., Waters, K. F. A. and Bale, J. S. 1993. The ecology of insect overwintering. / Cambridge Univ. Press. Masaki, S. 1967. Geographic variation and climatic adaptation in a field cricket (Orthoptera: Gryllidae). / Evolution 21: 725 /741. Merila¨, J. et al. 2000. Plasticity in age and size at metamorphosis in Rana temporaria / comparison of high and low latitude populations. / Ecography 23: 457 /465. Mousseau, T. A. and Roff, D. A. 1989. Adaptation to seasonality in a cricket: patterns of phenotypic and genotypic variation in body size and diapause expression along a cline in season length. / Evolution 43: 1483 /1496. Nylin, S. 1994. Seasonal plasticity and life-cycle adaptation in butterflies. / In: Danks, H. V. (ed.), Insect lifecycle polymorphism: theory, evolution and ecological consequences for seasonality and diapause control. Kluwer, pp. 41 /67. Orr, M. 1996. Life-history adaptation and reproductive isolation in a grasshopper hybrid zone. / Evolution 50: 704 /716. Pickford, R. 1966. The influence of date of oviposition and climatic conditions on hatching of Camnula pellucida (Acridiae). / Can. Entomol. 98: 1145 /1159. Roff, D. 1980. Optimizing development time in a seasonal environment: the ‘ups and downs’ of clinal variation. / Oecologia 45: 202 /208. Roff, D. A. 2002. Life history evolution. / Sinauer. Rowe, L. and Ludwig, D. 1991. Size and timing of metamorphosis in complex life cycles: time constraints and variation. / Ecology 72: 413 /427. Schlichting, C. D. and Pigliucci, M. 1998. Phenotypic evolution: a reaction norm perspective. / Sinauer. Schluter, D., Price, T. D. and Rowe, L. 1991. Conflicting selection pressures and life history trade-offs. / Proc. R. Soc. Lond. B 246: 11 /17. Slatkin, M. 1987. Gene flow and the geographic structure of natural populations. / Science 236: 787 /792. Stearns, S. C. 1992. The evolution of life histories. / Oxford Univ. Press. Tanaka, S. 1994. Diapause as a pivotal factor for latitudinal and seasonal adaptation in Locusta migratoria in Japan. / In: Danks, H. V. (ed.), Insect life-cycle polymorphism: theory, evolution and ecological consequences for seasonality and diapause control. Kluwer, pp. 173 /190. Tanaka, S. and Brookes, V. J. 1983. Altitudinal adaptation of the life cycle in Allonemobius fasciatus DeGeer (Orthoptera: Gryllidae). / Can. J. Zool. 61: 1986 /1990. Tatar, M., Gray, D. W. and Carey, J. R. 1997. Altitudinal variation for senescence in Melanoplus grasshoppers. / Oecologia 111: 357 /364. Tauber, M. J., Tauber, C. A. and Masaki, S. 1986. Seasonal adaptations of insects. / Oxford Univ. Press. Taylor, F. and Karban, R. 1986. The evolution of insect life cycles. / Springer. Telfer, M. G. and Hassall, M. 1999. Ecotypic differentiation in the grasshopper Chorthippus brunneus : life history varies in relation to climate. / Oecologia 121: 245 /254. Thorens, P. and Nadig, A. 1997. Atlas de distribution des Orthopte`res de Suisse. / Centre suisse de cartographie de la faune (CSCF), Documenta Faunistica Helvetiae 16. Wingerden, W. K. R. E. van, Musters, J. C. M. and Maaskamp, F. I. M. 1991. The influence of temperature on the duration of egg development in west European grasshoppers. / Oecologia 87: 417 /423.

ECOGRAPHY 27:6 (2004)


ECOGRAPHY 26: 456–466, 2003

Unique elevational diversity patterns of geometrid moths in an Andean montane rainforest Gunnar Brehm, Dirk Su¨ssenbach and Konrad Fiedler

Brehm, G., Su¨ssenbach, D. and Fiedler, K. 2003. Unique elevation diversity patterns of geometrid moths in an Andean montane rainforest. – Ecography 26: 456– 466. Alpha-diversity of geometrid moths was investigated along an elevational gradient in a tropical montane rainforest in southern Ecuador. Diversity was measured using 1) species number, 2) extrapolated species number (Chao 1 estimator), 3) rarefied species number, and 4) Fisher’s alpha. When applied to the empirical data set, 1 and 2 strongly depended on the sample size, whereas 3 and 4 were suitable and reliable measures of local diversity. At single sites, up to 292 species were observed, and extrapolation estimates range from 244 to 445 species. Values for Fisher’s alpha are among the highest ever measured for this moth family, and range from 69 to 131 per site. In contrast to theoretical assumptions and empirical studies in other regions of the world, the diversity of geometrid moths remained consistently high along the entire gradient studied. Diversity measures correlated with neither altitude nor ambient temperature. The large subfamily Ennominae has previously been assumed to be a group that occurs mainly at low and medium elevations. However, no decline in diversity was found in the study area. The diversity of the other large subfamily, Larentiinae, even increased from the lowest elevations and was highest at elevations above 1800 m. The roles of a decreasing diversity of potential host-plants, decreasing structural complexity of the vegetation, increasingly unfavourable climatic conditions and possible physiological adaptations in determining herbivore species richness are discussed. A relatively low predation pressure might be an advantage of high-altitude habitats. The physiognomy of the Andes (folded mountains, large areas at high altitudes) might also have allowed speciation events and the development of a species-rich high-altitude fauna. There is evidence that the species-richness of other groups of herbivorous insects in the same area declines as altitude increases. This emphasises difficulties that are associated with biodiversity indicator groups, and calls for caution when making generalisations from case studies. G. Brehm (gunnar – brehm@yahoo.com), D. Su¨ssenbach and K. Fiedler, Dept of Animal Ecology I, Uni6. of Bayreuth, D-95440 Bayreuth, Germany.

Biodiversity research in tropical regions has largely focussed on lowland rainforests, whereas montane rainforests have so far largely been neglected. This is particularly true with regard to studies on arthropods. For example, none of the 89 studies on canopy arthropods reviewed by Basset (2001) was conducted in a tropical montane habitat. The northern Andes have been recognised as a ‘‘hyper hotspot’’ on earth for vascular plants and vertebrates (Henderson et al. 1991, Myers et al. 2000). However, apart from a few exceptions, little is

known about patterns of insect diversity in Andean montane forest habitats. The dominant environmental gradient in mountainous habitats is altitude, which is directly related to decreasing temperature and various other abiotic and biotic environmental factors. Significant changes in the vegetation along altitudinal gradients in the Andes were first documented by Alexander von Humboldt (von Humboldt and Bonpland 1807). Later studies yielded increasingly refined pictures (e.g. Grubb et al. 1963,

Accepted 6 November 2002 Copyright © ECOGRAPHY 2003 ISSN 0906-7590

456

ECOGRAPHY 26:4 (2003)


Gentry 1988). Elevational changes in diversity have also been documented for a variety of animals in the Neotropical region (references in Brehm 2002). In a review on species richness patterns along elevational gradients, Rahbeck (1995) found that approximately half of the studies showed a continuous decline of species numbers with increasing altitude, whereas the other half detected a peak at medium elevations. Although some authors suggested that the occurrence of mid-elevational diversity peaks might be a sampling artefact (e.g. Wolda 1987), there is now substantial evidence for the existence of such peaks in a broad range of organisms (McCoy 1990, Grytnes and Vetaas 2002 with references therein). Among insects, several families of Lepidoptera in south east Asia have been reported to exhibit their greatest diversity at medium elevations, between 600 and 1000 m a.s.l. (Holloway 1987, Holloway and Nielsen 1999). However, only a few exceptions to an overall declining diversity at altitudes higher than 1000 m have been documented for arthropods. Examples include tropical larentiine moths (a subfamily of Geometridae), which are most speciesrich at high altitudes, and also increase in relation to other subfamilies with latitude in temperate regions (Brehm and Fiedler 2003). Schulze (2000) reported that diversity of the moth families Geometridae and Arctiidae in Borneo attained richness maxima at altitudes between 1200 and 2000 m. Above this altitude, all available studies indicate that diversity of insect assemblages decreases. This suggests that environmental conditions within and above the cloud-forest zone of tropical mountains become so harsh that the number of ectothermic insect species that are able to cope with such circumstances decreases steadily above this level. The moth family Geometridae is one of the three largest clades of Lepidoptera and currently includes \21000 described species (Scoble 1999), with some 6400 (30%) occurring in the Neotropical region. Compared to most arthropod taxa, the taxonomy of geometrid moths is relatively advanced, and they are a suitable group for biodiversity studies in tropical forests (Intachat and Woiwod 1999). However, prior to the present study, no detailed ecological studies of geometrids in the Neotropical region have ever been conducted. Samples taken from tropical arthropod communities provide a methodological challenge for diversity measures. They are almost always incomplete and the numbers of specimens available for analysis often diverge considerably between sites (e.g. Schulze and Fiedler 2003). Moreover, tropical arthropod communities are characterised by a high proportion of rare species that cannot be excluded as artefact or a group of marginal importance (Price et al. 1995, Novotny and Basset 2000). From the plethora of available measures of alpha-diversity (e.g. Hayek and Buzas 1997, Southwood and Henderson 2000), four were selected and their ECOGRAPHY 26:4 (2003)

sample size dependence was tested. Suitable diversity measures should be able to discriminate between samples of different diversity, but at the same time be independent of sample size, in order to avoid misleading bias in the results. In this first study of its kind in the Neotropical region, we aimed to investigate diversity patterns of Geometridae as model organisms of a ‘‘mega-diverse’’ group of herbivorous insects in a montane Andean rainforest. An analysis of changes of faunal composition and beta diversity of geometrid moths was provided by Brehm and Fiedler (2003) and Brehm et al. (2003). In this paper, we tested the expectation, derived from all comparable studies, that species diversity should decline in the upper part of the elevational gradient.

Methods Study area, sampling, identification, temperature measurements The study area in southern Ecuador is situated within the East Cordillera of the Andes and belongs to the province Zamora-Chinchipe (Reserva Biolo´ gica San Francisco, 3°58%S, 79°5%W, and adjacent fractions of the Podocarpus National Park). It is covered with undisturbed to slightly disturbed montane rainforest. The vegetation in the upper part of the study area was described by Bussmann (2001) and Paulsch (2002). Moths were sampled manually using portable weak light-traps (2 ×15 W). Traps consisted of a white gauze cylinder (height 1.60 m, diameter 0.60 m) and were placed at the forest floor. We selected 22 sampling sites at 11 elevational levels between 1040 and 2677 m a.s.l. Sites were numbered from 1 to 11 (a + b) according to their elevational order. A detailed description of the sites and a discussion of the sampling methods was provided by Brehm (2002). Sampling occurred during three field periods (April –May 1999, October 1999 – January 2000, and October –December 2000). Lighttraps were run between 18:30 and 21:30 local time. Between two and four nightly catches from each site were pooled and analysed. Specimens were first sorted to morphospecies level and later determined in the Zoologische Staatssammlung, Munich and the Natural History Museum, London. Fifty-two percent of the taxa and 67% of the specimens were determined to species level (Brehm 2002). In parallel to the light trapping, the temperature during three to eleven nights per site was measured every 30 min from 18:30 to 21:30 using a Casio alti-thermo twin sensor at 1.60 m above ground level. Braun (2002) and Brehm (2002) showed that temperature in the study area declines linearly with increasing elevation. Thus, for each site the average of all available nightly measures taken at 20:00 was used as standard. 457


Alpha-diversity measures The analysis was restricted to four selected measures and was performed separately for Geometridae and the two largest subfamilies Ennominae and Larentiinae. The remaining subfamilies (Desmobathrinae, Geometrinae, Oenochrominae, and Sterrhinae) were not analysed separately because of their low representation in the samples.

1) Species number Measurement of species richness by complete census is only feasible for a few organisms. For most organisms, measurement means sampling (Colwell and Coddington 1994). However, species numbers are still widely used as a measure of diversity. Misleading results and biases must be expected with incomplete samples that differ in size (Gotelli and Colwell 2001). 2) Extrapolation According to Colwell and Coddington (1994), if certain assumptions are not violated, the ‘‘true’’ number can be estimated using extrapolation methods. These authors recommended the use of non-parametric estimators, such as Chao 1, as promising quantitative techniques. Analyses were performed using the computer program EstimateS 6.0b1 (Colwell 2000), with the bias-corrected formula. Since all samples contained at least 380 specimens, the use of ‘‘Chao 1’’ appeared to be justified. However, a certain dependence on sample size was expected, because the recorded number of species is an integral part of the formula of the estimator.

mens is usually recommended for calculating Fisher’s alpha (Hayek and Buzas 1997). This number was not reached in only one sample of the subfamily Larentiinae (site 1a: 65 specimens). All diversity measures were tested for correlation with specimen numbers (i.e. sample size dependence), altitude, and temperature. Since relationships between data in this study are non-linear, the Spearman rank correlation coefficient was used. Multiple tests of significance were Bonferroni-corrected according to Hochberg (1988). All standard statistical analyses were performed using the software package Statistica 5.5 (Anon. 1999).

Results A total of 14348 specimens were collected from 22 sites. Four hundred and ten specimens (2.9% of the total catch) could not be reliably sorted and had to be discarded. Figure 1A shows the fluctuating number of specimens per catch and site. Specimen numbers cannot be directly compared because of the variable number of nightly catches analysed. The minimum and maximum numbers of specimens per site are 384 and 1200, respectively. Figure 1B shows the total number of species per site, partitioned into the subfamilies. The number of observed species per site ranged from 134 (site 1a) to 292 (site 7a), but actual numbers were expected to be distinctly higher (see next section). Raw species counts showed no relationship with temperature or altitude for Geometridae and Ennominae, but strongly increased at higher elevations for Larentiinae.

3) Rarefaction This method is particularly useful if assemblages are sampled with different intensity or success (Hurlbert 1971, Gotelli and Colwell 2001). The geometrid samples were rarefied to a shared number of specimens using a program developed by Kenney and Krebs (2000). The program also provided standard deviations. Rarefied expected species numbers were calculated at the level of 350 specimens for Geometridae, of 100 specimens for Ennominae, and of 50 specimens for Larentiinae. This measure is expected to be independent of sample size since samples are standardised to an equal level. 4) Fisher’s alpha of the log-series In contrast to other diversity indices, Fisher’s alpha has been shown to be efficient in discriminating between sites, and is mainly influenced by the frequencies of species of medium abundance (Kempton and Taylor 1974). The fit of the log-series distribution was tested using a program developed by Henderson and Seaby (1998). Fisher’s alpha and standard deviations (according to Fisher’s original formula, Colwell pers. comm.) were calculated with the program EstimateS 6.0b1 (Colwell 2000). A minimum number of at least 100 speci458

Fig. 1. A) Number of specimens collected at 22 sites. Sites are sorted by altitude and are partitioned according to the nightly catch and the sampling period. Sp 1999 – Spring 1999 (April – May), Au 1999 – Autumn 1999 (October 1999 – January 2000), Au 2000 – Autumn 2000 (October – November 2000). B) Number of species at 22 sites (sorted by altitude and partitioned across subfamilies). Ster – Sterrhinae, Geom – Geometrinae, Lare – Larentiinae, Enno – Ennominae. Due to the chosen scale, Desmobathrinae and Oenochrominae are not visible. Desmobathrinae: one species at sites 1a and 1b. Oenochrominae: two species at sites 1b and 7b, one species at sites 1a, 2a, 5a, 7a, 7b, 8a, 10a, and 11b. ECOGRAPHY 26:4 (2003)


Table 1. Spearman rank correlations between four measures of alpha-diversity (species number, extrapolated species number (Chao 1), rarefied species number, and Fisher’s alpha) and 1) specimen number, 2) altitude, and 3) mean ambient temperature (at 20:00 local time). 1 Rarefied sample size for which the expected species number was calculated. Species numbers and extrapolated species numbers are highly significantly correlated with specimen numbers and are thus unreliable measures of local diversity. Only two (unreliable) measures in the subfamily Larentiinae are significantly correlated with altitude and temperature. ns not significant, ** pB0.01, *** pB0.005, **** pB0.001. Printed in bold are results that remain significant after sequential Bonferroni correction according to Hochberg (1988).

1) Correlation between specimen number and species number extrapolated species number (Chao 1) rarefied species number Fisher’s alpha 2) Correlation between altitude and species number extrapolated species number (Chao 1) rarefied species number Fisher’s alpha (3) Correlation between temperature and species number extrapolated species number (Chao 1) rarefied species number Fisher’s alpha Rarefaction level 1

Geometridae

Ennominae

0.91 0.58 −0.07 0.15

**** *** ns ns

0.93 0.54 −0.09 0.29

**** ** ns ns

0.94 0.83 0.32 0.11

**** **** ns ns

0.13 0.18 0.01 0.08

ns ns ns ns

−0.39 −0.30 0.57 0.11

ns ns ** ns

0.78 0.65 0.31 0.16

**** *** ns ns

−0.08 −0.14 0.09 0.11

ns ns ns ns

0.42 0.29 −0.54 −0.07

ns ns ** ns

−0.75 −0.60 −0.29 −0.17

**** *** ns ns

350

Extrapolation Extrapolated species numbers are on average 75% larger than the observed numbers in all three taxa. The estimate ranges from 29% (Larentiinae, site 11a) to 218% (Ennominae, site 8b) larger than the recorded species number. Figure 2A shows extrapolation results for Geometridae, Ennominae and Larentiinae for all 22 sites. For Geometridae, estimates range between 244 (site 2b) and 445 (site 7a) expected species per site. There is no relationship between the extrapolated number of species and either altitude or temperature for geometrid moths as a whole and Ennominae, while a distinct increase in species numbers for Larentiinae with altitude is notable (Table 1). However, these estimates have to be regarded with caution since the Chao 1 estimator is not a fully reliable measure of diversity in this particular data set because of its sample size dependence (see below).

Rarefaction Figure 3 shows rarefaction curves of Geometridae for all 22 sites. All curves lie within a relatively narrow band and no clear altitudinal pattern is visible. Figure 2B shows the expected species numbers for geometrid samples rarefied to a standard size of 350 specimens plotted against altitude. While one site (1a at 1040 m) has a significantly lower rarefied species number (122) than all other sites, these sites again range in a continuous band between 135 and 168 expected species and show no tendency along the altitudinal and temperature gradient (Table 1). ECOGRAPHY 26:4 (2003)

100

Larentiinae

50

Changes do occur at the subfamily level. Figure 2B shows rarefied species numbers for the subfamilies Ennominae and Larentiinae at rarefied sample sizes of 100 and 50 specimens, respectively. The patterns resemble each other. Surprisingly, rarefied species numbers of Ennominae even increase with altitude (Table 1), although this is not significant after sequential Bonferroni correction. A conspicuous difference between both subfamilies is visible at sites 3a, 3b, 4a, and 4b (at 1800 – 1875 m). While rarefied species numbers in Ennominae tended to be lower than in the neighbouring sites, those of Larentiinae tended to be higher.

Fisher’s alpha of the log-series A significant deviation from the log-series occurred in six ensembles of Geometridae, and in four ensembles of Ennominae, but not in Larentiinae (Table 2). However, after performing the sequential Bonferroni procedure suggested by Hochberg (1988), the log-series distribution was rejected only in Geometridae and Ennominae at the same single site (number 1b, Table 2). In all cases of deviation, the observed number of species in the first abundance class was larger than expected, i.e. there were ‘‘too many’’ rare species in the samples. Figure 2C shows Fisher’s alpha values for Geometridae at all 22 sites. They range from 69.3 95.4 to 130.6 9 10.4 and are among the highest ever measured for local geometrid ensembles. There is no significant consistent change in Fisher’s alpha along the altitudinal gradient. Alpha values for Ennominae range from 38.2 9 3.4 to 66.79 6.5, and for Larentiinae from 20.1 9 4.1 to 64.29 6.2. In both subfamilies, values of Fisher’s alpha 459


Fig. 2.

460

ECOGRAPHY 26:4 (2003)


Fisher’s alpha are all independent of sample size. Despite their differences with regard to sample size dependence, all four measures are highly associated (Kendall’s coefficient of concordance: Geometridae rK = 0.95, Ennominae rK = 0.91, Larentiinae rK = 0.98, all p B 0.001).

Discussion Diversity within a world context

Fig. 3. Rarefaction curves for Geometridae at all 22 sites. For clarity, standard deviation curves are omitted. The vertical line indicates the standardised sample size (350 specimens), see Fig. 2B. The lowest rarefied species number was calculated for site 1a (1040 m).

are neither correlated with altitude nor with temperature (Table 1). However, in Larentiinae, Fisher’s alpha is lowest at the lowest elevations (Fig. 2C).

Sample size dependence of diversity measures and concordance Table 1 shows correlations between specimen numbers and different measures of alpha-diversity for Geometridae, Ennominae, and Larentiinae across the 22 study sites. In all three taxa, the recorded species number is strongly correlated with the number of specimens collected. Extrapolated species numbers also show significant correlations in all taxa, although the relationship is less pronounced than with recorded specimen numbers. In contrast, values of rarefied species numbers and Table 2. Nominally significant deviations of samples from the log-series distribution. Provided are p-values from x2-tests after arranging species-abundances in octaves (between 3 and 4 degrees of freedom) (Henderson and Seaby 1998). Printed in bold: significant after sequential Bonferroni correction (Hochberg 1988). Samples from all other sites (levels 2, and 5–11) and all Larentiinae samples did not deviate from the log-series. A complete list of sampling sites and their geographical coordinates was provided by Brehm (2002) and Brehm and Fiedler (2003). Site number

Altitude (m)

Geometridae

Ennominae

1a 1b 3a 3b 4a 4b

1040 1040 1800 1800 1850 1875

0.038 B0.001 0.027 0.015 0.010 0.015

– 0.003 0.048 – 0.006 0.035

Values of Fisher’s alpha of up to 131 per site for the species-rich family Geometridae are among the highest ever measured in the world. While most geometrid ensembles in temperate regions reach values between 10 and 35 (Barlow and Woiwod 1989, Thomas and Thomas 1994, Thomas 2002), the highest values to date have been found in rainforests of Peninsular Malaysia and Borneo at elevations between 200 and 700 m, with alpha scores of 127 and 128, respectively (Barlow and Woiwod 1989, Beck et al. 2002). An all-year-round sampling would probably lead to even higher values of Fisher’s alpha because the sites in Ecuador were only sampled between two and four times. Since species numbers at single sites have been shown to be highly dependent on sample size, and samples are incomplete, raw species counts cannot be compared directly with other data. However, the total number of 1010 nocturnal geometrid moth species collected in the study area is the highest ever counted in a small geographic area (Brehm 2002).

Altitudinal patterns of different geometrid taxa All diversity measures reveal similar results along the altitudinal gradient (all Kendall’s rK \ 0.9). In all three taxa considered here (Geometridae, subfamilies Ennominae and Larentiinae) diversity either remained more or less constantly high or increased over an elevational range from 1040 to 2677 m. The patterns are very remarkable, because a decline of insect diversity towards higher altitudes was expected (Wolda 1987, a review by McCoy 1990, Hanski and Niemela¨ 1990, Bru¨ hl et al. 1999). Insect diversity often does not decrease steadily along elevational gradients, but peaks at medium elevations (Janzen et al. 1976, Holloway 1987, Olson 1994, Chey 2000). Our study is the first that revealed such an extended elevational diversity maximum of a very species-rich insect group at very high altitudes. So far, a similar pattern has only been observed in the geometrid subfamily Larentiinae in south

Fig. 2. Diversity of Geometridae (left), and the largest subfamilies Ennominae and Larentiinae (right) along an altitudinal gradient ranging from 1040 to 2677 m in a montane rainforest in southern Ecuador. Diversity measured A) by extrapolation with the estimator Chao 1, B) by rarefaction (the level of specimens to which samples have been rarefied is indicated), and C) with Fisher’s alpha. Error bars indicate 9 1 SD. ECOGRAPHY 26:4 (2003)

461


east Asia (Holloway et al. 1990, Schulze 2000). However, above 2000 m, even in Larentiinae only declining diversity has been detected so far. The high diversity of the subfamily Ennominae at high elevations was unexpected because this taxon was previously thought to be most diverse at lower montane zones (Holloway et al. 1990). We assume that the diversity of geometrid moths in the Andes declines only at very high altitudes in the transition zone between montane cloud rainforest and the pa´ ramo vegetation. Indeed, sporadic sampling in the summit region of the study area (ca 3100 m) revealed minimal numbers of geometrid moths. However, effects of isolation and small habitat area might have been responsible for this (see below). Further studies with an extended elevational coverage in neighbouring or other areas could reveal more detailed patterns.

Diversity patterns and environmental factors The exceptional, unique distributions of geometrid moths including their two largest subfamilies Ennominae and Larentiinae, call for explanations. Lawton et al. (1987) listed four hypotheses for declining diversity of herbivorous insects at higher altitudes: 1) reduction of habitat area, 2) reduction of resource diversity, 3) reduction of primary productivity, 4) increasingly unfavourable environments. Despite these constraints, diversity of geometrids remained constant or was even lower at low altitudes. How can this be explained? The points listed by Lawton et al. seem to be either not applicable in the study area (1 and 2), or might be compensated for by other factors (3 and 4).

Habitat area Habitat area is not expected to be a limiting factor in this study, because the highest site is situated ca 500 m below the mountain summit and large areas of similar forested habitat exist next to the study area. However, area might also come into play as a limiting factor at higher altitudes in the Andes. Rahbeck (1997) and Ko¨ rner (2000) emphasised the importance of decreasing area at high altitudes, and Holloway (1987) discussed a higher diversity of moths in montane Papua New Guinea compared to Borneo, as a consequence of greater land area situated above 2000 m. The Andes are folded mountains and provide a habitat area at high altitudes that is by far larger than on relatively isolated mountains such as Mount Kinabalu in Borneo, where the high altitude fauna is less species-rich. Furthermore, the biogeographical conditions in the Andes support the isolation of local populations (e.g. during glacial periods), subsequent speciation events, and a later co-existence of species. These fea462

tures might explain a considerable part of the exceptional diversity of geometrid moths in Ecuador.

Resource diversity The possible reduction of resource diversity, including spatial niches, is difficult to assess because of widely lacking information describing which resources are actually used by Neotropical geometrid moths. However, some conclusions can be drawn from available information on host-plant use (Brehm 2002), and from vegetation data covering the study area. The structural complexity of the forest clearly declines along the altitudinal gradient (Paulsch 2002). Upper montane forests in the study area provide a far poorer offer of structural niches, e.g. because of the lower height and the absence of lianas. As a consequence, structural niches do not seem to be a limiting factor in the diversity of geometrids. The level of floristic diversity is more difficult to interpret, but there is evidence that the diversity of potential host-plants of geometrid moths generally decreases along altitudinal gradients in the Neotropical region (e.g. Gentry 1988, Lieberman et al. 1996). In a few cases there are indications of specialism towards certain host plant groups, and these geometrid groups indeed decline towards high altitudes (Brehm 2002). For example, the ennomine tribes Cassymini and Macariini are specialised Fabaceae feeders and are not present at the highest sites in the study area. Only very few Fabaceae species have been found in the study area (Homeier pers. comm.). Lianas also decrease along the altitudinal gradient in the study area, whereas in herbaceous vines this trend is less pronounced (Matezki pers. comm.). The number of tree species ( \5 cm diameter at breast height, 400 m2 plots) in ridge forests in the study area decreases from ca 30 species at 1850 m to ca 20 species at 2450 m (Homeier et al. 2002). In contrast, the diversity of shrubs might be constant or even increase with increasing altitude. This latter resource (e.g. shrub-like Asteraceae) might play an important role for the very species-rich larentiine genus Eupithecia. There are also indications of a constant and high diversity of epiphytic plants (Werner 2002), which is generally known to be very high in Neotropical montane forests (Gentry and Dodson 1987, Nieder et al. 2001). However, most epiphytic vascular plants are monocotyledons and ferns (Rauer and Rudolph 2001). Both these plant groups are hardly exploited by Neotropical Geometridae (Brehm 2002). Overall, the total diversity of hosts that are actually used can be expected to decrease with altitude, but the extent of this decrease remains unknown. Irrespective of these uncertainties, the diversity of geometrids does not appear to be limited by a reduction of potential resource diversity. ECOGRAPHY 26:4 (2003)


Primary productivity Primary productivity usually decreases along altitudinal gradients (Brujinzeel and Veneklaas 1998, Waide et al. 1998, but see Singh et al. 1994). According to Tanner et al. (1998), nutrient limitation is widespread in montane soils and foliar nitrogen decreases with increasing altitude. Significant changes in soil properties have also been documented along the elevational gradient in the study area (Schrumpf et al. 2001). They reported decreasing pH values and nitrogen availability with rising altitude. Given these constraints, diversity of herbivorous geometrid moths remained unaffected.

Climate and physiology Obviously, the moths are able to resist the cool and humid climate at high altitudes. The monthly average temperature decreases linearly by ca 10 K along the gradient, and precipitation more than doubles from ca 2000 mm to ca 5500 mm per annum (Hagedorn 2001, Emck pers. comm.). Geometrid moths appear to be physiologically predisposed towards such cool conditions. Heinrich (1993) and Rydell and Lancaster (2000) reported that many geometrid moth species were able to fly with lower thoracic temperatures than most other Lepidoptera species do. This could be a major energetic advantage and might allow geometrid moths to colonise habitats that are unsuitable for most other insects. However, the knowledge about the flight physiology of the vast majority of species is still unknown. Further investigations could test this hypothesis. We expect that members of the subfamily Larentiinae are particularly cold-adapted because of their dominance at high altitudes and latitudes.

Are high-altitude habitats an enemy-free space? Predation can have impacts on the diversity of herbivorous insects. For example, Williams et al. (2001) pointed out that resources may often be less important than natural enemies in determining the distribution of herbivores. On the one hand, predators might regulate prey populations and prevent the dominance of single species. On the other hand, low predation pressure might allow an unconstrained radiation of herbivorous insects in nearly enemy-free space. Since predation pressure probably strongly declines with increasing altitude (see below), we find little support for the first hypothesis because neither diversity nor dominance values significantly change along the elevational gradient (Brehm 2002). The diversity of insectivorous species of bats, birds and ants in the Andes markedly decreases with altitude. Up to 38 species of insectivoECOGRAPHY 26:4 (2003)

rous bat species co-occur in lowland rainforests in Panama (Kalko 1997), whereas only eight occur above 1800 m and four above 2800 m in the study area in Ecuador (Matt 2001). Mixed species flocks of birds that forage in rainforests are expected to have a large impact on leaf-chewing insects (Braun 2002) and occur more prominently in lowland rather than montane forests (Rahbeck 1997, Thiollay 1999). Ants are the most prevalent invertebrate predators in many tropical forests (Ho¨ lldobler and Wilson 1990). However, they strongly decrease in diversity as altitude increases (Stork and Brendell 1990, Bru¨ hl et al. 1999). At higher altitudes in the Ecuadorian study area (above ca 2000 m), only very few ant species occur (unpubl.). Therefore, habitats are indeed a nearly enemy-free space with regard to this otherwise very important group of potential predators (Novotny et al. 1999, Floren et al. 2002). The knowledge on parasitoids and their role for Neotropical geometrid moths is still extremely scarce. However, parasitoids appear not to be a dominant factor since Brehm (2003) found only a low percentage of parasitised larvae in the study area.

Are the results representative of other groups? This study has demonstrated exceptional altitudinal patterns of one major group of herbivorous insects. Further sampling would be required to confirm whether the results of this study are also applicable for other groups. According to Holloway (1987) the relative contribution of geometrids to local moth assemblages increases with altitude, suggesting that diversity patterns should be discordant even among moths as a guild. Indeed moth taxa in the Ecuadorian study area such as Pyraloidea and Arctiidae exhibit completely different altitudinal diversity patterns (Su¨ ßenbach 2003). Beccaloni and Gaston (1995) found a relatively constant ratio of species of the nymphalid subfamily Ithomiinae among all butterflies, and Longino (1994) reported a number of tropical invertebrate ‘‘focal taxa’’ that might represent suitable ‘‘survey taxa’’. The transfer of results from one group to others is part of the controversial debate about the usefulness of biodiversity indicators. Although several studies have established parallels between diversity patterns of different groups of organisms (Wolda 1996, Kerr et al. 2000), others found that there were none (Lawton et al. 1998, Ricketts et al. 2002). Simberloff (1998) criticised the concept of biodiversity indicators because of a lacking consensus as to what indicators should indicate and which organisms are the most versatile. If various taxa exhibit fundamentally different diversity patterns even among the herbivorous Lepidoptera, there is no reason to assume that patterns of, for example, detrivorous or predatory insects are better reflected. 463


Choice of measurement and sample size dependence Our results confirm that unless it is possible to achieve complete inventories, the recorded species number is an unreliable measure of diversity because of its extreme dependence on the number of specimens collected (correlation coefficients all \0.9, p B0.001). As expected, it has to be rejected as a meaningful measure of diversity for the purpose of discriminating between samples that are incompletely sampled (Gotelli and Colwell 2001). The estimator Chao 1 has also been shown to be significantly sample size dependent, though not to the same extent as species number. It is very probable that the true local richness is still substantially underestimated at most sites. This is illustrated by the very high ratios of singletons at single sites, i.e. species that were collected only once (41 –60% of the species). A very high ratio of rare species is typical for samples of tropical arthropods. For example, Novotny and Basset (2000) found very similar singleton rates of 45% in samples of herbivorous insects in New Guinea. Underestimation occurs if samples are too sparse (Colwell and Coddington 1994). This study shows that even samples of at least 134 species and 384 individuals can be ‘‘too sparse’’ for extrapolation in extremely rich moth ensembles. According to Colwell and Coddington (1994), estimators correlate with sample size until about half the total fauna is observed and thereafter become gradually independent of sample size. Obviously, this level has not been reached at many sites because they could not be sampled more than two to four times. Rarefied species numbers were independent of sample size. The measure can overestimate diversity if species have clumped distributions (Achtziger et al. 1992), but this is of relatively little importance in large samples and does not affect the results presented in this study. Fisher’s alpha values did not correlate with specimen numbers and were thus independent of sample size. However, it is possible that values of Fisher’s alpha would further increase with an increasing number of samples (Wolda 1987, Intachat and Holloway 2000, Schulze and Fiedler 2003). Since goodness-of-fit of the log-series model was not always satisfactory, it does not seem to be appropriate to rely solely on this measure. The use of several different measures can be recommended since they complement each other in different aspects of diversity as well as in the mathematical assumptions underlying their usage. Acknowledgements – We are indebted to the taxonomists who were kind enough to allow access to collections under their care, and provided advice and literature: Linda M. Pitkin, Malcolm J. Scoble, and David J. Carter at the Natural History Museum, London, and Axel Hausmann, Manfred Sommerer and Robert Trusch at the Zoologische Staatssammlung, Munich. Paul Emck, Ju¨ rgen Homeier and Steffen Matezki provided unpublished results of their investigations on climate and vegetation of the study area. Jennifer Kay, Adrienne Hogg, Rita Schneider and Teresa Baethmann helped to pre-

464

pare and database the moths. Giovanni Onore and Christoph L. Ha¨ user offered administrative support. The Ministerio del Medio Ambiente del Ecuador granted research permits, NCI (Loja, Ecuador) allowed access to parts of the study area, and the Deutsche Forschungsgemeinschaft financed the project (Fi 547/5-1/3, FOR 402/1-1).

References Achtziger, R., Nigmann, U. and Zwo¨ lfer, H. 1992. Rarefaction-Methoden und ihre Einsatzmo¨ glichkeiten bei der zooo¨ kologischen Zustandsanalyse und Bewertung von Biotopen. – Z. O8 kol. Natursch. 1: 89 – 105. Anon. 1999. Statistica 5.5. for Windows. – StatSoft, Tulsa. Barlow, H. S. and Woiwod, I. P. 1989. Moth diversity of a tropical forest in Peninsular Malaysia. – J. Trop. Ecol. 5: 37 – 50. Basset, Y. 2001. Invertebrates in the canopy of tropical rain forests – how much do we know? – Plant Ecol. 153: 87 – 107. Beccaloni, G. W. and Gaston, K. J. 1995. Predicting the species richness of Neotropical forest butterflies: Ithomiinae (Lepidoptera: Nymphalidae) as indicators. – Biol. Conserv. 71: 77 –86. Beck, J. et al. 2002. From forest to farmland: diversity of geometrid moths along two habitat gradients on Borneo. – J. Trop. Ecol. 17: 33 – 51. Braun, H. 2002. Die Laubheuschrecken (Orthoptera, Tettigoniidae) eines Bergregenwaldes in Su¨ d-Ecuador – faunistische, bioakustische und o¨ kologische Untersuchungen. – Ph.D. thesis, Univ. of Erlangen-Nu¨ rnberg. Brehm, G. 2002. Diversity of geometrid moths in a montane rainforest in Ecuador. – Ph.D. thesis, Univ. of Bayreuth. Brehm, G. 2003. Host-plant records and illustrations of the larvae of 19 geometrid moths from a montane rainforest in Ecuador. – Nachr. Entomol. Ver. Apollo. 24, in press. Brehm, G. and Fiedler, K. 2003. Faunal composition of geometrid moths changes with altitude in an Andean montane rainforest. – J. Biogeogr. 30: 431 – 440. Brehm, G., Homeier, J. and Fiedler, K. 2003. Beta diversity of geometrid moths (Lepidoptera: Geometridae) in an Andean montane rainforest. – Divers. Distrib. 9, in press. Bru¨ hl, C. A., Mohamed, M. and Linsenmair, K. E. 1999. Altitudinal distribution of leaf litter ants along a transect in primary forest on Mount Kinabalu, Sabah, Malaysia. – J. Trop. Ecol. 15: 265 – 277. Brujinzeel, L. A. and Veneklaas, E. J. 1998. Climatic conditions and tropical montane forest productivity: the fog has not lifted. – Ecology 79: 3 – 9. Bussmann, R. W. 2001. The montane forests of Reserva Biolo´ gica San Francisco (Zamora-Chinchipe, Ecuador). – Erde 132: 9 – 25. Chey, V. K. 2000. Moth diversity in the tropical rain forest of Lanjak-Entimau, Sarawak, Malaysia. – Malayan Nat. J. 54: 305 – 318. Colwell, R. K. 2000. EstimateS: statistical estimation of species richness and shared species from samples, ver. 5.1. – Bhttp://viceroy.eeb.uconn.edu/estimates\. Colwell, R. K. and Coddington, J. A. 1994. Estimating terrestrial biodiversity through extrapolation. – Philos. Trans. R. Soc. Lond. B 345: 101 – 118. Floren, A., Biun, A. and Linsenmair, K. E. 2002. Arboreal ants as key predators in tropical lowland rainforest trees. – Oecologia 131: 137 – 144. Gentry, A. H. 1988. Changes in plant community diversity and floristic composition on environmental and geographical gradients. – Ann. Mo. Bot. Gard. 75: 1 – 34. Gentry, A. H. and Dodson, C. 1987. Contribution of nontrees to species richness of a tropical rainforest. – Biotropica 19: 149 – 156. ECOGRAPHY 26:4 (2003)


Gotelli, N. J. and Colwell, R. K. 2001. Quantifying biodiversity: procedures and pitfalls in the measurement and comparison of species richness. – Ecol. Lett. 4: 379 – 391. Grubb, P. J. et al. 1963. A comparison of montane and lowland rain forest in Ecuador I. The forest structure, physiognomy, and floristics. – J. Ecol. 51: 567 – 601. Grytnes, J. A. and Vetaas, O. R. 2002. Species richness and altitude: a comparison between null models and interpolated plant species richness along the Himalayan altitudinal gradient, Nepal. – Am. Nat. 159: 294 – 304. Hagedorn, A. 2001. Extent and significance of soil erosion in southern Ecuador. – Erde 132: 75 –92. Hanski, I. and Niemela¨ , J. 1990. Elevational distributions of dung and carrion beetles in northern Sulawesi. – In: Knight, W. J. and Holloway, J. D. (eds), Insects and the rain-forests of south east Asia (Wallacea). The Roy. Entomol. Soc., pp. 145 –152. Hayek, L.-A. and Buzas, M. A. 1997. Surveying natural populations. – Columbia Univ. Press. Heinrich, B. 1993. The hot-blooded insects. – Springer. Henderson, A., Churchill, S. P. and Luteyn, J. L. 1991. Neotropical plant diversity. – Nature 351: 21 –22. Henderson, P. A. and Seaby, R. M. H. 1998. Species diversity and richness, ver. 2.65. – Pisces Conserv., Pennington. Hochberg, Y. 1988. A sharper Bonferroni procedure for multiple tests of significance. – Biometrika 75: 800 – 802. Ho¨ lldobler and Wilson, 1990. The ants. – Springer. Holloway, J. D. 1987. Macrolepidoptera diversity in the IndoAustralian tropics: geographic, biotopic and taxonomic variations. – Biol. J. Linn. Soc. 30: 325 –341. Holloway, J. D. and Nielsen, E. S. 1999. Biogeography of the Lepidoptera. – In: Kristensen, N. P. (ed.), Lepidoptera, moths and butterflies, Vol. 1, part 35. De Gruyter. Holloway, J. D., Robinson, G. S. and Tuck, K. R. 1990. Zonation of the Lepidoptera of northern Sulawesi. – In: Knight, W. J. and Holloway, J. D. (eds), Insects and the rain-forests of south east Asia (Wallacea). The Roy. Entomol. Soc., pp. 153 –166. Homeier, J., Dalitz, H. and Breckle, S.-W. 2002. Waldstruktur und Baumartendiversita¨ t im montanen Regenwald der Estacio´ n Cientı´fica San Francisco in Su¨ decuador. – Ber. Reinh.-Tu¨ xen-Ges. 14: 109 –118. Hurlbert, S. H. 1971. The nonconcept of species diversity: a critique and alternative parameters. – Ecology 52: 577 – 586. Intachat, J. and Woiwod, I. P. 1999. Trap design for monitoring moth biodiversity in tropical rainforests. – Bull. Entomol. Res. 89: 153 –163. Intachat, J. and Holloway, J. D. 2000. Is there stratification in diversity or preferred flight height of geometroid moths in Malaysian lowland tropical forest? – Biodiv. Conserv. 9: 1417 – 1439. Janzen, D. H. et al. 1976. Changes in arthropod community along an elevational transect in the Venezuelan Andes. – Biotropica 8: 193 –203. Kalko, E. V. K. 1997. Diversity in tropical bats. – In: Ulrich, H. (ed.), Tropical biodiversity and systematics. Proc. of the Int. Symp. on Biodiv. and Syst. in tropical ecosystems. Zool. Forschungsinst. und Mus. Alexander Koenig Bonn, pp. 13– 43. Kempton, R. A. and Taylor, L. R. 1974. Log-series and log-normal parameters as diversity discriminants for the Lepidoptera. – J. Anim. Ecol. 43: 381 –399. Kenney, A. J. and Krebs, C. J. 2000. Programs for ecological methodology, ver. 5.2. – \ http://www.zoology.ubc.ca/ krebs \ . Kerr, J. T., Sugar, A. and Packer, L. 2000. Indicator taxa, rapid biodiversity assessment, and nestedness in an endangered ecosystem. – Conserv. Biol. 14: 1726 –1734. Ko¨ rner, C. 2000. Why are there global gradients in species richness? Mountains might hold the answer. – Trends Ecol. Evol. 15: 513 –514. ECOGRAPHY 26:4 (2003)

Lawton, J. H., MacGarvin, M. and Heads, P. A. 1987. Effects of altitude on the abundance and species richness of insect herbivores on bracken. – J. Anim. Ecol. 56: 147 – 160. Lawton, J. H. et al. 1998. Biodiversity inventories, indicator taxa and effects of habitat modification in tropical forest. – Nature 391: 72 – 75. Lieberman, D. et al. 1996. Tropical forest structure and composition on a large-scale altitudinal gradient in Costa Rica. – J. Ecol. 84: 137 – 152. Longino, J. T. 1994. How to measure arthropod diversity in a tropical rainforest. – Biol. Internat. 28: 3 – 13. Matt, F. 2001. Pflanzenbesuchende Flederma¨ use im tropischen Bergregenwald: Diversita¨ t, Einnischung und Gildenstruktur – Eine Untersuchung der Fledermausgemeinschaften in drei Ho¨ henstufen der Andenostabdachung des Podocarpus Nationalparks in Su¨ decuador. – Ph.D. thesis, Univ. of Erlangen-Nu¨ rnberg. McCoy, E. D. 1990. The distribution of insects along elevational gradients. – Oikos 58: 313 – 322. Myers, N. et al. 2000. Biodiversity hotspots for conservation priorities. – Nature 403: 853 – 858. Nieder, J., Prosperı´, J. and Michaloud, G. 2001. Epiphytes and their contribution to canopy diversity. – Plant Ecol. 153: 51 – 63. Novotny, V. and Basset, Y. 2000. Rare species in communities of tropical insect herbivores: pondering the mystery of singletons. – Oikos 89: 564 – 572. Novotny, V. et al. 1999. Predation risk for herbivorous insects on tropical vegetation: a search for enemy-free space and time. – Aust. J. Ecol. 24: 477 – 483. Olson, D. M. 1994. The distribution of leaf litter invertebrates along a Neotropical altitudinal gradient. – J. Trop. Ecol. 10: 129 – 150. Paulsch, A. 2002. Development and application of a classification system for undisturbed and disturbed tropical montane forests based on vegetation structure. – Ph.D. thesis, Univ. of Bayreuth. Price, P. W. et al. 1995. The abundance of insect herbivore species in the tropics: the high local richness of rare species. – Biotropica 27: 468 – 478. Rahbeck, C. 1995. The elevational gradient of species richness: a uniform pattern? – Ecography 18: 200 – 205. Rahbeck, C. 1997. The relationship among area, elevation, and regional species richness in Neotropical birds. – Am. Nat. 149: 875 – 902. Rauer, G. and Rudolph, D. 2001. Vaskula¨ re Epiphyten eines westandinen Bergregenwaldes in Ecuador. – In: Nieder, J. and Barthlott, W. (eds), The flora of the Rio Guajalito mountain rain forest. Books on Demand, pp. 323 – 469. Ricketts, T. H., Daily, G. C. and Ehrlich, P. R. 2002. Does butterfly diversity predict moth diversity? Testing a popular indicator taxon at local scales. – Biol. Conserv. 103: 361 – 370. Rydell, J. and Lancaster, W. C. 2000. Flight and thermoregulation in moths were shaped by predation from bats. – Oikos 88: 13 – 18. Schrumpf, M. et al. 2001. Tropical montane rain forest soils. – Erde 132: 43 – 59. Schulze, C. H. 2000. Auswirkungen anthropogener Sto¨ rungen auf die Diversita¨ t von Herbivoren – Analyse von Nachtfalterzo¨ nosen entlang von Habitatgradienten in Ost-Malaysia. – Ph.D. thesis, Univ. of Bayreuth. Schulze, C. H. and Fiedler, K. 2003. Vertical and temporal diversity of a species-rich moth taxon in Borneo. – In: Basset, Y. et al. (eds), Arthropods of tropical forests – spatio-temporal dynamics and resource use in the canopy. Cambridge Univ. Press, pp. 69 – 85. Scoble, M. J. (ed.) 1999. Geometrid moths of the world – a catalogue (Lepidoptera: Geometridae). – CSIRO Publ., Collingwood. Simberloff, D. 1998. Flagships, umbrellas, and keystones: is a single-species management passe´ in the landscape era? – Biol. Conserv. 83: 247 – 257.

465


Singh, S. P., Adhikari, B. S. and Zobel, D. B. 1994. Biomass, productivity, leaf longevity, and forest structure in the central Himalaya. – Ecol. Monogr. 64: 401 –421. Southwood, T. R. and Henderson, P. A. 2000. Ecological methods. – Blackwell. Stork, N. E. and Brendell, M. J. D. 1990. Variation in the insect fauna of Sulawesi trees with season, altitude, and forest type. – In: Knight, W. J. and Holloway, J. D. (eds), Insects and the rain-forests of south east Asia (Wallacea). The Roy. Entomol. Soc., pp. 173 – 190 Su¨ ßenbach, D. 2003. Diversita¨ t von Nachtfaltergemeinschaften entlang eines Ho¨ hengradienten in Su¨ decuador (Lepidoptera: Pyraloidea, Arctiidae). – Ph.D. thesis, Univ. of Bayreuth. Tanner, E. V. J., Vitousek, P. M. and Cuevas, E. 1998. Experimental investigation of nutrient limitations of forest growth on wet tropical mountains. – Ecology 79: 10 – 22. Thiollay, J. M. 1999. Frequency of mixed species flocking in tropical forest birds and correlates of predation risk: an intertropical comparison. – J. Avian Biol. 30: 282 – 294. Thomas, A. W. 2002. Moth diversity in a northeastern North America, red spruce forest, II. The effect of silvicultural practices on geometrid diversity (Lepidoptera: Geometridae). – Can. For. Serv., Fredericton. Thomas, A. W. and Thomas, G. M. 1994. Sampling strategies

466

for estimating moth species diversity using a light trap in a northeastern softwood forest. – J. Lepidop. Soc. 48: 85 – 10. von Humboldt, A. and Bonpland, A. 1807. Essai sur la geographie des plantes – accompagne´ d’un tableau physique des re´ gions e´ quinoxiales, fonde´ sur des mesures exe´ cute´ es, depuis le dixie`me degre´ de latitude bore´ ale jusqu’au dixie`me degre´ de latitude australe, pendant les anne´ es 1799, 1800, 1801, 1802 et 1803. – Levrault et Schoell, Paris. Waide, R. B., Zimmerman, J. K. and Scatena, F. N. 1998. Controls of primary productivity: lessons from the Luquilo mountains in Puerto Rico. – Ecology 79: 31 – 37. Werner, F. 2002. Ecology of vascular epiphytes in a montane forest and on remnant trees of adjacent pastures. – Diploma thesis, Univ. of Bonn. Williams, I. S., Jones, H. and Hartley, S. 2001. The role of resources and natural enemies in determining the distribution of an insect herbivore population. – Ecol. Entomol. 26: 204 – 211. Wolda, H. 1987. Altitude, habitat and tropical insect diversity. – Biol. J. Linn. Soc. 30: 313 – 323. Wolda, H. 1996. Between-site similarity in species composition of a number of Panamanian insect groups. – Miscella`nia Zool. 19: 39 – 50.

ECOGRAPHY 26:4 (2003)


Ecography 33: 425434, 2010 doi: 10.1111/j.1600-0587.2009.06016.x # 2010 The Authors. Journal compilation # 2010 Ecography Subject Editor: Jens-Christian Svenning. Accepted 5 June 2009

Change within and among forest communities: the influence of historic disturbance, environmental gradients, and community attributes Windy A. Bunn, Michael A. Jenkins, Claire B. Brown and Nathan J. Sanders W. A. Bunn (Windy_Bunn@nps.gov), C. B. Brown and N. J. Sanders, Dept of Ecology and Evolutionary Biology, Univ. of Tennessee, 569 Dabney Hall, Knoxville, TN 37996, USA. (Present address of W. A. B.: USDOI, National Park Service, Grand Canyon National Park, P. O. Box 129, Grand Canyon, AZ 86023, USA.)  M. A. Jenkins, USDOI, National Park Service, Great Smoky Mountains National Park, 1314 Cherokee Orchard Road, Gatlinburg, TN 37738, USA. (Present address of M. A. J.: Purdue Univ., Dept of Forestry and Natural Resources, 715 West State Street, West Lafayette, IN 47907, USA.)

Understanding how ecological communities change over time is critical for biodiversity conservation, but few long-term studies directly address decadal-scale changes in both the within- and among-community components of diversity. In this study, we use a network of permanent forest vegetation plots, established in Great Smoky Mountains National Park (USA) in 1978, to examine the factors that influence change in community composition within and among communities. In 2007, we resampled 15 plots that were logged in the late 1920s and 15 plots that had no documented history of intensive human disturbance. We found that understory species richness decreased by an average of 4.3 species over the 30-yr study period in the logged plots, but remained relatively unchanged in the unlogged plots. In addition, tree density decreased by an average of 145 stems ha 1 in the logged plots, but was relatively stable in the unlogged plots. However, we found that historic logging had no effect on within-community understory or tree compositional turnover during this time period. Instead, sites at lower elevations and sites with lower understory biomass in 1978 had higher understory compositional turnover than did sites at higher elevations and sites with higher understory biomass. In addition, sites with lower soil cation exchange capacity (CEC) and with lower tree basal area in 1978 had higher tree compositional turnover than did sites with higher soil CEC and higher tree basal area. Among-community similarity was unchanged from 1978 to 2007 for both the logged and unlogged plots. Overall, our results indicate that human disturbance can affect plant communities for decades, but the extent of temporal change in community composition may nevertheless depend more on environmental gradients and community attributes.

Both within- and among-community attributes can change over time, and understanding these changes often requires long-term empirical data (Magnuson 1990, Wardle et al. 2008). Within a community, the total number of species present as well as the abundance of particular species can change with time. These within-community changes can lead to temporal differences in similarity among communities within a region (Loreau 2000). Despite the potential for within-community changes to influence similarity among communities, few studies directly measure longterm changes in both the within- and among-community components of diversity (but see Chalcraft et al. 2004). In this study, we use long-term monitoring data from Great Smoky Mountains National Park (GSMNP) in eastern Tennessee to examine changes in forest understory plant communities and tree communities across 30 yr. Specifically, we examine the factors that influence compositional change within communities and change in compositional similarity among communities.

Human-caused disturbance can lead to large and persistent differences in understory communities in disturbed forests compared with undisturbed forests (Meier et al. 1995, Flinn and Vellend 2005, Harrelson and Matlack 2006). Furthermore, disturbance may also influence the magnitude of compositional change through time in disturbed versus undisturbed communities (Collins and Smith 2006). While general models of forest development (Oliver and Larson 1996) describe forest communities as undergoing rapid changes in the short term after disturbance, the long-term influence of disturbance on withincommunity change is less clear. In addition to disturbance, environmental gradients can influence variation in within-community compositional turnover. For example, the extent of compositional turnover through time in low-elevation sites may be higher than that of high elevation sites (Aplet and Vitousek 1994, Selmants and Knight 2003, Taverna et al. 2005), and a number of factors that are often associated with 425


elevation  precipitation, soil fertility, species richness, and primary productivity-have been shown or hypothesized to affect temporal change within communities (Peet and Christensen 1988, Chase and Leibold 2002, Verheyen et al. 2003, Yurkonis and Meiners 2004, Taverna et al. 2005, Smart et al. 2006, White et al. 2006, Anderson 2008). However, the relative influence of these factors on compositional turnover within plant communities is poorly understood and likely varies with community composition and structure. Over time, communities within a region can become either more similar or less similar to one another depending on the extent of change within individual communities of the region. Recent studies of decadal-scale change in forest communities in the eastern U.S. indicate that declines in species richness (Rooney et al. 2004, Taverna et al. 2005), shifts in plant community composition (Taverna et al. 2005), and changes in regional community similarity (Rooney et al. 2004) over time may be common. Landuse history can have large effects on forest communities (Foster et al. 1998, Vellend et al. 2007), and the legacy of human disturbance may therefore be important for understanding patterns in among-community similarity in these long-lived communities. In this study, we investigated changes in understory plant communities as well as tree communities in forests of GSMNP that were logged in the late 1920s and forests that were not logged. Specifically, we examined plant community data collected in 1978 (50 yr after logging) and in 2007 (80 yr after logging) to test four explicit hypotheses: 1) community composition of logged plots differs from that of unlogged plots both 50 and 80 yr after the logging event. 2) Historically logged plots have greater within-community compositional turnover than unlogged plots. 3) Elevation and associated edaphic and community attributes influence within-community compositional turnover. 4) The temporal change in similarity of logged communities to one another differs from the temporal change in similarity of unlogged communities to one another.

Methods Great Smoky Mountains National Park (GSMNP) is a 211 000 ha protected area that straddles the Tennessee North Carolina stateline. Elevations in GSMNP range from 271 to 2025 m, and climate and vegetation types vary

considerably along the elevational gradient. Mean annual rainfall in low elevation sites is 1400 mm with mean temperatures 128C, while annual rainfall is 2000 mm and temperature averages 68C at high-elevation peaks. GSMNP contains over 70 vegetation associations, varying from low- to mid-elevation mixed hardwood forests and xeric Pinus and Quercus forests to high-elevation PiceaAbies forests and heath balds. Prior to its establishment, ca 80% of the area that became GSMNP was subject to anthropogenic disturbance (Pyle 1988). Despite its history of disturbance, GSMNP is considered a center for diversity in North America. Plot selection and field methods In 2007, we resampled thirty 2050 m forest plots originally established in 1978. The plots were randomly distributed using a stratified design that divided watersheds into units based upon elevation, slope position, and aspect. Fifteen of the resampled plots were in historically logged forests and fifteen plots were in unlogged forests. Historically logged forests were defined using the ‘‘corporate logging’’ category of Pyle (1988) and included only those areas in which the use of railroads, mechanized skidding, non-selective cutting practices, and highly extensive cutting on slopes occured. We defined unlogged forests using Pyle’s ‘‘high in virgin forest attributes’’ and ‘‘big trees with diffuse disturbance’’ categories. We chose the thirty resampled plots from a pool of over 100 permanent plots. Since our primary goal was to evaluate the effects of historic disturbance and elevation on community dynamics, we used 1978 field data and 2007 pre-sampling surveys to exclude plots with high levels of recent disturbance. Toward this end, we resampled only plots dominated by hardwood species, sites without recent or frequent fires, and sites that were not (or have not been) influenced by Dendroctonus frontalis (southern pine beetle), Adelges piceae (balsam woolly adelgid) or Adelges tsugae (hemlock woolly adelgid). Because of this rigorous selection process, we consider the thirty resampled plots to be relatively free of disturbance in the thirty years between sampling events. The logged and unlogged plots were topographically similar to one another: mean elevation and percent slope of logged plots did not differ from the mean elevation and percent slope of unlogged plots (Table 1). The 15 logged plots ranged in elevation from 727 to 1402 m and

Table 1. Comparison of topographic and edaphic variables in historically logged and unlogged plots, using t-tests or Wilcoxon rank-sum tests to test for mean differences. Variable

Elevation (m) Slope (%) Soil pH Soil cation exchange capacity (meq 100 g 1) Soil K (ppm) Soil Ca (ppm) Soil Mg (ppm) Soil P (ppm) Soil organic matter (%)

426

Mean9SE

Logged vs unlogged

logged

unlogged

p-value

1104961.6 19.093.26 4.290.11 9.490.40 68.599.04 326971.9 44.996.63 14.391.70 5.190.28

1034967.3 19.892.42 4.590.12 7.190.16 64.894.93 165927.1 32.393.45 15.492.46 3.190.33

0.41 0.95 0.02 B0.0001 0.63 0.12 0.13 0.77 0.0002


occurred on north- (n 8), east- (n 4), and west- (n 3) facing slopes. The 15 unlogged plots ranged in elevation from 664 to 1400 m and occurred on north- (n 11), east(n 1), and west- (n 3) facing slopes. Both the logged and unlogged plots are characterized by well-drained loamy soils classified as either humic or typic dystrudepts (A. R. Khiel, NRCS, unpubl. report). Tree cores collected and analyzed by the GSMNP Vegetation Monitoring Program were available for a subset of the plots (Jenkins unpubl.). Dominant trees in unlogged plots were 150228 yr old (complete cores) or a minimum 132147 yr old (cores without pith) in 2007. As expected from logging history records, dominant trees in logged plots were 7580 yr old in 2007. In 1978, the 2050 m forest plots were permanently marked with rebar and witness tree tags, which allowed us to reestablish the plots in 2007. In the 2007, we used the same sampling design used in 1978. Within each 20 50 m plot, we recorded understory shrub and tree seedling species B1 m tall in 25 4-m2 subplots and understory herbaceous species in 25 1-m2 subplots nested within the shrub and seedling subplots. We defined the understory community as the shrub, seedling, and herbaceous species recorded in all 25 subplots within the 20 50 m plot. We also recorded all individual trees (]10 cm dbh) by species in the 20 50 m plot and used these data to characterize the tree community. In 2007, we sampled plots at roughly the same time of year that they were sampled in 1978. Sampling was conducted between 19 June and 26 August in 1978 and between 9 July and 26 August in 2007. Sampling of plots was paired, as best as possible, within seasons. That is, if a plot was sampled late in the field season in 1978, we attempted to sample that plot late in the field season of 2007. To characterize the sampling plots, we estimated a suite of topographic and edaphic parameters (Table 1). We estimated elevation using topographic maps and calculated percent slope by averaging three slope measurements taken at the two 20-m end lines and at the center of each plot facing downslope. Between 2002 and 2007, soil samples were collected from the top 10 cm of soil at five locations throughout each of the 30 plots with a hand spade. The five subsamples were combined into one composite sample per plot, dried at 438C for at least 8 h, and sieved through a 2 mm mesh. The samples were analyzed for pH, cation exchange capacity (CEC), total phosphorus (P), potassium (K), calcium (Ca), and magnesium (Mg), and percent organic matter by A&L Analytical Laboratories, Memphis, TN (see Jenkins et al. 2007 for details of cation extraction procedures). Analysis of the influence of historic logging on community composition We used nonparametric, permutational multivariate analysis of variance (PERMANOVA; Anderson 2001) based on Bray-Curtis similarity values of species abundances to compare understory and tree community composition between logged and unlogged plots in both 1978 and 2007. PERMANOVA compares the variability in species similarity between plots within a treatment to the

variability in species similarity between plots from different treatments and is performed using the FORTRAN program PERMANOVA (Anderson 2005). The test statistic for PERMANOVA is the pseudo F-ratio, where a large pseudo F-ratio indicates that logged plots are closer to one another in multivariate space than they are to unlogged plots and that the logged and unlogged plots differ in community composition. The significance of the pseudo F-ratio is tested using a permutation test that randomly shuffles the sample labels within and among treatment groups and calculates the pseudo F-ratio for 9999 arbitrary reassignments of the data. The pseudo F-ratios of these randomly assigned communities are then compared to the pseudo F-ratio of the observed community to calculate the significance level of the test (Anderson 2001). For the understory communities, we calculated species abundance as the percentage of the 25 subplots in which the species occurred. For the tree communities, we used the number of individual trees of a species as the abundance value. We chose to use understory frequency and tree density as the abundance measures rather than understory cover and tree basal area because these measures are more repeatable between observers and across years. The choice to use these frequency-based abundance measures did not change the results. Since the scale of abundance values in our study was small (ranging from 0 to 100 for understory species and 0 to 52 for tree species), we did not transform the data to reduce the influence of abundant species. A significant pseudo F-ratio from the PERMANOVA can indicate a difference in community composition between treatments due either to differences in the location of the treatment communities in multivariate space or to differences in dispersion of communities in multivariate space within the treatments (Anderson 2001). To confirm that compositional differences between logged and unlogged communities were due to location differences rather than to dispersion differences, we used permutational analysis of multivariate dispersions (PERMDISP; Anderson et al. 2006) performed in the FORTRAN program PERMDISP (Anderson 2004). PERMDISP calculates the centroid of each treatment (logged or unlogged) in multivariate space based on the chosen similarity measure (in this case, Bray-Curtis), and then calculates the distance of each plot within the treatment from the treatment centroid. To compare average dispersion values between treatments (logged understory communities vs unlogged understory communities; logged tree communities vs unlogged tree communities), PERMDISP performs a permutational ANOVA and calculates a pseudo F-statistic and p-value in the same manner as the PERMANOVA described above. A significant pseudo F-ratio from the PERMANOVA and a non-significant difference in dispersion between logged and unlogged plots from the PERMDISP analysis would suggest that logged and unlogged communities differ in multivariate composition and do not differ in variation around the mean composition within logged and unlogged communities. We tested whether particular species accounted for the observed differences in community composition between logged and unlogged communities with indicator species analysis (Dufreˆne and Legendre 1997) using PC-ORD 5.0 427


(MjM Software Design, Gleneden Beach, OR). The indicator analysis uses the relative abundance of each species (for example, percent cover of the species in the logged plots divided by percent cover of the species in all plots) and the relative frequency of the species within each group (for example, the number of logged plots in which the species occurs out of the 15 total logged plots) to calculate an indicator value that ranges from 0 to 100. An indicator value of 100 indicates that the species was observed in only one group (in this case, logged or unlogged plots) and that each plot within that group contained at least one individual of that species. In other words, a species with an indicator value of 100 for logged plots occurs in every logged plot and no unlogged plots, and is thus, a good indicator of plots that have been logged. A Monte Carlo test based on Bray-Curtis distance was used to test the significance of the indicator value (Dufreˆne and Legendre 1997). Analysis of change within communities We analyzed within-plot change in species richness and tree density with paired t-tests, where the species richness and density values for a plot in 1978 were compared with the species richness and density values for the same plot in 2007. For the analyses of tree species richness, we used both the observed number of species present in a plot and an estimate of species richness generated by individual-based rarefaction (PRIMER, ver. 6, PRIMER-E, Plymouth Marine Laboratory, Plymouth, UK). Rarefaction allowed us to correct species richness values for differences in the number of individuals sampled in each plot by using a resampling procedure to generate estimated species richness values based on the number of individuals sampled in the plot with the fewest trees (Gotelli and Colwell 2001). To estimate compositional turnover within communities (i.e. change within a plot over the 30-yr period), we calculated the similarity of each plot in 1978 to itself in 2007 using the ChaoSørensen incidence-based index and the BrayCurtis index in EstimateS (Colwell 2005). The ChaoSørensen incidence-based index (Linc) is a modified form of the traditional Sørensen similarity index that accounts for both the frequency of individual species in the community and for the effects of ‘‘unseen shared species’’ (species that are missing from the sample data but are likely present in the community) on community similarity (Chao et al. 2005). The ChaoSørensen index is useful for assessing similarity between diverse communities that contain many rare species, such as the forest understory plant community. The BrayCurtis index (also referred to as the Sørensen quantitative index or the Czekanowski coefficient; CN) is widely used to assess similarity between two communities (Magurran 2004). The ChaoSørensen and BrayCurtis indices produced qualitatively similar results for the understory community, so we chose to focus on only the ChaoSørensen index in the results and discussion. Since we did not expect ‘‘unseen shared species’’ in the tree community, we used the Bray Curtis index to assess tree compositional turnover. For both the BrayCurtis and the ChaoSørensen indices, values near 1 indicate nearly identical community composition between time periods and values near 0 428

indicate that communities have very little compositional overlap between time periods. We defined turnover as the degree of compositional dissimilarity between 1978 and 2007 within an individual plot. Therefore, we calculated turnover as 1-Linc for the understory community and 1-CN for the tree community. To evaluate whether historic logging influenced compositional turnover, we performed t-tests to compare mean logged and unlogged community turnover values. We then used mixed stepwise multiple regression (a combination of forward and backward steps; a 0.1) to determine whether elevation, edaphic, or community attributes influenced variation in compositional turnover. We used JMP 6.0 (SAS Inst., Cary, NC, USA) for all analyses of within-plot compositional turnover. Analysis of change among communities To examine whether similarity among communities (i.e. how similar plots were to one another within a sampling period) changed over time in logged and unlogged plots, we used a test for homogeneity of multivariate dispersions (Anderson et al. 2006) based on BrayCurtis dissimilarity. Among-community similarity is the average distance among plots within a group to the group centroid in multivariate space (i.e. multivariate dispersion as in Anderson et al. 2006) and is statistically tested for differences in amongcommunity similarity between years with a permutational ANOVA (described above) in the PERMDISP program. Using this approach, a significant p-value indicates that plots within a treatment (logged or unlogged) became either more homogeneous (had lower multivariate dispersion in 2007 than in 1978) or more dissimilar to one another (had higher multivariate dispersion in 2007 than in 1978) over time. For our study plots, average distance of individual plots to the group centroid is directly comparable to traditional measures that calculate mean similarity of each plot to all other plots within the group. For understory communities, average BrayCurtis similarity was highly correlated with average distance to the group centroid in 1978 (r 0.99, p B0.001) and in 2007 (r 0.99, pB0.0001). For tree communities, average BrayCurtis similarity was highly correlated with average distance to the group centroid in 1978 (r 0.99, pB0.001) and in 2007 (r 0.87, p B0.0001).

Results In 1978 (50 yr after logging), historically logged plots contained a total of 132 understory species and 29 tree species while unlogged plots contained 157 understory species and 25 tree species. In 2007 (80 yr after logging), historically logged plots contained a total of 110 understory species and 24 tree species while unlogged plots contained 134 understory species and 26 tree species. Across both sampling periods, historically logged plots contained 25 understory species and 5 tree species that were not found in unlogged plots, and unlogged plots contained 48 understory species and 2 tree species that were unique to unlogged plots (Supplementary material Table S1). Overall, 39 species recorded in 1978 were not


seen in 2007, and 9 new species were encountered in 2007 that were not recorded in 1978 (Supplementary material Table S2).

F1, 28 2.11, p 0.03) and 2007 (PERMANOVA: F1, 28 2.37, p 0.02). These differences were due to differences in the location of the logged and unlogged plots in multivariate space rather than to differences in the relative dispersion of plots within the logged and unlogged groups (1978 PERMDISP: F 0.02, p 0.89; 2007 PERMDISP: p 0.36, P 0.59). Betula lenta and Prunus serotina had significantly higher indicator values in logged plots than in unlogged plots in both years. In addition, Prunus pensylvanica had a significantly higher indicator value in logged plots than in unlogged plots in 1978 and Magnolia fraseri had a significantly higher indicator value in logged plots than in unlogged plots in 2007. Acer saccharum Marsh. was the only tree species with a significantly higher indicator value in unlogged plots than in logged plots and was an indicator of unlogged plots in only 2007 (Table 2).

Influence of historic logging on community composition We found slight differences in understory community composition between logged and unlogged plots in both 1978 (PERMANOVA: F1, 28 1.84, p0.05) and 2007 (PERMANOVA: F1, 28 1.80, p0.05). These differences were due to differences in the location of the logged and unlogged plots in multivariate space rather than to differences in the relative dispersion of plots within the logged and unlogged groups (1978 PERMDISP: F 0.54, p0.52; 2007 PERMDISP: F 0.40, p0.59). Seven understory species had significantly higher indicator values (a combination of relative abundance and relative frequency) in unlogged plots than in logged plots in 1978 and eight understory species were significant indicators of unlogged plots in 2007 (Table 2). Five understory species had significantly higher indicator values in logged plots in 1978, but only one understory species was still an indicator of logged plots in 2007 (Table 2). Indicators of unlogged plots included tree seedlings, small shrubs, and slowdispersing forest interior herbs, such as Trillium spp., Viola hastata, Arisaema triphyllum, and Eurybia divaricata. Four of the five significant indicators of logged plots were woody seedlings or shrubs. Composition of tree communities differed between the logged and unlogged plots in both 1978 (PERMANOVA:

Change within communities Understory species richness in individual logged plots was, on average, 13% lower in 2007 than in 1978 (t 2.35, DF 14, p0.03). However, richness did not change in the unlogged plots over the 30-yr study period (t 1.06, DF 14, p 0.31). The decrease in overall understory species richness in the logged plots resulted from decreased richness of herbaceous species (Supplementary material Fig. S1). Shrub, seedling, and tree species richness did not change over time in either logged or unlogged plots (p 0.08 in all cases; Supplementary material Fig. S1). Stem density of trees decreased by an average of 21% in the historically logged plots (t 6.14, DF 14, p B0.0001)

Table 2. Indicator species analysis for compositional differences between logged and unlogged plots. Indicator values (IV) represent the degree to which a species is an indicator of the listed group, with 100 representing perfect indication. Understory includes herbs, shrubs, and seedlingsB1 m tall and tree includes trees ]10 cm dbh. Species with significant IV in at least one year are listed alphabetically. Species name

1978

2007

Group

IV

p

Group

IV

p

Understory Acer saccharum Amphicarpaea bracteata Arisaema triphyllum Athyrium filix-femina Betula alleghaniensis Betula lenta Calycanthus floridus Collinsonia canadensis Dioscorea villosa Eurybia divaricata Liriodendron tulipifera Osmunda claytoniana Prunus serotina Quercus rubra Rhododendron maximum Rubus spp. Thalictrum thalictroides Trillium spp. Viola hastata

unlogged unlogged unlogged unlogged logged unlogged unlogged unlogged unlogged unlogged unlogged logged logged unlogged logged logged unlogged unlogged unlogged

44.6 31.7 63.6 12.6 44.9 10.0 53.3 60.0 48.0 51.2 51.2 40.0 72.5 58.9 42.9 55.7 22.2 29.2 49.6

0.16 0.04 0.001 0.91 0.03 0.87 0.002 0.002 0.01 0.16 0.02 0.02 0.001 0.01 0.02 0.02 0.23 0.15 0.07

unlogged unlogged unlogged unlogged logged unlogged unlogged unlogged unlogged unlogged unlogged  logged unlogged logged unlogged unlogged unlogged unlogged

55.0 18.7 45.1 44.6 20.0 40.0 37.0 6.7 20.0 59.3 37.2  51.4 31.7 49.9 33.7 36.7 72.3 58.4

0.03 0.45 0.02 0.02 0.25 0.02 0.06 1 0.25 0.04 0.29  0.15 0.63 0.01 0.97 0.05 0.002 0.003

Tree Acer saccharum Betula lenta Magnolia fraseri Prunus pensylvanica Prunus serotina

unlogged logged logged logged logged

39.2 68.3 40.0 33.3 56.9

0.24 0.005 0.05 0.03 0.02

unlogged logged logged  logged

62.2 60.4 43.9  50.0

0.03 0.02 0.05  0.02

429


430

0.0005 10.06

0.003

2, 27

7.22

B0.0001 0.001 0.03

0.23 0.13

0.32 0.11

0.43

2, 26

0 0.57 0.33

B0.0001 0.02 0.03 0 0.40 0.36

0.65 0.03 0.003

0.32 0.0001 0.001

Intercept CEC 1978 tree basal area

R

0.36

p F DF

Model

2

Partial R2 p Standardized estimate Parameter estimate

Intercept Elevation 1978 understory cover Tree compositional turnover

Logging history affected change in understory species richness and tree density within communities. However, logging history did not affect the extent of withincommunity compositional turnover for either the understory or tree community. Instead, factors such as elevation, soil properties, and community biomass explained withincommunity compositional change. Among-community similarity was unchanged from 1978 to 2007 in the understory communities and in the tree communities of both the logged and unlogged plots.

Understory compositional turnover

Discussion

Predictor(s)

Among-community similarity was unchanged from 1978 to 2007 in both the logged and unlogged plots (Table 4). We found no differences in the multivariate dispersion of understory communities in 2007 compared with 1978 in either the logged (PERMDISP: F 0.18, p 0.74) or unlogged (PERMDISP: F 0.19, p0.69) plots. Similarly, multivariate dispersion of tree communities did not change between 1978 and 2007 in logged (PERMDISP: F 0.06, p0.82) or unlogged (PERMDISP: F 1.17, p0.35) plots.

Dependent variable

Change among communities

Table 3. Results of stepwise multiple regressions for change in species richness and community composition over 30 yr. Understory includes herbs, shrubs, and seedlings B1 m tall and trees includes trees]10 cm dbh.

but did not change in unlogged plots (t 0.30, DF 14, p0.77) (Supplementary material Fig. S2). Tree basal area did not change over 30 yr in either the historically logged plots (t 2.0, DF 14, p0.07) or the unlogged plots (t 1.07, DF 14, p0.30). Historic logging had no effect on within-plot understory compositional turnover (t0.08, DF 28, p0.94) or on within-plot tree compositional turnover (t 1.4, DF 27, p0.17). A model containing elevation and 1978 understory biomass (estimated using percent cover values) accounted for 36% of the variation in within-plot understory compositional turnover (i.e. how similar a plot was to itself over the 30-yr period; p 0.003; Table 3). Overall, plots at higher elevations had lower understory compositional turnover over the 30-yr period than did plots at lower elevations (Fig. 1a). With the exception of one statistical outlier (determined using Cook’s D and hat matrix analyses) that contained ca 55% cover of two fern species (Phegopteris hexagonoptera and Dennstaedtia punctilobula), plots with high understory biomass in 1978 had lower understory turnover than did plots with low understory biomass in 1978 (Fig. 1b). Change in understory community composition was not related to tree compositional turnover (r 0.19, p0.32). Variation in tree compositional turnover was best explained by a model containing soil cation exchange capacity (CEC) and 1978 tree biomass (R2 0.43, p 0.0005; Table 3). Plots with high CEC had lower compositional turnover in the tree community than did plots with low CEC (Fig. 2a). In addition, plots with high tree biomass (estimated using stand basal area) in 1978 had lower tree turnover than did plots with low tree biomass in 1978 (Fig. 2b).


Figure 1. Correlation between understory compositional turnover across 30 yr and elevation (a) and 1978 understory percent cover (b). Filled circles represent logged plots and unfilled circles represent unlogged plots. Arrow in panel b points to an outlying data point that was excluded from the correlation.

Influence of historic logging on community composition In 1978 (50 yr after logging), understory community composition differed only slightly between historically logged and unlogged plots. Historically logged and unlogged plots also differed only slightly in understory composition in 2007 (80 yr after logging). These results are similar to some chronosequence studies comparing understory communities in recently logged forests with understory communities of older forests (Gilliam et al. 1995, Ford et al. 2000). However, other studies have found larger and more persistent differences in the understory communities of anthropogenically disturbed

Figure 2. Correlation between tree compositional turnover and soil cation exchange capacity (a) and 1978 tree basal area (b). Filled circles represent logged plots and unfilled circles represent unlogged plots.

and undisturbed forests (Meier et al. 1995, Flinn and Vellend 2005, Harrelson and Matlack 2006). In our study site, the relatively small differences we observed in community composition between logged and unlogged plots could be due to the short duration of logging activities or to our focus on late-season species. Logging activities lasted for four years in our study area (Schmidt and Hooks 1994), and understory species may have persisted in the soil seedbank and subsequently recolonized logged areas or survived as scattered mature individuals in the logged areas. In addition, our 1978 and 2007 sampling data were collected between mid-June and late-August and included primarily late-season understory species, which may be more resistant to logging effects than vernal species. 431


Table 4. Among-community similarity in logged and unlogged plots in 1978 and 2007. Mean distance of plots within a group from the group centroid (multivariate dispersion) in multivariate space is defined by BrayCurtis dissimilarity. The pseudo F-statistics and p-values were generated from permutational ANOVA tests of differences in among-community similarity between 1978 and 2007. Abundance variable

Understory frequency1 2

Tree density

Group

logged unlogged logged unlogged

Mean9SE distance to centroid 1978

2007

53.0592.14 50.9292.09 51.7092.22 51.1692.58

51.6892.39 49.5992.21 50.8792.41 47.7291.87

F

p-value

0.18 0.19 0.06 1.17

0.74 0.69 0.82 0.35

1

Understory frequency is an abundance measure based on the percentage of the 25 subplots within a 2050 m forest plot in which each species occurred. Tree density is an abundance measure based on the number of individual trees of a species within a 2050 m forest plot.

2

Overall, six understory species that were indicators of unlogged plots in 1978 were no longer indicators in 2007 as these species began to recolonize logged plots. Likewise, four understory species that were indicators of logged plots in 1978 were no longer indicators in 2007 as earlycolonizing species became less abundant in the logged plots. Seven understory species became indicators of unlogged plots between 1978 and 2007. Three of the seven species are spring ephemerals that generally increase in abundance as forests mature and their inclusion as indicator species of unlogged plots in 2007 but not in 1978 could be a result of variation in phenology between sample years. The remaining species that became indicators of unlogged plots in 2007 either became newly established in unlogged plots or remained unchanged in unlogged plots while occurring in fewer logged plots over time. This indicates that after 150 to 200 yr, the unlogged plots are still undergoing measureable changes in community composition. In addition, it suggests that some species associated with older forests may become established and subsequently lost then regained as disturbed forests mature. We note that although spatial autocorrelation may have contributed to differences between logged and unlogged plots, the stratified random sampling design, wide spatial distribution of plots within disturbance types, and physical similarity of logged and unlogged sample areas likely reduced its effects. Factors influencing change within communities Even though change in understory species richness and tree density differed between logged and unlogged plots, the occurrence of historic logging did not affect the extent of within-community compositional turnover in the understory community or the extent of within-community tree compositional turnover during our 30-yr study period. We had expected greater compositional change in the understory of logged plots than in unlogged plots during this period due to the changing understory light environment as the logged plots moved through the stem exclusion and understory reinitiation stages of development (Oliver and Larson 1996) and tree density decreased. It may be that we found no difference in the extent of compositional change in the understory of logged and unlogged plots because logging events that occurred 50 yr ago no longer affect understory dynamics in these forests. Also contrary to our expectations, the extent of tree compositional turnover in 432

unlogged plots was similar to tree turnover in logged plots. Since tree density in the logged plots was almost twice as high as tree density in the unlogged plots in 1978, the loss of individual trees likely had a smaller effect on compositional turnover in the logged plots than in the unlogged plots, which could explain the similar extent of compositional turnover we observed in logged and unlogged plots. Elevation and community biomass best explained compositional turnover within forest understory communities. Consistent with other studies (Aplet and Vitousek 1994, Selmants and Knight 2003), we found that the extent of turnover in the understory community decreased along the elevational gradient. Mean annual temperature decreases ca 48C and soils become more acidic over the 740 m elevational gradient in our study site (Garten and Hanson 2006). The lower temperatures at higher elevations could slow decomposition, decrease nutrient availability, and reduce overall plant growth (Vitousek et al. 1992, Aplet and Vitousek 1994), which might result in lower compositional turnover at high elevations. In Great Smoky Mountains National Park, decomposition rates generally decrease with elevation, but nitrogen availability increases due to low soil C-to-N ratios at high elevations (Garten 2004). We did not measure decomposition rates or nitrogen availability in our study plots; therefore, it is unclear whether slower ecosystem processes at high elevations are responsible for the smaller changes in understory composition over time that we observed. In addition to temperature, soil pH also varies with elevation in our study sites, with more acidic soils at higher elevation sites. In the acidic soils of southeastern U.S. forests, higher soil pH can indicate greater nutrient availability to plants. In the understory, greater nutrient availability could lead to increased compositional turnover either by increasing the likelihood that newly arriving species will establish in a community (Peet and Christensen 1988) or by increasing the growth of dominant species that could out-compete other species in the community. In addition, the large regional pool of species that favor high pH sites (Peet et al. 2003) could increase the chance that new species would colonize these sites over time or that more species would be present in the initial community. A larger pool of potential colonizers could increase understory compositional change in high pH communities compared to low pH communities. Unlike understory compositional turnover, tree compositional turnover was not related to elevation, but was related to soil CEC. Because turnover of individual tree


stems can be a function of elevation, latitude, and productivity in some forest systems (Phillips et al. 2004, Stephenson and van Mantgem 2005), we expected higher compositional turnover in the tree community in low elevation plots than in high elevation plots. However, elevation was not correlated with within-site change in tree community composition, stem density, or tree species richness over time in our study. We did not directly measure turnover of individual trees and cannot say whether the rate of stem recruitment or mortality changed with elevation. For trees, it may be that fertile sites allow faster tree growth regardless of elevation. Faster tree growth could increase recruitment into the overstory or increase mortality through competitive exclusion. Either increased recruitment or mortality could lead to greater tree compositional turnover in more fertile plots. Understory biomass in 1978 was correlated with understory compositional turnover, and tree biomass in 1978 was correlated to tree compositional turnover. Plots with high understory percent cover (an estimate of understory biomass; Gilliam and Turrill 1993) in 1978 had lower compositional turnover than did plots with low understory cover in 1978. Similarly, plots with high tree basal area (a surrogate for tree biomass; Wardle et al. 2008) in 1978 had lower tree compositional turnover than did plots with low tree basal area in 1978. Understory percent cover and tree basal area in 1978 were not correlated with any measured topographic or edaphic factors (Supplementary material Table S3). Sites with higher biomass may have low within-site turnover because in higher biomass sites, a larger proportion of the plot is occupied by established species. The proportion of a plot initially occupied might affect community change in these forests over time due to resident species excluding new species from establishing or limiting the population growth of other resident species. Influence of historic logging on change among communities We documented a wide range of changes in species composition within communities, but these withincommunity changes did not translate into a change in among-community similarity. We found no change in among-community similarity from 1978 to 2007 in either the logged and unlogged plots. Evidence that amongcommunity similarity is lower in forests formerly disturbed by agriculture compared with older forests (Christensen and Peet 1984, Vellend et al. 2007) suggests that younger forests may become more similar to one another over time. However, we found no change in among-community similarity in either understory or tree communities of the logged plots. We also found no change in amongcommunity similarity in plant communities of unlogged plots over this same 30-yr time period. The theoretical expectation for temporal change in among-community similarity in undisturbed forests is unclear. It may be that the 30-yr study period here was a relatively stable period within larger cycles of amongcommunity heterogeneity in forest development. Additionally, among-community heterogeneity may have reached a

static point where it will remain unchanged in the absence of further disturbance (Rejma´nek and Rose´n 1992).

Conclusions The extent of change in community composition was not related to historic disturbance, and among-community similarity did not change over time in either historically disturbed or undisturbed plots. Our results indicate that the extent of change in community composition over time may depend more on environmental gradients and community attributes than on the legacy of large-scale, but short-lived historic disturbances such as logging. In addition, variation in turnover within communities may not necessarily translate into changes in compositional similarity among communities over time. Additional long-term studies that directly measure temporal change both within and among communities are needed in order to increase our understanding of the factors that control multi-scale diversity across time and space.

Acknowledgements  We are grateful to A. Classen, D. Simberloff, H. H. Bruun, and two anonymous reviewers for helpful comments on the manuscript. L. Souza assisted with the 2007 data collection and J. Rock assisted with plant identification. Funding for the resampling portion of this project was provided to W.A.B. by the Dept of Ecology and Evolutionary Biology at the Univ. of Tennessee.

References Anderson, M. J. 2001. A new method for non-parametric multivariate analysis of variance.  Austral Ecol. 26: 3246. Anderson, M. J. 2004. PERMDISP: a FORTRAN computer program for permutational analysis of multivariate dispersions (for any two-factor ANOVA design) using permutation tests.  Dept of Statistics, Univ. of Auckland, New Zealand. Anderson, M. J. 2005. PERMANOVA: a FORTRAN computer program for permutational multivariate analysis of variance.  Dept of Statistics, Univ. of Auckland, New Zealand. Anderson, M. J. et al. 2006. Multivariate dispersion as a measure of beta diversity.  Ecol. Lett. 9: 683693. Anderson, T. M. 2008. Plant compositional change over time increases with rainfall in Serengeti grasslands.  Oikos 117: 675682. Aplet, G. H. and Vitousek, P. M. 1994. An age-altitude matrix analysis of Hawaiian rain forest succession.  J. Ecol. 82: 137147. Chalcraft, D. R. et al. 2004. Scale dependence in the speciesrichness-productivity relationship: the role of species turnover.  Ecology 85: 27012708. Chao, A. et al. 2005. A new statistical approach for assessing similarity of species composition with incidence and abundance data.  Ecol. Lett. 8: 148159. Chase, J. M. and Leibold, M. A. 2002. Spatial scale dictates the productivity-biodiversity relationship.  Nature 416: 427430. Christensen, N. L. and Peet, R. K. 1984. Convergence during secondary forest succession.  J. Ecol. 72: 2536. Collins, S. L. and Smith, M. D. 2006. Scale-dependent interaction of fire and grazing on community heterogeneity in tallgrass prairie.  Ecology 87: 20582067.

433


Colwell, R. K. 2005. EstimateS: statistical estimation of species richness and shared species from samples.  /<purl.oclc.org/ estimates/>. DufreË&#x2020;ne, M. and Legendre, P. 1997. Species assemblages and indicator species: the need for a flexible asymmetrical approach.  Ecol. Monogr. 67: 345366. Flinn, K. M. and Vellend, M. 2005. Recovery of forest plant communities in post-agricultural landscapes.  Front. Ecol. Environ. 3: 243250. Ford, W. M. et al. 2000. Stand-age, stand characteristics, and landform effects on understory herbaceous communities in southern Appalachian cove-hardwoods.  Biol. Conserv. 93: 237246. Foster, D. R. et al. 1998. Land-use history as long-term broadscale disturbance: regional forest dynamics in central New England.  Ecosystems 1: 96119. Garten, C. T. 2004. Potential net soil N mineralization and decomposition of glycine-C-13 in forest soils along an elevation gradient.  Soil Biol. Biochem. 36: 14911496. Garten, C. T. and Hanson, P. J. 2006. Measured forest soil C stocks and estimated turnover times along an elevation gradient.  Geoderma 136: 342352. Gilliam, F. S. and Turrill, N. L. 1993. Herbaceous layer cover and biomass in a young versus a mature stand of a central Appalachian hardwood forest.  Bull. Torrey Bot. Club 120: 445450. Gilliam, F. S. et al. 1995. Herbaceous-layer and overstory species in clear-cut and mature central Appalachian hardwood forests.  Ecol. Appl. 5: 947955. Gotelli, N. J. and Colwell, R. K. 2001. Quantifying biodiversity: procedures and pitfalls in the measurement and comparison of species richness.  Ecol. Lett. 4: 379391. Harrelson, S. M. and Matlack, G. R. 2006. Influence of stand age and physical environment on the herb composition of secondgrowth forest, Strouds Run, Ohio, USA.  J. Biogeogr. 33: 11391149. Jenkins, M. A. et al. 2007. Impacts of an exotic disease and vegetation change on foliar calcium cycling in Appalachian forests.  Ecol. Appl. 17: 869881. Loreau, M. 2000. Are communities saturated? On the relationship between alpha, beta and gamma diversity.  Ecol. Lett. 3: 7376. Magnuson, J. J. 1990. Long-term ecological research and the invisible present.  Bioscience 40: 495501. Magurran, A. E. 2004. Measuring biological diversity.  Blackwell. Meier, A. J. et al. 1995. Possible ecological mechanisms for loss of vernal-herb diversity in logged eastern deciduous forests.  Ecol. Appl. 5: 935946. Oliver, C. D. and Larson, B. C. 1996. Forest stand dynamics.  Wiley. Peet, R. K. and Christensen, N. L. 1988. Changes in species diversity during secondary forest succession on the North

Download the Supplementary material as file E6016 from / <www.oikos.ekol.lu.se/appendix/>.

434

Carolina Piedmont.  In: During, H. J. et al. (eds), Diversity and pattern in plant communities. SPB Academic Publ., pp. 233245. Peet, R. K. et al. 2003. Variation in species richness and species pool size across a pH gradient in forests of the southern Blue Ridge Mountains.  Folia Geobot. 38: 391401. Phillips, O. L. et al. 2004. Pattern and process in Amazon tree turnover, 19762001.  Phil. Trans. R. Soc. B 359: 381407. Pyle, C. 1988. The type and extent of anthropogenic vegetation disturbance in the Great Smoky Mountains before National Park Service acquisition.  Castanea 53: 183196. Rejma´nek, M. and Rose´n, E. 1992. Cycles of heterogeneity during succession  a premature generalization.  Ecology 73: 23292331. Rooney, T. P. et al. 2004. Biotic impoverishment and homogenization in unfragmented forest understory communities.  Conserv. Biol. 18: 787798. Schmidt, R. G. and Hooks, W. S. 1994. Whistle over the mountain: timber, track & trails in the Tennessee Smokies: an historical and field guide to the Little River Lumber Company and the Little River Railroad in the Great Smoky Mountains National Park in Tennessee.  Graphicom Press. Selmants, P. C. and Knight, D. H. 2003. Understory plant species composition 3050 years after clearcutting in southeastern Wyoming coniferous forests.  For. Ecol. Manage. 185: 275289. Smart, S. M. et al. 2006. Biotic homogenization and changes in species diversity across human-modified ecosystems.  Proc. R. Soc. B 273: 26592665. Stephenson, N. L. and van Mantgem, P. J. 2005. Forest turnover rates follow global and regional patterns of productivity.  Ecol. Lett. 8: 524531. Taverna, K. et al. 2005. Long-term change in ground-layer vegetation of deciduous forests of the North Carolina Piedmont, USA.  J. Ecol. 93: 202213. Vellend, M. et al. 2007. Homogenization of forest plant communities and weakening of species-environment relationships via agricultural land use.  J. Ecol. 95: 565573. Verheyen, K. et al. 2003. Herbaceous plant community structure of ancient and recent forests in two contrasting forest types.  Basic Appl. Ecol. 4: 537546. Vitousek, P. M. et al. 1992. The Mauna-Loa environmental matrix  foliar and soil nutrients.  Oecologia 89: 372382. Wardle, D. A. et al. 2008. The response of plant diversity to ecosystem regression: evidence from contrasting long-term chronosequences.  Oikos 117: 93103. White, E. P. et al. 2006. A comparison of the species-time relationship across ecosystems and taxonomic groups.  Oikos 112: 185195. Yurkonis, K. A. and Meiners, S. J. 2004. Invasion impacts local species turnover in a successional system.  Ecol. Lett. 7: 764769.


ECOGRAPHY 29: 111 /119, 2006

Impact of warming and timing of snow melt on soil microarthropod assemblages associated with Dryas-dominated plant communities on Svalbard Rebecca Dollery, Ian D. Hodkinson and Ingibjo¨rg S. Jo´nsdo´ttir

Dollery, R., Hodkinson, I. D. and Jo´nsdo´ttir, I. S. 2006. Impact of warming and timing of snow melt on soil microarthropod assemblages associated with Dryas- dominated plant communities on Svalbard. / Ecography 29: 111 /119. Open Top Chambers (OTCs) were used to measure impacts of predicted global warming on the structure of the invertebrate community of a Dryas octopetala heath in West Spitsbergen. Results from the OTC experiment were compared with natural variation in invertebrate community structure along a snowmelt transect through similar vegetation up the adjacent hillside. Changes along this transect represent the natural response of the invertebrate community to progressively longer and potentially warmer and drier growing seasons. Using MANOVA, ANOVA, Linear Discriminant Analysis and x2 tests, significant differences in community composition were found between OTCs and controls and among stations along the transect. Numbers of cryptostigmatic and predatory mites tended to be higher in the warmer OTC treatment but numbers of the aphid Acyrthosiphon svalbardicum , hymenopterous parasitoids, Symphyta larvae, and weevils were higher in control plots. Most Collembola, including Hypogastrura tullbergi , Lepidocyrtus lignorum and Isotoma anglicana , followed a similar trend to the aphid, but Folsomia bisetosa was more abundant in the OTC treatment. Trends along the transect showed clear parallels with the OTC experiment. However, mite species, particularly Diapterobates notatus, tended to increase in numbers under warming, with several species collectively increasing at the earlier exposed transect stations. Overall, the results suggest that the composition and structure of Arctic invertebrate communities associated with Dryas will change significantly under global warming. R. Dollery and I. D.Hodkinson (correspondence: i.d.hodkinson@ljmu.ac.uk), School of Biological & Earth Sciences, Liverpool John Moores Univ., Byrom St., Liverpool, L3 3AF, UK. / I. S. Jo´nsdo´ttir, Univ. Centre in Svalbard (UNIS), Post box 156, Longyearbyen, N-9171 Svalbard, Norway.

Many field studies have been conducted on the potential effects of global warming on the growth and community structure of vegetation in polar/alpine ecosystems. These experiments have frequently used some form of passive warming such as a cloche, miniature glasshouse or opentop chamber (OTC) to raise the ambient temperature (Strathdee and Bale 1993, Kennedy 1995, Chapin et al. 1995, Marion et al. 1997, Day et al. 1999, Hollister and Webber 2000, Convey et al. 2002, Sjursen et al. 2005).

Such studies are often standardised in design and form part of a broader geographical network of sites (e.g. The International Tundra Experiment, ITEX) that allow comparison of responses on a circumpolar scale (Molau and Mølgaard 1996). By contrast, equivalent experimental studies on invertebrates are sparse and comparative studies across sites, despite relatively good baseline data derived from the efforts of the International Biological Programme Tundra Biome (Ryan 1981), are

Accepted 19 September 2005 Copyright # ECOGRAPHY 2006 ISSN 0906-7590 ECOGRAPHY 29:1 (2006)

111


few. Detailed sampling programmes for invertebrates, particularly within the soil, have rarely been incorporated into the initial design of these experiments. Consequently, excellent opportunities have been lost and our understanding of how animal communities might respond to changing climate and their impact on soil nutrient dynamics is based on limited data for relatively few sites. Most studies on the effects of climate warming on polar/alpine invertebrate communities have focused on communities associated with particular broad types of vegetation, such as polar semi-desert, tundra heath, montane mixed herb or moss communities (Ryan 1981, Kennedy 1994, Hodkinson et al. 1996, 1998, Rae 2003) or on particular above-ground herbivores associated with selected plant species (Strathdee et al. 1993, 1995, Roy et al. 2004). However, invertebrate communities living in the soil beneath different plant species in tundra, including Dryas octopetala , despite their similar low species richness, may differ in their structure, particularly the relative abundance of species (Coulson et al. 2003). This is probably explained by subtle differences in the physico-chemical environment, including soil temperature and moisture content, below different plant species. Despite this clear link between the soil fauna, vegetation and microclimate, few attempts have been made to exploit the numerous temperature manipulation experiments on plants within polar or montane habitats to measure simultaneously parallel changes in the soil fauna. Where it has been done soil invertebrates have usually been sampled retrospectively (Ruess et al. 1999a,b, Convey and Wynn-Williams 2002, Convey et al. 2002, Sjursen et al. 2005). Experiments to date suggest that different groups of invertebrates respond differently to climate warming. Soft-bodied arthropods, with permeable cuticles, such as soil-dwelling Collembola and some prostigmatic mites, tend to be desiccation susceptible and their populations are apt to decline within experimental chambers as temperatures rise (Hodkinson et al. 1996, Convey and Wynn-Williams 2002, Convey et al. 2002, Sjursen et al. 2005). Cryptostigmatic mites, by contrast, tend to be desiccation resistant and appear less affected, although their slow rate of population turnover makes them potentially slow responders to rising temperature (Hodkinson et al. 1996, Webb et al. 1998). Some above ground herbivores, including the Dryas -feeding aphid Acyrthosiphon svalbardicum responded rapidly within one season, producing a 20-fold increase in the number of overwintering eggs (Strathdee et al. 1993, 1995). Snowmelt gradients reflect in many ways natural conditions that are analogous to those changes that are likely to occur over time in an artificial warming experiment. These include a lengthening of the growing season, warmer mean temperatures and changes in soil moisture levels. This in turn is reflected in differences in 112

plant phenology, growth, flowering and reproduction and distribution, which, together with changes in the rate of soil microbial processes, has consequences for both the above-ground insect herbivores and the soil invertebrates (Galen and Stanton 1995, Walker et al. 1995, Stanton et al. 1997, Sinclair and Sjursen 2001, Heegaard 2002, Totland and Alatalo 2002, Inouye et al. 2003, Van Wijk et al. 2003, Gerber et al. 2004, Stenstro¨m and Jo´nsdo´ttir 2005). This paper reports the results of an Open Top Chamber (OTC) experiment designed to test and measure the impacts of predicted global warming on the structure and species composition of the invertebrate communities of a Dryas octopetala heath in West Spitsbergen, Svalbard. Results and trends from the OTC experiment are compared with variation in invertebrate community structure measured at four stations along a natural snowmelt transect through similar Dryas vegetation up the adjacent hillside. Changes along this transect represents the natural response of the invertebrate community to progressively longer and potentially warmer average growing seasons. The two complementary experiments thus allow comparison of spatial variation in the invertebrate community with respect to local microclimate and the long-term community response to externally imposed climatic variation. Dryas octopetala dominated plant communities are widespread throughout the eastern Arctic where they tend to occupy drier, better-drained sites, often on limestone or other basic substrates (Bliss and Matveyeva 1992, Anon. 2003). They are common and broadly distributed on Svalbard (Elvebakk 1997).

Materials and methods Experimental site The study was conducted on a southeast-facing hillside in Endalen, West Spitsbergen Svalbard (15845?E, 78811?N) where Dryas octopetala heath formed the dominant vegetation type. In addition to the dominant D. octopetala , Salix polaris, Saxifraga oppositifolia , Bistorta vivipara and Carex rupestris also occurred with high frequency. Sampling was conducted along and adjacent to a 135 m snowmelt belt transect (width 20 m) extending directly up the hillside, with transect station 1 situated at the top and station 4 at the bottom. Stations 2 and 3 were spaced equally between. The direction of spring snowmelt was from the bottom to the top of the transect and represented a gradient of decreasing length of growing season and cooler spring temperatures. Late snowmelt, however, ensures that soil moisture remains higher for longer during the early growing season. The OTC experiment straddled the mid point stations (2 and 3) of the snowmelt transect. During 2003 snow clearance dates at transect stations 4, 3, 2, ECOGRAPHY 29:1 (2006)


and 1 was 31 May, 3 June, 7 June and after 10 June, respectively. Snow melt on the main treatment plots was ca 10 d earlier than on the equivalent control plots.

Animal sampling Five randomly located experimental plots were covered with standard hexagonal ITEX design OTCs with a base diameter of 1.5 m (Molau and Mølgaard 1996) at the beginning of the 2001 growing season. A further 8 plots of similar dimensions were identified as controls. The original purpose of the experiment was to examine the response of the plant community to warming and to this end each OTC enclosed a small central inner 0.75 /0.75 m plot, which was sampled continuously but non-destructively for vegetation. Samples for animals, comprising both soil core and pitfall trap samples, were taken randomly in summer 2003 within each chamber but avoiding the central plot. They were thus taken from the broad outer band between the central plot and the edge of the larger enclosing chamber. Five soil core samples per plot (each 4 cm diameter, depth 4 /5 cm) were taken between 3 and 8 July. Equivalent samples were taken from each control plot, where the same spatial pattern of sampling was replicated as in the treatment plots. Core samples were extracted for soil invertebrates using Macfadyen high gradient extractors (Leinaas 1978), collected into benzoic acid solution and then stored in 70% alcohol for subsequent identification. Within each treatment and control plot five 4-cm diameter pitfall traps containing water with a drop of surfactant were layed out on 8 July 2003 and emptied on 11 and 26 July. A further set of soil cores, as above, were taken and extracted from the snowmelt transect between 8 and 21 July. Extractor availability prevented all cores being taken simultaneously and sampling was equally distributed between sites on each date. At each of the four stations samples were taken at 1 m intervals (n /20) along a line extending horizontally 10 m either side of central spot defining the position of that station. Nomenclature of animals follows Fjellberg (1994), Coulson et al. (2003) and Coulson and Refseth (2004).

Environmental measurements Temperature was measured in OTC treatment and control plots using Tinytag Plus data loggers with standard microthermistor probes. Temperatures were recorded in the air immediately above the soil surface, within the top 1 cm of soil and at a height of 1.5 m above the soil. Soil moisture was measured on treatment and control plots and at the various stations along the snowmelt transect, using a HH Soil Moisture Meter with a ML2x/d Theta probe. ECOGRAPHY 29:1 (2006)

Statistical analyses General Linear Model MANOVA, on Log (n/1) transformed invertebrate population data, was used to test for overall differences between treatments, sample plots and replicates in both the OTC and transect experiments. Wilks’ lambda was used as the test statistic. General Linear Model ANOVA, using adjusted sums of squares, was then used to analyse differences in individual species population densities between treatments in the OTC experiment. One-way ANOVA, with post-hoc Tukey and Fisher pairwise comparisons, were applied to compare invertebrate community data for stations along the transect. In these comparisons, involving a large number of tests on individual species, it is recognised that there is a random chance of the occasional spurious significant result. The distinctness of the invertebrate community samples in the OTC treatment and controls and among transect stations was further tested using Linear Discriminant Analysis of individual samples and x2 analyses testing for differences in relative species abundances between treatments in the OTC experiment and among communities at different stations along the transect. In these tests less common species were grouped under the collective heading ‘‘minor taxa’’. For the x2 tests, the null hypothesis was that the numbers of individuals in each species as a proportion of the number of individuals in the whole community did not differ between treatments. The overall x2 value was thus used to indicate significant differences between the treatments/transect stations while the contributing x2 values for individual species was used to identify species that made a significant contribution to the overall x2 value.

Results Microclimate effects Over the summer period 18 June /5 August hourly mean air temperature inside the OTC was 1.28C higher than in the controls and 1.68C higher than the temperature at 1.5 m above the ground (Fig. 1). Soil temperature within the OTC was enhanced by 1.48C compared with the control plot. Soil moisture was variable over short distances and this masked possible trends along the transect and between OTC treatment and control plots.

Invertebrate community composition The full list of species captured and their distribution across sampling sites and stations (Appendix 1) shows a few rare species with highly restricted distributions. These included the Collembolans Micranurida pygmaea and Arrhopalites principalis found occasionally 113


Fig. 1. Mean air and soil temperatures within OTC and control plots in Endalen during the summer growing season 2003.

at transect station 1 or on control plots. By contrast the weevil Isochnus flagellum appeared only towards the bottom of the transect. Most species, however, were broadly distributed, albeit at varying densities, across the sites. MANOVA across all taxa showed significant treatment effects in the OTC core (Wilks, F21,31 /3.82, pB/ 0.001), the OTC pitfall (Wilks, F30,22 /12.53, pB/0.001) and in the transect core samples (Wilks, F75,99 /1.59, pB/0.001). There was an additional significant between sample plot effect for OTC core samples (Wilks, F147,217 /1.47, pB/0.01), but not for OTC pitfall or transect station core samples. The Linear Discriminant Analysis model for the relative abundance of species in soil core samples in the OTC chamber experiment correctly allocated samples to the OTC treatment or control groups with 92 and 87% accuracy respectively. This suggests a high level of predictability with respect to the distinctness of the invertebrate communities in the two treatments. Similarly, a x2 analysis, used to test the hypothesis that the relative species abundance within the invertebrate community of the OTC treatment soil core samples (based on total numbers caught) differed from that in the control samples, was highly significant (x2 /118, pB/0.001, DF /18) (Table 1). The equivalent analysis for pitfall samples was correspondingly highly significant (x2 /1146, pB/0.001, DF/19) (Table 1). The five species making the highest contributions to the overall x2 value in each of these tests are highlighted. In the OTC experiment, the collembolan Folsomia bisetosa was more abundant than expected (treatment vs control) in soil core samples whereas the collembolans Hypogastrura tullbergi and Folsomia quadrioculata were underrepresented. By contrast, the mites Diapterobatus notatus and Prostigmata were overrepresented in pitfall treatment samples. Numbers of the aphid Acyrthosiphon svalbardicum were consistently lower than expected in both soil core and pitfall treatment samples. Results of the GLM analysis for individual species within the OTC experiment showed a number of 114

significant effects, with differences being more accentuated for species in the pitfall than the core samples (Table 2). For pitfall samples, the cryptostigmatic mites D. notatus and Hermannia reticulata , the prostigmatic mites and the gamasid mites were significantly more abundant in OTC samples than in controls. The aphid A. svalbardicum , the collembolans Isotoma anglicana , H. tullbergi, the hymenopteran parasitoid Stenomacrus groenlandicus and Symphyta larvae were most abundant in the controls. The coleopteran I. flagellum and total aphid parasitoids data were similarly close to significance, with greater numbers of individuals caught in the control treatment (see Fig. 2). For core samples, total cryptostigmatic mites, but not individual species alone, were significantly more abundant in the OTCs whereas A. svalbardicum , H. tullbergi , Tetracanthella arctica and Symphyta larvae were more abundant in controls. In addition to the main treatment effect there were also significant among-plot effects for Lepidocyrtus lignorum , I. anglicana , H. tullbergi , Oligophura ursi , H. reticulata and Symphyta larvae and among-sample effects in L. lignorum and H. reticulata . The Linear Discriminant Analysis model of the invertebrate relative species abundance data for the transect allocated samples from the upper and lower limits of the gradient to their correct station with a high degree of certainty (80%), indicating that the communities at the gradient extremes were distinct (Table 3). Communities at the intermediate stations again showed relatively high distinctness, with a majority of samples (60 or 75%) being correctly classified but with higher overlap with other stations. Misclassified samples in most cases were placed in the group for an adjacent station. These differences were again reflected in the highly significant results of a x2 analysis (Table 1) (x2 /118, pB/0.001, DF /18) used to test the hypothesis that the relative species abundances of invertebrates (based on total individuals captured) was identical in samples from different stations along the transect. Taxa displaying a tendency to be more abundant than expected towards the upper end of the transect (Table 1) included Enchrytraeidae, A. svalbardicum , F. quadrioculata and L. lignorum . Species appearing more abundantly towards the bottom of the transect included the mites D. notatus and H. reticulata and the collembolan Tetracanthella arctica . The mite Camisia anomia uniquely lacked a recognizeable trend in abundance. Taxa showing a general trend of decrease from the top (station 1) to the bottom (station 4) in the ANOVA of snowmelt transect data included L. lignorum , H. tullbergi , A. principalis, Areneae and Enchytraeidae (Table 4). Other species, such as F. quadrioculata and Folsomia bisetosa , displayed maximum density at the mid point of the transect, peaking at stations 2 or 3, with the former showing lowest population density at the bottom of the transect. Populations of other species, ECOGRAPHY 29:1 (2006)


F. quadrioculata *** ( /) C. anomia ** ( /) D. notatus (/) T. arctica (/) H. reticulata (/)

4

F. bisetosa (/) A. svalbardicum *** ( /) Enchytraeidae ( /) Other minor taxa ( /) D. notatus (/) F. quadrioculata (/) C. anomia ( /) H. tullbergi ( /) H. reticulata ( /) Cryptostigmata indet. ( /) Enchytraeidae (/) F. bisetosa ( /) D. notatus ( /) A. svalbardicum (/) L. lignorum (/) A. svalbardicum *** ( /) L. lignorum *** ( /) Adult Chironomidae*** ( /) Prostigmatic mites** (/) D. notatus ** (/) H. tullbergi * ( /) A. svalbardicum (/) F. bisetosa (/) F. quadrioculata ( /) Chironomidae larvae ( /) 1 2 3 4 5

OTC pitfalls

1

2

Transect cores

3

Discussion

OTC soil cores

Table 1. Summary of the five species that made the greatest contribution, in descending order of importance, to the significance of each of the x2 tests outlined in the text. (/) indicates greater than expected numbers, ( /) indicates lower than expected numbers. In the case of the OTC experiment this applies to numbers in the OTC treatment versus the control. For the transect it represents deviation from a uniform distribution of species abundances along the gradient. Species that on their own exceed the critical probability value for the overall x2 total in a particular test are indicated by asterisks * p B/0.05, ** pB/0.01, *** p B/0.001. ‘‘Other minor taxa’’ is the grouped total number of individuals in the rarer species and includes some insects, collembolans and mites. ECOGRAPHY 29:1 (2006)

notably Camisia anomia , rose abruptly at the bottom site. Overall these results suggest significant changes in invertebrate community structure along the snowmelt gradient that can be identified in samples taken at successive stations.

One of the main criticisms of retrospective sampling of manipulation experiments, as applied in this and several previous studies, is that equivalent sampling was not carried out at the time that the experiment was established, i.e. there is no absolute baseline at time zero for comparison. Thus, there is a small probability that differences observed among treatments, after a period of elapsed time, existed by chance at the time the experiment was initiated. Furthermore, it is possible that exogenous factors, acting over the duration of the experiment, have produced parallel changes in both treatment and control plots. The parallel trends observed in the transect and OTC experiments suggest, however, that this is not the case. Soil core samples and pitfall traps provide different but complementary information on the invertebrate community. Core samples give a quantitative estimate of invertebrates living in a unit area of soil. Pitfall traps catch many more animals, primarily those moving around on the soil surface, but they also catch small insects that are flying among the low vegetation such as several parasitoid Hymenoptera species. The difference between methods is strongly reflected in the Collembola where pitfalls tend to catch the large surface active species such as L. lignorum and I. anglicana that are more rarely seen in core samples. Conversely, true soil dwelling species such as F. quadrioculata and P. ursi are rarely found in pitfalls. Aphids appear to be efficiently extracted from the base of Dryas in the heated core samples but they also fall into pitfalls in large numbers. Both the OTC and transect experiments revealed significant changes or differences in the structure of the invertebrate community in response to rising ambient temperatures or extended growing season, clearly indicating the potential response of these communities to global warming. Thus, despite broad similarities in the species present, significant differences in the invertebrate community composition between treatments were established for both core and pitfall samples in the OTC experiment and among stations along the snowmelt transect using MANOVA. Differences were sufficient for individual samples in both experiments to be classified as to their origin with a high degree of probability (60%/) using a Linear Discriminant Analysis model and for the significant distinctiveness of the communities between treatments or sites to be demonstrated using x2 analyses. 115


Table 2. Results of General Linear Model ANOVA summarising significant effects for different taxa in the OTC chamber experiment. Probability values for soil core and pitfall trap effects are listed separately. * Signifies close to significant. (/) indicates significantly more in the OTC treatment than in the control, (/) indicates significantly fewer than in the control. Pitfall traps source of variation Taxon A. svalbardicum Total aphid parasitoids L. lignorum I. anglicana H. tullbergi O. ursi T. arctica F. quadrioculata Total Collembola

Between treatments

Among plots

B/0.001 ( /) 0.08* B/0.05 ( /) B/0.01 ( /)

Between treatments

Among plots

Among samples

B/0.05 ( /) B/0.01 B/0.01

B/0.01 B/0.001 ( /)

B/0.01 B/0.05

B/0.05 ( /) B/0.05 B/0.01

Prostigmastic mites Gamasid mites Total predatory mites

B/0.01 (/) B/0.001 (/) B/0.01 (/)

D. notatus H. reticulata Total cryptostigmatic mites

B/0.05 (/) B/0.05 (/) B/0.05 (/)

B/0.05 B/0.05

S. groenlandicus Symphyta larvae I. flagellum

B/0.05 ( /) B/0.001 ( /) 0.06*

B/0.05

The main trends in the OTC experiment tended to be higher numbers of both cryptostigmatic and predatory mites in the warmed treatment plots and higher numbers of aphids, hymenopterous parasitoids, Symphyta larvae,

Fig. 2. Comparison of the captures of the aphid A. svalbardicum (numbers per trap9/SE), its mite predators and hymenopterous parasitoids in pitfall traps placed within OTC and control plots. Data for the two sampling periods are presented separately to illustrate the consistency of the trends. Significant differences in ANOVA tests between control and treatment plots are indicated by ***pB/0.001, **pB/0.01, *pB/0.05.

116

Among samples

Soil cores source of variation

B/0.05 B/0.001 (/)

B/0.05

B/0.001 ( /)

B/0.001

and weevils in the control plots. Most Collembola, including H. tullbergi , L. lignorum and I. anglicana followed a similar trend to the aphid but F. bisetosa was commoner in the OTC treatment, probably reflecting its broader thermal and humidity tolerances (Fjellberg 1994). These results suggest that future warming is likely to produce a significant shift in the composition and structure of the invertebrate communities associated with Dryas. The OTC chambers, as expected, appeared artificially to produce effects that mimicked those that occurred naturally along the snow melt transect. These included an extended growing season, warmer soil temperatures and possibly lower soil moisture, to which the soil invertebrate fauna is collectively responding. Soil animal species responded in analogous ways across experiments, with trends along the transect showing several parallels with the OTC experiment. Aphids and the moistureloving Enchytraeidae, taxa that decreased in the OTC treatment, were more abundant at the later exposed transect stations. Collembola species similarly tended to follow the same negative response to warming as in the OTC experiment, although F. bisetosa increased in abundance in the OTCs and was underrepresented in the site 1 community. Mite species, particularly D. notatus, by contrast, tended to increase in numbers under warming with several species collectively tending to increase in abundance at the earlier exposed transect stations. Our results show similar trends to climatic temperature manipulation experiments elsewhere and conform ECOGRAPHY 29:1 (2006)


Table 3. Results of Linear Discriminant Analysis showing the allocation of soil core samples from the four stations (n/20 per station) along the snowmelt transect to their correct or incorrect group. True group Allocated group Station 1 Station 2 Station 3 Station 4 % correct

Station 1

Station 2

Station 3

Station 4

16 2 2 0 80

2 12 6 0 60

1 3 15 1 75

1 1 2 16 80

Localised variation in some extraneous factor such as soil moisture or distribution of a particular plant species may provide an explanation. Similar spatial aggregation, independent of treatment, occurred in soil arthropod communities in chamber experiments on Anvers Island, West Antarctic Peninsula (Convey et al. 2002). A major contrast in response to climate warming is seen in the aphid A. svalbardicum in the Endalen experiment compared with previous studies at ˚ lesund where enclosed cloches produced a 20-fold Ny-A increase in population over one year compared with controls (Strathdee et al. 1993, 1995). Initial nonquantitative casual observations during year 1 suggested that this was happening also in Endalen. However, the populations at Endalen within the OTCs were significantly lower after the second season, as measured by both core and pitfall sampling, than in the controls. Furthermore, populations of aphids were higher towards the upper cooler end of the transect. There are several possible explanations. Endalen is significantly warmer ˚ lesund. It is known that enhanced temperathan Ny-A tures produced a stimulatory effect on growth and reproduction at low temperatures. However, it may be that once temperatures exceed an optimum then

to predictions by Hodkinson et al. (1998) that under conditions of climate warming, without increased moisture, desiccation-susceptible Collembola numbers would tend to decrease as habitats became warmer and drier but that numbers of desiccation resistant Acarina would increase. Coulson et al. (1996) found declining numbers of Collembola in cloches after 3 yr, a trend repeated in similar OTC experiments in Antarctica (Convey et al. 2002) and subarctic Sweden (Sjursen et al. 2005). Kennedy (1994), by contrast, found higher numbers of the dominant collembolan Cryptopygus antarcticus and other arthropods in cloche treatments in similar experiments on Signy Island, Antarctica, a fact that might be explained by increased moistureretaining ground vegetation cover in a longer term (8 yr) experiment. Both the overall MANOVA for the OTC core samples and the corresponding GLM ANOVAs for core and pitfall trap samples for individual species showed evidence for a significant among-plots effect as well as the main treatment effect. This suggests that for those species involved, primarily Collembola such as L. lignorum and H. tullbergi , there was an element of patchiness in their abundance operating at the plot scale.

Table 4. Summary of significant differences for individual species among stations along the transect derived from a one-way ANOVA of core sample data using Fisher’s (F) and Tukey’s (T) pair wise comparisons. Overall significant differences along the transect are indicated by critical probability values. Some species are included where pair wise comparisons indicate significance between at least two transect stations but the overall ANOVA is only approaching significance. In these cases the exact probability value is stated. Transect station 1 Species Collembola total L. lignorum H. tullbergi F. quadrioculata A. principalis F. bisetosa T. arctica C. anomia H. reticulata Araneae Enchytraeidae

ECOGRAPHY 29:1 (2006)

F

2 T

4 3,4 3 2,3,4 4 3,4

3,4

3

F

T

F

4

4

4 1 4 4 4 4 3

4

4 1 1 4 1

4

4 4 2 1

4 T

4

4 1

F

T

1,2, 3 1

2

2,3 1 2 2 1,2,3 2,3

2,3

1

1

2,3

Overall p B/0.01 0.07 0.15 B/0.001 0.10 0.07 0.10 B/0.01 0.07 0.13 B/0.01

117


warming begins to produce deleterious effects. This could partially explain the numbers both in the plots and along the transect. A second explanation might lie in the increased numbers of predators, particularly prostigmatic and gamasid mites, and host-specific aphid parasitoids within the OTC when compared with controls. Numbers of predatory mites were significantly higher within the OTCs (Fig. 2) and numbers of parasitoids were similarly approaching significance. This points to a delayed build up of predator/parasitoid numbers within the OTCs that was subsequently impacting significantly on the aphid population. It is also notable that treat˚ lesund, were enclosed, thereby ment cloches at Ny-A excluding flying parasitoids. The OTC’s in Endalen were also less efficient at raising the temperature/1.2 degrees compared with 2.8 degrees for the cloches at Ny˚ lesund. A It appears that desiccation susceptible predatory mites may be able to sustain body moisture by feeding on plant sap-sucking aphids within the OTCs, as commonly observed. Altenatively, these active mites may enter the OTC under the plastic wall and replenish declining populations. Physical exclusion may also explain why Symphyta larvae were common outside the OTCs but rare inside. Adult female sawflies are strongly dispersive and tend to fly immediately above the ground surface: the walls of the OTCs may thus form a significant barrier to colonisation. While parasitoids and predators may be exerting a top down effect on the population of some prey species, bottom up effects, induced by warmer conditions, may also be important, either directly or indirectly increasing food availability by changing the rate of microbial release of nutrients within the soil (Hedlund and Ohrn 2000). Several experiments have demonstrated effects of altered nutrient availability on soil invertebrates, including Enchytraeidae, nematodes and Collembola, under conditions of warming (Ruess et al. 1999a, b, Convey and Wynn-Williams 2002, Cole et al. 2002, Sjursen et al. 2005). The effect may also be physical, with improved or reduced vascular plant growth altering the available inter-plant habitat space occupied by surface-active species, such as several Collembola, or restricting the effect of incident UV-B (Convey et al. 2002). Our experiments indicate several significant changes in existing invertebrate community structure that are likely to occur in the short to medium term under conditions of climate warming. It fails inevitably, however, to account for the probability of immigration by more thermophilous species and the extinction of cold adapted species that is likely to occur over the longer term. Nevertheless, despite the obvious limitations, retrospective sampling on an ad hoc basis can provide valuable insights into changes that are occurring within invertebrate communities in response to environmental manipulation. 118

Acknowledgements / We thank Arne Fjellberg for checking the identity of the Collembola species. UNIS provided financial support and logistics for field work.

References Anon. 2003. Circumpolar Arctic vegetation map. Scale 1:7,500,000. Conservation of Arctic Flora and Fauna (CAFF)-Map No. 1. / U.S. Fish and Wildlife Service, Anchorage, AK. Bliss, L. C. and Matveyeva, N. V. 1992. Circumpolar arctic vegetation. / In: Chapin, F. S. et al. (eds), Arctic ecosystems in a changing climate. An ecophysiological perspective. Academic Press, pp. 59 /89. Chapin, F. S. et al. 1995. Responses of arctic tundra to experimental and observed changes in climate. / Ecology 76: 694 /711. Cole, L. et al. 2002. Enchytraeid worm (Oligochaeta) influences on microbial community structure, nutrient dynamics and plant growth in blanket peat subjected to warming. / Soil Biol. Biochem. 34: 83 /92. Convey, P. and Wynn-Williams, D. D. 2002. Antarctic soil nematode response to artificial climate amelioration. / Eur. J. Soil Biol. 38: 255 /259. Convey, P. et al. 2002. Response of antarctic terrestrial microarthropods to long-term climate manipulations. / Ecology 83: 3130 /3140. Couslon, S. J. and Refseth, D. 2004. The terrestrial and freshwater invertebrate fauna of Svalbard (and Jan Mayen). / In: Prestrud, P., Strøm, H. and Goldman, H. V. (eds), A catalogue of the Svalbard terrestrial and marine animals: invertebrates, fishes, birds and mammals. Skrifter 201, Norwegian Polar Inst., Tromsø, pp. 57 /122. Coulson, S. J. et al. 1996. Effects of experimental temperature elevation on high-arctic soil microarthropod populations. / Polar Biol. 16: 147 /153. Coulson, S. J., Hodkinson, I. D. and Webb, N. R. 2003. Microscale distribution patterns in high Arctic soil microarthropod communities: the influence of plant species within the vegetation mosaic. / Ecography 26: 801 /809. Day, T. A. et al. 1999. Growth and reproduction of Antarctic vascular plants in response to warming and UV radiation reductions in the field. / Oecologia 119: 24 /35. Elvebakk, A. 1997. Tundra diversity and ecological characteristics of Svalbard. / In: Wielgolaski, F. E. (ed.), Polar and Alpine tundra. Ecosystems of the world 3. Elsevier, pp. 347 / 359. Fjellberg, A. 1994. The Collembola of the Norwegian arctic islands. / Medd. Norskpolarinst. 133: 1 /57. Galen, C. and Stanton, M. L. 1995. Responses of snowbed plant-species to changes in growing-season length. / Ecology 76: 1546 /1557. Gerber, J. D., Baltisberger, M. and Leuchtmann, A. 2004. Effects of a snowmelt gradient on the population structure of Ranunculus alpestris (Ranunculaceae). / Bot. Helv. 114: 67 /78. Hedlund, K. and Ohrn, M. S. 2000. Tritrophic interactions in a soil community enhance decomposition rates. / Oikos 88: 585 /591. Heegaard, E. 2002. A model of alpine species distribution in relation to snowmelt time and altitude. / J. Veg. Sci. 13: 493 /504. Hodkinson, I. D. et al. 1996. Can high Arctic soil microarthropods survive elevated summer temperatures? / Funct. Ecol. 10: 314 /321. Hodkinson, I. D. et al. 1998. Global change and Arctic ecosystems: conclusions and predictions from experiments with terrestrial invertebrates on Spitsbergen. / Arct. Alp. Res. 30: 306 /313. ECOGRAPHY 29:1 (2006)


Hollister, R. D. and Webber, P. J. 2000. Biotic validation of small open-top chambers in a tundra ecosystem. / Global Change Biol. 6: 835 /842. Inouye, D. W., Saavedra, F. and Lee-Yang, W. 2003. Environmental influences on the phenology and abundance of flowering by Androsace septentrionalis (Primulaceae). / Am. J. Bot. 90: 905 /910. Kennedy, A. D. 1994. Simulated climate-change / a field manipulation study of polar microarthropod community response to global warming. / Ecography 17: 131 /140. Kennedy, A. D. 1995. Temperature effects of passive greenhouse apparatus in high-latitude climate-change experiments. / Funct. Ecol. 9: 340 /350. Leinaas, H. P. 1978. Sampling of soil microarthropods from coniferous forest podzol. / Norw. J. Entomol 25: 57 /62. Marion, G. M. et al. 1997. Open-top designs for manipulating field temperature in high-latitude ecosystems. / Global Change Biol. 3 (Suppl. 1): 20 /32. Molau, U. and Mølgaard, P. 1996. ITEX manual. / Danish Polar Centre, Copenhagen. Rae, D. A. 2003. Plant and invertebrate community responses to species interaction and microclimatic gradients in alpine and Arctic environments. / Ph.D. thesis, Dept of Biology, Norwegian Univ. of Science and Technology. Roy, B. A., Gusewell, S. and Harte, J. 2004. Response of plant pathogens and herbivores to a warming experiment. / Ecology 85: 2570 /2581. Ruess, L., Michelsen, A. and Jonasson, S. 1999a. Simulated climate change in subarctic soils: responses in nematode species composition and dominance structure. / Nematology 1: 513 /526. Ruess, L. et al. 1999b. Simulated climate change affecting microorganisms, nematode density and biodiversity in subarctic soils. / Plant Soil 212: 63 /73. Ryan, J. K. 1981. Invertebrate faunas at IBP tundra sites. / In: Bliss, L. C., Heal, O. W. and Moore, J. J. (eds), Tundra ecosystems a comparative analysis. Cambridge Univ. Press, pp. 517 /539.

Download the appendix as file E 4366 from B/www.oikos.ekol.lu.se/appendix /.

ECOGRAPHY 29:1 (2006)

Sinclair, B. J. and Sjursen, H. 2001. Terrestrial invertebrate abundance across a habitat transect in Keble Valley, Ross Island, Antarctica. / Pedobiologia 45: 134 /145. Sjursen, H., Michelsen, A. and Jonasson, S. 2005. Effects of long-term soil warming and fertilisation on microarthropod abundances in three sub-arctic ecosystems. / Appl. Soil Ecol. 30: 148 /161. Stanton, M. L., Galen, C. and Shore, J. 1997. Population structure along a steep environmental gradient: consequences of flowering time and habitat variation in the snow buttercup, Ranunculus adoneus. / Evolution 51: 79 / 94. Stenstro¨m, A. and Jo´nsdo´ttir, I. S. 2005. Effects of simulated climate change on phenology and life history traits in Carex bigelowii . / Nord. J. Bot., in press. Strathdee, A. T. and Bale, J. S. 1993. A new cloche design for elevating temperature in polar terrestrial ecosystems. / Polar Biol. 13: 577 /580. Strathdee, A. T. et al. 1993. Effects of temperature elevation on a field population of Acyrthosiphon svalbardicum (Hemiptera, Aphididae) on Spitsbergen. / Oecologia 96: 457 /465. Strathdee, A. T. et al. 1995. Climatic severity and the response to temperature elevation of Arctic aphids. / Global Change Biol. 1: 23 /28. Totland, O. and Alatalo, J. M. 2002. Effects of temperature and date of snowmelt on growth, reproduction, and flowering phenology in the arctic/alpine herb, Ranunculus glacialis. / Oecologia 133: 168 /175. Van Wijk, M. T. et al. 2003. Interannual variability of plant phenology in tussock tundra: modelling interactions of plant productivity, plant phenology, snowmelt and soil thaw. / Global Change Biol. 9: 743 /758. Walker, M. D., Ingersoll, R. C. and Webber, P. J. 1995. Effects of interannual climate variation on phenology and growth of two alpine forbs. / Ecology 76: 1067 /1083. Webb, N. R. et al. 1998. The effects of experimental temperature elevation on populations of cryptostigmatic mites in high Arctic soils. / Pedobiologia 42: 298 /308.

Subject Editor: John Spence.

119


Ecography 30: 193  208, 2007 doi: 10.1111/j.2007.0906-7590.04818.x Copyright # Ecography 2007, ISSN 0906-7590 Subject Editor: John Spence. Accepted 5 February 2007

From forest to pasture: an evaluation of the influence of environment and biogeography on the structure of dung beetle (Scarabaeinae) assemblages along three altitudinal gradients in the Neotropical region Federico Escobar, Gonzalo Halffter and Lucrecia Arellano F. Escobar (federico.escobarf@gmail.com), G. Halffter and L. Arellano, Dept de Biodiversidad y Comportamiento Animal, Inst. De Ecolog覺織a A.C., Apartado Postal 63, 91000 Xalapa, Veracruz, Mexico. (Present address of F. E.: Dept of Zoology and Entomology, Univ. of Pretoria, 0002 Pretoria, South Africa.)

The objective of this study is to evaluate the effect of environmental (associated with the expansion of cattle ranching) and biogeographical factors on the diversity of dung beetle (Scarabaeinae) assemblages along three altitudinal gradients in the Neotropical region. One gradient is located in the Mexican Transition Zone, on the Cofre de Perote mountain, the other two are in the northern Andes (the Chiles Volcano and the R覺織o Cusiana Basin). For the three gradients, the number of species decreased as altitude increased. On the Cofre de Perote, regardless of altitude, the number of species and of individuals was similar in both forest and pasture, while species composition was different between habitats. On this mountain, species turnover in pastures was characterized by the addition of new species as altitude increased. In the northern Andes, species diversity was always greater in the forest than in the pasture, and species turnover between habitats was notably influenced by species loss with increasing altitude. As such the pasture fauna of the northern Andes was an impoverished derivative of the fauna present in the forests at the same altitude characterized by species of Neotropical affinity with a limited capacity for colonizing open, sunnier habitats. The opposite occurs in the areas used by cattle on the Cofre de Perote. This habitat has its own fauna, which is mainly comprised of Holarctic and Afrotropical species adapted to the prevailing environmental conditions of areas lacking arboreal vegetation. These results suggest that the impact on beetle communities caused by human activities can differ depending on the geographic position of each mountain and, particularly, the biogeographical history of the species assemblage that lives there.

A decrease in species richness and changes in the composition of the flora and fauna with increasing altitude have frequently been described in the literature (Huston 1994). However, it is not possible to describe a general relationship between changes in diversity and increasing altitude (Rahbek 1995), because the ecological factors (current environmental conditions, e.g. land use) and biogeographical factors differ in their relative effect depending on the mountain system being studied (Brown 2001). This is emphasized by the fact that we still do not understand how these factors vary and interact as altitude changes. For this reason comparative studies using the same taxonomic group

along altitudinal gradients on different mountains (such as this one), and those that use several taxonomic groups along the same altitudinal gradient of the same mountain are very useful for explaining this co-variation (Lomolino 2001). Studying the altitudinal variation in dung beetle assemblages on different mountains around the world, Lobo and Halffter (2000) proposed two different biological and interrelated processes to explain the conformation of the biota on mountains, the patterns of species richness and variations in composition: horizontal colonization by elements originating from lineages inhabiting higher altitudes, and vertical

193


colonization by lineages from surrounding lower lands at the same latitude. The relative effect of both processes depends on the orientation and location of the mountains, and on their degree of isolation and biogeographical history, as these characteristics greatly influence the refuge and ‘‘corridor’’ capacity of mountain areas (Lobo and Halffter 2000). Two hypotheses for explaining the changes in diversity with altitude emerge from the relative influence of these processes: 1) the mountain fauna is composed of a lesser number of phylogenetically related species relative to the fauna of lower altitudes (colonization vertical model), and 2) the mountain fauna is composed of elements with different evolutionary histories and origins compared to the lowlands fauna (horizontal colonization model). According to the first hypothesis, we would expect to find mountains that, owing to their geographic isolation, limited extension or recent geological formation, have yet to accumulate their own species assemblages. Therefore we would expect species substitution to be slow and species richness to notably decrease with increasing altitude. This would be the consequence of the environmental restrictions imposed by high altitudes on the fauna from warmer altitudes, especially in tropical regions (Janzen 1967). This appears to be the case for the northern Andes (Escobar et al. 2005) and southeastern Asia (Hanski 1983). In contrast, in the horizontal colonization model (second hypothesis), geographical and historical factors become especially relevant. This situation has been described for the mountains located in the central Iberian Peninsula (Martin-Piera et al. 1992), those of southern France (Errouissi et al. 2004) and of the Mexican Transition Zone (MTZ, Halffter 1987). Mountains that owing to their geographic location and orientation have served as refuge and speciation areas for lineages from more northern regions during the climatic changes of the Plio-Pleistocene. They consequently host a fauna adapted to cold environment and annual fluctuations in climate (Halffter 1987, Martin-Piera et al. 1992). When horizontal colonization is the main process governing the establishment of mountain fauna, one would expect to find greater phylogenetic diversity, fast species substitution and a less pronounced decrease in species richness with increasing altitude (Lobo and Halffter 2000). Our ability to understand contemporary biogeographical patterns also relies on our understanding of human impact, and specifically on how human impact affects natural ecosystems, modifying the spatial distribution of species, community and population structure throughout large geographic areas (Lomolino and Perault 2004). One such activity is cattle ranching in the mountains of the Neotropical region, which has resulted in a continuous increase in the area covered by pasture and is responsible for the homogenization of the

194

mountain landscape (Kappelle and Brown 2001). During this process, sunnier open areas are created and the environmental conditions there become much more severe. Additionally, the quantity of dung, mainly from cattle, is greater, which has been shown to modify the structure of the dung beetle assemblages at the local and landscape levels in the different tropical and subtropical regions of the world (Halffter 1991, Nichols et al. in press). From these studies, the general conclusion is that vegetation cover determines species abundance and richness of this group of beetles. The association of the species with a given type of habitat appears to be related to its micro-climatic requirements (i.e. temperature, relative humidity, light intensity) and to its close dependence on mammalian dung for feeding and reproduction (Halffter and Matthews 1966). It has also been observed, however, that the preferences of species for certain habitats, both natural and anthropogenic, varies with altitude in different ways depending on the geographic position of the mountain and, therefore, the biogeographical affinity of the fauna that live there (Halffter et al. 1995, Davis et al. 1999, Romero-Alcaraz and Avila 2000, Errouissi et al. 2004). From the above, it might be expected that the effects of expanding cattle ranching on dung beetle assemblage diversity would be different in each mountain region. To evaluate this prediction we studied the diversity patterns of dung beetles along three altitudinal gradients that are ecologically similar but have different biogeographical histories. One is in the MTZ and two are in Colombia, located on opposite slopes of the northern Andes. In both mountain systems the main change in the landscape is a notable conversion of the original forest into cattle pasture and, to a lesser degree, agricultural crop fields. Given that the pastures are a relatively new type of vegetation, and the mountains vary with respect to the influence of the vertical vs horizontal colonization processes, we expect: a) the fauna of the pastures in the northern Andes to be a rarefied subset of the species found in the surrounding forested areas, comprised of species belonging to genera of Neotropical affinity with a limited capacity for colonizing open areas (Amat et al. 1997, Escobar 2004), and b) the fauna of the MTZ pastures to be a mixture of species of wide ecological tolearance, capable of leaving the forested sites, and species from nonNeotropical lineages adapted to the prevailing conditions of the areas without any arboreal cover (Halffter et al. 1995, Arellano and Halffter 2003). In this study we address the following questions: 1) how does dung beetle diversity changes with respect to altitude along three altitudinal gradients? 2) How do changes in the dung beetle assemblages occur in forests and pastures at each altitude, along the altitudinal gradient, and according to the geographical position of the mountain? 3) How do environmental and biogeographical factors


influence any differences detected (points 1 and 2) between the MTZ gradient and the two northern Andean gradients?

coffee plantations and at high altitudes, seasonal agriculture (wheat, oats and potato) and intensive dairy cattle ranching (Challenger 1998).

Materials and methods

Northern Andes, the Chiles Volcano and the Rı´o Cusiana

MTZ, Sierra Madre Oriental-Sistema Volca´nico Transversal Halffter (1987) defined the MTZ as a complex and varied region extending from northern Mexico to southern Nicaragua in which Nearctic and Neotropical biota overlap. To the north, there are Nearctic elements that gradually decrease towards the south. The northern lineages have dispersed through the mountains, which in the MTZ have a generally N-S orientation, facilitating horizontal colonization. In contrast, the coastal plains and tropical lowlands are the penetration route for Neotropical elements. This double occupation of the territory that occurs latitudinally in the MTZ has an altitudinal equivalent in the mountains: higher altitudes are occupied by lineages of northern affinity, lower altitudes by lineages of neotropical affinity and intermediate altitudes are characterized by an overlap of these two lineages and strong in situ speciation, particularly of those lineages with a long evolutionary history in the zone. The altitudinal gradient is located on the eastern slope of the Cofre de Perote volcano in the state of Veracruz (19819?19839?N, 96824?97812?W; Fig. 1) and on the eastern end of the Sistema Volca´nico Transversal where it meets the Sierra Madre Oriental. The Sierra Madre Oriental is comprised of a series of NNW-SSE folds, with an average width of 80 to 600 km each. It starts in the north on the Texas platform and is interrupted by the Sistema Volca´nico Transversal. The latter is a complex mountain chain that runs W-E and is 950 km long, and 50150 km wide. It is considered one of the youngest mountain systems in the country (Pliocene-Quaternary, 2 3 million years BP; Ferrusquı´a-Villafranca 1993). This altitudinal gradient covers different types of vegetation: tropical deciduous forest in the lowlands (B1000 m a.s.l.; temperature: 228248C; annual precipitation: 1500 2000 mm), cloud forest, oak forest and pine-oak forest at intermediate altitudes (1000  2000 m a.s.l.; 128 188C; 2000 3000 mm), pine forest and oyamel fir forests at 2000 3000/3500 m a.s.l. (58 128C; 1000 1800 mm). Above this altitude, there are high natural pastures (2858C; B1200 mm). Land use varies with altitude. In the lowlands extensive cattle ranching is the main land use, but there are also irrigation agriculture, sugar cane and fruit crops (orange, mango and tamarind). At intermediate altitudes there are corn crops, dairy farming and especially

The northern Andes belong to a mountain system that extends from northern Peru (the Huancabamba Depression) to Venezuela. In Colombia, the Andes represent an enormous mass of mountains that occupies ca 30% of the country, and diverges into three branches or ranges: the Western, the Central and the Eastern, each extending in a general south-north direction. The Western Range is considered one of the oldest ranges of the Colombian Andes (Oligocene, 38 million years BP). This range is ca 650 km long and is the narrowest of the three at B50 km wide and altitudes no higher than 4800 m a.s.l. The Eastern Range, with a length of 1200 km, is 200 km at its widest and reaches altitudes 5500 m a.s.l. It is considered the main mountain chain of the northern Andes. It originated in the Miocene (18 million years BP) and its final formation occurred in the PliocenePleistocene (2.5 million years BP; van der Hammen and Hooghiemstra 2001). The slopes of the altitudinal gradients studied in the Colombian Andes have different aspects. The Chiles Volcano (0810? 1817?N, 78815?77811?W; Narin˜o Department) is on the western slope of the Eastern Range, facing the Pacific Plain. The other gradient is in the Rı´o Cusiana Basin (5826?5823?N, 72841? 728 42?W, Boyaca´ Department), on the eastern slope of the Eastern Range, facing the Amazonia-Orinoquia region. The following vegetation types are found in the northern Andes: tropical lowland forest (up to 1000/ 1250 m a.s.l.; 228 268C; 4000 8000 mm), tropical sub-Andean forest (1000/12502000/2300 m a.s.l.; 168228C; 2000 4000 mm); high Andean forest (2000/23003200/3600 m a.s.l.; 68128C; 1000 1500 mm) and pa´ramos (above 3200/3900 m a.s.l.; 3868C; 500 1000 mm). The limits of the vegetation zones vary depending on the topography and local climate, as indicated by the values given in parentheses (van der Hammen 1995). On the Chiles Volcano, in addition to extensive cattle ranching, the lowlands are mostly used for cultivating African palm, bananas and corn. At intermediate altitudes there are coffee plantations, sugar cane and cattle ranching; and the highest altitudes (above 2500 m a.s.l.) are used for dairy farming and potato crops. In the Rı´o Cusiana basin, human activites have reduced the forested areas to remnants of varying sizes along the entire altitudinal gradient. The lower part of the mountain is dominated by pastures and small parcels with corn crops, intermediate altitudes are used for banana and sugar cane

195


Forest

Pasture

(a)

20

20

16

16

12

12

8 4

30 m 900 m 1800 m

450 m 1360 m 2000 m

2340 m

3300 m

8 4

0

0

Cumulative number of species

0 20

200

400

600

0

800

100

200

300

20

(b)

16

16

12

12 8

8 4

50 m

520 m

1000 m

1350 m

4

1800 m

0

0 0 20

200

400

600

0

800

50

100

150

200

20

(c)

16

16

12

12

8

4

450 m

900 m

1250 m

1450 m

1750 m

2000 m

8

4

2250 m 0

0 0

100

200

300

400

0

20

40

60

80

100

Cumulative number of individuals Fig. 1. Smoothed species accumulation curves using the number of individuals collected as a substitute for the sampling effort applied at each forest and pasture site along each altitudinal gradient: (a) Cofre de Perote; (b) Chiles Volcano (except the pasture at 1000 m a.s.l.) and, (c) R覺織o Cusiana (except the pasture at 2000 m a.s.l.). For those sites where abundance values were 52 individuals it was not possible to estimate species richness, such as at 2600 and 3300 m a.s.l. for the Chiles Volcano.

crops, and extensive cattle ranching, and finally the highest altitudes are mostly used for dairy farming and potato, barley and wheat crops. In both the MTZ and the northern Andes mountain systems, the transformation of the forest into pastures for cattle began with the arrival of the Spaniards 500 yr ago, and has dramatically modified native ecosystem throughout tropical America (Murgueitio 2003). It is currently estimated that ca 33% (602 million ha) of this

196

region is covered by permanent cattle pastures (Anon. 2002). In the mountainous regions of the Americas, the most conspicuous changes to the landscape occurred at the beginning of the 20th century, particularly form the 1950s onwards and the transformation process continues at an alarming pace to this day (Challenger 1998, Kapelle and Brown 2001). In these mountains the cattle management systems vary widely and largely depend on the climatic and topographic conditions. They also vary


in size and range from 1 to 500 ha (Murgueitio 2003). In the lowlands there are 1 8 cows ha1. At intermediate altitudes there is greater variation (1 15 ha1) and at very high sites where natural pastures are used, cattle density is 5 8 cows ha 1 (Anon. 2002). Sampling Sampling was carried out in two contrasting habitat types occurring along each of the three altitude gradients: forested areas and induced pastures used for cattle. Beetles were caught with buried pitfall traps (top flush with the soil) with two types of bait (excrement and carrion). The bait was wrapped in muslin and suspended from a wire right above the trap. The volume of the traps was ca 1000 ml (13 mm deep and 11 mm in diameter) and a mixture of water and detergent was placed inside to prevent caught beetles from leaving. On the Cofre de Perote, 16 sites were sampled at eight altitudes between 50 and 3000 m a.s.l., from May to October 1994. At 450 m a.s.l., the sites were sampled in April and May of 1993. At each site, a line of traps was set with alternating bait of fresh excrement (human and cattle mixed) and decomposing squid. Eight to 17 traps were set per forest site (mean9SD: 12.893.0) and 15 traps were set in the pastures (12.591.9). Traps were placed 2530 m apart and left in the field for one day and one night (24 h) before being collected. On the Chiles Volcano, 13 sites were sampled at seven altitudes between 50 and 3300 m a.s.l. during April and September 1993. At each site we placed a line of 12 traps with 2530 m between traps. Similarly, in the Rı´o Cusiana basin, 13 sites were sampled at seven altitudes between 450 and 2500 m a.s.l. during May and June 1997. At each site we placed a line of 10 traps, with the traps 2530 m apart. In both cases, traps were alternately baited with fresh human excrement and decomposing meat. Baited traps were left in place for two days and two nights (ca 48 h) at each site before being collected. For the gradients in the northern Andes it was not possible to collect from the pastures at 1000 m a.s.l. (on the Chiles Volcano) or at 2000 m a.s.l. (Rı´o Cusiana) owing to the lack of sites appropriate for sampling. Data analysis Given that sampling effort was different in each mountain region, we used accumulation curves with the number of individuals collected, rarefaction based on individuals, as a measure of sampling effort. For each site, the number of species observed was obtained995% confidence interval (Colwell et al. 2004). As an estimate of species richness, we used the Michaelis-Menten equation (MM), one of the curvi-

linear asymptotic functions most commonly used in the evaluation of diversity inventories and adequate for a small number of samples (Colwell and Coddington 1994). The smoothed accumulation curves were obtained by repeated random reordering (500 times) of the samples using v. 7.5.0 of EstimateS program (Colwell 2005). Analyses were carried out for three levels of comparison: a) total diversity (Gamma diversity, g) defined in this case as the cumulative number of species by habitat type along each altitudinal gradient, b) local diversity between habitats (Alpha diversity, a) along each gradient, using the total number of species recorded at each site (St) and the mean number of species per trap (Sm) and, c) species turnover (Beta diversity, b). To compare the process of g diversity accumulation in each habitat type along each gradient, we calculated the slope of the linear regression of altitude (independent variable) against the observed cumulative number of species (dependent variable). We used a Student’s t to test whether the slopes (forest vs pasture) were significantly different between habitats (Zar 1996). In order to determine the relationship between local species richness (quantified as St and Sm) and habitat along each gradient, we used an analysis of covariance (ANCOVA) with altitude as the covariable. For all cases, the model fit to the data Y mHabitat AltitudeHabitat Altitudeo. For St, we obtained the complete model assuming a Poisson distribution of errors (link function Log; Crawley 2002). For Sm, error distribution was assumed to be normal. In both cases, the model was verified by examining the standardized residuals vs the fit values, in addition to the graphical distribution of errors. Species turnover along each gradient was analyzed in two ways: a) between adjacent altitudes for each habitat type along each altitudinal gradient and, b) between habitat types (forest vs pasture) at each altitudinal level. Wilson and Shmida’s (1984) index (bt) was used: (bt) (ac)/(2abc), where a is the number of species found at two sites and b and c are the number of species lost and gained in each comparison. Values of bt vary between 0 and 1, with 1 indicating the greatest degree of dissimilarity between sites. This index produces results similar to those of other indices of b diversity and is one of the most recommended since it provides a direct expression of species turnover when the samples are arranged along an environmental gradient and because it is independent of a diversity (Wilson and Shmida 1984). Since the indices of b diversity do not reveal whether turnover values are a product of the loss or gain of species, and in order to understand the relative influence of each process, we calculated the number of species lost and gained for each comparison.

197


The abundance distribution of species in each habitat was compared using range-abundance curves. These curves can also be used to describe the changes in community structure (Magurran 1988). In order to determine how different observed changes are from random differences in the structure of the beetle community when forest is replaced by pasture at each altitude, we used the test developed by Solow (1993, available in the program developed by Henderson and Seaby 2002). This randomization test can be used together with any other species abundance-based measure of community structure. We used Simpson’s index (D) (in its reciprocal form 1/D) to evaluate the change in diversity between sites: D ani (ni  1)=[N(N1)]; where ni is the number of individuals of species i and N ani : Simpson’s index represents the probability that two individuals randomly selected from a sample belong to different species, and in its reciprocal expression is a measure of dominance (Magurran 1988). In Solow’s test (1993), the observed change (d) in 1/D is compared with the values obtained from 10000 random partitions of the total sample of individuals in a set of samples similar in size to the observed. The statistical significance of the observed value of d can be evaluated by its position relative to those of the ordered values of d obtained randomly. In this test, the value of probability for a two-tailed test is given by the proportion of partitions where the simulated value ½d½ is greater than the observed value ½d½.

Results On the Cofre de Perote, we captured a total of 3245 individuals belonging to 40 species. The number of species and individuals caught in the forest was similar to that of the pasture (Table 1). For this mountain the rate of accumulation of g diversity was not different between habitats (bforest 8.1 species/1000 m, bpasture 7.5 species/1000 m; t1.62, DF 12, p 0.13). In contrast, the dung beetle diversity in the northern Andes was remarkably different between habitats. On the Chiles Volcano 1746 individuals belonging to 37 species were caught: 89% of the species (87% of individuals) were from the forest and 57% of the species (13% of individuals) were from the pasture (Table 1). Although there was no significant difference in the rate of accumulation of g diversity with increasing altitude, the value was higher in the forest than in the pasture (bforest 5.4 species/1000 m, bpasture 3.4 species/1000 m; t 1.79, DF 9, p  0.10). At Rı´o Cusiana, 1518 individuals belonging to 49 species were caught. Of these 88% (90% of individuals) were caught in the forest and 45% (10% of individuals) were caught in the pasture (Table 1)

198

while g diversity accumulated at a greater rate in the forest than in the pasture (bforest 17.2 species/1000 m, bpasture 9.3 species/1000 m; t 7.36, DF 9, p B 0.0001). In spite of the limited variation in the total number of tribes, genera and species between mountains, on the Cofre de Perote, 58% of species and 53% of the individuals caught were Neotropical. On this mountain the species belonging to genera with an Afrotropical affinity (Digitonthophagus ) were captured more often in pastures, while those of Holarctic affinity (Copris and Onthophagus ) were captured equally in the forest and in the pasture (Table 1). In contrast, in the northern Andes ca 92% of species and85% of the individuals belonged to genera of Neotropical affinity and these were caught more frequently in forested areas (Table 2). In these mountains, no genus was clearly dominant in the pastures and only two species (Anisocanthon villosus and Canthon sp. 1) were exclusive to areas used for cattle located at low altitudes (Appendix 1). The complete list of species by habitat along each altitudinal gradient is given in Appendix 1. Species richness reliability Visual comparison of the species accumulation curves for the forest and the pasture at each altitude indicate that at the Cofre de Perote most of the sites reached an asymptote (Fig. 1). Estimated species richness values (MM) indicate that a large proportion of the species (80%) present at each site were captured (Table 2). In contrast, for the two northern Andes gradients, regardless of altitude the species accumulation curves for the pastures rarely reached the asymptotic phase (Fig. 1), due in part to the low number of individuals captured and to the high dominance of a few species. In this environment, the percentage of species captured was 70% in only a few cases and the estimated species richness values were highly variable, ranging from 37 to 88% (Table 2). On the other hand, at the forest sites the species accumulation curves were clearly asymptotic (Fig. 1b, c), and the proportion of species captured at each site was 76 95% (Table 2). The analysis of the entire gradient by habitat type indicates that at the Cofre de Perote 90% of the species present in each type of habitat were captured. For the northern Andes, and similar to the results found at the sites level, the proportion of species captured was lower in the pasture than in the forest (Table 2). The analysis for each mountain shows that between 88% (Rı´o Cusiana) and 95% (Cofre de Perote) of all the species present on each gradient were captured (Table 2).


Table 1. Composition at the levels of tribe and genus: number of species and individuals (in parentheses) captured in the forest and pasture along each altitudinal gradient. Biogeographical affinity of each genus: AFRAfrotropical, HOLHolarctic, NEONeotropical. *Endemic to the Neotropical region. The observed species richness (Sobs9995% CI) was calculated using the Analytical formula proposed by Colwell et al. (2004); a indicates the proportion of species observed relative to Michaelis-Menten (MM) that was used as an estimate of expected richness. Tribe

Genus

Cofre de Perote Forest

Canthonini

Coprini Dichotomiini

Eurysternini Phanaeini

Onthophagini Sysiphini Total number of tribes Total number of genera Total number of species (Sobs995%CI) Michaelis-Menten (MM) Estimate(%)a Total number of individuals

Anisocanthon NEO* Canthon NEO Cryptocanthon NEO* Deltochilum NEO* Scybalocanthon NEO* Copris HOL Ateuchus NEO* Canthidium NEO* Dichotomius NEO* Ontherus NEO* Scatimus NEO* Uroxys NEO* Eurysternus NEO* Coprophanaeus NEO* Phanaeus NEO Oxysternon NEO* Sulcophanaeus NEO* Digitonthophagus AFR Onthophagus HOL Sisyphus AFR/HOL

Pasture

Chiles Volcano Both

Forest

Pasture

R覺織o Cusiana Both

7 (723)

6 (321)

7 (1044)

2 (3)

2 (120)

3 (123)

3 (287)

2 (36)

3 (323)

1 (8)

3 (89)

3 (97)

4 (285) 1 (164) 1 (4)

2 (5) 1 (23) 1 (2)

4 (290) 1 (187) 1 (6)

1 (7) 3 (77) 1 (27)

1 2 1 1

1 3 1 1 1 1 2 3

3 (29) 2 (6)

2 (4) 3 (202) 3 (259)

(31) (13) (54) (9)

1 (45) 2 (30) 2 (16)

11 (608) 1 (21) 6 11 3393.0 36.2 91.1 1849

1 (2) 1 (33) 2 (4) 1 (53) 10 (751) 6 12 3194.4 34.3 90.4 1396

(38) (90) (81) (9) (45) (2) (63) (20)

1 (53) 12 (1359) 1 (21) 7 14 4092.7 42.2 94.8 3245

2 (4) 3 (173) 3 (253) 6 4 1 1 2 2

(400) (29) (5) (20) (21) (86)

1 (4)

7 4 1 1 2 2

3 (72)

2 (12)

3 (84)

6 14 3592.8

6 11 2195.4

6 14 3792.7

38.2 91.6 1519

2 (5) 4 (19) 1 (2)

27.6 76.0 227

(405) (48) (7) (20) (21) (90)

40.5 91.3 1746

Forest 3 (197) 1 (46) 5 (100)

Pasture 1 (5) 1 (4) 1 (1)

1 4 1 5

(5) (201) (46) (101)

1 10 6 2

(1) (129) (210) (279)

5 5 1 3

(63) (179) (3) (30)

1 10 6 2

(1) (129) (193) (231)

3 (17) 2 (48)

5 4 1 2

(56) (153) (2) (26)

3 4 1 2

(7) (26) (1) (4)

1 (30) 2 (199) 5 13 4393.8 48.7 88.3 1363

Both

1 (30) 4 (42) 5 10 2293.9 27.9 78.8 155

4 (241) 5 14 4993.8 55.4 88.4 1519

199


Table 2. Observed and estimated species richness (forest/pasture) in each of the sites along each altitudinal gradient. *Denotes those sites where the estimates were lower than 80%. For the Chiles Volcano at 2600 m a.s.l., it was not possible to calculate the estimated value of species richness because abundance was52 individuals. Mountain/elevation (m a.s.l.)

Forest/pasture No. individuals

Cofre de Perote 50 450 900 1340 1860 2000 2340 3000 Chiles Volcano 50 520 1000 1350 1800 2600 3300 R覺織o Cusiana 450 900 1250 1500 1750 2000 2500

MM

Estimate(%)

306/226 721/125 389/250 102/236 57/47 44/186 216/240 14/68

1093.3/1192.5 1593.5/1094.5 1092.4/1494.0 1090.9/692.3 591.2/593.0 491.3/892.4 391.2/390.0 290.0/392.2

11.6/13.4 16.9/11.7 11.1/18.2 11.3/6.6 5.7/5.9 4.2/8.3 3.5/3.6 2.8/3.2

86.2/89.1 88.7/85.4 90.1/76.2* 88.5/90.9 87.7/84.7 95.2/96.4 85.7/83.3 71.4*/93.7

216/181 85/21 357/ 684/14 175/6 2/1 0/4

1791.5/992.4 1295.1/693.1 1291.8/ 1390.0/592.3 891.3/492.9

19.5/10.2 14.5/8.8 12.3/ 13.7/8.5 8.8/10.7

87.2/88.2 82.7/68.1* 97.5/ 94.9/58.8* 90.9/37.3*

13/17 338/33 280/57 187/20 309/20 53/  65/16

Variation in Alpha diversity The number of species decreased with increasing altitude on the three gradients (Fig. 2). However, the pattern of change in species richness at the local level (St and Sm) at each habitat type was different for each mountain system. On the Cofre de Perote values of St were similar for the forest and the pasture regardless of altitude (mean9SD: forest 7.6590.81, pasture  7.7590.80, p0.92; Table 3), and there were even some altitudes for which St was greater in the pasture (Fig. 2a). On this mountain, altitude explained 74% of the variation and habitat type explained 1%. In contrast, on the Chiles Volcano St values were always higher for the forest than for the pasture. (forest  9.090.61, pasture4.390.66, p0.001; Fig. 2b). On this mountain, altitude explained 62% of the variation and habitat type accounted for 17% (Table 3). Similarly, at R覺織o Cusiana, St was also consistently greater in forest than in pasture (forest 10.5891.4, pasture 5.891.53, p0.002; Fig. 2c). The model explained 48% of the total variation, much of which was associated with habitat type (40%), while altitude accounted for the remaining 8% (Table 3). The lack of fit of the model, particularly for the forest, is a result of the increase in St at intermediate altitudes. The analyses

200

Sobs.995% IC

0/190.0 1092.5/390.9 1691.3/692.4 1294.8/1195.1 1392.7/391.2 1492.7/391.2 591.2/ 591.3/490.9

/1.5 11.2/7.16 18.3/13.7 14.5/15.6 14.9/3.4 18.4/3.4 5.5/ 5.9/5.7

/66.6* 89.3/41.9* 87.4/43.8* 82.7/70.5* 87.2/88.2 76.1*/88.2 90.9/ 84.7/70.1*

did not detect significant differences in the three altitudinal gradients regarding the rate at which species are lost as altitude increases for either type of habitat (Table 3). The analysis using mean species richness (Sm) produced a pattern similar to that described for St (Table 3). Variation in Beta diversity Species turnover patterns between adjacent levels were different on each mountain according to habitat type (Fig. 3). On the Cofre de Perote for the forest the values of bt reached a maximum between 900 and 1340 m a.s.l. and then slowly decreased with increasing altitude (bt: 0.90.6). In this habitat the values of bt were influenced by the loss of species (Fig. 3a). In contrast, for pastures species turnover increased rapidly up to 1860 m a.s.l. (bt: 0.45 0.82) and stayed constant above 2000 m a.s.l. For the pasture of the Cofre de Perote, values of bt were the result of gaining species at certain altitudes (Fig. 3b). The comparison between forest and pasture habitats for each altitude shows that although species turnover between habitats tends to decrease with increasing altitude (bt: 0.52 0.2), the values of bt reflect the gain of species, particularly above 1800 m a.s.l. (Fig. 3c).


12

Table 3. Summary of the ANCOVA results. For St (total number of species per site) a Generalized linear model (GLM) was used with a Poisson error distribution (link functionLog). Deviance values (ca x2) are given as a measure of the model’s fit. Sm denotes the mean number of species per trap. In both cases, the fit model was YmHabitatAltitudeHabitatAltitudeo. * pB0.05*; pB0.01**; pB0.001***; ns not significant.

10

Factor

DF

St Deviance (x2 approx.)

Sm F

Cofre de Perote Habitat Altitude HabitatAltitude Error

1 1 1 12

0.03ns 26.73*** 0.29ns 8.76

1.69ns 17.39*** 1.0ns

Chiles Volcano Habitat Altitude HabitatAltitude Error

1 1 1 9

10.10*** 38.32*** 0.09ns 12.85

8.69** 9.65** 1.63ns

Rı´o Cusiana Habitat Altitude HabitatAltitude Error

1 1 1 9

12.53*** 2.16ns 0.75ns 15.75

16.40** 2.75ns 2.67ns

20

(a)

Forest: a = 13.3; b = -3.9 Pasture: a = 12.7; b = -3.3

18 16 14

8 6 4 2 0 0 20

500

1000

1500

2000

(b)

2500

3000

3500

Forest: a = 17.0; b = - 5.3 Pasture: a = 8.2; b = - 3.4

18

Number of species

16 14 12 10 8 6 4 2 0 0

20

500

1000

1500

2000

(c)

2500

3000

3500

Forest: a = 16.1; b = -3.7 Pasture: a = 5.9; b = -0.4

18 16 14 12 10 8 6 4 2 0 0

500

1000

1500

2000

2500

3000

Altitude (m a.s.l.)

Fig. 2. Variation in total species richness (St) in each type of habitat as elevation increases: (a) Cofre de Perote, (b) Chiles Volcano, (c) Rı´o Cusiana. The black dots represent forest sites and the open rectangles represent the pastures. Lines indicate the fitted curve (forest continuous line; pasture dotted line).

Species turnover between adjacent altitudes on the Chiles Volcano, both in the forest and the pasture, show a similar pattern, decreasing at intermediate

altitudes and then reaching maximum values at the upper end of the gradient, although for the pasture the increase was more gradual (bt forest: 0.28 1.0; bt pasture: 0.55 1.0). In both habitats the values of bt were affected by the constant loss of species with increasing altitude (Fig. 3d, e). Likewise, when species turnover was compared between the forest and the pasture at each altitude, above 1800 m a.s.l. the similarity between habitats was lower and the values of bt reflect the loss of species (Fig. 3f). As for t he Chiles Volcano, at Rı´o Cusiana, species turnover between adjacent altitudinal levels in both forest and pasture decreases from low to intermediate altitudes around 1250 m a.s.l., and then rapidly increases towards the top of the gradient the (bt forest:. 0.24 0.9; bt pasture: 0.52 1.0). On Rı´o Cusiana, in contrast to what we observed on the other altitudinal gradients, the values of bt for both the forest and the pasture reflect the gain of species at the beginning of the gradient between 450 and 1250 m a.s.l. Above this the values of bt reflect the loss of species (Fig. 3g, h). The comparison between the forest and the pasture at each altitude indicate a strong species turnover between habitats, especially above 1500 m a.s.l., as well as values of bt that are strongly influenced by the loss of species along the entire altitudinal gradient (Fig. 3i). Variation in abundance and dominance On comparing the distribution of species abundance between habitats, above 1340 m a.s.l. on the Cofre

201


Cofre de Perote

Chiles Volcano 1 0 .8

12

Río Cusiana

(d)

16

1 0.8

12

0 .6 8

8

0

Number of species

50

0 -45

0 45

12

8 0.4

4

0

0

0

0

520 000 350 600 800 300 500-1 0-1 0-1 0-2 0-3 52 135 100 260 180

1

16

0 .8

12

(e)

0.8

50

16

-45

0 45

0-9

8 0.4

0. 8

12

30

450

900 1340 1860 2000 2340 3000

0.4

0

0

20

0-1 52

350

0-1 135

800

000 600 0-3 0-2 260 180

(f)

16

1 0.8

12

0.2 0

0-9 45

00 900

50 -12

50 12

00 -15

50 00 -17 -25 00 50 15 17

(i)

16

1 0.8

12

0.6

0.6

8

8

0

8

0

0.4 4

0.8 0.6

4

0. 6 8

1

12

0.2

5 50-

1

(h)

16

4

00 000 -2340 -3000 340 86 0 0-1 40-1 60-2 40 00 90 20 13 23 18

(c)

0

0.6

0 .2 0

0.2

00 50 00 00 50 00 0-9 -12 -15 -25 -17 -20 45 50 00 50 00 900 12 20 15 17

1

0. 4

0

0.4

0.2

8 4

0.6

0. 2

0. 6

0.8

12

4

0 860 340 000 -2340 -3000 -90 0-1 40-1 60-2 00 40 90 13 18 20 23

(b)

16

1

0.6

0 .4 4

(g)

16

0.4

0.4

0.2

4

0

0

βt

(a)

16

0.2 0 50

0 52

0 100

0 135

0 180

0 260

0 330

4

0.2 0

0 450

900

50 12

00 15

17

50

00 20

00 25

Altitude (m a.s.l.)

Fig. 3. Changes in bt diversity (Wilson and Shmida index; black dots) between adjacent altitudes in the forest (a, d, g) and pasture (b, e, h) along each altitudinal gradient. The bottom figures (c, f, i) show species turnover between habitats (forest vs pasture) at each altitudinal level. The filled bars indicate the number of species lost and the open bars indicated the number of species gained in each comparison.

de Perote, abundance was greater in pastures (Table 2). High on this mountain, the areas used for cattle were dominated by very few species (Fig. 4), while between 50 and 900 m a.s.l. the distribution of abundances was more even and the number of individuals captured was greater in the forest than in the pasture. On this mountain, Simpson’s index was greater for pastures for some altitudes and the change in diversity between habitats was significantly different from a random distribution (Table 4). In contrast, for the northern Andes independent of altitude, abundance was always greater in the forest (Table 2). For both the Chiles Volcano and Rı´o Cusiana, the range-abundance curves for the forest show a more even distribution than the open areas used for cattle do; areas where the dominance of only a few species increased (Fig. 4). On these mountains, Simpson’s index was generally greater at forest or was similar between habitats and, the changes in diversity were not significantly different from a random distribution (Table 4).

202

Discussion On the three altitudinal gradients studied, we observed a decrease in species richness as altitude increased; this phenomenon has been reported for different taxonomic groups on different mountains (Rahbek 1995 and references therein). However, the lack of comparative studies between taxonomic groups, and between natural and anthropogenic environments hinders efforts to properly contrast patterns of diversity with altitude (Lomolino 2001). In the case of dung beetles, this study indicates that in the northern Andes the rate of species loss with increasing altitude was much more pronounced in the forest than in the pasture. While on the Cofre de Perote (MTZ) the decrease was very similar for both habitat types. In these pastures, the decrease in species richness was lower owing to the presence of a set of species of northern affinity (Paleoamerican Montane and Nearctic Patterns) or of those that evolved on the Mexican High Plain (Altiplano Distribution Pattern). The latter is


Cofre de Perote 1000

50 m

900 m

450 m

2340 m

1340 m 2000 m

100

3000 m

1860 m 10

1

Chiles Volcano

1000

Abundance (Log 10)

1350 m 50 m 100

1000 m*

1800 m

520 m 10

1

RĂ­o Cusiana

1000

900 m

1500 m

1250 m 100

1750 m 2500 m 2000 m*

450 m

10

1

Abundance Range

Fig. 4. Dominance-diversity curves comparing the distribution of abundance for forest (black dots) and pasture (open squares) at each elevation along each gradient. Altitudes marked with an asterisk * indicate sites where it was not possible to collect in pastures (1000 m a.s.l. on the Chiles Volcano and 2000 m a.s.l. on the RĹ´o Cusiana).

comprised of species with a heliophile habit that are of South American origin and were isolated in the MTZ a very long time ago (having migrated during the Oligocene-Miocene period, Halffter 1987). Consistent with the rate of species loss with increasing altitude, the opposite process  species accumulation or g diversity  in each type of habitat along each altitudinal gradient exhibited a similar pattern. For the Cofre de Perote, both habitats accumulated a similar number of species; while in the northern Andes species accumulation was

always lower for pasture than for the forest. As such, a first conclusion that can be drawn from these results is that the degree to which species richness decreases and the degree to which species are added in each type of habitat as altitude increases, indicate that the impact of cattle pasturing is different for each gradient, and results from the biogeographical differences between the mountain systems studied. At the local level, the pastures of the northern Andes always were less diverse than sites at equivalent altitudes

203


Table 4. Simpson’s index values (1/D) obtained for each type of habitat along each altitudinal gradient; d is the difference between forest and pasture in the Simpson’s index (see Data analysis). * Denotes elevations on the Chiles Volcano where it was not possible to calculate 1/D because abundance was52 individuals. Elevation (m a.s.l.)

Forest

Pasture

d

Number of times ½d½ simulated½d½ observed

p

Cofre de Perote 50 450 900 1340 1860 2000 2340 3000

2.13 6.07 2.94 6.00 3.03 2.71 1.28 1.57

5.41 3.90 6.29 1.44 1.60 4.27 1.94 1.06

3.28 2.17 3.35 4.56 1.43 1.56 0.66 0.51

0 2 0 0 42 10 0 215

B0.0001 0.0002 B0.0001 B0.0001 B0.0042 0.001 B0.0001 0.02

7.67 4.70 4.58 5.68 5.90

2.19 3.33  4.79 5.00

5.48 1.37  0.89 0.90

0 3091  5788 7897

B0.0001 0.31  0.57 0.78

4.18 2.43 5.30 2.39 2.32 2.78 2.44

3.5 4.16 4.47 1.38 3.00  2.85

0.68 1.73 0.83 1.01 0.68  0.41

199 7992 3894 1566 2378  5860

0.02 0.79 0.39 0.15 0.23  0.58

Chiles Volcano 50 520 1000 1350 1800 2600* 3300* Rı´o Cusiana 450 900 1250 1500 1750 2000 2500

on the Cofre de Perote (MTZ). On the Cofre de Perote differences were accentuated at lower altitudes and less conspicuous on the upper parts of the altitudinal gradient. The latter was particularly notable at altitudes over 1800 m a.s.l. owing to the influence of the previously mentioned Holarctic and High Plain elements. This agrees with findings for different mountain regions in Europe where dung beetle richness and abundance, especially in open habitats, are not negatively correlated with altitude (Mene´ndez and Gutie´rrez 1996, Romero-Alcaraz and A´vila 2000). In temperate climates, such as the upper part of the Cofre de Perote, Scarabaeinae are restricted to or dominate open environments. This is a general phenomenon characteristic of the northern hemisphere, both latitudinally and altitudinally (Martin-Piera et al. 1992), and one that does not occur or is very limited in the northern Andes. There, the regional set of species than is adapted to forest conditions has more species that the set that is adapted to open habitats. The study by Amat et al. (1997) shows that the forests of the Sabana de Bogota´, Eastern Range (2800 2900 m a.s.l.) has up to 11 species while the pastures only have three, and these belong to genera that are widely diversified in the lowland forests. Based on these results it is also possible to conclude that if the reduction of forest in the upper slopes of the northern Andes continues as a conse-

204

quence of creating cattle pastures, the beetle fauna will become isolated in small forest remnants, as has been documented for birds and amphibians in fragmented areas above 1500 m a.s.l. in the mountains of Colombia and Ecuador (Kattan and A´lvarez-Lo´pez 1996, Marsh and Pearman 1997). Therefore, the integrity of these communities depends to a large extent on the connectivity of the forests along the altitudinal gradient, and on alternative land uses, both of which can buffer the impacts of cattle ranching on biodiversity (Pineda et al. 2005). The comparison of fauna composition at the tribe and genus level for each mountain and between habitats reveals notable differences: 32.5% of the species and 43% of the individuals on the Cofre de Perote belong to Onthophagini, while in the northern Andes this tribe represents no more than 8% of the species and 15% of the individuals. The present day distribution of Onthophagus (Onthophagini) is the result of an ancient process of invasion of the Americas from Asia, followed by intense diversification in North America, including Mexico. With 2000 species described, this genus is considered one of the most modern of the dung beetles and is supposed to have diversified during the Oligocene (ca 23 33 million years BP), a diversification that coincided with the expansion of pastures and the spread of mammals (Davis et al. 2002). Currently, the


representatives of this genus are ubiquitous members of beetle communities in areas where the forest has been cut at different altitudes in both Mexico and Central America (Halffter et al. 1995, Horgan 2002). In South America, however, this genus is restricted to distinct habitat types below 2000 m a.s.l. with few species at higher altitude in the mountains (Zunino and Halffter 1997). The other member of tribe Onthophagini found on the Cofre de Perote, Digitonthophagus gazella , is a notable example of the modern expansion process on the American continent and could serve as a model for understanding how Paleo-American tropical lineages expanded on this continent. This Afrotropical species was introduced in Texas in 1972 and in little more than 30 yr has made its way down to southern Nicaragua (Montes de Oca and Halffter 1998). According to Halffter et al. (1995), D. gazella is found in open habitats and is markedly associated with cattle dung. Its dispersal has been favored by deforestation and by the change in the use of large tracts of land to cattle pasture. Consequently, the invasion of introduced species is indicative of possible habitat deterioration (Kennedy et al. 2002). The results of this study and others (Halffter et al. 1995, Arrellano and Halffter 2003) indicate that the pastures of the Cofre de Perote have served as altitudinal dispersal routes for heliophile and thermophile species such as D. gazelle , and Euonicitellus intermedius (another species introduced in North America) from tropical lowland landscapes. So, the expansion of areas used for pasturing cattle in the mountains appears to have facilitated the expansion of those species adapted to open environments previously present in the region. This may be contributing to the homogenization of the fauna along altitudinal gradients, as reported for amphibians, reptiles and birds in Costa Rica’s mountains where this process is favored by an increase in temperature at higher altitudes as a consequence of global climate change (Pounds et al. 1999). Although we do not have a definitive image of horizontal colonization by beetles in the northern Andes mountain ranges, studies of plants and birds allows us to illustrate its relevance; as one moves up these mountains, the proportion of genera originating outside the tropics increases (Vuilleumier 1986, Gentry 2001). However, for the dung beetles and butterflies, colonization is mainly vertical (Decimon 1986, Escobar et al. 2006). The fauna of the intermediate and high altitudes on these mountains is a derivative of the found in the neighboring lowlands. This also occurs in the mountains of southeast Asia (Hanski and Niemela¨ 1990), Ecuador (Celis et al. 2004) and in some of the mountains of Costa Rica (Halffter and Reyes-Castillo unpubl.). Therefore for historical-biogeographical reasons, in the northern Andes species turnover between

adjacent altitudes mainly results from the loss of species. This is a product of the process of vertical colonization and can be explained by the restrictions imposed by altitude in environmental terms (decreasing temperature) and the reduction in food availability; conditions that require physiological adjustments if the higher altitudes of the mountains are to be colonized (Chown et al. 2002). There is an important difference in the species turnover between the two Andean transects. On the Chiles Volcano (located on the western slope of the Eastern Range), species turnover was dominated by the loss of species, while on the Rı´o Cusiana transect (located on the eastern slope of the Eastern Range), there was a gain of species below 1250 m a.s.l.  the contact zone between the lowland fauna and the mountain forests. According to Lomolino (2001), the degree of overlap or juxtaposition between adjacent communities along altitudinal gradients contributes to explaining the type of relationship between species richness and altitude (monotonic model vs hump-shape model) and particularly, the degree of faunistic turnover between adjacent altitudinal bands. Precisely this was observed for five altitudes gradients between 08 and 78 North latitude on the Eastern cordillera of the Andes, Colombia (Escobar et al. 2005). Therefore, the differences in the species turnover patterns for opposite slopes in the northern Andes could result from the fact that their lowland dung beetle faunas differ in diversity and composition. This has been documented for the less diverse lowlands of the Pacific Plains on the western slope of the Andes (Peck and Forsyth 1982, Medina and Kattan 1996) and for the locations of the Amazonia-Orinoquia that are richest in species on the eastern slope of the Andes (Howden and Nealis 1975, Pulido et al. 2003). There appears to be an altitudinal gradient with respect to abundance, and it depends on habitat type. In general, abundance was much greater in forests than in pastures, while dominance increased in the pastures. However, the differences in abundance between the pastures of the different mountains above altitudes of 1750 m a.s.l. are marked. On the Cofre de Perote (MTZ) pastures were home to 39% of all the individuals collected, while in the northern Andes the proportion of individuals found in pastures was never higher than 10%. High total abundance values (biomass) have been recorded in the higher zones where species richness decreases in the mountains of Europe (Lumaret and Stiernet 1991) and in Asia (Hanski and Krikken 1991). In all of these studies, the dominance of a few species was found to increase in what seemed to be a compensating mechanism for adjusting populations to the available resources (Hanski and Cambefort 1991).

205


Although cattle dung can be abundant resource in the pastures of many tropical mountains, in this environment it dries out quickly and this modifies its microenvironmental and nutritional characteristics (Halffter 1991). Once changed, this dung can only be used by some species; species that are physiologically and behaviorally adapted to using the dung of medium sized and large herbivores in open areas. This is the case for many species of Holarctic (Onthophagus chevrolati , O. incensus) and Afrotropical (D. gazella and E. intermetus ) affinity in the MTZ or those of Neotropical affinity that exhibit wide ecological tolerance and are present in both mountain systems, such as Dichotomius colonicus , Ontherus mexicanus and Scatimus ovatus in the MTZ (Halffter et al. 1995, Arellano and Halffter 2003) and Dichotomius satanas , D achamas , Ontherus kirchii , O. brevicollis , Eurysternus marmoreus and E.caribaeus in the northern Andes (Amat et al. 1997, Escobar 2004). The limited presence of the Scarabaeinae in the pastures of the northern Andes highlights the following paradox: in spite of the abundance of food  cow dung  this environment is not available and effectively does nor exist for the majority of species that inhabit the native forest. This supports the concept that in Tropical America plant shade and its influence on microclimatic conditions on the ground are more important than a greater abundance of food (Halffter 1991).

Conclusion Although it was not possible to control factors such as the size of the mountains, the topography, the climatic variation, the configuration of the landscape and food supply, the results are revealing. The change in beetle community attributes that we observed in the face of an ecological change resulting from the transformation of forest to cattle pastures were clearly different along each altitudinal gradient. This suggests that processes of disturbance caused by human activity along altitudinal gradients can impact communities in different ways, depending on the geographic position of each mountain and particularly the biogeographical history of the group of species that inhabits it. This study contributes to the understanding that the response of communities to human activities (such as replacing forest with pastures, habitat loss and fragmentation) cannot be extracted from their regional context. Nor can they be understood without considering the biogeographical patterns and the evolutionary restrictions (e.g. habitat specialization) of species that belong to these communities (Ewer and Didham 2006).

Acknowledgements  We are grateful to the subject editor of Ecography for valuable remarks and critiques. We thank

206

Bianca Delfosse for translating the article into English and Ute Kryger for offering valuable suggestions to the last version. Research in Colombia was supported by the Financiera Ele´ctrica Nacional (FEN) and by the Inst. Colombiano para el Desarrollo de la Ciencia y la Tecnologı´a (COLCIENCIAS, project 2245-13-306-97). In Mexico, this research was financed by the Consejo Nacional de Ciencia y Tecnologia de Me´xico (CONACYT, project 37514-V), by the Comisio´n Nacional para el Uso y Conocimiento de la Biodiversidad (CONABIO, projects 093-01 and EE005) and by Ministerio del Medio Ambiente y Recursos Naturales y el Consejo Nacinal de Ciencia y Tecnologia de Me´xico (SEMARNAT-CONACyT, project 2004-56-A1). Finally, the first author appreciates support from Univ. of Pretoria (Postdoctoral Fellowship Programme) to allow writing the last version of this article.

References Amat, G. et al. 1997. Patrones de distribucio´n de escarabajos copro´fagos (Coleoptera: Scarabaeidae) en relictos del bosque altoandino, cordillera Oriental de Colombia.  Caldasia 19: 191 204. Anon. 2002. Statistical database for agriculture.  FAO, Rome. Arellano, L. and Halffter, G. 2003. Gamma diversity: derived from and a determinant of alpha diversity and beta diversity: an analysis of three tropical landscapes.  Acta Zool. Mex. (n.s.) 90: 27 76. Brown, J. H. 2001. Mammals on mountainsides: altitudinal patterns of biodiversity.  Global Ecol. Biogeogr. 10: 101 109. Celis, J. et al. 2004. Dung beetles (Coleoptera: Scarabaeinae) diversity in an altitudinal gradient in the Cucutu´ Range, Morona Santiago, Ecaudorian Amazon.  Lyonia 7: 37  52. Challenger, A. 1998. Utilizacio´n y conservacio´n de los ecosistemas terrestres de Me´xico: Pasado, presente y futuro.  CONABIO, Inst. de Biologı´a, UNAM y Sierra Madre, Me´xico. Chown, S. L. et al. 2002. Physiological variation in insects: large-scale patterns and their implications.  Biochem. Mol. Ecol. 531: 587 602. Colwell, R. K. 2005. EstimateS: statistical estimation of species richness and shared species from samples, v. 7.5.0.  Bhttp://viceroy.eeb.uconn.edu/estimates . Colwell, R. K. and Coddington, J. A. 1994. Estimating the extent of terrestrial biodiversity through extrapolation.  Phil. Trans. R. Soc. B 345: 101 118. Colwell, R. K. et al. 2004. Interpolating, extrapolating and comparing incidence-based species accumulation curves.  Ecology 85: 2717 2727. Crawley, M. J. 2002. Statistical computing.  Wiley. Davis, A. L. V. et al. 1999. Species turnover, community boundaries, and biogeographical composition of dung beetle assemblages across an altitudinal gradient in South Africa.  J. Biogeogr. 26: 1039 1055. Davis, A. L. V. et al. 2002. Historical biogeography of Scarabaeinae dung beetles.  J. Biogeogr. 29: 1217  1256.


Decimon, H. 1986. Origins of lepidopteran faunas in the high tropical Andes.  In: Vuilleumier, F. and Monasterio, M. (eds), High altitude tropical biogeography. Oxford Univ. Press, pp. 500 533. Escobar, F. 2004. Diversity and composition of dung beetles (Scarabaeinae) assemblages in a heterogeneous Andean landscape.  Trop. Zool. 17: 123 136. Escobar, F. et al. 2005. Altitudinal variation of dung beetle assemblages in the Colombian Andes.  Global Ecol. Biogeogr. 14: 337 347. Escobar, F. et al. 2006. Assessing the origin of Neotropical mountains dung beetle assemblages (Scarabaeidae: Scarabaeinae): the comparative influence of vertical and horizontal colonization.  J. Biogeogr. 33: 1793 1803. Errouissi, F. et al. 2004. Composition and structure of dung beetle (Coleoptera: Aphodiidae, Geotrupidae, Scarabaeidae) assemblages in mountain grasslands of the southern Alps.  Ann. Entomol. Soc. Am. 97: 710 209. Ewer, R. M. and Didham, R. 2006. Confounding factors in the detection of species responses to habitat fragmentation.  Biol. Rev. Camb. Philos. Soc. 81: 117 142. Ferrusquia-Villafranca, I. 1993. Geology of Mexico: a synopsis.  In: Ramamoorthy, T. P. et al. (eds), Biological diversity of Mexico: origins and distribution. Oxford Univ. Press, pp. 3 107. Gentry, A. H. 2001. Patrones de diversidad y composicio´n floristica en los bosques de las montan˜as tropicales.  In: Kappelle, K. and Brown, A. D. (eds), Bosques Nublados del Neotro´pico. Inst. Nacional de la Biodiversidad, pp. 63 84. Halffter, G. 1987. Biogeography of the mountain entomofauna of Mexico and Central America.  Annu. Rev. Entomol. 32: 95 114. Halffter, G. 1991. Historical and ecological factors determining the geographical distribution of beetles (Coleoptera: Scarabaeidae: Scarabaeinae).  Folia Entomol. Mex. 82: 195 238. Halffter, G. and Matthews, E. G. 1966. Natural history of dung beetles of the subfamily Scarabaeinae (Coleoptera: Scarabaeidae).  Folia Entomol. Mex. 12 14: 195  238. Halffter, G. et al. 1995. Spatial distribution of three groups of Coleoptera along an altitudinal transect in Mexican Transition Zone and its biogeographical implications.  Elytron 9: 151 185. Hanski, I. 1983. Distributional ecology and abundance of dung and carrion-feeding beetles (Scarabaeidae) in tropical rain forest in Sarawak, Borneo.  Actas Zool. Fenn. 167: 1 45. Hanski, I. and Niemela¨, J. 1990. Elevational distributions of dung and carrion beetles in north Sulawesi.  In: Knight, W. J. and Holloway, J. D. (eds), Insects and the rain forest of southeast Asia (Wallacea). Roy. Entomol. Soc., pp. 145 152. Hanski, I. and Cambefort, Y. 1991. Dung beetle ecology.  Princeton Univ. Press. Hanski, I. and Krikken, J. 1991. Dung beetles in tropical forest in southeast Asia.  In: Hanski, I. and Cambefort, Y. (eds), Dung beetles ecology. Princeton Univ. Press, pp. 179 197.

Henderson, P. A. and Seaby, R. M. 2002. Species diversity and richness v. 3.0.  Pisces Conservational, Pennington. Horgan, F. G. 2002. Shady field boundaries and the colonization of dung by coprophagous beetles in central American pastures.  Agricult. Ecosyst. Environ. 91: 25  36. Howden, H. F. and Nealis, V. 1975. Effects of deforestation clearing in a tropical rain forest on the composition of the coprophagous scarab beetle fauna (Coleoptera).  Biotropica 7: 77 83. Huston, M. A. 1994. Biological diversity.  Cambridge Univ. Press. Janzen, D. H. 1967. Why mountain passes are higher in the tropics.  Am. Nat. 101: 233 249. Kappelle, K. and Brown, A. D. 2001. Bosques Nublados del Neotro´pico.  Inst. Nacional de la Biodiversidad-INBio, Santo Domingo de Heredia, Costa Rica. Kattan, G. and .A´lvarez-Lo´pez, H. 1996. Preservation and management of biodiversity in fragmented landscape in the Colombian Andes.  In: Schelhas, J. and Greenberg, R. (eds), Forest patches in tropical landscapes. Island Press, pp. 3 18. Kennedy, T. A. et al. 2002. Biodiversity as a barrier to ecological invasion.  Nature 417: 636 638. Lobo, J. M. and Halffter, G. 2000. Biogeographical and ecological factors affecting altitudinal variation of mountainous communities of coprophagous beetles (Coleptera, Scarabaeoidea): a comparative study.  Ann. Entomol. Soc. Am. 93: 115 126. Lomolino, M. V. 2001. Elevation gradients of species-density: historical and prospective views.  Global Ecol. Biogegr. 10: 3 13. Lomolino, M. V. and Perault, D. R. 2004. Geographic gradients of deforestation and mammalian communities in a fragmented, temperate rain forest landscape.  Global Ecol. Biogeogr. 13: 55 64. Lumaret, J. P. and Stiernet, N. 1991. Montane dung beetles.  In: Hanski, I. and Cambefort, Y. (eds), Dung beetles ecology. Princeton Univ. Press, pp. 242 254. Magurran, A. E. 1988. Ecological diversity and its measurement.  Princeton Univ. Press. Marsh, D. M. and Pearman, P. B. 1997. Effect of habitat fragmentation on the abundance of two species of Leptodactylid frogs in an Andean montane forest.  Conserv. Biol. 11: 1323 1328. Martin-Piera, F. et al. 1992. Ecology and biogeography of dung-beetle communities (Coleoptera, Scarabaeoidea) in an Iberian mountain range.  J. Biogeogr. 19: 677 691. Medina, C. A. and Kattan, G. H. 1996. Diversidad de coleo´pteros copro´fagos (Scarabaeidae) de la reserva forestal de Escalerete.  Cespedesia 21: 89 102. Mene´ndez, R. and Gutie´rrez, D. 1996. Altitudinal effects on habitat selection of dung beetles (Scarabaeoidea: Aphodiidae) in the northern Iberian Peninsula.  Ecography 19: 331 317. Montes de Oca, E. and Halffter, G. 1998. Invasion of Mexico by two dung beetles previously introduced into the United States.  Stud. Neotrop. Fauna Environ. 33: 37 45. Murgueitio, E. 2003. Environmental impact of milk production systems in Colombia and alternative solutions.  Livestock

207


Research for Rural Development 15: 1010, Bhttp:// www.cipay.org.co/Irrd/Irrd15/10/Murg1510.htm. Nichols, E. et al. in press. Dung beetle response to tropical forest modification and fragmentation: a quantitative literature review and meta-analysis.  Biol. Conserv. Peck, S. B. and Forsyth, A. 1982. Composition, structure and comparative behaviour in a guild of Ecuadorian rain forest dung beetles (Coleoptera, Scarabaeidae).  Can. J. Zool. 60: 1624 1634. Pineda, E. et al. 2005. Biodiversity in cloud forest and shade coffee: analysis of three indicator groups.  Conserv. Biol. 19: 400 410. Pounds, J. A. et al. 1999. Biological response to climate changes on a tropical mountain.  Nature 398: 611 615. Pulido, L. A. et al. 2003. Escarabajos copro´fagos (Coleoptera: Scarabaeidae: Scarabaeinae) del parque nacional natural Sierra del Chiribiquete, Caqueta´, Colombia.  In: Onore, G. et al. (eds), Escarabeidos de Latinoame´rica: estado actual del conocimiento. Monografias Tercer Milenio, Vol. 3. Sociedad Entomolo´gica Aragonesa, pp. 51 58. Rahbek, C. 1995. The altitudinal gradient of species richness: a uniform pattern?  Ecography 18: 200 205. Romero-Alcaraz, E. and .A´vila, J. M. 2000. Effect of altitude and type of habitat on the abundance and diversity of Scarabaeoid dung beetles (Scarabaeoidea) assemblages in a

Download the appendix as file E4818 from Bwww.oikos.ekol.lu.se/appendix.

208

Mediterranean area southern Iberian Peninsula.  Zool. Stud. 39: 351 359. Solow, A. R. 1993. A simple test for change in community structure.  J. Anim. Ecol. 62: 191 193. van der Hammen, T. 1995. Global change, biodiversity, and conservation of Neotropical montane forest.  In: Churchill, S. P. et al. (eds), Biodiversity and conservation of Neotropicsl montane forests. Proc. of the Neotropicsl Montane Forest Biodiversity and Conservation Symp. New York Botanical Garden, pp. 603 607. van der Hammen, T. and Hooghiemstra, H. 2001. Historia y paeloecologia de los bosques montanos andinos tropicales.  In: Kappelle, K. and Brown, A. D. (eds), Bosques Nublados del Neotro´pico. Inst. Nacional de la Biodiversidad INBio, pp. 63 86. Vuilleumier, F. 1986. Origins of the tropical avifauna in the high Andes.  In: Vuilleumier, F. and Monasterio, M. (eds), High altitude tropical biogeography. Oxford Univ. Press, pp. 586 622. Wilson, M. V. and Shmida, A. 1984. Measuring beta diversity with presence-absence data.  J. Ecol. 72: 1055 1064. Zar, J. H. 1996. Biostatistical analysis, 2nd ed.  Prentice Hall. Zunino, M. and Halffter, G. 1997. Sobre Onthophagus Latreille, 1802 americanos.  Elytron 11: 157 168.


ECOGRAPHY 29: 919 927, 2006

Elevational patterns of frog species richness and endemic richness in the Hengduan Mountains, China: geometric constraints, area and climate effects Cuizhang Fu, Xia Hua, Jun Li, Zheng Chang, Zhichao Pu and Jiakuan Chen

Fu, C., Hua, X., Li, J., Chang, Z., Pu, Z. and Chen, J. 2006. Elevational patterns of frog species richness and endemic richness in the Hengduan Mountains, China: geometric contraints, area and climate effects.  Ecography 29: 919 927. We studied frog biodiversity along an elevational gradient in the Hengduan Mountains, China. Endemic and non-endemic elevational diversity patterns were examined individually. Competing hypotheses were also tested for these patterns. Species richness of total frogs, endemics and non-endemics peaked at mid-elevations. The peak in endemic species richness was at higher elevations than the maxima of total species richness. Endemic species richness followed the mid-domain model predictions, and showed a nonlinear relationship with temperature. Water and energy were the most important variables in explaining elevational patterns of non-endemic species richness. A suite of interacting climatic and geometric factors best explained total species richness patterns along the elevational gradient. We suggest that the mid-domain effect was an important factor to explain elevational richness patterns, especially in regions with high endemism. C. Fu (czfu@fudan.edu.cn), X. Hua, J. Li, Z. Chang, Z. Pu and J. Chen, Ministry of Education Key Laboratory for Biodiversity Science and Ecological Engineering, and Inst. of Biodiversity Science, Fudan Univ., Shanghai 200433, China.

Understanding variation in species richness has been a central aim of community ecology for decades. Although the latitudinal gradient in species richness is one of the most studied diversity patterns, research on montane biotas has continued to play a prominent role in understanding the distribution of organisms (Lomolino et al. 2006). Rahbek (1995, 2005) recognized three basic types of elevational diversity patterns, i.e. species richness decrease with elevation, peak at low elevation plateaus, and peak at mid-elevation. Area, climate and geometric constraints are the most frequently cited explanation for both a linear and a humped relationships between species richness and elevation (Rahbek 1995, 1997, Brown 2001, Lomolino 2001, McCain 2004, 2005, 2006, Bhattarai et al. 2004, Cardelu´s et al. 2006, Fu et al. 2006, Kluge et al. 2006).

The elevational gradient of species richness may be intricately related to species-area relationships (Lomolino 2001). Some studies found that area of elevational belts explained a large proportion of the variation in species richness (Rahbek 1997, Kattan and Franco 2004, Bachman et al. 2004, Fu et al. 2004). Elevational patterns of diversity were also commonly explained by water and energy (Bhattarai et al. 2004, McCain 2006, Fu et al. 2006). Recent studies demonstrated that the mid-domain effect (MDE) is also a powerful explanatory variable in elevational diversity patterns (McCain 2004, Cardelu´s et al. 2006, Kluge et al. 2006, Watkins et al. 2006). The mid-domain effect arises from geometric constraints on species ranges within a bounded domain (Colwell and Lees 2000, Colwell et al. 2004). Based on chance alone, the likelihood of the

Accepted 5 October 2006 Copyright # ECOGRAPHY 2006 ISSN 0906-7590 ECOGRAPHY 29:6 (2006)

919


elevational ranges of many species overlapping is higher at mid-elevations than for the lower and higher elevations (Colwell et al. 2004, Cardelu´s et al. 2006). MDE theory predicts that endemic species will show more of MDE peak than non-endemics (Lee et al. 1999, McCain 2004, Colwell et al. 2004) and that small-ranged species will show less of an MDE peak than large-ranged species (Lee et al. 1999, Cardelu´s et al. 2006, Dunn et al. 2006, Kluge et al. 2006, Watkins et al. 2006). Biogeographical variation in species richness and endemic richness is critical to our understanding and conservation of biological diversity (Grytnes and Vetaas 2002). Areas with high species richness may also have a large number of endemic species, but patterns in richness and endemism are not necessarily positively related (Whittaker et al. 2001). Recent studies have found that species richness patterns across elevational gradients for total species and endemic species were different for plants (Kessler 2002, Grytnes and Vetaas 2002) and freshwater fishes (Fu et al. 2004). Geographic range size affects determinants of geographic patterns in species richness (Jetz and Rahbek 2002, Mora and Robertson 2005). Recent studies have found that there were different patterns and processes for elevational diversity of large-ranged and small-ranged species (Cardelu´s et al. 2006, Kluge et al. 2006). In this study, we delineated elevational patterns of frog species richness and endemic richness in the Hengduan Mountains of China, and assessed the ability of area, climate, and geometric constraints to explain the elevational diversity patterns.

Methods Study area The Hengduan Mountains (238 338N, 978 1038E) of China occur within an ecotone between the Oriental (Indo-Malayan) region and the Palearctic region. They span east of Tibet, west of Sichuan province and northwest of Yunnan Province, and are a part of the Qinghai-Tibet Plateau (Fig. 1). The total area of the mountains is ca 0.56 million square kilometers, and elevation range from ca 200 to 7300 m a.s.l. based on a global digital elevation model (DEM) from B/http:// lpdaac.usgs.gov/gtopo30/gtopo30.asp /. The altitude of this area declines from northwest to southeast (Fig. 1). Most parts of the area are characterized by a series of paralleled mountain ranges and rivers from south to north, with a sharp altitudinal differentiation.

Richness data We compiled frog distribution and altitudinal data from secondary sources (Appendix 1). Species checklists and 920

the altitudinal limits of occurrences were listed in Appendix 2. Species accumulation curves and species richness estimates showed that sampling was adequate (see the details in Appendix 3). One species, Paa feae is excluded from the analyses because of lacking information on their elevational distribution. Endemic species are defined according to their distribution being limited to the Hengduan Mountains.

Area data To calculate the area at each elevational band in the Hengduan Mountains, we used a global digital elevation model (DEM), GTOPO30, with a horizontal grid spacing of 30 arc-seconds (ca 1 km2) from B/http:// lpdaac.usgs.gov/gtopo30/gtopo30.asp/. First, we extracted the map, which contains elevational information of Hengduan Moutains, from the global GTOPO30 map. This map was converted to Lambert-Azimuthal equal area projection map, and rasterized at 1 /1 km grid cells. Finally, we counted the number of grid cells (1 km2/grid cell) within each 200-m elevational interval based on the elevational value of each grid cell, and summed up to the area. The relationship between area and elevation is shown in Appendix 4.

Climatic variables In this study, two climatic variables, annual mean temperature and annual mean precipitation, were included, selected because they have been shown to be important correlates of broad-scale richness gradients. Temperature and precipitation data were obtained from 104 climate stations with a record length of 30 yr (1951 1980) from Climate Resource Database (B/http:// www.data.ac.cn/zrzy/g03.asp /) and Meteorological Administration of local or central goverment in China. Six 100-m elevation intervals lack climate stations, including those at 600 700, 800 900, 2400 2500, 2700 2800, 3500 3600, and 4000 4100 m. In these 100-m elevation zones, temperature and precipitation data were interpolated from the mean value of the nearest adjacent upper and lower climatic station’s record. The relationships between climatic variables and elevation are shown in Appendix 4.

Geometric constraints: the mid-domain effect To test the mid-domain effect (MDE), diversity patterns were compared to null model predictions using a novel, discrete MDE model that does not necessitate the use of interpolated ranges (Dunn et al. 2006). The simulation program was implemented in RangeModel software, ver. 5 (Colwell 2006). For each data set, expected mean ECOGRAPHY 29:6 (2006)


Fig. 1. The sketch map of the Hengduan Mountains in China.

richness and its 95% confidence interval over the domain based on 50 000 simulations sampled were used to assess the impact of spatial constraints on the elevational diversity gradients.

Data analysis We divided the range of elevation into 200-m bands between 400 and 5000 m, and calculated the total number of species in each band to examine the relationship between species richness and elevation. Following the methods of Rahbek (1997), we tested four models of the relationships between species richness (S) and area (A), i.e. S/A, S/log A, log S/A, and log S/log A, and selected the best fit S/A model based on comparisons of r values. To account for multicollinearity, we used a combination of multi-regression techniques recommended by Graham (2003). These include single and multiple ordinary least squares models (OLS models), single and multiple conditional autoregressive models (CAR models), and generalized linear models (GLM models). Recent studies have advocated using the CAR models to identify the predictive power of the hypotheses for explaining geographic richness pattern of different taxa (Jetz and Rahbek 2002, Mora and Robertson 2005). ECOGRAPHY 29:6 (2006)

The GLM models have also been used to relate species richness to explanatory variables along the elevational gradients (Bhattarai et al. 2004, Kluge et al. 2006). GLM models were performed with S-Plus 7.0. OLS and CAR models were performed in SAM 1.0 (Rangel et al. 2006). We used Akaikeâ&#x20AC;&#x2122;s information criteria (AIC) to compare the fit of the OLS and CAR models, and smaller AIC values indicated a better fit. We repeated all analyses for large and small ranges, separately, to compare the role of candidate predictors for range size classes (following Jetz and Rahbek 2002). We divided each dataset (the overall-species set and the taxon subsets of endemics and non-endemics) into the 50% of species with large ranges and the 50% of species with the small ranges. We used STATISTICA (ver. 6.0) for the graph representation.

Results Frog fauna Our synthesis found 94 frog species from the Hengduan Mountains (Appendix 2), distributed among 7 families and 29 genera. Among these, there are 37 species endemic to the mountains. The most species-rich 921


families are the Megophryidae (8 genera and 28 species), the Ranidae (10 genera and 35 species), and the Rhacophoridae (4 genera and 12 species). The frogs are distributed between 400 and 5000 m a.s.l.

Elevational diversity patterns Overall species of total, endemic and non-endemic frogs showed a humped relationship between species richness and elevation (Fig. 2). Peak in endemic overall species richness (2400 2600 m elevational zones, Fig. 2b) was at higher elevations than the maxima of non-endemic or total overall species richness (1200 1400 m elevational zones, Fig. 2a, c).

Partitioning species into range-size categories highlighted disparate contributions to overall species richness patterns (Fig. 2). There were higher correlation coefficients between large-ranged species richness and overall species richness (total, r2 /0.96; endemics, r2 / 0.97; non-endemics, r2 /0.96) than between smallranged species richness and overall species richness (total, r2 /0.86; endemics, r2 /0.90; non-endemics, r2 /0.95).

Species richness patterns and explained variables Total species richness was strongly correlated with temperate and precipitation across the data sets of overall, large-ranged and small-ranged species (Table 1,

Fig. 2. Species richness patterns along elevational gradients (black circles and lines) in the Hengduan Mountains including 95% simulation limits (lines only) of the mid-domain analyses from 50 000 simulations samples for overall species and separated for large-ranged and small-ranged species of (a) total, (b) endemic, and (c) non-endemic frogs in the Hengduan Mountains.

922

ECOGRAPHY 29:6 (2006)


Table 1. Relationships between species richness of total, endemic and non-endemic frogs and explained variables using simple ordinary least squares (OLS) and simple conditional autoregressive (CAR) models for overall species and separated for large-ranged and small-ranged species in the Hengduan Mountains. Variables

Total

Overall species

Large-ranged species

Small-ranged species

Endemics

Overall species

Large-ranged species

Small-ranged species

Non-endemics

Overall species

Large-ranged species

Small-ranged species

MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT MDE Area MAP MAT

OLS models

CAR models 2

t

AIC

R

t

AIC

R2

2.33* /0.97 6.53** 5.50** 3.05** /0.51 4.82** 4.42** 1.15 /1.84 10.60** 6.95** 8.73** /0.01 0.63 0.90 10.44** /0.27 0.68 0.90 3.90** 0.50 0.49 0.82 0.42 /1.19 11.69** 7.16** 1.36 /0.72 7.70** 6.19** /0.40 /1.66 11.54** 6.32**

137 141 117 122 117 125 108 110 92 90 51 66 66 101 100 100 41 83 82 82 40 52 52 52 133 132 87 105 105 106 76 83 99 96 53 74

0.21 0.04 0.66 0.59 0.31 0.01 0.53 0.48 0.06 0.14 0.84 0.70 0.78 B/0.01 0.02 0.04 0.84 B/0.01 0.02 0.04 0.42 0.01 0.01 0.03 B/0.01 0.06 0.87 0.71 0.08 0.02 0.74 0.65 B/0.01 0.12 0.86 0.66

2.56* /0.39 5.59** 4.74** 3.30** 0.07 4.11** 3.86** 1.28 /1.33 8.45** 5.55** 7.90** /0.06 0.28 0.76 9.53** /0.43 0.37 0.76 3.23* 0.59 0.15 0.71 0.39 /0.43 9.70** 5.88** 1.42 0.06 6.41** 5.25** /0.4 /0.96 10.00** 5.06**

127 144 134 134 102 127 122 121 89 96 65 76 45 102 102 103 11 84 85 85 41 55 56 56 133 135 102 112 100 109 94 94 102 100 38 75

0.61 0.15 0.46 0.46 0.72 0.15 0.33 0.35 0.35 0.15 0.77 0.64 0.93 0.19 0.17 0.15 0.97 0.19 0.16 0.14 0.52 0.13 0.11 0.09 0.24 0.17 0.79 0.69 0.43 0.17 0.56 0.55 0.13 0.18 0.95 0.73

Model fit was assessed using the Akaike information criteria (AIC), smaller values indicate a better fit. *p B/0.05; **p B/0.01. MDE, the mid-domain effect; MAP, mean annual precipitation; MAT, mean annual temperature.

Appendix 5). The best fit multiple models also identified MDE as a strong predictor of overall and large-ranged species, whereas the explanatory power of MDE was low for small-ranged species (Table 2, Fig. 2a). Area was the weakest predictors of total richness (Table 1 and 2, Appendix 5). Endemic richness was strongly correlated only with MDE, whereas all other variables showed weak linear correlations across the data sets of overall, large-ranged and small-ranged species (Table 1, Fig. 2b). The explanatory power of MDE was stronger for largeranged species than for small-ranged species (Table 1 and 2). Endemic richness also showed a nonlinear relationship with temperature (Appendix 5). Non-endemic species richness was strongly correlated with temperate and precipitation, whereas MDE and area were weak predictors of non-endemic species richness across the data sets of overall, large-ranged and small-ranged species (Table 1 and 2, Fig. 2c, Appendix 5). ECOGRAPHY 29:6 (2006)

Discussion Elevational diversity patterns Amphibians and reptiles generally show a monotonic decline in species richness with increasing altitude (Brown and Alcala 1961, Duellman 1988, Fauth et al. 1989, Nathan and Werner 1999), although a humped relationships between species richness and elevation have been observed in particular habitats (Heyer 1967, Fischer and Lindenmayer 2005, Fu et al. 2006). However, growing evidence suggested that mid-elevational peaks in species richness for a wide variety of taxa are perhaps more general (Rahbek 1995, 2005, Lomolino 2001). This study showed a diversity peak at midelevations for frog species richness in the Hengduan Mountains. Other studies from the adjacent regions of the Hengduan Mountains, plant diversity in the Nepal Himalaya and the Indian Western Himalaya, and small mammal diversity in the Mt. Qilian also reported similar elevational diversity patterns (Grytnes and Vetaas 2002, 923


924 Table 2. Multiple ordinary least squares (OLS) and multiple conditional autoregressive (CAR) regressions for explained variables and species richness of total frogs, endemics and nonendemics including overall species and separated for large-ranged and small-ranged species in the Hengduan Mountains. Model A included all explained variables, model B all variable except the mid-domain effect (MDE). Magnitudes of t -values indicate variable importance in the models. Overall species

Total

Endemics

Non-endemics

MDE Area MAP MAT AIC Model fit MDE Area MAP MAT AIC Model fit MDE Area MAP MAT AIC Model fit

t t t t R2 t t t t R2 t t t t R2

Large-ranged species

OLS models

CAR models

OLS models

A

B

A

B

A

4.9** 0.6 2.9** 0.3 105 0.87 10.1** /1.0 /1.8 2.0 65 0.86 0.8 1.3 4.7** /0.5 91 0.89

/ 1.3 1.8 0.6 121 0.70  0.7 /0.8 1.1 105 0.08  1.6 4.7** /0.4 88 0.89

4.9** 0.7 3.0** 0.8 98 0.93 9.4** /1.1 /1.9 2.1 34 0.97 0.8 1.4 4.8** 0.001 100 0.89

/ 1.2 1.9 0.9 139 0.51  0.5 /0.9 1.2 109 0.20  1.5 4.7** 0.1 104 0.84

5.7** 1.4 2.1 0.8 91 0.86 12.9** /2.1* /0.8 1.1 38 0.91 2.8* 2.0 2.9** 0.6 72 0.86

B  1.7 0.9 1.0 111 0.59  0.4 /0.6 0.8 88 0.06  2.3* 2.1* 0.9 77 0.80

Small-ranged species

CAR models

OLS models

CAR models

A

B

A

B

A

B

 1.7 0.9 1.4 126 0.42  0.03 /0.6 0.9 92 0.18  2.4* 2.2* 1.5 95 0.67

2.2* /1.2 4.5** /1.1 54 0.88 4.3** 0.2 /2.4* 2.5* 42 0.59 2.4* 0.1 7.6** /2.8* 49 0.93

 /0.4 4.3** /1.0 56 0.85  1.4 /1.2 1.6 55 0.15  /0.6 6.7** /2.6* 52 0.90

2.1 /1.3 4.4** /0.6 60 0.89 3.9** 0.3 /2.6* 2.7* 41 0.71 2.0 0.05 7.3** /2.2* 62 0.91

 /0.6 4.3** /0.7 74 0.76  1.4 /1.5 1.8 60 0.21  /0.5 6.7** /2.0 43 0.95

5.8** 1.6 2.1 1.3 78 0.94 11.9** /2.3* /0.8 1.1 /5 0.98 2.8* 2.3* 2.9* 1.2 72 0.90

Model fit was assessed using the Akaike information criteria (AIC), smaller values indicate a better fit. *p B/0.05; **p B/0.01. Abbreviations expressed as in Table 1.

ECOGRAPHY 29:6 (2006)


Li et al. 2003, Bhattarai et al. 2004, Oommen and Shanker 2005). More studies on local and region diversity of amphibians and reptiles are needed to make generalizations on elevational richness patterns of these groups at global scale. In the Hengduan Mountains, species richness patterns across the elevational gradient for total frogs and endemics were different, and the maxima in diversity for endemic frogs peaked at higher elevations. Similar patterns were also reported in plants (Kessler 2002, Grytnes and Vetaas 2002) and freshwater fishes (Fu et al. 2004). Kessler (2002) thought that reduced surface area and more divided topography lead to more isolated populations and hence higher speciation rates with increasing elevation. Increased isolation and reduced dispersal might have resulted in increased differentiation and higher endemism with increasing elevation (Brown 2001). Recent studies reported that small-ranged and largeranged species showed markedly different species richness patterns, and geographic patterns in species richness were mainly based on large-ranged species because their larger number of distribution records had a disproportionate contribution to the species richness counts (Lee et al. 1999, Jetz and Rahbek 2002, Lennon et al. 2004, Cardelu´s et al. 2006, Dunn et al. 2006, Kluge et al. 2006, Kreft et al. 2006, Watkins et al. 2006). Va´zquez and Gaston (2004) further found that this pattern was most clearly demonstrated by endemic species. In this study, we also found that large-ranged species contributed more to overall richness pattern than small-ranged species in the data sets of total species and endemics. However, there were equal contributions for largeranged and small-ranged non-endemics to overall nonendemic richness. Below we discuss how area, climate and MDE may influence elevational patterns of frog species richness in the Hengduan Mountains.

Area effects Area may be a crucial parameter determining elevational diversity patterns because area generally declines with increasing elevation (Rahbek 1997, Lomolino 2001). The influence of area in determining regional species richness in altitudinal ranges has been shown for different taxa (Kattan and Franco 2004, Bachman et al. 2004, Fu et al. 2004). In this study, we found that area as a single predictor was less important to explaining richness patterns along an elevational gradient. The reasons for this may be from the special relationships between area and elevation. Area showed a bimodal relationship with elevation in the Hengduan Mountains (Appendix 4). ECOGRAPHY 29:6 (2006)

The mid-domain effect Many studies reported that the mid-domain effect (MDE) was an important variable to explain species richness patterns along elevational gradients (McCain 2004, Bachman et al. 2004, Oommen and Shanker 2005, Cardelu´s et al. 2006, Kluge et al. 2006, Watkins et al. 2006). The frog datasets in the Hengduan Moutains also showed that MDE accounted for a significant proportion of elevational patterns of total species richness. Endemic frog richness in the Hengduan Moutains was strongly correlated with MDE, whereas MDE was the weak predictor of non-endemic frog richness. It confirmed that MDE theory predicts that endemic species will show more of MDE peak than non-endemics (Lee et al. 1999, McCain 2004, Colwell et al. 2004). In this study, we found that the explanatory power of MDE was stronger for large-ranged species than for small-ranged species in the datasets of total frogs and endemics. It confirmed that MDE theory predicts that small-ranged species will show less of an MDE peak than large-ranged species (Lee et al. 1999, Jetz and Rahbek 2002, Colwell et al. 2004, 2005, Mora and Robertson 2005, Cardelu´s et al. 2006, Dunn et al. 2006, Kluge et al. 2006, Watkins et al. 2006).

Climatic effects A few studies have shown that water inputs played a prominent role in richness patterns of amphibian in Iberia (Schall and Pianka 1977) and frogs in Australia and the United States (Schall and Pianka 1978), and others found that energy input constrainted amphibian richness in North America (Currie 1991, Allen et al. 2002). And a recent study found that water-energy dynamics determined amphibian richness pattern in Europe (Rodrı´guez et al. 2005). Lomolino (2001) hypothesized that many components of climate and local environments varied along the elevational gradients and ultimately created the variation in species richness. In this study, climatic variables, annual mean precipitation and annual mean temperature were also found to be important predictors for frog species richness when total species and non-endemic species were considered. Other studies also observed that water and temperature were correlated with elevational richness in plants (Bhattarai et al. 2004, Kro¨mer et al. 2005) and reptiles (Fu et al. 2006). For endemic frogs in the Hengduan Mountains, water and temperature were only weakly correlated with patterns of species richness.

Conclusions This study has provided some answers to the questions presented at the outset. 1) We could reject the argument 925


that frog species richness has a monotonic decreasing relationship along the elevational gradient and replaced this with an alternative unimodal hypothesis. 2) Elevational richness patterns for total frogs, endemics and non-endemics exhibited different patterns, and endemic richness peaked at higher elevations. 3) Land area explained a small amount of the variation in species richness of total frogs, endemics and non-endemics. 4) Endemic species richness followed the mid-domain model predictions, and showed a nonlinear relationship with temperature. 5) Water and energy were the most important variables in explaining elevational patterns of non-endemic species richness. 6) A suite of interacting climatic and geometric factors best explained total species richness patterns along the mountains elevational gradient. Acknowledgements  This study was financially supported by ‘‘Sustaining project of career development of young teachers’’.

References Allen, A. P. et al. 2002. Global biodiversity, biochemical kinetics, and the energetic-equivalence rule.  Science 297: 1545 1548. Bachman, S. et al. 2004. Elevational gradients, area and tropical island diversity: an example from the palms of New Guinea.  Ecography 27: 299 310. Bhattarai, K. R. et al. 2004. Fern species richness along a central Himalayan elevational gradient, Nepal.  J. Biogeogr. 31: 389 400. Brown, J. H. 2001. Mammals on mountainsides: elevational patterns of diversity.  Global Ecol. Biogeogr. 10: 101 109. Brown, W. C. and Alcala, A. C. 1961. Populations of amphibians and reptiles in the submontane and montane forest of Cuernos Negros, Philippine Islands.  Ecology 42: 628 636. Cardelu´s, L. C. et al. 2006. Vascular epiphyte distribution patterns: explaining the mid-elevation richness peak.  J. Ecol. 94: 144 156. Colwell, R. K. 2006. RangeModel: a Monte Carlo simulation tool for assessing geometric constraints on species richness. Ver. 5.  User’s guide and application published at B/http:/ viceroy.eeb.uconn.edu/RangeModel/. Colwell, R. K. and Lees, D. C. 2000. The mid-domain effect: geometric constraints on the geography of species richness.  Trends Ecol. Evol. 15: 70 76. Colwell, R. K. et al. 2004. The mid-domain effect and species richness patterns: what have we learned so far?  Am. Nat. 163: E1 E23. Colwell, R. K. et al. 2005. The Mid-Domain effect: there’s a baby in the bathwater.  Am. Nat. 166: E149 E154. Currie, D. J. 1991. Energy and large-scale patterns of animaland plant-species richness.  Am. Nat. 137: 27 49. Duellman, W. E. 1988. Patterns of species diversity in anuran amphibians in the American tropics.  Ann. Missouri Bot. Gard. 75: 79 104. Dunn, R. R. et al. 2006. The river domain: why are there more species halfway up the river?  Ecography 29: 251 259. Fauth, J. E. et al. 1989. Elevational patterns of species richness, evenness, and abundance of the Costa Rican leaf litter herpetofauna.  Biotropica 21: 178 185. Fischer, J. and Lindenmayer, D. B. 2005. The sensitivity of lizards to elevation: a case study from south-eastern Australia.  Div. Distrib. 11: 225 233.

926

Fu, C. et al. 2004. Patterns of diversity, altitudinal range and body size among freshwater fishes in the Yangtze River basin, China.  Global Ecol. Biogeogr. 13: 543 552. Fu, C. et al. 2006. Elevational gradients of diversity for lizards and snakes in the Hengduan Mountains, China.  Biodiv. Conserv., in press. Graham, M. H. 2003. Confronting multicollinearity in ecological multiple regression.  Ecology 84: 2809 2815. Grytnes, J. A. and Vetaas, O. R. 2002. Distribution of vascular plant species richness and endemic richness along the Himalayan elevation gradient in Nepal.  Global Ecol. Biogeogr. 11: 291 301. Heyer, W. R. 1967. A herpetofauna study of an ecological transect through the Cordillera de Tilaran, Costa Rica.  Copeia 1967: 259 271. Jetz, W. and Rahbek, C. 2002. Geographic range size and determinants of avian species richness.  Science 297: 1548  1551. Kattan, G. H. and Franco, P. 2004. Bird diversity along elevational gradients in the Andes of Colombia: area and mass effects.  Global Ecol. Biogeogr. 13: 451 458. Kessler, M. 2002. The elevational gradient of Andean plant endemism: varying influences of taxon-specific traits and topography at different taxonomic levels.  J. Biogeogr. 29: 1159 1165. Kluge, J. et al. 2006. What drives elevational patterns of diversity? A test of geometric constraints, climate and species pool effects for pteridophytes on an elevational gradient in Costa Rica.  Global Ecol. Biogeogr. 15: 358  371. Kreft, H. et al. 2006. The significance of geographic range size for spatial diversity patterns in Neotropical palms.  Ecography 29: 21 30. Kro¨mer, T. et al. 2005. Diversity patterns of vascular epiphytes along an elevational gradient in the Andes.  J. Biogeogr. 32: 1799 1809. Lee, D. C. et al. 1999. A null model for species richness gradients: bounded range overlap of butterflies and other rainforest endemics in Madagascar.  Biol. J. Linn. Soc. 67: 529 584. Lennon, J. J. et al. 2004. Contribution of rarity and commonness to patterns of species richness.  Ecol. Lett. 7: 81 87. Li, J. S. et al. 2003. Elevational gradients of small mammal diversity on the northern slopes of Mt. Qilian, China.  Global Ecol. Biogeogr. 12: 449 460. Lomolino, M. V. 2001. Elevational gradients of species diversity: historical and prospective views.  Global Ecol. Biogeogr. 10: 3 13. Lomolino, M. V. et al. 2006. Biogeography, 3rd ed.  Sinauer. McCain, C. M. 2004. The mid-domain effect applied to elevational gradients: species richness of small mammals in Costa Rica.  J. Biogeogr. 31: 19 31. McCain, C. M. 2005. Elevational gradients in diversity of small mammals.  Ecology 86: 366 372. McCain, C. M. 2006. Could temperature and water availability drive elevational richness patterns? A global case study for bats.  Global Ecol. Biogeogr., in press. Mora, C. and Robertson, D. R. 2005. Causes of latitudinal gradients in species richness: a test with fishes of the tropical eastern Pacific.  Ecology 86: 1771 1782. Nathan, R. and Werner, Y. L. 1999. Reptiles and breeding birds on Mt. Hermon: patterns of altitudinal distribution and species richness.  Israel J. Zool. 45: 1 33. Oommen, M. A. and Shanker, K. 2005. Elevational species richness patterns emerge from multiple local mechanisms in Himalayan woody plants.  Ecology 86: 3039 3047. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern?  Ecography 18: 200 205. Rahbek, C. 1997. The relationship between area, elevation, and regional species richness in Neotropical birds.  Am. Nat. 149: 875 902. ECOGRAPHY 29:6 (2006)


Rahbek, C. 2005. The role of spatial scale and the perception of large-scale species-richness patterns.  Ecol. Lett. 8: 224  234. Rangel, T. F. L. V. B. et al. 2006. Towards an integrated computational tool for spatial analysis in macroecology.  Global Ecol. Biogeogr. 15: 321 327. ´ . et al. 2005. Energy, water and large-scale Rodrı´guez, M. A patterns of reptile and amphibian species richness in Europe.  Acta Oecol. 28: 65 70. Schall, J. J. and Pianka, E. R. 1977. Species densities of reptiles and amphibians on the Iberian Peninsula. Don˜ana.  Acta Vertebr. 4: 27 34. Schall, J. J. and Pianka, E. R. 1978. Geographical trends in numbers of species.  Science 201: 679 686.

Va´zquez, L. and Gaston, K. J. 2004. Rarity, commonness and patterns of species richness: the mammals of Mexico.  Global Ecol. Biogeogr. 13: 535 542. Watkins, J. E. Jr et al. 2006. Vascular epiphyte distribution patterns: explaining the mid-elevation richness peak.  J. Ecol. 94: 144 156. Whittaker, R. J. et al. 2001. Scale and species richness: towards a general, hierarchical theory of species diversity.  J. Biogeogr. 28: 453 470.

Subject Editor: Carsten Rahbek.

Download the appendix as file E4802 from B/www.oikos.ekol.lu.se/appendix /.

ECOGRAPHY 29:6 (2006)

927


ECOGRAPHY 15: 177-183. Copenhagen 1992

Latitudinal and elevational variation in fruiting phenology among western European bird-dispersed plants Marcelino Fuentes

Fuentes, M 1992 Latitudinal and elevational variation in fruiting phenology among western European bird-dispersed plants - Ecography 15 177-183 I try to test the prediction that bird-dispersed plants should produce fruits when fruit-eating birds are most abundant by reviewing some phenological data of fleshy fruit production in westem Europe The prediction that fruit ripening dates m populations of the same species should occur later at lower latitudes and elevations, to coincide with the maximum abundance of fruit-eating birds, is not supported by the data The patterns of seasonal variation in the total number and biomass of fruits, but not in the proportion of species in fruit, in communities at different latitudes and elevations do coincide with patterns of seasonal abundances of avian frugivores 1 suggest that this coincidence is due to the greater relative abundance (and contribution to total fruit production) in each locality of species that fruit at times of the year when birds are most abundant These species may have achieved a demographic advantage by getting more seeds dispersed than species that ripen fruits in other seasons M Fuentes, Area de Ecologia, Facuttad de Biologia, Umv de Santiago, E-15071 Santiago de Compostela, Spam

The second seasonal pattern of bird abundance is found in the Mediterranean basin lowlands, where large In areas where strong seasonal fluctuations of bird numbers of migrants (including many frugivorous speabundance occur, it is expected that, m general, bird- cies) coming from northern and higher elevation areas dispersed plants should tend to produce fruits at times join sedentary populations m autumn-winter, producing of the year when fniit-eating birds are most abundant a peak in bird numbers dunng this season (see Herrera (Snow 1971, Thompson and WiUson 1979, Stiles 1980) 1982, 1984, Costa 1984, Jordano 1985, Cuadrado 1986 Two distinct patterns of seasonal variation in the for southem Spain, Luis and Purroy 1980 for the Baabundance of fruit-eating birds have been reported for leanc Islands, Debussche and Isenmann 1989 for southwestern Europe. One is typical of high latitudes and em France, Lovei 1989 for southern Italy) In these elevations and is characterized by minimum abundances wintenng grounds fruit-eatmg bird species make up a from September-October until spnng, as a consequence larger proportion of bird communities than in more of birds migratmg towards warmer regions (see Hogstad northem and mountainous areas (Jordano 1985, Telle1%7, Alatalo 1978 for Scandinavia, Eybert 1973 for ria et al. 1988, see also Willson 1986 for eastern North northern France; Frochot 1973, Cordonnier 1976 for America). A pattern similar to the first one has been found for central France; Purroy 1975, Santos and Suarez 1983 for northern Spam; and Pens Alvarez 1984, Guitian 1984, the northeastern part of North America and one similar Obeso IWI for mountainous areas of the Ibenan Penin- to the second one for the southeastem part of North sula). In more southem localities or at lower elevations, America and around the Gulf of Mexico (Thompson the autumn decrease m bird numbers occurs later, since and WiUson 1979, Stiles 1980, Skeate 1987). Some authors have proposed that, to enhance the the climatic conditions are more favorable.

Introduction

Accepted 24 January 1991 ECOORAPHY 12 ECOGRAPHY 15 2 (1992)

177


Spnng migrant birds scarcely eat these fruits (see above references), which may be seen as surplus of winter production I do not think their case is worth discussing in the context of the predictions tested here For objective (2) I have used all the phenological studies of western European plant communities I could find For objective (3) I have represented all the published data I could find on the phenology of total fruit abundance, except those offered by Guitian (1984) and Jordano (1984) Their data are only refered to in the text because from the information provided by them, I was not able to draw graphics similar to the ones used throughout this paper Obeso (1985) found similar fruit production phenologies in two study sites in the same locality I have also found similar phenologies in my two sites of La Barosa For brevity, I have represented only one of each data sets I estimated the curves for the phenology of total fruit biomass by combining data of individual species phenologies and average masses for each fruit species When fruit masses were not given in the onginal sources I used data from Herrera (1987) I could not construct biomass curves for Obeso's (1985) data because of the difficulty m computing individual Sfjecies phenologies from his graphics Different authors used different methods to record fruiting phenology and, theoretically, there is a possibility of obtaining patterns that are but artifacts derived from inconsistent methodologies The authors used difMethods ferent methods to calculate the dates of peaks and of For objective (1), above, I have compared the dates whole periods of npe fruit availability, l e fruit counts when there is a maximum of npe fruit among different in tagged plants, fruit counts in permanent plots, or populations of the same species I have considered all qualitative observations as in Debussche et al (1987) published information, provided the dates were given and Snow and Snow (1988) Most authors considered a with sufficient accuracy (e g month), as well as some species to be "in fruit" when some npe fruit were preunpublished data that I collected m two northwestern sent in the plants, while others required that a minimum Spanish spiny shrublands (La Barosa) These shrub- proportion be npe, e g 10% of the total crop Delands consisted mainly of Crataegus monogyna, Rosa bussche et al (1987) and Snow and Snow (1988) commicrantha, Rubus ulmifolius, Quercus ilex and Lonicera bined information from wide regions over several years, etrusca For latitudinal variation I have compared local- while the rest of the authors obtained their data in ities at low elevations (< 600 m a s I) For elevational limited areas with relatively homogeneous vegetation, variation I have compared different sites withm re- and gave fruit npenmg dates for different years sepstricted regions (particularly the Sierra de Cazorla area, arately The data reviewed to test prediction (3) were in southeastern Spam). In the Results, I have consid- obtained using very similar methods, except that Soreered that two npt-travi availability peak dates were nsen's (1981) study site was more extensive and indifferent when they were separated by at least a fort- ternally more diverse than those of the rest of the authors. Sorensen (1981) and Herrera (1984 for El Viso) night calculated the abundance of the different fruit species Fruit consumption by birds is well documented for all by combining phenological data and fruit crop estimates the plant species considered (see Herrera 1984, Jordano from tagged plants, and cover estimates for each plant 1984, Guitian 1984, Obeso 1985, Snow and Snow 1988, species. Herrera (1984 for Hoyos de Munoz) periodFuentes 1990, and references therein), except for Rus- ically counted all fruits present in permanent plots. The cus aculeatus, which is nevertheless included because it other studies (mcluding my own unpubl. one of La presents all the typical characteristics of the fruits eaten Barosa) combmed phenological data from tagged by birds (Herrera 1987,1989). Mammals are considered plants, and abundance data from fruit counts m ranto have a minor impact, m comparison to birds, on the domly placed plots. In all cases, I believe that the methseed dispersal of most plant species reviewed (see above odological differences between authors are very slight references) Sometimes, fruits of certam winter-npen- and certainly not hkely to produce, per se, the kind of mg species remain in good condition until early spnng. probability of seed dispersal by frugivorous birds, plants in the northern regions should fruit in early autumn, when many birds switch from an insectivorous diet to a mixed or almost completely frugivorous diet, and before most birds leave for southern areas (Morton 1973. Thompson and Willson 1979, Stiles 1980, Herrera 1982, 1984) Maximum fruit production should occur gradually later at lower latitudes, and take place in winter in bird wintering grounds such as the Mediterranean basin or around the Gulf of Mexico (Snow 1971, Thompson and Willson 1979, Stiles 1980, Skeate 1987, Snow and Snow 1988) A similar reasoning can be applied to elevational gradients fruit ripening should occur later at lower elevations In this paper I will try to test this hypothesis by reviewing the phenological data of fleshy fruit production in western Europe Specifically, I will test the predictions by examining 1) the fruit ripening dates in populations of the same species, 2) the patterns of seasonal variation in the proportion of species bearing npe fruit m different plant communities, and 3) the patterns of seasonal vanation m total fruit abundance (measured in number and biomass of fruits) in different plant communities, situated at different points along latitudinal and elevational gradients

178

ECOGRAPHY 15.2 (1992)


Table 1 Latitude and elevation (in m a s 1 ) of the localities considered Locality

Latitude

Elevation (m a s I )

5r45'

< 200 < 100

Snow and Snow 1988 Sorensen 1981

43''40'

50-950

Debussche and Isenmann 1985 Debussche et al 1987

42''37' 42''50'

600 1350

Fuentes 1990 Guitian 1984

37°26' 37" 9' 37° 1' 37°59' 37''56'

100 10 10 1150 1350

Herrera Jordano Herrera Herrera Obeso

South England Aylesbury Whytham South France Montpellier Cedex North Spain La Barosa Ancares South Spam El Viso Hato Raton Donana Hoyos de Munoz Roblehondo

latitudinal and elevational patterns expected to support or refute the predictions stated above The latitude and elevation of the different localities and the author references are given m Table 1 The annual cycles of abundance of fruit-eatmg birds m the regions considered agree with the patterns outlined in the Introduction, l e high numbers in summer and early autumn, and low in winter, in the more northern regions and mountains, and the reverse in circummediterranean lowland sites (Snow 1971, Herrera 1984, Jordano 1985, Obeso 1985, Debussche and Isenmann 1989, Fuentes unpubl data but see Guitian 1984)

Reference

1984 1984 1986 1984 1985

Therefore, the set of localities is reasonably adequate for testing the hypothesis

Results Fruit ripening dates in different populations of the same species Only 3 {Asparagus acuttfohus, Crataegus monogyna and Ruscus aculeatus) out of 20 species considered suppiort the prediction that southern populations would ripen

Table 2 Fruiting peaks of populations of single species growing at different latitudes The bars separate dates of different years The number indicates first or second fortnight of^ the month

Asparagus acuttfohus Bryonia cretica Cornus sanguinea Crataegus monogyna Daphne gnidium Euonymus europaeus Hedera helix Lomcera periclymenum Osyrts alba Phillyrea angusttfolia Pistacm lentiscus Prunus spinosa Rhamnus alaternus Rosa canina Rubia peregnna Rubus ulnufoltus Ruscus aculeatus Sambucus nigra Snulax aspera Tamus communis

South England (Whytham)

South France (Montpellier)

North Spam (La Barosa)

South Spain (El VISO) (Hato Raton)

_

Oct Jul Sep Sep

_ -

_ — Sep

2Oct Sep* 1 Oct -

1 Dec 1 Apr 2 Sep — _

2 Oct _

2 Jan 2 Sep + lOct _

lNov

Sep Nov Jan Sep Oct Nov Oct Jul Oct Oct Aug Oct Aug Oct Sep

2 Aug 1 Sep/1 Oct 2 Oct —

1 Sep — -

2 Aug — —

2 Nov/1 Nov 2 Sep — 2 Sep

Dec Jun — Oct

2 Nov/1 Dec 2 Sep/1 Aug

-

1 Dec/1 Nov 2 Aug/2 Sep -

2 Aug/1 Jul 2 Sep/1 Sep 2 Jul/2 Aug 2 Nov/2 Oct — -n Sep 2 Aug/1 Aug 2 Nov/1 Dec 1 Oct/1 Oct 2 Aug/1 Jul

' Aylesbury + Rubus fhtticosus n* ECOGRAPHY 15 2 (1992)

179


Table 3 Fruiting peaks of populations of single species growing at different elevations in southern Spain The bars separate dates of different years The number indicates first or second fortnight of the month

Crataegus monogyna Daphne gnidium Juniperus oxycedrus* Lonicera implexa Phillyrea angustifolia Pistacia lentiscus Rhamnus alaternus Rosa canina Rubia peregrma Rubus utmifolius Smilax aspera

Roblehondo

H Mufioz

1 Sep -

_ Nov

Aug-Sep

Nov-Feb

— -

2 Aug 2 Oct

Oct Oct Nov Nov

-

Dec

2 Sep

Oct Nov

-

El Viso

Hato Raton 1 Dec/1 Nov 2 Aug/2 Sep

-

Aug-Oct

Sep Dec Jun Oct

2 Jul/2 Aug 2 Nov/2 Oct

-

-/I Sep 2 Aug/1 Aug 1 Oct/1 Oct

* the fruiting periods, not the peaks, are given

their fruits later than northern ones (Table 2) In 14 species, the southern populations ripen their fruits earlier than the northern ones Two species do not show latitudinal variation in fruiting phenology and one (Ptstacta lenttscus) shows no clear pattern

In 7 out of 11 species, the populations from higher elevation sites npen their fruits later than those of lower elevation ones, thus contrary to initial expectation (Table 3) Only m 3 cases does the reverse occur, and in one no clear pattern emerges In Rubus ulmtfohtis.

LOWLAND LOCALITIES

SOUTH ENGLAND (/Veotxry)

SOUTH ENGLAND (W»v»wni)

SOUTH FRANCE (MontpeHer)

NORTH SPAIN ( U Barosa)

SOUTH SWUN (Doflana)

SOUTH gptm {B vho)

HIGHLAND LOCALITIES

100

CC NORTH SBMN (Arcanis)

aOUTH SFMN (H. M l A » )

SOUTH SFMN (RoUahondD)

100

50 0

J J A S O N D J F M A M SOUTH SmN {Hmi Rattn 1881/82)

J J A S O N D J F M A M

J J A S O N D J F M A M

SOUTH s m N (Han Rat6n 1882/83)

J J A S O N O J F M A M

Fig 1 Curves of the number of species (expressed as percent of the total number of species) beanng npe fruit m communities situated at different latitudes and elevations Thin hnes are the fruiting curves (number of fruits expressed as proportion relative to the peak) of all species combined See Table 1 for geographic information about these communities ECOGRAPHY 15 2 (1992)


100 A t

\ / /

, \ \ Y

LATITUDINAL VARIATION (Number)

50

sites of comparable latitude A similar trend was found by Herrera (1985, Fig 1) for three localities in the Sierra de Cazorla situated at 1150, 1350 and 1800 m a s 1 However, at < 100 m a s 1 the peaks occur on similar or earlier dates than at the 1150 m site

Ul

o

i z

100 LATITUDINAL VARIATION (Biomass)

UJ

o

Dates of maximum total fruit abundance

n

50

UJ

0 Z

100

g

ELEVATIONAL VARIATION (Number)

o Q.

otr

50

Q.

J J A S O N D J F M A M Fig 2 (A) Number of npe fruits, of all species combined, expressed as proportion relative to the maximum abundance of ripe fruits, m three lowland localities situated at different latitudes ( ) Whytham, southern England, (â&#x20AC;&#x201D;) La Barosa, northern Spam, ( ) El Viso, southern Spain (B) Biomass of ripe fnuts, of all species combined, expressed as proportion relative to the maximum quantity (in biomass) of npe fruits, in the same three locahties as in A (C) Number of npe fruits, all species combined, expressed as proportion relative to the maximum abundance of npe fnuts, in three localities situated at different elevations in southern Spain ( ) Roblehondo (1350 m), (â&#x20AC;&#x201D;) Hoyos de Munoz (1150 m), ( ) El Viso (100 m)

maximum npe fruit availability occurs at the same time in highland (Sierra de Ancares) and lowland (La Barosa) populations in northern Spain

Dates of manmum proportion of species in fruit in different communities No clear pattern emerges from latitudinal comparisons of the dates of maximum proportion of species in fruit (Fig 1). Furthermore, the vanability of dates of maximum proportion of species in fruit is often greater withm the same latitude tfian between latitudes. On the other hand, the maximum proportion of species m fruit occurs earlier in tfie two highest sites in the north and south of Spain than in the two lower elevation ECOORAPHY

When curves of the total fruit availability (in number. Fig 2A, and estimated biomass. Fig. 2B), of all species combmed, are compared for three localities at different latitudes, a pattern consistent with mitial predictions is seen The clearest differences are between southern Spam and southern England These differences also hold true when data from a Donana (southern Spam) scrubland are compared with data from England (Jordano 1985, Fig 1) The dates of maximum fruit number m the northern Spanish shrublands fall in an intermediate position between those of more extreme latitudes This also holds true for the dates of maximum fruit biomass in one of the sites, but not in the other, in which maximum biomass occurs on the same dates as in England (Fig 2B) The maximum number of fruit in a 1350 m a s 1 locahty in the Cazorla mountains occurs, earlier than in a 1150 m locality in the same region, as predicted However, there is no time span between the later and a 100 m elevation site m southwestern Spam (Fig 2C) In a northwestern Spanish highland site (Sierra de Ancares, 1400 m) the maximum fruit availability also occurs earlier than in a nearby lowland locality (La Barosa, 600 m) (see Guitian 1984, Fig 0 17)

Discussion Some {xjssible shortcomings of this study (such as the relatively small number of localities for which information IS available, the short duration of studies, or the broad variation observed among nearby locations and among years) make the conclusions somewhat provisional and prompt the need of further studies. However, some of the patterns revealed by this study deserve consideration. The theoretical expectation that, within species, fruiting peaks should occur later at lower latitudes is not supported by the data for western Europe Actually, the prevailing trend observed runs contrary to expectations, most species ripen their fruits earlier at lower latitudes. This contrasts with the assertion of Stiles (1980) for eastern North America. An analogous prediction for elevational vanation (later npenmg at lower elevations) IS also not supported by the data, and the reverse again app>ears to be the rule There is a possibility that np)enmg dates are adapted in each population to particular local conditions, such as abundance cycles of certain

181


frugivores that could depart from the rather general abundance patterns outhned m the Introduction However, this does not explain why fruit ripening tends to occur earlier in lower latitudes and elevations This finding may be better explained by some factor that vanes predictably with latitude and elevation Thus, the variation in fruiting phenology may be caused by differences in flowenng phenology (Pnmack 1985, 1987, Rathcke and Lacey 1985) or m the timing of fruit development and/or maturation (Rathcke and Lacey 1985, Duke 1990) induced by latitudinal and elevational climatic variation No clear pattern emerges when comparing the dates of maximum proportion of species in fruit among localities situated at different latitudes, contrary to the pattern reported for eastern North America (Thompson and Willson 1979, Skeate 1987), but peaks tend to occur earlier at higher elevations than at lower ones in southern Spam This indicates that many species that produce fruits when fruit-eating birds are relatively scarce are nevertheless able to persist successfully in plant communities The variation in the dates of maximum fruit availability among communities situated along latitudinal and elevational gradients agrees with the variation in the dates of maximum abundance of avian fruit consumers This match between fruiting and bird abundance is not accomplished, as we have seen, either by intraspecific vanation, or by variation m the frequencies of the species ripening their fruits at different times of the year (at least for latitudinal variation) A possible explanation is that the abundance of fruit-eating birds has shaf>ed the fruit availability curves through a demographic, rather than evolutionary, process At a given locality, the shape of the fruit abundance curve is usually determined by the disproportionate influence of a single, very abundant sf)ecies, which produces most fruits in the habitat In northern and high elevation communities the curve is dominated by an early-fruiting species {Sambucus ntgra in southern England (Sorensen 198f), Berberts hispantca in Sierra de Cazorla highlands (Obeso 1985)) and in southern and low elevation localities by a late-npenmg one (Ptstacta lenttscus in southern Spanish lowlands (Herrera 1984, Jordano 1984)) The greater abundance of these species may be, at least partly, due precisely to their fruiting when birds are most abundant m their particular localities These species may gain a demographic advantage by getting more seeds dispersed than species that njien fruits m other seasons, and thus they shift the community fniitmg curve in a way that matches fnigivore abundance phenology (see also Herrera 1985). This kind of ecological sorting of populations of different speaes, and not adaptive change withm species, may underly many allegedly coevolutionary adjustments among animals and plants (Janzen f985).

182

Acknowledgements - I thank J Guitian. C M Herrera. P Jordano and M F WiUson, for commenting upon earlier versions of the manuscript C M Herrera and L F Turnes kindly helped to improve my English 1 thank my parents for their financial support This study was funded by grants from Xunta de Gahcia (Tercer Cido and XUGA-8030789) and the Spanish Mimsterio de Educacion y Ciencia (FPI and PB 86-0453)

References Alatalo, R V 1978 Bird community energetics in a boreal coniferous forest - Holarct Ecol 1 367-376 Cordonnier. P 1976 Etude du cycle annuel des avifaunes par la methode des "points d'ecoute" - Alauda 44 169-180 Costa. L 1984 Composicion de las comunidades de aves en pmares del Parque Nacional de Donana (suroeste de Espana) - Donana Acta Vert 11 151-183 Cuadrado. M 1986 La comumdad de aves de un acebuchar del sur de Espana durante el penodo mvernal y de cria Donana Acta Vert 13 71-85 Debussche, M and Isenmann, P 1985 Frugivory of transient and wintering European robins Erithacus rubecula in a Mediterranean region and its relationship to ornithochory - Holarct Ecol 8 157-163 - and Isenmann. P 1989 Fleshy fruit characters and the choices of bird and mammal seed dispersers in a Mediterranean region - Oikos 56 327-338 - . Cortez. J and Rimbault, I 1987 Vanation in fleshy fruit composition in the Mediterranean region the importance of npenmg season, life-form, fruit type and geographical distribution - Oikos 49 244-252 Duke, N C 1990 Phenological trends with latitude in the m&ngxo\c XKS Avicennia marina - J Ecol 78 113-133 Eybert, M C 1973 Le cycle annuel des oiseaux dans trois stades evolutifs d'une pmede de Bretagne - Terre Vie 27 507-522 Frochot, B 1973 L'evolution seasonniere de l'avifaune dans Una futaie de chenes en Bourgogne - Terre Vie 25 145182 Fuentes, M 1990 Relaciones entre pajaros y frutos en un matorral del norte de Espana variaciones estacionales y diferencias con otras areas geograficas - Ardeola 3753-66 Guitidn, J 1984 Ecologia de la comumdad de passenformes de un bosque montano de la Cordillera Cant^brica occidental - Ph D Thesis, Univ Santiago Herrera, C M 1982 Seasonal vanation in the quahty of fruits and diffuse coevolution between plants and avian dispersers - Ecology 63 773-785 - 1984 A study of avian frugivores. bird-dispersed plants, and their interaction m Mediterranean scrublands - Ecol Monogr 54 1-23 - 1985 Habitat-consumer interactions m frugivorous birds In Cody, M L (ed ), Habitat selection in birds Academic Press, Orlando, FL, pp 341-365 - 1987 Bird-dispersed plants of the Iberian Peninsula a study of fruit characteristics - Ecol Monogr 57 305-331 - 1989 Frugivory and seed dispersal by carnivorous mammals, and assoaated frtut characteristics, in undisturbed Mediterranean habitats - Oikos 55 250-262 Herrera, J 1986 Flowenng and fruiting phenology in the coastal shniblands of Donana, south Spam - Vegetatio 68 91-98. Hogstad, O 1%7 Seasonal fluctuation in bird populations within a forest area near Oslo (southern Norway). - Nytt Mag Zool 15:81-96 Janzen, D.H 1985 On ecological ftttmg -Oikos45 308-310 ECOGRAPHY 15 2 (1992)


Jordano, P 1984 Relaciones entre plantas y aves frugi'voras en el matorral mediterraneo del area de Donana - Ph D Thesis, Umv Sevilla - 1985 El ciclo anual de los pasenformes frugi'voros en el matorral mediterraneo del Sur de Espana lmportancia de su invernada y vanaciones interanuales - Ardeola 32 69-94 Lovei, G L 1989 Passerine migration between the Palaearctic and Africa - Current Ornithology 6 143-174 Luis, E and Purroy, F J 1980 Evolucion estacional de las comunidades de aves en la lsla de Cabrera (Baleares) Studia Oecol 1 181-223 Morton, E S 1973 On the evolutionary advantages and disadvantages of fruit eating in tropical birds - Am Nat 107 8-22 Obeso, J R 1985 Comunidades de Passenformes y frugivorismo en altitudes medias de la Sierra de Cazorla - Ph D Thesis, Umv Oviedo - 1987 Comunidades de passeriformes en bosques mixtos de altitudes medias de la Sierra de Cazorla - Ardeola 34 37-59 Pens Alvarez, S J 1984 Avifauna mvernante y nidificante en la Sierra de Bejar (Sistema Central, Provincia de Salamanca) - Studia Oecol 3 219-230 Primack, R B 1985 Patterns of flowering phenology in communities, populations, individuals, and single flowers - In White, J (ed ), The population structure of vegetation Dr W Junk, Dordrecht, The Netherlands, pp 571-593 - 1987 Relationships among flowers, fruits, and seeds - Ann Rev Ecol Syst 18 409-430

ECOGRAPHY 15 2 (1992)

Purroy, F J 1975 Evolucion anual de la avifauna de un bosque mixto de coniferas y frondosas en Navarra - Ardeola 21 669-697 Rathcke, B and Lacey, E P 1985 Phenological patterns of terrestrial plants - Ann Rev Ecol Syst 16 179-214 Santos, T and Suarez, F 1983 The bird communities of the heathlands of Palencia The effects of coniferous plantations - Proc VII Int Cong Bird Census Work, pp 172179 Skeate, S T 1987 Interactions between birds and fruits in a northem Florida hammock community - Ecology 68 297309 Snow, B and Snow, D 1988 Birds and berries - T & AD Poyser, Waterhouses, England Snow, D W 1971 Evolutionary aspects of fruit-eatmg by birds -Ibis 113 194-202 Sorensen, A E 1981 Interactions between birds and fruits in a British woodland - Oecologia (Berl ) 50 242-249 Stiles, E W 1980 Patterns of fruit presentation and seed dispersal in bird-dissemmated woody plants in the eastern deciduous forest - Am Nat 116 670-688 Telleria, J L , Santos, T and Carrascal, L M 1988 La invernada de los pasenformes (Orden Passenformes) en la Peninsula Ibenca - In Telleria, J L (ed ), Invernada de aves en la Peninsula Ibenca SEO, Madrid, pp 153-166 Thompson, J N and Willson, M F 1979 Evolution of temperate fruit/bird interactions phenological strategies Evolution 33 973-982 WiUson, M F 1986 Avian frugivory and seed dispersal in eastern North Amenca - Current Ornithology 3 223-279

183


Ecography 34: 8593, 2011 doi: 10.1111/j.1600-0587.2010.06250.x # 2011 The Authors. Journal compilation # 2011 Ecography Subject Editor: Francisco Pungnaire. Accepted 18 January 2010

Demographic processes of upward range contraction in a long-lived Mediterranean high mountain plant Luis Gime´nez-Benavides, Marı´a Jose´ Albert, Jose´ Marı´a Iriondo and Adria´n Escudero L. Gime´nez-Benavides (luis.gimenez@urjc.es), M. J. Albert, J. M. Iriondo and A. Escudero, A´rea de Biodiversidad y Conservacio´n, Univ. Rey Juan Carlos-ESCET, Tulipa´n s/n. ES-28933 Mo´stoles, Madrid, Spain.

We analyzed demographic data of a long-lived high mountain Mediterranean plant, Silene ciliata Poirret, over a 4-yr period. Selected populations were located at contrasting altitudes at the southernmost margin of the species (Sierra de Guadarrama, central Spain), representing a local altitudinal range at the rear edge of its overall distribution. Previous studies have suggested that differences in the reproduction and performance of individuals at upper and lower populations may have implications for population dynamics. We used matrix analysis to assess their demographic behaviour. Life Table Response Experiments were used to identify the life history stages most relevant to observed differences in population growth rates between populations. Transition matrices revealed great spatio-temporal variability in demographic traits. Seedling recruitment was very low each year in all populations. Maximum longevity of S. ciliata individuals in the lower peripheral population was much lower compared to the central population, probably due to higher adult mortality. Population growth rate (l) showed a declining trend at the lowest altitude and a relatively stable trend at the central population. Long-term simulations also indicated a great risk of quasi-extinction at the lowest population. Our results suggest that rear edge populations of S. ciliata at Sierra de Guadarrama are suffering demographic processes that may be leading to the latitudinal displacement of the species’ range.

Peripheral populations of plant species are important reservoirs of intraspecific genetic diversity and evolutionary potential (Lesica and Allendorf 1995, Hampe and Petit 2005, Jump and Pen˜uelas 2005), as well as functional drivers of ecosystem stability (Eriksson 2000). Studies of individual plant performance across a species’ range frequently find lower survival and/or reduced fecundity at range margins compared to the range center (Jump and Woodward 2003, Gime´nez-Benavides et al. 2007a, b, Marcora et al. 2008). However, a major concern is whether reductions in fitness components really affect population growth and persistence. In fact, many times the persistence of plant species is not crucially dependent on reproductive success and seedling establishment (Pico and Riba 2002, Garcı´a 2008, Iriondo et al. 2008). Differences in life-history traits, such as life-span, will largely determine the species dependence on sexual regeneration. In long-lived perennial species, the impact of limited seed output on population maintenance is difficult to determine due to the complexity of recruitment, but a tradeoff between sexual regeneration and persistence of already established individuals has been suggested (Garcı´a and Zamora 2003). Typical examples of persistence due to longevity and/or vegetative reproduction are more frequent in stressful and unstable environments such as arid, alpine

and rocky habitats (Grime 2001, Garcı´a and Zamora 2003) where geographical limits of reproduction do not necessarily coincide with actual range limits (Gaston 2003). Therefore, to get a complete view of the factors shaping geographical range limits, the components of individual performance must be integrated into population dynamic models across species’ distributions (Angert 2006, Foden et al. 2007). This task is especially relevant today, when ongoing climate warming and other anthropogenic impacts, such as habitat fragmentation and changes in land use, are currently threatening peripheral populations. Despite this, there are few detailed comparisons of population dynamics of central vs marginal populations (Nantel and Gagnon 1999, Stokes et al. 2004, Angert 2006, Samis and Eckert 2007). Moreover, while demographic and evolutionary traits underlying the expansion of species at their leading edge have been more extensively studied during the last few decades (Petit et al. 2004), population dynamics responsible for range contractions at the rear margins of species’ distributions have not received sufficient attention (Hampe and Petit 2005). This general lack of mechanistic studies is even greater in high mountain environments, even though they provide an excellent opportunity for the study of range margins (Angert 2006, Ko¨rner 2007).

85


In mountain plants, conditions for regeneration and survival are hierarchically arranged within their distribution range. Firstly, they are more suitable in the latitudinal centre of their distribution area than in the periphery. Secondly, they also appear structured within each mountain island. Similar to latitudinal range displacements, recent altitudinal shifts in the abundance and distribution of species inhabiting mountain environments have been documented during the last few decades. Several studies have revealed contemporary alterations of species richness in high summits (Grabherr et al. 1994, Gottfried et al. 1999, Virtanen et al. 2003, Walther et al. 2005, Pauli et al. 2007, Erschbamer et al. 2009). The migration of lowland plant species to higher elevations forces subsequent displacements of alpine species (Theurillat and Guisan 2001). Altitudinal shifts in vegetation belts and distribution ranges of species have already been documented (Walther et al. 2002, Klanderud and Birks 2003, Pen˜uelas and Boada 2003, Lesica and McCune 2004). Therefore, there is an urgent need for accurate forecasting of the consequences of this process. Important progress on species’ distribution modelling has recently been made, but most of these models do not explicitly take into account the essential mechanisms operating at individual and population scales (Thuiller et al. 2008, Morin and Thuiller 2009). These factors may cause important bias and inaccuracy in current projection models, and are a probable cause of the divergence found among coarse and fine resolution models when compared (Trivedi et al. 2008). At least at the population level it seems basic to monitor and model demographic trends at rear populations. Despite this necessity, demographic studies of high mountain plants are still scarce compared to those of lowland species, and few studies have documented the populationlevel dynamics driving altitudinal displacements (Doak and Morris 1999, Diemer 2002, Angert 2006). In the present work, we analyzed demographic data of a long-lived Mediterranean high mountain plant, Silene ciliata (Caryophyllaceae), at different altitudes within its southernmost margin of distribution (central Spain). During the last 45 yr, mean air temperature has increased by 1.88C in this area, and days of snowcover per year have decreased by 19.7 d (Gime´nez-Benavides et al. 2007a). In addition to direct impacts of climate warming on the species’ performance, the area has suffered a substantial bottom-up shrub encroachment (Sanz-Elorza et al. 2003) with potential consequences for the persistence of S. ciliata rear populations. The final objective of the present work is to assess whether the reproductive and recruitment failure observed at the rear edge of the species (Gime´nez-Benavides et al. 2007a, 2008) results in regressive population dynamic in the lower population compared to higher altitude populations. We argue that the breakdown of sexual regeneration at lower limits could only be balanced out by a long lifespan and reduced adult mortality. Otherwise, the species could suffer a high risk of peripheral extinction and altitudinal range contraction under the present global warming context. We analyzed demographic performance of the species over a 4-yr period using transition matrix models and long-term simulations. Specifically, the questions addressed were: 1) are S. ciliata populations at the rear altitudinal edge experiencing a declining population trend? 2) Are the upper and lower populations along an altitudinal gradient driven by the 86

same demographic processes? And, if they differ, 3) what are the vital rates responsible for the observed differences in population growth rates at different altitudes? 4) Do populations at different altitudes differ in their probability of quasi-extinction in the long term?

Methods Plant species and study site Silene ciliata (Caryophyllaceae) is a long-lived perennial plant that grows in main mountain ranges of the Balkan Peninsula, the Appenines, the Massif Central in France and the northern half of the Iberian Peninsula, covering a latitudinal range from 408N to 468N (Tutin et al. 1995). In mountain ranges of central Spain, where the species reaches its southernmost margin, it grows from 1900 m (treeline zone) up to the highest summits (ca 2600 m). The species typically grows in a compact cushion-shape. Its flowering period extends from late June to early-mid September. Flowering stems (133 per adult plant) are 15 cm in height and bear 15 flowers. Hand-crossing experiments indicate that S. ciliata is a self-compatible species. However, passive autogamy is restricted by a pronounced protandry so it requires pollinators (Gime´nez-Benavides et al. 2007a). Although many alpine species are highly clonal (Forbis 2003), no evidence of vegetative propagation was observed in this species when several individuals were dug up (Gime´nez-Benavides unpubl.). The study area was in the Sierra de Guadarrama (Pen˜alara Natural Park), a mountain range located in central Spain, 50 km north of Madrid city (408N, 38W). Mean annual precipitation at Navacerrada Pass weather station (1800 m, 8 km southwest of the study site) is 1350 mm, and is concentrated from late autumn to early winter. A marked drought season occurs from late May to October (Fig. 1). Snowfall generally begins in October and the snow-free season begins in MayJune (Palacios et al. 2003). This work is part of a broader study of factors controlling the distribution and performance of S. ciliata along the local altitudinal gradient. Three populations were selected for this demographic approach. The first population was located in the vicinity of Laguna Chica (hereafter Laguna), a small glacial lake situated in a moraine deposit in the treeline zone (1970 m). Vegetation is dominated by a dense shrub cover of Cytisus oromediterraneus and Juniperus communis subsp. alpina intermingled with a low-dense stand of Pinus sylvestris. Here, S. ciliata is displaced by the shrub species and only grows in small, isolated pasture patches dominated by Festuca curvifolia. The second population was on the Dos Hermanas peak (hereafter Dos Hermanas), a summit flat area situated at 2250 m, dominated by a Cytisus-Juniperus shrub formation and patchy xerophytic fellfields of Festuca curvifolia. This fellfield community bears extreme winds and a relatively short snowcover period, and is characterized by the abundance of cushion plants (Escudero et al. 2004). The third population was located at the summit of the highest peak of the mountain range, the Pen˜alara peak (hereafter Pen˜alara), at 2440 m. This area is dominated by the Festuca fellfield and shrub species are scarce. As a consequence of


Figure 1. Climatic data at the Navacerrada Pass weather station (40846?N, 4819?W; 1860 m, located 8 km south-west of study sites). Columns and lines represent monthly mean precipitation and temperature, respectively, during the period 19462006 and in the study years, 2003 to 2006.

reported temperature increase and reduction of snow cover, probably combined with a moderate reduction in livestock grazing, the Cytisus-Juniperus shrub belt is encroaching and replacing the Festuca cryophilic pastures colonized by S. ciliata (Sanz-Elorza et al. 2003). Census scheme Data were collected from 2003 to 2006. One permanent plot was established in each population for demographic monitoring. Plot size varied between populations due to differences in microhabitat characteristics and plant density but populations were larger enough and similar in slope and orientation to be considered representative at each altitude considered. At Laguna (1970 m), we initially monitored all plants available within a small population (128 individuals in a 9 m2 plot), while at Dos Hermanas (2250 m) and Pen˜alara (2440 m) we tagged 266 (7 m2 plot) and 168 individuals (10 m2 plot), respectively, within a larger, continuous population. All plants found within each plot were mapped to allow subsequent location. Plants were monitored every year at the end of the reproductive season (SeptemberOctober). Plant size was estimated as maximum cushion diameter. Total number of inflorescences per plant was counted in a single visit as an estimate of reproductive output. Previous studies suggest that inflorescence number is a good surrogate of fruit production (Pearson’s r790, 549 and 580, pB0.0001, for Laguna, Dos Hermanas and Pen˜alara respectively, Gime´nez-Benavides et al. 2007a).

Silene ciliata seedlings emerge at the beginning of the growing season, suffering extremely high mortality during summer (Gime´nez-Benavides et al. 2007b). Thus, all seedlings found within the plots at the end of the growing season were registered as a basis for estimating annual recruitment rates. Unfortunately, the plot at Pen˜alara was vandalized during the second year of study, so parameter estimations of only one transition could be obtained. The possible uncertainty in plant population structure and demographic parameters derived from the establishment of a single plot per population was assessed by comparing them with five extra plots of 5 m2 randomly placed at each population. Plant size of every plant in these extra plots was measured in 2005, and size structures of the permanent plots were compared with those of the respective extra plots from the same altitude by cross-tabs and chi-square test. Stage classification We established five stage classes, one seedling class and four reproductive classes. The first corresponded to seedlings of about 1 cm in diameter that germinated in spring. Seedlings that survived into the next growth period grew into a reproductive class. Reproductive classes were obtained by classification of individuals by k-means clustering (Hartigan 1975), except seedlings, using pooled data from all populations: small (plants of 1.52.5 cm diameter), medium (34.5 cm), large (58 cm) and extra-large (]8.5 cm). 87


Matrix construction We constructed Lefkovitch matrices for each population and time interval using estimates of inflorescence production, seedling recruitment and transition probabilities between reproductive stages. Transitions were built from the underlying vital rates (survival, growth and fertilitity), following Morris and Doak (2002). When sample size of some stages was small (n B13)  mainly seedlings and small reproductive plants at Laguna  survival rates were obtained from average transition frequencies across all years for each population (Menges and Dolan 1998, Angert 2006). The reproduction terms in the matrix were estimated as follows. Mean number of inflorescences per class was used to calculate the proportional contribution of each adult class to total reproductive effort. Thus, the reproduction term for each reproductive class in each transition was estimated following the equation: Fi;t ;t 1 

sdlt 1  Ri;t ; 4 X (Ri;t  ni;t ) i1

where Fi,t,t1 is the reproduction matrix element of class i (small, medium, large or extra-large) for the period t to t  1, sdlt1 is the total number of seedlings censused in the population in time t 1, Ri,t is the proportional contribution of class i to total reproductive effort, and ni,t is the number of individuals in class i surviving at time t. Thus, seedlings censused in the following year were allocated among the four reproductive classes according to their proportional reproductive effort in the previous year. These estimations assumed that seedlings in time t 1 germinated from seeds produced in time t, as occurs in the absence of a permanent soil seed bank. Field and lab assays

showed that germination capacity of S. ciliata seeds can reach 100% over a one year period (Gime´nez-Benavides et al. 2005, 2007b). Thus, we assumed that soil seed bank does not play an important role in the dynamics of S. ciliata populations. A recent seedbank study conducted on this fellfield community also supports the absence of a permanent seedbank in this species (GarcĹ´a-Camacho 2009). Transition matrix models project population size according to the equation: x(t 1)Ax(t ); where x is a vector of the number of individuals in different plant stages (stage distribution at time t) and A is a matrix of probabilities and fertilities that defines the survival and reproduction of individuals in each stage between time t and time t1 (i.e. matrix elements). The transition matrix A is derived from a life cycle graph that shows the possible transitions between stages (Caswell 2001). The life cycle graph for our model system is shown in Fig. 2. The dominant eigenvalue for each transition matrix was used to calculate the finite rate of increase, l. Bootstrapped matrices were generated by randomly sampling individuals with replacement within stage classes, using a Matlab routine (Matlab 7.0, MathWorks). Two-thousand bootstrapped transition matrices were used to obtain bias-corrected 95% confidence intervals for l. Maximum plant longevity was estimated for each population separately following Forbis and Doak (2004). A starting vector of one seedling and zero reproductive individuals was multiplied by the mean matrix with all fecundities set to zero (Caswell 2001). Year by year, the resulting vector was multiplied by the mean matrix until the summed probability of survival for all stage classes reached 0.01.

Figure 2. Life cycle of Silene ciliata. Each arrow represents a one-year transition. S denotes survival (stasis in the same class, and growth or regression to a different class) and F fertility.

88


Life Table Response Experiments

X (l ) (dh) 1l (ai;j ai;j ) l(l ) :l(dh)  1ai;j A m i;j

j

where l(dh) is the population growth rate in Dos Hermanas, and Am is the mean matrix from both populations. The first factor of the sum (in parenthesis) denotes the differences, and the result of multiplying them by the appropriate sensitivity are the contributions (Caswell 2001). We carried out an LTRE analysis for each time interval. All matrix analyses were performed with Matlab 7.0. Probability of quasi-extinction We calculated the probability of reaching a quasi-extinction threshold for each population by computer simulations. We used the Matlab routine developed by Morris and Doak (2002), which estimates the quasi-extinction time cumulative distribution function for a structured population in a stochastic environment. Proportions of individuals were equal to the initial stage structure of each population. Environmental stochasticity was implemented by using the three available transition matrices for each population. It was based on the climatic variability among the periods of time considered. The years 2004, 2005 and 2006 had milder temperatures and summer rainfall (data not shown, Navacerrada Pass weather station, 40846?N, 4819?W, 1860 m), whereas in 2003 the most severe European heatwave took place (Scha¨r et al. 2004) (Fig. 1). Maximum temperatures of the complete climate series were reached in summer, no rainy days were recorded in July, and only one in August (Gime´nez-Benavides et al. 2007a). Therefore, two contrasting scenarios were considered: in the ‘‘realistic’’ scenario the probability of occurrence of an extremely warm and dry year (first transition matrix) was fixed to 0.1 whereas in the ‘‘warming’’ scenario this probability was substantially increased by making the three transition matrices equally probable. We set the quasiextinction threshold at 50 individuals and time horizon at 250 yr. Pen˜alara was not included in the simulations because only one transition was available. The starting population density was the number of individuals of the initial plots, i.e. 128 and 266 individuals for Laguna and Dos Hermanas, respectively.

40 Individuals (%)

We used Life Table Response Experiments (LTREs) to identify the matrix elements most relevant to the observed differences in population growth rate between populations (Caswell 2001). We focused on a one-way fixed design where the two populations (Laguna and Dos Hermanas) were of interest in themselves. We chose the Dos Hermanas matrix as the reference matrix, i.e. a baseline for measuring population effects. Thus, population growth rate in Laguna was defined as:

50

Laguna (Low) Dos Hermanas (Middle) Peñalara (High)

30

20

10

0 Seedling

Small

Medium

Large

Extra-large

Stage classes

Figure 3. Size structure of S. ciliata populations according to stage classes established in transition matrices. Data from the three permanent plots set in 2003. n128 individuals in Laguna, 266 individuals in Dos Hermanas and 168 individuals in Pen˜alara.

large and extra-large classes. On the contrary, Dos Hermanas and Pen˜alara populations were consisted of smaller plants, with a majority of medium and large plants and B15% of extra-large individuals. The seedling class was scarcely represented in all populations. The results obtained in our permanent plots did not differ in terms of size structure from the surrounding extra plots (low: x2  13.4, p0.640, medium: x2 22.2, p 0.332, high: 19.4, p0.248), thus the permanent plots could be considered representative of their corresponding population. Matrix analyses Transition matrices also revealed great differences in demographic traits between populations (Supplementary material Table S1). Adult survival was higher at Dos Hermanas for all stage classes and all time intervals. Mean survival per class increased at Dos Hermanas from small to extra-large plants, while mean survival was higher for large plants at Laguna. Seedling survival showed high variability between years and populations. The climatically extreme year (20032004 transition) showed higher mortality only for seedling and extra-large plants in Laguna (Supplementary material Table S1). Population growth rates (l) showed a declining trend at the lowest altitude and relatively stable population dynamics at the central and highest population. Bootstrapping confidence intervals of l were much wider at Laguna than at Dos Hermanas (Fig. 4). Estimated maximum plant longevity inferred from average transition matrices was 147 yr at Dos Hermanas and only 23 yr at Laguna. Life Table Response Experiments

Results Plant size structure showed great differences between populations (Fig. 3). Laguna population was dominated by large plants, with 70% of individuals belonging to

LTRE analyses revealed greater differences in matrix elements between populations for the seedling stage, in addition to greater variability in differences (Fig. 5a). However, the higher seedling survival at Laguna slightly contributed to differences in l between populations (Fig. 5b). Lower 89


1.10 Laguna (Low) Dos Hermanas (Middle) Peñalara (High)

0.8 0.6

2003–04 2004–05 2005–06

0.4 LTRE differences

Median population growth rate

1.05

(a)

1

0.95

0.2 0 –0.2 –0.4 –0.6

0.9

–0.8

(b)

0.85

Seedling

Small

Medium

Large

Extra-large

0.03 Fecundity Survival

0.02

0.75 2003–04

2004–05

2005–06

Time intervals

LTRE contributions

0.01

0.8

0 –0.01 –0.02 –0.03 –0.04 –0.05

Figure 4. Lambda values and 95% confidence intervals (from 2000 bootstrapped transition matrices) for each population and time interval. Dashed line highlights a stable population (l 1).

–0.06

Seedling

Small

Medium

Large

Extra-large

–0.07 Stage classes

Probability of quasi-extinction The probability of reaching a threshold density of 50 individuals was very different between Laguna and Dos Hermanas under both scenarios. Such probability reached 100% after only ca 16 yr at Laguna, irrespective of the scenario considered. However, at Dos Hermanas 100% probability of quasi-extinction was reached in about 240 yr under the ‘‘realistic’’ scenario, and was reached 50 yr earlier under the ‘‘warming’’ scenario (Fig. 6).

Discussion Demographic processes at contrasting altitudes Our study clearly revealed that the S. ciliata population at the lowest altitude presented a pronounced decline and showed a demographic behaviour highly different from those at the central and highest altitudes. These differences were great enough to suggest that the studied peripheral population of S. ciliata at its lower margin is not able to withstand present conditions, which may lead to its local extinction. This scenario was supported by the long-term simulations showing great quasi-extinction hazard at the lowest altitude, irrespective of the climatic scenario projected (Fig. 6). In the central population, finite rate of increase was relatively stable, and time to quasi-extinction was much longer than at the rear edge, indicating that 90

Figure 5. LTRE differences (a) and contributions (b) to differences in l between populations, per each stage class. Contributions were grouped per vital rate (survival and fecundity). Bars represent the mean value9SD for all time intervals. The Dos Hermanas matrix was used as the reference matrix.

population declining is not likely to be occurring at this altitude. However, quasi-extinction probability under a warming scenario (with a higher chance of extreme years) showed that this population is also vulnerable to changing conditions. The larger confidence intervals of l at Laguna 1 Cumulative probability of quasi-extinction

survival of all reproductive stages at Laguna most greatly contributed to differences in l. Fecundity contributions increased with stage class but were always substantially lower than survival contributions (Fig. 5b).

0.9 Laguna (Low) 0.8

Dos Hermanas (Middle)

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

50

100

150

200

250

Years into the future

Figure 6. Average of ten simulated cumulative distribution functions of 5000 simulations, for the time to reach a quasiextinction threshold of 50 individuals for both populations of S. ciliata. The plot only shows the ‘‘warming’’ scenario, in which all time intervals are equally probable.


compared to those at Dos Hermanas reflected higher uncertainty in population growth estimates, as a consequence of higher variability in plant fates. These results were expected because marginal populations are usually near the limit of their physiological tolerance and, consequently, are more vulnerable to environmental stochasticity (Gaston 2003). Although we could only provide one transition, the most favourable trend is expected at the highest population (Pen˜alara peak), based on the positive finite rate of increase (l 1.059) and the significantly higher values of adult survival (100% for all reproductive stages). Seedling recruitment observed in the field was highly variable and mean values were very low for all the studied years. This is a common feature in long-lived, high mountain species (Forbis 2003, Forbis and Doak 2004). Furthermore, recruitment in Mediterranean mountains is seriously limited by high seedling mortality during summer drought (Castro et al. 2004, 2005, Cavieres et al. 2005). The climatic diagram presented in Fig. 1 precisely showed that mean monthly temperature during the growing season of all the studied years was 1.528C over the mean of the last 60 yr. Moreover, mean monthly precipitation was severely reduced just in July, the harsher month of the summer period. The studied period is therefore representative of the warming trend occurred in the last half of the century in this area (Wilson et al. 2005, Gime´nez-Benavides et al. 2007a). Previous results in S. ciliata showed that recruitment was mainly limited by low seed production and low seedling emergence and survival, probably due to environmental harshness in summer, in contrast to biotic factors such as flower and fruit predation, which had a minor effect on the probability of plant recruitment (Gime´nez-Benavides et al. 2008). Although a reciprocal sowing experiment detected some evidence of local adaptation in seedling establishment along this gradient (Gime´nez-Benavides et al. 2007b), the probability of recruitment (estimated as the probability of an ovule becoming a 2-yr-old plant) was 20 to 40-fold higher in medium and higher populations compared to the lower population (Gime´nez-Benavides et al. 2008). Reduced fecundity at species’ distribution limits has been widely observed, and may be the main factor responsible for lower population densities and aged population structures (Garcı´a et al. 2000, Dorken and Eckert 2001, Jump and Woodward 2003, Marcora et al. 2008). As predicted by life history theory (Grime 2001), extended longevity is expected to allow long-term persistence of remnant populations in harsh environments with high interannual climate variation, while waiting for eventual recruitment episodes (Morris and Doak 1998, Garcı´a and Zamora 2003, Garcı´a 2008). In fact, our LTRE analyses suggested that observed differences in l between populations, much more pronounced at the seedling stage, were only slightly explained by the fecundity term (Fig. 5b). Thus, fecundity didn’t seem an important factor explaining differences in population growth rate between altitudes. For this reason, demographic trends of long-lived plants at their rear edge populations cannot simply be inferred from their current recruitment rates, as they are more determined by adult mortalities (Hampe and Petit 2005). Moreover, long-lasting longevity have been proved to buffer the variability in vital rates associated with the variability in climate, hence reducing the vulnerability of long-lived species to climate change (Morris et al. 2008).

Interestingly, the effects of plant size on survival did not follow the same trend at both altitudes. A decrease in plant survival of extra-large plants occurred at the lower population, while this stage reached 100% probability of survival at the central population. Size-dependence of demographic fates defined as transition probabilities has been commonly observed in a great variety of plants and environments (Hortvitz and Schemske 1995). Greater adult survival has also been observed in other alpine cushion-form Caryophyllaceae such as Silene acaulis (Morris and Doak 1998), Minuartia obtusiloba and Paronychia pulvinata (Forbis and Doak 2004). Smaller size stages are expected to be more vulnerable to losses of above-ground tissues during adverse environmental conditions (e.g. seasonal drought), and consequently are expected to show higher rates of mortality. However, our results suggest that environmental conditions at the lower limit may also be affecting survival of large plants by seriously reducing their life span. Indeed, maximum longevity of S. ciliata individuals, inferred from transition matrices, was very low at Laguna population (23 yr) compared to Dos Hermanas (147 yr). In general, maximum longevity of S. ciliata is relatively low when compared to S. acaulis, another cushion plant from arcticalpine habitats (Morris and Doak 1998). The lifespan of S. acaulis may extend over 300 yr and demographic studies carried out in this species did not detect mortality among large plants (Morris and Doak 1998). The results found at Pen˜alara peak  the highest population of the Guadarrama mountain range  were more in accordance with this pattern (100% adult survival, resulting in a maximum longevity of over 350 yr). Evidence of changes in plant life-history along altitudinal gradients has been reported previously (Ko¨rner 2007). Von Arx et al. (2006) detected significantly older plants and lower growth rates at higher altitudes by means of herb-chronology in three forb species, corresponding to a more conservative life-history. Population dynamics of Silene ciliata: towards an altitudinal range contraction Our results highlight the relevance of survival and longevity for dissecting the processes that may be driving such distinct population dynamics. Together with reproductive limitations (Gime´nez-Benavides et al. 2007a, 2008), rear edge populations of S. ciliata at Sierra de Guadarrama are suffering other demographic processes, resulting in low adult survival, which may force them to an altitudinal range shift. Populations inhabiting the rear edge will become completely extirpated if current demographic processes prevail. As upward shift is unviable, since the species actually colonizes the highest summits of the major mountain ranges in this southern margin, range shifts will therefore be irremediably associated with a decrease in habitat area. Biotic causes, apart from direct climatic effects on survival and reproduction, are undoubtely involved in the expected habitat contraction. The encroachment of the high mountain xerophytic pastures (the main niche of S. ciliata) by montane shrub species already detected in the area (Sanz-Elorza et al. 2003), is probably one of the major sources of risk. Remnant patches of suitable habitat are currently colonized by small-sized populations dominated 91


by adult and senescent plants with extremely low proportions of seedlings and juveniles and a high degree of isolation. Reduced population sizes and isolation are common factors limiting individual reproductive performance (Leimu et al. 2006), leading to a feedback process towards local extinction. Under this scenario, long-term persistence would only be possible by the longevity of established individuals, but our results highlight that longevity is also seriously eroded at this range margin. These findings contrast with the assumption that populations of many alpine species are not likely to be affected substantially by climate warming due to their long lifespan (Steinger et al. 1995, Diemer 2002). Evolutionary implications in response to climate change As noted above, extreme longevity coupled with occasional recruitment episodes may support the demographic stasis and even growth of perennial plant populations. However, a more important consequence of population dynamics governed by extreme longevity arises in the context of global change. The evolutionary adaptation of populations to changes in environmental conditions varies over both space and time as a consequence of natural selection operating on fitness components, and eventually fixed by sexual regeneration. Rapid climate change may act as a potent agent of natural selection within populations and, in this context, the adaptive potential of a given population will be partially ruled by the frequency of sexual regeneration, being annual plants the quickest to adapt because of their short generation time (Jump and Pen˜uelas 2005). By contrast, in long-lived perennials with delayed regeneration time, the lag of adaptation will be significantly longer. Moreover, in S. ciliata an extremely low recruitment rate is coupled to size-dependent reproduction in stressful years, especially in its lowland rearing edge (Gime´nez-Benavides et al. 2007a, 2008). In years of extreme summer drought, small-sized individuals have a much lower flowering probability, seriously limiting opportunities for adaptive selection. In conclusion, although high longevity is the last strategy to assure the long-term persistence of remnant populations of S. ciliata, it is also critically reduced at its lowland range limit. This situation is affecting the population growth rate, eventually forcing the upward shift and the contraction of its regional distributional range. Further demographic studies are required to gather the necessary data to move from simple bioclimatic niche models to process-based models that take into account both climate change and population dynamics. Acknowledgements  The authors especially thank Pedro QuintanaAscencio for his help with Matlab programming and the staff of Parque Natural de las Cumbres, Circo y Lagunas de Pen˜alara who gave them permission to work in the area. They also thank Nuria Ortega, Vera Ortega and Rau´l Garcı´a-Camacho who helped with the field work and Lori De Hond for her linguistic assistance. This work was supported by projects ISLAS (CGL2009-13190-C0301), SIL-HAD (CGL2009-08755) and LIMITES (CGL2009-

92

07229) funded by the Ministerio de Ciencia e Innovacio´n (Spain) and REMEDINAL2 funded by Comunidad de Madrid.

References Angert, A. L. 2006. Demography of central and marginal populations of monkeyflowers (Mimulus cardinalis and M. lewisii).  Ecology 87: 20142025. Castro, J. et al. 2004. Seedling establishment of a boreal tree species (Pinus sylvestris) at its southernmost distribution limit: consequences of being in a marginal, Mediterranean habitat.  J. Ecol. 92: 266277. Castro, J. et al. 2005. Alleviation of summer drought boots establishment success of Pinus sylvestris in a Mediterranean mountain: an experimental approach.  Plant Ecol. 181: 191 202. Caswell, H. 2001. Matrix population models: construction, analysis, and interpretation.  Sinauer. Cavieres, L. A. et al. 2005. Nurse effect of the native cushion plant Azorella monantha on the invasive non-native Taraxacum officinale in the high-Andes of central Chile.  Perspect. Plant Ecol. Evol. Syst. 7: 217226. Diemer, M. 2002. Population stasis in a high-elevation herbaceous plant under moderate climate warming.  Basic Appl. Ecol. 3: 7784. Doak, D. and Morris, W. 1999. Detecting population-level consequences of ongoing environmental change without long-term monitoring.  Ecology 80: 15371551. Dorken, M. E. and Eckert, C. G. 2001. Severely reduced sexual reproduction in northern populations of a clonal plant, Decodon verticillatus (Lythraceae).  J. Ecol. 89: 339350. Eriksson, O. 2000. Functional roles of remnant plant populations in communities and ecosystems.  Global Ecol. Biogeogr. 9: 443450. Erschbamer, B. et al. 2009. Short-term signals of climate change along an altitudinal gradient in the South Alps.  Plant Ecol. 202: 7989. Escudero, A. et al. 2004. Patch dynamics and islands of fertility in a high mountain Mediterranean community.  Arct. Antarct. Alp. Res. 36: 518527. Foden, W. et al. 2007. A changing climate is eroding the geographical range of the Namib Desert tree Aloe through population declines and dispersal lags.  Divers. Distrib. 13: 645653. Forbis, T. A. 2003. Seedling demography in an alpine ecosystem.  Am. J. Bot. 90: 11971206. Forbis, T. A. and Doak, D. F. 2004. Seedling establishment and life history trade-offs in alpine plants.  Am. J. Bot. 91: 1147 1153. Garcı´a, D. and Zamora, R. 2003. Persistence and multiple demographic strategies in long-lived mediterranean plants.  J. Veg. Sci. 14: 921926. Garcia, D. et al. 2000. Geographical variation in seed production, predation and abortion in Juniperus comunis throughout its range in Europe.  J. Ecol. 88: 436446. Garcia, M. B. 2008. Life history and population size variability in a relict plant. Different routes towards long-term persistence.  Divers. Distrib. 14: 106113. Garcı´a-Camacho, R. 2009. Evaluacio´n del e´xito reproductive de Armeria caespitosa en el contexto de cambio clima´tico en la alta montan˜a mediterra´nea.  PhD thesis, Univ. Rey Juan Carlos, Madrid, Spain. Gaston, K. J. 2003. The structure and dynamics of geographical ranges.  Oxford Univ. Press. Gime´nez-Benavides, L. et al. 2005. Seed germination of high mountain Mediterranean species: altitudinal, interpopulation and interannual variability.  Ecol. Res. 20: 433444.


Gime´nez-Benavides, L. et al. 2007a. Reproductive limits of a lateflowering high-mountain Mediterranean plant along an elevational climate gradient.  New Phytol. 173: 367382. Gime´nez-Benavides, L. et al. 2007b. Local adaptation enhances seedling recruitment along an altitudinal gradient in a high mountain Mediterranean plant.  Ann. Bot. 99: 723734. Gime´nez-Benavides, L. et al. 2008. What shapes the altitudinal range of a high mountain Mediterranean plant? Recruitment probabilities from ovule to seedling stage.  Ecography 31: 731740. Gottfried, M. et al. 1999. A fine-scaled predictive model for changes in species distribution patterns of high mountain plants induced by climate warming.  Divers. Distrib. 5: 241 251. Grabherr, G. et al. 1994. Climate effects on mountain plants.  Nature 369: 448. Grime, J. P. 2001. Plant strategies, vegetation processes, and ecosystem properties.  Wiley. Hampe, A. and Petit, R. J. 2005. Conserving biodiversity under climate change: the rear edge matters.  Ecol. Lett. 8: 461 467. Hartigan, J. A. 1975. Clustering algorithms.  Wiley. Horvitz, C. C. and Schemske, D. W. 1995. Spatiotemporal variation in demographic transitions for a neotropical understory herb: projection matrix analysis.  Ecol. Monogr. 65: 155192. Iriondo, J. M. et al. 2008. Conservation of plant populations. Myths and paradigms.  In: Valladares, F. et al. (eds), Unity in diversity. Reflections on ecology after the legacy of Ramon Margalef. Fundacio´n BBVA, pp. 247267. Jump, A. S. and Woodward, F. I. 2003. Seed production and population density decline approaching the range-edge of Cirsium species.  New Phytol. 160: 349358. Jump, A. S. and Pen˜uelas, J. 2005. Running to stand still: adaptation and the response of plants to rapid climate change.  Ecol. Lett. 8: 10101020. Klanderud, K. and Birks, H. J. B. 2003. Recent increases in species richness and shifts in altitudinal distributions of Norwegian mountain plants.  Holocene 13: 16. Ko¨rner, C. 2007. The use of ‘altitude’ in ecological research.  Trends Ecol. Evol. 22: 569574. Leimu, R. et al. 2006. How general are positive relationships between plant population size, fitness and genetic variation?  J. Ecol. 94: 942952. Lesica, P. and Allendorf, F. W. 1995. When are peripheral populations valuable for conservation?  Conserv. Biol. 9: 753760. Lesica, P. and McCune, B. 2004. Decline of arcticalpine plants at the southern margin of their range following a decade of climatic warming.  J. Veg. Sci. 15: 679690. Marcora, P. et al. 2008. The performance of Polylepis australis trees along their entire altitudinal range: implications of climate change for their conservation.  Divers. Distrib. 14: 630636. Menges, E. S. and Dolan, R. W. 1998. Demographic viability of Silene regia in midwestern prairies and relationships with fire management, genetics, geography, population size, and isolation.  J. Ecol. 86: 6378. Morin, X. and Thuiller, W. 2009. Comparing niche- and processbased models to reduce prediction uncertainty in species range shifts under climate change.  Ecology 90: 13011313. Morris, W. F. and Doak, D. F. 1998. Life history of the long-lived gynodioecious cushion plant Silene acaulis (Caryophyllaceae), inferred from size-based population projection matrices.  Am. J. Bot. 85: 784793.

Morris, W. F. and Doak, D. F. 2002. Quantitative conservation biology. Theory and practice of population viability analysis.  Sinauer. Morris, W. F. et al. 2008. Longevity can buffer plant and animal populations against changing climatic variability.  Ecology 89: 1925. Nantel, P. and Gagnon, D. 1999. Variability in the dynamics of northern peripheral versus southern populations of two clonal plant species, Helianthus divaricatus and Rhus aromatica.  J. Ecol. 87: 748760. Palacios, D. et al. 2003. Distribution and effectiveness of nivation in Mediterranean mountain, Pen˜alara (Spain).  Geomorphology 54: 157178. Pauli, H. et al. 2007. Signals of range expansions and contractions of vascular plants in the high Alps: observations (19942004) at the GLORIA master site Schrankogel, Tyrol, Austria.  Global Change Biol. 13: 147156. Pen˜uelas, J. and Boada, M. 2003. A global change-induced biome shift in the Montseny mountains (NE Spain).  Global Change Biol. 9: 131140. Petit, R. J. et al. 2004. Ecology and genetics of tree invasions: from recent introductions to Quaternary migrations.  For. Ecol. Manage. 197: 117137. Pico´, X. F. and Riba, M. 2002. Regional-scale demography of Ramonda myconi: remnant population dynamics in a preglacial relict species.  Plant Ecol. 161: 113. Samis, K. E. and Eckert, C. G. 2007. Testing the abundant center model using range-wide demographic surveys of two coastal dune plants.  Ecology 88: 17471758. Sanz-Elorza, M. et al. 2003. Changes in the high-mountain vegetation of the central Iberian Peninsula as a probable sign of climate warming.  Ann. Bot. 92: 273280. Scha¨r, C. et al. 2004. The role of increasing temperature variability in European summer heatwaves.  Nature 427: 332336. Steinger, T. et al. 1995. Long-term persistence in a changing climate: DNA analysis suggests very old ages of clones of alpine Carex curvula.  Oecologia 105: 9499. Stokes, K. E. et al. 2004. Population dynamics across a parapatric range boundary: Ulex gallii and Ulex minor.  J. Ecol. 92: 142155. Theurillat, J.-P. and Guisan, A. 2001. Potential impact of climate change on vegetation in the European Alps: a review.  Clim. Change 50: 77109. Thuiller, W. et al. 2008. Predicting global change impacts on plant species’ distributions: future challenges.  Perspect. Plant Ecol. Evol. Syst. 9: 137152. Trivedi, M. R. et al. 2008. Spatial scale affects bioclimate model projections of climate change impacts on mountain plants.  Global Change Biol. 14: 10891103. Tutin, T. et al. 1995. Flora Europaea Vol. 15.  Electronic dataset supplied by Pankhurst, R. J. Virtanen, R. et al. 2003. Long-term changes in alpine plant communities in Norway and Finland.  In: Nagy, L. et al. (eds), Alpine biodiversity in Europe. Ecological Studies 167. Springer, pp. 411421. von Arx, G. et al. 2006. Evidence for life history changes in high-altitude populations of three perennial forbs.  Ecology 87: 665674. Walther, G.-R. et al. 2002. Ecological responses to recent climate change.  Nature 416: 389395. Walther, G.-R. et al. 2005. Trends in the upward shift of alpine plants.  J. Veg. Sci. 16: 542548. Wilson, R. J. et al. 2005. Changes to the elevational limits and extent of species ranges associated with climate change.  Ecol. Lett. 8: 11381146.

Download the Supplementary material as file E6250 from <www.oikos.ekol.lu.se/appendix>. 93


ECOGRAPHY 26: 291–300, 2003

Species-richness patterns of vascular plants along seven altitudinal transects in Norway John Arvid Grytnes

Grytnes, J. A. 2003. Species-richness patterns of vascular plants along seven altitudinal transects in Norway. – Ecography 26: 291– 300. Altitudinal richness patterns were investigated along altitudinal gradients located in northern Norway (two transects) and along a west– east gradient in southern Norway (five transects). The transects were sampled for vascular plant species richness using a uniform sampling method. Each transect consisted of 38 – 48 5 ×5 m sample plots regularly spaced from sea level or valley bottom to a local mountain top. In five transects species richness peaked at mid-altitudes, whereas in the two northern transects species richness decreased with altitude. The observations were qualitatively evaluated in relation to the influence of the area of the species pool, hard boundaries, temperature and precipitation, and mass effect. The observed patterns cannot be fully accounted for by any of these factors. However, the altitude of the peak in species richness was above the forest-limit for all the humped relationships, which may suggest that species richness above the forest-limit might be enhanced by a mass effect from forest taxa. The two monotonic relationships found in the north may be caused by the relatively low number of alpine species at these sites. The monotonic pattern may result from a decrease in ‘‘forest species’’ towards the mountain tops. J. A. Grytnes ( jon.grytnes@bot.uib.no), Dept of Botany, Uni6. of Bergen, Alle´gaten 41, N-5007 Bergen, Norway.

Understanding altitudinal species-richness patterns is important for the management of species diversity in a world that may become warmer due to human impact. Large environmental variation within a small geographical area makes altitudinal gradients ideal also for investigating several ecological and biogeographical hypotheses (Ko¨rner 2000). Altitudinal gradients have therefore become increasingly popular for investigating patterns in species richness in recent years. At present it is not clear if there are any common patterns of species richness and altitude. Rahbek (1995) reviewed studies on altitudinal species-richness patterns and found that approximately half the studies revealed a hump-shaped pattern. Also monotonically decreasing species richness with altitude or no relationship before a decrease in richness with altitude are common altitudinal richness patterns. Studies since this review confirm that both unimodal (e.g. Rahbek 1997, Fleishman et al. 1998, Heaney 2001, Md. Noor 2001) and monotonically

decreasing (e.g. Patterson et al. 1998, Odland and Birks 1999, Ohlemu¨ller and Wilson 2000, Austrheim 2002) patterns of richness occur with altitude. Rahbek (1995) suggests that differences between studies may at least partially result from failure to correct for area and sampling effort (see also Terborgh 1977, McCoy 1990, Whittaker et al. 2001). There is a major lack of studies on altitudinal richness that are directly comparable and this makes it difficult to assess whether the variation in patterns are real or due to sampling artefacts. Lomolino (2001a) calls for studies comparing altitudinal trends between mountain ranges using the same sampling regimes for all transects and standardised plot sizes within transects. In this study I investigate the pattern of vascular plant species richness along seven altitudinal gradients. All are sampled with the same method and identical plot size, ensuring that the patterns both within and between transects are not a result of different sampling

Accepted 28 October 2002 Copyright © ECOGRAPHY 2003 ISSN 0906-7590 ECOGRAPHY 26:3 (2003)

291


strategies. The main aim of the study is to describe statistically the altitudinal pattern in species richness in different areas in Norway. In addition a discussion of some selected potential causes for the observed pattern is presented by setting up tentative predictions for what is expected if each of these factors is important in determining the richness pattern.

Methods and study area Transects were sampled along seven hillsides at different geographical locations in Norway. Two transects were in northern Norway, and the remaining five make a gradient from west to east in southern Norway. The transects were chosen after looking at detailed maps (scale 1:50 000) to find transects with a relatively long continuous hillside with little variation in aspect. Transects were selected to be in areas of homogeneous bedrock type (Holtedahl and Dons 1960). Some important characteristics of the selected transects are summarised in Table 1. Sampling was done during the summers of 1999, 2000, and 2001. Plots were placed along the transects starting as far down as possible with the criterion that plots should have a similar aspect throughout the transect, and not be placed in clear-cuts or other areas clearly influenced by humans. All transects are, however, influenced by grazing by domestic animals, mainly sheep. The highest altitude sampled is either a local top, or where vegetation cover became discontinuous. In one case weather conditions (30 cm of fresh snow) and time limitations prevented sampling at higher altitudes even though the vegetation was continuous for at least another 100 altitude m (Kvitingskjølen).

Plot size (or grain size) is 5 × 5 m. All vascular plants were recorded in each plot. Nomenclature and taxonomy follows Lid and Lid (1994) with the exception that Taraxacum and Hieracium species were only identified to genus level. Depressions and ridges were avoided when the plots were placed, as were mires. This was done to avoid sampling extreme wet and extreme dry places. Generally, distance between plots is 20 vertical metres with some deviations depending on the length of the gradient, steepness of hillside, etc. The number of plots in a transect varies from 38 (Lynghaugtinden) to 48 (Gra˚ heivarden and Tronfjellet), which is considered sufficient for a statistical evaluation of the altitudinal species richness pattern for each transect. The relatively small grain size compared to other studies on altitudinal variation in species richness was used because it was possible to have a large number of samples within each transect and to sample several transects. The small plot size also ensures a fairly complete list of species from each plot. Sampling at scales not usually sampled may also give extra information and suggest how processes may influence the patterns. However, there may be other processes explaining species richness patterns if larger areas are sampled than if smaller grain sizes are used (Rahbek and Graves 2001, Whittaker et al. 2001). In addition, the fine scale of the sampling area makes the estimate of species richness sensitive to the number of individuals (Gotelli and Colwell 2001). To evaluate if the small grain size created a different pattern than would have been detected with larger grain sizes, two approaches were used. First, testing the altitudinal pattern using larger sample sizes in four of transects, and second, using a nested sampling to evaluate if the altitudinal richness pattern is dependent on sampling area.

Table 1. A summary of some important characteristics for each transect. Location of each transect is indicated by county, latitude, and longitude (see also Fig. 1). The lowest and highest points locally corresponding to the hard boundaries along the hillside where the transect is sampled are given. Note that this does not always correspond to the highest and lowest points sampled (cf. Fig. 2). A crude description of the bedrock geology is taken from Holtedahl and Dons (1960). Approximate forest limit as observed in the field is indicated. Aspect of the sampled transect is given. Annual precipitation and mean January and July temperature from the nearest station are from Førland (1993) and Aune (1993), respectively (altitude of the climatic station is given in brackets). Mountain

Trollan

County

Latitude, Lowest– Bedrock Longitude highest point (m)

Bø in Vestera˚ len 68°46%N, 14°33%E Lynghaugtinden Bø in Vestera˚ len 68°40%N, 14°32%E Horndalsnuten Voss 60°39%N, 06°38%E Gra˚ heivarden Jondalen 60°13%N, 06°15%E Grjothøi Skja˚ k 61°52%N, 08°08%E Kvitingskjølen Lom 61°45%N, 08°43%E Tronfjellet Tynset/Alvdal 62°11%N, 10°41%E

292

0–543 0–504 300–1461 0–1271 450–1952 360–2060 600–1666

Forest Aspect Annual Jan°C limit precipitation (mm (m))

Acidic 20 granite Acidic 200 granite Acidic 700 supracrustal Acidic 600 granite Acidic 950 gneiss Basic plu1000 tonic Basic Cam- 900 bro-Silurian

S

1505 (3)

Jul°C

−1.4 (12)

12.1

WNW 1017 (12)

−1.4 (12)

12.1

N

1555 (590)

−4.6 (590) 11.8

N

2200 (342)

N

278 (372)

−9.4 (378) 13.9

NNW

321 (382)

−9.4 (378) 13.9

E

500 (485)

−11.4 (485) 12.5

0.8 (1)

14.3

ECOGRAPHY 26:3 (2003)


For the first approach larger areas were sampled in addition to the ordinary 5 ×5 m plots in four transects. In one transect (Grjothøi) a plot of 10 × 10 m surrounding each of the ordinary plots was used. In three other transects (Horndalsnuten, Kvitingskjølen, and Tronfjellet) a 25 m walk to each side of the main plot perpendicular to the slope of the hillside was made for approximately half the ordinary plots. All species within one metre were noted and added to the 5 × 5 m plot, i.e. the richness of the large-scale plots are the sum of the 5×5 m plots plus the extra species in the 50 × 2 m walk. The second approach is evaluating if the altitudinal richness pattern is plot-size sensitive or not along one of the seven transects (Grjothøi). If the emerging richness pattern is a result of different number of individuals in the plots along the altitudinal gradient (Gotelli and Colwell 2001), this will be clearest at small grain sizes (fewest number of species). At larger grain sizes the number of individuals will increase in all plots and due to the relationship between the number of individuals and the number of species, the effect of the number of individuals on the altitudinal richness pattern will be dampened. In one transect (Grjothøi), a nested design of samples were used to allow a test of the sensitivity of the altitudinal richness pattern on sampling area. Quadrat samples of 0.25 m2 (four contiguous samples), 1 m2, 25 m2 (four contiguous samples), and 100 m2 were sampled at each altitudinal level (a total of ten samples at each altitude level) and multiple regression was used to detect if any significant change in the altitude pattern could be found with sampled area. This was statistically evaluated by testing if the interaction between sampled area and altitude was significant after including both variables in multiple regression. Area was log-transformed prior to analyses.

Statistical methods A Generalised Linear Model (GLM; McCullagh and Nelder 1989) was made for each individual transect. The most common patterns described for altitudinal richness, and the main interest in this study, are whether richness is monotonically or unimodally related to altitude. GLM using first- or second-order polynomials was therefore tested against no relationship and against each other. However, a GLM with a first- or second-order polynomial is restricted to be linear or symmetrically unimodal, whereas several authors describe a plateau before a decrease or an asymmetric hump. Therefore a more complex pattern described by non-parametric regression may give valuable additional information when inspecting the graphical output. A Generalised Additive Model (GAM; ECOGRAPHY 26:3 (2003)

Fig. 1. Map showing the geographical location of the seven transects.

Hastie and Tibshirani 1990) with a cubic smoother spline using 3, 4, or 5 degrees of freedom was tested against the best GLM model. As the response variable is counts (number of species) a Poisson distribution is assumed and a log link is used for all regressions. This gave satisfactory results when inspecting diagnostic plots. F-tests were used to evaluate statistical significance.

Predictions Numerous hypotheses have been proposed to explain both a linear and humped relationship between richness and altitude (recently reviewed by Brown and Lomolino 1998, Brown 2001, Lomolino 2001a). Hard boundaries (Colwell and Hurtt 1994, Colwell and Lees 2000, Grytnes and Vetaas 2002), area (MacArthur 1972, Rahbek 1995, 1997, Odland and Birks 1999), climate (Odland and Birks 1999), and mass effect (Shmida and Wilson 1985, Kessler 2000) are commonly discussed when altitudinal species richness patterns are considered. Although no quantitative measurements are available to test the different hypotheses directly, a qualitative evaluation of the predictions made for each of these is possible. This section therefore makes tentative predictions for the different transects based on these four factors. Note that even if this section sets up predictions and these predictions are evaluated in the Discussion section, the transects are far too few to consider this as a rigorous statistical evaluation of the different factors included here. 293


Hard boundaries Random placement of species optima and species ranges between an upper hard boundary (mountain top) and a lower hard boundary (sea level or valley bottom) will give a humped richness pattern along altitude. This effect has been demonstrated by simulations and analytical modelling in several studies (Colwell and Hurtt 1994, Willig and Lyons 1998, Koleff and Gaston 2001, Grytnes and Vetaas 2002). The simulations show that hard boundaries result in a symmetrical humped relationship in the middle of the gradient. It is also evident from these simulations that species richness decreases most steeply as the boundaries are approached. In this study the valley bottom or sea level and the mountain top define the hard boundaries (cf. Table 1). Following the results of simulations, humped relationships are predicted for all transects and maximum species richness is predicted to occur at or near the middle of the gradient (Table 1). Richness should also decrease most strongly as the hard boundaries are approached.

Area of species pool The relationship between species richness and area is well known (Rosenzweig 1995, Lomolino 2001b). In this study the sampled area is the same along the transect, hence the direct effect of plot area is cancelled out. However, as several studies have demonstrated, the number of species in a sample is dependent on the number of species in the species pool (Ricklefs 1987, Cornell 1999). The concept of species pool has been used with various meanings (Grace 2001). Some take the species present in a larger area around the focus area to be the species pool for the focus area (Caley and Schluter 1997), whereas other incorporates the constraints set by the environment (Gough et al. 1994, Dupre´ 2000, Grytnes and Birks 2003). Taking the latter approach here gives a potential role for the area through the effect on the number of species in the species pool. This interpretation is in accordance with Terborgh (1973) and Taylor et al. (1990), who were among the first to discuss the importance of species pools (although Terborgh (1973) did not use this term). They argue that higher species richness will be found in plots that have the most common habitat. The main environmental gradient in this study coincides with the altitudinal gradient. The areas covered by an altitudinal interval vary along the gradient. This area may then influence plot species richness through two steps. First, the area of an altitude interval influences the number of species at large scale (the species pool) and, second, the plot species richness depends on the size of the species pool (see also Lomolino 2001a). 294

The transects differ in the distribution of area for the altitude intervals as the mountains are not equally shaped. An accurate estimate of area per altitudinal interval is not available. Therefore only a qualitative inspection of maps at a scale of 1:50 000 is used here. Some of the mountains are more or less dome-shaped in a matrix of large area lowlands, which gives a monotonically decreasing area with altitude at both the local and the regional scale (Ko¨ rner 2000). These mountains are Lynghaugtinden, Trollan, and Tronfjellet. The other mountains are hillsides going down to a narrow valley or to the sea in an area dominated by high mountain plateaux. The area of the altitudinal bands in these mountains therefore depends on the slope of the hillside. Gentle-sloped hillsides result in large areas per altitude metre, and steep hillsides result in small areas per altitude metre. Horndalsnuten and Kvitingskjølen have approximately the same slope all along the transect and go down to a rather narrow valley. The area is therefore approximately equal for each interval along these two transects. At Gra˚ heivarden the hillside is steep and remains steep until it goes into the fjord. The slope becomes gentler, and hence the areas with similar climatic environments become larger above 1000 m. At Grjothøi the steepest slopes are towards the lowest altitudes while the slopes are gentler between 1200 and 1500 m. The predictions from this would therefore be that Lynghaugtinden, Trollan, and Tronfjellet should have a monotonically decreasing richness with altitude, Kvitingskjølen and Horndalsnuten should have no trend in richness, and Grjothøi and Gra˚ heivarden should have no trend until 1200 and 1000 m, respectively, and above these altitudes richness should increase.

Temperature and rainfall Temperature linearly decreases with altitude whereas precipitation has a more erratic pattern with altitude, but generally increases towards higher altitudes. Several studies have demonstrated that temperature or precipitation may influence species richness at broader scales (Currie and Paquin 1987, Leathwick et al. 1998, Odland and Birks 1999, Grytnes et al. 1999, Ohlemu¨ ller and Wilson 2000). Assuming that these variables have a positive effect on species richness along altitude in the transects in this study a monotonically decreasing trend for richness with altitude is expected (not necessarily linear if precipitation is important). The pattern should be similar for all transects. The two climatic variables in combination have often been seen as an indirect estimate of productivity (Currie 1991, O’Brien 1993, Odland and Birks 1999). Focussing on the effect that temperature and precipitation have on productivity, and assuming a positive relationship between productivity and species richness (for a discusECOGRAPHY 26:3 (2003)


sion of this see Rahbek (1997) and Waide et al. (1999)), will give different predictions for the continental versus the oceanic transects. In the oceanic transects (Lynghaugtinden, Trollan, Horndalsnuten, and Gra˚ heivarden), precipitation is probably not limiting in any part of the gradient (annual precipitation between 1017 and 2200 mm; Table 1). This means that temperature probably controls the richness in these areas more than precipitation does and a linearly decreasing richness pattern with altitude can be expected. In the continental transects (Grjothøi, Kvitingskjølen, and Tronsfjellet), however, the precipitation may be limiting in the lower part of the transect (annual precipitation between 278 and 500 mm; Table 1) and temperature in the upper part as precipitation increases and temperature decreases. In these cases a unimodal pattern of species richness can be expected. With the present data it is impossible to say where the optimum richness should be expected along the gradient if a combination of temperature and precipitation is important for the continental transects.

Mass effect Mass effect is the establishment of species in sites where a self-maintaining population cannot exist (Shmida and Wilson 1985). Other closely related terms that have been used to explain the same effect include source – sink effect (Pulliam 1988, 2000) and rescue effect (Brown and Kodric-Brown 1977, Stevens 1989). The most obvious way the mass effect can influence the altitudinal richness pattern is through a feedback among zonal communities. This will increase species richness in transition zones between two bordering communities (Lomolino 2001a). The most noticeable transition along altitudinal gradients is the transition from forest to alpine communities. Many species have their distribution more or less restricted to above or below the forest-limit (Hofgaard and Wilmann 2002). The interchange of species between forest and alpine

communities at the forest limit may cause richness to increase at altitudes corresponding to the forest-limit. The prediction is therefore that species richness will peak at or near the forest-limit, which differs between transects (Table 1). Only the Trollan transect has no forest along the whole transect and here no hump is predicted from the transition-zone or mass effect hypotheses.

Results The results of the regressions are summarised in Table 2 and show that when relating species richness to altitude using the 25 m2 plots a statistically significant second-order polynomial model was found in five of the seven transects (Horndalsnuten, Gra˚ heivarden, Grjothøi, Kvitingskjølen, Tronfjellet, Fig. 2). In none of these five transects was the first-order polynomial model statistically significant when tested against the null model of no relationship (lowest p-value for the first-order polynomial is 0.064 at Horndalsnuten). The two transects in northern Norway (Lynghaugtinden and Trollan) showed a linear relationship with altitude and were the only transects where a second-order polynomial model did not statistically significantly improve the fit. Generally the models explain a large part of the deviance (Table 2). For the four transects where larger grain sizes were tried in addition to the 25 m2 plots, the pattern for the larger grain sizes confirms the patterns from the smaller grain size. In this study both the general pattern and the placement along altitude where maximum species richness is estimated are independent of the grain sizes used (cf. Fig. 2 with Fig. 3). The GAM models significantly improved the fit over the GLM models in four of the transects (Table 2). The Lynghaugtind transect clearly has a break point at 250 m where species richness suddenly drops from ca 22 species to ca 12 species. This is captured by the GAM and not by the GLM model. The reason for a signifi-

Table 2. Summary of the regression models between species richness and altitude for each transect (LS indicates large-scale plots). GLM and GAM models with the respective degrees of freedom for each model given and the p-value of each model (given in brackets) refer to a test against no relationship for the GLM models and for the GAM models the p-value refers to a test against the given GLM model. NS = not significant (p \ 0.05). Transect

Null deviance

GLM model

Lynghaugtinden Trollan Horndalsnuten Horndalsnuten (LS) Gra˚ heivarden Grjothøi Grjothøi (LS) Kvitingskjølen Kvitingskjølen (LS) Tronfjellet Tronfjellet (LS)

80.17 74.93 32.52 35.27 67.50 104.26 140.41 65.11 50.47 311.20 212.23

1 1 2 2 2 2 2 2 2 2 2

ECOGRAPHY 26:3 (2003)

( \ 0.001) (B0.001) (0.02) (0.027) ( \ 0.001) (0.0018) ( \ 0.001) (0.039) (\ 0.001) ( \ 0.001) (\ 0.001)

GLM residual deviance

GAM model

GAM residual deviance

34.80 30.74 26.77 19.29 33.75 78.50 95.84 56.01 20.50 119.30 42.37

5 ( \ 0.001) 4 (0.0045) NS Not applied 5 ( \ 0.001) 5 (0.048) Not applied NS Not applied NS Not applied

15.79 21.80 – – 18.58 65.27 – – – – –

295


Fig. 2. Scatter plots of species richness in relation to altitude in the ordinary 5 Ă&#x2014;5 m plots for the seven transects. The unbroken line is the statistically best GLM model of a monotonic (two transects) or unimodal (five transects) relationship. Where a GAM model statistically improved the GLM model a broken line describing this model is added (see Table 2). The vertical broken line indicates the altitude of the forest-limit for each transect, and the midpoint between valley bottom (or sea level) and mountaintop is indicated by * on the altitude axis.

296

ECOGRAPHY 26:3 (2003)


Fig. 3. Scatter plots of species richness along altitude for the large-scale plots. In Grjothøi a plot of 10 ×10 m is used whereas in the three remaining transects the species in an area of 50 × 2 m is added to the ordinary 5 ×5 m plots. The line refers to the best fitted GLM model (see Table 2).

cant improvement of the GAM models over a linear GLM model in the Trollan transect is the plateau in richness from 300 –400 m (probably due to some species-poor plots around 300 m). The GAM model for the Gra˚ heivarden transect indicates a sharp peak in richness rather than a smooth unimodal curve indicated by the GLM model. In the Grjothøi transect the GAM model indicates a bimodal rather than the unimodal altitudinal pattern. This is due to some species-poor plots in the dense belt of Betula nana between 1100 and 1200 m. The GAM models were not applied for the large-scale plots because there are few samples at this scale. Testing the effect of the interaction term between sampling area and altitude on species richness in the Grjothøi transect after accounting for altitude and area indicates that the richness pattern is independent of sampling area within the scale sampled here (F = 0.45, p=0.64, n =480). Even though the samples at each altitude level are not strictly independent the result is so clear that there is no question about whether the pattern is independent of scale within the range of scales used. ECOGRAPHY 26:3 (2003)

Discussion Even if a standard sampling regime with a standard plot size is used, both humped and monotonically decreasing trends of richness appear. Five of the seven transects have a humped altitudinal richness pattern, whereas the two transects in northern Norway show a monotonically decreasing richness with altitude. This is in accordance with the studies reviewed by Rahbek (1995) who finds that a unimodal pattern is the most common and that a linear decreasing trend with altitude is also frequently observed. Previously, differences in altitudinal richness patterns have been attributed to differences in sampling methods (McCoy 1990, Rahbek 1995). In the present study the difference in patterns cannot be due to differences in sampling or plot sizes. The relatively small grain size used in this study may result in the patterns observed here having different causes than in other studies where larger grain sizes were used. For example, the importance of local biological interactions may be more important at smaller grain sizes, or the pattern may be a result of different numbers of individuals at different altitudes (Gotelli and Colwell 2001). While this may be true, the fact that 297


the pattern is the same when sampling larger areas suggests that the pattern is not due to small grain sizes and is therefore comparable to other studies using larger grain sizes. It is probable that small grain size may only obscure the relationship between altitude and richness. This is indicated by the larger explanatory power of the regressions at the large scale compared to the small scale (seen by comparing the fraction of explained deviance of the two scales derived from Table 2). In other words, a humped relationship is evident in spite of the small grain size rather than because of the small grain size. The independence of grain size is also demonstrated for the Grjothøi transect when testing if the pattern changes significantly when different grain sizes are used. The monotonic altitudinal richness patterns are found in two altitudinal transects with many similar features. The geographical distance between the two transects is short (Fig. 1). This implies similar climatic conditions, similar geology, and similar species pool. In addition, both mountains have short altitudinal gradients (Table 1). The finding of monotonic patterns in both transects strengthens confidence that the observed pattern is real. On the other hand, when comparing these two transects with the others and discussing the predictions, the similarities between the transects imply that the two transects might be treated as pseudoreplicates (Hurlbert 1984), hence we should be careful not to treat them as two independent pieces of ‘‘evidence’’ for or against a specific hypothesis. The predictions made from assuming hard boundaries as the only important factor are that species richness peaks at intermediate altitudes, i.e. intermediate between valley bottom (or sea level) and mountain top. This intermediate peak is marked by an asterisk on the graphs (Fig. 2). Inspection of the figures shows that the transects with unimodal relationships usually have the maximum estimated species richness close to the intermediate altitudinal point. Moreover, species richness at Tronfjellet (where richness is sampled from the valley bottom to the mountain top) seems to decrease more steeply as the hard boundaries are approached. This supports the hypothesis of hard boundaries as an important factor. However, at the other two transects where richness is sampled close to both hard boundaries the pattern is linear and does not correspond to the predictions from hard boundaries. At the remaining four transects with a humped pattern, richness is not sampled close enough to the hard boundaries to evaluate if richness decreases more steeply towards these boundaries. However, the hump at Gra˚ heivarden is clearly due to a peak between 650 and 850 m, and does not behave as the predictions from simulations that assume hard boundaries (e.g. Grytnes and Vetaas 2002). To summarise, indications in support of hard boundaries are found in four transects, but indications contradicting the importance of hard 298

boundaries are found in the remaining three transects (Fig. 3). The only transects where species richness behaves as predicted if area was the only important factor are the two northern transects, namely Lynghaugtinden and Trollan. Here a monotonically decreasing trend is expected and is observed. For the other transects, the predicted pattern and observed pattern do not coincide. The importance of species-pool area (gamma richness sensu Whittaker 1960) for plot species richness is therefore probably relatively minor. Lomolino (2001a) reaches a similar conclusion when discussing alternative explanations for altitudinal richness patterns. The predictions made assuming that temperature or rainfall is important assume that all richness should have a monotonic trend with altitude for all transects. This is not found. Assuming that a combination of the two variables are important and that precipitation is limiting for richness at the lower part of the transects in the continental areas and not in the oceanic is partly supported as the two linear relationships observed are in oceanic areas and all the continental transects have a unimodal relationship. However, the two oceanic transects in southern Norway (Horndalsnuten and Gra˚ heivarden) contradict the predictions as they show a unimodal relationship between richness and altitude whereas a linear relationship is expected. These transects are also the transects with the highest annual precipitation (Table 1). Hence the predictions are only partly supported. Mass effect may also enhance species richness in transition zones, especially in the transition zone defined by the forest-limit. The forest-limit is indicated on the figures and, except for the two monotonic patterns, the estimated maximum species richness is generally well above the forest-limit. This contradicts the interpretation of feedback effects among communities as outlined in the Predictions section. However, only a few alpine species are able to survive in shaded conditions, whereas forest species may be able to survive in the more open areas above the forest-limit. Field observations support this as several species with their main distribution below the forest-limit were found in the plots above the forest-limit, but typically not growing as vigorously as lower down. Examples of such species include Geranium syl6aticum, Geum urbanum, Trientalis europaea, and also small specimens of Betula pubescens and Pinus syl6estris. One of the two transects where a monotonic pattern is found is predicted to show a linear pattern as the forest-limit is almost at sea level (Trollan), whereas for the other transect (Lynghaugtinden) the expected humped pattern does not correspond to the observed linear pattern. However, for both these transects, an alternative hypothesis can be proposed for the lack of a humped pattern. In these northern oceanic parts the ECOGRAPHY 26:3 (2003)


species pool for the highest altitudes may be small, partly due to the effect of the oceanic climate (Moe 1995), but also due to the relative low maximum altitudes of the mountain tops in the area. The plant species on the top of these mountains may therefore be mainly species with their main distribution in the lowlands. To evaluate this possibility I counted the number of presences of each species at the highest 200 m, and compared with the lowest 300 m (340 m for Trollan). Only 9 species clearly have more presences in the upper part compared to the lower part for Trollan (Arctostaphylos alpinus has 13 occurrences in the 19 plots at the highest part and only 1 at the lower part, Juncus trifidus also has 13 above and 1 below, Salix herbacea 18:1, Silene acaulis 5:0, Cerastium alpinum 5:1, and Carex bigelowii 6:0, and three species have one or two occurrences in the upper part and none in the lower part). For the other 57 species there is no difference or more individuals of each species are found in the lower part. In Lynghaugtinden 10 species have more presences in the upper 200 m than in the lower 300 m (Silene acaulis has 13 occurrences in 16 plots above and 0 below, Loiseleuria procumbens 3:0, Antennaria dioica, 4:0, Rubus chamaemorus 4:1, Juncus trifidus 12:6, Arctostaphylos alpinus 6:2, Carex brunnescens 6:0, and three species have one or two occurrences above and none below 300 m), whereas 60 species have no difference or have their main distribution in the lower part. This means that, on average, ‘‘alpine species’’ (main distribution above 300/340 m) makes up a small part of the richness in any plot and hence the pattern of ‘‘forest species’’ (main distribution below 300/340 m) completely masks any trend in ‘‘alpine species’’. The total pattern of richness therefore corresponds to the pattern of ‘‘forest species’’ richness. Another possibly important factor, especially when plots are as small as used in this study, is environmental heterogeneity, especially heterogeneity of moisture within the plot. At low altitudes above-ground water is mostly concentrated in large rivers. At high altitudes small streams run almost everywhere, especially early in the growing season when snow is melting. This may enhance heterogeneity in two ways at high altitudes. First, very different moisture regimes may occur within small distances, and second, small streams create local disturbances and different soil conditions (e.g. different soil depth and grain size differences in soil). As species at high altitudes are generally of a relatively small size, fine-scaled heterogeneity created in this way may be an important determinant for differences in richness. Contrary to the importance of heterogeneity on species richness Austrheim (2001) found an increase in soil nutrient heterogeneity at higher altitudes in semi-natural grasslands in Norway, but this was not reflected in species richness. The moderate disturbance made by small streams may itself also enhance richness in plots at higher altitudes. The importance of moderate disturECOGRAPHY 26:3 (2003)

bance in enhancing species richness has been discussed by Huston (1979, 1994). The main conclusion from this study is that altitudinal richness patterns vary between transects even if the same sampling regime is used between different transects. Five of the seven altitudinal transects studied have a unimodal relationship between species richness and altitude, whereas the two remaining transects have a negative linear relationship. The qualitative evaluation of four causal hypotheses considering all seven transects together gave strongest support for the importance of mass effect, whereas there are only weak support for the area of species pool hypothesis. Further studies increasing the number of transects with similar methods should be done to give clearer support or rejection of the possible causal hypotheses. Acknowledgements – I am grateful to Gunnar Austrheim, H. J. B. Birks, Einar Heegaard, Carsten Rahbek, Ole R. Vetaas, and Vigdis Vandvik for reading the manuscript and giving valuable comments. Thanks are also due to Hans H. Blom, Louise Lindblom, Einar Heegaard, Per G. Ihlen, Anne E. Bjune, and Halvard H. Eggen for assistance in the field. Financial support was given by Grolle Olsen’s legat and NFR-grant no. 127 594/720.

References Aune, B. 1993. Temperaturnormaler normalperiode 1961 – 1990. – Det Norske Meteorologiske Institutt, Universitestforlaget, Oslo. Austrheim, G. 2001. Heterogeneity in semi-natural grasslands: the importance of the elevational gradient. – Nord. J. Bot. 21: 291 – 303. Austrheim, G. 2002. Plant diversity patterns in semi-natural grasslands along an elevational gradient in southern Norway. – Plant Ecol. 161: 193 –205. Brown, J. H. 2001. Mammals on mountainsides: elevational patterns of diversity. – Global Ecol. Biogeogr. 10: 101 – 109. Brown, J. H. and Kodric-Brown, A. 1977. Turnover rates in insular biogeography: effect of immigration and extinction. – Ecology 58: 445 – 449. Brown, J. H. and Lomolino, M. V. 1998. Biogeography, 2nd ed. – Sinauer. Caley, M. J. and Schluter, D. 1997. The relationship between local and regional diversity. – Ecology 78: 70 – 80. Colwell, R. K. and Hurtt, G. C. 1994. Nonbiological gradients in species richness and a spurious Rapoport effect. – Am. Nat. 144: 570 – 595. Colwell, R. K. and Lees, D. C. 2000. The mid-domain effect: geometric constraints on the geography of species richness. – Trends Ecol. Evol. 15: 70 – 76. Cornell, H. W. 1999. Unsaturation and regional influences on species richness in ecological communities: a review of the evidence. – Ecoscience 6: 303 – 315. Currie, D. J. 1991. Energy and large scale patterns of animaland plant-species richness. – Am. Nat. 137: 27 – 49. Currie, D. J. and Paquin, V. 1987. Large-scale biogeographical patterns of species richness of trees. – Nature 329: 326 – 327. Dupre´ , C. 2000. How to determine a regional species pool: a study in two Swedish regions. – Oikos 89: 128 – 136. Fleishman, E., Austin, G. T. and Weiss, A. D. 1998. An empirical test of Rapoport’s rule: elevational gradients in montane butterfly communities. – Ecology 79: 2482 – 2493.

299


Førland, E. J. 1993. Nedbørnormaler normalperiode 1961 – 1990. – Det Norske Meteorologiske Institutt, Universitestforlaget, Oslo. Gotelli, N. J. and Colwell, R. K. 2001. Quantifying biodiversity: procedures and pitfalls in the measurement and comparison of species richness. – Ecol. Lett. 4: 379 – 391. Gough, L., Grace, J. B. and Taylor, K. L. 1994. The relationship between species richness and community biomass: the importance of environmental variables. – Oikos 70: 271 – 279. Grace, J. B. 2001. Difficulties with estimating and interpreting species pools and the implications for understanding patterns of diversity. – Folia Geobot. 36: 71 –83. Grytnes, J. A. and Vetaas, O. R. 2002. Species richness and altitude: a comparison between simulation models and interpolated plant species richness along the Himalayan altitude gradient, Nepal. – Am. Nat. 159: 294 – 304. Grytnes, J. A. and Birks, H. J. B. 2003. The influence of scale and species pool on the relationship between vascular plant species richness and cover in an alpine area in Norway. – Plant Ecol., in press. Grytnes, J. A., Birks, H. J. B. and Peglar, S. M. 1999. Plant species richness in Fennoscandia: evaluating the relative importance of climate and history. – Nord. J. Bot. 19: 489 – 503. Hastie, T. J. and Tibshirani, R. J. 1990. Generalized additive models. – Chapman and Hall. Heaney, R. H. 2001. Small mammal diversity along elevational gradients in the Philippines: an assessment of patterns and hypotheses. – Global Ecol. Biogeogr. 10: 15 –39. Hofgaard, A. and Wilmann, B. 2002. Plant distribution pattern across the forest-tundra ecotone: the importance of treeline position. – Ecoscience 9: 375 –385. Holtedahl, O. and Dons, J. A. 1960. Geological map of Norway (bedrock). – Norges Geologiske Undersøkelse, No. 208. Hurlbert, S. H. 1984. Pseudoreplication and the design of ecological field experiments. – Ecol. Monogr. 54: 187 – 211. Huston, M. A. 1979. A general hypothesis of species diversity. – Am. Nat. 113: 81 –101. Huston, M. A. 1994. Biological diversity. The coexistence of species on changing landscapes. – Cambridge Univ. Press. Kessler, M. 2000. Upslope-directed mass effect in palms along an Andean elevational gradient: a cause for high diversity at mid-elevations? – Biotropica 32: 756 – 759. Koleff, P. and Gaston, K. J. 2001. Latitudinal gradients in diversity: real patterns and random models. – Ecography 24: 341 – 351. Ko¨ rner, C. 2000. Why are there global gradients in species richness? Mountains might hold the answer. – Trends Ecol. Evol. 15: 513 –514. Leathwick, J. R., Burns, B. R. and Clarkson, B. D. 1998. Environmental correlates of tree alpha-diversity in New Zealand primary forests. – Ecography 21: 235 – 246. Lid, J. and Lid, D. T. 1994. Norsk flora, 6th ed. – Det Norske Samlaget, Oslo. Lomolino, M. V. 2001a. Elevational gradients of species-density: historical and prospective views. – Global Ecol. Biogeogr. 10: 3 –13. Lomolino, M. V. 2001b. The species –area relationship: new challenges for an old pattern. – Prog. Phys. Geogr. 25: 1 – 21. MacArthur, R. H. 1972. Geographical ecology, patterns in the distribution of species. – Harper and Row.

300

McCoy, E. D. 1990. The distribution of insects along elevational gradients. – Oikos 58: 313 – 322. McCullagh, P. and Nelder, J. A. 1989. Generalized Linear Models, 2nd ed. – Chapman and Hall. Md. Nor, S. 2001. Elevational diversity patterns of small mammals on Mount Kinabalu, Sabah, Malaysia. – Global Ecol. Biogeogr. 10: 41 – 62. Moe, B. 1995. Studies of the alpine flora along an east-west gradient central-western Norway. – Nord. J. Bot. 77 – 89. O’Brien, E. M. 1993. Climatic gradients in woody plant species richness: towards an explanation based on an analysis of southern Africa’s woody flora. – J. Biogeogr. 20: 181 – 198. Odland, A. and Birks, H. J. B. 1999. The altitudinal gradient of vascular plant species richness in Aurland, western Norway. – Ecography 22: 548 – 566. Ohlemu¨ ller, R. and Wilson, J. B. 2000. Vascular plant species richness along latitudinal and altitudinal gradients: a contribution from New Zealand temperate rainforests. – Ecol. Lett. 3: 262 – 266. Patterson, B. D. et al. 1998. Contrasting patterns of elevational zonation for birds and mammals in the Andes of southeastern Peru. – J. Biogeogr. 25: 593 – 607. Pulliam, H. R. 1988. Sources, sinks, and population regulation. – Am. Nat. 132: 652 – 661. Pulliam, H. R. 2000. On the relationship between niche and distribution. – Ecol. Lett. 3: 349 – 361. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern? – Ecography 18: 200 – 205. Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in neotropical birds. – Am. Nat. 149: 875 – 902. Rahbek, C. and Graves, G. R. 2001. Multiscale assessment of patterns of avian species richness. – Proc. Natl. Acad. Sci. USA 98: 4534 – 4539. Ricklefs, R. E. 1987. Community diversity: relative roles of local and regional processes. – Science 235: 167 – 171. Rosenzweig, M. L. 1995. Species diversity in space and time. – Cambridge Univ. Press. Shmida, A. and Wilson, M. W. 1985. Biological determinants of species diversity. – J. Biogeogr. 12: 1 – 20. Stevens, G. C. 1989. The latitudinal gradient in geographic range: how so many species coexist in the tropics. – Am. Nat. 132: 240 – 256. Taylor, D., Aarssen, L. and Loehle, C. 1990. On the relationship between r/K selection and environmental carrying capacity: a new habitat templet for plant life history strategies. – Oikos 58: 239 – 250. Terborgh, J. 1973. On the notion of favorableness in plant ecology. – Am. Nat. 107: 481 – 501. Terborgh, J. 1977. Bird species diversity on an Andean elevational gradient. – Ecology 58: 1007 – 1019. Waide, R. B. et al. 1999. The relationship between productivity and species richness. – Annu. Rev. Ecol. Syst. 30: 257 – 300. Whittaker, R. H. 1960. Vegetation of the Siskiyou Mountains, Oregon and California. – Ecol. Monogr. 30: 279 – 338. Whittaker, R. J., Willis, K. J. and Field, R. 2001. Scale and species richness: towards a general hierarchical theory of species diversity. – J. Biogeogr. 28: 453 – 470. Willig, M. R. and Lyons, S. K. 1998. An analytical model of latitudinal gradients of species richness with an empirical test for marsupials and bats in the New World. – Oikos 81: 93 – 98.

ECOGRAPHY 26:3 (2003)


ECOGRAPHY 28: 209 /222, 2005

The elevational gradient in Andean bird species richness at the local scale: a foothill peak and a high-elevation plateau Sebastian K. Herzog, Michael Kessler and Kerstin Bach

Herzog, S. K., Kessler, M. and Bach, K. 2005. The elevational gradient in Andean bird species richness at the local scale: a foothill peak and a high-elevation plateau. / Ecography 28: 209 /222. A monotonic decline in species richness with increasing elevation has often been considered a general pattern, but recent evidence suggests that the dominant pattern is hump-shaped with maximum richness occurring at some mid-elevation point. To analyse the relationship between species richness and elevation at a local scale we surveyed birds from lowlands to timberline in the Bolivian Andes. We divided the transect into 12 elevational belts of 250 m and standardized species richness in each belt with both individual- and sample-based rarefaction and richness estimation. The empirical data were then correlated to four explanatory variables: 1) area per elevational belt, 2) elevation (also representing ecosystem productivity), 3) a middomain effect (MDE) null model of geometrically constrained empirical range sizes, and 4) a hump-shaped model derived empirically for South American birds representing the regional species pool hypothesis. Local species richness peaked at ca 1000 m elevation, declined sharply to ca 1750 m, and then remained roughly constant. Elevation was the best single predictor, accounting for 78 /85% of the variance in the empirical data. A multiple regression model with elevation, area, and MDE explained 85 /90% of the variance. Monte Carlo simulations showed that the richness peak at 1000 m is the result of an overlap of two distinct avifaunas (lowland and highland) and that the correlation to MDE in the multiple regression was likely spurious. We recommend complementing correlation analyses involving MDE predictions with an examination of the distribution of range midpoints. The steep decline at mid-elevations was mainly due to a rapid loss of lowland species. The high-elevation plateau is striking and unexpected, but has also been found previously. It cannot be explained at present and exemplifies that despite several decades of research elevational gradients are still not well understood. S. K. Herzog (skherzog@compuserve.com), Inst. fu¨r Vogelforschung ‘‘Vogelwarte Helgoland’’, An der Vogelwarte 21, D-26386 Wilhelmshaven, Germany. / M. Kessler and K. Bach, Albrecht-von-Haller-Inst. for Plant Sciences, Systematic Botany, Untere Karspu ¨ le 2, D-37073 Go¨ttingen, Germany.

The relationship between species richness and elevation has received considerable attention in the ecological literature. Equivalent to the latitudinal gradient in species richness, a monotonic decline in the number of species with increasing elevation has often been considered a general pattern (Brown and Gibson 1983, Begon et al. 1990, Rohde 1992, Stevens 1992).

However, Rahbek (1995) argued that this generalization is largely a result of too much emphasis on, and misinterpretation of, a few early studies (Terborgh 1977) combined with ‘‘citation inbreeding’’. In an extensive literature review including studies on a wide range of taxonomic groups, biomes, and spatial scales, Rahbek (1995) found empirical support for several

Accepted 17 November 2004 Copyright # ECOGRAPHY 2005 ISSN 0906-7590 ECOGRAPHY 28:2 (2005)

209


different elevational richness patterns, with a dominance of hump-shaped relationships where maximum species richness occurs at some mid-elevation point. However, Rahbek (1995) also noted that most studies had methodological problems because they did not account for the effect of sampling effort and/or area on patterns of species richness. Failure to standardize data to account for sampling effort (Gotelli and Colwell 2001) and area (Rahbek 1995, 1997) can cause artefactual results. To overcome such pitfalls, Rahbek (1997) analysed a data set largely devoid of sampling biases on the elevational distribution of neotropical land birds. After controlling for the surface area of each elevational belt, the emerging regional-scale relationship between species richness and elevation was indeed hump-shaped with a maximum at ca 1000 m. Stotz et al. (1996) approached the problem from a different angle. As not only surface area but also habitat diversity is greater in lowland Amazonia than on adjacent Andean slopes, Stotz et al. (1996: 37) only considered those species among lowland birds that inhabit terra firme forest and obtained a similar result as Rahbek (1997), namely a hump-shaped specieselevation curve with a maximum in the foothill zone between 500 and 1000 m. Thus, both analyses indicate that the enormous bird species richness of the Amazonian lowland is largely a consequence of its huge surface area and increased habitat diversity, and that if one controls for those factors, the lower Andean slope emerges as the zone of highest bird species richness. Recent work on other groups of organisms also documented hump-shaped relationships between species richness and elevation, e.g. for trees in Costa Rica (Lieberman et al. 1996), for several plant groups in Bolivia (Kessler 2000c, 2001b), for ants in the western United States (Sanders 2002), and for non-volant small mammals in the Philippines (Heaney 2001), Malaysia (Md. Nor 2001), and Costa Rica (McCain 2004). However, several of those studies interpolated species presence between maximum and minimum observed elevations, which may cause an artificial hump in the species-elevation curve (Grytnes and Vetaas 2002). By contrast, Patterson et al. (1998) found a trough-shaped, but non-significant, relationship for murid rodents in Peru with richness maxima in the lowlands and in the highlands and a minimum at mid-elevations. For bats and birds, Patterson et al. (1998) reported a smooth, slightly curvilinear decline of richness with elevation, but their data were not standardized for area or habitat diversity. Patterson et al. (1998) remarked, however, that preliminary analyses of the same bird data set revealed a hump-shaped pattern once the effect of increased Amazonian habitat diversity was removed. These studies seemingly illustrate that mountainsides do not simply mirror the latitudinal diversity gradient (Brown 2001) and that there is no universal agreement in 210

the shape of the elevational pattern. Brown (2001) further argued that synthetic theories explaining the common patterns of species richness on elevational gradients have not been developed yet. Recent advances in null model theory, however, have added a new twist to an old story. Colwell and Hurtt (1994) proposed several models that simulate range size and randomise range placement within one-dimensional bounded geographical domains (e.g. an elevational gradient). Based entirely on stochastic processes these null models produce symmetrical curves with a mid-domain peak in species richness. These simulations are the ancestor of analytical null models (Willig and Lyons 1998, Lees et al. 1999, Jetz and Rahbek 2001) and analytical-stochastic null models (Colwell 2000, applied by Sanders 2002, McCain 2003, 2004) that have been tested against empirical data and extended to two-dimensional domains (Jetz and Rahbek 2001). These studies provide strong support for what has come to be called the mid-domain effect (MDE; Colwell and Lees 2000, Colwell et al. 2004). By contrast, other authors (Bokma and Mo¨nkko¨nen 2000, Koleff and Gaston 2001, Bokma et al. 2001, Hawkins and Diniz-Filho 2002, Diniz-Filho et al. 2002, Zapata et al. 2003) have dismissed a significant influence of MDE on spatial richness patterns, criticizing MDE model assumptions as unrealistic, conceptually flawed, or internally inconsistent. Colwell et al. (2004) contend that much of the MDE criticism is based on misunderstandings of the nature of null models and of MDE models in particular, the use of inappropriate frequency distributions of geographical range sizes, and an all-ornothing approach searching for single-factor explanations. To properly evaluate the role of MDE in shaping species richness patterns Colwell et al. (2004) recommend that studies should a) quantitatively assess the relative importance of MDE instead of simply testing it as a null hypothesis, b) use re-sampling of the empirical range size frequency distribution (rather than theoretical range size frequency distributions or empirical range midpoints) for making MDE predictions, and c) apply multivariate analyses considering candidate explanations for richness patterns in addition to MDE. Of the 16 onedimensional MDE studies found by Colwell et al. (2004) in an exhaustive literature review only one (McCain 2004) meets all three criteria. In the present study, we follow the recommendations of Colwell et al. (2004) in analysing bird survey data from a perhumid elevational gradient from lowlands to timberline in central Bolivia. Lowland surveys were restricted to terra firme forest, and data analyses controlled for survey effort. Using multiple regression analysis we correlated the resulting pattern of species richness with the following explanatory variables: 1) area, 2) elevation (also representing potential evapotranspiration and mean canopy tree height), 3) MDE using the empirical version of the Colwell and Lees ECOGRAPHY 28:2 (2005)


(2000) constrained range-size null model, and 4) the regional species pool using Rahbek’s (1997) humpshaped model for South American birds.

Study area We surveyed birds in and adjacent to Carrasco National Park on the east-Andean slope in the central Bolivian department of Cochabamba (Fig. 1). The area spans an elevational gradient from 200 to 4800 m. Its natural vegetation consists of evergreen, perhumid forest that once extended up to elevations of ca 4200 m. Due to centuries of human impact in the high Andes, the current timberline is lowered and generally found around 3400 m, but remnant forest patches remain at higher elevations (Kessler and Herzog 1998, Kessler 1999, 2000b). Foothill and lowland forests have been cleared extensively in recent decades, primarily for timber extraction, coca plantations, and road construction (Henkel 1995). Data on the vegetation of the study transect can be found in Ibisch (1996), Navarro (1997), and Kessler (2000a, 2001b). We studied a continuous elevational gradient at 950 / 3400 m in the Serranı´a de Callejas (Fig. 1) using a ca 4-m wide gravel road abandoned in the 1980s. The adjacent forest was almost entirely in a natural state; few disturbed areas largely had reverted to montane forest, and the road partly had been invaded by woody plants. Human activity in the Serranı´a de Callejas increased drastically below 1000 m, and we collected data in two foothill area’s. Between the villages of El Palmar and Villa Tunari (Fig. 1; 300 /900 m) we selected several sites that contained some of the area’s best-preserved primary or mature secondary forest. The second area at Rı´o Ichoa-Cerro Len˜e (Fig. 1; 300 /750 m) contained pristine forest and was reached by helicopter. A seismic line (a straight trail just wide enough for one person) established several weeks before represented the only visible anthropogenic disturbance; no timber extraction or hunting had occurred. We surveyed lowland forest in the Valle del Sacta (Fig. 1; 220 m), a 5600-ha area of

Fig. 1. Location of Carrasco National Park in Bolivia and distribution of survey sites within the study area. ECOGRAPHY 28:2 (2005)

primary and mature secondary terra firme forest. This area is separated from Andean forests by a ca 10-km wide belt of human settlements. It is located in the southernmost extension of Amazonian evergreen lowland forest, which forms a 10 /30-km wide wedge at the base of the central Bolivian Andes. Further to the northeast evergreen forest is replaced by seasonally flooded savannah vegetation (Ribera et al. 1996). Geologically, the bedrock mainly consists of sandstones, lutites, and quartzitic rocks of Devonian and Ordovician age (Montes de Oca 1997, Ergueta and Go´mez 1997) with limited areas of granitic intrusives and calcareous rocks at 2000 /2200 m along the study road and some white sand areas at 450 /500 m. At Villa Tunari the geological substrate is composed of nutrientpoor white sandstones, whereas it consists of clayey red alluvial deposits in flat areas and more sandy soils on ridges in the Valle del Sacta. Mean annual precipitation at the region’s only reliable climatic station in Villa Tunari is 5676 mm (10 yr data; Ibisch 1996). Most precipitation falls from November to May, but even from June to October every month receives /100 mm. Ibisch (1996) estimated /3500 mm mean annual precipitation at 2200 m in the somewhat sheltered valley of Sehuencas (Fig. 1). Mid-elevation (1000 /3000 m) slopes directly exposed to incoming clouds in the northwest of the park likely receive /8000 mm (Kessler 1999). Mean annual precipitation at Sacta is /3000 mm (Acebey pers. comm.), with additional moisture provided by frequent morning fog. Mean annual temperatures are 24.68C at Villa Tunari and 12 /158C at 2200 m in Sehuencas (Ibisch 1996), with an annual variability of monthly means of ca 58C. Nocturnal frosts occur down to 2000 m (Ibisch 1996, unpubl.), especially during periodic influxes of southern polar winds during the austral winter (Fjeldsa˚ et al. 1999).

Methods Field surveys We studied the continuous Serranı´a de Callejas transect (950 /3400 m) from June to September 1996 and sections of it (950 /1500 m, 3000 /3400 m) again in October 1997. The El Palmar-Villa Tunari area (300 /900 m) and the Valle del Sacta (220 m) were studied in September and October 1996, and Rı´o Ichoa-Cerro Len˜e (300 /750 m) in September 1997. No surveys were conducted above timberline (3400 m). To factor out the effect of increased lowland habitat diversity on species richness, lowland field work was restricted to terra firme forest. The survey method is detailed in Herzog et al. (2002) and only briefly summarized here. While walking slowly and quietly from dawn to mid-day and often again from late afternoon to after dusk along roads, trails and 211


212

55 65 64 59 70 69 53 61 58 55 64 62 62 82 73 63 74 74 b

a

Number of 10-species lists (samples) compiled. Mean number of individuals per 10-species list (sample)9/SD. c Observed total species richness. d MMMean estimate of species richness after the maximum number of samples. e Sample-based MMMean estimate of species richness standardized for survey effort. f Observed species richness after 337 individuals. g Individual-based MMMean estimate after maximum number of individuals. h Individual-based MMMean estimate after 337 individuals.

82 121 109 103 162 146 100 139 139

85 125 111

72 68 82 78 71 60 76 66 88 71 84 79 129 120 174 158 149 149

132 119

20 18.59/6.0 369 56 20 18.59/6.8 369 61 21 19.99/6.5 417 57 20 20.59/8.2 409 58 36 17.69/6.4 633 76 20 16.99/6.9 337 63 49 15.69/4.9 765 106 56 14.79/3.9 821 110 51 14.19/3.2 719 133 27 12.59/1.9 337 100

Lists 45 80 12.69/2.7 14.79/14.0 Individuals per listb Total individuals 569 1175 Scobs 119 141 Sample-based rarefaction d 153 161 MMMean max MMMean stde 141 136 Individual-based rarefaction Sobs stdf 98 96 146 156 MMMean maxg 137 132 MMMean stdh

3000 /3249 1000 /1249 1250 /1499 1500 /1749 1750 /1999 2000 /2249 2250 /2499 2500 /2749 2750 /2999 500 /749 300-499 220

We divided the transect into 14 elevational belts of 250 m (B/250 m, 250 /499 m, 500 /749 m, etc.; Table 1). We restricted surveys to forest areas, but included data from small areas of natural (e.g. landslides) or anthropogenic (e.g. small coca fields) disturbance. Species occurring naturally at forest edge were included in all analyses; species depending on aquatic habitats were excluded. For analysing the relationship between species richness and elevation, data from the El Palmar-Villa Tunari area were excluded for two reasons (but they were included in the analysis of range midpoints). First, human disturbance probably introduced biases that we were unable to control for. Second, individual survey sites with suitable habitat were spaced relatively far apart and measured species richness included a strong component of beta diversity (sensu Whittaker 1972), rather than alpha diversity as on the remaining transect. Thus, the Rı´o Ichoa-Cerro Len˜e data were used for the 300 / 499-m and 500 /749-m belts, and no data were available for the 750 /999-m belt. Although surveys extended to 3400 m, the accessible area in the 3250 /3499-m belt was too small for meaningful data analysis and it was excluded. For the Serranı´a de Callejas only data collected in 1996 were used in this step of the analysis. Due to topographic complexity and variations in the amount of accessible habitat, it was not feasible to standardize survey area across all elevational belts. Survey distance (length of transect line) varied from 3.9 km (2750 /2999 m) to 10.3 km (300 /499 m) with a mean of 5.6 km.

a

Transect subdivision

Table 1. Observed and estimated species richness values of elevational belts studied along the Carrasco transect on the east Andean slope in central Bolivia. Only those belts included in the analysis of the relationship between species richness and elevation are shown. Elevation in m.

through the habitat where feasible, SKH continuously recorded all visual and acoustical observations of birds (including numbers of individuals per species) within 50 m of the observer. The observer’s movement rate largely depended on the level of bird activity. When spending longer periods in one spot and during very occasional resampling of an area (the latter occurred to approximately the same degree in all elevational belts and thus did not introduce a systematic error), repeated counts of obviously territorial individuals were avoided. Tape recordings were made extensively to supplement observations and to identify unknown voices, and were integrated into the master list of temporally consecutive observations. Fjeldsa˚ (1999) quantitatively compared this approach with standardized point counts. Its main advantages compared to any timed species-count method (e.g. point counts) are time efficiency and relative observer independence (Fjeldsa˚ 1999, Herzog et al. 2002).

ECOGRAPHY 28:2 (2005)


175

Standardisation of survey effort for species richness estimation

ECOGRAPHY 28:2 (2005)

Cumulative species richness

32.8%

125 100 75

67.2%

50 25 0

0

5

10

15

20

25

30

35

40

100

B Cumulative species richness

Survey effort was not exhaustive and varied between elevational belts, precluding the use of raw species counts in our analyses. Therefore, we used three methods to standardize species richness values for survey effort (Table 1). First, we used a modified version of the ‘‘m-species-list method’’ (MacKinnon and Phillipps 1993, Poulsen et al. 1997) following the recommendations of Herzog et al. (2002) (method 1). We divided the master list of temporally consecutive bird observations in each 250-m belt into lists of 10 species: the first list consists of the first 10 species observed, the second list includes the following 10 species and may contain species already found on the first list, and so on. We then plotted cumulative species number as a function of list number, treating each 10-species list as a separate sample. By randomising sample accumulation order 50 times using EstimateS 5.0.1 (Colwell 1997), we obtained samplebased rarefaction curves and estimated total species richness in each belt with the MMMean statistic (Fig. 2; Raaijmakers 1987, Keating and Quinn 1998). For species-rich neotropical bird data sets, MMMean was the least biased of nine estimators evaluated by Herzog et al. (2002), but a drawback of MMMean is that no statistically sensible variance estimator exists (Colwell and Coddington 1994, Colwell pers. comm.). Ideally, the MMMean curve for each elevational belt would quickly reach an asymptote after 10 /15 lists. As this was not the case, we standardized all data sets for survey effort following the procedure in Herzog et al. (2002) (Fig. 2). Gotelli and Colwell (2001) suggested that x-axes of sample-based rarefaction curves should be rescaled from samples to individuals because datasets may differ systematically in the number of individuals per sample. Thus, our second standardisation method (method 2) is derived from individual-based rarefaction. Accumulation order of individuals was randomised 50 times using EstimateS 5.0.1 (Colwell 1997), and we used the observed species richness after 337 individuals (the lowest number of individuals recorded in any elevational belt; Table 1) as the standardized cut-off point. In the third method, we determined MMMean values of estimated species richness after 337 individuals from the same individual-based rarefaction curves using EstimateS 5.0.1 (Colwell 1997) (method 3). Individual-based rarefaction probably is the least biased method for comparing species richness via species-accumulation curves (Gotelli and Colwell 2001), but it is not without problems. In cases with strongly differing sampling intensities, such as here (Table 1), it implies a loss of much valuable data. It also is sensitive to biases in the quantification of the number of individuals per species. Such biases are

A

100% 150

100%

80

32.8%

60

40

67.2% 20

0

0

5

10

15

20

25

30

35

40

Number of 10-species lists

Fig. 2. Sample-based rarefaction curves for bird data sets from two elevational belts (A: 500 /749 m; B: 2000 /2249 m) in Carrasco National Park, Bolivia. Observed (Sobs; circles) and estimated species richness using the MMMean statistic (squares) are expressed as a function of the number of 10species lists. Sample accumulation order of all curves was randomized 50 times, and each point represents the mean of the resulting 50 values. To control for the confounding effects of survey effort, we determined a standardized cut-off point from the relation between the Sobs and the MMMean curve: for each data set, every Sobs value was expressed as the proportion of the respective MMMean value (Herzog et al. 2002). Survey effort was lowest in the 500 /749-m belt (A), where Sobs comprised 67.2% of the estimated richness (149 species) at maximum sample size (27 10-species lists). The equivalent cut-off point was determined for all other data sets as illustrated for the 2000 /2249-m belt (B), where a standardized estimate of 71 species was obtained.

inevitable when studying birds in tropical forests, where many birds are only seen briefly and where many species move in large mixed-species flocks. Also, the presence (or absence) of a few large single-species flocks of some species, e.g. swifts or parrots, can strongly influence total individual numbers and thereby the slope of the accumulation curve, artificially inflating survey effort. Finally, whereas species richness values obtained by individual-based rarefaction may accurately reflect relative differences in species richness between sites, they almost invariably underestimate true species richness and usually are lower than the total observed species numbers. 213


Species richness estimation based on samples such as m-species lists is less sensitive to these problems by including all or most of the collected data, by counting a species only once when two or more individuals are registered together, and by providing species richness values somewhat above the observed values. However, this method also has three important weaknesses (Gotelli and Colwell 2001). First, different estimation methods provide different results, and although there is some empirical data to suggest which statistic might be suitable for a given data set, no general test exists to choose the least biased estimator (of course this also applies to individual-based richness estimation). In general, the response of estimators to differences in species abundance distributions, species richness, etc. is poorly understood. Second, the mean number of individuals per list tends to be higher in species-poor habitats, simply because more birds have to be recorded to accumulate 10 (or five, or 20) species. This leads to relatively steeper accumulation curves in species-poor habitats, and hence an overestimation of species richness relative to species-rich habitats. Finally, and perhaps most importantly, the arbitrary decision on how many species to include in each list will influence the estimated values. High species numbers per list will reduce the number of lists and lead to relatively steeper accumulation curves and higher estimates (Herzog et al. 2002). All three error sources will be stronger in cases with more pronounced differences in actual species richness between sites. Given the potential biases of both approaches for obtaining comparable, standardised species richness values, here we chose to apply both sample- and individual-based rarefaction. As the elevational richness patterns obtained by both approaches are highly correlated (Spearman rank correlation, values of standardization method 1 vs method 2: r /0.98, pB/0.0001; method 1 vs 3: r /0.99, pB/0.0001; method 2 vs 3: r /1.00, p/ 0), we conclude that our raw data are fairly robust to standardization, and that both standardization approaches accurately reflect the richness pattern captured by our field data. On the other hand, since the raw species counts show lower correlation values with the standardized data (Spearman rank correlation, raw data vs method 1: r/0.85; raw data vs method 2: r /0.88; raw data vs method 3: r /0.87; pB/0.001 in all cases), we conclude that standardization is appropriate.

Explanatory variables Area / due to the conical shape of mountains, land surface area decreases steadily with increasing elevation (Graves 1988, Rahbek 1997). The effect of the area of elevational belts on species richness at the regional scale was illustrated by Rahbek (1997), but the procedures 214

used therein to control for area are not applicable to our local-scale data set. Rahbek (1997) also used belts of greater amplitudes than analysed here. As a proxy for area we measured the horizontal width of each 250-m belt in central Bolivia on topographic maps (scale 1:50 000) issued by the Inst. Geogra´fico Militar, La Paz, Bolivia. For the lowland belt we measured only the area covered by evergreen terra firme forest as surveys were restricted to this habitat, excluding open savannah further east. We took five measurements for each belt (one in the study area, two to the north, and two to the south at 10 km intervals), and the mean was calculated and rounded to the closest 100 m (Table 2). As species richness does not increase linearly with area, we multiplied the mean horizontal width of each belt by an areadependent factor, assuming a slope of z /0.13 in a double-log species-area plot, which corresponds to the mean z value obtained for birds in the tropical Andes by Rahbek (1997: Fig. 2c). Elevation / this represents the model of a monotonic decline of species richness with elevation (Stevens 1992). We used the mean elevation of each belt, except for the lowland belt, where we used the actual elevation (Table 2). Ecosystem productivity / Kessler (2001b) determined potential evapotranspiration (PET) calculated after Thornthwaite and Mather (1957) and mean canopy tree height in mature forest as indices of ecosystem productivity (Rosenzweig 1968, Lieth 1975) for our study area. Within the boundaries of the present gradient, both PET (Pearson correlation: r2 /1.00, p/0) and mean canopy tree height (Pearson correlation: r2 /0.95, p B/0.0001) closely correlate negatively with elevation. Hence, only elevation is used in the regression analysis. MDE / we used the richness values predicted by the empirical version of the Colwell and Lees (2000: Box 5) constrained range-size null model, shuffling empirical range size distributions by random mid-point Monte Carlo simulations using RangeModel 3.1 (Colwell 2000: Model 4). Upper and lower domain limits were defined as the natural upper limit of humid forest on the eastern slope of the central Andes (ca 4200 m) and the lower limit of Amazonian evergreen forest (sea level), respectively. Only those species whose elevational distribution throughout the neotropics falls entirely within these domain limits were included in the model. Upper and lower elevational range limits were taken from standard references on neotropical birds (Ridgely and Tudor 1989, 1994, FjeldsaË&#x161; and Krabbe 1990, Hoyo et al. 1992 /2002, Stotz et al. 1996, Isler and Isler 1999, Ridgely and Greenfield 2001, Hennessey et al. 2003). Unlike Lees et al. (1999), who defined their domain as the present day extent of forest cover on Madagascar, we did not use the present elevation of the closed upper timberline (ca 3400 m) as the upper domain limit because remnant forest patches of varying sizes still ECOGRAPHY 28:2 (2005)


c

b

a

Horizontal width in km determined from topographical maps (mean of five measurements rounded to the closest decimal). Mean elevation of each belt, representing both elevation per se as well as PET and mean canopy tree height as indices of ecosystem productivity. Values predicted by the empirical version of the Colwell and Lees (2000: Box 5) and Colwell (2000: Model 4) constrained range-size MDE null model for 548 species. e Values predicted by Rahbek’s (1997) regional hump-shaped model for South American birds.

0.4 3125 268 351 0.5 2875 318 407 0.5 2625 353 463 0.5 2375 376 512 0.6 2125 387 561 0.6 1875 384 606 0.7 1625 364 638 0.7 1375 334 666 0.8 1125 288 682 1.6 625 174 662 3.0 400 116 630

3000 /3249 2750 /2999 2500 /2749 2250 /2499 2000 /2249 1750 /1999 1500 /1749 1250 /1499 1000 /1294 500 /749 300 /499 220

15.0 220 70 589 Area Elevationb MDEc Species poole

a

Table 2. Values of the explanatory variables for each elevational belt included in the analysis of the relationship between species richness and elevation. Elevation in m.

ECOGRAPHY 28:2 (2005)

exist up to elevations of 4200 m in the study area, in part only a few kilometers from our actual study sites (see, e.g. photos in Kessler 2000b). While these patches were not accessible to us, they certainly provide suitable habitat for the local forest-based avifauna, and the upper elevational distribution of the bird species is consequently not limited to the forest line at 3400 m along the study road. Extensive ornithological surveys in Carrasco National Park and adjacent lowland forests in recent years by a number of field workers (especially R. MacLeod, J. Fjeldsa˚, and collaborators) have resulted in a nearly complete inventory of the area’s avifauna (Herzog et al. unpubl.). With the exception of species depending primarily on aquatic habitats and lowland species not found in terra firme forest, all recorded species that inhabit humid forest were included in the null model computations. An additional 19 species that have gone unrecorded so far but that are expected to occur in the area also were included, resulting in a total of 548 species. Monte Carlo simulations were replicated 20 times and mean richness values were computed for 60 domain divisions (the default of RangeModel 3.1). Predicted values for each 250-m belt were taken from the resulting curve (Table 2). As some of the MDE debate has focused on whether to use theoretical or empirical range size frequency distributions for MDE predictions (Colwell et al. 2004), we also computed the bivariate uniform model of Colwell and Lees (2000: Box 2), which is equivalent to the Colwell and Hurtt (1994) null model 2 of bounded random geographical ranges, for 548 species using RangeModel 3.1 (Colwell 2000: Model 1). This theoretical model takes no account of the empirical range size frequency distribution. As above, Monte Carlo simulations were replicated 20 times and mean richness values were computed for 60 domain divisions. As both null models were highly correlated (Pearson correlation: r2 / 0.99, pB/0.0001; however, the constrained range size model predicted higher absolute values than the bivariate uniform model) only the constrained range size model is used in the analysis. Regional species pool / the regional species pool has been proposed as a significant determinant of local species richness as it places an upper limit on the number of species potentially able to colonise local habitats (Cornell and Lawton 1992, Caley and Schluter 1997). We considered Rahbek’s (1997) regional hump-shaped model for South American birds as the most appropriate regional source pool model and used the richness values predicted by this model (Fig. 4 in Rahbek 1997; values taken from the fitted curve) for each of our elevational belts (Table 2). It may be argued that this model is incorrect to some degree as it is based on partly interpolated species-distribution data (see Grytnes and Vetaas 2002). However, at the broad spatial scale 215


considered by Rahbek (1997) we regard the data on South American bird distributions as sufficiently complete so as to render the influence of interpolation on the species-elevation curve negligible.

Multiple regression analysis We first performed bivariate linear regressions of the empirical species richness pattern (separately for each of the three standardization methods) against each of the four explanatory variables individually, followed by multiple regression analysis. Due to the spatial proximity of most elevational belts, estimates of species richness violate statistical assumptions of independence, and the present analysis therefore focuses on the proportion of variance explained rather than on probability values. All regression and correlation analyses were performed using STATISTICA for Windows (ver. 5.1, Anon. 1997).

Monte Carlo simulations: range midpoints and turnover To asses whether species are randomly distributed along the elevational gradient or show clear zonations, we conducted Monte Carlo simulations with a program written by KB in Visual Basic within EXCEL. Observed elevational ranges of all recorded species were randomly placed along the elevational gradient treating the gradientâ&#x20AC;&#x2122;s end points as hard boundaries not to be crossed by any speciesâ&#x20AC;&#x2122; range. Total species number per belt was constrained in two ways, according to the MMMean-standardized observed richness pattern, and to the richness pattern predicted by the MDE constrained range size model. Early simulation trials showed that not all species could be accommodated in the simulations using the observed standardized species richness. This was a result of tight species packing in the empirical data, whereas in the randomised data many gaps between species could not be filled because they were narrower than the species ranges remaining to be placed. Therefore, the observed standardized data was multiplied by the factor 1.5. The MDE model values were high enough that this correction was not necessary. Randomisations were repeated 1000 times, and we calculated the mean and 95% confidence intervals for three parameters (midpoints per belt, number of upper/ lower elevational limits per belt, species turnover between adjacent belts) explained below. These values were compared to the observed data, assuming that empirical and predicted values were significantly different if the observed value fell outside the 95% confidence intervals of the predicted value. Midpoints: as hump-shaped patterns that may correlate to MDE predictions can arise from a variety of 216

different processes, we compared not only the observed and predicted species richness patterns but also the empirical and predicted range midpoint distributions, counting the number of midpoints per belt (Grytnes 2003). Upper and lower limits: to asses whether species turnover along the gradient is homogeneous, we separately compared the number of upper and lower elevational limits per belt of empirical vs simulated data. Turnover: as a further test for species zonation, species turnover along the gradient was measured with the Wilson-Shmida index (Wilson and Shmida 1984). The index is calculated for pairs of adjacent elevational belts as follows: b(bc)=(2abc) where a is the number of species recorded in both belts, and b and c the number of species lost and gained, respectively. The more dissimilar two belts are, the higher is the index, reaching a maximum of 1 at total dissimilarity. The Wilson-Shmida index produces results comparable to those of other b-diversity indices (Davis et al. 1999, Koleff et al. 2003).

Results We recorded a total of 449 bird species, 38 of which were found only in the El Palmar-Villa Tunari area. Species richness peaked around 1000 m and slightly decreased towards the lowlands (Fig. 3, Table 1). Above 1250 m species richness decreased quickly to a minimum at 2500 m, followed by a slight increase towards the highest elevation. In general, however, variation above 1750 m was slight and richness remained surprisingly constant, generating a high-elevation plateau (Fig. 3, Table 1). Examining the congruence between empirical and model data (Fig. 3, Table 3), elevation (also representing ecosystem productivity) was the best predictor variable explaining 78 /85% of the variance in the empirical species richness pattern. The regional species pool was a moderate predictor of the empirical pattern, explaining 59 /63% of its variance. At low elevations it was largely in accordance with the empirical curve, but above 1250 m observed species richness decreased much faster than predicted by this model (Fig. 3D). Area had little explanatory power, concurring with the empirical pattern only in the lowlands (Fig. 3A). MDE predictions were negatively correlated to and strongly contrasted with the observed species richness pattern. The former reached its peak at elevations where the latter approached its minimum (Fig. 3C), and the maximum richness predicted by the MDE model is over twice as high as the empirical richness peak (Table 1). The best multiple regression model was elevation combined with area and MDE, explaining 85 /90% of ECOGRAPHY 28:2 (2005)


100

A

90

Relative species richness

Fig. 3. Comparison of the empirical richness pattern (circles; standardized for survey effort by MMMean estimates for samples) and the patterns predicted by the four explanatory models examined (squares): (A) area, (B) elevation, (C) mid-domain effect, and (D) regional species pool. Empirical data were not standardized for area since area was included as an explanatory model. Curves were fitted by distance-weighted least-squares smoothing.

100

80

80

70

70

60

60

50

50

40

40

30

30 20

20 10

B

90

0

500

1000

1500

2000

2500

100

3000

C

90

10

80

70

70

60

60

50

50

40

40

30

30

20

20 0

500

1000

1500

2000

2500

3000

500

1000

1500

2000

2500

10

3000

D

90

80

10

0

100

0

500

1000

1500

2000

2500

3000

Elevation (m)

the variance in the empirical species richness pattern (Table 3). The multiple regression of richness vs. regional species pool and area combined also obtained high determination coefficients of 0.72 /0.79 (Table 3). The proportion of observed range midpoints was significantly higher at 250 /750 m and 3000 /3250 m than predicted by the Monte Carlo simulations (Fig. 4). Thus, the empirical richness peak at ca 1000 m results at least partly from a large number of species restricted to low elevations. The significant surplus of elevational midpoints at 3000 /3250 m might be a sampling artefact as species mainly occurring at higher elevation and barely entering our study gradient, are falsely considered as having their mid-points in the highest survey belt.

Further, the results of the analyses of range midpoints may be biased by two shortcomings in our data set, i.e. the sampling gap at 750 /999 m and the fact that the area from 300 to 750 m was surveyed in a separate location. Both shortcomings may potentially inflate the number of midpoints at elevations of 300 /750 m because species actually reaching up to 750 /999 m may be misinterpreted as having their upper elevational limits at lower elevations, and because species found only at the separate survey location could not be recorded above 750 m. However, we doubt that these potential biases are a major problem, for two reasons. First, the survey data from the El Palmar-Villa Tunari area were included in determining the elevational ranges of species. As a result,

Table 3. Linear determination coefficients (r2) between observed species richness standardized by three methods and four explanatory model response variables. The upper four lines show the individual regression values, the lower six lines multiple regressions combining two and three parameters (starting with elevation as a fixed variable because it has the highest single explanatory power). Since Rahbekâ&#x20AC;&#x2122;s (1997) regional species pool model is corrected for area, we also performed a multiple regression combing species pool and area. Note the consistency in results between the three richness estimation methods. Sample-based standardized MMMean estimate Areaa Elevation (E)b MDEc Species poole E and area E and MDE E and species pool E and area and MDE E and area and species pool Species pool and area

0.37 0.78 0.43 0.59 0.81 0.78 0.79 0.85 0.82 0.72

Sobs after 337 individuals 0.43 0.85 0.46 0.63 0.87 0.85 0.86 0.90 0.88 0.79

Individual-based standardized MMMean estimate 0.43 0.84 0.47 0.60 0.85 0.84 0.84 0.88 0.86 0.77

a

Horizontal width in km determined from topographical maps (mean of five measurements rounded to the closest decimal). Mean elevation of each belt, representing both elevation per se as well as PET and mean canopy tree height as indices of ecosystem productivity. c Empirical version of the Colwell and Lees (2000: Box 5) and Colwell (2000: Model 4) constrained range-size MDE null model for 548 species. e Rahbekâ&#x20AC;&#x2122;s (1997) regional hump-shaped model for South American birds. b

ECOGRAPHY 28:2 (2005)

217


100

50

A

80

Number of species

Percentage of species

60

40

30

20

60

40

20

10

0

0 0

500

1000

1500

2000

2500

0

3000

Elevation (m)

empirical range midpoints are affected only by an elevational gap of 50 m (900 /950 m), and it is unlikely that such a negligible a gap will inflate the number of range midpoints below it. Second, with regard to the separate location of the area from 300 to 750 m, of the 127 species that had their midpoints within this elevational range only 16 (13%) were exclusive to that area. For these 16 species, the mean upper elevational limit in Bolivia (based on Hennessey et al. 2003) is 1440 m. This implies that their elevational midpoints would be located at ca 700 m elevation. As a result, even if surveys at the separate site had extended to higher elevation and recorded the full elevational ranges of these species, their range midpoints would still be located well below the elevation predicted by the MDE model. Regarding species zonation, significantly more lower limits than expected by chance occurred at 220 /500 m, 1000 /1250 m, and 3000 /3250 m, and significantly fewer lower limits at 500 /750 m (Fig. 5A). Upper elevational limits showed significantly higher values at 500 /750 and 1500 /1750 m and significantly lower values at 250 /500 m, 1750 /2000 m, and above 2500 m (Fig. 5B). Turnover was significantly higher than expected from the Monte Carlo simulations at 750 /1250, 1750 and 2500 m, and lower at 200 /500, 2000, and 3000 m (Fig. 6). Taken together, these analyses show that the change of species composition along the elevational gradient is not smooth. Rather, areas of rapid change and others of limited turnover exist. However, the location of peaks of upper and lower elevational species limits and of peaks in species turnover are not concordant and it is difficult to point out distinct elevational zones. 218

1000

1500

2000

2500

3000

B 50

Number of species

Fig. 4. Proportion of empirical range midpoints (open circles) and those predicted by Monte Carlo simulations restricting species richness per elevational belt according to the empirical richness pattern (closed circles) and MDE model predictions (squares). Vertical bars indicate 95% confidence intervals of the predicted values based on 1000 randomizations.

500

60

40

30

20

10

0 0

500

1000

1500

2000

2500

3000

Elevation (m)

Fig. 5. Number of lower (A) and upper (B) elevational range limits observed (open circles) and predicted by Monte Carlo simulations restricting species richness per elevational belt according to the empirical richness pattern (closed circles). Vertical bars indicate 95% confidence intervals of the predicted values based on 1000 randomizations.

Discussion The elevational pattern of bird species richness at the local scale documented here shows three distinct zones: a slight increase from the lowlands to a maximum at ca 1000 m, a sharp decline at 1250 /1750 m, and roughly constant values up to the highest survey elevation at 3250 m. Of course, at even higher elevations at and above timberline species richness will decrease. Among the four models examined, this pattern best fits the elevation model, although there are noticeable discrepancies. Neither the slightly lower values in the lowlands, nor the sharp decline at mid-elevations, nor the richness plateau at high elevations are adequately predicted by this model. In the multiple regression model, these divergences are partly accounted for by the additional factors included. Area predicts fairly constant and high values at higher elevations, thus modelling the highelevation plateau. The MDE model, by contrast, predicts a mid-elevation peak and, although negatively correlated with the observed pattern when compared directly, in the multiple regression model, it explains the ECOGRAPHY 28:2 (2005)


0.4

Tur nover

0.3

0.2

0.1

0.0 0

500

1000

1500

2000

2500

3000

Elevation (m)

Fig. 6. Species turnover between adjacent elevational belts calculated with the Wilson-Shmida index (Wilson and Shmida 1984) for the empirical data (open circles) and predicted by Monte Carlo simulations restricting species richness per elevational belt according to the empirical richness pattern (closed circles). Vertical bars indicate 95% confidence intervals of the predicted values based on 1000 randomizations.

peak in the foothills and the decrease of species richness towards the lowlands. The rather good correlation between the observed richness pattern and elevation indicates that local ecosystem properties may be involved in shaping the pattern. However, for two reasons we are unable to pinpoint which local factors might be responsible. First, most local factors co-vary closely along the gradient and cannot be disentangled. Potential evapotranspiration and canopy height decline almost linearly with elevation (see Methods). Other factors not quantified by us, e.g. insect abundance and habitat complexity, very likely decline in a similar fashion with elevation. Which of these factors or factor combinations are ultimately the most important can only be determined by comparing different transects where these variables show different distributions. Second, there is no consensus about the general relationship between richness and productivity. Both monotonic and hump-shaped relationships have been documented (Waide et al. 1999, Mittelbach et al. 2001). As yet, no comprehensive theoretical framework exists for the mechanisms involved in productivityrichness relationships (Currie et al. 1999, Ricklefs 2004). Regardless of the causes for the correlation between the empirical pattern and elevation, we are still faced with explaining the three deviations outlined above, i.e. the foothill hump, the sharp mid-elevation decline, and the high-elevation plateau. Among the models considered here, the foothill peak is best predicted by the regional species pool model. Rahbek (1997) concluded that the regional skewed mid-elevation peak is best explained by geometric boundary constraints, i.e. MDE, whereas the details of the pattern are governed by other factors. However, our Monte Carlo simulations suggest that the contribution of MDE to the multiple ECOGRAPHY 28:2 (2005)

regression model is likely spurious. If this model were to explain the foothill richness peak, we would also expect a peak of elevational midpoints at the same elevation. This simply is the result of the nature of the mid-domain null model. Species are randomly placed along the gradient, accumulating at mid-elevations because species with large elevational amplitudes must occur in the middle of the gradient (Colwell and Lees 2000, Colwell et al. 2004). As shown by Laurie and Silander (2002), Grytnes (2003), and our simulations (Fig. 4), this leads to a slight mid-domain peak of range midpoints. By contrast, in our empirical data midpoints are accumulated at low elevations. In combination with the zonation simulations (Figs 5 and 6), this can be interpreted as reflecting a prominent low-elevation avifauna. This avifauna can be detected by the high number of lower elevational limits at 220 /500 m, the surplus of midpoints at 250 /750 m, and the high number of upper elevational limits at 500 / 1250 m. Thus, the peak of species richness at 1000 m apparently results from an overlap of a distinct lowland with a distinct highland avifauna, and coincides with the significant peak of species turnover at 750 /1250 m. When assigning all recorded species to mutually exclusive categories of lowland species (present below 300 m) and highland species restricted to elevations above 300 m, this overlap and richness accumulation becomes apparent (Fig. 7). Although this seems trivial at first sight, there is no a priori reason to assume that the overlap between the two regional species pools leads to a hump-shaped richness distribution. It is equally conceivable that the decline of lowland species is sufficiently strong and/or the increase of highland species so weak that the resulting overall richness curve peaks in the lowlands. The sharp decline of species richness at 1250 /1750 m is largely caused by the rapid loss of lowland species. This decline is evidenced in the large number of upper elevational limits at this elevation (Fig. 5B) and causes high species turnover at 1750 m (Fig. 6). The highelevation plateau at 1750 /3250 m is puzzling, although we are not the first to document this pattern. Local bird species richness in Manu National Park of south-central Peru (Patterson et al. 1998: Fig. 1a) shows a similar plateau with almost identical richness values, but this phenomenon is not discussed. For geometrid moths in southeast Ecuador, Brehm et al. (2003) also found roughly constant values at 1800 /2700 m. A highelevation richness plateau is not predicted by Rahbekâ&#x20AC;&#x2122;s (1997) hump-shaped model or by the geometric null model, and strongly disagrees with the productivity hypothesis. Currently we have no plausible explanation for it. Although area apparently accounts for the plateau to a certain degree in our multiple regression model, we cannot conceive any causal relationship. Comparing our results with those of other avian transect studies in the Neotropics, we find substantial 219


200

All species Montane species Lowland species

Sp ec i es r i c h n es s

160

120

80

40

0

0

500

1000

1500

2000

2500

3000

3500

Elevation (m)

Fig. 7. Changes in species composition along an elevational gradient on the east Andean slope in central Bolivia. Montane species were defined as restricted to areas above 300 m and absent from the lowlands, whereas lowland species may or may not occur in montane areas. Elevational ranges of species were interpolated to all elevations between the recorded minimum and maximum of each species. Since interpolation added no species to the lowest and highest elevational belt, values for these belts are negatively biased, whereas a sampling bias accounts for the minimum at 750 /999 m. Note the nearly constant number of montane species between 1750 and 3250 m.

similarities. When removing the effect of increased lowland habitat diversity from their Manu data set, Patterson et al. (1998) found that species richness peaked around 1000 m rather than in the lowlands, i.e. exactly at the same elevation as in our study. Stotz et al. (1996: Fig. 3.3) reported similar cases. Whereas absolute richness values of these studies and ours coincide above about 2000 m, our values are much lower on the remainder of the transect. The avifauna of lowland terra firme forest in our study area is depauperate due to its semi-isolated location at the end of a narrow (10 /30-km wide), ca 500km long band of lowland forest wedged in between the Andes and open savannah habitats. Consequently, many bird species typical of Amazonian forest are lacking despite ecological conditions that in principle appear suitable. Blake and Loiselle (2000) found a diversity peak at 500 m in Costa Rica based on mist net captures, but point count data from the same transect indicated highest richness at 50 m. Although Blake and Loiselle (2000) standardized their data for sampling effort, they did not account for the effect of area. A reanalysis controlling for area seems worthwhile and we suspect that a hump-shaped pattern would result regardless of the survey method. Terborgh (1977: Fig. 5) also found a richness maximum at mid-elevation in Peru when controlling his data for sampling effort, but as noted by Rahbek (1995), only Terborgh’s non-standardized graph depicting a monotonic decline of species richness is usually cited even in recent papers (Heaney 2001). In conclusion, when excluding birds of seasonally flooded forests, a maximum of bird species richness in the foothills appears to be a general pattern in the Neotropics. 220

Taking into account other taxonomic groups along the same gradient can be an effective way to determine general species richness patterns (Heaney 2001). Two other groups of organisms have been studied in Carrasco at largely the same sites as birds. Ko¨hler (2000) surveyed amphibians at 500 /2500 m, and species richness peaked at 1500 m. However, amphibian species richness is about four to five times higher in adjacent lowland forests than at 1500 m (Ko¨hler and Reichle pers. comm.). This probably reflects the influence of temperature on the diversity of ectothermic organisms and the absence of lentic waters in montane forests required for the reproduction of many amphibians (Ko¨hler 2000). By contrast, alpha diversity of most of the six plant groups sampled by Kessler (2001a, b) on the Carrasco transect did show a mid-elevation peak (except for a monotonic decline in Melastomataceae), but individual maxima varied from elevations of 500 to 3000 m. Thus, while most taxa appear to show a humped richness-elevation relationship, the actual elevation of the peaks varies to such a degree that no single explanation seems plausible. In conclusion, the hump-shaped relationship between elevation and species richness of neotropical birds appears to be caused by the overlap between two large regional species pools, i.e. lowland and highland. Although at a first glance the resulting pattern may appear to be driven by MDE, at least in our study this is not the case. As outlined by Grytnes (2003) close examination of the elevational distribution of empirical range midpoints in indispensable for correctly interpreting any correlation between observed richness and MDE predictions. We suspect that some of the elevational gradient studies that have supported MDE models simply on the basis of correlations of species numbers without considering range midpoints would have to be re-assessed. It also is important in this context to distinguish latitudinal from elevational patterns. Elevational gradients typically cover much shorter spatial distances, and local population-level processes, such as source-sink dynamics (Wiens and Rotenberry 1981, Pulliam 1988, Grytnes 2003), can therefore have a stronger effect on richness patterns. For example, it would be interesting to examine the reproductive output of bird populations along elevational gradients to establish if high species richness at the overlap between the lowland and highland faunas corresponds to an accumulation of sink populations. The high-elevation plateau documented by us further exemplifies that despite several decades of ecological research, elevational gradients are still not well understood and other unexpected patterns may remain to be discovered. Acknowledgements / We thank J. Balderrama, A. Green, B. Rios, A. Rojas, R. Soria, and L. Tangara for help in the field and S. Mayer for assistance with the identification of tape recordings. We are grateful to J. Aparicio and C. Quiroga, Coleccio´n Boliviana de Fauna, La Paz, for logistical support, to the Direccio´n Nacional de Conservacio´n de la Biodiversidad, ECOGRAPHY 28:2 (2005)


La Paz, for work permits, to I. Da´valos for access to Carrasco N.P., and to the Univ. Mayor de San Simo´n, Cochabamba, for access to Valle del Sacta. SKH would like to thank F. Bairlein for guidance and support. We also thank R. K. Colwell, N. Gotelli, C. Rahbek, and T. Romdal for helpful discussions and for providing unpublished manuscripts. F. Bairlein, R. K. Colwell, J. A. Grytnes, and C. Rahbek made valuable comments on earlier drafts of this paper. Financial support was provided to SKH by the Gesellschaft fu¨r Tropenornithologie and to MK by the Deutsche Forschungsgemeinschaft and the DIVA project under the Danish Environmental Programme.

References Anon. 1997. STATISTICA for Windows. / Statsoft, Tulsa, OK. Begon, M., Harper, J. L. and Townsend, C. R. 1990. Ecology: individuals, populations and communities. / Blackwell. Blake, J. G. and Loiselle, B. A. 2000. Diversity of birds along an elevational gradient in the Cordillera Central, Costa Rica. / Auk 117: 663 /686. Bokma, F. and Mo¨nkko¨nen, M. 2000. The mid-domain effect and the longitudinal dimension of continents. / Trends Ecol. Evol. 15: 288 /289. Bokma, F., Bokma, J. and Mo¨nkko¨nen, M. 2001. Random processes and geographic species richness patterns: why so few species in the north. / Ecography 24: 43 /49. Brehm, G., Suessenbach, D. and Fiedler, K. 2003. Unique elevational diversity patterns of geometrid moths in an Andean montane rainforest. / Ecography 26: 456 /466. Brown, J. H. and Gibson, A. C. 1983. Biogeography. / Mosby. Brown, J. H. 2001. Mammals on mountainsides: elevational patterns of diversity. / Global Ecol. Biogeogr. 10: 101 /109. Caley, J. and Schluter, D. 1997. The relationship between local and regional diversity. / Ecology 78: 70 /80. Colwell, R. K. 1997. EstimateS: Statistical estimation of species richness and shared species from samples. Version 5. / User’s guide and application published at: B/http:// viceroy.eeb.uconn.edu/EstimateS /. Colwell, R. K. 2000. RangeModel: a Monte Carlo simulation tool for assessing geometric constraints on species richness. Version 3. / User’s guide and application published at: B/http://viceroy.eeb.uconn.edu/RangeModel /. Colwell, R. K. and Coddington, J. A. 1994. Estimating terrestrial biodiversity through extrapolation. / Philos. Trans. R. Soc. Lond. B 345: 101 /118. Colwell, R. K. and Hurtt, G. C. 1994. Nonbiological gradients in species richness and a spurious Rapoport effect. / Am. Nat. 144: 570 /595. Colwell, R. K. and Lees, D. C. 2000. The mid-domain effect: geometric constraints on the geography of species richness. / Trends Ecol. Evol. 15: 70 /76. Colwell, R. K., Rahbek, C. and Gotelli, N. J. 2004. The middomain effect and species richness patterns: what have we learned so far? / Am. Nat. 163: E1 /E23. Cornell, H. V. and Lawton, J. H. 1992. Species interactions, local and regional processes, and limits to the richness of ecological communities: a theoretical perspective. / J. Anim. Ecol. 61: 1 /12. Currie, D. J., Francis, A. P. and Kerr, J. T. 1999. Some general propositions about the study of spatial patterns of species richness. / EcoScience 6: 392 /399. Davis, A. L. V., Scholtz, C. H. and Chown, S. L. 1999. Species turnover, community boundaries and biogeographical composition of dung beetle assemblages across an altitudinal gradient in South Africa. / J. Biogeogr. 26: 1039 /1055. Diniz-Filho, J. A. F. et al. 2002. Null models and spatial patterns of species richness in South American birds of prey. / Ecol. Lett. 5: 47 /55. Ergueta S. P. and Go´mez C. H. (eds) 1997. Directorio de Areas Protegidas de Bolivia. / CDC, Bolivia. ECOGRAPHY 28:2 (2005)

Fjeldsa˚, J. 1999. The impact of human forest disturbance on the endemic avifauna of the Udzungwa Mountains, Tanzania. / Bird Conserv. Int. 9: 47 /62. Fjeldsa˚, J. and Krabbe, N. 1990. Birds of the high Andes. / Apollo Books. Fjeldsa˚, J., Lambin, E. and Mertens, B. 1999. Correlation between endemism and local ecoclimatic stability documented by comparing Andean bird distributions and remotely sensed land surface data. / Ecography 22: 63 /78. Gotelli, N. J. and Colwell, R. K. 2001. Quantifying biodiversity: procedures and pitfalls in the measurement and comparison of species richness. / Ecol. Lett. 4: 379 /391. Graves, G. L. 1988. Linearity of geographic range and its possible effect on the population structure of Andean birds. / Auk 105: 47 /52. Grytnes, J. A. 2003. Ecological interpretations of the middomain effect. / Ecol. Lett. 6: 883 /888. Grytnes, J. A. and Vetaas, O. R. 2002. Species richness and altitude: a comparison between null models and interpolated plant species richness along the Himalayan altitudinal gradient, Nepal. / Am. Nat. 159: 294 /304. Hawkins, B. A. and Diniz-Filho, J. A. F. 2002. The mid-domain effect cannot explain the diversity gradient of Nearctic birds. / Global Ecol. Biogeogr. 11: 419 /426. Heaney, L. R. 2001. Small mammal diversity along elevational gradients in the Philippines: an assessment of patterns and hypotheses. / Global Ecol. Biogeogr. 10: 15 /39. Henkel, R. 1995. Coca (Erythroxylum coca ) cultivation, cocaine production, and biodiversity in the Chapare region of Bolivia. / In: Churchill, S. P. et al. (eds), Biodiversity and conservation of neotropical montane forests. The New York Botanical Garden, NY, pp. 551 /560. Hennessey, A. B., Herzog, S. K. and Sagot, F. 2003. Lista anotatda de las aves de Bolivia. Quinta edicio´n. / Asociacio´n Armonı´a/BirdLife International. Herzog, S. K., Kessler, M. and Cahill, T. M. 2002. Estimating species richness of tropical bird communities from rapid assessment data. / Auk 119: 749 /769. Hoyo, J. del, Elliott, A. and Sargatal, J. (eds) 1992 /2002. Handbook of the birds of the world, Vol. 1-7. / Lynx Edicions. Ibisch, P. L. 1996. Neotropische Epiphytendiversita¨t / das Beispiel Bolivien. / Archiv naturwissenschaftlicher Dissertationen 1, Martina Galunder-Verlag, Germany. Isler, M. I. and Isler, P. R. 1999. Tanagers. / Christopher Helm. Jetz, W. and Rahbek, C. 2001. Geometric constraints explain much of the species richness pattern in African birds. / Proc. Nat. Acad. Sci. USA 98: 5661 /5666. Keating, K. A. and Quinn, J. F. 1998. Estimating species richness: the Michaelis-Menten model revisited. / Oikos 81: 411 /416. Kessler, M. 1999. Plants species richness and endemism during natural landslide succession in a perhumid montane forest in the Bolivian Andes. / Ecotropica 5: 123 /136. Kessler, M. 2000a. Altitudinal zonation of Andean cryptogam communities. / J. Biogeogr. 27: 275 /282. Kessler, M. 2000b. Observations on a human-induced fire event at a humid timberline in the Bolivian Andes. / Ecotropica 6: 89 /93. Kessler, M. 2000c. Elevational gradients in species richness and endemism of selected plant groups in the central Bolivian Andes. / Plant Ecol. 149: 181 /193. Kessler, M. 2001a. Pteridophyte species richness in Andean forests in Bolivia. / Biodiv. Conserv. 10: 1473 /1495. Kessler, M. 2001b. Patterns of diversity and range size of selected plant groups along an elevational transect in the Bolivian Andes. / Biodiv. Conserv. 10: 1897 /1920. Kessler, M. and Herzog, S. K. 1998. Conservation status in Bolivia of timberline habitats, elfin forest and their birds. / Cotinga 10: 50 /54. Ko¨hler, J. 2000. Amphibian diversity in Bolivia: a study with special reference to montane forest regions. / Bonner Zool. Monogr. 48.

221


Koleff, P. and Gaston, K. J. 2001. Latitudinal gradients in diversity: real patterns and random models. / Ecography 24: 341 /351. Koleff, P., Lennon, J. J. and Gaston, K. J. 2003. Are there latitudinal gradients in species turnover? / Global Ecol. Biogeogr. 12: 483 /498. Laurie, H and Silander, J. A. 2002. Geometric constraints and spatial pattern of species richness: critique of range-based models. / Div. Distribut. 8: 351 /364. Lees, D. C., Kremen, C. and Andriamampianina, L. 1999. A null model for species richness gradients: bounded range overlap of butterflies and other rainforest endemics in Madagascar. / Biol. J. Linn. Soc. 67: 529 /584. Lieberman, D. et al. 1996. Tropical forest structure and composition on a large-scale altitudinal gradient in Costa Rica. / J. Ecol. 84: 137 /152. Lieth, H. 1975. Modeling the primary productivity of the world. / In: Lieth, H. and Whittaker, R. H. (eds), Primary productivity of the biosphere. Springer, pp. 237 /263. MacKinnon, S. and Phillipps, K. 1993. A field guide to the birds of Borneo, Sumatra, Java and Bali. / Oxford Univ. Press. McCain, C. M. 2003. North American desert rodents: a test of the mid-domain effect in species richness. / J. Mammal. 84: 967 /980. McCain, C. M. 2004. The mid-domain effect applied to elevational gradients: species richness of small mammals in Costa Rica. / J. Biogeogr. 31: 19 /31. Md. Nor, S. 2001. Elevational diversity patterns of small mammals on Mount Kinabalu, Sabah, Malaysia. / Global Ecol. Biogeogr. 10: 41 /62. Mittelbach, G. G. et al. 2001. What is the observed relationship between species richness and productivity? / Ecology 82: 2381 /2396. Montes de Oca, I. 1997. Geografı´a y recursos naturales de Bolivia. / EDOBOL, Bolivia. Navarro, G. 1997. Contribucio´n a la clasificacio´n ecolo´gica y florı´stica de los bosques de Bolivia. / Revista Boliviana de Ecologı´a y Conservacio´n Ambiental 2: 3 /38. Patterson, B. D. et al. 1998. Contrasting patterns of elevational zonation for birds and mammals in the Andes of southeastern Peru. / J. Biogeogr. 25: 593 /607. Poulsen, B. O. et al. 1997. A rapid assessment of Bolivian and Ecuadorian montane avifaunas using 20-species lists: efficiency, biases and data gathered. / Bird Conserv. Int. 7: 53 / 67. Pulliam, H. R. 1988. Sources, sinks and population regulation. / Am. Nat. 132: 652 /661. Raaijmakers, J. G. W. 1987. Statistical analysis of the MichaelisMenten equation. / Biometrics 43: 793 /803. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern? / Ecography 18: 200 /205.

222

Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in neotropical birds. / Am. Nat. 149: 875 /902. Ribera, M. O. et al. 1996. Vegetacio´n de Bolivia. / In: Mihotek, K. (ed.), Comunidades, territorios indı´genas y biodiversidad en Bolivia. CIMAR-UAGRM, Bolivia, pp. 169 /222. Ricklefs, R. E. 2004. A comprehensive framework for global patterns in biodiversity. / Ecol. Lett. 7: 1 /15. Ridgely, R. S. and Tudor, G. 1989. The birds of South America. Vol. 1. / Oxford Univ. Press. Ridgely, R. S. and Tudor, G. 1994. The birds of South America. Vol. 2. / Oxford Univ. Press. Ridgely, R. S. and Greenfield, P. J. 2001. The birds of Ecuador. Status, distribution and taxonomy. / Christopher Helm. Rohde, K. 1992. Latitudinal gradients in species diversity: the search for the primary cause. / Oikos 65: 514 /527. Rosenzweig, M. L. 1968. Net primary productivity of terrestrial communities: prediction from climatic data. / Am. Nat. 102: 67 /74. Sanders, N. J. 2002. Elevational gradients in ant species richness: area, geometry, and Rapoport’s rule. / Ecography 25: 25 /32. Stevens, G. C. 1992. The elevational gradient in altitudinal range: an extension of Rapoport’s latitudinal rule to altitude. / Am. Nat. 140: 893 /911. Stotz, D. F. et al. 1996. Neotropical birds: ecology and conservation. / Univ. of Chicago Press. Terborgh, J. 1977. Bird species diversity on an Andean elevational gradient. / Ecology 58: 1007 /1019. Thornthwaite, C. B. and Mather, J. R. 1957. Instructions and tables for computing potential evapotranspiration and the water balance. / Drexel Inst. of Technology, Laboratory of Climatology, Publ. on Climatology 10: 181 /311. Waide, R. B. et al. 1999. The relationship between productivity and species richness. / Annu. Rev. Ecol. Syst. 30: 257 /300. Whittaker, R. H. 1972. Evolution and measurement of species diversity. / Taxon 21: 213 /251. Wiens, J. A. and Rotenberry, J. T. 1981. Censusing and the evaluation of avian habitat occupancy. / Stud. Avian Biol. 6: 522 /532. Willig, M. R. and Lyons, S. K. 1998. An analytical model of latitudinal gradients of species richness with an empirical test for marsupials and bats in the New World. / Oikos 81: 93 /98. Wilson, M. V. and Shmida, A. 1984. Measuring beta diversity with presence-absence data. / J. Ecol. 72: 1055 /1064. Zapata, F. A., Gaston, K. J. and Chown, S. L. 2003. Middomain models of species richness gradients: assumptions, methods and evidence. / J. Anim. Ecol. 72: 677 /690.

ECOGRAPHY 28:2 (2005)


ECOGRAPHY 26: 411–420, 2003

Area, altitude and aquatic plant diversity J. Iwan Jones, Wei Li and Stephen C. Maberly

Jones, J. I., Li, W. and Maberly, S. C. 2003. Area, altitude and aquatic plant diversity. – Ecography 26: 411– 420. Several explanations have been given for the decline in species richness with altitude. However, separating the influences of altitude, area, and isolation is difficult because of the conical shape of mountains. We used species lists of aquatic plants from \ 300 lakes in a small geographical area to investigate the influence of altitude on species richness. Altitude and/or surface area were better predictors of species richness than any measure of water chemistry. The surface area and depth of individual lakes were not related to altitude, neither was the degree of isolation from other waterbodies. Although species range size increased with altitude, range sizes of all but the rarer species (in the data set) encompassed the lowest altitudes, indicating species loss rather than turnover and no influence of the Rapoport rescue effect. Nevertheless we found a decline in species richness with altitude, additive to the effect of area. Species were ascribed to attribute groups according to an a priori classification based on morphological and life-history traits. The number of attribute groups and number of species within each group increased with area, suggesting both increased diversity and coexistence within niches. With altitude, the number of attribute groups declined, but the number of species per group increased, consistent with a loss of richness and reduced competition. The species remaining at high altitudes were characterised by stress tolerant traits, associated with sites of low productivity. Our results suggest an absolute effect of altitude on species richness, irrespective of other influences and consistent with a decline in productivity. J. I. Jones ( j.i.jones@qmul.ac.uk), School of Biological Sciences, Queen Mary, Uni6. London, London, U.K. E1 4NS. – W. Li, Lab. of Aquatic Plant Biology, Wuhan Inst. of Botany, The Chinese Academy of Sciences, Wuhan, 430074, People’s Republic of China. – S. C. Maberly, CEH Windermere, The Ferry House, Ambleside, Cumbria, U.K. LA22 0LP.

The relationship between area and species richness is probably one of the few general laws of ecology (Lawton 1999), with a weight of evidence and theoretical background supporting it (MacArthur and Wilson 1967, Rosenzweig 1995). In contrast, the decline in species richness with latitude, although widely applicable (Lawton 1999) and long recognised (Wallace 1878), still does not have a universally accepted mechanism at its root (Stevens 1989, Begon et al. 1990, Rohde 1992, Rosenzweig 1995, Willig and Lyons 1998, Chown and Gaston 2000, Lambers et al. 2002). Although studies of the influence of latitude on species richness abound (Rosenzweig 1995, Chown and Gaston 2000), fewer have contemplated the influence of

altitude (Rahbek 1995, Ko¨rner 2000). Nevertheless, a natural comparison exists between latitude and altitude, both representing a gradient of worsening climate. As with latitude, there are, however, several confounding factors. The conical shape of mountains means that the influence of altitude is compounded by one of area. The area contained within altitudinal zones is progressively reduced towards the summit and the species characteristic of higher altitudes influenced by a restricted area of habitat available to them. Studies of variation in species richness with altitude which have acknowledged this problem have sought to overcome it by including the area of arbitrary altitudinal bands (i.e. not necessar-

Accepted 20 January 2003 Copyright © ECOGRAPHY 2003 ISSN 0906-7590 ECOGRAPHY 26:4 (2003)

411


ily corresponding to species distributions, e.g. every 50 m) as a covariable in their models, with some success (Rahbek 1997, Odland and Birks 1999, Grytnes and Vetaas 2002). Nevertheless, due to covariation (the tops of mountains are always smaller than the lower parts), altitude and area are not entirely separable, and these models can only seek to identify any effect of area additional to that of altitude. The potential hard boundaries presented to species distributions at the upper (top of a mountain) or lower (bottom of a valley or sea level) extremes of the altitudinal gradient represent geometric constraints to random distributions of species. Incorporation of such hard boundaries in random simulation models leads to a humped response of species richness to altitude as a null model (Willig and Lyons 1998, Grytnes and Vetaas 2002). This humped response is accentuated if species richness is calculated from regional descriptions of species distributions and species assumed to be present between their highest and lowest sightings (e.g. Stevens 1992, Rahbek 1997, Fleishman et al. 1998). Calculated this way, species richness near the altitudinal extremes consists only of actual sightings, whereas at other altitudes species richness is inflated by interpolations in addition to actual observations (Grytnes and Vetaas 2002). Further complications arise from an increasing degree of isolation with altitude, with populations of those species restricted to mountaintop ‘‘islands’’ separated from one another by an ocean of valleys, which is not true for the converse. The species of lakes suffer from isolation also, surrounded by terrestrial habitats largely inhospitable to aquatic organisms. Lakes are easily separable from the surrounding terrestrial matrix and their area estimated, and because of this discrete nature they have frequently been used to test theories of island biogeography (e.g. Browne 1981, Tonn and Magnuson 1982, Fryer 1985). However, a number of studies have stressed the importance of local environmental conditions in determining species richness and composition in lakes, particularly those related to water chemistry and productivity (Spence 1967, Seddon 1972, Rørslett 1991, Vestergaard and Sand-Jensen 2000, Jeppesen et al. 2000, Heegaard et al. 2001). Here we present a study of species richness of aquatic plants in lakes that cover a range of altitude and surface area, yet within a restricted geographical region. We have used lakes in order to isolate the effects of area, range size and isolation, from altitude, and species richness directly from point samples to avoid those errors that result from the use of distributions interpolated from regional descriptions of species. We have also incorporated the environmental characteristics of each lake (where possible) to include local influences on species richness. 412

Methods Between 1973 and 1980 the amateur naturalist R. Stokoe conducted an extensive survey of aquatic plants in 316 lakes, ponds and tarns in Cumbria. At each water body he compiled a list of plant species present, based on one or several visits. Specialists verified voucher specimens at the time. Upon his untimely death his data were donated to and published by the Freshwater Biological Association (Stokoe 1983), and it is these data which comprise the species lists used here. We used a more strict definition of aquatic plant species than that of Stokoe (1983), who included all plants found in damp and wet habitats surrounding the water bodies, namely those species which are described in the texts of Moore (1986), Cook (1990) or Preston and Croft (1997). We adopted this protocol to provide an objective boundary between aquatic and terrestrial habitats, which in reality is transitional and fluctuates temporally. Applying these criteria reduced the recorded number of species from 233 to 130. Many of those excluded were terrestrial and only occurred at a single site. We disregarded all taxonomy below the species level. Records that were not determined below genus were included where they could be attributed exclusively to an aquatic species. We followed the taxonomy of the above texts. Although the species used in our analysis cover a wide taxonomic range (green algae, bryophytes, pteridophytes and angiosperms), they are all macrophytes (large photosynthetic organisms typically rooted in a permanently submerged substrate) constrained by the selective pressures of living in an aquatic environment. Data describing the environmental conditions in the lakes were derived from various sources (Table 1). Where repeat measurements were available, we used Table 1. Sources used to derive data describing environmental conditions in 316 Cumbrian water bodies. Sources Altitude Area Depth Inflows Outflow Dam Distance to nearest standing waterbody Major ions Alkalinity pH Nitrate Phosphorus Eastings Northings

1, 2, 2, 6, 6, 6, 6 2, 2, 2, 2, 2, 6 6

2 ,6, 7 3, 6 3, 5, 7, 8 7 7 7 8 4, 8 8 8 8

1. Stokoe (1983). 2. Carrick and Suttcliffe (1982). 3. Smyly (1958). 4. Knudson (1954). 5. Talling (1999). 6. Ordnance Survey 1:25 000 maps. 7. The original Stokoe record cards held at the FBA library, The Ferry House, Windermere. 8. unpubl. ECOGRAPHY 26:4 (2003)


means encompassing those most proximal to the date of Stokoe’s survey in preference. Where they were unavailable from other sources we calculated the surface area of water bodies using 1:25000 Ordnance Survey maps. To assess the proximity of the nearest standing water body, we measured the distance from each lake to the nearest standing water shown on 1:25000 Ordnance Survey maps. The most complete sets of data were those describing morphometry and geographical setting. We calculated species altitudinal ranges as the highest and lowest sites where the species occurred, and the mid-point as the average of these values. To determine the effect of temporal incongruence and missing data on the results, we also conducted analyses using only those sites where a complete set of water chemistry data from the 1970s was available, which restricted the number of sites to 67 and species to 86. Initially, we tested for any covariation among the environmental variables using correlation, or logistic regression in the case of presence absence data. Subsequently, we used linear regression with stepwise selection to determine any relationships between species richness (as number of species) and the available environmental variables using SAS (Anon. 1989). To remove heteroscedasticity we log transformed the data where appropriate, after examination of the residuals. For variables without complete sets of data, we repeated analyses excluding the variable to determine the influence of missing data on the results. We investigated the relationship between the occurrence of individual species and environmental conditions using canonical correspondence analysis (CCA), with rare species downweighted using Canoco (ter Braak 1987). Initial tests indicated that the first axis was \ 2 standard deviations in length indicating that unimodal methods were more appropriate than linear ones. In order to determine how different kinds of species responded to changing conditions, we classified plant species, a priori, into the 20 attribute groups (groups of species sharing similar suites of morphological and life history traits) of Willby et al. (2000), together with a further 2 nominal groups, bryophytes and charophytes. We allocated the two hybrids whose parent species spanned attribute groups to the attribute group of the most morphologically similar parent species. For each water body the number of groups, mean number of species per group, and occurrence of groups were used in analyses.

Results The varied geology of Cumbria and the comprehensive survey of Stokoe resulted in the water bodies used ECOGRAPHY 26:4 (2003)

covering a wide range of environmental conditions from coastal lagoons to mountain corries, farm ponds to glacial lakes. Altitude of the water bodies ranged from 2 to 837 m a.s.l., area from 100 m2 to 14.77 km2, alkalinity from − 16 to 4405 meqv L − 1. Naturally there was considerable covariation in the measures of environmental conditions (Fig. 1), with ion poor lakes predominant at higher altitudes and ion rich lakes at the very lowest (caused by the influence of the sea). Many of the ions (except nitrate, sulphate and sodium) were correlated with alkalinity, and each other (Fig. 1). Area was not correlated with altitude (Fig. 2, p = 0.15): although the largest lakes were found at altitudes B 250 m a.s.l., small and moderate sized water bodies were found at all altitudes. Variance in the size of the water bodies was greatest at lower altitudes. Area was correlated with depth (p B 0.0001), presence (p = 0.012) and number of inflows (pB0.0001), presence of an outflow (p B 0.0001) and nitrate concentration (p = 0.024) (Figs 1 and 3). Altitude was not correlated with presence (p =0.47) or number of inflows (p= 0.20), the presence of an outflow (p= 0.68) or the distance to the nearest standing water shown on a 1:25000 map (p = 0.41) (Fig. 3). Of all the variables tested altitude and area were the most significant predictors of species richness, and their effect was additive (no interaction as indicated by GLM). Species richness declined with altitude, and increased with area (Fig. 2). When both these variables were combined the predictive power of the model increased significantly (log species richness + 1=0.1739 0.018(Log area) −0.0010198.8e − 5(altitude) + 0.5489 0.077, F =126.4, p B 0.0001, adj R2 = 0.451), indicating that the distribution of lake area with altitude did not

Fig. 1. Biplot showing correlations between the environmental variables used in the 316 sites. The angle between variables indicates the extent of correlation: acute = correlated, 90°= not correlated, 180°=inversely correlated.

413


sively steeper with increasing numbers of observations, and thus an increasing accuracy with which range size could be estimated. With an increasing number of observations, the slope of this relationship approached 2 (Fig. 5); i.e. where the range of a species could be accurately estimated it included the lowest altitudes. There were many upper elevational limits to species’ distributions but not lower elevational limits. The number of attribute groups per water body increased with area and declined with altitude (Fig. 6a, b), whilst the mean number of species per group increased in both cases (Fig. 6c, d). Other variables were significantly correlated with the number of attribute groups per water body and the mean number of species per group (e.g. number of inflows, alkalinity, chloride concentration) but again these did not have a significant effect when the better predictors, altitude or area, were included in the model. Naturally, attribute groups responded to the environmental variables in a similar fashion to their constituent species (Fig. 7). Notably group 20 (characterised by species with traits of stress tolerance) were associated with higher altitude (Fig. 7), although the constituent species were found throughout the range.

Fig. 2. Relationships between a) altitude and area (p = 0.15), b) species richness and area (F = 71.8, p B0.0001), and c) species richness and altitude (F = 83.97, p B0.0001). Including both altitude and area produced the model, log species richness + 1 = 0.173 9 0.018(Log area) −0.001019 8.8e − 5(altitude) + 0.548 9 0.077, F =126.4, p B0.0001, adj R2 = 0.451.

influence these relationships. Species richness was not correlated with total phosphorus or nitrogen, either independently or in combination with altitude and/or area. Although species richness was also correlated with other variables (e.g. number of inflows, alkalinity, chloride concentration) these variables covaried with altitude or area (Fig. 1) and were less good at predicting richness than either of these factors, singly or in combination. The size of the range of altitudes over which species occurred increased with increasing altitude (Fig. 4, p B 0.0001). However, there did not appear to be any zonation of species (Fig. 5). Although rare species (those occurring in only 1 or 2 sites) were more frequent at low altitudes (Fig. 4), species of high altitudes were found throughout the gradient of altitude (Figs 4 and 5). Estimated altitudinal range size was influenced by the frequency of occurrence of species within the data set (Fig. 4). The slope of the relationship between the size of species’ altitudinal ranges and the altitudes at which their mid-points occurred became progres414

Discussion The influence of water chemistry on the distribution of aquatic plants has long been known (Iversen 1929), with alkalinity and nutrient availability variously described as major predictors of species distributions (Seddon 1972, Kadono 1982, Tiovonen and Huttunen 1995, Vestergaard and Sand-Jensen 2000, Heegaard et al. 2001). As with all correlative studies of species distributions, these descriptions suffer from an inability to separate the influence of naturally associated variables (e.g. calcium and alkalinity) without considerable effort in site selection. Here we used sites from a small but varied geographical area, rich in water bodies, to investigate environmental influences on aquatic plant distribution. Together with many other studies of species distributions this is a retrospective analysis of a large data set collected for other purposes, and fraught with the difficulties of covariation. Although it was not possible to attribute sole influence to many of the water chemistry variables, notably alkalinity and phosphorus, the influence of surface area was separable from these variables and from altitude (Fig. 1). The relationship between species richness and area is well founded in island biogeography (MacArthur and Wilson 1967). Aquatic habitats lend themselves well to such studies because of their easily prescribed boundaries, and several descriptions in support of the species-area relationship have come from such habitats (e.g. Browne 1981, Tonn and Magnuson 1982). Here ECOGRAPHY 26:4 (2003)


Fig. 3. Influence of altitude and area on isolation from other waterbodies, measured as a) presence of at least one inflow (altitude, p =0.47, area, p = 0.012), b) presence of an outflow (altitude, p = 0.68, area, p B0.0001), c) number of inflows (altitude, p =0.20, area, p B 0.0001), d) distance to the nearest standing waterbody visible on a 1:25000 map (altitude, p =0.41, area, p =0.21). ECOGRAPHY 26:4 (2003)

415


Fig. 4. Relationship between the size of the range of altitudes over which species occurred and the altitudes at which their mid-points occurred. Species are categorized according to frequency of occurrence within the data set (number of sites where that species was observed). According to ANCOVA, altitude F1,123 = 760, p B 0.0001, frequency of occurrence F3,123 = 79.5, p B 0.0001, altitude Ă&#x2014; occurrence F3,123 =30.1, p B 0.0001. Influence of occurrence on mid-point of range F3,127 = 8.67, p B 0.0001.

we found an increase in species richness with area, which was not attributable to an increase of elodeid species in alkaline lakes (Vestergaard and Sand-Jensen 2000). The slope of the relationship between species richness and area (z = 0.1739 0.018) was comparable to that found by other workers for aquatic plants (Vestergaard and Sand-Jensen 2000) and other aquatic organisms (Browne 1981). Even though our lakes spanned a range from oligotrophic to hyper-eutrophic (Anon. 1982) neither total phosphorus nor nitrate had any effect on species richness. The number of species did increase logarithmically with alkalinity and several other ions (e.g. chloride, potassium). Nevertheless altitude was a better predictor of species richness than any of these other variables. Over the altitudinal gradient used here, the largest altitudinal range in England, we found a linear decline in species richness. The influence of altitude was additive to that of area (from GLM), with these two variables explaining 45% of the variation in species richness.

Fig. 5. Highest and lowest altitude at which species occurred, ranked according to the highest altitude of occurrence.

416

ECOGRAPHY 26:4 (2003)


Fig. 6. Relationship between number of attribute groups per lake and either a) area (F= 87.85, p B0.0001) or b) altitude (F = 80.5, p B 0.0001), and relationship between mean number of species per attribute group and either c) area (F =19.7, p B 0.0001) or d) altitude (F = 14.2, p = 0.0002). Species were ascribed to attribute groups according to an a priori classification based on morphological and life-history traits (Willby et al. 2000) together with a further 2 nominal groups, bryophytes and charophytes, giving a total of 21 groups.

The likelihood of an inflow or outflow being present, and the number of inflows increased with area, indicating that isolation and area were inversely correlated and that larger lakes were more likely to receive an influx of colonists from other waterbodies. Depth was highly correlated with area also. Zonation of aquatic plants with depth is well known, with species having a characteristic position along this gradient (Spence 1982). An increase in depth will result in an increase in the number of niches available along this gradient, and hence species richness. In addition to using species per se we allocated the species to the attribute groups of Willby et al. (2000) classified a priori according to morphological and reproductive characteristics. The number of attribute groups bore a strong relationship with area, with both the number of groups and the average number of species within each group increasing with area. Species within attribute groups share similar sets of traits and are thus adapted to similar environmental conditions. Thus, an increase in area not only results in more different kinds of species, but more similar species also. If we accept that these attribute groups are likely to correspond with niches (Grime et al. 1988), then an increase in area results in an increase in the number of niches and also the density of species within each niche. ECOGRAPHY 26:4 (2003)

The former could be explained by an increase in habitat diversity (microhabitats) or a more complete exploitation of those available, whilst the latter could be a consequence of increased coexistence within niches (niche packing) or a decrease in niche size. Although the largest lakes were found B 250 m a.s.l., there was no effect of altitude on area (Fig. 2); moderate and small sized water bodies were found at all altitudes. Depth, which has the potential of increasing species coexistence by increasing the number of niches within a water body, was not correlated with altitude either. Typically, the influence of altitude is not separable from that of area. Mountains are conical, and the species of higher altitudes are influenced by a restricted area of habitat. Whereas other workers have dealt with this by incorporating the area of altitudinal zones in their models (Rahbek 1997, Odland and Birks 1999, Grytnes and Vetaas 2002), here we separated this effect of declining habitat area with altitude by using species from a prescribed habitat found at a range of altitudes, and the areas of these patches of habitat. Furthermore, altitudinal zonation of species was not apparent in our data. There were many upper elevational limits to speciesâ&#x20AC;&#x2122; distributions but not lower elevational limits (Figs 4 and 5). As a consequence, those species that 417


Fig. 7. Biplot obtained through canonical correspondence analysis of the distributions of attribute groups and environmental variables. Attribute groups correspond to those described in Willby et al. (2000) where details of the constituent species are given, together with a further 2 nominal groups, bryophytes (group 21) and charophytes (group 22). Group 1 = Alisma plantago-aquatica type, 2 = Glyceria fluitans type, 3 = Nymphaea alba type, 4 =Potamogeton polygonifolius type, 5 =Apium inundatum type, 6 =Persicaria amphibia type, 7 = Callitriche hamulata type, 8 =Ranunculus peltatus type, 9 = Lythrum portula type, 10 = Ranunculus omiophyllus type, 11 = Ranunculus flammula type, 12 =Potamogeton berchtoldii type, 13 = Elodea canadensis type, 14 =Utricularia minor type, 15 = Potamogeton crispus type, 16 =Myriophyllum alterniflorum type, 17 =Zannichellia palustris type, 18 = Lemna minor type, 20 =Juncus bulbosus type. No representative species of group 19 were found.

were present at higher altitudes had a larger total area of habitat available to them than those restricted to lower altitudes, the converse of what is usually found for montane vegetation. Thus, the decline in species richness with altitude is not attributable to a decline in the total area of habitat available, and not caused by differences in the species susceptibility to metapopulation sizes. The distribution of the species with altitude also has consequences with respect to the Rapoport effect (Stevens 1992). (The Rapoport rescue effect maintains that where ranges are narrow there is an increased likelihood of transient populations of species not native to that range being sustained by repeated colonisation, thus inflating species richness.) Although the size of speciesâ&#x20AC;&#x2122; altitudinal range did increase with altitude, species were found throughout the altitudinal gradient with no zonation or turnover of species and hence no potential for the rescue effect inflating the numbers of species at lower altitudes. The Rapoport effect on species richness (Stevens 1989, 1992), therefore, does not 418

contribute to the decline in species with altitude seen in this data set. Unlike other montane vegetation, the degree of isolation was not related to altitude. Neither the presence of an inflow nor an outflow were correlated with altitude, indicating that water bodies at lower altitudes were not more likely to receive a supply of colonists from upstream or downstream (Johansson and Nilsson 1993). The degree of isolation in the varied landscape of Cumbria is more dependent on geographical location than altitude. Measures of isolation had no effect on species richness additional to those of area and altitude, suggesting that metapopulation effects on species richness were not apparent. Although altitude is correlated inversely with human population density and hence with the degree of human influence (as vectors for plant propagules), previous workers have found that population density only influences the number of alien species and not species richness (Roy et al. 1999). In Cumbria, there is the additional complication of tourists who visit the area with the express purpose of climbing to high altitude. Using this methodology to separate the effects of altitude, area and isolation, we found a linear decrease in species richness which can be ascribed solely to increasing altitude, probably due to an indirect effect on temperature and duration of growing season. Mean temperatures decline at a rate of ca 0.65°C 100 m â&#x2C6;&#x2019; 1, with a resultant reduction in the length of the growing season. Precipitation and wind speed also increase with altitude (Woodward 1993), but these are unlikely to have a substantial direct effect on the growth of submerged plants. Together with species richness, the number of attribute groups declined with altitude (Fig. 6b); not only were there fewer species but fewer kinds of species, with group 20, the stress tolerators, predominant at higher altitudes. This was not simply a consequence of sampling effort (fewer species =fewer groups) as the mean number of species per group increased with altitude (Fig. 6d). An increase in species per group with altitude implies increased coexistence (within niches) and reduced competition. Such an effect is typical of low productivity environments (Begon et al. 1990), and is consistent with stress tolerant species being found at altitude, competition being less between such species (Grime 1979). The response of the species within this attribute group to other co-varying environmental variables cannot be discounted, many are typical of oligotrophic ion-poor conditions which predominate at high altitude, yet the distribution of these species encompassed all altitudes, including the lowest and most ion-rich conditions. This group of species is also predominant at higher latitudes of both Britain (Preston and Croft 1997) and Finland (Virola et al. 2001), a pattern usually linked with geology, a lower human population density and more oligotrophic conditions (Palmer et al. 1992, Preston and Croft 1997). Although ECOGRAPHY 26:4 (2003)


it is not possible to ascribe these patterns solely to the effect of altitude or latitude, because of colinearity with many variables, the species of this attribute group display many traits associated with low productivity and poor growth conditions (e.g. slow growing, evergreen, low stature) consistent with the short growing season and low temperatures found at high altitude and latitude. Although we cannot comment on the evolutionary history, the species used in our analysis cover a wide taxonomic range (green algae, bryophytes, pteridophytes and angiosperms) of varying antiquity, and the area covered by Stokoe’s survey is one shaped by past glaciations. Furthermore, we have separated the effects of area, isolation and range size from that of altitude, and still found a decline in species richness. This decline is attributable solely to altitude, consistent with an effect of declining productivity. Contrary to Rosenzweig’s hypothesis (Rosenzweig 1995), that the latitudinal gradient of species richness is one of area, a consequence of the tropics occupying a larger contiguous area than other climatic regions, our data indicate an effect of climate on species richness, irrespective of area. Acknowledgements – We dedicate this paper to the late Ralph Stokoe and family, without whose efforts this work would not have been possible. We are indebted to Freshwater Biological Association, for their kind permission to use the data. Thanks also to Nigel Willby for assistance with the attribute groups, and to Robert Ricklefs for his useful comments on the manuscript. J. I. Jones is supported by NERC Fellowship GT5/98/21/CB, and W. Li’s visit to the U.K. funded by the Royal Society and the Chinese Academy of Sciences (the Hundred Talents Program).

References Anon. 1982. Eutrophication of waters: monitoring, assessment and control. – Tech. Rep., Environmental Directorate, Organisation for Economic Co-operation and Development. Anon. 1989. SAS/STAT. – SAS Inst. Begon, M., Harper, J. L. and Townsend, C. R. 1990. Ecology: individuals, populations, and communities. – Blackwell. Browne, R. A. 1981. Lakes as islands: biogeographic distribution, turnover rates, and species composition in the lakes of central New York. – J. Biogeogr. 8: 75 – 83. Carrick, T. R. and Suttcliffe, D. W. 1982. Concentrations of major ions in the lakes and tarns of the Lake District (1953 – 1978). – Occas. Publ. No. 6, Freshwater Biol. Assoc. Chown, S. L. and Gaston, K. J. 2000. Area, cradles and museums: the latitudenal gradient in species richness. – Trends Ecol. Evol. 15: 311 –315. Cook, C. D. K. 1990. Aquatic plant book. – SPB Academic Publishing. Fleishman, E., Austin, G. T. and Weiss, A. D. 1998. An empirical test of Rapoport’s rule: elevational gradients in montane butterfly communities. – Ecology 79: 2482 – 2493. Fryer, G. 1985. Crustacean diversity in relation to the size of water bodies: some facts and problems. – Freshwater Biol. 15: 347 – 361. ECOGRAPHY 26:4 (2003)

Grime, J. P. 1979. Plant strategies and vegetation processes. – Wiley. Grime, J. P., Hodgson, J. G. and Hunt, R. 1988. Comparative plant ecology: a functional approach to common British species. – Unwin Hyman. Grytnes, J. A. and Vetaas, O. R. 2002. Species richness and altitude: a comparison between null models and interpolated plant species richness along the Himalayan altitudinal gradient, Nepal. – Am. Nat. 159: 294 – 304. Heegaard, E. et al. 2001. Species – environmental relationships of aquatic macrophtes in Northern Ireland. – Aquat. Bot. 70: 175 – 223. Iversen, J. 1929. Studien u¨ ber pH-verha¨ ltnisse da¨ nischer gewa¨ sser und ihren einfluss auf die hydrophyten-vegetation. – Bot. Tidsskr. 40: 277 – 333. Jeppesen, E. et al. 2000. Trophic structure, species richness and biodiversity in Danish lakes: changes along a phosphorus gradient. – Freshwater Biol. 45: 201 – 218. Johansson, M. E. and Nilsson, C. 1993. Hydrochary, population-dynamics and distribution of the clonal aquatic plant Ranunculus lingua. – J. Ecol. 81: 81 – 91. Kadono, Y. 1982. Distribution of Japanese Potamogeton. – Bot. Mag. Tokyo 95: 63 – 76. Knudson, B. M. 1954. The ecology of the genus Tabellaria in the English Lake District. – J. Ecol. 42: 345 – 358. Ko¨ rner, C. 2000. Why are there global gradients in species richness? Mountains might hold the answer. – Trends Ecol. Evol. 15: 513 – 514. Lambers, J. H. R., Clark, J. S. and Beckage, B. 2002. Densitydependent mortality and the latitudinal gradient in species diversity. – Nature 417: 732 – 735. Lawton, J. H. 1999. Are there general laws in ecology? – Oikos 84: 177 – 192. MacArthur, R. H. and Wilson, E. O. 1967. The theory of island biogeography. – Princeton Univ. Press. Moore, J. A. 1986. Charophytes of Great Britain and Ireland. – Bot. Soc. of the British Isles. Odland, A. and Birks, H. J. B. 1999. The altitudinal gradient of vascular plant richness in Aurland, western Norway. – Ecography 22: 548 – 566. Palmer, M. A., Bell, S. L. and Butterfield, I. 1992. A botanical classification of standing waters in Britain: applications for conservation and monitoring. – Aquat. Conserv. Mar. Freshwater Ecosyst. 2: 125 – 143. Preston, C. D. and Croft, J. M. 1997. Aquatic plants in Britain and Ireland. – Harley Books. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern? – Ecography 18: 200 – 205. Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in neotropical birds. – Am. Nat. 149: 875 – 902. Rohde, K. 1992. Latitiudinal gradients in species diversity: the search for the primary cause. – Oikos 65: 514 – 527. Rørslett, B. 1991. Principal determinants of aquatic macrophyte richness in northern European lakes. – Aquat. Bot. 39: 173 – 193. Rosenzweig, M. L. 1995. Species diversity in space and time. – Cambridge Univ. Press. Roy, D. B., Hill, M. O. and Rothery, P. 1999. Effects of urban land cover on the local species pool in Britain. – Ecography 22: 507 – 515. Seddon, B. 1972. Aquatic macrophytes as limnological indicators. – Freshwater Biol. 2: 107 – 130. Smyly, W. J. P. 1958. The Cladocera and Copepoda (Crustacea) of the tarns of the English Lake District. – J. Anim. Ecol. 27: 87 – 103. Spence, D. H. N. 1967. Factors controlling the distribution of freshwater macrophytes with particular reference to the lochs of Scotland. – J. Ecol. 55: 147 – 170. Spence, D. H. N. 1982. The zonation of plants in freshwater lakes. – Adv. Ecol. Res. 12: 37 – 125. Stevens, G. H. 1989. The latitudinal gradient in geographical range: how so many species coexist in the tropics. – Am. Nat. 133: 240 – 256.

419


Stevens, G. H. 1992. The elevational gradient in altitudinal range: an extension of Rapoport’s latitudinal rule to altitude. – Am. Nat. 140: 893 –911. Stokoe, R. 1983. Aquatic macrophytes in the tarns and lakes of Cumbria. – Occas. Publ. No. 18, Freshwater Biol. Assoc. Talling, J. F. 1999. Some English lakes as diverse and active ecosystems: a factual summary and source book. – Freshwater Biol. Assoc. ter Braak, C. J. F. 1987. Canoco – a FORTRAN program for Canonical Community Ordination by [Partial] [Detrended] [Canonical] Correspondence Analysis, Principal Components Analysis and Redundancy Analysis (ver. 2.1). Tiovonen, H. and Huttunen, P. 1995. Aquatic macrophytes and ecological gradients in 57 small lakes in southern Finland. – Aquat. Bot. 51: 197 –221. Tonn, W. M. and Magnuson, J. J. 1982. Patterns in the species composition and richness of fish assemblages in northern Wisconsin lakes. – Ecology 63: 1149 –1166.

420

Vestergaard, O. and Sand-Jensen, K. 2000. Aquatic macrophyte richness in Danish lakes in relation to alkalinity, transparency, and lake area. – Can. J. Fish. Aquat. Sci. 57: 2022 – 2031. Virola, T. et al. 2001. Geographic patterns of species turnover in aquatic plant communities. – Freshwater Biol. 46: 1471 – 1478. Wallace, A. R. 1878. Tropical nature and other essays. – Macmillan. Willby, N. J., Abernethy, V. J. and Demars, B. O. L. 2000. Attribute-based classification of European hydophytes and its relationship to habitat utilization. – Freshwater Biol. 43: 43 – 74. Willig, M. R. and Lyons, S. K. 1998. An analytical model of latitudinal gradients of species richness with an empirical test for marsupials and bats in the New World. – Oikos 81: 93 – 98. Woodward, F. I. 1993. The lowland-to-upland transition – modelling plant responses to environmental change. – Ecol. Appl. 3: 404 – 408.

ECOGRAPHY 26:4 (2003)


ECOGRAPHY 22: 659-673. Copenhagen 1999

Assemblage structure and quantitative habitat relations of small mammals along an ecological gradient in the Colorado Desert of southern California Douglas A. Kelt

Kelt, D. A. 1999. Assemblage structure and quantitative habitat relations of small mammals along an ecological gradient in the Colorado Desert of southern California. Ecography 22: 659-673. Ecological gradients have intrigued ecologists for many years. In southern California the Deep Canyon Transect spans a range of habitats and elevations from Lower Sonoran Desert sand dunes and creosote scrub to Upper Transition coniferous forest, where relict species typical of the Sierra Nevada are found. I sampled a 1050 nci elevational range in this transect to evaluate the ecological distributions of small mammals and to better characterize community structure. Results complement and substantially extend a previous study of this fauna, and provide insights into the habitat associations of species in this complex fauna. Assemblage structure changed greatly between summer and winter, largely due to reduced presence of pocket mice Chuetodipus in winter. Additionally, the distribution of abundanee and species richness was different than reported earlier, suggesting that patterns across this gradient may be temporally variable, and strongly influenced by local dynamics. Most taxa exhibited significantly nonrandom use of a large number of habitat variables but this was not a simple consequence of the elevational gradient. A mid-elevation bulge in species richness was indicated, but likely is not a consequence of mass etTects since a number of animals captured in intermediate regions were reproductively active. D. A. Kelt (dakc'lt{iv,u(dciiis.i'du). Dcpl oJ Wildlijc. Fish, and Consenaiion Univ. of Ccilijornici. Dini.s, CA 95616, USA.

Many species reach their ecological litnits across ecological gradiertts. where they come into contact wit:h taxa from separate biotas. Elevational gradients a:re particularly interesting as these often are rooted at either end by species adapted to very dilTerent environmental conditions (e.g. thermal or precipitation regimes). As a result, numerous studies have been conducted in recent years on patterns of distribution and abundance of various taxa across elevational gradients (e,g,, Whitaker and Neiring 1965. Kikkawa and Williams 1971. Terborgh 1977, Patterson et al, 1989, 1990, Rickart et al, 1991, Kelt et al, 1999), In a recent review of species richness across elevational gradients. Rahbek (1995) concluded that species richness tends to decline with elevation, but often not monotonically.

Biology,

Interestingly, almost half of the studies reviewed (including 36 of 73 studies in tropical regions and eight of 17 studies in non-tropical regions) demonstrated a hump-shaped pattern, with greater species richness at intermediate elevations than at sites located at either extreme. Hypotheses to explain such elevated richness within ecotonal regions generally invoke increased resource availability at intertnediate regions or the overlap of species from faunas at either end of the transition. If environmental conditions are more favorable within ecological transitions then elevated abundance could refiect a higher carrying capacity. Increased habitat heterogeneity should favor elevated species richness, while not necessarily favoring greater abundance. Shmida and Wilson (1985) argued that in

Accepted 10 March 1999 Copyright r) ECOGRAPHY 1999 ISSN 0906-7590 Printed in Ireland - all rights reserved ECOGRAPHY 22:() (l')')91

659


some cases intermediate sites are populated by individuals that have emigrated from more optimal habitats. Such "mass effects" would influence richness more than abundance, and have been demonstrated in some systems (see Kunin 1998), The scale of observation may also influence interpretations. Across a substantial ecological transition in southern South America, several authors noted a hump-shaped distribution of species richness for birds (Ralph 1985) and small mammals (Pearson and Pearson 1982), However, closer inspection of the mammal fauna there indicated that the increased richness was caused by increased beta diversity between habitats at intermediate sites, but that at any point across this transition had about equal numbers of species (Kelt 1996). Thus, there is reason to remain skeptical about such generalities until further studies evaluate patterns

at multiple spatial and temporal scales (see Wiens 1989, Levin 1992, Karieva 1994), In southern California the Colorado Desert yields to the Peninsular mountain ranges across a relatively short distance (Burk 1988, Thorne 1988; Fig, 1). In a span of < 15 km the biota changes substantially; faunal and floral elements of the Colorado Desert yield to elements of the San Diegan District, and near the crests of the Peninsular Ranges a subset of Sierra Nevadan elements are found amongst coniferous forest (Fig. 2), Faunal lists taken from Ryan's (1968) study on the mammalian fauna of this transition (Fig. 3) suggest that species richness of both rodent species and all terrestrial mammals are distributed in a bimodal fashion across this elevational gradient. Richness was relatively low in the sand dunes of the central Coachella Valley, increased in the Deep Canyon floodplain, and declined

M-^ 2,000â&#x20AC;&#x201D;3,000 m D 500â&#x20AC;&#x201D;1,000 m ! I Sea levelâ&#x20AC;&#x201D;200 tn below sea level I

Fig. 1. Map of southern California. The Deep Canyoti Transect is indicated by the rectangle that extends froin the western fringe of the Colorado Desert into the Santa Rosa Mountains, part of the Peninsular Ranges that extend south into Baja California, Mexico.

660

ECOGRAPHY 22:6 (t999)


Fig. 2. Idealized cross section of the Deep Canyon Transect, showing the principal habitats encountered. The present study spans the region from Creosote-Palo Verde to Pifion-Juniper habitats. Figure modified slightly from Ryan (1968) with permission of The Palm Springs Desert Museum.

-2500 m Transition Life Zone Coniferous Forest

Chaparral

Upper Sonoran Life Zone

Pihon-Jurtiper

-1000 Lower Sonoran Life Zone

Agave-Oootillo

Rooky Slopes Cholla-Palo Verde

-500

Creosote-Palo Verde Mesquite Sand Dunes 8, Creosote

-'Si'^ail^lil^^.

again in non-floodplain valley floor habitats (Ryan's "cholla/palo verde" habitat). Richness then increased on rocky slopes, had a primary peak in agave/ocotillo habitat (which overlooks the rocky slopes), and declined through pifion-juniper, chaparral, and coniferous forests. Interestingly, however, rodent density (number/acre) was unimodally distributed (Fig, 3), declining both above and below the rocky slopes habitat; although the rocky slopes had only moderate species richness, this habitat had greater abundances than all other habitats sampled. In fact none of these relationships differ from a uniform distribution across elevation (linear regression, all p > 0.40, r-<0,12; no improvement in fit was obtained with a polynomial regression). However, these nascent patterns occur at or near the interface between two biotas, begging questions of local community structure and assembly across ecological frontiers. At intermediate sites along this gradient we might expect changes in species composition and relative abundances, atid consequent shifts in foraging strategies, spatial activity patterns, etc. In order to better understand the factors underlying the observed distributions of the matnmal fauna across an ecological transition. I surveyed small matrtmal communities at ten sites encompassing all major habitats across 1050 m of the Deep Canyon transect, spanning the peaks in both richness and abundance noted by Ryan (1968), Foraging and spatial ecology are not addressed in this report. Rather, I censused ECOGRAPHY 22.6 (19991

small mammal communities with live traps and markrecapture methods, and recorded extensive habitat metrics during both the summer (1996) and winter (1996/97) to evaluate elevational and seasonal shifts in richness, abundance, and habitat use.

Materials and methods Mammals Small mammals were censused with folding Sherman live traps placed singly in 7 x 7 trapping grids (15 m

Fig. 3. Mammal speeies richness (symbols) and total density (bars) reeorded by Ryan (1968) aeross the Deep Canyon elevational transeet. Labels on the abseissa correspond to habitat types used by Ryan. Numbers at the base of vertical bars give the area sampled for density estimates.

661


spacing). Assuming a boundary region of \ inter-trap distance these grids measured ca 1.1 ha in area. Traps were baited with millet seed and were opened in the early evening and checked before dawn. Animals were individually marked with numbered ear tags or by clipping fur, and species, sex, age (based on pelage), reproductive condition, and weight were recorded before releasing the animals at their point of capture. Grids were run for three consecutive nights in summer (June and July) 1996, and winter (December and January) 1996/97,

Habitats sampled Sampling effort was divided among the valley floor (two grids), rocky slopes (two grids), agave/ocotillo (two grids), and both lower and upper pifion-juniper (two grids each) habitats. The lowest end of the Deep Canyon Transect lies within the Coachella Valley, where agricultural and urban development precluded meaningful sampling. Because of a combination of fire hazard and difficulty of access, 1 was unable to sample chaparral and conifer habitats which occur at higher elevations. Brief descriptions of the sites studied follow.

very hot site, and is dominated by barrel cactus and small numbers of creosote. Many brittlebush are here as well, although these are deciduous and in the summer form mere skeletons of plants. Station site (T6S, R5E, SW I Sec. 17320 m) is located on a north facing slope with rocky soils. The dominant plants there are creosote and brittlebush, with scattered agave Agave descrii and ocotillo Fouqueria splendens.

Agave/Ocotillo Two sites (Agave Hill east, T6S, R6E, NW \ Sec. 19: Agave Hill west, T6S, R5E, NW 5 Sec, 19; both sites at 745 m) in this habitat type were located only several hundred meters apart, immediately above the steep walls that form the boundaries of Deep Canyon. These sites are dominated by ocotillo Fouqueria splendens, brittlebush, goldenhead Vigiiieria deltoidea, four o'clock Mirabiiis tenuiloba, galleta grass Hillaria rigida. and clumps of agave. Additionally, there are some barrel cacti as well as jumping O. bigelovii and deerhorn chollas.

Lower Pifion-Juniper (Lower plateau of Zabriskie 1979) Valley Floor - Cholla/Palo Verde habitat Two sites were sampled. These are not intended to be true replicates, as they were dominated by very different plant species, and had very different soils, Palo Verde site (T6S, R5E, NW 5 Sec, 16255 m) was located in a sandy wash dominated by palo verde Cercicliiim fiorideum. cat-claw acacia Acacia greggii, and small amounts of mesquite Prosopsis glandulosa, smokethorn Dalea spinosa, and creosote Larrea tridentata. Small numbers of barrel cactus Ferocactus acanthodes and deerhorn cholla Opwitia acanthocarpa round out the most important plant species found here, Cholla site (T6S, R5E, NW | Sec. 16, 260m) was dominated by a variety of plants including creosote Larrea tridentata and brittlebush Eiuelia farino.sa, indigo bush Dalea schottii, burrobosh Ambrosia dumosa, jojoba Simmondsia chinensis, and beavertail O. basilaris, pencil O. ramasissima, deerhorn O. acanthodes, and golden O. echinocarpa chollas.

Rocky slopes As above, the two sites on rocky slopes were designed to yield information on the diversity of habitats occurring in the rocky slope of Deep Canyon, and are not intended to serve as statistical replicates, Chuckwalla Hill (T6S, R5E, NW \ Sec. 17320 m) is a south facing slope covered with large boulders. As a result it is a 662

Two plots here were dominated by scattered piiion pine Pinus monophylta and California juniper Jimipcrus californicus. Other common plants included sugar bush Rhus ovata, scrub oak Quercus turbinata. crucillo Ziziphus parryi. western bernardia Bernardia incana, desert apricot Primus fremontii. snakeweed Guttierczia sarothrae. Virgin river encelia Fncelia virginen.sis. and buckwheat Eriogonum fasciculatum. Piiion Crest north (T6S, R5E, SW 5 Sec, 26, 1250 m) was located about 200 m from Piiion Crest south (T6S, R5E, SW | Sec. 26, 1250 m).

Upper Pinon-Juniper habitat (Upper plateau of Zabriskie 1979) One site (Observatory site, T6S, R5E, NW \ Sec. 34, 1280 m) was located near the Cecil and Ida Green Piiion Elat Observatory, operated by Univ, of California, San Diego. Dominant plants here were piiion pine and scrub oak, with good representation by California juniper, manzanita Arctostaphylos glauca. sugar bush, antelope bush Purshia tridentata, rabbit bush Crysothamnus teretifolius, and brickelia Brickelia oblongifolia. Other notable plants include golden cholla, pancake cactus O. chlorotica, prickly pear O. phaeacantha. and yuccas Nolina parryi and A', wotfi. The second site (Microwave site, T6S, R5E, SE 5 Sec, 34, 1295 m) was located near a microwave tower at the tXOGRAI'llY 22;(i (IW9)


Pinon Flat Observatory, Species composition here was similar to that at the previous site, with the addition of ribbonwood Adenostoma sparsifoliunu deerhorn cholla, and more grass cover under a slighly denser tree canopy.

Habitat metrics Thirty-one variables, reflecting the general structure and diversity of eaeh trapping point (Table 1), were recorded at each trapping point during both trapping sessions. Variables were evaluated for normality and transformed when possible. Slope, Aspect, # shrub sp. Shrub ht, and Cv-ground did not require transformation, D-tree was arcsin transformed, and # cholla, # holes, D-cholla, D-agave, D-shrub, Soil hardness, Cv-dead shrub, Cv-live shrub, Cv-herb, Cv-rock, Cvgrass, and Shrub area were log transformed. Other variables could not be normalized; these were used for analysis of habitat selectivities, which does not require normally distributed variables, but they were excluded from analyses requiring normally distributed data.

Analyses Local species richness (S) and densities (N = number of animals captured) were evaluated for patterns across elevation. Species diversity was calculated using the Shannon-Weiner index, H' = — Z pi In Pj, where P| is the proportion of the community comprised by species i. Habitat variables were entered into a principal components analysis to reduce the dimensionality of the data set and to increase the sample size/variable ratio. Principal components axes with eigenvectors greater than unity were retained for subsequent analyses. Mammal species distributions were analyzed in PC space by means of multivariate analysis of variance, and two a posteriori tests (Student-Newman-Kuels test and Scheffe's test) were applied to evaluate species differentiation on these axes. Habitat selectivities provide a distribution-free means of evaluating the extent to which species use nonrandom subsets of the available environment (see Patterson et al, 1990, Kelt et al, 1994, 1999), Because the absolute deviation from the available background is highly sensitive to sample size, a randomization routine was used to evaluate the significance of observed departures from random, using a two-tailed significance criterion of 5"A> (2,5% in each tail: see Kelt et al, 1994 for details of the randomization routine).

Results A total of 495 individuals of 11 species were captured ECOGRAPHY 22:6 119991

Table I. Habitat metrics recorded at Deep Canyon. Variables preceded by asterisks could be normalized, and these were used in the principal components analysis. Direct count measurements: Traploc

Trap location: arboreal, by trunk, litter, among herbs, by log, under shrub. * Aspect Coded numerically by cardinal compass points: 1 = N, 2 = NE, etc. *Slope Coded as 1 = < 5°, 2 = 5-20°, 3 = 2045°, 4 = >45°. * # shrub sp Number of shrub species within 3 m of trap marker. #herb sp Number of herbaceous species within 3 m of trap marker. #cacti Number of barrel cacti within 3 m of trap marker. * # cholla Number of cholla cacti within 3 m of trap marker. #agave Number of agave plants within 3 m of trap marker. * # hole Number of obvious burrows within 3 m of trap marker. #tree Number of dead trees within 3 m of trap marker. #sm log Number of small logs (7.5-15 cm diameter) within 3 m of trap marker. #lg log Number of large logs ( > 15 cm diameter) within 3 m of trap marker. Herb ht Mean height of herbaceous vegetation. *Shrub ht Mean height of shrubs. *D-cholla Distance to nearest cholla cactus. *D-agavc Distance to nearest agave. *D-shrub Distance to nearest shrub. Lxw-shrub Length x Width of nearest shrub. (Shrub area was calculated as pi*(0.5*LenShrub*Wid-Shrub). * D-tree Distance to nearest tree. Dbh Diameter at breast height of nearest tree. Litter depth Litter depth in each of four quadrats delineated by the line transects. *Soil hardness Soil hardness in each of four quadrats, using soil penetrometer. Measurements recorded along two 2-m transects (oriented N S & E W): *Cv-live shrub Cover of live shrubs, *Cv-dead shrub Cover of dead shrubs. *Cv-ground Cover by bare ground. *Cv-herb Cover by herbaceous plants. *Cv-rock Cover by rock. *Cv-grass Cover by grass. Cv-moss Cover by moss. Cv-log Cover by logs. Cv-water Cover by water. Estimated by eye: Canopy cover Canopy cover.

(overall \l.l"/u trap success; Table 2), including 308 captures of ten species in the summer (21'%. trap success), and 187 captures of nine species in the winter (14'Xi trap success). These included six species of heteromyid rodents {Cluwtodipus falki.x Merriam, 1889, C, foriiiosus Merriam, 1889, C. spinatus Merriam, 1889, Perognathus kmgimembris (Coues, 1875), Dipodomys agilis Gambel, 1848, D. merriami Mearns, 1890) and tive species of sigmodontine rodents (Neotoma lepida 663


QJ 3^

3 O n " . ' — ^ O ' ^ 0 » 0 ' — ' y^t

___^______O

px'

—— O O

(NOOr'-iOO

O -

Tt O O r^ - ^ - ^ O O ^+ i/~i './^ ^^^ ^^^ '^^ '•^^ ^^^ o^j f ^ l

O

u

^"^ f "^ t ' 1 o ^ f' ^ < ;

^^^ V

5;

OO'H S .E

o . bJ: ; — OJ O O

'E-a-5 o c i,

a. '^ -~ O ^ K .tt.C^i,.

— in — •

^ 0-

oooooooooo —

£

oooooor^oo ; =

j OOOO

-

OrsOOOO

2

8 =o

^2d

O O O O O O O — OO

ooooooooo

= 00000000

o o o o o o — r-oo

1^,11 u '-'

OOOt

OJ ' — . :

^

\O — O O O O O O

^ DO — ; O O O O

OOOOOOO

1 ^ 1 •= 't O O—O O O O - ^ °

— OO

n-, O O O O (^^ O O ^

: - ?: D.

S S § ? CJ OJ CO JZ nj QJ o id

E

CJ

:d

QJ

r" ioi

QJ

r-j tu

yon.

H

•evia

o a

S

X) QJ aj

5 (^ iJ QJ OJ O-

i.l

X) d

664

ECOGRAPHY 22:6 (1999)


Cholla/ Palo Vertie

Rocky Slopes

Agave/ Ocotillo

Lower P^

Upper P-J

X X ^

X

X^-:X

\ —

/

a

X ^ —^

\

\

40

\

Cholia/ Paio Verde

Rocky Slopes

Agave/ Ocotillo

\

Lower V~i

V

X 30

Fig. 4. Numbers of rodent species captured at Deep Canyon, California, in summer 1996 and winter 1996/97.

b

/ ( \

25

~'~~T \

20 -

\ ^ \

15 -

Distribution of species and abundances Species richness across the gradient appeared unimodal in winter (Fig. 5a), but was equally well described by linear and polynomial regression against elevation (Table 2), Summer data also appeared mildly unimodal, but neither regression models yielded significant results (Table 2), When summer and winter data were pooled, however, only a polynomial model approached significance (linear, F = 3,09, p > 0,10, r- = 0,30; polynomial, F = 3,78, 0.05 < p < 0,10, r^ = 0,35), Any incipient peak in richness was largely due to the very low richness in upper piiion-juniper habitat (Table 2). The distribution of population abundances differed greatly between seasons (Fig, 5b), As with species richness, the evidence for a peak at middle elevations was ambiguous. Winter data regressed signifieantly against elevation with both linear and polynomial models (Table 2), In contrast, summer data were poorly fit using a linear model, but approached significance (p < 0,10) with a polynomial model. Summer abundances E C O G R A P H Y 22:6 (1999)

10 5 0

Choiiay Paio Verde

Rocky Siopes

Agave/ Ocotiiio

Lower P^

Upper P-J

c

pecies Diversity (H'|

Thomas, 1893, Feromyscus boylii (Baird, 1855), P. crinitus (Merriam, 1891), P. eremicus (Baird, 1858), and P. truei (Shufeldt, 1885)), The most abundant species was C, falkix, followed by C. formosus and C spinatus (Fig, 4). Together, these three species constituted just over half of all individuals captured, although their dominance was highly seasonal ((yl"/n in summer vs 'ilVu in winter), reflecting lowered levels of activity during the cooler winter months. However, only two species (Peromyscus eremicus and Neotoma lepida) were more abundant in winter than in summer, suggesting that spring/early summer recruitment may have played a role in the numerical differences observed across these seasons. Two species (P. longimembris and P. boylii) were trapped only in summer, and one species (P. iruei) was captured only in winter.

Upper P,J

35

01

05

—T - Summer -&, Winter

00 %

"c

\

%„

*o \

Fig. 5. Rodent species richness and density (number captured per census using standardized trapping methods) in summer 1996 and winter 1996/97 across the Deep Canyon elevational transect. Labels on the abscissa are Grid names, while boxes at the top of each panel give the corresponding habitat types used by Ryan. The "X" symbols in the upper panel reflect the total number of species captured across both seasons.

were relatively high throughout much of the transect, with a possible peak in the agave/ocotillo habitat and rapid decline at higher sites. Winter abundances were unimodally distributed with a peak in the rocky slopes. The seasonal difference appears to reflect greatly reduced abundances at most sites in the winter - only sites in the rocky slopes and the upper pifion-juniper habitat (with very low numbers) remained similar in both seasons. The distribution of abundances paralleled that for species richness in winter but not in summer. Abundances were greatest in the agave/ocotillo habitat during the summer, but during the lower overall abundances of winter the peak shifted to the rocky slopes. The seasonal shift in abundances occurred across most species, but was particularly strong for the pocket mice 665


Table 3. Results of regression between elevation and both species and PC axes. Degrees of freedom are 1,8 for summer and 1,7 for winter. E-statistics in bold font indicate analyses that were significant after sequential Bonferoni adjustment. Species

Linear F

C. formosu.s C. fallax C. .spinatu.s P. longimenibris D. njerrianii D. agili.s P. boylii P. eriniius P. ereniieii.s P. iruci N. lepida

47.620 0.425 8.222 1.774 1.747 9.294 1.085 4.204 0.709 1.085 0.176 Principal component axes PC I 28.652 1.059 PC 11 2,030 P C III 5.844 PC IV PC V 0.005 0.594 PC VI PC VII 0.291

2 -

Polynomial ]â&#x20AC;˘-

F

0.737 0.024 0.326 0.094 0.093 0.354 0.060 0.198 0.040 0.060 0.010

33,209 0.171 7.849 1,874 1.154 11.196 1.144 6.055 1.101 1.144 0.645

0.661 0.010 0.316 0.099 0.064 0.397 0.063 0.263 0.061 0.063 0.037

0.628 0.059 0.107 0.256 0.0003 0.034 0.017

35.217 0.349 3.244 4.600 0.053 0.290 0.198

0.674 0.020 0.160 0.213 0,003 0.017 0,012

(Fig. 4), which ai-e much less active iti wititer, perhaps reflecting seasotial hibertnation or use of torpor. With equal trapping effort, nurnbers of the three principal pocket mice (C. formosus, C. fallax, and C. .spinatu.s) declined by 35-79'%i, Dipodomys agilis and P. criuitus also declined about 3()'yii, and Ncotonui increased by Ab%, but these species only occurred at relatively low numbers such that these proportional changes did not greatly influence comtnunity-wide patterns. Species diversity (H') also was best described with a polynotnial regression for both seasons, but was only marginally significant in summer (Table 2, Fig, 5c). Somewhat unexpectedly, only one speeies (C. fonno.sii.s)

Agave HillE

Agave HillW

Observatory

-

3

-

2

-

1

0

1

2

3

4

PCI

Fig. 6. Distribution of sampling grids in the first two principal component axes using summer habitat metrics. Sites were distributed very similarly in winter except that Palo Verde was not sampled. regressed significantly against elevation (Table 3). Three other species (C. spinatu.s, D. agilis, P. crinitiis) regressed significantly before critical values were Bonlerroni adjusted. Low satnple sizes tnay have limited statistical power on these analyses.

Habitat tnetrics Principal components analysis yielded five informative tnetries (A > 1.0; Table 4), The first PC axis (PC I) appears lo be largely an axis of ground cover and soil hardness, being strongly influenced by ground cover variables, with eover by rocks loading positively and that by open ground loading negatively. Soil hardness also loaded positively, as did slope and distance to the nearest tree. PC II refleeted shrub eover and diversity.

Table 4. Principal eigenvectors from a Principal Components Analysis on vegetative and habitat metrics collected at Deep Canvon in summer 1996 and winter 1996 97.

Slope Aspect Shrub sp #chola #hole Shrub ht D-cholla D-agave D-shrub D-tree Soil hardness Cv-dead shrub Cv-live shrub Cv-ground Cv-herb Cv-rock Cv-grass Shrub area

666

PRINl 4.43

PRIN2 2.51

PR1N3 1.75

PRIN4 1.47

PRIN5 1.02

0.389891 -0.088027 0.053391 -0.077257 0.294069 -0.134137 0.187678 0.174159 -0.127821 0.325383 0.415829 -0.000580 0.050200 -0.313307 -0.170808 0.421107 -0.212715 -0.117608

0.084283 -0.212492 0.424451 0.298966 -0.103130 0.009448 - 0 ^75^50 -0.282289 -0.388254 0.109233 0.015029 0.301708 0.395211 -0.063070 0.191485 0.048808 0.161103 -0.053293

0.071150 0.057760 0.194242 -0.376089 0.224001 0.424542 0.360743 0.118710 -0.177537 -0.156708 -0.109747 0.054214 0.314705 -0.142655 0.017972 -0.115740 0.198825 0.440554

-0.159784 -0,270308 0.031630 0.259086 -0.114017 0.399275 -0.241807 0.355634 0.080798 0.273086 0.008148 -0.075169 0.068950 0.227589 -0.276142 0.110567 -0,306130 0.387743

0.105393 -0.351383 -0.109495 0.000629 -0.006937 0.228271 -0.161089 -0.194626 0.477208 0.012904 0.152975 0.164781 -0.308004 -0.384939 0.147318 0.112811 0.369024 0.212949

tiCOGRAPHY 22:6 (1999)


with number of shrub species and cover by both live and dead shrubs loading positively and distance to cholla and to shrubs loading negatively. PC III loaded positively with shrub area and shrub height, as well as with distance to cholla and with live shrub cover, and negatively with cholla cover. Thus, this axis also appears dominated by shrub structural elements, PC IV appeared to reflect a shrub/grass gradient, being strongly influenced by shrub height and area, distance to agave (all positive), and ground cover by grass (negative). F'C V also was influenced by shrubs and ground cover, with strong loadings by distance to shrubs, ground cover by grass (positive loadings), and by bare ground cover, aspect, and live shrub cover (negative). Because interpretation of these patterns rapidly became difficult, discussion is generally limited to the first two PC axes; these axes described just under 40'^;. of the total variance in the data set. 3 , PEBO

PEEI

2 . 1 .

PGLO

1 O

a

>

CHFA

'

i 1 NELE

0-

1

PECR CHSP —1

-1 -

1

1 1 DIAG

1

1> DIME

1

Total CHFO

-2 •

-3 .

PCI

2 • PEER 1 = 1

CHFA 1 1 (> PETR

O Q.

1

1 Total

NELE •1 PECR

0 DIME

-in

CHFO

''

CHSP

-1 DIAG -2 -

-3 .

Trapping sites were arrayed in a counter-clockwise pattern in PC space (Fig, 6), indicating a non-linear but readily interpretable distribution of data in PC space. As a result of this orientation, some sites clustered together that were not located near to each other along the transect, reflecting similar habitat characteristics at spatially disjunct parts of the transect, Chuckwalla Hill, from the rocky slopes, was characterized as extremely rocky and barren of much vegetation, and this site polarized the first PC axis. This site was most closely allied with the Station plot (also from the rocky slopes), which was similar in having a steep aspect and limited vegetation. Two sites from the valley floor (Palo Verde and Cholla) grouped more closely to the highest elevation sites (Observatory and Microwave) than to other low elevation sites (Station and Chuckwalla). This likely reflected the relatively soft soils and the large percentage of cover by open ground at these sites. Additionally, these sites all had a fairly limited diversity of shrub species which tended to be located near to each other. When small mammal captures are superimposed onto principal components space the apparent incongruities in the distribution of sites become clearer, as species are arrayed in an ecologically reasonable sequence. It also is clear that species overlap extensively in PC space (Fig, 7), although segregation is apparent on the first four PC axes (ANOVA, p < 0.0001 for axes 1, 2, and 4, p = 0,0167 for axis 3). A posteriori tests on each PC axis demonstrated habitat segregation by species (Table 5), On axis I, C. spinatus. P. criniius. and A', lepida are distinct from D. agilis, P. triiei, P. boytii, and P. tonginienibris. Varying levels of similarities are exhibited by the remaining species. The clarity of these distinctions diminished on subsequent axes, although segregation of PC space remains apparent at a fairly coarse level. There was a general congruence in species placement in PC space between summer and winter data sets (Fig, 8). Nonetheless, on the first three PC axes, Neotoma, P. criniius. and C. .y^natus converged in the winter, whereas C. Jatla.x and P. eremicus diverged on axes I and II. When viewed in PC Il/III space, D. nierriami and C. fonnosus converged in winter, as did P. crinitus and

Neotonia. Finally, only the first principal component axis regressed significantly with elevation (Table 3), suggesting that it captured elements of environmental variation that reflected changes across the gradient. All other principal component axes were unrelated to elevation.

PCI Fig. 7. Distribution of rodent species in the first two principal component axes, using summer a) and winter b) data. Abbreviations as in Table 2: CHFA, C/kicto(lipiis fiillax: CHFO, C. formosus; CHSP, C. spiiuitu.s- DIAG, Dipoc/onns agiii.v. DIME, D. luciriaiui: NELF, Neotoiiui Icplila: PFBO, Pciomysciis hoyiii: PECR, P. ainiius: PEER, P. cremiciis: PETR, P. tnici: PGLO, Pcrogiuitiiiis longimembris. Total, all species combined. ECOGR.APHY 22:6 (1999)

Habitat selectivities Most taxa were non-randomly distributed on various habitat axes (Fig. 9). Additionally, the number of significant deviations was positively related to number of captures for summer (y = 0.1065x + 11,32, r - = 0,42, p < 0.025, N = 308 captures of 10 species) but not winter 667


Table 5, Results of a posteriori tests of the distribution of species on the ftrst four priticipal comportent axes. Species are arranged vertically, and species sharing letters are not significantly different on a given PC axis. Results for both SNK and Scheffes tests are presented.

CHFA CHFO CHSP DIAG DIME NELE PEBO PECR PEER PGEO PETR

Prin4 1,47

Prin3 1,75

Prin2 2,51

Prinl 4,43 SNK

Scheffe

SNK,

Seheffe

SNK

Scheffe

SNK

Scheffe

bed ab a d cd a d a abc d d

abed abc a d bed ab cd a abed ed d

ab b b b b b a b ab ab ab

ab ab ab b b ab a ab ab ab ab

a

a

a

a a a a

a a a a

ab ab ab b a ab ab ab ab ab ab

a

a

a a a a a a

(p>0,10, N = 187 captures of 9 species): the greater number of captures in sumtrter likely provided for greater resolution than during the winter. Habitat selectivities generally agreed with the limited and generally anecdotal information published on habitat use by these speeies. Perhaps the most useful means of viewing these data is to evaluate divergence among functionally similar speeies.

a a a

a

a

a a a

a a a a a a

ceous growth in the summer, but with taller shrubs and low cover by logs and herbs in the winter, Chaetodipus fallax was found at sites with low grass cover and close proximity to oeotillo in the summer but taller grasses and greater distanees to oeotillo in 'winter. Finally, sites

Pocket mice All four pocket miee exhibit a large number of significant deviations from the available habitat. Three Chaetodipus were found at sites with more cacti, lower eanopy cover, less grass and shrub cover, greater distance to trees, and shallower litter than expected by chance, Perognathus differed frotn Chaetodipus in selecting sites with greater eanopy cover than random, whereas Chaetodipus species selected sites with significantly less eatiopy cover than random. Additionally. Perognathus was found much closer to trees, at sites with greater litter cover and less shrub cover, and with fewer cacti and more grass, than the three Chaeotodipus. Perognathus and C, falla.x shared some selectivities. such as looser soils, cover by live shrubs, greater proximity to both chollas and agaves, distance to shrubs, and ground cover by roek, Chaetodipus spinatus was found further than expected from shrubs and agaves, and at sites with tall herbs and little cover by logs, Chaetodipus formosus was found with less live shrub cover than other pocket mice in the summer. The most abundant rodent at Deep Canyon. C. falla.x. evidently favored sites with more ground cover of looser soils, but also more herbaceous cover, and less open rock, greater cover by shrubs, fewer agaves and cholla. and fewer apparent holes, than other pocket mice. There were several interesting seasonal shifts in habitat use (Fig, 9), Chaetodipus formo.ms seleeted sites with low shrub height and high cover by logs and herba668

o Q.

PCI PETR o

b

o \

PGLO \

\

. PEBO

\pEER

DIAG

f

\ CHFA

P PECR

Totari \

DIME/

CHSP

CHFO ^NELE

o

PCI Fig, 8, Shifts in centroid positions in principal components spaee between summer (filled symbols) and winter (open sytnbols). Abbreviations as in Fig, 6, ECOGRAPHY 22;6 (t999)


SUMMER

WINTER

C, formosus N = so, 33

C. faltax N = 101, 24

C. spinatus N = 41, 20 ']

p.

*

*l^

longimembris mbris U •—j» 1 __• «__,»_ t^=9,o \ B ^ ^ T o i P * " " ^ " ' ' ' - ^ — ^ ^ ~

*

*

* * *

* * * *

*

*

in

I

D. agilis N = 16, 16

D. merriami N = 15, 13

P. boytii N=1,0

P. crinitis N = 16, 18

P. eremicus N = 12, 43

p. truei N =

0,3

Neotoma lepida N = 17, 17

Fig, 9, Habitat selectivities of small mammals across the Deep Canyon elevational transect in summer 1996 and winter 1996,97. Use of measured habitat characteristics are presented as ratios of mean use to the mean value of these metrics across all stations. Asterisks indicate variables for which species differ significantly from the available range of a given characteristic. Ticks on the ordinate axes represent unit deviation. Habitat characteristics are labeled on the ordinate.

with C spinatus were characterized as havitig relatively tall herbs in the sumtner but low herbs in winter. Whereas some seasonal shifts might reflect changing distribution of environmental features (e,g,, grasses or herbaceous eover, shrub height), the distribution of logs did not chatige between these seasons, suggesting that this was a real shift in mierohabitat use by C. formosus.

Kangaroo rats Differences between these species are expected and largely reflect their general association with Lower and ECOGRAPHY 22:6 |I999|

Upper Sonoran (D. merriami) and Transition (D. agilis) life zones. Nonetheless these species exhibited many more similarities than differences in terms of the habitat tnetrics recorded. Of the 10 and 11 metrics on which both taxa deviated significantly in summer and winter, respectively, they deviated in the same direction on 9 and 7 of these (ca 15% similarity). These species differ in terms of canopy cover and litter depth in both seasons (not significant in winter for D. merriami), and distance to shrubs in the winter (all favored by D. agilis. disfavored by D. merriami). Additionally, both speeies were found at sites with signifieantly less ground eover by roeks, greater distances to oeotillo. 669


low cover by herbs and shrubs, low slope, and loose soils. Both were found at sites rnoderately (but signifieantly) close to shrubs, with less herbaceous ground cover than expected in summer but more in winter, greater log eover in summer but less in winter, and with fewer shrub species than expected in winter.

Sigmodontine rodents Only three sigmodontine rodents (P. crinitus, P. eremicus, and Neotoma) were captured in both seasons, and the remaining two species {P. boylii, P. truei) were captured at very low numbers. Many more similarities than differences were apparent in these species' use of habitat. In both seasons all three speeies favored sites with low canopy cover, close to oeotilio, low litter cover, and relatively hard soils. Additionally, all three of these species seleeted sites with low grass cover, close proximity to shrubs, and moderate but significant distanee to trees in the summer. All used sites low in herbaceous cover, except for Neotoma in summer. High slopes were used by these species, exeept for P. eremieus in summer, and they all selected sites with low ground cover by logs, except for P. crinitus in the summer, which favored sites with high log cover. In winter both P. crinitus and Neotoma were found relatively close to shrubs and with short herbaeeous growth, whereas P. eremicus favored sites with tall herbs located away from shrubs: interestingly, however, the former species also seleeted sites with relatively low cover by live shrubs, and the latter were found at sites with greater shrub cover, suggesting that the distribution of shrub sizes may be inversely related to shrub abundanee along this transect. It is worth noting that some of the significant results obtained in this analysis are trivial consequences of different macrohabitat use. For example, C. spinatus occurs on rocky slopes where logs are not likely to oecur, and D. merriami and all three Chaetodipns typically occupy sites completely or nearly laeking in trees. Nonetheless these results quantify important features of the ecological gradient that are important to the mammal species foutid there.

Discussion In spite of many years of study substantial gaps remain in our knowledge of the ecology of small mammals. The ecological distribution of many species, and the parameters that influence their distributions, have generally been studied only superficially, and much of our understanding is based on observations recorded by field workers early in this century. In this study I have documented distribution patterns of small terrestrial mammals from two faunal districts across a major

670

elevational gradient, in terms of species richness, population abundance, and habitat associations. Habitat associations repotted here supplement state-; ments found in the literature, and provide insights into the tneans by which some species may be segregating resources along this gradient. For example, although the literature reports that the three Cluietodipus species reported here generally have distinct macrohabitat preferences (C. spinatus, rocky habitats; C. formosus, hardpacked gravel, stable alluvium; C. falki.x, often near rocks but also into silt and fine sand habitats; Miller and Stebbins 1964, Ryan 1968, Hoffmeister 1986) they all may be captured sympatrically and even syntopically, albeit not commonly. It is not clear, however, whether this syrnpatry is a consequence of seasonal dispersal (and therefore eonstitutes a fleeting snapshot that would re-sort within a period of days), is an epiphenomenon of resource abundance (lasting until populations approach carrying capacity and again competitively sort among habitats), or if some alternate explanation exists. These speeies differ slightly in body size, but if this were sufficient to allow sympatry then we should see co-occurrence more frequently, and this is not the case. Abramsky et al, (1990) demonstrated that species with similar habitat preferences may be peteeived as seleeting different habitats if one of the speeies is a superior competitor, such that the second speeies is rnore abundant in less preferred habitat. Data presented here suggests that syntopy among these pocket mice rnay occur only ephemerally, such as during dispersal following spring recruitment. The role of interspecific competition vs true habitat preferenee, however, is not clear. Habitat assoeiations in principal components space also reflected the general associations that are observed between speeies, supporting the relevance of PCA to provide insights into patterns of habitat segregation. For example, Peromvsetis boylii, P. truei, Perognathus longimembris, and Dipodomys agilis (San Diegan taxa typical of shrubby habitats) anchored one end of PC axis I, whereas Peromvseus erinitus and Chaetodipus spinatus (desert taxa generally found in exposed rocky habitat) clustered at the opposite pole of this axis, Cluietodipus Jalla.x and Peromyseus eremicus are relatively generalist in their habitat requirements, and they lie at the middle of this axis, while the modei ate habitat specialists Neotoma, C. formosus, and D. merriami fall between the polar habitat specialists and the centrally located generalists. The distribution of richness and abundance was somewhat different from that reported by Ryan (1968) for rodents in the same habitats at Deep Canyon, Speeies riehness was lower at every point in the transition than reported by Ryan, and it peaked at a lower section of the gradient. Differences in richness likely reflect the scale of observation; Ryan listed taxa found in broad habitat types, whereas I have reported animals ECOGRAPHY 22:6 (1999)


captured at specific trapping grids. Although Ryan included diurnal rodents {Ammospermophilus, Spermophilus) in his report, inclusion of these ta.xa in the present report would not bring the observed richness to the level reported by Ryan. Differences in richness may also reflect the use of different trapping methodologies. To estimate abundances Ryan used a single trapping "quadrat" in each of the macrohabitats he studied, but these quadrats varied in size and in the distribution of traps according to the density of vegetation and of rodent burrows. Thus, quadrats ranged in size from 0.52 acres (pifion/juniper) to 1.08 acres (cholla/palo verde). and although traps were placed in rectangular arrays the inter-trap distance was largely set by "placing traps at approximately one per rodent burrow, or clump of plants containing a burrow, so that the area was saturated by the 100 traps used" (Ryan 1968, p. 126). While this is a perfectly justifiable protocol it also results in different sized sample units and difficult interpretation. I have chosen to use sample units of constant area (ca 1.1 ha) to avoid this potential confusion. Total abundances were distributed somewhat differently than reported by Ryan (1968), with a moderate peak at a higher elevation than in Ryan's study. This could also be a consequence of different trapping methods, but it might reflect real changes in the distribution of abundances since Ryan's surveys. The substantial variation in total population abundances and, to a lesser degree, in species richness, between seasons, underscores the temporal variability of such metrics. It seems reasonable to conclude, therefore, that patterns of distribution across this gradient may be dynatnic in time and that local effects (e.g., variations in shrub cover, forage availability, moisture, etc.) may substantially influence these patterns. For example, the El Nilio southern oscillation event of 1998 (a year after this study was conducted) resulted in greatly elevated precipitation throughout southern California, and small mammal numbers rose concomitantly. Habitats that are of low value to some species in years of normal precipitation may become suitable after events such as El Nifio, and emigration may substantially alter local composition and ecological dynamics. Elevated species richness or abundances at intermediate points along ecological gradients have been noted by many authors. Such bulges range from familiar edge effects at small spatial scales (e.g., Harris 1988, Yahner 1988, see also Heske 1995) to patterns over larger ecological transitions (Shmida and Wilson 1985, Patterson et al. 1989, Kelt 1994, Shepherd and Kelt in press) or even biogeographic scales (e.g., Owen 1988, Rosenzweig 1992). Factors resulting in such bulges generally are assumed either to reflect increased resource availability at intermediate regions, or overlap of faunas from either end of a transition. Under the former scenario one would predict a gradual addition of species towards intermediate points along the gradient. ECOGRAPHY 22:6 (1999)

and that titness would be greatest at intermediate sites along these gradients. In the latter scenario intermediate locations may be only marginally suited to the species present such that fitness is predicted to be lower for most species, or populations may only survive in intermediate regions by immigration (e.g., Shmida and Wilson's 1985 mass effects model; see Kunin 1998). It is unlikely that mass effects (sensu stricto) are responsible tor the patterns observed at Deep Canyon, as a number of animals within the regions of highest species richness were observed to have descended testes or to be pregnant, lactating, or otherwise reproductively active. Additionally, the distribution of abundances (especially in summer) supports an argument for declining resource availability at elevations above the agave/ocotillo zone, but only minimal evidence for similar declines at lower elevations. Finally, the small mammal fauna at the richest sites is almost entirely comprised of desert taxa; San Diegan faunal elements only become apparent in the lower and upper piiion-juniper habitats (Ryan reported D. agilis and P. loiii;iiiwmhri.s in agave/ocotillo habitat, but this was not confirmed in this study). Thus, although further research on this would be useful, it seems most likely that reduced resource availability is responsible for diminishing rodent abundances above the agave/ocotillo habitat. If abundance is positively related to species richness (e.g., Preston 1962) then this might also explain the distribution of species richness across this transition. This is a different interpretation, however, than offered by Shepherd and Kelt (in press) for the entire mammal fauna across the same elevational transect. Basing their arguments on the data reported by Ryan (1968) they noted that intermediate sites with the greatest number of species also contained a number of sites typical of both Sonoran and San Diegan faunas. This suggests that faunal overlap may be responsible for elevated richness of larger mammals, but does not preclude the possibility that this in turn is a consequence of a more hospitable habitat. The habitat metrics recorded along this transition suggest that the distributional patterns observed are not simple responses to changes in elevation. Selectivities by many species reflect differential use of habitat characteristics, many of which change markedly across the gradient, providing a broad gradient in habitat types and vegetative structure. Many variables (e.g., number of herbaceous species, agaves, trees, and both small and large logs; shrub height and area; distance to shrubs and ocotillo; litter depth; canopy cover; ground cover by bare ground, logs, grass, herbaceous plants, and by both live and dead shrubs) were significantly positively associated with elevation, whereas other variables (e.g., number of cacti and of holes; distance to agave, cholla, and trees; soil hardness; ground cover by rocks) were significantly negatively associated with elevation (all 671


p < 0,0001 after Bonferoni adjustment). However, most of these regressions explained very little of the variance in these relationships (r- > 0,50 for distance to tree and for ground cover by rock; r- > 0,20 for distance to agave, soil hardness, and ground cover by grass; r~ < 0,20 for all other analyses). Thus, significant associations with the habitat characteristics reported here do not imply gradients with elevation. Rather, these likely reflect selection for habitat types that partially transcend elevations, and it is clear that many species are more common in some habitats than others. For example, species that require loose soils and open habitats, such as D. merriami, are unlikely to fare well in the rocky soils characteristic of the rocky slopes and the agave/ocotillo habitats, but two individuals were captured at higher elevations in the lower pifion-juniper habitat, Chaetodipus spinatus and P. crinitus are relatively saxicolous and therefore unlikely to occur in either upper or lower pifion-juniper or valley floor habitats, although Ryan (1968) reported the latter species from flood plain habitats. It is worth reiterating that selectivities recorded here do not necessarily imply true preferences. Competitive exclusion by the larger C. formosus could be excluding C. spinatus from preferred habitat on the valley floor, resulting in apparent preference for sites with rocky substrate, similar to the situation documented for two gerbil species by Abramsky et al, (1990), Whether the three Chaetodipus that occur in the valley of Deep Canyon occupy different macrohabitats by preference or subsistence is not certain, therefore, and would require experimental manipulations which have not been implemented. Because of access limitations and concerns about fire hazard, this study was not able to sample the entire Deep Canyon Transect, At the lower end of the transect, sand dunes in the Coachella Valley have been largely altered by development or agriculture. At the other end of the transect, limited access and fire hazard prevented us from sampling chaparral and coniferous forest. Data presented by Ryan (1968) suggest that inclusion of these habitats would accentuate the hump in species richness by adding depauperate sites at either end of the gradient (Fig, 3), although this is not a certain outcome. The observed seasonal changes in community composition suggest that species at Deep Canyon may be focusing their activities in those seasons in which they are particularly efficient at foraging. Seasonal rotation in foraging efficiency has been documented among rodents in the Sonoran Desert (Brown 1989, see also Kotler and Brown 1988), but not in the Negev Desert (Brown et al, 1994), Further work on seasonal changes in small mammal communities would be a fruitful area of work. Across a non-elevational transition from forest to steppe in southern South America, Kelt (1996) noted 672

the general uniformity of population and community characteristics of small mammals, leading him to suggest that local processes (e,g,, resource availability) may compensate for geographical dynamics (e,g,, faunal overlap) to yield surprisingly uniform patterns of species richness, community evenness, and biomass across the transition. Such constancy does not characterize elevational gradients in southern Chile (Patterson et al, 1989, Keh et al, 1999) or southeast Asia (Philippines, Rickart et al, 1991; Malaysia, Langham 1983) and does not appear to characterize the elevaticMial gradient studied here. It may be that patterns and perhaps even processes operating across elevational transects are fundamentally different than those operating across "horizontal" ecological gradients involving minimal change in elevation. Since temperature and precipitation generally covary with elevation, and in turn influence ecological processes from rates of soil mineralization and decomposition to potential evapotranspiration, it may not be surprising that gradients lacking such pervasive influences would be structured very differently from gradients exhibiting such processes. The dynamics of ecological transitions continue to provide insightful natural experiments into the factors that structure ecological communities. Further research should be directed towards quantifying the differences between elevational and non-elevational gradients, and how these differ among different groups (e,g,, mammals, birds, insects, plants, etc), Aci<mm!edgemcnts This study was paitially funded by Faculty Research Grants from the Univ, of California. Al Muth and Mark Fisher provided critical logistical support at Deep Canyon, and Patrick Aldrich, Victoria Dye, Chantal Green, and Rodolfo Santos all assisted with collection of data. Burt Kotler reviewer provided useful critiques of this paper.

References Abramsky, Z. et al. 1990. Habitat selection: an experimental field test with two gerbil species. - Ecology 71: 2358-2369. Brown, J. S. 1989. Coexistence on a seasonal resource. - Am. Nat. 133: 168-182. Brown, J. S., Kotler, B. P. and Mitchell, W. A. 1994. Foraging theory, patch use, and the structure of a Negev Desert granivore community. - Ecology 75: 2286-2300. Burk, J. H. 1988. Sonoran Desert vegetation. - In: Barbour, M. G. and Major, J. (eds). Terrestrial vegetation of California, expanded ed, California Native Plant Soc. Spec, Publ. 9: 869-889. Harris, L. D. 1988. Edge effects and conservation of biotic diversity. - Conserv. Biol. 2: 330 332. Heske, E. J. 1995. Mammalian abundances on forest-farm edges versus forest interiors in southern Illinois: is there an edge effect? J. Mammal. 76: 562-568. Hoffmeister, D. F. 1986. Mammals of Arizona. - Univ. of Arizona Press. Karieva, P. 1994, Space: the final frontier for ecological theory. - Ecology 75: 1. Kelt, D. A. 1996. Ecology of small mammals across a strong environmental gradient in southern South America. - J. Mammal. 77: 205 219. ECOGRAPHY 22:6 (1999)


Kelt, D, A,, Meserve, P, L, and Lang, B. K. 1994, Quantitative habitat associations of small mammals in a temperate rainforest in southern Chile: empirical patterns and the importance of ecological scale, - J, Mammal, 75: 890-894. Kelt, D, A, et al, 1999, Scale dependence and scale independence in habitat associations of small mammals in southern temperate rainforest, - Oikos 85: 320-334, Kikkawa, J, and Williams, E, E, 1971. Altitude distribution of land birds in New Guinea, - Search 2: 64-65, Kotler, B. P. and Brown, J. S. 1988. Environmental heterogeneity and the coexistence of desert rodents. - Annu. Rev. Ecol. Syst. 19: 281 307. Kunin, W. E. 1998. Biodiversity at the edge: a test of the importance of spatial "mass effects" in the Rothamsted Park grass experiments. - Proc. Nat, Acad. Sci. U.S.A. 95: 207-212. Langham, N. 1983. Distribution and ecology of small mammals in three rain forest localities on peninsular Malaysia with particular reference to Kedah Peak, - Biotropica 15: 199-206. Levin, S, A, 1992. The problem of pattern and scale in ecology. - Ecology 73: 1943-1967. Miller, A. H. and Stebbins, R. C, 1964. The lives of desert animals in Joshua Tree National Monument. - Univ. of California Press. Owen, J. G. 1988, On productivity as a predictor of rodent and carnivore diversity. - Ecology 69: 1161-1165, Patterson, B. D., Meserve, P. L. and Lang. B, K, 1989, Distribution and abundance of small mammals along an elevational transect in temperate ramforests of Chile. - J. Mammal. 70: 67-78. Patterson, B. D., Meserve, P. L. and Lang. B. K. 1990. Quantitative habitat associations of small mammals along an elevational transect in temperate rainforests of Chile. J. Mammal, 71: 620-633, Pearson, O. P, and Pearson, A. K. 1982. Ecology and biogeography of the sothern rainforests of Argentina. - In: Mares, M. A. and Genoways, H. H. (eds). Mammalian biology in South America. Special Publ. Ser., Pymatuning Lab of Ecology, Univ. of Pittsburgh, Vol. 6: 129-142

ECOGRAPHY 22:6 (1999)

Preston, F. W. 1962. The canonical distribution of commonness and rarity. - Ecology 43: 410-432. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern? - Ecography 18: 200-205. Ralph, C. J. 1985, Habitat association patterns of forest and steppe birds of northern Patagonia, Argentina. - The Condor 87: 471-483. Rickart, E. A., Heaney, L. R. and Utzurrum, R. C. B. 1991. Distribution and ecology of small mammals along an elevational transect in southeastern Luzon, Philippines. J. Mammal. 72: 458-469. Rosenzweig, M. L. 1992. Species diversity gradients: we know more and less than we thought. - J. Mammal. 73: 715730. Ryan, R. M. 1968. Mammals of Deep Canyon, Colorado Desert, California. - The Desert Museum, Palm Desert, CA. Shepherd, U, L. and Kelt, D. A. 1999. Mammalian species richness and morphological complexity along an elevational gradient in the arid southwest. - J. Biogeogr. 26: 843-855. Shmida, A. and Wilson, M. V. 1985. Biological determinants of species diversity. - J. Biogeogr. 12: 1-20. Terborgh, J. 1977. Bird species diversity on an Andean elevational gradient. - Ecology 58: 1007-1019. Thorne, R. F. 1988. Montane and subalpine forests of the Transverse and Peninsular Ranges. - In: Barbour, M. G. and Major, J. (eds). Terrestrial vegetation of California, expanded ed. California Native Plant Soc. Spec. Publ. 9: 537-557. Wiens, J. A. 1989. Spatial sealing in ecology. - Funct. Ecol. 3: 385-397. Whitaker, R. H. and Neiring, W. A. 1965. Vegetation of the Santa Catalina Mountains, Arizona: a gradient analysis of the sotith slope. - Ecology 46: 429-452, Yahner, R. H. 1988. Changes in wildlife communities near edges. - Conserv. Biol. 2: 333-339. Zabriskie, J. G. 1979. Plants of Deep Canyon and the eentral Coachella Valley, California, - Univ. of California Press.

673


Ecography 34: 123131, 2011 doi: 10.1111/j.1600-0587.2010.06371.x # 2011 The Authors. Journal compilation # 2011 Ecography Subject Editor: Nathan J. Sanders. Accepted 6 April 2010

The impact of sterile populations on the perception of elevational richness patterns in ferns Michael Kessler, Sandra Hofmann, Thorsten Kro¨mer, Daniele Cicuzza and Ju¨rgen Kluge M. Kessler (michael.kessler@systbot.uzh.ch), D. Cicuzza and J. Kluge, Systematic Botany, Univ. of Zurich, Zollikerstrasse 107, CH-8008 Zurich, Switzerland.  S. Hofmann, Albrecht-von-Haller-Inst. of Plant Sciences, Univ. of Go¨ttingen, Untere Karspu¨le 2, DE37073 Go¨ttingen, Germany.  T. Kro¨mer, Centro de Investigaciones Tropicales, Univ. Veracruzana, Interior de la Exhacienda Lucas Martı´n, Privada de Araucarias s/n, Col. 21 de Marzo, C.P. 91019 Xalapa, Veracruz, Me´xico.

Dispersal may influence the spatial distribution of species richness through mass or source-sink effects, but the extent of sink populations at the community level remains largely unknown due to difficulties of identifying such populations. We compared the richness patterns of ferns in 333 plots along six tropical elevational gradients in America, the Mascarenes, and southeast Asia, using sterile populations as an indication of sink populations. First, we tested whether sterile fern records were more common towards the elevational range limits of the individual species, but found this pattern in only one out of ten cases. It is therefore uncertain if sterile records correspond to marginal sink populations. Second, we compared the elevational richness patterns of sterile and fertile species. In several cases, elevational trends for sterile and fertile records were quite similar, but in others they differed distinctly. The percentage of sterile records per plot decreased with elevation among epiphytic ferns along all six transects, whereas terrestrials showed mixed results (decrease, increase, and U-shaped patterns). The percentage of sterile species records per plot relative to the number of species per plot recovered four significant patterns among the twelve cases analysed: higher percentages at higher species numbers among terrestrial ferns on two transects and lower percentages among epiphytes on two others. Despite the problems with equating sterile records to sink populations, we thus found distinct elevational patterns of sterile records that clearly affected our perception of the overall richness patterns. Ignoring the impact of population dynamics on diversity patterns is thus liable to result in misinterpretations of the diversity patterns.

Geographical patterns of species richness are determined by a wide range of factors at different spatial and temporal scales (Mittelbach et al. 2007). These factors involve such disparate aspects as habitat area (Rosenzweig and Ziv 1999, Lomolino 2001), evolutionary and historical factors (Wiens and Donoghue 2004, Ricklefs 2005), and local ecological conditions (Hawkins et al. 2003, Currie et al. 2004, Evans et al. 2005). Many of these aspects have been extensively studied in the last decades. However, the number of species co-occurring at a site is also influenced by dispersal, an aspect that has received less attention because of the difficulties involved in documenting and quantifying dispersal (Myers and Harms 2009). At the local scale, dispersal can influence patterns of species richness either through dispersal limitation, where potentially suitable habitats are not occupied because propagules fail to reach these sites (Eriksson and Ehrlen 1992, Tilman 1997, Turnbull et al. 2000), or due to mass or source-sink effects (Shmida and Wilson 1985, Pulliam 1988, 2000, Dias 1996). The latter occurs when diaspores of a species are dispersed to suboptimal habitats where the species survive but where they are unable to produce enough offspring to maintain

self-sustaining populations. Although sink populations have been documented in numerous individual species (Gilpin and Hanski 1991, Wilson 1992, Leibold et al. 2004), the extent of such populations at the community level and therefore their influence on richness patterns remains largely unexplored (Myers and Harms 2009). One geographical system along which the influence of such source-sink dynamics may be studied are elevational gradients (Kessler 2009). Elevational patterns of species richness have received increased attention in the last two decades (Rahbek 1995, 2005, Kessler 2001a, Lomolino 2001, McCain 2009), partly because they represent unambiguous gradients that are replicated in different mountain systems. Dispersal has been invoked as a factor to explain elevational patterns of species richness via two different mechanisms. Stevens (1992) proposed that lower elevations have increased richness because species occurring under less extreme, tropical environmental conditions have narrower ecological ranges, resulting in narrower geographical ranges and hence a higher frequency of range margins. This in turn increases the potential for dispersal beyond the source areas. In contrast, maximum species 123


In the present study, we compared the richness patterns of ferns along six tropical elevational gradients in America, the Mascarenes, and southeast Asia. Our aim was to assess the potential impact of source-sink dynamics on elevational richness patterns of ferns, using sterile populations as an indication of sink populations. Ferns are particularly suitable for this kind of study because they are distributed worldwide with a large, but manageable number of species, thus allowing for replicated, quantitative studies (Barrington 1993). Furthermore, in humid tropical mountain forests with their limited climatic seasonality, the majority of fern species is fertile during all or most of the year, reducing the potential impact of the timing of the surveys on the recorded patterns (Sharpe and Mehltreter 2010).

richness at mid-elevations has repeatedly been hypothesized to be due to dispersal of species from both lower and higher elevations, resulting in highest overlap of sink populations at mid-elevations (Rahbek 1997, Kessler 2000b, Lomolino 2001, Grytnes and Vetaas 2002, Grytnes 2003a, b, Kattan and Franco 2004). Already in 1961, van Steenis observed that comparatively low mountains in Java lacked plants at higher elevations that were present at similar elevations on higher mountains. He suggested that on high mountains these species dispersed downwards from the highest peaks to lower elevations and were thus absent on mountains lacking such source habitats (van Steenis 1961). Beyond such descriptive accounts, however, quantitative studies are very rare. One problem is that conducting detailed population-level studies of numerous species at multiple elevations is prohibitively expensive and time consuming (Dias 1996, Diffendorfer 1998). Typically, source populations have higher reproductive rates than sink populations (Pulliam 1988, Robinson et al. 1995, Hansen and Rotella 2002). This implies that, unless vegetative reproduction is involved, populations consisting entirely of sterile individuals are more likely to represent sink populations than populations with fertile individuals. Grytnes et al. (2008) showed that along several Norwegian elevational gradients sterile populations of individual plant species were significantly more frequently found at the upper and lower ends of the elevational ranges of these species. Thus, the distinction of sterile and fertile populations represents a suitable first approach to understanding the potential impact of source-sink dynamics on elevational richness patterns. This approach has been employed in two recent studies. In Bolivia, sterile individuals of numerous palm species are found at higher elevations than their fertile conspecifics, probably representing sink populations located above the elevations where these species can reproduce (Kessler 2000b). In Norway, along several elevational gradients, sterile plant populations are clumped at mid-elevations, emphasizing the hump-shaped richness patterns (Grytnes et al. 2008). These two studies suggest that source-sink dynamics may indeed modify elevational richness patterns, but they also differ from each other in one important aspect. Whereas among the Bolivian palms sterile populations were only found towards the upper elevational limits of the species, in Norway sterile plant populations were documented at both the upper and lower limits of the individual species ranges. Thus, in Bolivia the sterile populations lead to upwards shift in the palm species richness pattern whereas in Norway sterile populations accumulated at the middle of the gradients.

Material and methods Study sites Ferns were studied along six elevational transects, two in Bolivia and one each in Costa Rica, Mexico, La Re´union, and Indonesia (Table 1). Site description are available in previous publications for transects in Bolivia (Kessler 2000a, 2001a), Costa Rica (Kluge and Kessler 2006, Kluge et al. 2006, 2008), and Mexico (Kro¨mer and Acebey 2007, Acebey and Kro¨mer 2008). The island of La Re´union (2512 km2) is a volcanic island about three million years old. Our study was conducted from 300 m elevation in the southeast corner of the island up to the upper limit of closed forest at 2100 m on Piton de Neiges (3071 m). In our study region, mean annual precipitation increases from ca 4000 mm in the lowlands to ca 6000 mm at 550 m and then decreases to ca 2500 mm near tree line. June to September are the driest months. Mean annual temperatures are ca 248C at sea level and 128C at 2000 m. On the island of Sulawesi, Indonesia, we worked in Lore Lindu National Park in the centre of the island. Our transect extended from 300 m in the extreme northwest of the park to 2350 m near the summit of Gunung Rorekatimbu (2450 m). The geological substrate of the area consists of tertiary acid intrusives (Whitten et al. 2002). Mean annual precipitation increases from about 2000 mm in the lowlands to 4000 mm on the higher mountain slopes. There is no well-defined dry season. Field sampling Along all transects we used the same sampling method, detailed in Kessler and Bach (1999), Kessler (2000a,

Table 1. Overview of the study sites and key characteristics of field data and number fern of species. Site

Elevational extent

No. plots

Bolivia, Carrasco NP Bolivia, Masicurı´ Costa Rica, Braulio Carrillo NP La Re´union Indonesia, Sulawesi, Lore Lindu NP Mexico, Los Tuxtlas

2203950 5002500 1003400 1002750 2502450 1401670

81 51 96 18 45 42

124

No. species Source (time of field work in italics) 369 146 426 96 284 91

Kessler 2001a (MK 6-10.1996) Kessler 2000a (MK 7.1995, 5-6.1996) Kluge and Kessler 2005, Kluge et al. 2006 (JK 8.2002-9.2003) MK 3-4.2008 MK, SH, DC, JK 6-8.2007 Kro¨mer and Acebey 2007, Acebey and Kro¨mer 2008 (TK 4-12.2005)


2001a, b), and Kluge et al. (2006), and summarized here. Sampling was conducted in plots of 20 20 m2 each or of plots of similar area but different shape if local conditions did not allow the establishment of square plots. This plot size is small enough to be ecologically homogeneous within while containing a representative sample of the local fern flora (Kessler and Bach 1999). Plots were located in natural forests on slopes, excluding secondary or disturbed habitats as well as ridge or valley-bottom habitats. All fern species present in each plot were recorded, separating epiphytic and terrestrial records, and annotating whether at least one individual of a species was in reproductive condition (fertile records) or whether all individuals were sterile (sterile records). Epiphytic species were sampled by climbing lower parts of trees and by searching for fallen-down branches as well as with the use of trimming poles. In addition, large species, where fertility can also be assessed at a distance, were checked through binoculars. All species present in a study area (but not all species from each plot) were collected at least in triplicate and deposited at the herbaria GOET, UC, and the respective national herbaria. Data analysis Following the approach of Grytnes et al. (2008), we conducted two successive sets of analyses, albeit with different statistics due to different data structures. In the first step of the analysis, we tested whether sterile fern records were more common towards the elevational range limits of the respective species, in order to assess if sterile records indeed reflect sink populations. This was not possible for the La Re´union transect as it included too few plots for a statistically meaningful analysis. We further only included species with at least nine records in a given transect. For each species, we divided its elevational amplitude into quartiles (steps of 25% of the total elevational amplitude of the species), and calculated the percentage of sterile records separately for each quartile. Because the overall percentage of sterile records differed greatly between species (from 10 to 90%), percentages of sterile records in each quartile were transformed as to sum up to 100% for each species. We then used one-way two-tailed KruskalWallis tests to assess whether the percentage of sterile species records in the upper- and lowermost quartiles differed from those in the two central quartiles. Because for many species the upper- or lowermost records corresponded to the extreme points of the transects and may therefore not reflect the potential distributional limits of the species, in a second run of the test we only included range limits that were at least 500 elevational meters from the boundaries of the transects. In this way, for some species we only included either the upper or lower limits, whereas others were fully excluded. In the second part of the analysis, we compared the elevational richness patterns of sterile and fertile species. Elevational patterns of species richness were plotted separately for all species, epiphytes, and terrestrials, as well as for sterile and fertile records for the six study transects, and trend lines fitted by 2nd-order polynomial regressions. Additionally, to reduce the influence of variations in overall species richness, we plotted the percentage of species with

sterile records in each plot against elevation. Finally, to assess whether the distribution of sterile records influences the observed richness patterns, we plotted the percentage of sterile species records per plot against species number per plot. All analyses were conducted with R (R Development Core Team 2008).

Results The first series of tests, assessing whether sterile records are more frequent at the elevational range margins of the species, all resulted in non-significant results (Table 2). When the analyses were redone including only species boundaries at least 500 m from the transect limits to exclude potential biases by considering range limits determined by the transect extent rather than the biology of the species, no significant results were obtained either (data not shown). Comparing the elevational richness patterns of the ferns, five transects showed hump-shaped patterns (Los Tuxtlas only weakly so, because the transect is cut off at 1700 m), whereas on Sulawesi fern richness increased continuously with elevation (Fig. 1). When the records were divided into epiphytic and terrestrial life forms as well as by sterile and fertile records, the patterns became more complex. Epiphytes showed more pronounced hump-shaped patterns than terrestrials, except again in Sulawesi. In several cases (Los Tuxtlas, La Re´union) trend lines for sterile and fertile records were visually quite similar, whereas in others (especially Carrasco) they had distinctly different peaks. Among terrestrials, the differences between sterile and fertile records was even more pronounced, especially in Costa Rica where sterile records showed hump-shaped patterns whereas fertile records increased continuously with elevation, while this pattern is opposite in Masicurı´. In Los Tuxtlas, fertile records even showed an inversed hump-pattern. Focussing on the percentage of sterile species records per plot, epiphytic ferns along all six transects showed decreasing percentages with elevation, in two cases combined with slight increases at highest elevations (Fig. 2). Three of these patterns were statistically significant (Table 3). Among the terrestrial species, the patterns were more diverse, significantly decreasing with elevation in two cases (and not significantly in another one), increasing in one case, and with U-shaped patterns in two others. Table 2. Results of the Kruskall-Wallis tests comparing the frequency of sterile species records in the upper- and lower-most quartiles of the species distributions with that in the two intermediate quartiles. Study area

DF

K value

p value

Epiphytes

Bolivia, Carrasco Bolivia, Masicurı´ Costa Rica Mexico, Los Tuxtlas Sulawesi

2, 2, 2, 2, 2,

210 30 288 60 180

5.78 1.55 2.55 2.14 3.76

0.053 0.415 0.261 0.298 0.172

Terrestrials

Bolivia, Carrasco Bolivia, Masicurı´ Costa Rica Mexico, Los Tuxtlas Sulawesi

2, 2, 2, 2, 2,

132 48 156 24 102

1.24 4.88 2.78 2.34 1.77

0.450 0.083 0.193 0.276 0.376

125


All species

Terrestrials

80

40

60

30

40

20

20

10

Epiphytes 50

40

30

20

Bolivia – Carrasco

10

0

0 0

1000

2000

3000

4000

0 0

30

30

25

25

20

20

15

15

10

10

5

5

1000

2000

3000

4000

0

1000

2000

3000

4000

20

15

10

5

0

0 0

500

1000 1500 2000 2500 3000

0 0

500

1000 1500 2000 2500 3000

0

70

30

60

60

25

50

20

40

15

30

10

20

5

10

50

Bolivia − Masicurí 500

1000 1500 2000 2500 3000

40

Number of species

30 20 10

Costa Rica 0

0 0

500 1000 1500 2000 2500 3000 3500

30

0 0

500 1000 1500 2000 2500 3000 3500

0

500 1000 1500 2000 2500 3000 3500

30

20

25

25 15

20

20

15

10

15

10

10 5

5

Mexico – Los Tuxtlas

5

0

0 0

500

1000

1500

2000

0 0

40

20

30

15

500

1000

1500

2000

0

500

1000

1500

2000

30 25 20

20

10

10

5

0

0

15 10 5

Réunion 0

500

60

1000 1500 2000 2500 3000

0 0

500

1000 1500 2000 2500 3000

500

1000 1500 2000 2500 3000

0

500

1000 1500 2000 2500 3000

60

25

50

0

50

20

40

40 15

30

30 10

20

20 5

10

10

Sulawesi 0

0 0

500

1000 1500 2000 2500 3000

0 0

500

1000 1500 2000 2500 3000

Elevation (m)

Figure 1. Elevational patterns of fern species richness along the six elevational study transects for all species (left), terrestrials (centre), and epiphytes (right). Black dots and black lines mark total species richness, dark grey dots and wide dashed lines fertile species plot records, and light grey dots and dotted lines sterile species plot records. Lines are 2nd-order polynomial regression lines.

126


100

100

80

80

60

60

40

40

20

20

0

0 0

1000

2000

3000

4000

0

100

100

80

80

60

60

40

40

20

20

10

20

30

40

50

Bolivia − Masicurí

0

0 0

Percent of sterile species

Bolivia – Carrasco

500

1000 1500 2000 2500 3000

100

100

80

80

60

60

40

40

20

20

0

5

10

15

20

25

30

0

10

20

30

40

50

60

Costa Rica 0

0 0

500 1000 1500 2000 2500 3000 3500

100

100

80

80

60

60

40

40

20

20

Mexico − Los Tuxtlas

0

0 0

500

1000

1500

2000

100

100

80

80

60

60

40

40

20

20

0

5

10

15

20

25

30

0

5

10

15

20

25

30

0

10

20

30

40

50

60

Réunion 0

0 0

500

1000 1500 2000 2500 3000

100

100

80

80

60

60

40

40

20

20

Sulawesi 0

0 0

500

1000 1500 2000 2500 3000

Elevation (m)

Number of species

Figure 2. Percentage of sterile species records relative to elevation (left) and number of species of the respective life form (right) along the six elevational transects. Data and regression lines (2nd-order polynomial regression lines) are given for terrestrial species (black) and epiphytic species (grey). Continuous lines mark regressions significant at the pB0.05 level, non-significant lines are dashed.

127


Table 3. Determination coefficients and significancy levels of polynomic models describing the relationship of percentage of sterile species records to elevation along the six study transects. p-valuesB0.05 marked in bold. Study area

R2 value

p value

Epiphytes

Bolivia, Carrasco Bolivia, Masicurı´ Costa Rica Mexico, Los Tuxtlas La Re´union Sulawesi

0.42 0.05 0.40 0.43 0.16 0.13

B0.001 0.370 B0.001 B0.001 0.352 0.071

Terrestrials

Bolivia, Carrasco Bolivia, Masicurı´ Costa Rica Mexico, Los Tuxtlas La Re´union Sulawesi

0.10 0.26 0.42 0.28 0.07 0.06

0.018 0.001 B0.001 0.002 0.629 0.273

Finally, relating the percentage of sterile species records per plot to the number of species per plot recovered four significant patterns among the twelve cases analysed (Fig. 2, Table 4). The significant cases documented higher percentages of sterile records at higher species numbers per plot among terrestrial ferns at Masicurı´ and in Costa Rica, and higher percentages of sterile records at lower species numbers per plot among epiphytic ferns in Los Tuxtlas and Sulawesi. The non-significant patterns showed four decreases among the terrestrial species, and three increases as well as one decrease among the epiphytic species.

Discussion Our study of the elevational distribution of sterile ferns addressed two complementary questions. First, we asked whether sterile species records are more frequent along species range margins and can hence be considered to represent sink populations. Second, we asked whether sterile records show distinct elevational richness patterns that influence our perception of the overall species richness of ferns. The first of these questions has to be answered negatively, whereas for the second question the answer is complex, with differences between life forms and study sites. Table 4. Determination coefficients and significancy levels of polynomic models describing the relationship of percentage of sterile species records per plot to number of species per plot along the six study transects. p-valuesB0.05 marked in bold. Study area

R2 value

p value

Epiphytes

Bolivia, Carrasco Bolivia, Masicurı´ Costa Rica Mexico, Los Tuxtlas La Re´union Sulawesi

0.03 0.02 0.02 0.19 0.12 0.12

0.140 0.418 0.126 0.005 0.216 0.023

Terrestrials

Bolivia, Carrasco Bolivia, Masicurı´ Costa Rica Mexico, Los Tuxtlas La Re´union Sulawesi

0.01 0.12 0.05 0.05 0.00 0.02

0.480 0.013 0.034 0.164 0.982 0.320

128

We were unable to document that sterile records are clumped at the elevational range margins of the species. This contrasts with the results of Grytnes et al. (2008) who showed that for flowering plants along several elevational transects in Norway sterile records were indeed more common in the upper- and lowermost quartiles than expected by chance. However, Grytnes et al. (2008) also observed that the ability to detect this pattern depended on the spatial size of the plots, with larger plots revealing more diffuse patterns. They believed that this was because species records were classified only as fertile or sterile, rather than reflecting the actual number of sterile and fertile plants. Thus, a plot record was considered as ‘‘fertile’’ if at least one individual was found in fertile condition, regardless of whether hundreds of other plants were sterile. The probability of finding such scattered fertile individuals is higher in larger plots, reducing the ability of this approach to detect sink populations. In our study we used the same method as Grytnes et al. (2008), and it is conceivable that our plot size of 400 m2 is also too large for this approach. In both studies, the data were originally gathered for a different suite of analyses, with fertility status only included as an additional parameter. Clearly, more detailed data are needed, including precise counts of sterile and fertile specimens per plot. Our difficulty in detecting increases of sterile records at the elevational range margins of the species may also have been determined by the dispersal mechanisms of ferns. Contrary to seed plants, where dispersal mostly takes place by seeds of varying but usually substantial size and accordingly relatively restricted dispersal distances, in ferns dispersal is by dust-like spores that have highly efficient wind dispersal (Barrington 1993). There are no comparative studies of the diaspore rain of ferns and angiosperms along elevational gradients, but it is certainly plausible that the dispersal kernels of ferns and angiosperms are different, and accordingly that the spatial positioning of sink populations may be different. In particular, ferns have an independent gametophytic life stage which we did not consider at all due the difficulty of sampling and identifying fern gametophytes in tropical forests. It is conceivable that much of the ecological filtering in fern sink populations takes place in this gametophytic phase and may thus no longer be evident in the sporophytic stage studied by us. Thus, the detection of sterile populations may require different approaches for ferns and angiosperms. Furthermore, sterile populations are certainly not an unambiguous measure of sink populations. Especially among long-lived organisms under extreme environmental conditions successful reproduction may only take place at long, sometimes decadal intervals. Moreover, sterile populations can result from processes other than source-sink dynamics. For example, many ferns have vegetative reproduction and such species are not randomly distributed with elevation (Kluge and Kessler 2007). However, detailed studies of the population dynamics along elevational gradients can realistically only be conducted for selected species and will probably never be achieved for entire, diverse communities. Thus, while these problems have to be taken into account, at present the distinction of sterile and fertile populations represents a suitable first approach to


understanding the potential impact of source-sink dynamics on elevational richness patterns. Despite the problems with equating sterile records to sink populations, in our study we found distinct elevational patterns of sterile records that clearly affected our perception of the overall richness patterns. This was most conspicuous for the terrestrial ferns at Los Tuxtlas and Costa Rica, where the fertile records showed completely different elevational patterns compared to the hump-shaped distribution of sterile records (Fig. 1). In these cases, the overall hump-shaped richness pattern of terrestrial ferns was clearly caused by the sterile records. Accordingly, our previous interpretations of such hump-shaped patterns (Kessler 2000a, 2001a, Kluge et al. 2006) actually attempted to explain the distribution of sterile populations, while unknowingly ignoring the fact that the fertile populations showed distinct patterns that might be explained in different ways. In such a situation, the explanatory factors that we considered, including climatic conditions, topography, or the mid-domain effect, may all have missed the actual point. However, not all transects showed such clearcut patterns. Indeed, we found a high variability of patterns along the different transects and for the two main life forms. Among epiphytes, all six transects showed a tendency towards decreasing percentages of sterile species records with elevation, although this was only significant for three transects. On the other hand, there was no clear relationship of total species richness per plot to the percentage of sterile records, suggesting that our overall perception of the richness patterns of epiphytic ferns along elevational transects is not strongly influenced by the sterile records. Terrestrial ferns, in contrast, showed a much more variable distribution of sterile records, with increasing, decreasing, and U-shaped patterns. The latter, found along the transects in Mexico and on La Re´union, are particularly interesting in that they show that sterile populations of terrestrial ferns can be most common in species-poor communities under (for ferns) stressful environmental conditions. The low diversity of ferns in tropical lowlands and at high elevations has repeatedly been interpreted as reflecting extreme environmental conditions, namely limited water availability in the lowlands and low temperatures in the highlands (Kessler 2001b, Bhattarai et al. 2004, Kluge et al. 2006). Our study suggests that under these conditions sterile populations may be more common, either because they represent sink populations derived from source populations at mid-elevations, or because under extreme environmental conditions vegetative reproduction may be more common. On the other hand, along the Masicurı´ and Costa Rica transects sterile records of terrestrial ferns were most common in the most species rich plots, suggesting that here potential sink populations were accumulated at midelevations, inflating the hump-shaped richness patterns. This observation is accordance with the hypotheses of numerous researchers (Rahbek 1997, Kessler 2000b, Lomolino 2001, Grytnes and Vetaas 2002, Grytnes 2003a, b, Kattan and Franco 2004) as well as with the observations of Grytnes et al. (2008) in Norway.

How might these contradictory results of different transects studies be reconciled? One possibility is the influence of topography in the different study areas. Both the Los Tuxtlas and La Re´union transects are located on isolated volcanic mountains without large mountain massifs in close proximity. Due to the conical shape of these mountains, their high-elevation habitats have a very limited spatial extent and accordingly species-poor floras. The young age of the mountains and their vegetation (last eruption at Volca´n San Martı´n Tuxtla was in 1793, Guevara et al. 2004) and their active volcanic history further limit their species pools. In such a situation, it is conceivable that most plant species found near the mountain tops are sink populations derived from sources lower down on the mountain, as indeed suggested by our data. In contrast, the transects in Bolivia and Costa Rica are located on older, more extensive mountain ranges with a well-developed high elevation flora. Other geographic, climatic, or historical factors might further influence the spatial distribution of source and sink plant populations on mountains. A comparative analysis of numerous mountains with contrasting conditions might bring us a long step forward in understanding the role of population dynamics in shaping elevational richness patterns. For now, we can conclude that sterile populations of especially terrestrial ferns show distinct elevational distribution patterns that influence our perception of elevational richness patterns among ferns. Future studies of such patterns should take into account the possibility that source-sink dynamics determine or at least modify the observed patterns, both for ferns and other organisms. This does not only apply to elevational gradients but also to other ecological or geographical gradients. It is likely that the impact of population dynamics will be more pronounced over distances within the scale of regular dispersal of the taxon under consideration and less so for extensive gradients exceeding the scale of usual dispersal such as the latitudinal gradient (Kessler 2009). In any case, ignoring the impact of population dynamics on diversity patterns is liable to result in misinterpretations of the diversity patterns, as shown in our study for the terrestrial ferns. Acknowledgements  First of all, we wish to thank A. R. Smith for tireless plant identification and J.-A. Grytnes for valuable comments on the manuscript. In Bolivia, we thank the Herbario Nacional de Bolivia, A. Acebey, K. Bach, S. G. Beck, M. Cusicanqui, S. K. Herzog, J. Gonzales, I. Jimenez, A. de Lima, R. de Michel, M. Moraes, the Direccio´n Nacional de Conservacio´n de la Biodiversidad, the park guards of Carrasco National Park, the mayor of Villa Tunari, Hotel El Puente, and the Facultad de Agricultura, Univ. Mayor de San Simo´n, Cochabamba. For the study in Costa Rica, we thank F. Corrales for invaluable help during the field work, the park rangers of the Sistema Nacional de Area de Conservaciones (SINAC) and the Area de Conservacio´n Cordillera Volca´nica Central (ACCVC) in Costa Rica, the staff of Biological Station La Selva and the Organisation for Tropical Studies OTS, and the National Herbarium in San Jose´ for logistical support. The study in Los Tuxtlas was supported by a postdoctoral grant to TK from the Univ. Nacional Auto´noma de Me´xico, who also thanks A. Acebey for fieldwork assistance, and the directors of the Los Tuxtlas Biological Research Station, R. Coates and M. Ricker, for logistical support. Field work on La

129


Re´union was funded by am ERASMUS grant to MK and would have been impossible without the help of D. Strasberg and C. Ah Peng. The Sulawesi data were collected within the field work of the STORMA project, whose framework conditions were of invaluable importance. Field work was partly funded by the Deutsche Forschungsgemeinschaft DFG (grants to MK) and the Deutscher Akademischer Austauschdienst DAAD (to JK).

References Acebey, A. and Kro¨mer, T. 2008. Diversidad y distribucio´n de las ara´ceas de la Reserva de la Biosfera Los Tuxtlas, Veracruz, Me´xico.  Rev. Mex. Biodiv. 79: 465471. Barrington, D. S. 1993. Ecological and historical factors in fern biogeography.  J. Biogeogr. 20: 275280. Bhattarai, K. R. et al. 2004. Fern species richness along a central Himalayan elevational gradient, Nepal.  J. Biogeogr. 31: 389400. Currie, D. J. et al. 2004. Predictions and tests of climate-based hypotheses of broad-scale variation in taxonomic richness.  Ecol. Lett. 7: 11211134. Dias, P. C. 1996. Sources and sinks in population biology.  Trends Ecol. Evol. 11: 326330. Diffendorfer, J. E. 1998. Testing models of source-sink dynamics and balanced dispersal.  Oikos 81: 417433. Eriksson, O. and Ehrlen, J. 1992. Seed and microsite limitation of recruitment in plant populations.  Oecologia 91: 360364. Evans, K. L. et al. 2005. Speciesenergy relationships at the macroecological scale: a review of the mechanisms.  Biol. Rev. 80: 125. Gilpin, M. E. and Hanski, I. A. 1991. Metapopulation dynamics: empirical and theoretical investigations.  Academic Press. Grytnes, J. A. 2003a. Ecological interpretations of the middomain effect.  Ecol. Lett. 6: 883888. Grytnes, J. A. 2003b. Species richness patterns of vascular plants along seven altitudinal transects in Norway.  Ecography 26: 291300. Grytnes, J. A. and Vetaas, O. R. 2002. Species richness and altitude: a comparison between null models and interpolated plant species richness along the Himalayan altitudinal gradient, Nepal.  Am. Nat. 159: 294304. Grytnes, J. A. et al. 2008. Can the mass effect explain the midaltitudinal peak in vascular species richness?  Basic Appl. Ecol. 9: 373382. Guevara, S. et al. 2004. Los Tuxtlas. El paisaje de la sierra.  Inst. de Ecologı´a, A.C., Xalapa. Hansen, A. J. and Rotella, J. J. 2002. Biophysical factors, land use, and species viability in and around nature reserves.  Conserv. Biol. 16: 11121122. Hawkins, B. A. et al. 2003. Energy, water, and broad-scale geographic patterns of species richness.  Ecology 84: 3105 3117. Kattan, G. H. and Franco, P. 2004. Bird diversity along elevational gradients in the Andes of Colombia: area and mass effects.  Global Ecol. Biogeogr. 13: 451458. Kessler, M. 2000a. Elevational gradients in species richness and endemism of selected plant groups in the central Bolivian Andes.  Plant Ecol. 149: 181193. Kessler, M. 2000b. Upslope-directed mass effect in palms along an Andean elevational gradient: a cause for high diversity at midelevations?  Biotropica 32: 756759. Kessler, M. 2001a. Patterns of diversity and range size of selected plant groups along an elevational transect in the Bolivian Andes.  Biodivers. Conserv. 10: 18971920.

130

Kessler, M. 2001b. Pteridophyte species richness in Andean forests in Bolivia.  Biodivers. Conserv. 10: 14731495. Kessler, M. 2009. The impact of population processes on patterns of species richness: lessons from elevational gradients.  Basic Appl. Ecol. 10: 295299. Kessler, M. and Bach, K. 1999. Using indicator groups for vegetation classification in species-rich Neotropical forests.  Phytocoenologia 29: 485502. Kluge, J. and Kessler, M. 2005. Inventory of pteridophytes along an elevational transect in Braulio Carrillo National Park, La Selva Biological Stationand Cerro de la Muerte, Costa Rica.  Brenesia 6364: 1134. Kluge, J. and Kessler, M. 2006. Fern endemism and its correlates: constribution from an elevational transect in Costa Rica.  Divers. Distrib. 12: 535545. Kluge, J. and Kessler, M. 2007. Morphological characteristics of fern assemblages along an elevational gradient: patterns and causes.  Ecotropica 13: 2743. Kluge, J. et al. 2006. What drives elevational patterns of diversity? A test of geometric constraints, climate, and species pool effects for pteridophytes on an elevational gradient in Costa Rica.  Global Ecol. Biogeogr. 15: 358371. Kluge, J. et al. 2008. Elevational distribution and zonation of tropical pteridophyte assemblages in Costa Rica.  Basic Appl. Ecol. 9: 3543. Kro¨mer, T. and Acebey, A. 2007. The bromeliad flora of the San Martı´n Tuxtla volcano, Veracruz, Mexico.  J. Bromeliad Soc. 57: 6269. Leibold, M. A. et al. 2004. The metacommunity concept: a framework for multi-scale community ecology.  Ecol. Lett. 7: 601613. Lomolino, M. V. 2001. Elevational gradients of species-density: historical and prospective views.  Global Ecol. Biogeogr. 10: 313. McCain, C. M. 2009. Global analysis of bird elevational diversity.  Global Ecol. Biogeogr. 18: 346360. Mittelbach, G. G. et al. 2007. Evolution and the latitudinal diversity gradient: speciation, extinction and biogeography.  Ecol. Lett. 10: 315331. Myers, J. A. and Harms, K. E. 2009. Seed arrival, ecological filters, and plant species richness: a meta-analysis.  Ecol. Lett. 12: 12501260. Pulliam, H. R. 1988. Sources, sinks, and population regulation.  Am. Nat. 132: 652661. Pulliam, H. R. 2000. On the relationship between niche and distribution.  Ecol. Lett. 3: 349361. R Development Core Team 2008. R: a language and environment for statistical computing.  R Foundation for Statistical Computing, Vienna. Rahbek, C. 1995. The elevational gradient of species richness: a uniform pattern?  Ecography 18: 200205. Rahbek, C. 1997. The relationship among area, elevation, and regional species richness in neotropical birds.  Am. Nat. 149: 875902. Rahbek, C. 2005. The role of spatial scale and the perception of large-scale species-richness patterns.  Ecol. Lett. 8: 224239. Ricklefs, R. E. 2005. Phylogenetic perspectives on patterns of regional and local richness.  In: Bermingham, E. et al. (eds), Tropical rainforest: past, present, and future. Univ. Chicago Press, pp. 1640. Robinson, S. K. et al. 1995. Regional forest fragmentation and the nesting success of migratory birds.  Science 267: 19871990. Rosenzweig, M. L. and Ziv, Y. 1999. The echo pattern of species diversity: pattern and process.  Ecography 22: 614628.


Sharpe, J. M. and Mehltreter, K. 2010. Ecological insights from fern population dynamics.  In: Mehltreter, K. and Sharpe, J. M. (eds), Fern ecology. Cambridge Univ. Press, in press. Shmida, A. and Wilson, M. W. 1985. Biological determinants of species diversity.  J. Biogeogr. 12: 120. Stevens, G. C. 1992. The elevational gradient in altitudinal range: an extension of Rapoportâ&#x20AC;&#x2122;s latitudinal rule to altitude.  Am. Nat. 140: 893911. Tilman, D. 1997. Community invasibility, recruitment limitation, and grassland biodiversity.  Ecology 78: 8192. Turnbull, L. A. et al. 2000. Are plant populations seed-limited? A review of seed sowing experiments.  Oikos 88: 225238.

van Steenis, C. G. G. J. 1961. An attempt towards an explanation of the effect of mountain mass elevation.  Proc. Kon. Ned. Akad. Wetensch. C64: 435442. Whitten, T. et al. 2002. Ecology of Sulawesi.  Periplus Editions. Wiens, J. J. and Donoghue, M. J. 2004. Historical biogeography, ecology and species richness.  Trends Ecol. Evol. 19: 639 644. Wilson, D. S. 1992. Complex interactions in metacommunities, with implications for biodiversity and higher levels of selection.  Ecology 73: 19842000.

131


Ecography 34: 372383, 2011 doi: 10.1111/j.1600-0587.2010.06460.x # 2011 The Authors. Ecography # 2011 Ecography Subject Editor: Jeremy T. Kerr. Accepted 5 July 2010 The first two authors (O. L. and N. L.) contributed to this paper equally and are listed alphabetically

Can we predict butterfly diversity along an elevation gradient from space? Oded Levanoni, Noam Levin, Guy Pe’er, Anne Turbe´ and Salit Kark O. Levanoni, G. Pe’er, A. Turbe´ and S. Kark (salit@hebrew.edu), The Biodiversity Research Group, Dept of Ecology, Evolution and Behavior, The Silberman Inst. of Life Science, The Hebrew Univ. of Jerusalem, IL-91904 Jerusalem, Israel. Present address of GP: Dept of Conservation Biology, UFZ  Helmholtz Centre for Environmental Research, Permoserstr. 15, DE-04318 Leipzig, Germany.  N. Levin, Dept of Geography, Faculty of Social Sciences, The Hebrew Univ. of Jerusalem, Mount Scopus, IL-91905 Jerusalem, Israel.

An important challenge in ecology is to predict patterns of biodiversity across eco-geographical gradients. This is particularly relevant in areas that are inaccessible, but are of high research and conservation value, such as mountains. Potentially, remotely-sensed vegetation indices derived from satellite images can help in predicting species diversity in vast and remote areas via their relationship with two of the major factors that are known to affect biodiversity: productivity and spatial heterogeneity in productivity. Here, we examined whether the Normalized Difference Vegetation Index (NDVI) can be used effectively to predict changes in butterfly richness, range size rarity and beta diversity along an elevation gradient. We examined the relationship between butterfly diversity and both the mean NDVI within elevation belts (a surrogate of productivity) and the variability in NDVI within and among elevation belts (surrogates for spatial heterogeneity in productivity). We calculated NDVI at three spatial extents, using a high spatial resolution QuickBird satellite image. We obtained data on butterfly richness, rarity and beta diversity by field sampling 100 m quadrats and transects between 500 and 2200 m in Mt Hermon, Israel. We found that the variability in NDVI, as measured both within and among adjacent elevation belts, was strongly and significantly correlated with butterfly richness. Butterfly range size rarity was strongly correlated with the mean and the standard deviation of NDVI within belts. In our system it appears that it is spatial heterogeneity in productivity rather than productivity per se that explained butterfly richness. These results suggest that remotely-sensed data can provide a useful tool for assessing spatial patterns of butterfly richness in inaccessible areas. The results further indicate the importance of considering spatial heterogeneity in productivity along elevation gradients, which has no lesser importance than productivity in shaping richness and rarity, especially at the local scale.

Elevation gradients and species diversity Ecologists have a long lasting interest in diversity patterns across spatial gradients (Rosenzweig 1995, Lomolino 2001). Many earlier studies have examined changes in biodiversity along elevation gradients, yet no single spatial pattern has been identified thus far (Shmida and Wilson 1985, Rahbek 1995, 2005, Lomolino 2001, Grytnes and McCain 2007, Nogue´s-Bravo et al. 2008). Therefore, despite the interest in predicting patterns along climatic gradients, such predictions remain challenging. This is especially true in remote areas that are difficult to access and to sample in the field, such as mountains (Levin et al. 2007). The current availability of satellite imagery at detailed spatial resolutions (Kark et al. 2008) has created an opportunity to study and gain information about remote areas (Levin et al. 2007). Several studies (Bawa et al. 2002, Oindo 2002, Kerr and Ostrovsky 2003, Foody and Cutler 2006, Levin et al. 2007, Gillespie et al. 2008) have suggested that plant species richness can be

372

effectively predicted using simple indices derived from remotely-sensed images, such as the Normalized Difference Vegetation Index (NDVI; Tucker 1979, Tucker and Sellers 1986). This predictive ability is likely related to the fact that both primary productivity and habitat heterogeneity, two of the major factors shaping biodiversity patterns, can be relatively easily estimated by calculating satellite-derived vegetation indices (Kerr and Ostrovsky 2003, Gillespie et al. 2008, Rocchini et al. 2010, for reviews on the use of satellite images in ecology and biodiversity research). While predicting animal diversity using vegetation indices is more challenging, because NDVI is based on vegetation-related variables, Kerr et al. (2001) have shown that remote sensing tools can accurately predict butterfly richness at a semi-continental scale using low spatial resolution satellite imagery (1 and 8 km) and biodiversity data at a coarse scale (2.58 grid cells). They found that satellite-derived heterogeneity measures of land cover were strongly correlated with butterfly richness when examined


across Canada. However, the potential of remotely-sensed estimates of spatial heterogeneity in productivity to predict animal diversity at smaller scales (covering a small area using high spatial resolution satellite imagery and high spatial resolution biodiversity data) is less well understood. Our study focuses on butterflies, which are often considered to be good surrogates of biodiversity, being tightly dependant on a range of plants. They are known to respond to various environmental factors, to vegetation changes (reviewed in Peâ&#x20AC;&#x2122;er and Settele 2008a) and to climate changes (Parmesan et al. 1999, Thomas et al. 2004). Butterflies are relatively easy to sample in the field (Nowicki et al. 2008, Peâ&#x20AC;&#x2122;er and Settele 2008b), and have been successfully used in studies of ecological gradients (Blair 1999, Fleishman et al. 2000) and in conservation and global change research (Samways 1989, Kremen 1992, Kim 1993, Parmesan et al. 1999, Thomas et al. 2004, Pin Koh and Sodhi 2005, Thomas 2005, Parmesan 2006). However, fewer studies have tested the productivityrichness relationship in butterflies using remotely sensed indices of vegetation (but see Kerr et al. 2001, Bailey et al. 2004, Seto et al. 2004). The relationship between spatial heterogeneity in productivity (as estimated using remotely sensed vegetation indices) and species richness along local gradients has especially remained under-explored (Bailey et al. 2004). This is surprising, since spatial environmental heterogeneity is hypothesized to be an important factor shaping ecological communities and is related with species richness (Rosenzweig 1995, Atauri and de Lucio 2001, Rocchini et al. 2010). Furthermore, spatial heterogeneity in productivity has received much attention in studies at large scales (Kerr and Packer 1997, Jetz and Rahbek 2002). Here, we aim to examine whether butterfly richness, rarity, and beta diversity along an elevation gradient (hereby termed diversity estimates, see Methods) can be accurately predicted using satellite-derived vegetation indices. We asked whether the mean NDVI and estimates of spatial variation in NDVI can successfully predict: 1) butterfly richness within elevation belts along the elevation gradient, 2) changes in species composition among elevation belts, and 3) changes in range size rarity along the elevation gradient. We predicted that the mean values of NDVI and the spatial heterogeneity in NDVI, both within and among 100 m elevation belts, will be useful predictors for butterfly richness, beta diversity and rarity along the elevation gradient. While we did not aim to examine the effect of all potential environmental variables affecting species diversity along the altitudinal gradient, we did examine the effect of two major factors that can potentially confound the NDVIrichness relationship along the elevation gradient, namely area and the mid-domain effect (Grytnes and McCain 2007). A mid-domain effect (a peak in richness at midelevations, or mid landmass, due to spatial geometric constraints), is predicted where landmass boundaries, such as mountain tops, restrict species ranges and the overlap of variously sized ranges creates a peak in species richness at mid-elevations (Colwell and Lees 2000, Colwell et al. 2004, Grytnes and McCain 2007).

Methods Study area Located in north-eastern Israel along the border with Lebanon and Syria, Mt Hermon (33.25?N, 35.48?E), is part of the Anti-Lebanon Mountains, which are isolated from the main mountain ranges of the Middle East, Asia and Europe (Shmida 1977). The parent material is homogeneous, hard Jurassic limestone and dolomite, forming Terra Rossa soils (Shmida 1977). The terrain is characterized by steep rocky limestone Karst slopes (Auerbach and Shmida 1993). Mt Hermon is an elongated anticline that extends NE-SW over 35 km and rises from 300 to 2814 m over a 13-km distance on its SW side in Israel, where our study area is located (the highest point is 2224 m). The climate is Mediterranean, with rainy or snowy winters, and hot dry summers. Precipitation ranges from 600 to 1500 mm yr1 along the mountain, and above an elevation of 1500 m consists mostly of snow. As in other mountains, temperature decreases, while solar radiation and precipitation increase with elevation. Snow usually begins to cover the slopes of Mt Hermon in the first half of January and lasts until April. Snow patches usually remain until June above 1900 m, mainly on SE facing slopes and in the valleys (Shmida 1980). Three main vegetation belts have been defined in earlier studies of Mt Hermon (Shmida 1977, 1980). These include: 1) evergreen Mediterranean maquis (300 1200 m); 2) the xero-montane open forest (12001900 m) and 3) the subalpine Tragacanthic belt (19002814 m) (Fig. 1). The part of the Hermon in Israel ranges approximately 7300 ha, most of which is a nature reserve since 1972 (Levin et al. 2007). Selection of study sites and sampling design Most studies of butterfly richness, as well as systematic monitoring schemes, rely on line transect sampling (van Swaay et al. 1997, Ku¨ hn et al. 2008, Nowicki et al. 2008), as described by Pollard (1977) and standardized by Pollard and Yates (1993). Quadrat sampling is less often used for butterfly sampling (but see Su et al. 2004, Grill et al. 2005). However, quadrat sampling enables one to concentrate higher sampling effort in given locations and provides comparative ability with sampling methodologies used for other taxa, such as quadrat sampling for plants or point-counts for birds. To examine changes in diversity both among and within elevation belts, we conducted both line transect and quadrat sampling. We divided the elevation range of our study area (500 2200 m) into elevation classes 100 m high, thus obtaining 17 elevation belts. The area of the elevation belts ranged between a minimum of 5.8 ha (21002200 m) to 117 ha (13001400 m). To reduce variability resulting from different surface aspects, all quadrats and transects were located on SW facing slopes (following Levin et al. 2007), which are the most common slopes in the study area, corresponding with the shape of the Mt Hermon anticline that extends from SW to NE (see Levin et al. 2007 for

373


Figure 1. Map showing NDVI values in the study area on Mt Hermon. The photos on the right show the three major vegetation belts, from top to bottom: subalpine Tragacanthic belt (above 1900), xero-montane open forest (12001900 m) and evergreen Mediterranean maquis (3001200 m). Photos by S.K.

further details). We marked two grid squares of 50 20 m (1000 m2) within each 100 m elevation belt, totalling 34 quadrats. This enabled us to examine changes along the elevation gradient and to obtain repetition within elevation belts. The grid squares were pre-selected using remote sensing tools, being those with NDVI values close to the median NDVI of the whole elevation belt in the SW slope of the study area, so as to assure that they indeed represent vegetation of their respective elevation belt. The same quadrats were used for plant sampling in an earlier study (Levin et al. 2007). Butterfly sampling We sampled the butterflies along the elevation gradient between 500 and 2200 m. Butterfly sampling was conducted during two years in the peak activity season of most Hermon butterflies (FebruarySeptember 2005, and MarchAugust 2006; sensu Benyamini 2002), with each of the vegetation belts visited on average over 15 separate dates. Butterfly sampling was conducted by two Lepidoptera experts capable of recognizing all species in the field (O.L. and G.P.). A butterfly expert and a note-taker walked along parallel lines inside each sampling quadrat for 20 min. The sampling duration of 20 min in the quadrats was determined based on species-accumulation curves generated in preliminary work, using EstimateS ver. 8.0 (Colwell 2006). These indicated that accumulation is reached, on average, after 1293 min of sampling (sampling ]95% of the species). The transect lines began from the corner of each quadrat and led to the next quadrat (i.e. they were located outside the quadrats). Each line transect was 300 m long (the length was determined based again on species-accumulation curves (Colwell 2006), indicating the accumulation of at least 95% of the species within 250950 m of sampling). The transect line was divided into sections 50 m long (Pollard and Yates 1993) in order to allow the calculation of species-accumulation curves with distance. The line transects were marked by metal rods 374

(0.5 m high) in order to ensure accurate repetition of transects during all sampling visits. Butterfly species recognition in both quadrats and transect lines was performed visually. All individuals first seen within a distance of 5 m from the observers were recorded. When needed, in order to verify identification, we captured the butterfly with a sweeping net and then immediately released it at same location. Line transect sampling and quadrat sampling were performed sequentially. All sampling was performed between 9:00 am (10:00 am in elevations 1500 m) and 15:00, when ambient temperatures were 208C, cloud cover was B50%, and wind speed was B4 km h1. Weather conditions were recorded using a hand-held Kestrel 4000 weather station. Repeated sampling visits to each 100 m elevation belt were organized so that there were different starting times for each belt in order to reduce potential biases that are related to the timing of sampling. Diversity estimates In order to increase statistical power, we pooled the data from the two quadrats and transects within each 100 m elevation belt. This was done after a preliminary analysis, which showed that the butterfly diversity estimates were rather similar for quadrats and for transects when analysed separately and since our goal in this paper was not to compare the different methodologies. We calculated values of richness, beta diversity and range size rarity as sampled in each 100 m elevation belt. Species richness (alpha diversity) was calculated by summing up all the species that were observed in the two quadrats and transects within each 100 m elevation belt. Following McCain (2004), species were assumed to be present at an elevation if they were detected at both higher and lower elevations adjacent to a given belt. In cases where larger elevation-belt gaps in appearance were found (over 100 m), we only ‘‘filled in’’ the occurrence of a species in a given elevation belt if the data was consistent with the known range of distribution of


the species in Mt Hermon based on the literature (Benyamini 1993, 2002). Filling in was done for 25 of the 83 butterfly species sampled in this study. For 15 of these 25 species only a single elevation belt was filled-in. Results were very similar when analysis was repeated without filling in. Various estimators for beta diversity have been suggested in the literature, as reviewed by Koleff et al. (2003). We adopted the bsim (beta sim) estimator, which was considered by Lennon et al. (2001) and Koleff et al. (2003) as a reliable estimate. Preliminary results indicated that it produced very similar results to the estimator bt used by Wilson and Shmida (1984) in their earlier study of plants in Mt Hermon. Beta sim (bsim) was calculated as follows: bsim 

min(b; c) min(b; c)  a

(1)

where for each two neighbouring 100 m elevation belts (X and Y): a the number of species observed in both X and Y, b the number of species in Y that are not observed in X, cthe number of species in X that are not observed in Y. High values of bsim indicate that there were few species shared between two adjacent elevation belts (i.e. a high turnover rate). Rarity has been defined and estimated in the literature using many different approaches (Gaston 1994, Izco 1998). Because we were interested in the relative range size rarity within the mountain area, rather than in rarity over the whole distribution range, we used an estimate that quantifies the confinement of species to a small number of elevation belts in the mountain range. This approach has been used in many recent spatial ecology and large-scale conservation studies (Myers et al. 2000), and in Mt Hermon in a study by Levin et al. (2007). Range size rarity (RSR) was calculated for each elevation belt as the sum of the inverse of the range sizes of all the species occurring in it (Williams et al. 1996, Williams 2000): X RSR  (1=Ci) (2) where Ci is the number of elevation belts occupies by species i. We estimated range size as the number of elevation belts in which the species occurred (of the seventeen 100 m altitude belts sampled). Remote sensing analyses We used a high spatial resolution QuickBird satellite image of the study area that was acquired during mid-spring (26 May 2004), when vegetation flowering is at its peak (Shmida 1977, 1980, Levin et al. 2007). The image has a spatial resolution of 2.4 m in its four spectral bands that cover the visible and near infrared spectral regions. We corrected the satellite image for atmospheric scattering and absorption and for topographic effects of shading using the atmospheric/topographic correction of multispectral sensors for rugged terrain as applied in ATCOR 3 ver. 7.1 (Richter 1998), which is considered a reliable model for atmospheric corrections (Ben-Dor et al. 2005). We used a Digital Elevation Model (DEM) obtained from the Survey of Israel at a spatial resolution of 25 m (Hall et al. 1999) for calculating the slope, aspect and the sky view factor (i.e. the

visible area of the sky as dependent upon the surrounding topography). We then calculated normalized difference vegetation index (NDVI), one of the earliest remotely sensed vegetation indices applied in the literature (Rouse et al. 1973, Tucker 1979). Its relationship with vegetation productivity is well established, and it is one of the most commonly used vegetation indices (Kerr and Ostrovsky 2003, Pettorelli et al. 2005, Levin et al. 2007), especially in biodiversity studies (Gillespie et al. 2008). NDVI was calculated as follows: NDVI (NIR R)=NIR R (3) where NIR reflectance in the near infrared band of an image pixel, R reflectance in the red band of an image pixel. Because NDVI is a ratio index shading effects have only a minor effect on it (Lillesand and Kiefer 1994). We also compared the results with three other remotely sensed vegetation indices designed for overcoming issues of variability in the soil background, atmospheric haze, and saturation of the NDVI in cases of dense vegetation, including the Soil Adjusted Vegetation Index (Huete 1988), the Enhanced Vegetation Index (Huete et al. 2002) and the percentage of tree cover (as in Levin et al. 2007). Because results were generally similar for the four satellite-derived vegetation indices and because correlations with diversity estimates were strongest for NDVI, we report here only the results for NDVI (detailed results for the three other indices are available from the authors upon request). In addition to calculating the mean and standard deviation (SD) of NDVI within the elevation belts, we quantified the change in NDVI among elevation belts along the elevation gradient. When examining changes in the values of NDVI along a gradient, multiple statistics can be used, such as the difference in NDVI among elevation belts and the ratio between adjacent belts (compare with Walker et al. 2003). Here, we initially calculated four different estimates for the change in NDVI along the elevation gradient. These include: rate of change in NDVI between each two neighbouring elevation belts (RC1): RC1X NDVIX =NDVIX1

(4)

degree of change in NDVI between each two neighbouring elevation belts (DC1): DC1X NDVIX NDVIX1

(5)

rate of change in NDVI between the elevation belt above and below the belt in focus (RC2): RC2X NDVIX1 =NDVIX1

(6)

degree of change in NDVI between the elevation belt above and below the belt in focus (DC2): DC2X NDVIX1 NDVIX1

(7)

where subscript x represents elevation belt x, subscript x1 stands for the elevation belt adjacent to and above elevation belt x, and subscript x 1 stands for the elevation belt adjacent to and below elevation belt x. RC2 and DC2 were used to examine elevation gradients at a somewhat larger vertical distance, one that is still relevant for butterflies (200 vs 100 m). Negative values of 375


(mean and SD of NDVI per belt, DC1, DC2, RC1, RC2) and the residuals from the relationship between area and butterfly diversity, which was used as the dependent variable. This was calculated for the total area (log transformed) at each of the three spatial extents considered in this study.

Results Butterfly diversity along the elevation gradient Overall, in a total of 120 km of line transect sampling and 116 h of quadrat sampling, we recorded 10 513 individual butterflies belonging to 83 species and six butterfly families. Butterfly species richness showed a bimodal pattern with elevation, peaking between 1300 and 1500 m (48 species within each of the two 100 m belts) and between 1800 and 1900 m (46 species; Fig. 2a). Range size rarity showed local maxima at two intermediate elevation ranges (9001000 and 13001400 m) and then increased sharply towards the highest elevation belts (Fig. 2b). Beta sim (bsim) diversity showed multiple peaks along the gradient, the largest of which was at the elevation range between 1900 and 50

(a)

376

40 35 30 25 20 9

(b)

8 7 RSR

6 5 4 3 2 0.14

(c)

0.12 0.10 0.08 0.06 0.04 0.02

2200

2100

2000

1900

1800

1700

1600

1500

1400

1300

1200

1100

900

1000

800

700

0.00 500

To examine the mid-domain effect, we ran 1000 Monte Carlo simulations without replacement. This was done in order to compare the observed species richness with the predictions of a null model simulating species richness based on empirical range sizes for each belt between 500 and 2200 m. The empirical range sizes were derived from the field data. We compared our results with 95% confidence intervals generated by the simulations. This was done based on analytical stochastic models (Colwell and Hurtt 1994) using the Mid-Domain-Null program, which takes into account the lower and upper elevations of the species ranges (McCain 2004). We calculated the linear regressions with the mean and SD of NDVI (within vegetation belts) or between adjacent belts (DC1, DC2, RC1 and RC2 among vegetation belts) as the independent variable (at the three spatial scales of our analysis) and each of the butterfly diversity estimates, including richness, bsim and range size rarity as the dependent variables. Variables were log-transformed when needed. To account for autocorrelation along the elevation gradient, we used the method developed by Dutilleul (1993), as applied in PASSAGE 1.1 (<www.passagesoft ware.net/>). To account for the potential effect of area of the different vegetation belts along the elevation gradient on the relationships and to examine whether the area of the elevation belts had a confounding effect on the observed relationships between NDVI and diversity estimates, we performed a partial correlation analysis, as well as a multiple regression analysis (using JMP 7.0 SAS Inst.). We examined whether statistical significance is maintained after calculating the relationship between each of the NDVI estimates

Beta SIM

Data analysis

Richness

45

600

DC indicate that NDVI values increase with elevation. We deliberately avoided using absolute values of DC, as the sign provided also an index for the directionality of changes (i.e. upwards or downwards) along the elevation gradient. This is especially important for butterflies, since some species perform directional hilltopping behaviour in which they ascend to mountain summits for the purpose of mating (Shields 1967, Alcock 1987, Ehrlich and Wheye 1988, Peâ&#x20AC;&#x2122;er et al. 2004). Because butterflies are usually not limited in their activity to a single quadrat, in order to calculate NDVI statistics comparing the different 100 m belts, we used three spatial extents (coverage areas) in the mountain. These included: 1) the total area within each of the 0.1 ha quadrats and a buffer zone of 5 m on both sides of the transects, within each of the 100 m elevation belts; 2) the total area of the SW facing aspect of each 100 m elevation belt in our study region. This was the aspect in which our sampling quadrats and transects were located; 3) the total area of the 100 m elevation belt in our study region (all aspects). At each of these spatial extents, we examined the relationship between the different butterfly diversity estimates and the mean NDVI, SD, RC1, RC2, DC1 and DC2. This enabled us to examine which spatial scale of NDVI best predicts changes in local butterfly diversity within and among elevation belts.

Elevation (m)

Figure 2. Changes in butterfly diversity along the elevation gradient in Mt Hermon showing: (a) species richness, (b) range size rarity (RSR) and (c) beta diversity (bsim).


2000 m (Fig. 2c). Range size rarity showed a significant positive relationship with bsim (R2 0.47, p B0.01, n 16), and a significant (weaker) positive relationship with species richness (R2 0.36, pB0.02, n 17). Changes in vegetation indices along the elevation gradient Mean NDVI showed a hump-shaped pattern along the elevation gradient, peaking between 900 and 1200 m and gradually declined to its minimum value at the highest elevation belts (Fig. 3). The SD of NDVI remained relatively constant up to 1200 m, above which it gradually declined with increasing elevation (Fig. 3). The values of DC1, DC2, RC1 and RC2, however, increased with elevation up to 1200 m and 1300 m, respectively, after which they declined (with a minor peak around 1850 m; Fig. 3; the changes in RC1 and RC2 with elevation are similar to those of DC1 and DC2, and are therefore not shown). Relationship between butterfly diversity and NDVI

0.7

Average DC1 DC2 Standard deviation

0.6 0.5 0.4 0.3 0.2 0.1 0 â&#x20AC;&#x201C;0.1

2200

2100

2000

1900

1800

1700

1600

1500

1400

1300

1200

1100

900

1000

800

700

600

â&#x20AC;&#x201C;0.2 500

NDVI values for 100 m elevation belts

Richness was strongly correlated with all variables estimating spatial heterogeneity in productivity, including DC1, DC2, RC1 and RC2 (Table 1, Fig. 4). However, the relationship between butterfly richness and mean and standard deviation (SD) of NDVI was weak and was not significant (Table 1). The relationship between richness and the rates of change in NDVI between elevation belts was strong, and remained significant after correcting for autocorrelation effects (Table 1). Unlike richness, range size rarity was significantly correlated with mean NDVI as well as with all the variables expressing the spatial heterogeneity in NDVI: SD, DC1, DC2, RC1, RC2 (Table 1, Fig. 5). Of the diversity estimates tested (richness, bsim and range size rarity), beta diversity (bsim) was in most cases the least strongly correlated with NDVI variables.

Elevation (m)

Figure 3. NDVI values (mean, standard deviation, DC1 and DC2) per 100 m elevation belt. DC1 stands for the difference between the mean of a given elevation belt and the elevation belt above it, while DC2 stands for the difference between the mean values of the elevation belts below and above each elevation belt in Mt Hermon.

When comparing NDVI estimates deriving from the three spatial extents examined (quadrats and transects, SW slope, and the area of the whole elevation belt within the study area), NDVI estimates deriving from the largest spatial extent (the entire elevation belt), were in most cases more strongly correlated with all butterfly diversity measures than those deriving from the quadrats and transects alone (Table 1). The area of the vegetation belts was not significantly correlated with butterfly richness at any spatial extent (Table 2). Butterfly rarity (RSR) was, however, significantly correlated with area (after log transformation) at the spatial scale of the SW aspect of the elevation belt (Table 2). However, in most cases, the correlations between NDVI and both richness and range size rarity remained statistically significant after taking into account the effect of the (log transformed) area of the elevation belts using a residual analysis (Table 3), as well as when performing a multiple regression analysis (Table 4). In most cases area was nonsignificant in the multiple regression analysis (Table 4). No mid-domain effect was detected. Species richness did not fall within the 95% prediction curves of the model based on the 1000 simulations of the Mid-Domain Null model (Fig. 6). When we included the effect of autocorrelation on the significance of the correlations, the significance of the regression model declined, as expected (Table 1). However, in some of the cases, the correlation remained significant between richness and spatial heterogeneity between the elevation belts (DC1, DC2, and RC1), also when spatial autocorrelation was taken into account.

Discussion We found that the NDVI was a strong predictor of butterfly richness along the elevation gradient in Mt Hermon, explaining up to 80% of the total variation in butterfly richness (Fig. 4). However, it was not the mean NDVI, but rather its variability among elevation belts, that best predicted butterfly richness within the elevation belts. Mean NDVI is considered a good surrogate for net primary productivity (Gillespie et al. 2008). Butterfly richness along the elevation gradient appears to be more strongly shaped by spatial heterogeneity in productivity than by productivity per se at the local spatial scale examined here. This suggests that the most commonly used remotely-sensed vegetation statistic, namely the mean NDVI is in some cases not the most efficient estimate if one aims to predict richness and diversity patterns along spatial gradients (e.g. elevation). In such cases, it may be more useful to study the spatial heterogeneity in NDVI. Here we show the importance of spatial heterogeneity in productivity at the small regional scale. The importance of heterogeneity in studies using remote sensing indices has been shown at much larger (e.g. continental) scales. For example, in their work on mammals, Kerr and Packer (1997) found that in the higher energy regions of North America, the best predictor of mammal richness was topographic heterogeneity and local variation in energy availability. At a regional scale, Atauri and de Lucio (2001) examined the relationships between landscape structure, land use and richness of birds, amphibians, reptiles and 377


Table 1. Pearson correlation coefficients between NDVI statistics calculated at three spatial extents and butterfly diversity estimates, including richness, range size rarity (RSR) and beta sim (bsim). The number sign (#) marks significance at the 0.05 level when taking into account the effect of autocorrelation. Correlation coefficients between each NDVI statistic and butterfly diversity estimates NDVI statistic

Spatial extent

Butterfly diversity

Entire 100 m elevation belt

SW aspect of the 100 m elevation belt

Quadratstransects within the 100 m elevation belt

0.07 0.68** 0.55* 0.17 0.76*** 0.64**

0.08 0.66** 0.41 0.25 0.63** 0.16

Average Log (average) Average Standard deviation Log (standard deviation) Standard deviation

Richness RSR Beta sim Richness RSR Beta sim

0.33 0.79*** 0.55* 0.46 0.86*** 0.67**

DC1

Richness Log RSR Beta sim

0.85*** 0.55* 0.04

#

0.78*** 0.56* 0.01

0.33 0.43 0.25

DC2

Richness Log RSR Beta sim

0.90*** 0.68** 0.03

#

0.81*** 0.70** 0.14

0.44 0.47 0.15

RC1

Richness Log RSR Beta sim

0.88*** 0.68** 0.07

#

0.81*** 0.68** 0.11

0.51* 0.28 0.05

RC2

Richness Log RSR Beta sim

0.90*** 0.81*** 0.18

0.79*** 0.82*** 0.27

0.59* 0.37 0.00

*pB0.05, **pB0.01, ***pB0.001.

(a)

9 8

Butterfly RSR

butterflies in a Mediterranean landscape (Madrid, Spain). They found that the response of species richness to land use heterogeneity varied depending on the group of species considered. The most important factor affecting bird and butterfly richness in their study was landscape heterogeneity, while other factors, such as the specific composition of land use, played a secondary role (Atauri and de Lucio 2001). In the montane ecosystem examined here, spatial heterogeneity in productivity between elevation belts explains butterfly richness better than mean productivity. What biological factors may lead to these results? One possibility is that the spatial heterogeneity in productivity estimated here represents the variety of habitat types available to the butterflies at local spatial scales and within relatively short distances (dozens of meters to kilometres). Such heterogeneity is particularly beneficial for adult

y = –4.866ln(x) + 1.1911

7

R2 = 0.6209

6 5 4 3 2 0.2

0.3

0.4

0.5

0.7

(b) 10 y = –5.18ln(x) –4.7109

9

50

R2 = 0.7441

40 y = 81.054x + 37.679

35

R2 = 0.8015

Butterfly RSR

8

45

Butterfly richness

0.6

Average NDVI

7 6 5 4 3

30

2 0.05

25

0.07

0.09

0.11

0.13

0.15

0.17

0.19

0.21

0.23

Standard deviation of NDVI 20 –0.2

–0.15

–0.1

–0.05

0

0.05

0.1

0.15

0.2

DC2

Figure 4. The relationship between DC2 (defined in eq. 7) and butterfly species richness in Mt Hermon.

378

Figure 5. (a) The relationship between average NDVI values within each 100 m elevation belt and butterfly range size rarity (RSR) in Mt Hermon; (b) the relationship between the standard deviation of NDVI within each 100 m elevation belt and butterfly range size rarity (RSR).


Table 2. Pearson correlation coefficients between the area of the three spatial extents examined in the study and butterfly diversity estimates, including richness, range size rarity (RSR) and beta sim (bsim). Correlation coefficients between area and butterfly diversity estimates

Spatial extent

Area

Butterfly diversity

Entire 100 m elevation belt

SW aspect of the 100 m elevation belt

Quadratstransects within the 100 m elevation belt

Area Log (area) Area Log (area) Area Log (area)

Richness Richness RSR RSR Beta sim Beta sim

0.39 0.23 0.32 0.55* 0.30 0.40

0.01 0.01 0.68** 0.74*** 0.52 0.55*

0.16 0.10 0.15 0.07 0.39 0.34

*pB0.05, **pB0.01, ***pB0.001.

butterflies, allowing them to utilize a variety of resources. While being highly dependent on particular host plants during larval development, adult butterflies often have different habitat requirements than those of the larvae (Benyamini 2002, Settele et al. 2009). Our results may also be partly shaped by the specific habitat and host plants of the butterfliesâ&#x20AC;&#x2122; larvae, but information about larval spatial distribution along the elevation gradient is not sufficient for analyzing such potential effect. Another explanation may be that the high spatial heterogeneity in productivity represents high turnover of habitats and changes in conditions, which allow more species to co-occur in transitional areas (Shmida and Wilson 1985). This supports findings from earlier studies, which suggest that areas of sharp environmental transition (ecotones) are especially rich both in species richness and Table 3. Partial correlations: Pearson correlation coefficients between NDVI statistics calculated at the spatial extent of the entire elevation belt and residuals of the butterfly diversity estimates (after predicting their values with area as the independent variable), including richness, range size rarity (RSR) and beta sim (bsim). Correlation coefficients between each NDVI statistic and residuals of butterfly diversity estimates

Spatial extent

NDVI statistic

Butterfly diversity estimate

Entire 100 m elevation belt

Average

Richness RSR Beta sim

0.60* 0.30 0.59*

Standard deviation

Richness RSR Beta sim

0.75*** 0.41 0.76***

DC1

Richness RSR Beta sim

0.75*** 0.71** 0.14

DC2

Richness RSR Beta sim

0.81*** 0.79*** 0.16

RC1

Richness RSR Beta sim

0.84*** 0.78*** 0.25

RC2

Richness RSR Beta sim

0.88*** 0.84*** 0.31

*pB0.05, **pB0.01, ***pB0.001.

in rare species because they serve as meeting areas between different communities and/or due to the unique environmental conditions found in ecotonal environments (reviewed in Kark and van Rensburg 2006). We found two peaks in beta diversity in the transition areas between Mt Hermonâ&#x20AC;&#x2122;s three vegetation belts. A peak in the betadiversity of plants was also found between 1200 and 1300 m on Mt Hermon, corresponding to the transition between a maquis and montane flora (Wilson and Shmida 1984). This supports the hypothesis that transition areas are zones of high turnover, where spatial heterogeneity is high (Shmida and Wilson 1985, Kark and Van Rensburg 2006). Our findings here also support the prediction and recent findings at continental and regional scales that areas with high turnover tend to show higher levels of rarity and local endemism (Kark et al. 2007, van Rensburg et al. 2009). Thus far, relatively few studies have examined the relationship between butterfly richness and satellite-derived vegetation indices that estimate productivity (but see Kerr et al. 2001). Some studies found relatively weak positive relationships, while others showed none (see Bailey et al. 2004, Seto et al. 2004 and references therein). Few studies have examined the relationship between butterfly richness, rarity and NDVI heterogeneity in space at local scales. Bailey et al. (2004) studied butterfly and bird richness and its correlation with NDVI heterogeneity using Simpsonâ&#x20AC;&#x2122;s diversity index. While heterogeneity in NDVI predicted the total species richness of birds (R2 0.75), no association occurred between NDVI heterogeneity and species richness of butterflies in any of the vagility classes tested in their work (Bailey et al. 2004). These included low vagility (an individual is likely to move on the order of dozens of meters in its lifetime); intermediate (an individual may move hundreds of meters); and high (an individual may move more than a kilometer) (Bailey et al. 2004). The authors suggested that for butterflies, NDVI may not be the best measure of environmental heterogeneity and that other measures (e.g. elevation) may be more appropriate (Bailey et al. 2004). However, the lack of relationship between spatial heterogeneity in productivity and butterfly richness may partly result from the estimates used to measure heterogeneity in NDVI, rather than from the lack of suitability of NDVI in predicting environmental heterogeneity. We propose that future studies should calculate spatial heterogeneity in productivity along the elevation gradient, quantifying changes in productivity between altitudinal gradients. This approach is more equivalent 379


Table 4. Multiple regression coefficients and the adjusted R2 between the area and NDVI variables as the independent variables and butterfly diversity (richness, RSR or bsim) at the spatial extent of the entire elevation belt. NDVI variable 2

R Coefficient of area NDVI variable R2 Coefficient of area

RC1

Standard deviation of NDVI within an elevation belt

Coefficient of NDVI

Richness

RSR

Beta-sim

0.77 *** 1.355 69.28 ***

0.71 *** 1.54 *** 13.32 ***

0.85 *** 0.24 ***

0.71 *** 0.01

0.43 ** 0.0002

42.14 ***

0.67 **

177.8 ***

0.15 0.02 0.12

*pB0.05, **pB0.01, ***pB0.001.

to beta diversity estimates used for estimating turnover of species in space, focusing on changes between neighbouring cells. The relatively weak correlations found here between productivity (mean NDVI) and butterfly richness may result from the fact that in some cases butterfly species are constrained by the identity of the plant species available, and particularly by the presence of specific host plants, rather than by total plant richness or vegetation cover (Kelly and Debinski 1998, Peâ&#x20AC;&#x2122;er and Settele 2008b). In such cases, productivity would not be a good predictor of butterfly richness, compared with, for instance, larger or more generalistic taxa that depend more directly on productivity (see Shochat 1999 for birds). Whereas plant richness above 2000 m amounted to B40% of the peak plant richness (found at about 1000 m; Levin et al. 2007), butterfly richness above 2000 m reached almost 80% of the peak butterfly richness (found at about 1400 m; Fig. 2a). We found higher levels of butterfly species richness at the high altitudes (above 2000 m), which are characterised by lower productivity, relatively low plant richness and harsher weather conditions (e.g. winds; Shmida 1977) compared with lower elevations (B900 m; Fig. 2a). This supports our knowledge that butterfly ranges are often limited by factors other than the diversity and distribution of plants (Dennis and Shreeve 1991, Dennis et al. 1991, Quinn et al. 1997, Hawkins and Porter 2003) or by temperature and rainfall (Pollard 1988). Observed species richness Lower 95%

70

Upper 95%

Species richness

60 50 40 30 20 10 0 400

600

800

1000 1200 1400 1600 1800 2000 2200

Elevation (m)

Figure 6. Mid domain analysis of the butterfly species richness. The upper and lower 95% confidence intervals were generated from 1000 Monte Carlo simulations.

380

Here, we examined the effect of the spatial extent at which the variability in NDVI was calculated. Interestingly, we found that the NDVI estimates that were based on the whole area of the elevation belt often provided better predictors of butterfly richness within each of the quadrats than the data from the quadrats themselves. It is not easy to conclude why this was found, but we can hypothesize that it results from the fact that scaling up when calculating NDVI better represented the relevant habitat as perceived by the butterflies (compare with Kumar et al. 2009). Because adult butterflies are mobile and move among host plants, which are not distributed in the area uniformly, this may better capture their preferences than local sampling of NDVI. Rowe and Lidgard (2009), following a detailed analysis of the effect of sampling methodology on patterns of elevation diversity, suggest that it may be advantageous to adopt more than a single spatial sampling method as empirical evidence because organisms relate to factors at a variety of spatial scales. This is especially true in the case of insects such as butterflies, in which the response of the different life stages may be quite different, with the adults being more mobile than the larval stages. Multiple factors affect species diversity in mountains and its spatial variation along the elevation gradient, such as climate, soil type, water availability, snow cover and topographic heterogeneity (reviewed in Grytnes and McCain 2007). Additional factors that have been studied are area and the mid-domain effect (Grytnes and McCain 2007). Given that many factors can affect changes in diversity along elevation gradient, we find it interesting that such a large portion of the variation in butterfly diversity, and especially in richness and rarity, was explained by satellite-derived vegetation indices. Mid-domain was not an important factor in this system. After removing the effect of area on the relationship between NDVI and the diversity indices examined, NDVI remained a strong predictor of butterfly diversity when using the degree of change among adjacent belts. This suggests that the variation among belts in their productivity is a good indicator (or even surrogate) of spatial heterogeneity or other processes that shape diversity patterns across the mountain. Range size rarity showed a strong negative correlation with both mean and SD of NDVI, whereas its relationship with the heterogeneity between elevation belts was somewhat weaker (Table 1). As we ascend the mountain into areas with lower vegetation cover, few or no trees, and with lower productivity, the proportion of rare species (those found in few elevation belts) increases. This is in


accordance with the prediction that the highest elevations, being more limited in area, more isolated from other areas and having conditions that require high levels of specialization (e.g. strong winds), will tend to show higher levels of rarity and endemism (this was also found for the range size rarity of plants on Mt Hermon; Levin et al. 2007). For locally rare species with smaller ranges in the mountain, both productivity and small-scale spatial heterogeneity in productivity are important. In our case, this may result from the fact that most of the locally-rare species are highaltitude butterflies that are not found elsewhere in Mt Hermon (Benyamini 2002, Pe’er and Benyamini 2008). Many butterfly species of the higher elevation belts reach the southern edge of their global distribution range on Mt Hermon, and comprise of peripheral populations (Benyamini 2002, Pe’er and Benyamini 2008). Lower NDVI values, indicating the lower productivity of highaltitude habitats and high local heterogeneity, are strongly correlated with the occurrence of unique species and high rarity. Species occurring in higher elevations are at higher risk in the face of climatic changes (Parmesan 2006 and references therein), as they occupy particularly small areas along mountain ranges. We should note, however, that our research was constrained to an elevation-range between 500 and 2200 m, and did not reach the summit of Mt Hermon, located in Syria at 2814 m. Thus, rarity patterns may be underestimated, as they are partly affected by the low number of elevation belts that were sampled above the tree-line ( 18501900 m). Indeed, Nogue´s-Bravo et al. (2008) have shown that the different sampling extents along the elevation gradient can affect the relationships found between richness and productivity, which may partly result from the effect of under-sampled rare species. Reports from the uppermost sections of the mountain indicate that the area near the peak of Mt Hermon actually harbours several additional rare butterfly species (Benyamini 1993, 2002). However, different land-use practices (e.g. cutting of trees and overgrazing) in parts of Mt Hermon located in Syria and in Lebanon, beyond the study area, likely lead to a reduction in butterfly diversity and rarity there. It would be interesting to collaborate across the political borders and sample the upper areas of Mt Hermon. In summary, we propose that estimates of local spatial heterogeneity in productivity based on remotely sensed vegetation indices may be useful in predicting butterfly richness along elevation gradients and should be examined in future studies. Such tools may be very useful in predicting and monitoring both plant richness (Levin et al. 2007) and animal richness in remote and inaccessible regions of high conservation importance. This is especially relevant in the face of the rapid climate changes and other environmental changes.

Acknowledgements  We would like to thank Ben Inbar, Hava Goldstein and Didi Kaplan from the Israel Nature and Park Authority for their cooperation, Dubi Benyamini for advice and Catharine van Maanen, Rivka Peretz and Michal Hen-Gal for their help in conducting the field work. Christy McCain provided advice on the mid-domain simulations. This research was supported by the Israel Science Foundation (grant no. 740/04 to

SK). GP was partly supported by a Raymond and Jenine Bollag post-doctoral fellowship at the Hebrew Univ. of Jerusalem.

References Alcock, J. 1987. Leks and hilltopping in insects.  J. Nat. Hist. 21: 319328. Atauri, J. A. and de Lucio, J. V. 2001. The role of landscape structure in species richness distribution of birds, amphibians, reptiles and lepidopterans in Mediterranean landscapes.  Landscape Ecol. 16: 147159. Auerbach, M. and Shmida, A. 1993. Vegetation change along an altitudinal gradient on Mt-Hermon, Israel  no evidence for discrete communities.  J. Ecol. 81: 2533. Bailey, S. A. et al. 2004. Primary productivity and species richness: relationships among functional guilds, residency groups and vagility classes at multiple spatial scales.  Ecography 27: 207217. Bawa, K. et al. 2002. Assessing biodiversity from space: an example from the Western Ghats, India.  Conserv. Ecol. 6: 7, <www.consecol.org/vol6/iss2/art7>. Ben-Dor, E. et al. 2005. Comparison between six model-based methods to retrieve surface reflectance and water vapor content from hyperspectral data: a case study using synthetic AVIRIS data.  Presented at International Conference on Optics and Optoelectronics  ICOL 2005, Dehradun, India. Benyamini, D. 1993. The butterflies of Mt Hermon (Lepidoptera, Rhopalocera and Hesperiidae).  Linn. Belg. 14: 167203. Benyamini, D. 2002. A field guide to the butterflies of Israel; including butterflies of Mt Hermon, Sinai and Jordan.  Keter Publ. House, Jerusalem. Blair, R. B. 1999. Birds and butterflies along an urban gradient: surrogate taxa for assessing biodiversity?  Ecol. Appl. 9: 164 170. Colwell, R. K. 2006. EstimateS: statistical estimation of species richness and shared species from samples. Version 8.0, user’s guide and application.  <http://purl.oclc.org/estimates>. Colwell, R. K. and Hurtt, G. C. 1994. Nonbiological gradients in species richness and a spurious Rapoport effect.  Am. Nat. 144: 570595. Colwell, R. K. and Lees, D. C. 2000. The mid-domain effect: geometric constraints on the geography of species richness.  Trends Ecol. Evol. 15: 7076. Colwell, R. K. et al. 2004. The mid-domain effect and species richness patterns: what have we learned so far?  Am. Nat. 163: E1E23. Dennis, R. L. H. and Shreeve, T. G. 1991. Climatic change and the British butterfly fauna: opportunities and constraints.  Biol. Conserv. 55: 116. Dennis, R. L. H. et al. 1991. A multivariate approach to the determination of faunal structures among European butterfly species (Lepidoptera, Rhopalocera).  Zool. J. Linn. Soc. 101: 149. Dutilleul, P. 1993. Modifying the t test for assessing the correlation between two spatial processes.  Biometrics 49: 305314. Ehrlich, P. R. and Wheye, A. 1988. Hilltopping butterflies revisited.  Am. Nat. 132: 460461. Fleishman, E. et al. 2000. Upsides and downsides: contrasting topographic gradients in species richness and associated scenarios for climate change.  J. Biogeogr. 27: 12091219. Foody, G. M. and Cutler, M. E. J. 2006. Mapping the species richness and composition of tropical forests from remotely sensed data with neural networks.  Ecol. Model. 195: 3742. Gaston, K. J. 1994. Rarity.  Chapman and Hall. Gillespie, T. W. et al. 2008. Measuring and modelling biodiversity from space.  Prog. Phys. Geogr. 32: 203221.

381


Grill, A. et al. 2005. Butterfly, spider, and plant communities in different land-use types in Sardinia, Italy.  Biodivers. Conserv. 14: 12811300. Grytnes, J. A. and McCain, C. M. 2007. Elevation patterns in species richness.  In: Levin, S. (ed.), Encyclopedia of biodiversity. Elsevier, pp. 18. Hall, J. K. et al. 1999. Test of the accuracy of the DEM of Israel  Geological Survey of Israel, Jerusalem. Hawkins, B. A. and Porter, E. E. 2003. Does herbivore diversity depend on plant diversity? The case of California butterflies.  Am. Nat. 161: 4049. Huete, A. R. 1988. A soil-adjusted vegetation index (SAVI).  Remote Sens. Environ. 25: 295309. Huete, A. et al. 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices.  Remote Sens. Environ. 83: 195213. Izco, J. 1998. Types of rarity of plant communities.  J. Veg. Sci. 9: 641646. Jetz, W. and Rahbek, C. 2002. Geographic range size and determinants of avian species richness.  Science 297: 1548 1551. Kark, S. and van Rensburg, B. J. 2006. Ecotones: marginal or central areas of transition?  Isr. J. Ecol. Evol. 52: 2953. Kark, S. et al. 2007. The role of transitional areas as avian biodiversity centers.  Global Ecol. Biogeogr. 16: 187196. Kark, S. et al. 2008. Global environmental priorities: making sense of remote sensing.  Trends Ecol. Evol. 23: 181182. Kelly, L. and Debinski, D. M. 1998. Relationship of host plant density to size and abundance of the regal fritillary Speyeria idalia Drury (Nymphalidae).  J. Lepidopterists’ Soc. 52: 262 276. Kerr, J. T. and Packer, L. 1997. Habitat heterogeneity as a determinant of mammal species richness in high-energy regions.  Nature 385: 252254. Kerr, J. T. and Ostrovsky, M. 2003. From space to species: ecological applications for remote sensing.  Trends Ecol. Evol. 18: 205299. Kerr, J. T. et al. 2001. Remotely sensed habitat diversity predicts butterfly species richness and community similarity in Canada.  Proc. Natl Acad. Sci. USA 98: 1136511370. Kim, K. C. 1993. Biodiversity, conservation and inventory  why insects matter.  Biodivers. Conserv. 2: 191214. Koleff, P. et al. 2003. Measuring beta diversity for presenceabsence data.  J. Anim. Ecol. 72: 367382. Kremen, C. 1992. Assessing the indicator properties of species assemblages for natural areas monitoring.  Ecol. Appl. 2: 203217. Ku¨ hn, E. et al. 2008. Getting the public involved in butterfly conservation: lessons learned from a new monitoring scheme in Germany.  Isr. J. Ecol. Evol. 54: 89103. Kumar, S. et al. 2009. Effects of spatial heterogeneity on butterfly species richness in Rocky Mountain National Park, CO, USA.  Biodivers. Conserv. 18: 739763. Lennon, J. J. et al. 2001. The geographical structure of British bird distributions: diversity, spatial turnover and scale.  J. Anim. Ecol. 70: 966979. Levin, N. et al. 2007. Predicting mountain plant richness and rarity from space using satellite-derived vegetation indices.  Divers. Distrib. 13: 692703. Lillesand, T. M. and Kiefer, R. W. 1994. Remote sensing and image interpretation.  Wiley. Lomolino, M. V. 2001. Elevation gradients of species-density: historical and prospective views.  Global Ecol. Biogeogr. 10: 313. McCain, C. M. 2004. The mid-domain effect applied to elevation gradients: species richness of small mammals in Costa Rica.  J. Biogeogr. 31: 1931.

382

Myers, N. et al. 2000. Biodiversity hotspots for conservation priorities.  Nature 403: 853858. Nogue´s-Bravo, D. et al. 2008. Scale effects and human impact on the elevation species richness gradients.  Nature 453: 216219. Nowicki, P. et al. 2008. Butterfly monitoring methods: the ideal and the real world.  Isr. J. Ecol. Evol. 54: 6988. Oindo, B. O. 2002. Predicting mammal species richness and abundance using multi-temporal NDVI.  Photogramm. Eng. Remote Sens. 68: 623629. Parmesan, C. 2006. Ecological and evolutionary responses to recent climate change.  Annu. Rev. Ecol. Evol. Syst. 37: 637669. Parmesan, C. et al. 1999. Poleward shifts in geographical ranges of butterfly species associated with regional warming.  Nature 399: 579583. Pe’er, G. and Benyamini, D. 2008. A template for publishing the ‘‘conservation chain’’ from problem identification to practical action, exemplified through the campaign for butterfly protection in Israel.  Isr. J. Ecol. Evol. 54: 1939. Pe’er, G. and Settele, J. 2008a. The rare butterfly Tomares nesimachus (Lycaenidae) as a bioinicator for pollination services and ecosystem functioning in the north of Israel.  Isr. J. Ecol. Evol. 54: 111136. Pe’er, G. and Settele, J. 2008b. Butterflies in and for conservation: trends and prospects.  Isr. J. Ecol. Evol. 54: 717. Pe’er, G. et al. 2004. Response to topography in a hilltopping butterfly and implications for modeling nonrandom dispersal.  Anim. Behav. 68: 825839. Pettorelli, N. et al. 2005. Using the satellite-derived NDVI to assess ecological responses to environmental change.  Trends Ecol. Evol. 20: 503510. Pin Koh, L. and Sodhi, N. S. 2005. Importance of reserves, fragments, and parks for butterfly conservation in a tropical urban landscape.  Ecol. Appl. 14: 16951708. Pollard, E. 1977. A method for assessing changes in the abundance of butterflies.  Biol. Conserv. 12: 115134. Pollard, E. 1988. Temperature, rainfall and butterfly numbers.  J. Appl. Ecol. 25: 819828. Pollard, E. and Yates, T. 1993. Monitoring butterflies for ecology and conservation.  Chapman and Hall. Quinn, R. M. et al. 1997. Abundancerange size relationships of macrolepidoptera in Britain: the effects of taxonomy and life history variables.  Ecol. Entomol. 22: 453461. Rahbek, C. 1995. The elevation gradient of species richness: a uniform pattern?  Ecography 18: 200205. Rahbek, C. 2005. The role of spatial scale and the perception of large-scale species-richness patterns.  Ecol. Lett. 8: 224239. Richter, R. 1998. Correction of satellite imagery over mountainous terrain.  Appl. Optics 37: 40044015. Rocchini, D. et al. 2010. Remotely sensed spectral heterogeneity as a proxy of species diversity: recent advances and open challenges.  Ecol. Inform. 5: 318329. Rosenzweig, M. 1995. Species diversity in space and time.  Cambridge Univ. Press. Rouse, J. W. et al. 1973. Monitoring vegetation systems in the Great Plains with ERTS.  In: Freden, S. C. et al. (eds), Third Earth Resources Technology Satellite-1 Symposium Technical presentations, section A, Vol. I. National Aeronautics and Space Administration, Washington, DC, pp. 309317. Rowe, R. J. and Lidgard, S. 2009. Elevation gradients and species richness: do methods change pattern perception?  Global Ecol. Biogeogr. 18: 163177. Samways, M. J. 1989. Insect conservation and the disturbance landscape.  Agric. Ecosyst. Environ. 27: 183194. Seto, K. C. et al. 2004. Linking spatial patterns of bird and butterfly species richness with Landsat TM derived NDVI.  Int. J. Remote Sens. 25: 43094324.


Settele, J. et al. 2009. Ecology of butterflies in Europe.  Cambridge Univ. Press. Shields, O. 1967. Hilltopping: an ecological study of summit congregation behavior of butterflies on a southern California hill.  J. Res. Lepidoptera 6: 69178. Shmida, A. 1977. A quantitative analysis of the Tragacanthic vegetation of Mt Hermon and its relation to environmental factors.  PhD thesis, The Hebrew Univ. of Jerusalem. Shmida, A. 1980. Vegetation and flora of Mt Hermon.  In: Shmida, A. and Livne, M. (eds), Mt Hermon  nature and landscape. Kibbutz Hameouchad, Tel Aviv, pp. 97158. Shmida, A. and Wilson, M. V. 1985. Biological determinants of species diversity.  J. Biogeogr. 12: 120. Shochat, E. 1999. The effect of scrub fragmentation by planted woods on bird communities in the northern Negev.  PhD thesis, Ben-Gurion Univ. of the Negev, Beer-Sheva. Su, J. C. et al. 2004. Beyond species richness: community similarity as a measure of cross-taxon congruence for coarsefilter conservation.  Conserv. Biol. 18: 167173. Thomas, J. A. 2005. Monitoring change in the abundance and distribution of insects using butterflies and other indicator groups.  Phil. Trans. R. Soc. B 360: 339357. Thomas, J. A. et al. 2004. Comparative losses of British butterflies, birds and plants and the global extinction crisis.  Science 303: 18791881.

Tucker, C. J. 1979. Red and photographic infrared linear combinations for monitoring vegetation.  Remote Sens. Environ. 8: 127150. Tucker, C. J. and Sellers, P. J. 1986. Satellite remote-sensing of primary production.  Int. J. Remote Sens. 7: 13951416. van Rensburg, B. J. et al. 2009. Spatial congruence between ecotones and range restricted species: implications for conservation biogeography at the national scale.  Divers. Distrib. 15: 379389. van Swaay, C. A. M. et al. 1997. Monitoring butterflies in the Netherlands and Flanders: the first results.  J. Insect Conserv. 1: 8187. Walker, S. et al. 2003. Properties of ecotones: evidence from five ecotones objectively determined from a coastal vegetation gradient.  J. Veg. Sci. 14: 579590. Williams, P. H. 2000. Some properties of rarity scores used in site quality assessment.  Br. J. Entomol. Nat. Hist. 13: 7386. Williams, P. H. et al. 1996. A comparison of richness hotspots, rarity hotspots, and complementary areas for conserving diversity of British birds.  Conserv. Biol. 10: 144174. Wilson, M. V. and Shmida, A. 1984. Measuring beta diversity with presenceabsence data.  J. Ecol. 72: 10551064.

383


Ecography 34: 364371, 2011 doi: 10.1111/j.1600-0587.2010.06629.x # 2011 The Authors. Ecography # 2011 Ecography Subject Editor: Ken Kozak. Accepted 16 August 2010

Elevational gradients in phylogenetic structure of ant communities reveal the interplay of biotic and abiotic constraints on diversity Antonin Machac, Milan Janda, Robert R. Dunn and Nathan J. Sanders A. Machac (A.Machac@email.cz), Dept of Zoology, Faculty of Science, Univ. of South Bohemia, Branisovska 31, CZ-37005 Ceske Budejovice, Czech Republic.  M. Janda, Biology Center, Czech Academy of Sciences and Faculty of Sciences Branisovska 31, CZ-37005 Ceske Budejovice, Czech Republic. (Present address of MJ: Museum of Comparative Zoology, Harvard Univ., Cambridge, MA 02138, USA.)  R. R. Dunn, Dept of Biology and Keck Center for Behavioral Biology, North Carolina State Univ., Raleigh, NC 27695, USA.  N. J. Sanders, Dept of Ecology and Evolutionary Biology, Univ. of Tennessee 28, Knoxville, TN 37996, USA, and Center for Macroecology, Evolution and Climate, Dept of Biology, Univ. of Copenhagen, DK-2100 Copenhagen, Denmark.

A central focus of ecology and biogeography is to determine the factors that govern spatial variation in biodiversity. Here, we examined patterns of ant diversity along climatic gradients in three temperate montane systems: Great Smoky Mountains National Park (USA), Chiricahua Mountains (USA), and Vorarlberg (Austria). To identify the factors which potentially shape these elevational diversity gradients, we analyzed patterns of community phylogenetic structure (i.e. the evolutionary relationships among species coexisting in local communities). We found that species at low-elevation sites tended to be evenly dispersed across phylogeny, suggesting that these communities are structured by interspecific competition. In contrast, species occurring at high-elevation sites tended to be more closely related than expected by chance, implying that these communities are structured primarily by environmental filtering caused by low temperatures. Taken together, the results of our study highlight the potential role of niche constraints, environmental temperature, and competition in shaping broad-scale diversity gradients. We conclude that phylogenetic structure indeed accounts for some variation in species density, yet it does not entirely explain why temperature and species density are correlated.

A fundamental pattern in biogeography is that both the number of species in a local community (i.e. species density; Gotelli and Colwell 2001) and the composition of communities vary, often systematically, along elevational gradients (Rahbek 2005, McCain 2009). The question, of course, is what drives that variation? Despite a growing number of ecological and evolutionary hypotheses to explain elevational diversity gradients (Sanders 2002, Colwell et al. 2004, Smith et al. 2007, Wiens et al. 2007, Kozak and Wiens 2010, and citations therein), the causes remain poorly understood. One promising approach to infer the underlying processes shaping spatial variation in community composition is the use of phylogenetic tools (Cavender-Bares et al. 2009). Modern approaches build on the earlier use of taxonomic similarity to understand the assembly of communities (Elton 1946, Simberloff 1970). For example, if species within the same genus are more functionally and ecologically similar to one another than distantly related species, observed genus-to-species ratios that are higher than expected (i.e. when compared to a null model) might indicate that competition structures communities. The availability of well-sampled phylogenies has allowed the development of a framework (Webb 2000, Webb et al. 364

2002) which combines the approach of Elton (1946) with the information now available from phylogenetic trees. The framework allows inferring the potential mechanisms that underlie community phylogenetic structure, i.e. phylogenetic relationships among species coexisting within a community. After the actual phylogenetic structure of local communities is assessed, it is compared with structure of communities randomly assembled (i.e. following a specific null model) from the larger, regional species pool. Webb et al. (2002) argued that if the species occurring in a local community are clustered in the phylogeny (i.e. more phylogenetically related than in the null model communities) then the underlying cause of structure is likely to be environmental filtering on shared physiological tolerances, assuming that niches are conserved (Webb et al. 2002, Losos 2008). Alternatively, when species are overdispersed in the phylogeny (i.e. species are less related than in the null model communities) then either interspecific competition or trait convergence is implicated as the structuring force. A lack of a phylogenetic structuring suggests that neutral processes shape the community (Kembel and Hubbell 2006). However, it is worth noting as a caveat that other processes have also been proposed to lead to patterns similar to clustering/overdispersion (e.g. density-dependent


interactions, facilitation during succession; Cavender-Bares et al. 2009). A growing number of studies have used community phylogenetic approaches to better understand spatial variation in community composition (Stevens 2006, Emerson and Gillespie 2008, Graham and Fine 2008, Algar et al. 2009, Cavender-Bares et al. 2009, Vamosi et al. 2009). Yet, only two studies, to our knowledge, have tested whether the phylogenetic structure of local communities might vary along elevational gradients or whether the drivers of diversity along elevational diversity gradients can be inferred by employing a community phylogenetics perspective (Bryant et al. 2008, Graham et al. 2009). Bryant et al. (2008) examined the phylogenetic structure of microbial and plant communities at five sites along a single elevational gradient in the Rocky Mountains, USA, and found that the microbial communities tended to be phylogenetically clustered throughout the entire elevational gradient, but the plant communities were overdispersed at higher elevations. Graham et al. (2009) examined the phylogenetic structure of 189 hummingbird communities in the Andes in Ecuador and found that communities were overdispersed in lowlands, suggesting an important role of interspecific competition. The community phylogenetics approach hinges on an important assumption  that closely related species share similar traits and functions. This assumption has been called phylogenetic conservatism, niche conservatism, or evolutionary stability (Losos 2008). Importantly, neither the Bryant et al. (2008) nor the Graham et al. (2009) study tested for phylogenetic conservatism. Moreover, the Bryant et al. (2008) and the Graham et al. (2009) focused on single elevational gradients such that the generality of the patterns they documented is hard to assess. In this study, we examine patterns of ant species density and community phylogenetic structure along three elevational gradients. We tested two predictions: 1) communities at high-elevation sites would be phylogenetically clustered, as would be expected if traits are conserved and only closely related species of a subset of lineages possessed the traits which allowed them to persist at cold, highelevation sites, and 2) communities at low-elevation sites would be phylogenetically overdispersed in the phylogeny, as would be expected if interspecific competition rather than environmental filtering shaped the composition of local communities. Finally, we assessed whether the elevational pattern in phylogenetic structure is sufficient to explain patterns in species density, or whether environmental gradients have effects on species density above and beyond the effects of phylogenetic structure.

Methods The data We obtained data on the identities and occurrences of species within local communities from two published studies (Chiricahua Mountains, USA: Andersen 1997; Vorarlberg Mountains, Austria: Glaser 2006) and our own work (southern Appalachian Mountains, USA: Sanders et al. 2007). Importantly, each of the datasets consists of samples from local communities along extensive elevational gradients (Table 1); the data are not interpolated ranges or derived from niche models. For detailed information on geography of montane systems and sampled sites, see Supplementary materials Appendix 1. Constructing phylogenies We constructed three phylogenies, one for each of the montane systems, based on published genus-level phylogenies (Brady et al. 2006, Moreau et al. 2006). We adopted the molecular datasets from these studies from the TreeBase database <www.treebase.org>. Nine of the 175 species considered here lacked species- and genus-level molecular data. In these few cases (5% of all species in this study), species were substituted with closely related taxa with relationships derived from Boltonâ&#x20AC;&#x2122;s (2003) classification. We extended the molecular dataset using 80 additional sequences (using the same genes as in the Brady et al. (2006) and Moreau et al. (2006) studies) available for particular species in GenBank in order to incorporate within-genus variability and to resolve some of the genuslevel polytomies (especially in the genera Pheidole and Camponotus). These additional sequences, their GenBank codes, as well as the substituted taxa are listed in the Supplementary materials Appendix 2. We aligned the edited sequences in MAFFT, ver. 6 (Katoh et al. 2002). To reconstruct the phylogenies, we employed a maximum likelihood approach with topology constraint in PAUP 4.0 (Swofford 1993). The tree topology, on which molecular data were forced, corresponded with the genus-level phylogeny of Bolton (2003), Brady et al. (2006), and Moreau et al. (2006). We estimated branch lengths on the basis of substitution rates in a combined molecular dataset. For more details, see Supplementary materials. Assessing phylogenetic structure of communities Prior to examining the phylogenetic structure of communities, we tested for niche conservatism/phylogenetic conservatism. Based on our understanding of the natural

Table 1. Location, number of sites, and elevational span sampled for the community data used in the analyses. The entire elevational extent of Chiracahua Mts: 11002900 m, Vorarlberg Mts: 3503000 m, Smoky Mts: 2502000 m. More information on geography of the montane systems and sites sampled is given in the Supplementary material Appendix 1. Author

Montane system

Location

No. of sites

Elevation range (m)

Andersen (1997) Glaser (2006) Sanders et al. (2007)

Chiricahua Mts Vorarlberg Smoky Mts

Arizona, USA Austria Tennessee/N Carolina, USA

9 sites 18 sites 29 sites

14002600 4002100 3791828

365


history of ants and previous studies, we tested for conservatism in three traits: habitat associations (woodlands, shrublands, meadows and grasslands), nest site (in soil, under rocks, mounds and ground nests, rotting wood, canopy and trees), and worker size (measured as the Weber’s length of thorax). Each of these three suites of traits represents a significant axis of the ecological niche of ants and can be related to interspecific competition (worker size, nest site) or environmental tolerance (habitat association, nest site). We employed the random tree-length distribution algorithm for discrete traits (Cubo et al. 2005) that randomly permutes taxa (and their character values) along the phylogeny, while holding the topology as well as the branch lengths constant. Each character is mapped on the phylogeny through the maximum parsimony procedure. Afterwards, the number of character steps along the actual phylogeny is contrasted with the distribution of the number of steps in the 10 000 randomly constructed phylogenies. In the case of continuous characters, we used squared-length parsimony (Cubo et al. 2005). We performed these analyses in Mesquite 2.7 (Maddison and Maddison 2002). Our results were not affected by the bias potentially introduced by phylogenetic signal in occurrence frequencies (i.e. closely related species appear in many communities as present) (Kembel 2009) because we employed a null model that takes into consideration the prevalence of species in the communities and samples them accordingly (see below, Gotelli 2000). After testing for niche/phylogenetic conservatism, we estimated the phylogenetic structure of each community from the three montane systems using two indices: mean phylogenetic distance (MPD) and mean nearest neighbor distance (MNND; Webb et al. 2002). MPD is an estimate of the average phylogenetic relatedness (on basis of branch lengths) between all possible pairs of taxa in a local community. MNND, in contrast, is an estimate of the mean phylogenetic relatedness between each taxon in a local community and its nearest relative. We then calculated standardized NRI and NTI indices. The NRI and NTI describe the difference between average phylogenetic distances (MPD and MNND, respectively) in the observed and randomly generated null communities, standardized by the standard deviation of phylogenetic distances in the null communities (Webb et al. 2008). We used R 2.8 to calculate NRI and NTI (Kembel et al. 2009). Since values of NRI and NTI were highly correlated (r 0.837, pB0.001), we report only NRI values in subsequent analyses. To assess whether the observed NRI values differed significantly from zero, we compared them to NRI values of null communities generated by Gotelli’s swap algorithm (Gotelli 2000); i.e. the occurrence matrix is randomized holding the number of species per sample and the frequency of occurrence of each species across samples constant. The phylogeny used for the calculation of NRI was not fully resolved and branch-length estimates were not available for all of the taxa. Therefore, we examined the impact of the phylogeny’s resolution (defined as branch lengths availability) on NRI by estimating three different distances: species-level distances, genus-level distances, and simple Grafen’s (1989) distances based on the tree topology. We then calculated new NRI values using these 366

distances for each of the three montane systems and mutually compared them. Environmental variables To examine the relationship between ant species density and climate, we extracted information on annual precipitation and annual mean temperature for each community in each montane region from the WorldClim v1.4 database (<www.worldclim.org>; Hijmans et al. 2005) using ArcView GIS (ver. 3.2, Esri 19922000; ESRI, Redlands, CA). WorldClim data pose some problems, especially in montane systems, because the resolution of the data is 1 km2. Considerable variation in temperature can occur within one square kilometer, especially in montane systems. As a check of the potential bias of using WorldClim data, we examined whether mean annual temperature data obtained from WorldClim were correlated with mean annual temperature data obtained from measurements of temperature from dataloggers arranged along the elevational gradient in the Smokies. Ideally, we would have measured temperature and precipitation data in each of the montane systems. However, because the WorldClim temperature data were correlated with the measured temperature data (r 0.998, pB0.001), and measured climate data were unavailable for two of the three gradients, we instead use WorldClim data. We also note that such an approach is common to other studies of elevational diversity gradients (McCain 2009) such that our work should be directly comparable. We chose these focal environmental variables because they are often strongly correlated with ant species density (Kaspari et al. 2000, Sanders et al. 2007, Dunn et al. 2009). Analyses We related elevation and the climate variables (mean annual temperature, annual precipitation) to species density (the number of species occurring in a local community) and to phylogenetic structure (NRI) of local communities using linear mixed-effect models. In the model, identity of a montane system was treated as a random effect, and elevation, mean annual temperature, annual precipitation, and all their combinations as explanatory variables. We used the maximum likelihood procedure to fit each model, and we compared those models via Bayesian information criterion (BIC). As temperature appeared to be the best predictor of both species density and community phylogenetic structure, we conducted further analyses to tease apart the mechanisms linking these variables. The effects of temperature on NRI and species density could operate in one of two ways. First, temperature could influence phylogenetic structure and phylogenetic structure could in turn influence species density. In this scenario, species density patterns are simply a consequence of species of a few clades possessing the traits necessary to persist in harsh conditions, such as the cold. Or second, temperature could influence species density via mechanisms independent of phylogenetic structure. Any of a variety of effects are possible, including effects on speciation rates (Davies et al. 2004) or abundance mediated


effects on extinction (Willig et al. 2003). To distinguish between these possibilities, we assessed whether the effect of phylogenetic structure (estimated as NRI) on species density was significant even when the effect of temperature was already included as an explanatory variable in the model predicting species density. This outcome would indicate that phylogenetic structure has a direct effect on species density. Alternatively, if phylogenetic structure is not related to species density after temperature is included in the model, the result implies that phylogenetic structure may influence patterns of species density, but is insufficient as a complete explanation for them. All the analyses were conducted in R 2.8 (Pinheiro and Bates 2000). We note that spatial autocorrelation within montane systems can inflate type I errors in statistical tests. However, because interpreting the coefficients from spatial regression can be challenging at best (Bini et al. 2009), we do not use spatial regression techniques in these analyses (e.g. SAM  Rangel et al. 2006) and instead rely on BIC and R2 values as estimates of goodness of fit.

Results The three elevational gradients we considered consisted of 56 local ant communities with 175 ant species from seven subfamilies (Supplementary material Appendix 3). Along each of the gradients, ant species density decreased with elevation (Fig. 1). Bayesian information criterion indicated that the best environmental predictor of ant species density was annual mean temperature (BIC 364.04, positive relationship; Table 2). Only slightly less plausible were the models including elevation (DBIC 0.11; Table 2) and temperatureprecipitation (DBIC 0.64; Table 2). We found strong evidence for niche conservatism for each of the three traits we examined. The evolutionary stability of niches (represented by habitat associations, nest site, and worker size) was consistent among the montane

Species density

(a)

(b)

(c)

20

25

15

30 15

10 10

5 500

Net relatedness index

systems for each of the examined traits (10 000 randomizations, p B0.05) (Table 4). In other words, not only were the traits examined here phylogenetically conserved, they were conserved everywhere. NRI was correlated with temperature (BIC 338.95, R2 0.36; Table 3); the correlation was negative in each of the three montane systems (Fig. 2). This would be expected if environmental stress (due to lower temperatures at higher elevations) acted as a filter on lineages at high elevations and competition structured communities at low elevations. Besides temperature, the next best model of NRI with only a minor difference in BIC comprised elevation (DBIC 0.68; Table 3). Low-elevation communities tended to be significantly overdispersed (4 sites), whereas communities at higher elevations tended to be significantly clustered (7 sites) (Fig. 2). Ant species density was significantly and positively correlated both with temperature (p B0.001, R2 0.55) and NRI (p 0.002, R2 0.17) in independent models (above). However, once temperature had been added in the model of species density, the contribution of NRI became insignificant (p 0.76, R2NRI B0.01). Conversely, even if the model of species density already comprised NRI, the effect of temperature remained significant (p B0.001, R2temp 0.44). These outcomes suggest that both the species density and community phylogenetic structure are mutually independent products of temperature variation. The estimates of NRI are robust to phylogenetic resolution for the Smoky Mountains and Vorarlberg ant communities (Table 5). NRI for the communities from the Chiricahua Mountains varied with phylogenetic resolution, however, perhaps because there were only 9 sites sampled in the Chiricahuas and few genera were monotypic (as opposed to the Smoky Mountains and Vorarlberg) such that it was possible for polytomies within genera to have a greater effect. It could be argued that the species level phylogeny for the Smoky Mountains and Vorarlberg (due to its lower resolution) approaches the genus-level phylogeny; thus, resulting in a tight correlation between the respective NRI values. To avoid this artifact and examine

1000

1500

5 500

1000 1500 2000

1400

1800

2200

2600

1400

1800

2200

2600

1.0 4 2

0.0

–1

0 –2

–1.0

–3 500

1000

1500

500

1000 1500 2000 Elevation (m)

Figure 1. Community characteristics for (a) Smoky Mountains, (b) Vorarlberg, (c) Chiricahua Mountains. Species density (i.e. number of species in a local community) and net relatedness index (NRI) are plotted against elevation. Each point is a site sampled for ants.

367


Table 2. Models of ant species density. The most parsimonious model was identified via Bayesian information criterion (BIC). Abbreviations refer to temperature (Temp), precipitation (Precip), and elevation (Elev).

Null model Elevation Temper