Page 1

SCIENTIFIC REPORT

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia Sandra Englhart, Peter Navratil, Uwe Ballhorn, Juilson Jubanski and Florian Siegert Kalimantan Forests and Climate Partnership


SCIENTIFIC PAPER Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia. Sandra Englhart, Peter Navratil, Uwe Ballhorn, Juilson Jubanski and Florian Siegert

Kalimantan Forests and Climate Partnership April 2014


ACKNOWLEDGEMENTS This report was prepared by RSS – Remote Sensing Solutions GmbH, for the Kalimantan Forests and Climate Partnership (KFCP). The aim of this study was to assess aboveground biomass (AGB) variability of different tropical forest types and degradation levels by large-scale airborne Light Detection and Ranging (LiDAR) measurements in Central Kalimantan, Indonesia. The analysis in the report was prepared by Dr Sandra Englhart, Dr Uwe Ballhorn, Dr Juilson Jubanski, Peter Navratil, Kristina Konecny and Prof. Dr Florian Siegert. The paper was reviewed by Dr Robert Waterworth, with support provided by Dr Rachael Diprose.

Copy editor: Lisa Robins Reviewer: Robert Waterworth Layout and publication: James Maiden and Nanda Aprilia

This research was carried out in collaboration with the governments of Australia and Indonesia, but the analysis and findings presented in this paper represent the views of the authors and do not necessarily represent the views of those governments. Any errors are the authors’ own. The paper constitutes a technical scientific working paper and, as such, there is potential for future refinements to accommodate feedback and emerging evidence.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page i


ABOUT THE AUTHORS RSS – Remote Sensing Solutions GmbH (www.rssgmbh.de), founded in 1999, is a company based in Germany focusing on earth observation specialised in satellite image processing and interpretation, aerial image interpretation and photogrammetry, application development for geographic information systems (GIS) and digital cartography. RSS experts have academic backgrounds in the fields of forestry, geography or cartography. Key personnel have more than 15 years experience in development cooperation and the management of national and international projects. Prof. Dr Florian Siegert Florian is managing director of RSS GmbH, and has a strong scientific background in ecology and more than 15 years’ experience in remote sensing. He is an accredited expert for the European Space Agency (ESA) and the German Space Agency (DLR) and was/is Principal Investigator for ERS, ENVISAT, TerraSAR-X, ALOS and RapidEye. In addition, he has extensive experience in international development cooperation and the management and execution of international projects with special focus on Asia and Africa. Dr Siegert is also a Professor at the University of Munich, Faculty of Biology, teaching biology and global change ecology at the GeoBio Center of the Ludwig-Maximilians-University and has successfully supervised and trained more than 15 PhD candidates and Master’s course students. His major scientific interest lies in studying ecology and natural resource functions of tropical rainforests and their role in global climate change. Dr Sandra Englhart Sandra is a biologist with expert knowledge in aboveground biomass modelling of tropical forests. Her research interests are multi-temporal and multi-sensoral (multispectral, SAR and LiDAR) environmental monitoring/change detection approaches, imaging as well as geo-statistics, GIS and environmental modelling. At RSS, she is responsible for the implementation of environmental and REDD+ projects and scientific projects in the context of RADAR applications. Dr Uwe Ballhorn Uwe is a forester and focuses on tropical forests and peatlands. His main interests include below and aboveground biomass modelling and carbon estimation using LiDAR and multi-spectral data, and the statistical assessment of spatial patterns in forests, especially in regard to deforestation and forest degradation. He is experienced in organising and conducting field surveys. At RSS he is responsible for the implementation of forestry, REDD+ projects and scientific projects in the context of LiDAR and forest applications. Dr Juilson Jubanski Juilson is a specialist in photogrammetry and geodesy and has wide practical experience in planning and conducting LiDAR surveys and LiDAR data processing, especially in the development of automated processing algorithms for the analysis of photogrammetric and LiDAR data. One of his focal fields is the development of LiDAR derived aboveground biomass models. At RSS, Dr Jubanski was responsible for the data processing of a full coverage 7000 km2 LiDAR dataset in the framework of KFCP.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page ii


Peter Navratil (Dip. Geog.) Peter has broad expertise in remote sensing and environmental monitoring with special focus on forestry and climate change mitigation. He has worked as a consultant for a wide variety of development cooperation and nature conservation projects in Central and Southeast Asia since 2005. Mr Navratil has a well-founded knowledge in digital image processing of optical and SAR satellite data, especially in automation of data processing, object-based classification and validation. He has wide practical experience in conducting aerial and field surveys and capacity building measures. He is qualified to conduct project management in accordance with the GIZ management model, ‘Capacity works’. At RSS he is responsible for the project management and technical implementation of forestry and climate change mitigation projects. Kristina Konecny (M.Sc.) Kristina is a cartographer specialising in remote sensing, GIS and photogrammetry. She is experienced in digital image processing and interpretation of optical satellite and aerial images, geometric analysis of LiDAR/photogrammetrically-derived 3D point clouds and geo-statistics. At RSS she is responsible for the analysis of LiDAR data. RSS – Remote Sensing Solutions GmbH Isarstr. 3 82065 Baierbrunn/München Germany www.rssgmbh.de info@rssgmbh.de

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page iii


EXECUTIVE SUMMARY Tropical peat swamp forests in Indonesia store significant amounts of carbon and are responsible for enormous carbon emissions every year due to forest degradation and deforestation. These forest areas are the focus of REDD+ (Reducing Emissions from Deforestation and forest Degradation) projects, which require accurate monitoring of their carbon stocks or aboveground biomass (AGB). Our study objective was to evaluate large-scale, airborne Light Detection and Ranging (LiDAR) measurements of tropical forest peatlands in an area spanning 693,806 ha in Central Kalimantan province. A Canopy Height Model (CHM) and AGB estimation model were created based on LiDAR height metrics. The variability in canopy height and carbon stocks for different forest types, former and recent logging areas, burned forest areas and natural drainage areas, were thoroughly investigated. The results showed that mean AGB and CHM values of each forest type differed significantly from each other. The analyses of the logging areas revealed that forests recover from logging impact, as higher mean AGB values are featured in older logging sites. The results of the investigated burned areas show that mean AGB values decrease with increasing number of fire events. The natural drainage area had a low mean AGB value of 168 t/ha, indicating high degradation, potentially due to forest accessibility via abandoned logging railways. These results demonstrate the great potential of airborne LiDAR measurements to analyse the high variability in forest carbon stocks, leading to a better understanding of the degradation and recovery rate.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page iv


TABLE OF CONTENTS ACKNOWLEDGEMENTS

i

ABOUT THE AUTHORS

ii

EXECUTIVE SUMMARY

iv

TABLE OF CONTENTS

v

LIST OF FIGURES

vi

LIST OF TABLES

vii

LIST OF ACRONYMS

viii

1

INTRODUCTION

1

2

MATERIALS AND METHODS

3

2.1

Project area

3

2.2

Data

4

2.2.1

Field inventory data

4

2.2.2

LiDAR data

5

2.3

Data analysis

5

2.3.1

LiDAR filtering and DTM generation

5

2.3.2

Development of the LiDAR AGB estimation model

6

2.3.3

Canopy Height Model

7

2.3.4

Analysis of AGB variability

7

3

LIDAR AGB ESTIMATION MODEL

10

4

EVALUATION OF AGB VARIABILITY

11

5

RESULTS

13

5.1

AGB variability of different forest types

13

5.2

AGB variability of recent and former logging areas

18

5.3

AGB variability of burned areas

21

5.4

AGB variability of natural drainage areas

23

5.5

AGB variability across the peat dome

25

6

7

DISCUSSION

27

6.1

Evaluation of the results

27

6.2

Uncertainties

28

6.3

Strengths and limitations for AGB and variability analyses

28

CONCLUSIONS AND RECOMMENDATIONS

REFERENCES

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

30 31

Page v


LIST OF FIGURES Figure 1: Location of the project area ............................................................................................................. 3 Figure 2: LiDAR derived DTM accuracy ........................................................................................................... 6 Figure 3: Managed logged peat swamp forest................................................................................................ 8 Figure 4: Small logging trails in peat swamp forests ....................................................................................... 8 Figure 5: Fire events ........................................................................................................................................ 9 Figure 6: LiDAR AGB estimation model ......................................................................................................... 10 Figure 7: Overview of analysed areas for AGB variability ............................................................................. 12 Figure 8: Overview of the different forest types ........................................................................................... 13 Figure 9: Statistics of AGB and CHM variability for different forest types .................................................... 14 Figure 10: LiDAR height and vegetation height profiles of low pole peat swamp forests (PSF) ................... 15 Figure 11: LiDAR height and vegetation height profiles of mixed peat swamp forests (PSF) ....................... 16 Figure 12: LiDAR height and vegetation height profiles of tall interior peat swamp forests (PSF) ............... 17 Figure 13: LiDAR height and vegetation height profiles of riparian forests .................................................. 18 Figure 14: Recent logging areas and LiDAR-derived AGB map...................................................................... 19 Figure 15: Statistics of AGB and CHM variability for former and recent logging areas ................................ 20 Figure 16: Overview of burned areas ............................................................................................................ 21 Figure 17: Statistics of AGB and CHM variability for number of fire events ................................................. 21 Figure 18: Statistics of AGB and CHM variability for the last fire events ...................................................... 22 Figure 19: Overview of the natural drainage area ........................................................................................ 23 Figure 20: Statistics of AGB and CHM variability for a natural drainage area............................................... 24 Figure 21: Transect across the peat dome .................................................................................................... 25

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page vi


LIST OF TABLES Table 1: Specifications of the LiDAR acquisition ............................................................................................. 5 Table 2: Characteristics of the different AGB variability categories and its classes...................................... 11 Table 3: Statistics of AGB and CHM variability for different forest types ..................................................... 14 Table 4: Statistics of AGB and CHM variability for former and recent logging areas.................................... 20 Table 5: Statistics of AGB and CHM variability for number of fire events. ................................................... 22 Table 6: Statistics of AGB and CHM variability for the last fire events ......................................................... 23 Table 7: Statistics of AGB and CHM variability for a natural drainage area .................................................. 24

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page vii


LIST OF ACRONYMS 3D

3-dimensional

AGB

Aboveground biomass

ALOS

Advanced Land Observing Satellite

C

Carbon

CH

Centroid height

CHM

Canopy height model

CO2

Carbon dioxide

DBH

Diameter at breast height

dGPS

Differential Global Positioning System

DSM

Digital surface model

DTM

Digital terrain model

ENVISAT

ENVIronmental SATellite

ERS

European Remote Sensing (satellites)

ESA

European Space Agency

GHG

Greenhouse gas

GIS

Geographic information system

GIZ

Gesellschaft f端r International Zusammenarbeit, the German Development Agency

Gt

Giga tonne

ha

Hectare

KFCP

Kalimantan Forests and Climate Partnership

km

Kilometre

LiDAR

Light detection and ranging

MODIS

Moderate Resolution Imaging Spectroradiometer

MRP

Mega Rice Project

MRV

Measurement, reporting and verification

NOAA

National Oceanic and Atmospheric Administration

PPR

Predictive power of the regression

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page viii


PSF

Peat swamp forest

QMCH

Quadratic mean canopy height

RADAR

Radio Detection and Ranging

REDD

Reducing emissions from deforestation and forest degradation

REDD+

Reducing emissions from deforestation and forest degradation and the role of conservation, sustainable management of forests and enhancement of forest carbon stocks in developing countries

RMSE

Root mean square error

RSS

Remote Sensing Solutions GmbH

SAR

Synthetic Aperture Radar

T

Tonne

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page ix


1 INTRODUCTION In 2008, worldwide greenhouse gas (GHG) emissions from deforestation and forest degradation were estimated to account for about 6–17% of the total anthropogenic carbon dioxide (CO2) emissions (van der Werf et al. 2009). In the period 1990–2005 approximately 13 million hectares (ha) of tropical forests were deforested annually and South and Southeast Asia had one of the highest annual deforestation rates (0.98%) between 2000 and 2005 (FAO 2006). One important measure to curb GHG emissions from this sector is reducing emissions from deforestation and forest degradation (REDD+) 1 programs (Campbell 2009). The focus of REDD+ is on tropical forests, which comprise approximately 40% of terrestrial carbon (C) (FAO 2009; Page et al. 2009). Furthermore, tropical peatlands accumulate additional carbon in thick below-ground peat deposits that are sustained by the above-ground, intact forests. In Southeast Asia, the largest known tropical peat deposits are found in Indonesia where around 55–58 Gt C is stored below ground and around 18.6 Gt C above ground in forests (Baccini et al. 2012; Jaenicke et al. 2008; Page et al. 2011). Indonesia is one of the largest carbon emitters worldwide; mostly caused by degradation and deforestation of its peatland areas and forests due to large-scale agricultural development and exploitation of forest timber resources (Miettinen and Liew 2012). As a result of these activities and consequential emission rates, Indonesia is a prime target for REDD+ projects. Aboveground biomass (AGB) loss and associated CO2 emissions resulting from forest degradation by logging and fire are difficult to assess and monitor because impacts can vary significantly. The most accurate method of AGB estimation is based on forest inventories, which can be either done by destructive sampling or tree measurements, from which AGB values may be extrapolated by using allometric equations. Destructive sampling provides more accurate AGB values because the total weight of the felled trees is oven-dried and weighted. However, trees and other vegetation need to be felled for the process, which can have a negative impact on forests and forest carbon storage. Therefore, destructive sampling is mainly used to set up allometric equations. Such equations cover a range of specificity, ranging from individual tree species to broader forest types (Chave et al. 2005; Steininger 2000). A sufficient number of field measurements is mandatory for developing an AGB estimation model and for evaluating the AGB estimation results. Although this approach provides precise AGB estimations, the biotic and structural complexity of tropical ecosystems make forest inventories difficult, time-consuming and expensive. Due to the inaccessibility of the forests, the most effective monitoring would be based on satellite or airborne observations (Gibbs et al. 2007). No remote sensing instrument can directly retrieve AGB. Therefore, a sufficient number of field inventory measurements is also mandatory for developing AGB estimation models based on satellite or airborne signals and for evaluating the AGB estimation results (Goetz and Dubayah 2011). The implementation of Light Detection and Ranging (LiDAR) measurements has rapidly grown in recent years due to its ability to precisely quantify the vertical structure of the forest and other attributes, such as canopy height distribution, tree height and crown diameter (Duncanson et al. 2010; Gleason and Im 2011; Jung and Crawford 2012; Vincent et al. 2012). The potential of airborne LiDAR to predict AGB in tropical forests has been demonstrated in many studies (Asner et al. 2010; Asner et al. 2012; Mascaro et al. 2011), whereby only few have been carried out in 1

together with conservation and sustainable management of forests and the enhancement of forest carbon stock.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 1


tropical peat swamp forest ecosystems (Ballhorn et al. 2011; Englhart et al. 2013; Jubanski et al. 2013; Kronseder et al. 2012; Meyer et al. 2013; Sweda et al. 2012). In these studies different point cloud statistics of the vegetation height and canopy cover were tested for their performance to predict peat swamp forest AGB using linear, multiple linear and power functions. It was found that appropriate parameters to estimate AGB were relative height quartiles (r2 = 0.71, RMSE = 115.2 t/ha or r2 = 0.7, RMSE = 28.6 t/ha) and other point cloud height distributions such as Quadratic Mean Canopy Height (QMCH) (r2 = 0.84) or Centroid Height (CH) (r2 = 0.75, RMSE = 20.5 t/ha and r2 = 0.88, RMSE = 106 t/ha) (Ballhorn et al. 2011; Jubanski et al. 2013; Kronseder et al. 2012; Meyer et al. 2013). The accuracy could be improved by taking the LiDAR point density into account for developing AGB models (Jubanski et al. 2013), The aim of this study is to assess AGB variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia. In Central Kalimantan, authorised and illegal logging increased in the 1990s as a direct result of the Mega Rice Project (MRP), initiated by the Indonesian Government in 1995, which resulted in the degradation of more than one million hectares of forest. Some 6000 km of drainage and irrigation channels were excavated into the peatlands, which allowed access into previously inaccessible peat swamp forests. In section 2 (‘Materials and methods’) the study area is introduced and the design of the field inventory and the method used to estimate AGB from field inventory data is described. Further airborne LiDAR data used and the processing of this data is explained. Finally the method used to derive LiDAR AGB models through correlating LiDAR point cloud statistics to field inventory AGB estimates is explained. Section 3 (‘LiDAR AGB estimation model’) describes the resulting LiDAR AGB regression model. In section 4 (‘Evaluation of AGB variability’) and section 5 (‘Results’) an overview is given of the area of investigation. AGB variability of the different forest types (low pole peat swamp forest, tall interior peat swamp forest, mixed peat swamp forest, and riparian forest) is evaluated. An analysis is also included of AGB estimations on recovery rates, recent and former (1990s) logging areas, vegetation in burned areas (based on number of fire events and last fire occurrence), and natural drainage areas (small rivers). Section 6 (‘Discussion’) presents the study results, and evaluates and compares these with other studies. Also the strengths and limitations of LiDAR to derive AGB and its variability are discussed. Further the benefits of LiDAR for baseline assessments, reducing uncertainty in upscaling, estimating recovery rates, determining proportions/ratios of timber extraction and the integration into a wider system are discussed and how these benefits are weighed up in the light of the costs of conducting LiDAR surveys. Finally, the strengths and limitations of allometric equations/modelling are evaluated and how this affects AGB estimates produced by LiDAR. Finally, section 8 (‘Conclusion and recommendations’) outlines the study’s overall conclusions and recommendations for future LiDAR surveys in tropical forests, especially tropical peat swamp forests, with the aim to estimate AGB values.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 2


2 MATERIALS AND METHODS 2.1

Project area

The study area (693,806 ha) is located in Central Kalimantan province, Indonesia, and covers selected parts of the former MRP (Figure 1). The area is characterised by peatlands, whereby the southern parts consist of agricultural and fallow land, and the northern part is covered primarily by peat swamp forests, which were logged in the early and mid1990s. The whole study area has been under severe anthropogenic pressure for the last three decades. The most severe impact was caused by the former MRP, conceptualised by the Indonesian Government in 1995 to convert an uncultivated area of 988,568 ha through the construction of about 6000 km of drainage and irrigation channels between 1996 and 1997. As a consequence of drainage and degradation, the forest of the peat area is very sensitive to fire during the dry season. Another driver of forest degradation in the area is illegal selective logging, which causes small-scale impacts in the forest canopy. Valuable trees are felled, cut into appropriate lengths and then dragged along narrow skid trails to the nearest river or channel where they are transported away. Figure 1: Location of the project area

Source: Basemap, ESRI; outlines provided by KFCP Note: Overview of the island of Borneo with the location of the project area (red) covered by airborne LiDAR in Central Kalimantan (yellow), Indonesia (left). The detailed map (right) shows the location of the project area within the Blocks (A–E) of the former Mega Rice Project. It also displays the location of the KFCP area.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 3


2.2 2.2.1

Data Field inventory data

Field inventory data was collected in the years 2010 and 2011 whereby inventory plots with different plot sizes were established in forested and regrowing areas. The sampling plot design was based on guidelines provided by Pearson et al. (2005). The plots were distributed within the LiDAR dataset based on accessibility and representativeness of the different land cover types and degradation stages, so that the whole biomass range (from woody regrowth to mature forest) was covered. Permanent inventory plots were not available due to the inaccessibility of the forests and ongoing deforestation. In forested areas, three circular nested plots with radii of 4 m, 14 m and 20 m were recorded. Inside each nest, trees of a certain diameter at breast height (DBH) were measured depending on degradation intensity: 2–10 cm or 5–20 cm (within the 4 m radius), 10–20 cm or 20–50 cm (within 14 m radius), and greater than 20 cm or 50 cm (within 20 m radius). In regrowing areas, rectangular plots of 20 x 50 m² were used and all saplings and trees within this area were recorded. Within both plot types, the following parameters were recorded: DBH, tree height and tree species to obtain estimates of wood density from databases provided by Chudnoff (1984), World Agroforestry Centre (2011) and IPCC (2006). If the tree species could not be identified an average specific wood density for Asian tropical trees of 0.57 Mg m-3 was applied (Brown 1997). DBH, tree height, and wood specific density for each tree were used to estimate AGB (t/ha) in a combination of allometric models from Hughes et al. (1999) for samplings (if DBH < 5 cm and height ≤ 1.3 m) or trees (if DBH < 5 cm and height > 1.3 m), and Chave et al. (2005) for moist tropical forest stands including DBH and tree height (if DBH ≥ 5 cm and height > 1.3 m):

10 −6 ⋅ e 4.7472+1.0915⋅ln ( DBH ² )  AGB = 1.14 ⋅ 10 −6 ⋅ e 4.9375+1.0583⋅ln( DBH ²) e −2.977+ln( ρ ⋅DBH ²⋅h ) 

if DBH < 5 cm and h ≤ 1.3 m if DBH < 5 cm and h > 1.3 m

(1)

if DBH ≥ 5 cm and h > 1.3 m

Where DBH is diameter at breast height, h is tree height and ρ is wood specific density. Altogether 55 plots were sampled in 2010 and 2011 that were inside the LiDAR tracks of 2011 and had an AGB range from 0.0 t/ha to 375 t/ha. These allometric models were chosen because these are ecological morphologic models based on DBH, tree height and wood specific density whereas other commonly used allometric models (Brown 1997; IPCC 2006) are derived from tree volume calculations and are solely based on an average wood density. As tree volume is an extrapolation of DBH and height, these other commonly used models include an additional source of error and therefore error propagation, and were not considered for AGB extrapolation in this study. In addition, the allometric model used in this study can be applied across a broad range of tropical forests, which is advantageous for integration on a national level for example in the Southeast Asia region. However, the use of generic allometric models may introduce errors in AGB estimations. Two study sites in Indonesia were included in the allometric model of Chave et al. (2005), which were categorised as moist tropical forests. This model is commonly used for extrapolating AGB from field inventory in Indonesia (Dossa et al. 2013; Griscom et al, 2014). Rutishauser et al. (2013) compared different regional and general allometric models to estimate AGB at an Indonesian site through destructive sampling and Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 4


concluded that the allometric model developed by Chave et al. (2005) provided the best AGB estimates when applied to the destructive samples.

2.2.2

LiDAR data

The airborne LiDAR dataset was acquired in a flight survey by Surtech between 15 August 2011 and 15 October 2011. During the survey, 693,806 ha was scanned (Figure 1). Further data specifications and acquisition details are provided in Table 1 and additional details on the LiDAR dataset can be found in Surtech (2011). Table 1: Specifications of the LiDAR acquisition Specification LiDAR system Optech Orion M200 Acquisition date 15 August – 14 October Power 100 Khz Nominal altitude 800 m Wavelength 1.064 µm Half scan angle ±11° Average point density 10.7 points/m2

2.3 2.3.1

Data analysis LiDAR filtering and DTM generation

A crucial step within the Digital Terrain Model (DTM) generation is the LiDAR filtering. A hierarchic robust filter was applied to the LiDAR point clouds, separating the ground and non-ground (vegetation) points (Pfeifer et al. 2001). The linear adaptable prediction interpolation (kriging) was utilised to generate the DTM with a resolution of 1 m. The accuracy of the DTM was evaluated on the basis of differential Global Positioning System (dGPS) measurements. Altogether, 93 dGPS measurements were available covering different land cover classes (peat swamp forest, open forest, burned areas, ferns and shrubs) and resulted in a RMSE of 0.16 m (Figure 2).

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 5


Figure 2: LiDAR derived DTM accuracy

Source: Remote Sensing Solutions GmbH Note: Scatterplot and accuracy of heights derived from the generated DTM and field dGPS measurements (n = 93).

2.3.2

Development of the LiDAR AGB estimation model

Previous studies revealed that the Centroid Height (CH) is an appropriate height parameter of the LiDAR point cloud to estimate AGB in tropical forests taking also the point distribution over the different vegetation layers into account (Ballhorn et al. 2011; Jubanski et al. 2012; Kronseder et al. 2012). LiDAR height histograms were calculated by normalising all points within a grid of 35 m (similar to the size of the largest nest of the field inventory plots) to the ground using the DTM as reference. A height interval of 0.5 m was defined and the number of points within this interval was stored in the form of a histogram. The first (lowest) interval was considered as ground return and therefore excluded from further processing. The CH of the height histogram was calculated by weighing each 0.5 m height interval with the relative number of LiDAR points stored within this interval. CH was related to field inventory estimated AGB and regression models were developed. Jubanski et al. (2012) showed that the accuracy of AGB estimations derived from LiDAR height histograms increased with higher point densities. For this reason, point density was also implemented in the regression as a weighting factor. The commonly used power functions resulted in significant overestimations in the higher biomass range within our study area (Asner et al. 2012; Jubanski et al. 2012). For this reason, a more appropriate AGB regression model was developed, which is a combination of a power function (in the lower biomass range up to a certain threshold CH0) and a linear function (in the higher biomass range). The threshold of CH0 was determined by increasing the value of CH0 in steps of 0.001 m through identifying the lowest RMSE.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 6


The linear function is the tangent through CH0 and was calculated on the basis of the first derivative of the power function:

a ⋅ CH b AGB =  ( b −1) b  a ⋅ b ⋅ CH 0 (CH − CH 0 ) + a ⋅ CH 0

(

)

if CH ≤ CH 0 if CH > CH 0

(2)

Where CH is the Centroid Height, CH0 is the threshold of function change and a, b are coefficients. The developed AGB regression model was independently validated by the Predictive Power of the Regression (PPR) as carried out by Asner et al. (2010). The PPR is the RMSE determined by running 1000 iterations of the regression and randomly omitting 10% of reference field inventory plots. 2.3.3

Canopy Height Model

The Canopy Height Model (CHM) with a resolution of 1 m was produced by calculating the difference between the elevation of the digital surface model (DSM) and the underlying terrain of the DTM. For the DSM the highest point was chosen within a grid of 1 m. Pixels containing no data were filled using the highest point of the neighbourhood and the morphological operator ‘closing’ (Dougherty et al. 2003). 2.3.4

Analysis of AGB variability

The AGB variability was analysed within different forest types, recent and former logging areas, previously burned areas, and natural drainage areas. 2.3.4.1

Different forest types

The AGB variability of different forest types was investigated on the basis of a land cover classification 2009 of the Kapuas District (Siegert et al. 2013), which included forest classes ‘peat swamp forest’ (PSF) and ‘riparian forest’. The PSF class was then further classified into the following classes: low pole PSF (mean tree height < 15 m), mixed PSF (15 m < mean tree height < 19 m) and tall interior PSF (mean tree height > 19 m) (after Shepherd et al. 1997), which was done on the basis of the CHM. 2.3.4.2

Recent and former logging areas

Former logging railways are clearly visible in historical Landsat imagery (Figure 3) and have been manually digitised. The railways were categorised into the years in which they were first visible in the Landsat image: 1991, 1997 and 2000.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 7


Figure 3: Managed logged peat swamp forest

Source: Remote Sensing Solutions GmbH Note: Landsat scene from 30 June 1991 (RGB: bands 543) (left) and an abandoned logging railway (right) (S. EnglhartŠ).

Figure 4: Small logging trails in peat swamp forests

Source: Remote Sensing Solutions GmbH Note: Aerial (left) and ground (right) photo of logging trails from a canal into the forest (Š P. Navratil).

Recent logging trails and small canals are of smaller scale and therefore not consistently visible in the available optical satellite imagery (Figure 3). They can be identified solely based on the high-resolution LiDAR-derived DTM. Since height differences between skid trails and the surrounding terrain are similar to those of the undulating microtopography of peatlands, an automatic approach was rejected. In terms of a visual interpretation, an appropriate local contrast stretching is essential. On the basis of field surveys, orthophotos and long-term experience within the study area, an area with a buffer of 50 m around skid trails and small canals was chosen for the investigation of AGB variability in logged areas. It is assumed that loggers do not remove trees that are further away from the logging trail than 50 m because new trails would then be constructed. This approach was previously used by Englhart et al. (2013).

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 8


2.3.4.3

Burned areas

The determination of burned areas was based on MODIS and NOAA hotspots, historical Landsat imagery and recent RapidEye scenes (see Figure 5). In a first step MODIS and NOAA hotspots were used to identify years of fire occurrence in the study area (1990, 1997, 2001, 2002, 2004, 2005, 2006, 2007, 2008, 2009 and 2011). For the years 1990–2008 an automatic object-based classification of Landsat scenes was performed. Gaps due to clouds and image artefacts were classified manually and designated as unburned in cases of uncertainty. Recently burned areas (2009–2011) were identified by visual interpretation of RapidEye satellite images since the automatic approach adapted to Landsat images was not feasible due to the limited number of spectral bands in this imagery. The higher spatial resolution allows a more precise delineation of the borders of the burned areas. In addition, edges of automatically-classified former burned areas were refined based on the RapidEye images in areas where they were still clearly visible. The time period of the 2011 burned area analysis was limited by the date of the LiDAR flight survey. Figure 5: Fire events

Source: Remote Sensing Solutions GmbH Note: Location with one fire event (left); location with three fire events (right) (© S. Englhart and P. Navratil).

2.3.4.4

Natural drainage areas

Natural drainage areas are forested areas with a natural small river without any visual anthropogenic influence. These areas were visually identified on the basis of historical and recent satellite imagery.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 9


3 LIDAR AGB ESTIMATION MODEL Figure 6 (left) shows the scatterplot of AGB and CH with the corresponding regression. The part of the power function in the lower biomass range is depicted in blue, and the linear part of the function in the higher biomass range is depicted in red. The dot size of the reference field inventory plots represents LiDAR point density, the higher the density, the larger the dot. Figure 6 (right) depicts the dependence of CH0, the threshold of function change, and RMSE for the AGB regression model. The lower the value of CHo, the lower the proportion of the power function; and the higher the value of CH0, the lower the proportion of the linear function. The value with the lowest RMSE is marked and was chosen as CH0. The AGB regression model was developed on the basis of 55 field inventory measurements resulting in a coefficient of determination r2 = 0.83 with PPR = 42.72 t/ha and CH0 = 9.131 m. Figure 6: LiDAR AGB estimation model

Source: Remote Sensing Solutions GmbH Note: Scatterplot of field derived AGB and CH (left). The most appropriate AGB regression model is based on a combined power and linear function (Equation 2). Dependence of CH0 and RMSE (right). The CH0 value with lowest RMSE was chosen as the threshold for function change.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 10


4 EVALUATION OF AGB VARIABILITY AGB variability was analysed in different forest types (riparian forest, low pole PSF, mixed PSF, and tall interior PSF), in recent and former logging areas, in burned forests and in a natural drainage area. The spatial extent of these areas is shown in Figure 7. In order to avoid bias in the analysis possibly introduced by the differing spatial extent of the classes, a specified number of random points was distributed within each class of a category. The characteristics of the categories and the different classes are listed in Table 2. Table 2: Characteristics of the different AGB variability categories and its classes Category Forest type analysis

Logging areas (buffer 50 m)

Burned areas (number of fire events)

Burned areas (last fire event)

Natural drainage area

Classes

Area (ha)

Number of points

Riparian forest Low pole PSF Mixed PSF Tall interior PSF Recent 1991 1997 2000 1 2 3 4 5 6 7 8 1990 1997 2001 2002 2004 2005 2006 2007 2009 2011

4,630.42 8,319.18 42,650.52 6,337.33 5,915.29 1,467.08 4,253.52 356.72 125,565.46 80,055.54 38,918.46 8,926.38 1,632.92 1,632.92 40.07 0.72 816.02 34,144.95 8,824.73 44,955.47 42,739.43 36,700.61 4,5150.9 392.13 40,901.21 777.08 235.75

10,000 10,000 10,000 10,000 500 500 500 500 500 500 500 500 500 500 500 500 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 1,000 5,000

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 11


Figure 7: Overview of analysed areas for AGB variability

Source: Remote Sensing Solutions GmbH

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 12


5 RESULTS 5.1

AGB variability of different forest types

An overview of the different forest types and the investigated areas is provided in Figure 8. Arrows mark degraded forest areas. Due to this high degradation, some parts of the mixed PSF were wrongly classified as low pole PSF because of the reduced tree height and were therefore excluded from the analysis. Figure 8: Overview of the different forest types

Source: Remote Sensing Solutions GmbH Note: Forest type classification (left) and AGB estimation with the forest types outlines (right). Disturbed areas are marked with arrows and are analysed in detail in section 5.2.

The mean AGB and CHM values of the different forest types differed significantly from each other at the 95% confidence interval. All results are listed in Figure 9 and in Table 3. Riparian forests had the lowest mean AGB with 108.95 t/ha (with a 95% confidence interval of 107.5–110.41 t/ha) and the lowest canopy height of 9.94 m (with a 95% confidence interval of 9.79–10.09 m). These results emphasise the degradation state of these riparian forest. There is hardly any intact riparian forest left in the study area; the remaining riparian forests are a dense mosaic of remnant forest patches, agroforestry and agriculture. The low pole PSF had the lowest biomass among the peat swamp forest types with 176.93 t/ha (and a 95% confidence interval of 176.52–177.34 t/ha) and a mean CHM of 14.24 m (with a 95% confidence interval of 14.17–14.32 m). The mixed PSF had a mean AGB of 215.73 t/ha (with a 95% confidence interval Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 13


of 214.77–216.69 t/ha) and a mean CHM of 16.24 m (with a 95% confidence interval of 16.14–16.35 m). The tall interior PSF had the highest AGB with 297.73 t/ha (with a 95% confidence interval of 296.94– 298.53 t/ha) and also the highest CHM of 19.97 m (with a 95% confidence interval of 19.85–20.09 m). The standard deviation indicates that there was relatively low variation in the low pole PSF. The standard deviation in mixed and tall interior PSF was higher than in the low pole PSF and the highest standard deviation was found in riparian forests. Figure 9: Statistics of AGB and CHM variability for different forest types

Source: Remote Sensing Solutions GmbH Note: The 95% confidence interval is depicted.

Table 3: Statistics of AGB and CHM variability for different forest types AGB (t/ha) Forest Type

Mean

Low pole PSF Mixed PSF Tall interior PSF Riparian forest

176.93 215.73 297.73

Confidence interval for the mean 176.52 177.34 214.77 216.69 296.94 298.53

108.95

107.5

110.41

CHM (m) Standard deviation

Mean

20.84 48.81 40.91

14.24 16.24 19.97

Confidence interval for the mean 14.17 14.32 16.14 16.35 19.85 20.09

75.02

9.94

9.79

10.09

Standard deviation 3.64 5.34 6.17 7.62

Three randomly-selected locations within each forest type were chosen, and LiDAR height profiles (with vegetation heights as well as LiDAR height histograms) were created. These illustrate the structural differences between the different forest types and impact of degradation. The LiDAR height profiles and histograms are depicted for low pole PSF, mixed PSF, tall interior PSF and riparian forest in Figures 10–13, respectively. The relatively low canopy height and degradation state of low pole forests can be clearly seen. In contrast, mixed and tall interior forests show higher canopy heights as well as higher degradation impact. As already stated above, the high degradation stage of riparian forests in the study area resulting in low canopy heights; few emergent trees and the high number of ground returns are also obvious in the height profiles and histograms.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 14


Figure 10: LiDAR height and vegetation height profiles of low pole peat swamp forests (PSF) Low pole PSF

Source: Remote Sensing Solutions GmbH

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 15


Figure 11: LiDAR height and vegetation height profiles of mixed peat swamp forests (PSF) Mixed PSF

Source: Remote Sensing Solutions GmbH

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 16


Figure 12: LiDAR height and vegetation height profiles of tall interior peat swamp forests (PSF) Tall interior PSF

Source: Remote Sensing Solutions GmbH

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 17


Figure 13: LiDAR height and vegetation height profiles of riparian forests Riparian forest

Source: Remote Sensing Solutions GmbH

5.2

AGB variability of recent and former logging areas

The overview of all areas under consideration in the logging analyses is depicted in Figure 7. Figure 14 depicts an area of recent logging activities. The AGB estimation map clearly shows the reduction of AGB in these areas.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 18


Figure 14: Recent logging areas and LiDAR-derived AGB map

Source: Remote Sensing Solutions GmbH

The mean AGB and CHM values of the different logging categories differed significantly from each other in terms of the confidence interval, except the CHM values of the 1991 and 1997 logging which did not significantly differ from each other. All results are depicted in Figure 15 and in Table 4. AGB and CHM values of the different categories were higher the longer the logging event that had taken place prior to this. The 1991 logging buffer zone had a mean AGB of 233.42 t/ha (confidence interval: 229.37– 239.20 t/ha) and a mean CHM of 16.90 m (confidence interval: 16.35–17.41 m). The 1997 logging buffer zone resulted in a mean AGB of 218.55 t/ha (confidence interval: 218.21–229.14 t/ha) and a mean CHM of 16.31 (confidence interval: 16.37–17.45 m). The 2000 logging buffer zone had a mean AGB of 208.43 t/ha (confidence interval: 203.65–211.47 t/ha) and a mean CHM of 15.85 m (confidence interval: 15.69– 16.68 m). The recent logging areas had a mean AGB of 172.66 t/ha (confidence interval: 172.03– 80.25 t/ha) and a mean CHM of 14.37 m (confidence interval: 13.62–14.65 m). The standard deviation of AGB and CHM values showed that there was high variation in all logging categories.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 19


Figure 15: Statistics of AGB and CHM variability for former and recent logging areas

Source: Remote Sensing Solutions GmbH Note: The 95% confidence interval is depicted.

Table 4: Statistics of AGB and CHM variability for former and recent logging areas AGB (t/ha) Logging Type

Mean

Railways 1991 Railways 1997 Railways 2000 Recent logging

234.29 223.67 207.56 176.14

Confidence interval for the mean 229.37 239.20 218.21 229.14 203.65 211.47 172.03 180.25

CHM (m) Standard deviation

Mean

54.71 62.43 41.67 49.60

16.88 16.91 16.19 14.14

Confidence interval for the mean 16.35 17.41 16.37 17.45 15.69 16.68 13.62 14.65

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Standard deviation 5.64 6.00 5.46 5.78

Page 20


5.3

AGB variability of burned areas

Figure 16: Overview of burned areas

Source: Remote Sensing Solutions GmbH Note: Number of fire events (left), year of the last fire (middle) and the respective AGB estimation map (right).

Burned areas were analysed for the number of fire events and for the last year of fire occurrence. Categorising the burned areas in number of fire events showed a clear trend. The more frequently areas burned, the lower the remaining biomass and vegetation height; although there was little variation between the third and fourth fire event, and after the fifth fire event. The standard deviations were relatively high compared to the mean values, which indicate a very high variability within single classes. The detailed results are shown in Figure 17 and in Table 5. Figure 17: Statistics of AGB and CHM variability for number of fire events

Source: Remote Sensing Solutions GmbH Note: The 95% confidence interval is depicted. Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 21


Table 5: Statistics of AGB and CHM variability for number of fire events. AGB (t/ha) Frequency of fires

Mean

1 2 3 4 5 6 7 8

16.42 7.87 3.15 2.35 1.47 1.14 1.74 0.66

Confidence interval for the mean 13.88 18.96 6.83 8.91 2.69 3.61 1.95 2.75 1.12 1.81 0.92 1.35 1.53 1.95 0.65 0.67

CHM (m) Standard deviation

Mean

30.23 12.73 6.81 4.33 3.64 2.58 1.93 0.12

2.92 1.85 1.14 0.76 0.68 0.60 0.41 0.25

Confidence interval for the mean 2.66 3.38 1.77 2.25 0.90 1.20 0.75 1.01 0.58 0.73 0.52 0.67 0.41 0.78 0.22 0.29

Standard deviation 3.89 2.72 1.86 1.32 1.06 0.92 1.57 0.40

The second categorisation was by the time of the last fire event. There was no clear trend identifiable and the standard deviations varied extremely across each category. The detailed results are depicted in Figure 18 and in Table 6. Figure 18: Statistics of AGB and CHM variability for the last fire events

Source: Remote Sensing Solutions GmbH Note: The 95% confidence interval is depicted.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 22


Table 6: Statistics of AGB and CHM variability for the last fire events AGB (t/ha) Last fire event

Mean

1990 1997 2001 2002 2004 2005 2006 2007 2009 2011

20.72 23.40 6.63 14.00 6.56 10.10 7.41 2.88 8.26 24.12

5.4

Confidence interval for the mean 19.17 22.27 20.49 26.30 5.91 7.35 12.88 15.13 5.86 7.26 9.20 11.00 6.58 8.25 2.34 3.42 7.11 9.42 21.41 26.84

CHM (m) Standard deviation

Mean

29.38 46.32 13.00 19.43 11.48 14.27 14.72 8.88 17.69 43.40

3.56 3.14 1.48 3.16 1.67 2.34 1.73 1.28 1.70 3.25

Confidence interval for the mean 3.31 3.82 2.86 3.41 1.33 1.62 2.94 3.38 1.51 1.84 2.15 2.52 1.57 1.90 1.18 1.39 1.49 1.91 2.93 3.57

Standard deviation 4.18 4.42 2.39 3.48 2.61 2.99 2.76 1.82 3.16 4.94

AGB variability of natural drainage areas

Figure 19: Overview of the natural drainage area

Source: Remote Sensing Solutions GmbH Note: Left: RapidEye image from 29 July 2012 (bands 453). Right: AGB estimation map.

Natural drainage areas resulted in a mean AGB of 168.41 t/ha (with a 95% confidence interval of 168.41â&#x20AC;&#x201C; 169.28 t/ha and a standard deviation of 30.55 t/ha) and a mean CHM of 13.84 m (with a 95% confidence interval of 13.70â&#x20AC;&#x201C;13.97 m and a standard deviation of 4.96 m). All results are shown in Figures 19 and 20 and in Table 7. It can clearly be seen that the mean AGB and CHM values of the natural drainage area is less than that for the surrounding mixed PSF. This fact emphasises the degradation due to facilitated access to the forests.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 23


Figure 20: Statistics of AGB and CHM variability for a natural drainage area

Source: Remote Sensing Solutions GmbH Note: The 95% confidence interval is depicted. The surrounding area is represented by mixed peat swamp forests (PSF).

Table 7: Statistics of AGB and CHM variability for a natural drainage area AGB (t/ha) Forest type

Mean

Natural drainage area

168.41

Confidence interval for the mean 167.54 169.28

CHM (m) Standard deviation

Mean

30.55

13.84

Confidence Standard interval for the deviation mean 13.70 13.97 4.96

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 24


5.5

AGB variability across the peat dome

Figure 21: Transect across the peat dome

Source: Remote Sensing Solutions GmbH Note: RapidEye image from 29 July 2012 (bands 452) (upper panel), AGB map (middle panel), transect of AGB values and DTM heights (lower panel).

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 25


Figure 21 depicts a transect across the peat dome showing AGB values and DTM heights. AGB values are very high at the beginning of the transect, as there is a tall interior PSF which contains high AGB values. In the middle of the peat dome, AGB is comparably low and slightly increases at the end of the transect. As shown in Figure 8, a low pole PSF is located in the middle of the transect, which explains the low AGB values. At the end of the transect, there is mixed PSF that was historically logged. According to Shepherd et al. (1997) there are no significant differences in chemical inputs to the peatland system away from the riverine zone where some nutrient influx may occur as a result of wet season river flooding. Adjacent to the beginning and the end of the transect are rivers that may be responsible for some nutrient supply, which in turn facilitates tree growth. Furthermore, Shepherd et al. (1997) stated that the main difference between the different PSF forest types is peat depth and the hydrological condition of the peat on which they are growing. In the tall interior forest, the water table drops to about 150 cm below the surface during the dry season. Therefore, the peat within this zone becomes aerated and organic matter decomposition is promoted, which supplies nutrients to the overlying vegetation. In contrast, a high water table is maintained in low pole PSF throughout the year, which prevents further nutrient supply (Shepherd et al. 1997).

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 26


6 DISCUSSION 6.1

Evaluation of the results

The four different forest types showed a significant difference between mean AGB and CHM values: •

The riparian forest had lower AGB and CHM values than expected. Firstly, the riparian forest is highly degraded and logged as a result of felling and burning and is replaced by a low growing vegetation (Shepherd et al. 1997). This is also evidenced by the low mean CHM value of 9.94 m, in contrast to the stated mean canopy height in the literature of between 25–35 m (Shepherd et al. 1997). Secondly, the classification of the riparian forest was based on Landsat imagery with a spatial resolution of 30 m. Misclassification as well as the difference of the spatial resolution (spatial resolution of AGB: 5 m, spatial resolution of CHM: 1 m) might result in the inclusion of a variety of land covers, including water and low vegetation areas (with a AGB and CHM value of around zero) in the riparian forest class. These facts explain the low AGB and CHM mean values as well as the high standard deviations, which indicate a very high variation of AGB and CHM values within this forest type. The low pole PSF, as expected, contains little AGB with a lower standard deviation and has the lowest canopy height of the PSF classes. The low standard deviation indicates that there is little variation within this class. This forest type is not prone to logging as it contains very few trees of commercial importance and few of these provide good quality timer (Shepherd et al. 1997). The mixed PSF class has intermediate AGB and CHM values within the PSF classes and with a high variability. The reasons for the high variability are various. As this forest type contains several commercial tree species, the high standard deviation is due to logging and the resulting degradation or because of a high diversity of species. Tall interior PSF contains a greater number of commercial tree species than the other forest types and is subject to the most intensive logging of any forest type apart from the mixed PSF (Shepherd et al. 1997). Due to the tall tree species, this forest type had the highest AGB and CHM values with high standard deviations as a result of logging activities.

The analysis of former and recently logged areas revealed that AGB and CHM values are higher the longer the logging event that has taken place prior to this. This indicates that these areas recover from the degradation by logging. The high standard deviation of the AGB and CHM values emphasises the high variability in each logging class, which may be due to different logging intensities and repeated logging as a result of accessibility to the forests via abandoned logging railways. The mean AGB (172.66 t/ha) and CHM (14.37 m) values are very similar to those retrieved by Englhart et al. (2013) (167 t/ha and 14.2 m, respectively). Although, both analyses are based on the same dataset, Englhart et al. (2013) analysed a very small area (113 ha) compared to this study (over 5900 ha), and used a different location in the study area. The analysis of vegetation in burned areas revealed that there is a relationship between AGB and CHM values and the number of fire events. The more often an area has burned, the lower the AGB and CHM mean values. There is no significant difference in AGB between areas that experienced three and four fire events, as well as between areas with five or more fire events. There was no discernable correlation between the time of the last fire event and mean AGB or CHM values; most likely because there was no Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 27


differentiation between the number of fire events. Future analysis should take the number of fire events into account when analysing the last fire event (e.g. investigating areas that burned only once categorised by the last fire event). It is expected that mean AGB is higher the longer the period of regeneration. Regarding the burn scar analysis, it should also be taken into account that not all fires have been detected due to lack of data (MODIS hotspots and Landsat imagery) between 1991 and 2001. Natural drainage areas showed a relatively low AGB and CHM in comparison to the different forest types. This could be a result of facilitated accessibility to the forests by boat, which makes logging activities easier. 6.2

Uncertainties

With regard to REDD+, it is important to take a closer look at the accuracy of AGB estimations. First, errors associated with field inventory work, such as measurements errors, or GPS accuracy may influence the AGB reference data. Additionally, allometric equations that are used to extrapolate field measurements to AGB values introduce a bias in AGB estimations, especially when using a generalised allometric model, as in this study (Chave et al. 2005). But the advantage of this allometric model is that it provides the most accurate AGB estimates in Indonesia compared to other regional and general allometric models (Rutishauser et al. 2013). This was tested for Dipterocarp forests and it is most likely that other forest types (such as kerangas forests, PSF, forests on limestone) would have given different results. But as Dipterocarp forest is the dominant vegetation type in Borneo, using the allometric equation of Chave et al. (2005) ensures a consistent method for large-scale applications. Another potential source of error is the time difference between field inventories and LiDAR acquisitions. These errors are hard to quantify in our analysis and were kept as low as possible by using only field inventory data of the years 2010 and 2011. Additionally, errors of the AGB regression model also influence the AGB estimation and the variability analysis. 6.3

Strengths and limitations for AGB and variability analyses

Implementation of REDD+ and other environmental projects require accurate carbon or AGB estimations. Airborne LiDAR measurements have the ability to provide very accurate AGB estimations and are therefore very suitable for large-scale aboveground carbon stock assessment. Relatively high operational costs for large-scale mapping and high data storage capacities limit airborne LiDAR applications. Conventional AGB estimation approaches, such as the stratify and multiply approach, do not capture the local variability in forest AGB (natural- and human-caused) and can lead to high uncertainties in the upscaling process due to comparatively low sample numbers. Both the consideration of variability and reduction of uncertainty are required for REDD+ measurement, reporting and verification (MRV) systems. Global estimates of biomass carbon stock for the tropics, based on a combination of in situ inventory plots, satellite LiDAR data and optical and microwave imagery (1 km resolution), showed that, especially for the peat forest areas of Central Kalimantan, the uncertainty in biomass carbon stock estimates was very high (> 45%) (Saatchi et al. 2011). This high uncertainty, the limited amount of in situ field measurements in Southeast Asian tropical forests (especially in tropical peat swamp forests), and the inability of high and medium multispectral resolution satellite instruments such as Landsat to quantify historic forest disturbance show the importance to derive more accurate AGB estimates in these inaccessible ecosystems.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 28


In combination with high-resolution satellite imagery, airborne LiDAR could be a cost-effective approach to derive more accurate regional maps on forest carbon densities (Asner et al. 2010). Furthermore, AGB estimations based on field inventory data and LiDAR metrics (as in this study) have the capability to improve current estimates on AGB spatial variability across different forest types and degradation levels also in other tropical biomes and to assist the efforts in upscaling LiDAR-derived AGB estimates to largescale geographic areas with low uncertainty. Englhart et al. (2013) showed that it is also possible to estimate AGB recovery rates based on repeated airborne LiDAR measurements. LiDAR surveys of 2007 and 2011 of this study area were evaluated for their ability to estimate AGB changes. Due to different acquisition parameters of the LiDAR data (such as point density), two independent regression models had to be developed in order to estimate AGB on the basis of CH of the LiDAR point cloud taking also the point density into account. Based on the AGB estimations for each year, AGB change over the four-year time period was then calculated. In unaffected peat swamp forests, the mean canopy height increased on average 2.3 m and accumulated 5 t/ha y−1. These unaffected forest areas included primary and secondary forests of different degradation levels as well as different peat swamp forest types. The growth and accumulation rates are an average of fast growth rates, for example, in previously-logged forests and slow growth rates and in primary forests. Sweda et al. (2012) and Boehm et al. (2012) found a canopy growth of 1.9 m and an AGB accumulation of 3.7 t ha−1 y−1.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 29


7 CONCLUSIONS AND RECOMMENDATIONS This study demonstrates the great potential of airborne LiDAR measurements to estimate AGB and its variability in carbon-rich tropical peat swamp forests in different degradation stages and forest types. In contrast to the commonly used stratify and multiply approach to derive forest AGB on the basis of land cover classifications, the current study emphasises the importance to consider the high variability of the forests. As this method generates a huge amount of AGB estimations for each land cover and forest type class, the estimation error is reduced and the standard deviation solely depicts the AGB variability within each class. An effective national carbon stock monitoring would therefore include a transect-based airborne LiDAR sampling over the national forest inventory plots, which is then upscaled by a detailed land cover and forest type classification map derived from, for example, high-to-medium resolution multispectral satellite imagery. This approach benefits from LiDAR-derived AGB estimations of focused areas and is therefore very cost efficient for a large area upscaling.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 30


REFERENCES Asner, G.P., G.V.N. Powell, J. Mascaro, D.E. Knapp, J.K. Clark, J. Jacobson, T. Kennedy-Bowdoin, A. Balaji, G. Paez-Acosta, E. Victoria, et al. 2010. High-resolution forest carbon stocks and emissions in the Amazon. Proceedings of the National Academy of Sciences of the United States of America 107:16738–16742. Asner, G.P., J. Mascaro, H.C. Muller-Landau, G. Vieilledent, R. Vaudry, M. Rasamoelina, J.S. Hall and M. van Breugel. 2012. A universal airborne LiDAR approach for tropical forest carbon mapping. Oecologia 168:1147–1160. Baccini, A., S. Goetz, W. Walker, N. Laporte, M. Sun, D. Sulla-Menashe, J. Hackler, P. Beck, R. Dubayah, M. Friedl, et al. 2012. Estimated carbon dioxide emissions from tropical deforestation improved by carbondensity maps. Nature Climate Change 2:182–185. Ballhorn, U., J. Jubanski and F. Siegert. 2011. ICESat/GLAS Data as a measurement tool for peatland topography and peat swamp forest biomass in Kalimantan, Indonesia. Remote Sensing 3:1957–1982. Boehm, V., V. Liesenberg, T. Sweda, H. Tsuzuki and S.H. Limin. 2012. Multi-temporal Airborne LiDARSurveys in 2007 and 2011 over Tropical Peat Swamp Forest Environments in Central Kalimantan, Indonesia. In: Proceedings of the 14th International Peat Congress, Stockholm, Sweden, 3–8 June 2012. Brown, S. 1997. Estimation biomass and biomass change of tropical forests: A primer. FAO Forestry Paper 134:1–55. Campbell, B.M. 2009. Beyond Copenhagen: REDD plus, agriculture, adaptation strategies and poverty. Global Environmental Change 19:397–399. Chave, J., C. Andalo, S. Brown, M.A. Cairns, J.Q. Chambers, D. Eamus, H. Folster, F. Fromard, N. Higuchi, T. Kira, et al. 2005. Tree allometry and improved estimation of carbon stocks and balance in tropical forests. Oecologia 145, 87–99. Chudnoff, M. 1984. Tropical timbers of the world. Agriculture Handbook 607. US Department of Agriculture, Forest Service, Forest Product Laboratory, Madison, WI, USA. Dossa, G. G. O., E. Paudel, J. Fujinuma, H. Yu, W. Chutipong, Y. Zhang, and R.D. Harrison. 2013. Factors determining forest diversity and biomass on a tropical volcano, Mt. Rinjani, Lombok, Indonesia. PloS One 8(7):e67720. doi:10.1371/journal.pone.0067720. Dougherty, E.R. and R.A. Lotufo. 2003. Hands-On Morphological Image Processing. Spie Bellingham, Bellingham, Washington, USA. Duncanson, L., K. Niemann and M. Wulder. 2010. Estimating forest canopy height and terrain relief from GLAS waveform metrics. Remote Sensing of Environment 114:138–154. Englhart, S., J. Jubanski and F. Siegert. 2013. Quantifying dynamics in tropical peat swamp forest biomass with multi-temporal LiDAR datasets. Remote Sensing 5:2368–2388. FAO. 2006. Global forest resources assessment 2005. FAO Forestry Paper 147. Food and Agriculture Organization of the United Nations, Rome, Italy. FAO. 2009. State of the world’s forests 2009. Food and Agriculture Organization of the United Nations, Rome, Italy.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 31


Gibbs, H.K., S. Brown, J.O. Niles and J.A. Foley. 2007. Monitoring and estimating tropical forest carbon stocks: Making REDD a reality. Environmental Research Letters 2:1–13. Gleason, C.J. and J. Im. 2011. A review of remote sensing of forest biomass and biofuel: Options for smallarea applications. GISci. Remote Sensing 48:141–170. Goetz, S. and R. Dubayah. 2011. Advances in remote sensing technology and implications for measuring and monitoring forest carbon stocks and change. Carbon Management 2, 231–244. Griscom, B., P. Ellis, and F.E. Putz. 2014. Carbon emissions performance of commercial logging in East Kalimantan, Indonesia. Global Change Biology 20(3): 923–937. Hughes, R.F., J.B. Kauffman and V.J. Jaramillo. 1999. Biomass, carbon, and nutrient dynamics of secondary forests in a humid tropical region of Mexico. Ecology 80:1892–1907. IPCC. 2006. 2006 IPCC guidelines for national greenhouse gas inventories, Prepared by the National Greenhouse Gas Inventories Programme. Eggleston, H.S., K. Buendia, Miwa, T. Ngara and K. Tanabe (eds.). Institute for Global Environmental Strategies (IGES), Kamiyamaguchi, Japan. Jaenicke, J., J.O. Rieley, C. Mott, P. Kimman and F. Siegert. 2008. Determination of the amount of carbon stored in Indonesian peatlands. Geoderma 147:151–158. Jubanski, J., U. Ballhorn, K. Kronseder, J. Franke and F. Siegert. 2013. Detection of large above ground biomass variability in lowland forest ecosystems by airborne LiDAR. Biogeosciences 10:3917–3930. Jung, J. and M.M. Crawford. 2012. Extraction of features from LIDAR waveform data for characterizing forest structure. IEEE Geosci. Remote Sensing Letters 9:492–496. Kronseder, K., U. Ballhorn, V. Böhm and F. Siegert. 2012. Aboveground biomass estimation across forest types at different degradation levels in Central Kalimantan using LiDAR data. International Journal of Applied Earth Observation and Geoinformation 18:37–48. Mascaro, J., M. Detto, G.P. Asner and H.C. Muller-Landau. 2011. Evaluating uncertainty in mapping forest carbon with airborne LiDAR. Remote Sensing of Environment 115:3770–3774. Meyer, V., S.S. Saatchi, J. Chave, J. Dalling, S. Bohlman, G.A. Fricker, C. Robinson and M. Neumann. 2013. Detecting tropical forest biomass dynamics from repeated airborne LiDAR measurements. Biogeosciences Discussions 10:1957–1992. Miettinen, J. and S.C. Liew. 2012. Two decades of destruction in Southeast Asia’s peat swamp forests. Frontiers In Ecology and the Environment 10:124–128. Page, S.E., A. Hoscilo, A. Langner, K. Tansey, F. Siegert, S. Limin and J.O. Rieley. 2009. Chapter 9: Tropical peatland fires in Southeast Asia. In: Cochrane, M. A. (ed.). Tropical fire ecology: Climate change, land use, and ecosystems dynamics. Springer-Praxis: Berlin, Germany.. Page, S.E., J.O. Rieley and C.J. Banks. 2011. Global and regional importance of the tropical peatland carbon pool. Global Change Biology 17:798–818. Pearson, T., S. Walker and S. Brown. 2005. Sourcebook for land use, land-use change and forestry projects. Winrock International, Little Rock, AR, USA. Pfeifer, N., P. Stadler and C. Briese. 2001. Derivation of digital terrain models in the SCOP++ environment. Proceedings of OEEPE workshop on ‘Airborne Laserscanning and Interferometric SAR for Detailed Digital Terrain Models’. Stockholm, Sweden, March 2001.

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 32


Rutishauser, E., F. Noor’an, Y. Laumonier, J. Halperin, K. Hergoualc’h, K. and L. Verchot. 2013. Generic allometric models including height best estimate forest biomass and carbon stocks in Indonesia. Forest Ecology and Management 307:219–225. Saatchi, S. S., N.L. Harris, S. Brown, M. Lefsky, M., E.T.A. Mitchard, W. Salas, B.R. Zutta, W. Buermann, S.L. Lewis, S. Hagan, S. Petrova, L. White, M. Silman, and A. Morel. 2011. Benchmark map of forest carbon stocks in tropical regions across three continents. Proceedings of the National Academy of Sciences of the United States of America 108(24):9899–904. Shepherd P.A., J.O. Rieley, and S.E. Page. 1997. The relationship between forest vegetation and peat characteristics in the upper catchment of Sungai Sebangau, Central Kalimantan. In: Rieley J.O. and S.E. Page (eds.) Biodiversity and sustainability of tropical peatlands, 191–207. Samara Publishing Limited, Cardigan, United Kingdom. Siegert, F., P. Navratil, J. Franke and K. Kronseder. 2013. Historical land cover classification and land cover change in the Kalimantan Forests and Climate Partnership (KFCP) site and the Kapuas district. IAFCP Technical Scientific Working Paper. Indonesia-Australian Forest Carbon Partnership (IAFCP), Jakarta, Indonesia. Steininger, M. K. 2000. Secondary forest structure and biomass following short and extended land-use in central and southern Amazonia. Journal of Tropical Ecology 16:689–708. Surtech. 2011. Report on topographic mapping by airborne laser scanning August to October 2011. Document Number G033 REP 02. Mimeo. Indonesia-Australian Forest Carbon Partnership (IAFCP), Jakarta, Indonesia. Sweda, T., H. Tsuzuki, Y. Maeda, V. Boehm and S.H. Limin. 2012. Above- and below-ground carbon budget of degraded tropical peatland revealed by multi-temporal airborne laser altimetry. Proceedings of the 14th International Peat Congress, Stockholm, Sweden, 3–8 June 2012. van der Werf, G.R., D.C. Morton, R.S. DeFries, J.G.J. Olivier, P.S. Kasibhatla, R.B. Jackson, G.J. Collatz and J.T. Randerson. 2009. CO2 emissions from forest loss. Nature Geoscience 2:737–738. Vincent, G., D. Sabatier, L. Blanc, J. Chave, E. Weissenbacher, R. Pélissier, E. Fonty, J.F. Molino and P. Couteron. 2012. Accuracy of small footprint airborne LiDAR in its predictions of tropical moist forest stand structure. Remote Sensing of Environment 125:23–33. World Agroforestry Centre 2011. Wood density database. Available online: http://www.worldagroforestrycentre.org/Sea/Products/AFDbases/WD/Index.htm (accessed on 1 March 2012).

Assessment of biomass variability of different tropical forest types and degradation levels by large-scale airborne LiDAR measurements in Central Kalimantan, Indonesia.

Page 33


S3 2 agb working paper 2014 02 17 final layout