remote sensing
Article
MappingChina’sGhostCitiesthroughthe CombinationofNighttimeSatelliteDataand DaytimeSatelliteData
HeliLu 1,* ID ,ChuanrongZhang 2 ID ,GuifangLiu 1,XinyueYe 3 andChanghongMiao 1
1 KeyLaboratoryofGeospatialTechnologyforMiddleandLowerYellowRiverRegionsofMinistryof Education&CollaborativeInnovationCenteronYellowRiverCivilizationofHenanProvince/KeyResearch InstituteofYellowRiverCivilizationandSustainableDevelopment&CollaborativeInnovationCenterof Urban-RuralCoordinatedDevelopmentofHenanProvince,HenanUniversity,Kaifeng475004,China; kf_guif@163.com(G.L.);chhmiao@henu.edu.cn(C.M.)
2 DepartmentofGeography&CenterforEnvironmentalSciencesandEngineering,UniversityofConnecticut, Storrs,CT06269-4148,USA;chuanrong.zhang@uconn.edu
3 DepartmentofGeographyandComputationalSocialScienceLab,KentStateUniversity, Kent,OH44242,USA;xye5@kent.edu
* Correspondence:hk_lhl@163.com;Tel.:+86-0371-2388-1850
Received:24May2018;Accepted:29June2018;Published:1July2018
Abstract: Oneoftheside-effectsgeneratedbymainlandChina’surbanizationprocessis “ghostcities”—generallydefinedasclustersofabandonedbuildingsorhousingstructures—butthere isanotablelackofstudiesonthebasiccharacteristicsrelatedtothisphenomenon,suchassize,growth, level,distribution,scale,intensity,patternanddeterminants.Throughacombinationofnighttime satellitedataanddaytimesatellitedataasausefulproxy,inthispaper,wepresentthespatialpattern andtemporalevolutionofChina’sghostcitiesoverthelasttwodecades.Nighttimelight’srateof changeinnewlybuiltareasisdevelopedbasedonDMSP/OLSandNormalizedDifferenceBuilt-up Indextoassessacity’sdarkness.Resultsshowthattheghostcityproblemisreal,but,atleastso far,confinedto22smallercities.However,furtheranalysisrevealsthatnighttimelightschange innewlybuiltareas,followinganinvertedU-curveforbigcitiesrepresentingareversionfrom positivetonegativevaluesforthetrendsinrecentyears.Themethodologythroughtheuseofthe complementarycharacteristicsintimebetweenDMSP/OLSandLandsatdatainourstudyproveto serveasdeposingthedirectevidencestoascertainandquantifysuchsocial-economicphenomenon.
Keywords: nighttimesatellitedata;daytimesatellitedata;ghostcities;China;urbanization
1.Introduction
UrbanizationinChinaisofparticularinterestbothbecauseofthelargenumbersinvolvedand becauseofthepaceofitsprogressintheurbanpopulation,urbansettlementsandurbanarea.In1978, about195millionChineselivedincities;todaythatnumberapproaches700million,withthenumber ofcitieshavingincreasedfrom223toover660[1–3].Althoughthisisarelativelylownumberin comparisonwiththeaverageEuropeanurbanpopulationofroughly70to80percentandmore than80percentintheUnitedStates,itisworthnotingthatChina’srateofurbanizationtooka significantlyshortamountoftime(1978–2012)torisefrom20percenttoabove60percent.Since the2000s,China’surbanizationhasbeenkickedintohighgearunderthe“SuggestionsforMaking the11th‘Five-YearPlan”whichrecognizedthaturbanizationwouldbeanimportantcontributing factorinthedevelopmentofabalancedeconomyinChina.Theannualaveragegrowthratehas exceeded3percentand30millionmorepeopleperyearentertheurbanpopulation.Theyear2011
markedthefirstyearthatChina’surbanpopulation(690million)wasgreaterthanitsruralpopulation (656million)[4].AnewstudyfromtheMcKinseyGlobalInstitute[5]projectsthatby2025China’s citieswillhave325millionmorepeople,includingabout230millionmigrants.Followingthecurrent trend,thecountry’surbanpopulationislikelytoreach926millionby2025andtop1billionby2030. Inthisscenario,Chinawillhave221citieswithpopulationsofoveronemillionbytheyear2025.
However,China’surbanizationisnotasimpleprocessofanincreasinglymobilepopulation movingtourbanareasbutacomplexrural-to-urbantransformationthatrequiresco-development withindustriesandtheentireeconomicsystemaswellascompatibilitywiththeavailabilityofsocial provisionsincludinghealthcare,education,socialprotection,employmentandinfrastructure[6,7]. Infact,Chinaiscurrentlyconfrontedbyahostofsocialandeconomicpolicychallengesthataccompany suchrapid,large-scaleurbanizationandthe“side-effects”generatedinthatprocess.Itconsiders anumberoftheseconcerns[8]suchasovercrowdedlivingconditions,povertyandinequality, environmentaldamage/degradation,deterioratinghealth,ecologicaldestruction,deteriorating infrastructures,intensivepressuresonemployment,socialsecurity,bubblesintherealestateindustry andghostcities.
Asoneoftheworld’seconomicpowerhouses,China’sghostcityphenomenonhasattracted attentionfromtheinternationalacademiccommunity.SoraceandHurst[9]arguethattheexistenceof ghostcitiesexposesdeeperpatternsofurbanizationpropelledbypoliticalimperativesandaesthetic norms,whichfollowlogicsdifferentfromthoseofpopulationpressuresormarketrationalities. Woodworth,M.D.andWallace,J.L.[10]triedtoexaminekeyquestionssuchaswhetherornotghost citiesaretemporaryanomaliesorstructuralfeaturesofChina’surban-ledeconomicgrowthmodel. HongYu[11],ontheotherhand,indicatesthattheemergenceofghostcitiesisrootedinlocalGDPism (worshipofgrossdomesticproductgrowth).Somelocalscholars[12]believethatlackofinvestment channelsandthesegmentationbetweenhousingrentalmarketsandhousesalesmarketresultinthe formationofghostcities.
Someresearcherstrytousestatisticaldatatodeterminetheexistenceofspecificghostcities. Basedonasurveyof609constructionprojectsin12Chinesecities,CreditLyonnaisSecurities Asia(CLSA),Asia-PacificMarketspresentedananalysisonChina’sghosttownsin2014[13]. ItfoundthattheaveragevacancyrateinChinaforpropertycompletedbetween2009and2014 was15percent—equivalentto10.2millionemptyunits—whichismorethanthe10percentrecorded intheU.S.andthe8–9percentrecordedinHongKongbetween1997and1998beforetheproperty bubbleburst.InFebruary2015,aHongKong-based SouthChinaMorningPost compiledaghostcity mapbasedonanexperimentalmodelusingtwoindicators:futuresupply-and-demandratioand measuringoversupplyorundersupplyagainstexistinghomesinonecity[14].
Morerecently,someresearchershavebeguntostudyChina’sghostcityusingnewdatasources basedonremotesensing.Usingpixel-based,time-seriesnight-timelighttrajectories,Yangetal.[15] foundseveralclustersatprefecture,cityandcountylevelswithhighproportionoftheurbanization acceleration.Inthelightofbig/opendata,Jinetal.[16]utilizedthenational-wideandmillion magnituderoadjunctions,pointsofinterestandlocationbasedservicerecordsof2014/2015for measuringthemorphological,functionalandsocialvitalityofresidentialprojects.Usingnighttime lightimagery,landcovertypeproductsandpopulationgrid,inYangtzeRiverDeltaZheng’s finding[17]impliedthat“ghostcities”wereprominentlyspatiallyclustered.Zhengetal.[18] establisheda“ghostcity”indextoquantifytheintensityofthephenomenoninthenortheastofChina.
Therewasprogressintheseresearches,whileitstilllackedconsistent,reliableandnationwide datarelatedtoghostcitiessofar.Forexample,informationisscarceonghostcities’size,level,pattern anddistribution.Furthermore,itremainsunclearhowghostcities’growth,scaleandintensitychanges orevolvesovertime.Thislackofreliabledataclearlyhindersstudiesonthetopic.Inthispaper, theresearchersemploynighttimeanddaytimesatellitedataasausefulproxyinanefforttoovercome thelackofreliabledatafromothersources.Bytracingspatialpatternsandtemporalevolutionduring thelasttwodecadesthroughtheuseofthecomplementarycharacteristicsintimebetweenDMSP/OLS
andLandsatdata,thepaperisabletosketchanoutlineofghostcitieslandscapesandthemajor determinantsintheirdevelopment.
2.MaterialsandMethods
AccordingtothetraditionaldefinitionoperativeinWesterncountries,aghostcityisatown orcitythathasbeenabandoned,whetherduetonature-inducedorhuman-induceddisasters[19]. GhosttownsdottedthewesternpartoftheUnitedStateswhereaneconomicorindustrialboomwas followedbyaperiodofdecline,suchasminingtowns,milltowns,ordryoil-welltowns.However, thispatterndoesnotholdtruefortheghosttownsofMainlandChina,whichisstillexperiencing rapideconomicandsocialdevelopmentatarapidpace.WadeShepard[20]emphasizesthat,unlike Westernghostcities,whicharetypicallytheresultofeconomic,natural,orsocialhardship,Chinese ghostcitiesreflectnewurbandevelopmentprojectsinruralareaornewcitiesbuiltwithinexisting cities.ThesignificanceofthisdistinctionisthatChina’sghostcitieshavenotyetcometolife,whereas thetraditionalghostcityhasbecomeeconomicallydefunctafteraperiodofprosperity.Assuch, theterminourstudyindicatescityareaswhichhavealreadybeenconstructedbutaresignificantly under-inhabitedowingtoconstraintsimposedbyurbanmanagementorothersocio-economicfactors. Intheseareas,promisingplanswerelaidandmodernizedbuildingswereerectedbuttodaytheystand empty,orverynearly.Asaresult,inthedaytimeitshowsverylittlehumanactivityandatnight,these areasaredark.
ToanalyzeChina’suniqueghostcities,twoparametersshouldbeaddressed:theconstruction yearsofthebuildingsandtheresidentialactivitiesinthebuildingsaftertheconstructionyear. LandsatdaytimeimageswerechosentodeterminetheconstructionyearofthebuildingsandDefense MeteorologicalSatelliteProgramOperationalLinescanSystem(DMSP/OLS)nighttimeimageswere chosentomeasureresidentialactivities.Theartificiallightingofbuildingsisavaluableindicatorfor ghostcitessinceitdirectlyreflectsresidentialactivity.Sustaineddeclineofartificialnighttimelight afterbuildingconstructionindicatesthataconstructedareahasbeendeserted—leftwithvacantor unfinishedbuildings;otherwisebuildingsarelargelypopulatedandutilizedastheirpopulations growandnighttimeimageryrevealsincreasingartificiallightuse.Thismethodologywasapplied totwogroups:GroupAincludes25smallercitiesandGroupB,15bigcities(including“thebig two”—BeijingandShanghai—twoTier1citiesand11Tier2cities)[21,22].
2.1.Data
2.1.1.NighttimeLightsData
ImagesofnighttimelightscollectedbytheUSAirForceDMSP/OLS,whichisknownforits remarkableabilitytodetecthumanactivity[23],wereusedinthestudy.Manystudieshaveusedthese datasetsfordeterminingthespatiotemporaldimensionsofsocio-economicfactorsincludingGDP, urbansprawl,impervioussurfacesandex-urbandevelopment.Morerecently,thesedatasetshave beenusedasproxiesforestimatingCO2 emissions[24].
Thelow-orbitingsatelliteofDMSP/OLSusesvisible/nearinfraredwavebands(0.4–1.1 µm) fordetectinglightsandthermalinfraredbands(10.5–12.6 µm)tofiltercloudcover.Thesatellite typicallypassesoverastudyareabetween8:30and9:30p.m.localtimeandannualglobal compositesoftemporallystablenighttimelightshavebeenproducedbytheNationalGeophysical DataCenter(NGDC).Therearethreeversionsofthedataavailablefordownloadincluding(1)“raw,” (2)“stablelights”and(3)“calibrated.”Thestablelightsversion,whichremovesclouds,gas flares,lightningandotherephemeralandextraneoussignalsusingtheproceduredescribedby Elvidgeetal.[25],wasusedinthisstudy.ImagesinthisstudycapturednighttimelightsoverMainland Chinafrom1992(theearliestavailableyear)to2012(themostrecentavailableyear).
NighttimelightingDMSP/OLSdatafromthesetwodecadesshouldbeinter-calibratedbefore theycanbeusedasreliableindicatorsforghostcitiesanalysisforatleasttworeasons.First,thetime
seriesdataarefromsixdifferentsatellites(F10,F12,F14,F15,F16andF18)acrosswhichOLSsensor settingsmayvary.Second,forthedatafromthesamesatellitemayvary,sincesensorsageandare subjecttonaturaldeteriorationovertimeandundocumentedgainadjustments.
Basedontheassumptionthatnighttimelightremainsstableovertimeinaparticulararea, Elvidgeetal.[26]havepresentedamodelofquadraticregressioninthereferenceareaviaanempirical procedure.Ourstudyappliedthismodeltothe21-yearDMSP/OLSdatasetsinceitisfullydocumented andcompleteandhasbeenusedasanacademicstandardindifferentpreviousstudies[27,28].Inthis model,new,inter-calibratedDigitalNumbers(DN)canbecalculateddirectlyfromtheoldnumbers throughthefollowingequation,where DNcalibrated isthenewinter-calibratedvalueand DN isthe original DN valueoftheimage:
Thethreecoefficientsinthemodel—a, b and c—areobtainedforeachyearandsatelliteby choosingJixiCityinChina’sHeilongjiangProvinceasthereferencearea(Table 1).Whencomparing thesocio-economicdataofGDPandbuilt-upareaforChina’smajorcitiesfrom1992to2012,Jixiis oneofthemoststablecitiesandshowshighconformityof DN valuesoverdifferentyears.Thisis inaccordwiththeassumptionthatnighttimelightsinthereferenceareashouldbestableovertime. UsingF162007asthereferenceimage,datafromothersatelliteyearswereadjustedtomatchthe F162007datarange[29].
Table1. Coefficients a, b and c ofthequadraticregressionmodel[29].
SatelliteYear abc R2
F1019920.00211.0297
1.12420.8977
19930.00251.1260 0.95440.9022
F1219940.00161.1312
1.13910.8902
19950.00820.63701.20330.8592
19960.00970.57781.49180.8272
F1419970.00091.07220.40250.8233
19980.00500.87290.52100.8538
19990.00071.09100.54100.9143
F1520000.00860.49862.17410.8809
20010.00121.0292
20020.00080.9713
2003 0.01261.7774
F162004 0.00101.1041
2005 0.00361.3178
2006 0.00561.3436
0.86520.9126
0.67400.9629
0.93330.9166
0.04500.9266
0.74410.9646
0.35140.9707
20070.00001.00000.00001.0000
20080.00120.92580.61220.9855
20090.00700.43602.35400.9030
F1820100.00350.74030.19450.9511
2011 0.00251.10730.10520.9584
20120.00850.22913.89710.9252
2.1.2.LandsatData
Sincetheproject’sinceptionin1972,Landsathasbeenlabeledthe“goldstandardofland observation.”Withitslong-termhistoricalrecordandmoderatespatialresolution,Landsatallowsa detailedstudyofnaturalandhuman-inducedchangesonthegloballandscape.TheLandsatdataused inthisresearchwerefromtheThematicMapper(TM)andEnhancedThematicMapperPlus(ETM+) sensoronboardLandsat5andLandsat7from1992to2012.ThelevelL1Tproductwasdownloaded
fromtheLandsatdataarchivedbytheUSGSEarthResourcesObservationandScienceCenterand redistributedbyGeospatialDataCloud(http://www.gscloud.cn/).Inordertoavoiddataset-specific errors,Landsatdataneedtobepre-processed/radiometricallycorrectedviaaradiometriccalibration procedureinordertoderivethecorrespondinggroundreflectancevaluesbeforecalculation.Thisphase comprisesthreemainsteps[30–33]:(1)conversiontoat-sensorspectralradiance;(2)conversionto top-of-atmosphere(TOA)reflectance;(3)conversiontogroundreflectance.Afterimplementingthe abovepre-processingschemeforalltheLandsatTM/ETM+images,adatasetwithcompatiblescenes thatareessentialformonitoringghostcitiesovertimewascreated.
dataset-specific errors, Landsat data need to be pre-processed/radiometrically corrected via a radiometric calibration procedure in order to derive the corresponding ground reflectance values before calculation. This phase comprises three main steps [30–33]: (1) conversion to at-sensor spectral radiance; (2) conversion to top-of-atmosphere (TOA) reflectance; (3) conversion to ground reflectance. After implementing the above pre-processing scheme for all the Landsat TM/ETM+ images, a data set with compatible scenes that are essential for monitoring ghost cities over time was created.
2.2.ConstructionYearoftheBuilding
2.2 Construction Year of the Building
Figure 1ashowstheLandsatimagecoveredbyPath123andRow32(falsecolorcompositein whichbands7,4and3aredisplayedasred,greenandblue).Figure 1bshowstheNDBIimages derivedfromtheradiometricallycorrectedimages.IntheNDBIimage,thebuilt-uplandfeaturesare greatlyenhancedwithalightgraytowhitetone,whilevegetationareasareconsiderablysuppressed withadarkgraytoblackshade.Thisowestotheenlargementofthecontrastbetweenbuilt-upland andvegetationbytheNDBI.Figure 1cistime-seriesNDBIvalueoftheexperimentalpoint,whichis locatedat116◦51
Figure 1a shows the Landsat image covered by Path 123 and Row 32 (false color composite in which bands 7, 4 and 3 are displayed as red, green and blue). Figure 1b shows the NDBI images derived from the radiometrically corrected images. In the NDBI image, the built-up land features are greatly enhanced with a light gray to white tone, while vegetation areas are considerably suppressed with a dark gray to black shade. This owes to the enlargement of the contrast between built-up land and vegetation by the NDBI. Figure 1c is time-series NDBI value of the experimental point, which is located at 116°51′18.6″ E and 40°23′16.8″ N.
Time series Landsat data cannot be used directly to identify the construction year of the buildings because the data had quality, quantity and timing errors. Generally speaking, Landsat
TimeseriesLandsatdatacannotbeuseddirectlytoidentifytheconstructionyearofthebuildings becausethedatahadquality,quantityandtimingerrors.Generallyspeaking,Landsatdatahavethe followingdrawbacks:(1)Landsatdataoftenexperiencetechnicalproblemswithseeminglyrandom dataloss,includingtheso-called“Christmastree”anomalyordroppedscansandsun-glintanomaly; (2)mostopticalremotesensingimagesincludecloudsandassociatedcloudshadowswhichobscure thedetectionoflandsurface,reducingtheavailabilityofusableLandsatdataandresultinginthe restrictionofthetimeseriesanalysisinthebuiltareasofthisstudy.Forexample,inthe40citiesin thestudyitwasfoundthatentirelycloud-freeimagesofanentirecityareaarerare,whilepartial coverageismorefrequent;(3)in2003,thepermanentfailureofacomponentoftheLandsat7ETM+ opticalscanningsystem(calledthescanlinecorrectoror“SLC”)resultedinawedge-shapedgapthat accountsforabout25percentofmissingdatapixelsperimage[34].Althoughsomedamagedimages canbesubstitutedwiththosefromLandsat5,otherslackanalternativeorareunavailableforother reasons;(4)accordingtotheEarthResourcesObservation&ScienceCenteroftheUSGeologicalSurvey (EROS-USGS),theremaybedatathatwerecollectedbythesensorbutnotarchivedordistributed. Thesealtogethermakeitdifficulttoconductthisstudyusingconventionalmethods.First,landcover mapscannotbeusedforchangedetectionbecauseerrorsfrom(1)and(2)willsignificantlydecrease thelikelihoodofdetectingtemporalchange[35]andthuslowertheaccuracy.Second,duetogaps causedby(3)and(4),thetemporallydiscontinuouslandcovermapswillreducetemporalresolution belowonceperyear.Tosolvesuchproblems,theconstructionyearofthebuildingsisidentifiedbased onaNormalizedDifferenceBuilt-upIndex(NDBI)logisticmodel.
NDBI isoneofseveralwidelyusedindiceswhichissensitivetothebuilt-uparea[36,37]. Theequationfor NDBI isasfollows:
where B5and B4areshortwaveinfraredband(SWIR,band5)andnear-infraredband(NIR,band4)of theLandsatimages,respectively.Theindexdevelopmentwasbasedontheuniquespectralcurvein built-upareas:thereflectanceintheSWIRwavelengthrangeishigherthanthatintheNIRwavelength range.ItcanbeseeninFigure 1bthatbuilt-upareasexperienceadrasticincrementintheirreflectance fromNIRtoSWIR,resultinginpositivevaluesforbuilt-uppixelswhichcanbeclearlyseparatedfrom theremainingcoversofcloseto0forvegetation.Unliketheconventionalthresholdapplicationof NDBI,whereacertainrangeofvaluesfromtheindexrasterisusedtofindatypicalfeatureofinterest (inmanycasestudies,thethresholdissetatzeroandthusthepositivevaluesof NDBI represent thebuilt-uparea),thisstudyutilizedcontinuousvaluesratherthandiscretizingtheoriginaldataset. ThegreatertheNDBIvalueofapixel,thehigherthepossibilityofthepixelbeingabuilt-uparea. Therefore,duringabuildingconstructioncycle,NDBIwillshiftfromasteadilylowvalueperiodand finallyreachahighvalueperiodandremainatthatlevel.Alogisticmodelwasusedtosimulateand reconstructNDBItimeseriessincesuchmodelbestfitsthevalues[38,39].Anotheruniqueattributeof thelogisticmodeloftimeisitsabilitytodeterminetheconstructionyearofthebuildingsthroughthe inflectionpointofthecurve,whichrepresentsthemaininversionphaseinthecycle(Figure 2).
BlacksolidpointsrepresentannualvaluesofNDBIinnewlybuiltareas(seeFigure 1).Thesolid bluelineisthefittedlogisticmodelandthedashedblacklineistherateofcurvaturechangeofthe model.Theredcircleistheminimumpointindicatingtheconstructionyear.
TimeseriesNDBIcanbemodeledusingthefollowingequation:
where t istheyear, y(t)istheNDBIvalueatyear t, a and b arefittingparameters, c + d isthemaximum NDBIvalueand d istheinitialbackgroundNDBIvalue.Thefittingparameters a and b weredetermined usingleast-squarefitting.
Figure 2. Detection of the construction year of the building
Detectionoftheconstructionyearofthebuilding.
Black solid points represent annual values of NDBI in newly built areas (see Figure 1). The solid blue line is the fitted logistic model and the dashed black line is the rate of curvature change of the model. The red circle is the minimum point indicating the construction year.
Time series NDBI can be modeled using the following equation:
BeforetheNDBIcouldbefittothelogisticmodeldescribedbyEquation(3),itwasnecessaryto smooththedatathroughaSavitzky-Golayfilter(SG),whichbothsmoothsandcomputesderivatives ofasetofconsecutivevalues.BasedonanasymmetricGaussianfunctionfitting,thedesignconcept ofSGistofindasuitablefiltercoefficienttobothremoverandomnoiseandprotectthehigh-order movement[40].Ratherthanaconstant,thelocalquadraticpolynomialisutilizedfortheleast-squares fittingwithinamovingwindow.Takingafixednumberofpointsinthevicinitypointtofitapolynomial, SGgivesthesmoothvalueofthevicinitypointaccordingtothepolynomialduringthefittingprogress. Theequationofthefittingprocesscanbegivenasfollows:
where t is the year, y(t) is the NDBI value at year t, a and b are fitting parameters, c + d is the maximum NDBI value and d is the initial background NDBI value. The fitting parameters a and b were determined using least-square fitting
Before the NDBI could be fit to the logistic model described by Equation (3), it was necessary to smooth the data through a Savitzky-Golay filter (SG), which both smooths and computes derivatives of a set of consecutive values. Based on an asymmetric Gaussian function fitting, the design concept of SG is to find a suitable filter coefficient to both remove random noise and protect the high-order movement [40]. Rather than a constant, the local quadratic polynomial is utilized for the least-squares fitting within a moving window Taking a fixed number of points in the vicinity point to fit a polynomial, SG gives the smooth value of the vicinity point according to the polynomial during the fitting progress. The equation of the fitting process can be given as follows:
where NDBI(t)isthefittedNDBIvalue, Ci isthecoefficientforthe ithNDBIpointofthefilter, N isthe numberofconvolutingintegersequaltothesmoothingwindowsize(2m+1)and j istherunning indexoftheoriginalordinatedatatable.Duringtheprocess,invalidpointsaffectedbyexternalfactors intheNDBItimeseriesshowabnormallyloworhighvaluesandthesewillbeeliminatedduringthe SGfilterprocess.
(4)
ThelocalminimumpointofthefittedlogisticmodeldividedNDBIintosustainedincreasing anddecreasingtrends.Sincetheconstructionyearofthebuildingcorrespondstothepointatwhich therateofchangeincurvatureexhibitsthelocalminimum,itisdefinedastheyearwhenthesecond derivativeshiftsfromnegativetopositivevalues[38].Thecurvature K forEquation(4)atyear t canbe computedby:
where NDBI(t) is the fitted NDBI value, Ci is the coefficient for the ith NDBI point of the filter, N is the number of convoluting integers equal to the smoothing window size (2 m + 1) and j is the
where z = ea+bt , a istheangleoftheunittangentvectoratyear t alongadifferentiablecurveand s isthe unitlengthofthecurve.Therateofchangeofcurvature K canbecomputedbythefollowingequation:
2.3.NighttimeLight’sRateofChangeinNewlyBuiltArea
Nighttimelight’srateofchangeinnewlybuiltarea(NLRCNBA)isusedtoassessacity’sdarkness accordingtothefollowingformula:
where i istheyearsequencefrom1992to2012; n isthenumberofyears; t istheconstructionyearof buildings; DN istheDMSP/OLSvalue;and NLRCNBAt denotesthenighttimelight’srateofchangein anewlybuiltareaforyear t.Since NLRCNBAt indicatesthemeantemporalchangeoftheinterannual variationoftheDMSP/OLSvalueinanewlybuiltareaforconstructionyear t,theyearly NLRCNBA changedirectlyreflectsthedynamicofdarknessinanewlybuiltarea.Therefore,the NLRCNBA trend, asaresultofresidentialactivitiesinanewlybuiltarea,isanappropriateindicatorforghostcities: ifthetrend>0,thecityisbrighterandtheintensityofresidentialactivitiesinnewbuildingsincreases; otherwise,thecitybecomesdarkerandtheintensityofresidentialactivitiesinnewbuildingsdecreases. Inthisstudy,thestatisticalsignificanceoftrendswasassessedbasedonMann-KendalltestandSen’s slopewasusedtoevaluatethe NLRCNBA trend.
TheMann-Kendall(MK)testisanon-parametricstatisticalprocedurethatiswellsuitedto analyzingtrendsindataovertime[41,42].Oneadvantageofthistestisthatthedataneednotconform toanyparticulardistribution.Thesecondadvantageofthetestisitslowsensitivitytoabruptbreaks duetoinhomogeneoustimeseries.Asaresult,thetimeseriesofNLRCNBAvaluesofcitieswere analyzedformonotonousincreasingordecreasingtrendsusingtheMKtest.
AlongwiththeMann-Kendalltest,Sen’sslopecalculationwasdetermined.Unlikelinear regressions,Sen’sslopetestisanonparametric,linearslopeestimatorthatworksmosteffectively onmonotonicdata.SimilartotheMKtest,theadvantageofSen’sslopeestimatoristhatitis largelyunaffectedbygrossdataerrors,outliers,ormissingdata.Therefore,itismorerigorousthan mostregressionslopesandprovidesamorerealisticmeasureofthetrendsintheNLRCNBAseries. Simplyspeaking,Sen’sslopeisthemedianofalldifferencesbetweensuccessivedatavalues[43,44].
2.4.Validation
Figure 3 showstwosampleplotsinBeijing,locatedat116◦51 18.6 Eand40◦23 16.8 Nand 116◦38 7.3 Eand40◦20 6.3 N,respectively.Photosweretakenduringfieldsurvey.Inthetimeseries Landsatcolorcompositeimages(falsecolorcompositeinwhichbands7,4and3aredisplayedasred, greenandblue)thesolidrectangleindicatesvegetatedareaandthedashedrectangleindicatesthat wasconvertedtobuilt-uparea.Theconstructionyearofthebuildingislabeledasredtext.
Ananalysisoftheachievedaccuracytodetecttheconstructionyearofthebuildingwasperformed byextractingasetofsampleplotsandvisuallyinspectingtheresultsinthoseplots.Thevalidation phasewasconductedbystepwiseprocedure,asfollows:(1)Sampleplotswereextractedusinga randomsamplingmethodtorepresentthevariabilitywithintheplotsample[45].Thenumberof sampleplotsrequiredforaccuracyassessmentdependsontheminimumlevelofaccuracyrequired. Jensendiscussesanequationsuitablefordeterminingaminimumnumberofpixelsrequiredfor differentlevelsofaccuracy[46],asfollows:
where N istotalnumbersampleplots, p isexpectedpercentaccuracyand E isallowableerror.Here p wasadoptedas80percentand E wasadoptedas ±5percentandthus N =256.Finally,300sample plotswereusedforaccuracyassessment.(2)Theresultsoftheconstructionyearwereassessedby comparisonwiththetimeseriesLandsatcolorcompositeimagesandfieldsurvey(Figure 4).
Toremovenoiseanddisturbance,cityareasweremaskedoutusingtwovectorpolygonswhich definetheurbanoutlinein1992and2012andonlythebuilt-uplandsinsidethe2012urbanregionand outsidethe1992urbanregionwereretainedasnewlybuiltarea.InBeijing,theoverallaccuracyofthe logisticmodelinidentifyingtheconstructionyearsofbuildingswas73.6percentwithdisagreements on79points.Thisdiscrepancymainlyoccurredinmixedurbanareas,whicharestronglyaffectedby thelowseparabilitybetweenagriculturalorvegetatedareasandbuildings.Suchfragmentationand varietycausedfalsityin51points.Theother28failurescanbeattributedtodatainconsistencyfrom time-seriesLandsatimages.Overall,thevalidationsshowthatthemethodcanprovidecorrectand reliableresultswithasatisfactorylevelofaccuracy.
3.Results
3.1.ACaseofNLRCNBATrend
Figure 4ashowsthenighttimelight’srateofchangefrom1992to2012inGuilinCity,Southwest China.Figure 4bshowsyearsagainstthenighttimelight’srateofchangeinnewlybuiltareas.Thenull hypothesis(Ho)isthatnochangeortrendhasoccurredinNLRCNBAovertime.Hencethealternate hypothesis(Ha)isthatasignificantchangehasoccurredovertime,orthatanincreasingordecreasing trendinNLRCNBAispresent.Inthiscase,thereisaverysmall p-valueof0.000046fromMann-Kendall test,whichismuchlessthanasignificancelevelof0.01(correspondingtoa99percentconfidence level),resultinginahighweightofevidenceagainstthenullhypothesis(Ho).So,Hoisrejected, oritisacceptedthatachangeofthenighttimelightrateofchangeinanewlybuiltareahasoccurred overtime.
TakingGuilinCityasacasestudy,thenighttimelight’srateofchangewaspositiveinalmost allareas,withtheresultthatthemapcolor(Figure 4a)ispredominantlyredoryellowratherthan blue.However,despitesuchgenerallyincreasingtrends,innewlybuiltareas,theslopeoftherateof
Thesignificancelevel(α)istheprobabilityofconcludingwhetherthereisaNLRCNBAtrend whennoneactuallyexists.Asignificancelevelof0.10isconsideredaconservativeapproachsothatthe analysisissensitivetopossiblechangesinNLRCNBAovertime.Significancelevels(α)weredivided intothreesub-levels:0.01,0.05and0.1.Alowersignificanceleveldecreasesthechancesoferroneous indicationofaNLRCNBAtrend,orindicatesthataNLRCNBAtrendismorelikelytooccur.Itwas foundthat,forthe15bigcitiesofGroupB,thesignificancelevelswerelessthan0.10,exceptChangsha andNanchang.Furthermore,thebigtwocitiesandTier1citiesalldippedbelowthe0.01leveland eightcities(72percentofthetotal)inTier2werelessthan0.05.ComparedtoGroupB,thesmaller citiesofGroupAshowedmorediversity:eightcitiespassedthe0.01level,sevencitiespassedthe 0.05levelandthreecitiespassedthe0.10level,accountingfor32percent,28percentand12percentof thetotal,respectively.
Figure 6aillustratesdifferencesofthecoefficientofdeterminations(R2)betweenthelinear regressionmodelandthenonlinearregressionmodelofsecond-orderpolynomialinbigcities; 6billustratesthepeakyearsindescendingorder;in6c,thesolidbluelinesarethesecond-order polynomialregressionmodelandthedashedredlinesarepeakyearsaccordingtotherootsofthe derivative;and6dillustratesNLRCNBAtrendsfor2007–2012basedontheirrealvalues.
Figure 6a illustrates differences of the coefficient of determinations (R2) between the linear regression model and the nonlinear regression model of second-order polynomial in big cities; 6b illustrates the peak years in descending order; in 6c, the solid blue lines are the second-order polynomial regression model and the dashed red lines are peak years according to the roots of the derivative; and 6d illustrates NLRCNBA trends for 2007–2012 based on their real values.
Figure 6. Regression model for NLRCNBA: (a) R2 for linear regression model and second-order polynomieal regression model; (b) peak year for second-order polynomial regression model; (c) NLRCNBA for cities; (d) NLRCNBA trends for 2007-2012.
Figure6. RegressionmodelforNLRCNBA:(a)R2 forlinearregressionmodelandsecond-order polynomiealregressionmodel;(b)peakyearforsecond-orderpolynomialregressionmodel; (c)NLRCNBAforcities;(d)NLRCNBAtrendsfor2007-2012.
TworegressionmodelsweretestedtodefinethedistributionofNLRCNBAintermsofyearsin bigcities:alinearregressionmodelandanonlinearregressionmodelofsecond-orderpolynomial (Figure 6).Itwasfoundthatthecoefficientofdeterminations(R2)inthenonlinearregressionmodel ofsecond-orderpolynomialweresignificantlyhigherthaninthelinearregressionmodel.Thiswas especiallytrueinlower-rankingcities.Forexample,thecoefficientofdeterminations(R2)increased significantlyfrom0.25to0.51forHangzhou,from0.19to0.75forChangshaandfrom0.14to0.68for Nanchang.HigherR2 levelssuggestthatthelinearregressionmodel,whereNLRCNBAisdirectly proportionaltotheamountofyears,isnolongerapplicableorappreciate.Thus,asecond-order polynomialregressionmodelbetterfitsthedata,resultinginhighcoefficientsofdeterminationvalues (R2)of0.94,0.86,0.74,0.51,0.74,0.85,0.75and0.68forthecitiesofBeijing,Shanghai,Guangzhou, Hangzhou,Wuhan,Xi’an,ChangshaandNanchang,respectively.Inthesecities,thismodelpresentsa maximumvalueforNLRCNBA.Becauseofthiseffect,theresponseofNLRCNBAagainstaspecificyear reachedafiniteamount.ThepeakyearscorrespondingtomaximumNLRCNBAvaluesweredetected byfindingtherootsofthederivatives.Itwasfoundthatwhenacity’srankingdecreased,itspeak yearaccordinglyadvanced(inthiscase,from2009,2007,2005to2003inFigure 6b).Asecond-order polynomialregressionmodelwillfallafterreachingthepeakyear,resultinginadescendingtrend. 2007wasregardedasastartingyearanditwasconfirmedthatNLRCNBAtrends,basedontheirreal values,werenegativeinrecentyears.
4.Discussion
Theghostcityproblemisrealbutsofarappearstobeconcentratedinsmallercitiesundergoing aneconomicshift,orwherethelocaleconomyisheavilydependentonnaturalresources,industries, tourism,orfarming.Forexample,thecityofUlanqabranksasthefirstnegativevalueonSen’sslope. Ulanqabhasaveryunbalancedeconomicsystem,with53.1percentofthepopulationrelyingon farmingandanimalhusbandry[47].Itisalsotooremoteandcoldtoattractimmigrants.Shizuishan, aresource-exhaustedcityrelyingonitscoalminesinNorthwestChina’sNingxiaHuiautonomous region,isrankedsecond.Otherresource-exhaustedcitiesincludeBenxi,Daqing,Hebi,Jincheng, Maanshan,TielingandYinkou.Thethirdcity,Lijiang,isapopulartouristdestination,resultingina largelytransientpopulation.OthertouristhotspotsincludeGuilin,Qinhuangdao,KaifengandWeihai.
Othercities,likeBaoji,Leshan,Nantong,Rizhao,Shiyan,Yueyang,ZhanjiangandZhenjiang, areimplementingarduousmeasurestocatchupwiththecountry’seconomicdevelopment.Allthese ghostcitiesaregenerallycharacterizedbyalargenumberofurbanrenewalprogramsandexpansion, orarebrandnewcitieswhoseeconomiesarenotdiversifiedenoughtoofferawiderangeof employmentopportunitiestoattractasustainableinflowofmigrants.
Theresultsfrombigcitiesshowacompletelydifferentsituation:trendsovertimeareallpositive. Suchpositivetrendscanbepartiallyverifiedbya2013CLSAsurveythatcovered800,000apartments in12citiesandfoundthatthevacancyrateinbigcitieswasjust10to16percent,asopposedto20to 30percentinsmallercities[13].
SimilartoaKuznetsmodel[48],bigcitiesinourstudyfollowedaninvertedUcurvethat graphsNLRCNBAagainsttime,overthecourseofeachcity’seconomicdevelopment.Inthisperiod, thedatashowsanincreaseinnighttimelight,whichpeaksandthenfalls.Thiscurvehasanumber ofimplications.Theleftsideofthecurveillustratesthatasaneconomydevelopsandcreatesmore jobsinacity,thepaceofhomesalesbeginstoaccelerate.Here,manypeoplearebuyingnewhouses inconjunctionwithlarge-scalemigration.ThisresultsinapositiveNLRCNBAtrend.Atthisstage, thelandboomleadstoarapidincreaseinpropertyvaluation.
Theright-handsideofthecurveshowsthedeclineofhomesalesafterthepeak.Whenvaluations ofrealpropertyreachunaffordablelevelsformostpeople,orthereisaneconomicslump,thenumber ofunsoldhomesishigherandhousingdemandsislower.Asaresult,theNLRCNBAtrendbeginsto showupnegative.ThisdescribesthecurrentsituationinChina’sbigcities,withasteadydropinthe numberofhomessold.
ThemiddleofthecurveshowswhenNLRCNBAreachesitspeak.Ourstudyhighlightshow thepeakyearsvaryasthecities’rankingvaries—thehigheracity’srank,thelateritspeakyear. ThismeansthathightiercitiesaretheconcentrationofChina’seconomicstrengthandthedemand forcommercialrealestateinhightiercitiesismuchstrongerthanthatinlowtiercities.TheNational BureauofStatisticsreportedthatin2015,averagehomepricesinChina’sfourlargestcitiesofBeijing, Shanghai,GuangzhouandShenzhen,wheresupplyisscarce,grewmuchfasterthantheother66 communitiessurveyed.Meanwhile,manysmallercommunitiesstillsufferfromlargeinventoriesof unsoldhomes[49].
Still,itremainsamajorchallengetoconducttheassessmentonmappingChina’sghostcities. Severallimitationsanduncertaintiesassociatedwiththisstudyareasfollows.Firstofall,thestudy wasrestrictedto25smallerand15largercitiesinChinafor1992–2012duetotheavailableyearof DMSP/OLSanddataqualityofLandsat.Second,amodelofquadraticregressionwasusedtocalibrate timeseriesnighttimelightsdatabuttherearestilluncertaintiesduringtheprocedure.Third,theloss intimeseriesNDBIcouldintroduceerrors,althoughalogisticmodelwithfittingparameterswasused toreconstructthevalues.Inaddition,thisstudycannotquantifythebuildingswithweaknighttime lightssincelightsfromthesebuildingsarebelowthethresholdofDMSP/OLS.
Approachesbasedonnewdatasourcesfordealingwiththeselimitationsanduncertainties couldbedevelopedinorderforthemeasurementandreportingofghostcitiestobemoresuccessful. TheVisibleInfraredImagingRadiometerSuite(VIIRS),whichenablesanewgenerationofoperational moderateresolution-imagingcapabilitiessinceNovember2011,collectsimageryandradiometric measurementsoftheland,atmosphere,cryosphereandoceansinthevisibleandinfraredbandsofthe electromagneticspectrum[50].SincetheVIIRSinstrumentcollectsthesamespectralbandsasLandsat pluslowlightimagingdatasimilartoDMSPfromasinglesensor,itmaybepossibletoproducethe mostrecent(2012–present)ghostcitymapsinasimpleway.Furthermore,toovercomedataquality barriersofLandsat,MODISdatacouldbeusedtocalculateNDBIbacktoyear2000.Asaresult, nationalwideghostcitymapcouldbeproducedthroughthecombinationofDMSP/OLSnighttime lightsandMODIS-basedNDBI.
5.Conclusions
AsChinesecitiespursueeconomicgrowththroughurbanizationandthusimplementambitious landdevelopmentprojects,someofthemhaveinadvertentlyturnedthemselvesintoghostcities. Traditionalsocial-economicstatisticaldatabasedonthehouseholdsurveysinChinaisinsufficient toconductin-depthresearchintosuchsocial-economicphenomenon.Anumberofresearchesapply remotesensingtocharacterizeurbanextent,monitorthedynamicsofurbanexpansionandevaluate associatedenvironmentalimpacts.Forexample,DMSP/OLSdatacoveringtwodecadesiswidely usedtodeterminethespatiotemporaldimensionsofsocio-economicfactors,whileLandsatsatellite imagesallowsadetailedstudyofnaturalandhuman-inducedchangesonthegloballandscape. Although,anyof thesedatasets,whenusedbythemselves,isnotthebestwaytosettlethisproblem, thecombinationofthesedatasetshasauniquecapacitytocapturethesize,growth,level,distribution, scale,intensityandpatternofChina’sghostcity.Asproveninthisresearch,residentialactivities declineinnewlybuiltareasofsmallercitiesinChina,leadingtowhatareknownasghostcities. Atnight,uninhabitedhousingunitswithfewlightsappeardark—asignthatfewpeoplelivethere. Oneparticularlyimportantresultfromthisworkisthatourresultsprovideanewperspective towardsincreasedusageofmultiplesourcesofremotelysenseddata,especiallythecomplementary datasourceinurbanarea.Previousresearchesfocusonthecomplementarycharacteristicsinspace, forexample,vegetationareaandbuiltupareabasedonMODISproductsandDMSP/OLSdata.Acase isthattheDMSP-OLSandTerraMODISNDVIdatawerecombinedtodevelopasettlementindex imagetoestimatefractionalsettlementsfromcoarsespatialresolutionimagesattheregionalscaleby combiningalimitednumberofmediumspatialresolutionimages[51].Otherresearch[52]indicates thatdespitetheuncertaintiesinsensorfusionandthecoarseresolutionofthedata,thecombination
ofMODISproductssuchasNPPwithnighttimedatacouldproviderapidassessmentofurbanland coverchangesandtheirimpactsonregionalecosystemresources.Inthisstudy,thecomplementary characteristicsintimebetweennighttimeremotelysenseddataanddaytimeremotelysenseddata cancontributetoenrich,extendandwidenthetopicandthereforewellprovidemorerepresentative characterizationoflanduseanddemographicgeospatialpatternsfromrapidurbanspatial-temporal changes,forexample,ghostcities.Itwasfoundthat22smallercitiescanbecategorizedasghostcities andnighttimelightschangeinnewlybuiltareasfollowinganinvertedU-curveforbigcities.
Thelogicalextensionofthisworkwillbetoapplythemethodologyinotherareas,especiallyin developingcountrieswhicharecurrentlyexperiencingassizeablegrowthratesofurbanconcentration duetorapidpopulationgrowthandeconomictransformationbyacombinationofrapidtechnological andpoliticalchange[24,53]asinChina.Howtoquantifysideeffectsofurbanizationataregional level,suchasghostcities,areessentialforpolicymakerstoputforwardcountermeasurestosolve them.Doingsowillrequireavarietyofreliableandwell-establishedinformation,includingtheir spatial-temporalpattern.Butlackofthefinancialand/orinstitutionalresourcestoconductall-round andin-depthcensusesinthesecountriesresultsinunavailabilityandpoorqualityofthestatisticaldata. Inparticular,ourmethodologicalapproachwouldwelldetectresidentialactivitieschangeinnewly builtareasvianighttimesatellitedataanddaytimesatellitedataandthusallowpolicymakerstotrack thespatialpatternandtemporalevolutionofghostcities.Thisisofgreatsignificanceinupgrading urbanmanagementlevel,implementingefficienturbanadministrationandachievingthestrategic goalofurbansustainabledevelopment.
AuthorContributions: Conceptualization,H.L.;Fundingacquisition,H.L.;Investigation,X.Y.andC.M.; Methodology,H.L.,C.Z.andG.L.;Writing-originaldraft,H.L.,C.Z.,G.L.,X.Y.andChanghongMiao; Writing-review&editing,H.L.,C.Z.,G.L.,X.Y.andC.M.
Funding: ThisresearchwasfundebyHASTIT(16HASTIT022),NSFC41371525,NSFC41430637,16IRTSTHN012, JOF201702,ScientificResearchStart-upFundingoftheProgramSupportingSpecialTalentZone(HenanUniversity; toZhaodongFENG)andScientificPromotionFundingofthePrioritizedAcademicDiscipline(Geography,Henan University;toZhaodongFENG).
Acknowledgments: Theauthorsthanktheanonymousreviewerswhosecommentsandsuggestionswerevery helpfulinimprovingthequalityofthispaper.Theauthorswouldliketothankthefollowingpeoplefortheir contributionswithtothisprojectindatadownload,dataprocessingandfieldwork:ZujinLiu,ZhengXu, YihangHu,TianweiFeng,ZhidanLiu,JingqiLu,PenghuiDuan,YunshengChen,PeijunLi,DandanLi, ShuaishuaiZhang,TianningZhang,DandanZhaoandBoyanZhou.
ConflictsofInterest: Theauthorsdeclarenoconflictofinterest.
References
1. Cartier,C.Territorialurbanizationandtheparty-stateinChina. TerritoryPolit.Gov. 2015, 3,294–320. [CrossRef]
2. Gu,C.;Wei,Y.D.;Cook,I.G.PlanningBeijing:Socialistcity,transitionalcityandglobalcity. UrbanGeogr. 2015, 36,905–926.[CrossRef]
3. Li,H.China’sUrbanizationRatetoReach60%in20Years. People’sDailyOnline.27December2013.Available online: http://en.people.cn/200105/17/eng20010517_70205.html (accessedon28November2015).
4. ChinaPopulationandDevelopmentResearchCenter(CPDRC). China’sUrbanPopulation;People’sUniversity ofChina:Beijing,China,2012.
5. Woetzel,J.;Mendonca,L.;Devan,J.;Negri,S.;Hu,Y.;Jordan,L.;Yu,F. PreparingforChina’sUrbanBillion; McKinseyGlobalInstitute:NewYork,NY,USA,2009.
6. DevelopmentResearchCenteroftheStateCouncil(DRCSC). AnAnalysisoftheSystemicCausesofChina’s UrbanPathologies;ReportNumber67;DRCSC:Beijing,China,2012;pp.1–13.
7. DevelopmentResearchCenteroftheStateCouncil;TheWorldBank. UrbanChina:TowardEfficient,Inclusive andSustainableUrbanization;WorldBankPublications:Washington,DC,USA,2014.
8. Young,O.;Guttman,D.;Qi,Y.;Bachus,K.;Belis,D.;Cheng,H.;Lin,A.;Schreifels,J.;VanEynde,S.; Wang,Y.;etal. Institutionalizedgovernanceprocesses:ComparingenvironmentalproblemsolvinginChina andtheUnitedStates. Glob.Environ.Chang.-Hum.PolicyDimens. 2015, 31,163–173.[CrossRef]
9. Sorace,C.;Hurst,W.China’sphantomurbanizationandthepathologyofghostcities. J.Contemp.Asia 2015, 46,304–322.[CrossRef]
10. Woodworth,M.D.;Wallace,J.L.Seeingghosts:ParsingChina’s“ghostcity”controversy. UrbanGeogr. 2017, 38,1270–1281.[CrossRef]
11. Yu,H.China’s“ghostcities”. EastAsianPolicy 2014, 6,33–43.[CrossRef]
12. Chuanchuan,Z.;Kun,J. HumbleAbodeinGhostTowns?IncomeInequalityandHousingBubble;ChinaEconomics NetworkWorkingPaperSeries;ChinaEconomicsNetwork:Beijing,China,2013.
13. AsiaOne.Nobody’sHome:China’sVacantPropertyRateSoars.Availableonline: http://business.asiaone. com/news/nobodys-home-chinas-vacant-property-rate-soars (accessedon28November2015).
14. Chen,Y.ChasingGhosts. SouthChinaMorningPost.13February2015.Availableonline: http://multimedia. scmp.com/china-ghost-towns/ (accessedon28November2015).
15. Ju,Y.;Dronova,I.;Ma,Q.;Zhang,X.AnalysisofurbanizationdynamicsinmainlandChinausingpixel-based night-timelighttrajectoriesfrom1992to2013. Int.J.RemoteSens. 2017, 38,6047–6072.[CrossRef]
16. Jin,X.;Long,Y.;Sun,W.;Lu,Y.;Yang,X.;Tang,J.Evaluatingcities’vitalityandidentifyingghostcitiesin Chinawithemerginggeographicaldata. Cities 2017, 63,98–109.[CrossRef]
17. Zheng,Q.;Zeng,Y.;Deng,J.;Wang,K.;Jiang,R.;Ye,Z.“Ghostcities”identificationusingmulti-source remotesensingdatasets:AcasestudyinYangtzeRiverDelta. Appl.Geogr. 2017, 80,112–121.[CrossRef]
18. Zheng,Q.;Deng,J.;Jiang,R.;Wang,K.;Xue,X.;Lin,Y.;Shahtahmassebi,A.R.Monitoringandassessing “ghostcities”inNortheastChinafromtheviewofnighttimelightremotesensingdata. HabitatInt. 2017, 70, 34–42.[CrossRef]
19. Baker,L. GhostTownsofTexas;UniversityofOklahomaPress:Norman,Oklahoma,1986.
20. Shepard,W. GhostCitiesofChina;ZedBooks:London,UK,2015.
21. JonesLangLaSalleIP,Inc.(JLL). China50:FiftyRealEstateMarketsthatMatter;JLL:Beijing,China,2012.
22. JonesLangLaSalleIP,Inc.(JLL). China60:FromFastGrowthtoSmartGrowth;JLL:Beijing,China,2015.
23. Elvidge,C.;Baugh,K.;Sutton,P.;Bhaduri,B.;Tuttle,B.;Ghosh,T.;Ziskin,D.;Erwin,E. Who’sintheDark: Satellite-BasedEstimatesofElectrificationRates;OakRidgeNationalLaboratory:OakRidge,TN,USA,2011.
24. Lu,H.;Liu,G.Spatialeffectsofcarbondioxideemissionsfromresidentialenergyconsumption: Acounty-levelstudyusingenhancednocturnallighting. Appl.Energy 2014, 131,297–306.[CrossRef]
25. Elvidge,C.;Baugh,K.;Kihn,E.;Kroehl,H.;Davis,E.Mappingcitylightswithnighttimedatafromthe DMSPOperationalLinescanSystem. Photogramm.Eng.RemoteSens. 1997, 63,727–734.
26. Elvidge,C.;Ziskin,D.;Baugh,K.;Tuttle,B.;Ghosh,T.;Pack,D.;Erwin,E.;Zhizhin,M.Afifteen-yearrecord ofglobalnaturalgasflaringderivedfromsatellitedata. Energies 2009, 2,595–622.[CrossRef]
27. Small,C.;Elvidge,C.Nightonearth:MappingdecadalchangesofanthropogenicnightlightinAsia. Int.J. Appl.EarthObs.Geoinf. 2013, 22,40–52.[CrossRef]
28. Elvidge,C.D.;Sutton,P.C.;Baugh,K.;Ziskin,D.C.;Ghosh,T.;Anderson,S.Nationaltrendsinsatellite observedlighting:1992–2009.In AGUFallMeetingAbstracts;AmericanGeophysicalUnion:Washington, DC,USA,2011;Volume1,p.3.
29. Han,P.;Huang,J.;Li,R.;Wang,L.;Hu,Y.;Wang,J.;Huang,W.MonitoringTrendsinLightPollutionin ChinaBasedonNighttimeSatelliteImagery. RemoteSens. 2014, 6,5541–5558.[CrossRef]
30. Chander,G.;Markham,B.;Helder,D.SummaryofcurrentradiometriccalibrationcoefficientsforLandsat MSS,TM,ETM+andEO-1ALIsensors. RemoteSens.Environ. 2009, 113,893–903.[CrossRef]
31. Song,C.;Woodcock,C.;Seto,K.;Lenney,M.;Macomber,S.Classificationandchangedetectionusinglandsat TMdata:Whenandhowtocorrectatmosphericeffects? RemoteSens.Environ. 2001, 75,230–244.[CrossRef]
32. Ahmed,O.;Franklin,S.;Wulder,M.Characterizingstand-levelforestcanopycoverandheightusingLandsat timeseries,samplesofairborneLiDARandtherandomforestalgorithm. ISPRSJ.Photogramm.RemoteSens. 2015, 101,89–101.[CrossRef]
33. Canty,M.;Nielsen,A.;Schmidt,M.Automaticradiometricnormalizationofmultitemporalsatelliteimagery. RemoteSens.Environ. 2004, 91,441–451.[CrossRef]
34. Zeng,C.;Shen,H.;Zhang,L.RecoveringmissingpixelsforLandsatETM+SLC-offimageryusing multi-temporalregressionanalysisandaregularizationmethod. RemoteSens.Environ. 2013, 131,182–194. [CrossRef]
2018, 10,1037
35. Friedl,M.;Sullamenashe,D.;Tan,B.;Schneider,A.;Ramankutty,N.;Sibley,A.;Huang,X.MODIScollection 5globallandcover:Algorithmrefinementsandcharacterizationofnewdatasets. RemoteSens.Environ. 2010, 114,168–182.[CrossRef]
36. Zha,Y.;Gao,J.;Ni,S.Useofnormalizeddifferencebuilt-upindexinautomaticallymappingurbanareas fromTMimagery. Int.J.RemoteSens. 2003, 24,583–594.[CrossRef]
37. Estoque,R.;Murayama,Y.Classificationandchangedetectionofbuilt-uplandsfromLandsat-7ETM+and Landsat-8OLI/TIRSimageries:Acomparativeassessmentofvariousspectralindices. Ecol.Indic. 2015, 56, 205–217.[CrossRef]
38. Zhang,X.;Friedl,M.;Schaaf,C.;Strahler,A.;Hodges,J.;Gao,F.;Reed,B.;Huete,A.Monitoringvegetation phenologyusingMODIS. RemoteSens.Environ. 2003, 84,471–475.[CrossRef]
39. White,K.;Pontius,J.;Schaberg,P.Remotesensingofspringphenologyinnortheasternforests:Acomparison ofmethods,fieldmetricsandsourcesofuncertainty. RemoteSens.Environ. 2014,97–107.[CrossRef]
40. Savitzky,A.;Golay,M.Smoothinganddifferentiationofdatabysimplifiedleastsquaresprocedures. Anal.Chem. 1964, 36,1627–1639.[CrossRef]
41. Lu,H.;Liu,G.Recentobservationsofhuman-inducedasymmetriceffectsonclimateinveryhigh-altitude area. PLoSONE 2014.[CrossRef][PubMed]
42. Silva,R.;Santos,C.;Moreira,M.;Cortereal,J.;Silva,V.;Medeiros,I.Rainfallandriverflowtrendsusing Mann–KendallandSen’sslopeestimatorstatisticaltestsintheCobresRiverbasin. Nat.Hazards 2015, 77, 1205–1221.[CrossRef]
43. Theil,H.Arank-invariantmethodoflinearandpolynomialregressionanalysis,I,II,III.In ProceedingsKNAW; KoninklijkeNederlandseAkademievanWetenschappen(KNAW):Cham,Switzerland,1950;Volume53, pp.386–392,512–525,1397–1412.
44. Sen,P.EstimatesoftheregressioncoefficientbasedonKendall’stau. J.Am.Stat.Assoc. 1968, 39,1379–1389. [CrossRef]
45. Stehman,S.Comparisonofsystematicandrandomsamplingforestimatingtheaccuracyofmapsgenerated fromremotely-senseddata. Photogramm.Eng.RemoteSens. 1992, 58,1343–1350.
46. Jensen,J. IntroductoryDigitalImageProcessing;Prentice-Hall:EnglewoodCliffs,NJ,USA,1986.
47. Zhang,P.Studyontheactionmechanismofeconomicdevelopmentandeco-environmentinUlanqabCity. J.AridLandResour.Environ. 2010, 3,001.
48. Kuznets,S.Economicgrowthandincomeinequality. Am.Econ.Rev. 1955, 49,1–28.
49. Michael,C.MajorCitiesContinuetoLeadChinaHousingRecoveryinJuly.Mingtiandi.Available online: http://www.mingtiandi.com/real-estate/china-real-estate-research-policy/major-cities-continueto-lead-china-housing-recovery-in-july/ (accessedon28November2015).
50. NationalAeronauticsandSpaceAdministration;GoddardSpaceFlightCenter;VisibleInfraredImaging RadiometerSuite(VIIRS).Availableonline: http://npp.gsfc.nasa.gov/viirs.html (accessedon30June2018).
51. Lu,D.;Tian,H.;Zhou,G.;Ge,H.RegionalmappingofhumansettlementsinsoutheasternChinawith multisensorremotelysenseddata. RemoteSens.Environ. 2008, 112,3668–3679.[CrossRef]
52. Milesi,C.;Elvidge,C.D.;Nemani,R.R.;Running,S.W.Assessingtheimpactsofurbanlanddevelopment onnetprimaryproductivityinthesoutheasternUnitedStates. RemoteSens.Environ. 2003, 86,401–410. [CrossRef]
53. Cohen,B.Urbanizationindevelopingcountries:Currenttrends,futureprojectionsandkeychallengesfor sustainability. Technol.Soc. 2006, 28,63–80.[CrossRef]
© 2018bytheauthors.LicenseeMDPI,Basel,Switzerland.Thisarticleisanopenaccess articledistributedunderthetermsandconditionsoftheCreativeCommonsAttribution (CCBY)license(http://creativecommons.org/licenses/by/4.0/).