Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Glob Change Biol 20191ndash15 wileyonlinelibrarycomjournalgcb emsp|emsp1copy 2019 John Wiley amp Sons Ltd
Received12September2018emsp |emsp Accepted20February2019DOI 101111gcb14611
P R I M A R Y R E S E A R C H A R T I C L E
Quantifying the biophysical and socioeconomic drivers of changes in forest and agricultural land in South and Southeast Asia
Xiaoming Xu1 emsp| Atul K Jain1 emsp| Katherine V Calvin2
1DepartmentofAtmosphericSciencesUniversityofIllinoisUrbanaIllinois2JointGlobalChangeResearchInstitutePacificNorthwestNationalLaboratoryCollegeParkMaryland
CorrespondenceAtulKJainandXiaomingXuDepartmentofAtmosphericSciencesUniversityofIllinoisUrbanaILEmailsjain1illinoiseduxuxmillinoisedu
Funding informationBiologicalandEnvironmentalResearchGrantAwardNumberDE-SC0016323
AbstractSouth andSoutheastAsia (SSEA)hasbeenahotspot for landuse and land coverchange(LULCC)inthepastfewdecadesTheidentificationandquantificationofthedriversofLULCCarecrucialforimprovingourunderstandingofLULCCtrendsSofarthebiophysicalandsocioeconomicdriversofforestchangehavenotbeenquantifiedattheregionalscaleparticularlyforSSEAInthisstudywequantifythebiophysicalandsocioeconomicdriversofforestchangeonacountry-by-countrybasisinSSEAusinganintegratedquantitativemethodologywhichsystematicallyaccountsforpre-viouslypublisheddriverinformationandregionaldatasetsWesynthesizemorethan200publicationstoidentifythedriversoftheforestchangeatdifferentspatialscalesinSSEASubsequentlywecollectspatiallyexplicitproxydatatorepresenttheidenti-fieddriversWequantifythedynamicsofforestandagriculturallandfrom1992to2015 using the Climate Change Initiative (CCI) land cover data developed by theEuropeanSpaceAgency(ESA)Ageographicallyweightedregressionmethodisem-ployedtoquantifythespatiallyheterogeneousdriversofforestchangeOurresultsshowthatsocioeconomicdriversaremoreimportantthanbiophysicaldriversfortheconversionofforesttoagriculturallandinSouthAsiaandmaritimeSoutheastAsiaIncontrastbiophysicaldriversaremoreimportantthansocioeconomicdriversfortheconversionofagriculturallandtoforestinmaritimeSoutheastAsiaandlessim-portant in SouthAsiaBothbiophysical and socioeconomicdrivers contribute ap-proximately equally to both changes in the mainland Southeast Asia region Byquantifyingthedynamicsofforestandagriculturallandandthespatiallyexplicitdriv-ersoftheirchangesinSSEAthisstudyprovidesasolidfoundationforLULCCmod-eling and projection
K E Y W O R D S
afforestationreforestationdeforestationdriversgeographicallyweightedregressionJohnsonrsquosRelativeWeightSouthandSoutheastAsia
1emsp |emspINTRODUC TION
Terrestrialecosystemshavebeenstronglyimpactedbyhumanac-tivitiesthroughchangesinlanduseandlandcover(LULCC)Since
preindustrialtime(ca1750)morethan50ofthegloballandsur-facehaschanged(GoldewijkBeusenDoelmanampStehfest2017)LULCC impacts the water cycle the carbon cycle climate andgreenhousegas(GHG)emissions(FoleyDefriesampAsner2005)
2emsp |emsp emspensp XU et al
Forexamplein2008ndash2017annualGHGemissionsfromland-usechangewas 15plusmn07GtCyr accounting for ~13of the globalanthropogenicGHGemissions(LeQuereetal2018)Manystud-ieshaveprovidedspatiallyexplicitLULCCdatasetsforquantifyingtheimpactofLULCCsondifferentfluxes(Goldewijketal2017Houghton et al 2012Meiyappan amp Jain 2012 Ramankutty ampFoley1999)
Inthefuturesocialdevelopmentandeconomicopportunitiescoulddrivepeopletofurtherchangetheland(Lambinetal2001)Changesinbiophysicalconditionssuchasclimatesoilandwaterconditionsmayalsoresult in landusechange(MustardDefriesFisherampMoran2012)LargeuncertaintiesexistinfutureLULCCprojections (Presteleet al 2016) partlydue to a limitedunder-standingofthespatialvariationinsocioeconomicandbiophysicaldriversQuantifyingthesocioeconomicandbiophysicaldriversofLULCCcouldimprovetheprojectionsoffuturelandusepatterns(Feddemaetal2005MeiyappanDaltonONeillampJain2014)Some studies have quantified LULCC drivers at different spa-tial scalesForexampleMonMizoueHtunKajisaandYoshida(2012)usedalogisticregressionmodeltoconcludethatthedefor-estationwasnegativelycorrelatedwithelevationanddistancetothenearesttowninthreereservedforestsinMyanmarHoweverthe results of these studies are not evaluated against publishedcasestudiesWhiletherearesomestudiessynthesizingpublishedcasestudiesinanefforttogeneralizeLULCCdrivers(vanVlietetal2016)thesestudiesaretypicallyqualitativeratherthanquan-titative Very few studies combine quantitative LULCC driverswithinformationfromcasestudies
SouthandSoutheastAsia(SSEA)hasoneofthelargestareasoftropicalforests(FAO2015)andisthemostpopulousregionoftheworld(Cervarichetal2016)Althoughbothdeforestationandaf-forestation(orreforestation)processesareobservedthenetfor-estareadecreasedfrom319millionhain1990to292millionhain2015(FAO2015)Thedriversofthesechangesarediversein-cludingsocioeconomicfactorssuchastheincreasingpopulationandeconomicgrowth(LopezampGalinato2005ShehzadQamerMurthyAbbasampBhatta2014)andbiophysicalfactorssuchasclimate change (IslamMiahamp Inoue 2016)However a quanti-tativespatiallyexplicitanalysisof the relationshipsbetween thebiophysicalandsocioeconomicdriversandLULCCattheregionalscaleinSSEAisstilllacking
Thereforetheobjectiveofthisstudyistoquantifythespatiallyexplicit relationships between the biophysical and socioeconomicdriversandLULCCattheregionalscaleinSSEATocarryoutsuchanalysisforthe16countriesintheSSEAregion(BangladeshBhutanBruneiCambodiaIndiaIndonesiaLaosMalaysiaMyanmarNepalPakistanPhilippinesSingaporeSriLankaThailandandVietnam)wesynthesizedcountry-specific case studies to identify themajorLULCCdrivers inSSEAanduseda spatiallyexplicit satellite-basedLULCCdatasetInthisstudywespecificallyfocusedontwoimport-antLULCCactivitiesnamelyfromforesttoagriculturallandandfromagriculturallandtoforest(weuseldquoforestchangerdquotoreferspecificallytothesetwochangesinthispaper)Toourknowledgethisstudyis
thefirstefforttoincorporatecasestudysynthesisandquantitativeanalysistoquantifythebiophysicalandsocioeconomicdriversoffor-estchangeinSSEATheresultsprovideaninsightintothecomplexLULCCprocessesandcouldcontributetomodelingandprojectionofLULCCinSSEAMoreoverthisstudycanserveasascientificbasisforstakeholderstoimprovelandmanagementinSSEAcountries
2emsp |emspMATERIAL S AND METHODS
Ourmethodology toquantify thedriversof forest change canbebroken down into seven steps (a) identify the drivers of forestchangebyanalyzingcasesstudies (b)collectdifferentbiophysicaland socioeconomic proxy data to represent the identified driversof forest change (c) compile satellite-based LULCCdata (d) iden-tifytheconcentratedregionsofforestchangeusingGetis-OrdGihotspotanalysistechnique(e)employprincipalcomponentanalysis(PCA)toaccountformulti-collinearityexistingintheproxydata(f)use a GeographicallyWeighted Regression (GWR)model to buildthe spatially explicit relationships between the proxy and forestchangeareaand(g)determinetherelativeimportanceofeachproxydrivercategoryusingtheJohnsonsrelativeweight(JRW)methodTheGetis-OrdGianalysis(step4)wasconductedinArcMap106(Redlands CA 2017) All other steps were performed in Matlab2017(NatickMA2017)FortheGWRanalysisweusedtheMatlabSpatialEconometricsToolbox (LeSageampPace2009)EachstepofthemethodologywasdescribedinbriefinthefollowingsectionsAdetaileddescriptionoftheindividualstepswithasamplecalculationcanbefoundintheSupplementarySectionTextS1
21emsp|emspSynthesis of the site level case studies
Inthisstudywecollected213publications(including65casestud-iesforIndiasynthesizedbyMeiyappanetal(2017))toidentifythedrivers of forest change at different spatial and temporal scalesWe ran an advanced search in all databases available inWeb ofSciencewith thequeryexpression ldquoTI=(DriversORdeterminantsORcausesORdynamics)ANDTS=(CountrynameANDland)ANDTS=(cropORforestORagriculORdeforORdegrad)rdquoforall16SSEAcountries Therewere in total 565publications thatmetthequery conditionsBy looking through the titles and abstractswemanually excluded 240 publications that did notmention anyLULCC driversWe carefully read the remaining 325 publicationsand excluded 112 publications that studied the drivers of LULCCother than forest changeFinallywedetermined213publicationsthatdiscussedeitherqualitativelyorquantitativelydriversforforestareagainorforestarealossWeusedthecriterionldquototalforestarealossandtotalforestareagainrdquoratherthantheexplicitchangesfromforesttoagricultural landandviceversabecauseveryfewstudieshaveexplicitlyinvestigatedthedriversforthelattercase(especiallythechangesfromagriculturallandtoforest)
Whenanalyzingthe213casestudieswerecordedtheLULCCtype (forest area loss or forest area gain) the study area (the
emspensp emsp | emsp3XU et al
coordinatesofthegeometriccenterforthestudyareaatdifferentspatialscale)timespanandfrequencyofthedriversmentionedinthepublications
WeidentifiedthemajordriversbythefollowingmethodIfthecase study stated important driverswe recorded all of these im-portant drivers mentioned in each publication (some publicationsdiscussedmultipledrivers)Otherwisewetreatedeachofthemen-tioneddrivers asmajor drivers In order to generalize the driverswecombinedsomeofthedriversthatwerecloselyrelatedDetailedinformationisprovidedinTable1
22emsp|emspSpatial proxy data of the drivers for forest change
Afteridentifyingthedriversfromthesynthesisofcasestudieswecollected 16 biophysical and 17 socioeconomic proxy datasets torepresent thesedrivers (Table2)Allof theseproxiesare spatiallygridded data which are used to build the quantitative relation-ships between the drivers and the forest change These proxieshavedifferent spatial and temporal resolutions (Table2)WekepttheoriginalspatialresolutionofeachdriverForexampleweused
TA B L E 1 emspGeneralizationoftheidentifieddriversandtheircorrespondingproxydata
Drivers identified from case studies Collected proxy data Remarks
bull Terrain(topographicalconditions) Terrainindex
bull Soilandotherenvironmentalconditions
Soilchemicalcompositiondepthdrainagefertilityandtexture
bull Wateravailability Distancetowaterbodies
bull Climate Meanrateofchangeandstandarddeviationofannual precipitation
Meanrateofchangeandstandarddeviationofannual temperature
Meanannualpotentialevapotranspiration
bull Fire Meanburnedareafraction
bull Othernaturaldisaster Distancetolandslideevents
bull Populationbull Labor
Meanandrateofchangeinurbanpopulationdensity ThelaboramountisdirectlyrelatedtopopulationMeanandrateofchangeinruralpopulationdensity
bull Urbanization Meanandrateofchangeinurbanareafraction
bull Migration Migration
bull Livestockbull Grazing
ChickenCattleSheepPigGoatandDuckcounts Grazingactivitiesarecorrelatedwithlivestock
bull Accessibilitybull Transportationbull Infrastructure
Marketaccessibilityindex Marketaccessibilityindexconsidersaccessibil-ityandinfrastructureparticularlythetransportationcondition
bull Marketinfluence(price)bull Economy developmentbull Plantationbull Agroforestbull Agricultureexpansion
GDP per capita PlantationagroforestdevelopmentandagricultureexpansionarerelatedtomarketandeconomyAllofthemcanbereflectedbyGDP per capita
bull Mining(industry) Distancetominingfacilities
bull Povertybull Incomedependencyonforestbull Livelihoodbull Benefit
Povertyindex Thedependencyonforestlivelihoodandbenefitareallcloselyrelatedwithpoverty
bull Shiftingagriculture(swidden)bull Forestmanagement(policy)bull Globalization(internationaltrade)bull Fuelwood(loggingandcharcoal)bull Tourismbull Culturebull Technologybull Farmsizebull Natural regenerationbull Educationorsocialawarenessbull Aquaculture
Spatiallyexplicitproxydataarenotavailable
4emsp |emsp emspensp XU et al
05degtimes05degCRUTSclimatedatawhichisrelativelycoarsecomparedto the01degtimes01deg resolutionof the forest changedata Thereforewhenweextractedtheclimatedatavaluesusingthecoordinatesofthe01degtimes01deggridsofLULCCdatathe01degtimes01deggridsinthesame05degtimes05deggridcellallhadthesamevalueAsaresultouranalysismaymisssomedetaileddriverinformationatsmallscalesHoweverbecausetheregioninvestigatedismuchlargerthan05degtimes05degtheclimatedatastillhavespatialgradientsfortheGWRanalysis
Iftheproxydatasetsdonotchangewithtime(egterrainsoilproperties)orareavailableintheliteratureforonlyonetimeperiod(eg distances towaterbodies landslideevents andmining facili-tiesaswellasmarketaccessibilityindex)weusethemdirectlyinthequantitativeanalysis(step6)Iftheproxiesaretime-seriesdata(egburned area fraction precipitation temperature rural and urbanpopulationdensityandurbanareafraction)weusethemeansanddynamics(rateofchangeandstandarddeviation)oftheminstep6Migrationdataareavailableatthedecadaltimescalecoveringfrom1970to2000weusedthe1990to2000dataasitisclosesttoourstudyperiodinstep6
23emsp|emspLand use data
We used the European Space Agency Climate Change Initiative(ESA-CCI) satellite data for land cover dynamics of forest and ag-riculturallandfrom1992to2015Theoriginalproductcategorizedlandcoverinto22classes(level1)(DefournyMoreauampBontemps2017)OurstudyusedthecorrespondingIPCCclasses(Defournyetal2017)ofagriculturallandandforestWederivethespatialdataof the forest changebyoverlaying the1992and2015 land covermapsfromtheESA-CCIdata(300times300mresolution)Thenweag-gregatedtheforestchangemapsto01degtimes01deg(~10kmtimes10km)anddetermined the fractionsof the changedarea from forest toagri-cultural landandagricultural landtoforest ineach01degtimes01deggrid(FigureS1)The01degtimes01degresolutionrepresentsatradeoffbetweenthe finer resolution of the LULCC data (300m) and the relativelycoarser resolution data for biophysical and socioeconomic proxydata (from01degtimes01degto05degtimes05deg)Wealsocalculatedthecoun-try-specificareasofthesetwochanges
Theareaschangedbetween1992and2015duetotheconver-sion of forest to agricultural land and vice versa are 185468km2 and89398km2whencalculatingbasedon1992and2015dataOntheotherhandtheareaschangedare191277km2and97298km2 whenaccumulatingtheyearlyareachangesofthesetwolandchangeactivitiesover theperiod1992ndash2015 (FigureS3)Theareasbasedtheformermethodare9696and9188oftheresultsfromthelattermethodindicatingthatforestchangedataestimatedbasedon1992and2015yearsofdataareabletocapturethemajorinforma-tionforthistimeperiod
24emsp|emspHotspot analysis
Thepurposeofthehotspotanalysiswastoexcludetheregionswithsmaller areas of forest and its change in the driver analysis The
inclusionofsuchlowvalueregionsmightdilutetheimportanceofthemajordriversWeusedtheGetis‐Ord Gianalysistechniquetoidentifythehotspotregionsforthechangesfromforesttoagricul-turallandandagriculturallandtoforestThistechniqueidentifiesstatistically significant spatial clustersofhigh (hot spots) and low(cold spots) valuesof changes (OrdampGetis 1995)A statisticallysignificanthotspothasagreaterareaofLULCCandissurroundedby other regionswith great areas of LULCCWemarked regionswithgt3 standarddeviations (at 99confidence level) as hotspotregionsWe analyzed the hotspot regions in each country at thedistrictlevel
25emsp|emspPrinciple component analysis (PCA)
Multi-collinearity isacommonproblemin landchangemodelingwhere one or more explanatory (or proxy) data are dependentoneachotherAhighdegreeofmulti-collinearityresults inhighstandarderrorsandspuriouscoefficientestimatesWeemployedthePCAmethodtoaccountforthemulti-collinearityWeselectedthePCswithcumulativecontributionrates(tothetotalvariationof all proxy data) greater than 85 (Deng Wang Deng amp Qi2008)
26emsp|emspGeographically weighted regression (GWR)
The GWRmodel constructs a distinct relationship between eachLULCC pixel and concomitant driver proxy data by incorporatingpixelsfallingwithinacertainbandwidthofthecenterLULCCpixel(Charlton Fotheringham amp Brunsdon 2009) Here we used theadaptiveGaussiankerneltodeterminethebandwidthandthelocalextent to estimate the regression coefficients The optimal band-widthsizeofthekernelwasdeterminedbymeansofcomparisonofAkaike InformationCriterion (AIC)withdifferentbandwidth sizesWeconductedtheGWRintheidentifiedhotspotdistrictsofeachcountry
27emsp|emspJohnsons relative weight (JRW)
The GWR results can represent the positive or negative impactof each individual proxy driver aswell as its relativemagnitudeHowever it isdifficulttoevaluatethecombinedeffectsofdiffer-entdrivercategories(seethesixcategoriesinTable2)InordertogeneralizetheforestchangedriversfromdifferentcategoriesweusedtheJohnsonsRelativeWeight(JRW)methodtoquantifytherelativeimportanceofeachindividualproxyandthensummeduptheimportancecoefficientsofalldrivercategoriesTheJRWanaly-sisfirstgeneratesaseriesoforthogonalvariablesthatarethelinearcombinationsofalloriginalproxydataandthenconductsthere-gressionanalysisbyusingthegeneratedorthogonalvariables(thisprocedureisthesameasthePCAanalysisinstep5)ThentheEqS5isusedtodeterminetherelativeimportanceoftheoriginalproxydata (Chao Zhao Kupper amp Nylander-French 2008 Johnson2000)
emspensp emsp | emsp5XU et al
TAB
LE 2
emspProxyvariablesoftheidentifydrivers
C
ateg
ory
Prox
y va
riabl
eTe
mpo
ral
reso
lutio
nSp
atia
l res
olut
ion
Uni
tSo
urce
Biophysical
variables
ITerrainsoil
and
wat
er1
Terrain
Constant
5primetimes5prime
(~10kmtimes10km)
Categoricaldatainto7
gradientclasses
GlobalAgro-ecologicalZones
(GAEZ)v30(FischerNachtergaele
ampPrieler2012)
2-6
Soilchemicalcomposition
depthdrainagefertilityand
texture
7Distancetowaterbodies
5primetimes5prime
kmCalculatedfromGlobalLakesand
WetlandsDatabase(GLWD)level2
data(LehnerampDoll2004)
IIC
limat
e8-10
Meanrateofchangeand
standarddeviationofannual
prec
ipita
tion
Yearly
05deg times
05
degdeg Cdeg C
yea
rClimaticResearchUnit(CRU)TS
401
11-13
Meanrateofchangeand
standarddeviationofannual
tem
pera
ture
mmmmyear
14
Meanannualpotential
evapotranspiration
mm
IIIN
atur
al
disaster
15
Meanburnedareafraction
Yearly(1997ndash2014)
025
deg times 0
25deg
GlobalFireEmissionsDatabase41
(GiglioRandersonampWerf2013)
16
Distancetolandslideevents
Constant
5primetimes5prime
kmCalculatedfromGlobalLandslide
Catalog(KirschbaumAdlerHong
HillampLerner-Lam2010)
Soci
oeco
nom
ic
variables
IVPo
pula
tion
and
urbanization
17-18
Meanandrateofchangein
urbanpopulationdensity
Yearly
5primetimes5prime
inhabitantskm
2 inhabitantskm
2 middotyea
rHYDE32(KleinGoldewijkBeusen
DoelmanampStehfest2017)
19-20
Meanandrateofchangein
ruralpopulationdensity
21-22
Meanandrateofchangein
urba
n ar
ea fr
actio
nyear
23
Migration
Dec
adal
(1970ndash2000)
05deg times
05
degNumberofmigrants
km2
GlobalEstimatedNetMigration
GridsByDecadev1(deSherbinin
etal2012)
VLivestock
24-29
ChickenCattleSheepPig
GoatandDuckcounts
Constant
1 km
times 1
km
Headkm
2GriddedLivestockoftheWorld
(GLW)version2(Robinsonetal
2014)
(Con
tinue
s)
6emsp |emsp emspensp XU et al
3emsp |emspRESULTS
31emsp|emspSynthesis of the case study
The case studies regarding the drivers of forest loss and gain arespreadacrossallcountriesandhavediversespatialscalesfromvil-lagetonationalscale(FigureS5)Thereare173studies(81ofthetotalcollectedstudies)studyingdriversbasedonfieldsurveys in-terviewsandliteraturereviewsMeanwhile40studies(19ofthecollectedstudies)haveusedquantitativeapproachessuchaslinearregressionlogisticregressionsystemdynamicsmodelingandcel-lularautomatamodelingtoreveal thedriversofLULCCHowevermostofthesequantitativestudiesareatsmallerspatialscales(vil-lage to state and province level) (31 studies) There are only ninequantitativestudiesatthenationalscale
We identify the drivers of forest change from collected casestudies(Figure1andTable1)Thebiophysicaldriversincludeterrainor topographicalconditions (egelevationaltitudeand theslopeof land) soil conditions (eg soil fertility texture and moisture)wateravailabilityclimate(egtemperatureprecipitationandtheirchanges)fireandothernaturaldisasters(eglandslide)Thesocio-economicdriversmainlyrelatetopopulationgrowthurbanizationlivestockorgrazing(egovergrazing)marketinfluenceoreconomydevelopment(egGDPpercapitaaccesstomarket)plantationoragroforestdevelopment agricultural expansionminingand indus-try accessibilityand infrastructure (eg transportationelectricityconnection)loggingandfuelwoodandpoverty
Fire and other natural disasters and terrain are the most fre-quentlymentionedbiophysicaldriversofforestarealoss(deforesta-tion) For example in India and Indonesia the two countrieswiththelargestforestareaamajorityofstudiessuggestterrainandfireand other natural disasters as the major biophysical drivers Oursynthesisofsite-levelcasestudiesalsoidentifiespopulationplanta-tionagriculturalexpansionaccessibilityandinfrastructureaswellasfuelwoodandloggingastheimportantsocioeconomicdriversofdeforestation
There are fewer studies on the drivers of forest area gain (af-forestationandreforestation)thanondeforestationAmongthemmoststudieshavefocusedonsocioeconomicdriversThefrequentlymentioned socioeconomic drivers include accessibility povertyeconomy and livestock and biophysical drivers include climateandterrainThemajordriversof forestchangeatdifferentspatialandtemporalscalesaredescribedinmoredetailinSupplementarySectionTextS2FiguresS6andS7
32emsp|emspForest change
Thecountry-specificandspatialdistributionofconvertingfromfor-esttoagriculturallandandagriculturallandtoforestoverthetimeperiod1992ndash2015isshowninFigure2andFigureS4MalaysiaandPhilippinesexperienced largeareasofdeforestationaswellasaf-forestation over 1992ndash2015 Cambodia and Singapore had moredeforestationthanafforestation
Cat
egor
yPr
oxy
varia
ble
Tem
pora
l re
solu
tion
Spat
ial r
esol
utio
nU
nit
Sour
ce
VI
Econ
omy
30
Marketaccessibilityindex
Constant
1 km
times 1
km
-VerburgEllisandLetourneau
(2011)
31
GD
P pe
r cap
itaDecadal(1980
ndash2010)
05deg times
05
degUSD2005
Globaldatasetofgriddedpopulation
andGDPscenariosMurakamiand
Yamagata(2016)
32
Distancetominingfacilities
Constant
5primetimes5prime
kmCalculatedfromMineralResources
On-LineSpatialDatabyUSGS
33
Povertyindex
Yearly(1992ndash2013)
5rsquotimes5rsquo
-Calculatedfrompopulation(HYDE
32)(KleinGoldewijkBeusen
DoelmanampStehfest2017)and
Nighttimelight(Version4
DMSP-OLSNighttimeLightsTime
Series)byfollowingthemethod
developedbyGhoshAnderson
ElvidgeandSutton(2013)
TAB
LE 2
emsp(Continued)
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
2emsp |emsp emspensp XU et al
Forexamplein2008ndash2017annualGHGemissionsfromland-usechangewas 15plusmn07GtCyr accounting for ~13of the globalanthropogenicGHGemissions(LeQuereetal2018)Manystud-ieshaveprovidedspatiallyexplicitLULCCdatasetsforquantifyingtheimpactofLULCCsondifferentfluxes(Goldewijketal2017Houghton et al 2012Meiyappan amp Jain 2012 Ramankutty ampFoley1999)
Inthefuturesocialdevelopmentandeconomicopportunitiescoulddrivepeopletofurtherchangetheland(Lambinetal2001)Changesinbiophysicalconditionssuchasclimatesoilandwaterconditionsmayalsoresult in landusechange(MustardDefriesFisherampMoran2012)LargeuncertaintiesexistinfutureLULCCprojections (Presteleet al 2016) partlydue to a limitedunder-standingofthespatialvariationinsocioeconomicandbiophysicaldriversQuantifyingthesocioeconomicandbiophysicaldriversofLULCCcouldimprovetheprojectionsoffuturelandusepatterns(Feddemaetal2005MeiyappanDaltonONeillampJain2014)Some studies have quantified LULCC drivers at different spa-tial scalesForexampleMonMizoueHtunKajisaandYoshida(2012)usedalogisticregressionmodeltoconcludethatthedefor-estationwasnegativelycorrelatedwithelevationanddistancetothenearesttowninthreereservedforestsinMyanmarHoweverthe results of these studies are not evaluated against publishedcasestudiesWhiletherearesomestudiessynthesizingpublishedcasestudiesinanefforttogeneralizeLULCCdrivers(vanVlietetal2016)thesestudiesaretypicallyqualitativeratherthanquan-titative Very few studies combine quantitative LULCC driverswithinformationfromcasestudies
SouthandSoutheastAsia(SSEA)hasoneofthelargestareasoftropicalforests(FAO2015)andisthemostpopulousregionoftheworld(Cervarichetal2016)Althoughbothdeforestationandaf-forestation(orreforestation)processesareobservedthenetfor-estareadecreasedfrom319millionhain1990to292millionhain2015(FAO2015)Thedriversofthesechangesarediversein-cludingsocioeconomicfactorssuchastheincreasingpopulationandeconomicgrowth(LopezampGalinato2005ShehzadQamerMurthyAbbasampBhatta2014)andbiophysicalfactorssuchasclimate change (IslamMiahamp Inoue 2016)However a quanti-tativespatiallyexplicitanalysisof the relationshipsbetween thebiophysicalandsocioeconomicdriversandLULCCattheregionalscaleinSSEAisstilllacking
Thereforetheobjectiveofthisstudyistoquantifythespatiallyexplicit relationships between the biophysical and socioeconomicdriversandLULCCattheregionalscaleinSSEATocarryoutsuchanalysisforthe16countriesintheSSEAregion(BangladeshBhutanBruneiCambodiaIndiaIndonesiaLaosMalaysiaMyanmarNepalPakistanPhilippinesSingaporeSriLankaThailandandVietnam)wesynthesizedcountry-specific case studies to identify themajorLULCCdrivers inSSEAanduseda spatiallyexplicit satellite-basedLULCCdatasetInthisstudywespecificallyfocusedontwoimport-antLULCCactivitiesnamelyfromforesttoagriculturallandandfromagriculturallandtoforest(weuseldquoforestchangerdquotoreferspecificallytothesetwochangesinthispaper)Toourknowledgethisstudyis
thefirstefforttoincorporatecasestudysynthesisandquantitativeanalysistoquantifythebiophysicalandsocioeconomicdriversoffor-estchangeinSSEATheresultsprovideaninsightintothecomplexLULCCprocessesandcouldcontributetomodelingandprojectionofLULCCinSSEAMoreoverthisstudycanserveasascientificbasisforstakeholderstoimprovelandmanagementinSSEAcountries
2emsp |emspMATERIAL S AND METHODS
Ourmethodology toquantify thedriversof forest change canbebroken down into seven steps (a) identify the drivers of forestchangebyanalyzingcasesstudies (b)collectdifferentbiophysicaland socioeconomic proxy data to represent the identified driversof forest change (c) compile satellite-based LULCCdata (d) iden-tifytheconcentratedregionsofforestchangeusingGetis-OrdGihotspotanalysistechnique(e)employprincipalcomponentanalysis(PCA)toaccountformulti-collinearityexistingintheproxydata(f)use a GeographicallyWeighted Regression (GWR)model to buildthe spatially explicit relationships between the proxy and forestchangeareaand(g)determinetherelativeimportanceofeachproxydrivercategoryusingtheJohnsonsrelativeweight(JRW)methodTheGetis-OrdGianalysis(step4)wasconductedinArcMap106(Redlands CA 2017) All other steps were performed in Matlab2017(NatickMA2017)FortheGWRanalysisweusedtheMatlabSpatialEconometricsToolbox (LeSageampPace2009)EachstepofthemethodologywasdescribedinbriefinthefollowingsectionsAdetaileddescriptionoftheindividualstepswithasamplecalculationcanbefoundintheSupplementarySectionTextS1
21emsp|emspSynthesis of the site level case studies
Inthisstudywecollected213publications(including65casestud-iesforIndiasynthesizedbyMeiyappanetal(2017))toidentifythedrivers of forest change at different spatial and temporal scalesWe ran an advanced search in all databases available inWeb ofSciencewith thequeryexpression ldquoTI=(DriversORdeterminantsORcausesORdynamics)ANDTS=(CountrynameANDland)ANDTS=(cropORforestORagriculORdeforORdegrad)rdquoforall16SSEAcountries Therewere in total 565publications thatmetthequery conditionsBy looking through the titles and abstractswemanually excluded 240 publications that did notmention anyLULCC driversWe carefully read the remaining 325 publicationsand excluded 112 publications that studied the drivers of LULCCother than forest changeFinallywedetermined213publicationsthatdiscussedeitherqualitativelyorquantitativelydriversforforestareagainorforestarealossWeusedthecriterionldquototalforestarealossandtotalforestareagainrdquoratherthantheexplicitchangesfromforesttoagricultural landandviceversabecauseveryfewstudieshaveexplicitlyinvestigatedthedriversforthelattercase(especiallythechangesfromagriculturallandtoforest)
Whenanalyzingthe213casestudieswerecordedtheLULCCtype (forest area loss or forest area gain) the study area (the
emspensp emsp | emsp3XU et al
coordinatesofthegeometriccenterforthestudyareaatdifferentspatialscale)timespanandfrequencyofthedriversmentionedinthepublications
WeidentifiedthemajordriversbythefollowingmethodIfthecase study stated important driverswe recorded all of these im-portant drivers mentioned in each publication (some publicationsdiscussedmultipledrivers)Otherwisewetreatedeachofthemen-tioneddrivers asmajor drivers In order to generalize the driverswecombinedsomeofthedriversthatwerecloselyrelatedDetailedinformationisprovidedinTable1
22emsp|emspSpatial proxy data of the drivers for forest change
Afteridentifyingthedriversfromthesynthesisofcasestudieswecollected 16 biophysical and 17 socioeconomic proxy datasets torepresent thesedrivers (Table2)Allof theseproxiesare spatiallygridded data which are used to build the quantitative relation-ships between the drivers and the forest change These proxieshavedifferent spatial and temporal resolutions (Table2)WekepttheoriginalspatialresolutionofeachdriverForexampleweused
TA B L E 1 emspGeneralizationoftheidentifieddriversandtheircorrespondingproxydata
Drivers identified from case studies Collected proxy data Remarks
bull Terrain(topographicalconditions) Terrainindex
bull Soilandotherenvironmentalconditions
Soilchemicalcompositiondepthdrainagefertilityandtexture
bull Wateravailability Distancetowaterbodies
bull Climate Meanrateofchangeandstandarddeviationofannual precipitation
Meanrateofchangeandstandarddeviationofannual temperature
Meanannualpotentialevapotranspiration
bull Fire Meanburnedareafraction
bull Othernaturaldisaster Distancetolandslideevents
bull Populationbull Labor
Meanandrateofchangeinurbanpopulationdensity ThelaboramountisdirectlyrelatedtopopulationMeanandrateofchangeinruralpopulationdensity
bull Urbanization Meanandrateofchangeinurbanareafraction
bull Migration Migration
bull Livestockbull Grazing
ChickenCattleSheepPigGoatandDuckcounts Grazingactivitiesarecorrelatedwithlivestock
bull Accessibilitybull Transportationbull Infrastructure
Marketaccessibilityindex Marketaccessibilityindexconsidersaccessibil-ityandinfrastructureparticularlythetransportationcondition
bull Marketinfluence(price)bull Economy developmentbull Plantationbull Agroforestbull Agricultureexpansion
GDP per capita PlantationagroforestdevelopmentandagricultureexpansionarerelatedtomarketandeconomyAllofthemcanbereflectedbyGDP per capita
bull Mining(industry) Distancetominingfacilities
bull Povertybull Incomedependencyonforestbull Livelihoodbull Benefit
Povertyindex Thedependencyonforestlivelihoodandbenefitareallcloselyrelatedwithpoverty
bull Shiftingagriculture(swidden)bull Forestmanagement(policy)bull Globalization(internationaltrade)bull Fuelwood(loggingandcharcoal)bull Tourismbull Culturebull Technologybull Farmsizebull Natural regenerationbull Educationorsocialawarenessbull Aquaculture
Spatiallyexplicitproxydataarenotavailable
4emsp |emsp emspensp XU et al
05degtimes05degCRUTSclimatedatawhichisrelativelycoarsecomparedto the01degtimes01deg resolutionof the forest changedata Thereforewhenweextractedtheclimatedatavaluesusingthecoordinatesofthe01degtimes01deggridsofLULCCdatathe01degtimes01deggridsinthesame05degtimes05deggridcellallhadthesamevalueAsaresultouranalysismaymisssomedetaileddriverinformationatsmallscalesHoweverbecausetheregioninvestigatedismuchlargerthan05degtimes05degtheclimatedatastillhavespatialgradientsfortheGWRanalysis
Iftheproxydatasetsdonotchangewithtime(egterrainsoilproperties)orareavailableintheliteratureforonlyonetimeperiod(eg distances towaterbodies landslideevents andmining facili-tiesaswellasmarketaccessibilityindex)weusethemdirectlyinthequantitativeanalysis(step6)Iftheproxiesaretime-seriesdata(egburned area fraction precipitation temperature rural and urbanpopulationdensityandurbanareafraction)weusethemeansanddynamics(rateofchangeandstandarddeviation)oftheminstep6Migrationdataareavailableatthedecadaltimescalecoveringfrom1970to2000weusedthe1990to2000dataasitisclosesttoourstudyperiodinstep6
23emsp|emspLand use data
We used the European Space Agency Climate Change Initiative(ESA-CCI) satellite data for land cover dynamics of forest and ag-riculturallandfrom1992to2015Theoriginalproductcategorizedlandcoverinto22classes(level1)(DefournyMoreauampBontemps2017)OurstudyusedthecorrespondingIPCCclasses(Defournyetal2017)ofagriculturallandandforestWederivethespatialdataof the forest changebyoverlaying the1992and2015 land covermapsfromtheESA-CCIdata(300times300mresolution)Thenweag-gregatedtheforestchangemapsto01degtimes01deg(~10kmtimes10km)anddetermined the fractionsof the changedarea from forest toagri-cultural landandagricultural landtoforest ineach01degtimes01deggrid(FigureS1)The01degtimes01degresolutionrepresentsatradeoffbetweenthe finer resolution of the LULCC data (300m) and the relativelycoarser resolution data for biophysical and socioeconomic proxydata (from01degtimes01degto05degtimes05deg)Wealsocalculatedthecoun-try-specificareasofthesetwochanges
Theareaschangedbetween1992and2015duetotheconver-sion of forest to agricultural land and vice versa are 185468km2 and89398km2whencalculatingbasedon1992and2015dataOntheotherhandtheareaschangedare191277km2and97298km2 whenaccumulatingtheyearlyareachangesofthesetwolandchangeactivitiesover theperiod1992ndash2015 (FigureS3)Theareasbasedtheformermethodare9696and9188oftheresultsfromthelattermethodindicatingthatforestchangedataestimatedbasedon1992and2015yearsofdataareabletocapturethemajorinforma-tionforthistimeperiod
24emsp|emspHotspot analysis
Thepurposeofthehotspotanalysiswastoexcludetheregionswithsmaller areas of forest and its change in the driver analysis The
inclusionofsuchlowvalueregionsmightdilutetheimportanceofthemajordriversWeusedtheGetis‐Ord Gianalysistechniquetoidentifythehotspotregionsforthechangesfromforesttoagricul-turallandandagriculturallandtoforestThistechniqueidentifiesstatistically significant spatial clustersofhigh (hot spots) and low(cold spots) valuesof changes (OrdampGetis 1995)A statisticallysignificanthotspothasagreaterareaofLULCCandissurroundedby other regionswith great areas of LULCCWemarked regionswithgt3 standarddeviations (at 99confidence level) as hotspotregionsWe analyzed the hotspot regions in each country at thedistrictlevel
25emsp|emspPrinciple component analysis (PCA)
Multi-collinearity isacommonproblemin landchangemodelingwhere one or more explanatory (or proxy) data are dependentoneachotherAhighdegreeofmulti-collinearityresults inhighstandarderrorsandspuriouscoefficientestimatesWeemployedthePCAmethodtoaccountforthemulti-collinearityWeselectedthePCswithcumulativecontributionrates(tothetotalvariationof all proxy data) greater than 85 (Deng Wang Deng amp Qi2008)
26emsp|emspGeographically weighted regression (GWR)
The GWRmodel constructs a distinct relationship between eachLULCC pixel and concomitant driver proxy data by incorporatingpixelsfallingwithinacertainbandwidthofthecenterLULCCpixel(Charlton Fotheringham amp Brunsdon 2009) Here we used theadaptiveGaussiankerneltodeterminethebandwidthandthelocalextent to estimate the regression coefficients The optimal band-widthsizeofthekernelwasdeterminedbymeansofcomparisonofAkaike InformationCriterion (AIC)withdifferentbandwidth sizesWeconductedtheGWRintheidentifiedhotspotdistrictsofeachcountry
27emsp|emspJohnsons relative weight (JRW)
The GWR results can represent the positive or negative impactof each individual proxy driver aswell as its relativemagnitudeHowever it isdifficulttoevaluatethecombinedeffectsofdiffer-entdrivercategories(seethesixcategoriesinTable2)InordertogeneralizetheforestchangedriversfromdifferentcategoriesweusedtheJohnsonsRelativeWeight(JRW)methodtoquantifytherelativeimportanceofeachindividualproxyandthensummeduptheimportancecoefficientsofalldrivercategoriesTheJRWanaly-sisfirstgeneratesaseriesoforthogonalvariablesthatarethelinearcombinationsofalloriginalproxydataandthenconductsthere-gressionanalysisbyusingthegeneratedorthogonalvariables(thisprocedureisthesameasthePCAanalysisinstep5)ThentheEqS5isusedtodeterminetherelativeimportanceoftheoriginalproxydata (Chao Zhao Kupper amp Nylander-French 2008 Johnson2000)
emspensp emsp | emsp5XU et al
TAB
LE 2
emspProxyvariablesoftheidentifydrivers
C
ateg
ory
Prox
y va
riabl
eTe
mpo
ral
reso
lutio
nSp
atia
l res
olut
ion
Uni
tSo
urce
Biophysical
variables
ITerrainsoil
and
wat
er1
Terrain
Constant
5primetimes5prime
(~10kmtimes10km)
Categoricaldatainto7
gradientclasses
GlobalAgro-ecologicalZones
(GAEZ)v30(FischerNachtergaele
ampPrieler2012)
2-6
Soilchemicalcomposition
depthdrainagefertilityand
texture
7Distancetowaterbodies
5primetimes5prime
kmCalculatedfromGlobalLakesand
WetlandsDatabase(GLWD)level2
data(LehnerampDoll2004)
IIC
limat
e8-10
Meanrateofchangeand
standarddeviationofannual
prec
ipita
tion
Yearly
05deg times
05
degdeg Cdeg C
yea
rClimaticResearchUnit(CRU)TS
401
11-13
Meanrateofchangeand
standarddeviationofannual
tem
pera
ture
mmmmyear
14
Meanannualpotential
evapotranspiration
mm
IIIN
atur
al
disaster
15
Meanburnedareafraction
Yearly(1997ndash2014)
025
deg times 0
25deg
GlobalFireEmissionsDatabase41
(GiglioRandersonampWerf2013)
16
Distancetolandslideevents
Constant
5primetimes5prime
kmCalculatedfromGlobalLandslide
Catalog(KirschbaumAdlerHong
HillampLerner-Lam2010)
Soci
oeco
nom
ic
variables
IVPo
pula
tion
and
urbanization
17-18
Meanandrateofchangein
urbanpopulationdensity
Yearly
5primetimes5prime
inhabitantskm
2 inhabitantskm
2 middotyea
rHYDE32(KleinGoldewijkBeusen
DoelmanampStehfest2017)
19-20
Meanandrateofchangein
ruralpopulationdensity
21-22
Meanandrateofchangein
urba
n ar
ea fr
actio
nyear
23
Migration
Dec
adal
(1970ndash2000)
05deg times
05
degNumberofmigrants
km2
GlobalEstimatedNetMigration
GridsByDecadev1(deSherbinin
etal2012)
VLivestock
24-29
ChickenCattleSheepPig
GoatandDuckcounts
Constant
1 km
times 1
km
Headkm
2GriddedLivestockoftheWorld
(GLW)version2(Robinsonetal
2014)
(Con
tinue
s)
6emsp |emsp emspensp XU et al
3emsp |emspRESULTS
31emsp|emspSynthesis of the case study
The case studies regarding the drivers of forest loss and gain arespreadacrossallcountriesandhavediversespatialscalesfromvil-lagetonationalscale(FigureS5)Thereare173studies(81ofthetotalcollectedstudies)studyingdriversbasedonfieldsurveys in-terviewsandliteraturereviewsMeanwhile40studies(19ofthecollectedstudies)haveusedquantitativeapproachessuchaslinearregressionlogisticregressionsystemdynamicsmodelingandcel-lularautomatamodelingtoreveal thedriversofLULCCHowevermostofthesequantitativestudiesareatsmallerspatialscales(vil-lage to state and province level) (31 studies) There are only ninequantitativestudiesatthenationalscale
We identify the drivers of forest change from collected casestudies(Figure1andTable1)Thebiophysicaldriversincludeterrainor topographicalconditions (egelevationaltitudeand theslopeof land) soil conditions (eg soil fertility texture and moisture)wateravailabilityclimate(egtemperatureprecipitationandtheirchanges)fireandothernaturaldisasters(eglandslide)Thesocio-economicdriversmainlyrelatetopopulationgrowthurbanizationlivestockorgrazing(egovergrazing)marketinfluenceoreconomydevelopment(egGDPpercapitaaccesstomarket)plantationoragroforestdevelopment agricultural expansionminingand indus-try accessibilityand infrastructure (eg transportationelectricityconnection)loggingandfuelwoodandpoverty
Fire and other natural disasters and terrain are the most fre-quentlymentionedbiophysicaldriversofforestarealoss(deforesta-tion) For example in India and Indonesia the two countrieswiththelargestforestareaamajorityofstudiessuggestterrainandfireand other natural disasters as the major biophysical drivers Oursynthesisofsite-levelcasestudiesalsoidentifiespopulationplanta-tionagriculturalexpansionaccessibilityandinfrastructureaswellasfuelwoodandloggingastheimportantsocioeconomicdriversofdeforestation
There are fewer studies on the drivers of forest area gain (af-forestationandreforestation)thanondeforestationAmongthemmoststudieshavefocusedonsocioeconomicdriversThefrequentlymentioned socioeconomic drivers include accessibility povertyeconomy and livestock and biophysical drivers include climateandterrainThemajordriversof forestchangeatdifferentspatialandtemporalscalesaredescribedinmoredetailinSupplementarySectionTextS2FiguresS6andS7
32emsp|emspForest change
Thecountry-specificandspatialdistributionofconvertingfromfor-esttoagriculturallandandagriculturallandtoforestoverthetimeperiod1992ndash2015isshowninFigure2andFigureS4MalaysiaandPhilippinesexperienced largeareasofdeforestationaswellasaf-forestation over 1992ndash2015 Cambodia and Singapore had moredeforestationthanafforestation
Cat
egor
yPr
oxy
varia
ble
Tem
pora
l re
solu
tion
Spat
ial r
esol
utio
nU
nit
Sour
ce
VI
Econ
omy
30
Marketaccessibilityindex
Constant
1 km
times 1
km
-VerburgEllisandLetourneau
(2011)
31
GD
P pe
r cap
itaDecadal(1980
ndash2010)
05deg times
05
degUSD2005
Globaldatasetofgriddedpopulation
andGDPscenariosMurakamiand
Yamagata(2016)
32
Distancetominingfacilities
Constant
5primetimes5prime
kmCalculatedfromMineralResources
On-LineSpatialDatabyUSGS
33
Povertyindex
Yearly(1992ndash2013)
5rsquotimes5rsquo
-Calculatedfrompopulation(HYDE
32)(KleinGoldewijkBeusen
DoelmanampStehfest2017)and
Nighttimelight(Version4
DMSP-OLSNighttimeLightsTime
Series)byfollowingthemethod
developedbyGhoshAnderson
ElvidgeandSutton(2013)
TAB
LE 2
emsp(Continued)
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp3XU et al
coordinatesofthegeometriccenterforthestudyareaatdifferentspatialscale)timespanandfrequencyofthedriversmentionedinthepublications
WeidentifiedthemajordriversbythefollowingmethodIfthecase study stated important driverswe recorded all of these im-portant drivers mentioned in each publication (some publicationsdiscussedmultipledrivers)Otherwisewetreatedeachofthemen-tioneddrivers asmajor drivers In order to generalize the driverswecombinedsomeofthedriversthatwerecloselyrelatedDetailedinformationisprovidedinTable1
22emsp|emspSpatial proxy data of the drivers for forest change
Afteridentifyingthedriversfromthesynthesisofcasestudieswecollected 16 biophysical and 17 socioeconomic proxy datasets torepresent thesedrivers (Table2)Allof theseproxiesare spatiallygridded data which are used to build the quantitative relation-ships between the drivers and the forest change These proxieshavedifferent spatial and temporal resolutions (Table2)WekepttheoriginalspatialresolutionofeachdriverForexampleweused
TA B L E 1 emspGeneralizationoftheidentifieddriversandtheircorrespondingproxydata
Drivers identified from case studies Collected proxy data Remarks
bull Terrain(topographicalconditions) Terrainindex
bull Soilandotherenvironmentalconditions
Soilchemicalcompositiondepthdrainagefertilityandtexture
bull Wateravailability Distancetowaterbodies
bull Climate Meanrateofchangeandstandarddeviationofannual precipitation
Meanrateofchangeandstandarddeviationofannual temperature
Meanannualpotentialevapotranspiration
bull Fire Meanburnedareafraction
bull Othernaturaldisaster Distancetolandslideevents
bull Populationbull Labor
Meanandrateofchangeinurbanpopulationdensity ThelaboramountisdirectlyrelatedtopopulationMeanandrateofchangeinruralpopulationdensity
bull Urbanization Meanandrateofchangeinurbanareafraction
bull Migration Migration
bull Livestockbull Grazing
ChickenCattleSheepPigGoatandDuckcounts Grazingactivitiesarecorrelatedwithlivestock
bull Accessibilitybull Transportationbull Infrastructure
Marketaccessibilityindex Marketaccessibilityindexconsidersaccessibil-ityandinfrastructureparticularlythetransportationcondition
bull Marketinfluence(price)bull Economy developmentbull Plantationbull Agroforestbull Agricultureexpansion
GDP per capita PlantationagroforestdevelopmentandagricultureexpansionarerelatedtomarketandeconomyAllofthemcanbereflectedbyGDP per capita
bull Mining(industry) Distancetominingfacilities
bull Povertybull Incomedependencyonforestbull Livelihoodbull Benefit
Povertyindex Thedependencyonforestlivelihoodandbenefitareallcloselyrelatedwithpoverty
bull Shiftingagriculture(swidden)bull Forestmanagement(policy)bull Globalization(internationaltrade)bull Fuelwood(loggingandcharcoal)bull Tourismbull Culturebull Technologybull Farmsizebull Natural regenerationbull Educationorsocialawarenessbull Aquaculture
Spatiallyexplicitproxydataarenotavailable
4emsp |emsp emspensp XU et al
05degtimes05degCRUTSclimatedatawhichisrelativelycoarsecomparedto the01degtimes01deg resolutionof the forest changedata Thereforewhenweextractedtheclimatedatavaluesusingthecoordinatesofthe01degtimes01deggridsofLULCCdatathe01degtimes01deggridsinthesame05degtimes05deggridcellallhadthesamevalueAsaresultouranalysismaymisssomedetaileddriverinformationatsmallscalesHoweverbecausetheregioninvestigatedismuchlargerthan05degtimes05degtheclimatedatastillhavespatialgradientsfortheGWRanalysis
Iftheproxydatasetsdonotchangewithtime(egterrainsoilproperties)orareavailableintheliteratureforonlyonetimeperiod(eg distances towaterbodies landslideevents andmining facili-tiesaswellasmarketaccessibilityindex)weusethemdirectlyinthequantitativeanalysis(step6)Iftheproxiesaretime-seriesdata(egburned area fraction precipitation temperature rural and urbanpopulationdensityandurbanareafraction)weusethemeansanddynamics(rateofchangeandstandarddeviation)oftheminstep6Migrationdataareavailableatthedecadaltimescalecoveringfrom1970to2000weusedthe1990to2000dataasitisclosesttoourstudyperiodinstep6
23emsp|emspLand use data
We used the European Space Agency Climate Change Initiative(ESA-CCI) satellite data for land cover dynamics of forest and ag-riculturallandfrom1992to2015Theoriginalproductcategorizedlandcoverinto22classes(level1)(DefournyMoreauampBontemps2017)OurstudyusedthecorrespondingIPCCclasses(Defournyetal2017)ofagriculturallandandforestWederivethespatialdataof the forest changebyoverlaying the1992and2015 land covermapsfromtheESA-CCIdata(300times300mresolution)Thenweag-gregatedtheforestchangemapsto01degtimes01deg(~10kmtimes10km)anddetermined the fractionsof the changedarea from forest toagri-cultural landandagricultural landtoforest ineach01degtimes01deggrid(FigureS1)The01degtimes01degresolutionrepresentsatradeoffbetweenthe finer resolution of the LULCC data (300m) and the relativelycoarser resolution data for biophysical and socioeconomic proxydata (from01degtimes01degto05degtimes05deg)Wealsocalculatedthecoun-try-specificareasofthesetwochanges
Theareaschangedbetween1992and2015duetotheconver-sion of forest to agricultural land and vice versa are 185468km2 and89398km2whencalculatingbasedon1992and2015dataOntheotherhandtheareaschangedare191277km2and97298km2 whenaccumulatingtheyearlyareachangesofthesetwolandchangeactivitiesover theperiod1992ndash2015 (FigureS3)Theareasbasedtheformermethodare9696and9188oftheresultsfromthelattermethodindicatingthatforestchangedataestimatedbasedon1992and2015yearsofdataareabletocapturethemajorinforma-tionforthistimeperiod
24emsp|emspHotspot analysis
Thepurposeofthehotspotanalysiswastoexcludetheregionswithsmaller areas of forest and its change in the driver analysis The
inclusionofsuchlowvalueregionsmightdilutetheimportanceofthemajordriversWeusedtheGetis‐Ord Gianalysistechniquetoidentifythehotspotregionsforthechangesfromforesttoagricul-turallandandagriculturallandtoforestThistechniqueidentifiesstatistically significant spatial clustersofhigh (hot spots) and low(cold spots) valuesof changes (OrdampGetis 1995)A statisticallysignificanthotspothasagreaterareaofLULCCandissurroundedby other regionswith great areas of LULCCWemarked regionswithgt3 standarddeviations (at 99confidence level) as hotspotregionsWe analyzed the hotspot regions in each country at thedistrictlevel
25emsp|emspPrinciple component analysis (PCA)
Multi-collinearity isacommonproblemin landchangemodelingwhere one or more explanatory (or proxy) data are dependentoneachotherAhighdegreeofmulti-collinearityresults inhighstandarderrorsandspuriouscoefficientestimatesWeemployedthePCAmethodtoaccountforthemulti-collinearityWeselectedthePCswithcumulativecontributionrates(tothetotalvariationof all proxy data) greater than 85 (Deng Wang Deng amp Qi2008)
26emsp|emspGeographically weighted regression (GWR)
The GWRmodel constructs a distinct relationship between eachLULCC pixel and concomitant driver proxy data by incorporatingpixelsfallingwithinacertainbandwidthofthecenterLULCCpixel(Charlton Fotheringham amp Brunsdon 2009) Here we used theadaptiveGaussiankerneltodeterminethebandwidthandthelocalextent to estimate the regression coefficients The optimal band-widthsizeofthekernelwasdeterminedbymeansofcomparisonofAkaike InformationCriterion (AIC)withdifferentbandwidth sizesWeconductedtheGWRintheidentifiedhotspotdistrictsofeachcountry
27emsp|emspJohnsons relative weight (JRW)
The GWR results can represent the positive or negative impactof each individual proxy driver aswell as its relativemagnitudeHowever it isdifficulttoevaluatethecombinedeffectsofdiffer-entdrivercategories(seethesixcategoriesinTable2)InordertogeneralizetheforestchangedriversfromdifferentcategoriesweusedtheJohnsonsRelativeWeight(JRW)methodtoquantifytherelativeimportanceofeachindividualproxyandthensummeduptheimportancecoefficientsofalldrivercategoriesTheJRWanaly-sisfirstgeneratesaseriesoforthogonalvariablesthatarethelinearcombinationsofalloriginalproxydataandthenconductsthere-gressionanalysisbyusingthegeneratedorthogonalvariables(thisprocedureisthesameasthePCAanalysisinstep5)ThentheEqS5isusedtodeterminetherelativeimportanceoftheoriginalproxydata (Chao Zhao Kupper amp Nylander-French 2008 Johnson2000)
emspensp emsp | emsp5XU et al
TAB
LE 2
emspProxyvariablesoftheidentifydrivers
C
ateg
ory
Prox
y va
riabl
eTe
mpo
ral
reso
lutio
nSp
atia
l res
olut
ion
Uni
tSo
urce
Biophysical
variables
ITerrainsoil
and
wat
er1
Terrain
Constant
5primetimes5prime
(~10kmtimes10km)
Categoricaldatainto7
gradientclasses
GlobalAgro-ecologicalZones
(GAEZ)v30(FischerNachtergaele
ampPrieler2012)
2-6
Soilchemicalcomposition
depthdrainagefertilityand
texture
7Distancetowaterbodies
5primetimes5prime
kmCalculatedfromGlobalLakesand
WetlandsDatabase(GLWD)level2
data(LehnerampDoll2004)
IIC
limat
e8-10
Meanrateofchangeand
standarddeviationofannual
prec
ipita
tion
Yearly
05deg times
05
degdeg Cdeg C
yea
rClimaticResearchUnit(CRU)TS
401
11-13
Meanrateofchangeand
standarddeviationofannual
tem
pera
ture
mmmmyear
14
Meanannualpotential
evapotranspiration
mm
IIIN
atur
al
disaster
15
Meanburnedareafraction
Yearly(1997ndash2014)
025
deg times 0
25deg
GlobalFireEmissionsDatabase41
(GiglioRandersonampWerf2013)
16
Distancetolandslideevents
Constant
5primetimes5prime
kmCalculatedfromGlobalLandslide
Catalog(KirschbaumAdlerHong
HillampLerner-Lam2010)
Soci
oeco
nom
ic
variables
IVPo
pula
tion
and
urbanization
17-18
Meanandrateofchangein
urbanpopulationdensity
Yearly
5primetimes5prime
inhabitantskm
2 inhabitantskm
2 middotyea
rHYDE32(KleinGoldewijkBeusen
DoelmanampStehfest2017)
19-20
Meanandrateofchangein
ruralpopulationdensity
21-22
Meanandrateofchangein
urba
n ar
ea fr
actio
nyear
23
Migration
Dec
adal
(1970ndash2000)
05deg times
05
degNumberofmigrants
km2
GlobalEstimatedNetMigration
GridsByDecadev1(deSherbinin
etal2012)
VLivestock
24-29
ChickenCattleSheepPig
GoatandDuckcounts
Constant
1 km
times 1
km
Headkm
2GriddedLivestockoftheWorld
(GLW)version2(Robinsonetal
2014)
(Con
tinue
s)
6emsp |emsp emspensp XU et al
3emsp |emspRESULTS
31emsp|emspSynthesis of the case study
The case studies regarding the drivers of forest loss and gain arespreadacrossallcountriesandhavediversespatialscalesfromvil-lagetonationalscale(FigureS5)Thereare173studies(81ofthetotalcollectedstudies)studyingdriversbasedonfieldsurveys in-terviewsandliteraturereviewsMeanwhile40studies(19ofthecollectedstudies)haveusedquantitativeapproachessuchaslinearregressionlogisticregressionsystemdynamicsmodelingandcel-lularautomatamodelingtoreveal thedriversofLULCCHowevermostofthesequantitativestudiesareatsmallerspatialscales(vil-lage to state and province level) (31 studies) There are only ninequantitativestudiesatthenationalscale
We identify the drivers of forest change from collected casestudies(Figure1andTable1)Thebiophysicaldriversincludeterrainor topographicalconditions (egelevationaltitudeand theslopeof land) soil conditions (eg soil fertility texture and moisture)wateravailabilityclimate(egtemperatureprecipitationandtheirchanges)fireandothernaturaldisasters(eglandslide)Thesocio-economicdriversmainlyrelatetopopulationgrowthurbanizationlivestockorgrazing(egovergrazing)marketinfluenceoreconomydevelopment(egGDPpercapitaaccesstomarket)plantationoragroforestdevelopment agricultural expansionminingand indus-try accessibilityand infrastructure (eg transportationelectricityconnection)loggingandfuelwoodandpoverty
Fire and other natural disasters and terrain are the most fre-quentlymentionedbiophysicaldriversofforestarealoss(deforesta-tion) For example in India and Indonesia the two countrieswiththelargestforestareaamajorityofstudiessuggestterrainandfireand other natural disasters as the major biophysical drivers Oursynthesisofsite-levelcasestudiesalsoidentifiespopulationplanta-tionagriculturalexpansionaccessibilityandinfrastructureaswellasfuelwoodandloggingastheimportantsocioeconomicdriversofdeforestation
There are fewer studies on the drivers of forest area gain (af-forestationandreforestation)thanondeforestationAmongthemmoststudieshavefocusedonsocioeconomicdriversThefrequentlymentioned socioeconomic drivers include accessibility povertyeconomy and livestock and biophysical drivers include climateandterrainThemajordriversof forestchangeatdifferentspatialandtemporalscalesaredescribedinmoredetailinSupplementarySectionTextS2FiguresS6andS7
32emsp|emspForest change
Thecountry-specificandspatialdistributionofconvertingfromfor-esttoagriculturallandandagriculturallandtoforestoverthetimeperiod1992ndash2015isshowninFigure2andFigureS4MalaysiaandPhilippinesexperienced largeareasofdeforestationaswellasaf-forestation over 1992ndash2015 Cambodia and Singapore had moredeforestationthanafforestation
Cat
egor
yPr
oxy
varia
ble
Tem
pora
l re
solu
tion
Spat
ial r
esol
utio
nU
nit
Sour
ce
VI
Econ
omy
30
Marketaccessibilityindex
Constant
1 km
times 1
km
-VerburgEllisandLetourneau
(2011)
31
GD
P pe
r cap
itaDecadal(1980
ndash2010)
05deg times
05
degUSD2005
Globaldatasetofgriddedpopulation
andGDPscenariosMurakamiand
Yamagata(2016)
32
Distancetominingfacilities
Constant
5primetimes5prime
kmCalculatedfromMineralResources
On-LineSpatialDatabyUSGS
33
Povertyindex
Yearly(1992ndash2013)
5rsquotimes5rsquo
-Calculatedfrompopulation(HYDE
32)(KleinGoldewijkBeusen
DoelmanampStehfest2017)and
Nighttimelight(Version4
DMSP-OLSNighttimeLightsTime
Series)byfollowingthemethod
developedbyGhoshAnderson
ElvidgeandSutton(2013)
TAB
LE 2
emsp(Continued)
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
4emsp |emsp emspensp XU et al
05degtimes05degCRUTSclimatedatawhichisrelativelycoarsecomparedto the01degtimes01deg resolutionof the forest changedata Thereforewhenweextractedtheclimatedatavaluesusingthecoordinatesofthe01degtimes01deggridsofLULCCdatathe01degtimes01deggridsinthesame05degtimes05deggridcellallhadthesamevalueAsaresultouranalysismaymisssomedetaileddriverinformationatsmallscalesHoweverbecausetheregioninvestigatedismuchlargerthan05degtimes05degtheclimatedatastillhavespatialgradientsfortheGWRanalysis
Iftheproxydatasetsdonotchangewithtime(egterrainsoilproperties)orareavailableintheliteratureforonlyonetimeperiod(eg distances towaterbodies landslideevents andmining facili-tiesaswellasmarketaccessibilityindex)weusethemdirectlyinthequantitativeanalysis(step6)Iftheproxiesaretime-seriesdata(egburned area fraction precipitation temperature rural and urbanpopulationdensityandurbanareafraction)weusethemeansanddynamics(rateofchangeandstandarddeviation)oftheminstep6Migrationdataareavailableatthedecadaltimescalecoveringfrom1970to2000weusedthe1990to2000dataasitisclosesttoourstudyperiodinstep6
23emsp|emspLand use data
We used the European Space Agency Climate Change Initiative(ESA-CCI) satellite data for land cover dynamics of forest and ag-riculturallandfrom1992to2015Theoriginalproductcategorizedlandcoverinto22classes(level1)(DefournyMoreauampBontemps2017)OurstudyusedthecorrespondingIPCCclasses(Defournyetal2017)ofagriculturallandandforestWederivethespatialdataof the forest changebyoverlaying the1992and2015 land covermapsfromtheESA-CCIdata(300times300mresolution)Thenweag-gregatedtheforestchangemapsto01degtimes01deg(~10kmtimes10km)anddetermined the fractionsof the changedarea from forest toagri-cultural landandagricultural landtoforest ineach01degtimes01deggrid(FigureS1)The01degtimes01degresolutionrepresentsatradeoffbetweenthe finer resolution of the LULCC data (300m) and the relativelycoarser resolution data for biophysical and socioeconomic proxydata (from01degtimes01degto05degtimes05deg)Wealsocalculatedthecoun-try-specificareasofthesetwochanges
Theareaschangedbetween1992and2015duetotheconver-sion of forest to agricultural land and vice versa are 185468km2 and89398km2whencalculatingbasedon1992and2015dataOntheotherhandtheareaschangedare191277km2and97298km2 whenaccumulatingtheyearlyareachangesofthesetwolandchangeactivitiesover theperiod1992ndash2015 (FigureS3)Theareasbasedtheformermethodare9696and9188oftheresultsfromthelattermethodindicatingthatforestchangedataestimatedbasedon1992and2015yearsofdataareabletocapturethemajorinforma-tionforthistimeperiod
24emsp|emspHotspot analysis
Thepurposeofthehotspotanalysiswastoexcludetheregionswithsmaller areas of forest and its change in the driver analysis The
inclusionofsuchlowvalueregionsmightdilutetheimportanceofthemajordriversWeusedtheGetis‐Ord Gianalysistechniquetoidentifythehotspotregionsforthechangesfromforesttoagricul-turallandandagriculturallandtoforestThistechniqueidentifiesstatistically significant spatial clustersofhigh (hot spots) and low(cold spots) valuesof changes (OrdampGetis 1995)A statisticallysignificanthotspothasagreaterareaofLULCCandissurroundedby other regionswith great areas of LULCCWemarked regionswithgt3 standarddeviations (at 99confidence level) as hotspotregionsWe analyzed the hotspot regions in each country at thedistrictlevel
25emsp|emspPrinciple component analysis (PCA)
Multi-collinearity isacommonproblemin landchangemodelingwhere one or more explanatory (or proxy) data are dependentoneachotherAhighdegreeofmulti-collinearityresults inhighstandarderrorsandspuriouscoefficientestimatesWeemployedthePCAmethodtoaccountforthemulti-collinearityWeselectedthePCswithcumulativecontributionrates(tothetotalvariationof all proxy data) greater than 85 (Deng Wang Deng amp Qi2008)
26emsp|emspGeographically weighted regression (GWR)
The GWRmodel constructs a distinct relationship between eachLULCC pixel and concomitant driver proxy data by incorporatingpixelsfallingwithinacertainbandwidthofthecenterLULCCpixel(Charlton Fotheringham amp Brunsdon 2009) Here we used theadaptiveGaussiankerneltodeterminethebandwidthandthelocalextent to estimate the regression coefficients The optimal band-widthsizeofthekernelwasdeterminedbymeansofcomparisonofAkaike InformationCriterion (AIC)withdifferentbandwidth sizesWeconductedtheGWRintheidentifiedhotspotdistrictsofeachcountry
27emsp|emspJohnsons relative weight (JRW)
The GWR results can represent the positive or negative impactof each individual proxy driver aswell as its relativemagnitudeHowever it isdifficulttoevaluatethecombinedeffectsofdiffer-entdrivercategories(seethesixcategoriesinTable2)InordertogeneralizetheforestchangedriversfromdifferentcategoriesweusedtheJohnsonsRelativeWeight(JRW)methodtoquantifytherelativeimportanceofeachindividualproxyandthensummeduptheimportancecoefficientsofalldrivercategoriesTheJRWanaly-sisfirstgeneratesaseriesoforthogonalvariablesthatarethelinearcombinationsofalloriginalproxydataandthenconductsthere-gressionanalysisbyusingthegeneratedorthogonalvariables(thisprocedureisthesameasthePCAanalysisinstep5)ThentheEqS5isusedtodeterminetherelativeimportanceoftheoriginalproxydata (Chao Zhao Kupper amp Nylander-French 2008 Johnson2000)
emspensp emsp | emsp5XU et al
TAB
LE 2
emspProxyvariablesoftheidentifydrivers
C
ateg
ory
Prox
y va
riabl
eTe
mpo
ral
reso
lutio
nSp
atia
l res
olut
ion
Uni
tSo
urce
Biophysical
variables
ITerrainsoil
and
wat
er1
Terrain
Constant
5primetimes5prime
(~10kmtimes10km)
Categoricaldatainto7
gradientclasses
GlobalAgro-ecologicalZones
(GAEZ)v30(FischerNachtergaele
ampPrieler2012)
2-6
Soilchemicalcomposition
depthdrainagefertilityand
texture
7Distancetowaterbodies
5primetimes5prime
kmCalculatedfromGlobalLakesand
WetlandsDatabase(GLWD)level2
data(LehnerampDoll2004)
IIC
limat
e8-10
Meanrateofchangeand
standarddeviationofannual
prec
ipita
tion
Yearly
05deg times
05
degdeg Cdeg C
yea
rClimaticResearchUnit(CRU)TS
401
11-13
Meanrateofchangeand
standarddeviationofannual
tem
pera
ture
mmmmyear
14
Meanannualpotential
evapotranspiration
mm
IIIN
atur
al
disaster
15
Meanburnedareafraction
Yearly(1997ndash2014)
025
deg times 0
25deg
GlobalFireEmissionsDatabase41
(GiglioRandersonampWerf2013)
16
Distancetolandslideevents
Constant
5primetimes5prime
kmCalculatedfromGlobalLandslide
Catalog(KirschbaumAdlerHong
HillampLerner-Lam2010)
Soci
oeco
nom
ic
variables
IVPo
pula
tion
and
urbanization
17-18
Meanandrateofchangein
urbanpopulationdensity
Yearly
5primetimes5prime
inhabitantskm
2 inhabitantskm
2 middotyea
rHYDE32(KleinGoldewijkBeusen
DoelmanampStehfest2017)
19-20
Meanandrateofchangein
ruralpopulationdensity
21-22
Meanandrateofchangein
urba
n ar
ea fr
actio
nyear
23
Migration
Dec
adal
(1970ndash2000)
05deg times
05
degNumberofmigrants
km2
GlobalEstimatedNetMigration
GridsByDecadev1(deSherbinin
etal2012)
VLivestock
24-29
ChickenCattleSheepPig
GoatandDuckcounts
Constant
1 km
times 1
km
Headkm
2GriddedLivestockoftheWorld
(GLW)version2(Robinsonetal
2014)
(Con
tinue
s)
6emsp |emsp emspensp XU et al
3emsp |emspRESULTS
31emsp|emspSynthesis of the case study
The case studies regarding the drivers of forest loss and gain arespreadacrossallcountriesandhavediversespatialscalesfromvil-lagetonationalscale(FigureS5)Thereare173studies(81ofthetotalcollectedstudies)studyingdriversbasedonfieldsurveys in-terviewsandliteraturereviewsMeanwhile40studies(19ofthecollectedstudies)haveusedquantitativeapproachessuchaslinearregressionlogisticregressionsystemdynamicsmodelingandcel-lularautomatamodelingtoreveal thedriversofLULCCHowevermostofthesequantitativestudiesareatsmallerspatialscales(vil-lage to state and province level) (31 studies) There are only ninequantitativestudiesatthenationalscale
We identify the drivers of forest change from collected casestudies(Figure1andTable1)Thebiophysicaldriversincludeterrainor topographicalconditions (egelevationaltitudeand theslopeof land) soil conditions (eg soil fertility texture and moisture)wateravailabilityclimate(egtemperatureprecipitationandtheirchanges)fireandothernaturaldisasters(eglandslide)Thesocio-economicdriversmainlyrelatetopopulationgrowthurbanizationlivestockorgrazing(egovergrazing)marketinfluenceoreconomydevelopment(egGDPpercapitaaccesstomarket)plantationoragroforestdevelopment agricultural expansionminingand indus-try accessibilityand infrastructure (eg transportationelectricityconnection)loggingandfuelwoodandpoverty
Fire and other natural disasters and terrain are the most fre-quentlymentionedbiophysicaldriversofforestarealoss(deforesta-tion) For example in India and Indonesia the two countrieswiththelargestforestareaamajorityofstudiessuggestterrainandfireand other natural disasters as the major biophysical drivers Oursynthesisofsite-levelcasestudiesalsoidentifiespopulationplanta-tionagriculturalexpansionaccessibilityandinfrastructureaswellasfuelwoodandloggingastheimportantsocioeconomicdriversofdeforestation
There are fewer studies on the drivers of forest area gain (af-forestationandreforestation)thanondeforestationAmongthemmoststudieshavefocusedonsocioeconomicdriversThefrequentlymentioned socioeconomic drivers include accessibility povertyeconomy and livestock and biophysical drivers include climateandterrainThemajordriversof forestchangeatdifferentspatialandtemporalscalesaredescribedinmoredetailinSupplementarySectionTextS2FiguresS6andS7
32emsp|emspForest change
Thecountry-specificandspatialdistributionofconvertingfromfor-esttoagriculturallandandagriculturallandtoforestoverthetimeperiod1992ndash2015isshowninFigure2andFigureS4MalaysiaandPhilippinesexperienced largeareasofdeforestationaswellasaf-forestation over 1992ndash2015 Cambodia and Singapore had moredeforestationthanafforestation
Cat
egor
yPr
oxy
varia
ble
Tem
pora
l re
solu
tion
Spat
ial r
esol
utio
nU
nit
Sour
ce
VI
Econ
omy
30
Marketaccessibilityindex
Constant
1 km
times 1
km
-VerburgEllisandLetourneau
(2011)
31
GD
P pe
r cap
itaDecadal(1980
ndash2010)
05deg times
05
degUSD2005
Globaldatasetofgriddedpopulation
andGDPscenariosMurakamiand
Yamagata(2016)
32
Distancetominingfacilities
Constant
5primetimes5prime
kmCalculatedfromMineralResources
On-LineSpatialDatabyUSGS
33
Povertyindex
Yearly(1992ndash2013)
5rsquotimes5rsquo
-Calculatedfrompopulation(HYDE
32)(KleinGoldewijkBeusen
DoelmanampStehfest2017)and
Nighttimelight(Version4
DMSP-OLSNighttimeLightsTime
Series)byfollowingthemethod
developedbyGhoshAnderson
ElvidgeandSutton(2013)
TAB
LE 2
emsp(Continued)
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp5XU et al
TAB
LE 2
emspProxyvariablesoftheidentifydrivers
C
ateg
ory
Prox
y va
riabl
eTe
mpo
ral
reso
lutio
nSp
atia
l res
olut
ion
Uni
tSo
urce
Biophysical
variables
ITerrainsoil
and
wat
er1
Terrain
Constant
5primetimes5prime
(~10kmtimes10km)
Categoricaldatainto7
gradientclasses
GlobalAgro-ecologicalZones
(GAEZ)v30(FischerNachtergaele
ampPrieler2012)
2-6
Soilchemicalcomposition
depthdrainagefertilityand
texture
7Distancetowaterbodies
5primetimes5prime
kmCalculatedfromGlobalLakesand
WetlandsDatabase(GLWD)level2
data(LehnerampDoll2004)
IIC
limat
e8-10
Meanrateofchangeand
standarddeviationofannual
prec
ipita
tion
Yearly
05deg times
05
degdeg Cdeg C
yea
rClimaticResearchUnit(CRU)TS
401
11-13
Meanrateofchangeand
standarddeviationofannual
tem
pera
ture
mmmmyear
14
Meanannualpotential
evapotranspiration
mm
IIIN
atur
al
disaster
15
Meanburnedareafraction
Yearly(1997ndash2014)
025
deg times 0
25deg
GlobalFireEmissionsDatabase41
(GiglioRandersonampWerf2013)
16
Distancetolandslideevents
Constant
5primetimes5prime
kmCalculatedfromGlobalLandslide
Catalog(KirschbaumAdlerHong
HillampLerner-Lam2010)
Soci
oeco
nom
ic
variables
IVPo
pula
tion
and
urbanization
17-18
Meanandrateofchangein
urbanpopulationdensity
Yearly
5primetimes5prime
inhabitantskm
2 inhabitantskm
2 middotyea
rHYDE32(KleinGoldewijkBeusen
DoelmanampStehfest2017)
19-20
Meanandrateofchangein
ruralpopulationdensity
21-22
Meanandrateofchangein
urba
n ar
ea fr
actio
nyear
23
Migration
Dec
adal
(1970ndash2000)
05deg times
05
degNumberofmigrants
km2
GlobalEstimatedNetMigration
GridsByDecadev1(deSherbinin
etal2012)
VLivestock
24-29
ChickenCattleSheepPig
GoatandDuckcounts
Constant
1 km
times 1
km
Headkm
2GriddedLivestockoftheWorld
(GLW)version2(Robinsonetal
2014)
(Con
tinue
s)
6emsp |emsp emspensp XU et al
3emsp |emspRESULTS
31emsp|emspSynthesis of the case study
The case studies regarding the drivers of forest loss and gain arespreadacrossallcountriesandhavediversespatialscalesfromvil-lagetonationalscale(FigureS5)Thereare173studies(81ofthetotalcollectedstudies)studyingdriversbasedonfieldsurveys in-terviewsandliteraturereviewsMeanwhile40studies(19ofthecollectedstudies)haveusedquantitativeapproachessuchaslinearregressionlogisticregressionsystemdynamicsmodelingandcel-lularautomatamodelingtoreveal thedriversofLULCCHowevermostofthesequantitativestudiesareatsmallerspatialscales(vil-lage to state and province level) (31 studies) There are only ninequantitativestudiesatthenationalscale
We identify the drivers of forest change from collected casestudies(Figure1andTable1)Thebiophysicaldriversincludeterrainor topographicalconditions (egelevationaltitudeand theslopeof land) soil conditions (eg soil fertility texture and moisture)wateravailabilityclimate(egtemperatureprecipitationandtheirchanges)fireandothernaturaldisasters(eglandslide)Thesocio-economicdriversmainlyrelatetopopulationgrowthurbanizationlivestockorgrazing(egovergrazing)marketinfluenceoreconomydevelopment(egGDPpercapitaaccesstomarket)plantationoragroforestdevelopment agricultural expansionminingand indus-try accessibilityand infrastructure (eg transportationelectricityconnection)loggingandfuelwoodandpoverty
Fire and other natural disasters and terrain are the most fre-quentlymentionedbiophysicaldriversofforestarealoss(deforesta-tion) For example in India and Indonesia the two countrieswiththelargestforestareaamajorityofstudiessuggestterrainandfireand other natural disasters as the major biophysical drivers Oursynthesisofsite-levelcasestudiesalsoidentifiespopulationplanta-tionagriculturalexpansionaccessibilityandinfrastructureaswellasfuelwoodandloggingastheimportantsocioeconomicdriversofdeforestation
There are fewer studies on the drivers of forest area gain (af-forestationandreforestation)thanondeforestationAmongthemmoststudieshavefocusedonsocioeconomicdriversThefrequentlymentioned socioeconomic drivers include accessibility povertyeconomy and livestock and biophysical drivers include climateandterrainThemajordriversof forestchangeatdifferentspatialandtemporalscalesaredescribedinmoredetailinSupplementarySectionTextS2FiguresS6andS7
32emsp|emspForest change
Thecountry-specificandspatialdistributionofconvertingfromfor-esttoagriculturallandandagriculturallandtoforestoverthetimeperiod1992ndash2015isshowninFigure2andFigureS4MalaysiaandPhilippinesexperienced largeareasofdeforestationaswellasaf-forestation over 1992ndash2015 Cambodia and Singapore had moredeforestationthanafforestation
Cat
egor
yPr
oxy
varia
ble
Tem
pora
l re
solu
tion
Spat
ial r
esol
utio
nU
nit
Sour
ce
VI
Econ
omy
30
Marketaccessibilityindex
Constant
1 km
times 1
km
-VerburgEllisandLetourneau
(2011)
31
GD
P pe
r cap
itaDecadal(1980
ndash2010)
05deg times
05
degUSD2005
Globaldatasetofgriddedpopulation
andGDPscenariosMurakamiand
Yamagata(2016)
32
Distancetominingfacilities
Constant
5primetimes5prime
kmCalculatedfromMineralResources
On-LineSpatialDatabyUSGS
33
Povertyindex
Yearly(1992ndash2013)
5rsquotimes5rsquo
-Calculatedfrompopulation(HYDE
32)(KleinGoldewijkBeusen
DoelmanampStehfest2017)and
Nighttimelight(Version4
DMSP-OLSNighttimeLightsTime
Series)byfollowingthemethod
developedbyGhoshAnderson
ElvidgeandSutton(2013)
TAB
LE 2
emsp(Continued)
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
6emsp |emsp emspensp XU et al
3emsp |emspRESULTS
31emsp|emspSynthesis of the case study
The case studies regarding the drivers of forest loss and gain arespreadacrossallcountriesandhavediversespatialscalesfromvil-lagetonationalscale(FigureS5)Thereare173studies(81ofthetotalcollectedstudies)studyingdriversbasedonfieldsurveys in-terviewsandliteraturereviewsMeanwhile40studies(19ofthecollectedstudies)haveusedquantitativeapproachessuchaslinearregressionlogisticregressionsystemdynamicsmodelingandcel-lularautomatamodelingtoreveal thedriversofLULCCHowevermostofthesequantitativestudiesareatsmallerspatialscales(vil-lage to state and province level) (31 studies) There are only ninequantitativestudiesatthenationalscale
We identify the drivers of forest change from collected casestudies(Figure1andTable1)Thebiophysicaldriversincludeterrainor topographicalconditions (egelevationaltitudeand theslopeof land) soil conditions (eg soil fertility texture and moisture)wateravailabilityclimate(egtemperatureprecipitationandtheirchanges)fireandothernaturaldisasters(eglandslide)Thesocio-economicdriversmainlyrelatetopopulationgrowthurbanizationlivestockorgrazing(egovergrazing)marketinfluenceoreconomydevelopment(egGDPpercapitaaccesstomarket)plantationoragroforestdevelopment agricultural expansionminingand indus-try accessibilityand infrastructure (eg transportationelectricityconnection)loggingandfuelwoodandpoverty
Fire and other natural disasters and terrain are the most fre-quentlymentionedbiophysicaldriversofforestarealoss(deforesta-tion) For example in India and Indonesia the two countrieswiththelargestforestareaamajorityofstudiessuggestterrainandfireand other natural disasters as the major biophysical drivers Oursynthesisofsite-levelcasestudiesalsoidentifiespopulationplanta-tionagriculturalexpansionaccessibilityandinfrastructureaswellasfuelwoodandloggingastheimportantsocioeconomicdriversofdeforestation
There are fewer studies on the drivers of forest area gain (af-forestationandreforestation)thanondeforestationAmongthemmoststudieshavefocusedonsocioeconomicdriversThefrequentlymentioned socioeconomic drivers include accessibility povertyeconomy and livestock and biophysical drivers include climateandterrainThemajordriversof forestchangeatdifferentspatialandtemporalscalesaredescribedinmoredetailinSupplementarySectionTextS2FiguresS6andS7
32emsp|emspForest change
Thecountry-specificandspatialdistributionofconvertingfromfor-esttoagriculturallandandagriculturallandtoforestoverthetimeperiod1992ndash2015isshowninFigure2andFigureS4MalaysiaandPhilippinesexperienced largeareasofdeforestationaswellasaf-forestation over 1992ndash2015 Cambodia and Singapore had moredeforestationthanafforestation
Cat
egor
yPr
oxy
varia
ble
Tem
pora
l re
solu
tion
Spat
ial r
esol
utio
nU
nit
Sour
ce
VI
Econ
omy
30
Marketaccessibilityindex
Constant
1 km
times 1
km
-VerburgEllisandLetourneau
(2011)
31
GD
P pe
r cap
itaDecadal(1980
ndash2010)
05deg times
05
degUSD2005
Globaldatasetofgriddedpopulation
andGDPscenariosMurakamiand
Yamagata(2016)
32
Distancetominingfacilities
Constant
5primetimes5prime
kmCalculatedfromMineralResources
On-LineSpatialDatabyUSGS
33
Povertyindex
Yearly(1992ndash2013)
5rsquotimes5rsquo
-Calculatedfrompopulation(HYDE
32)(KleinGoldewijkBeusen
DoelmanampStehfest2017)and
Nighttimelight(Version4
DMSP-OLSNighttimeLightsTime
Series)byfollowingthemethod
developedbyGhoshAnderson
ElvidgeandSutton(2013)
TAB
LE 2
emsp(Continued)
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp7XU et al
Theconversionfromforesttoagricultural landandagriculturallandtoforestco-existinmanyregions(FigureS4)Thehotspotre-gionsofthetwochangesovertheperiod1992ndash2015aremainlycon-centrated inKalimantan Sumatra East India and theHinduKushHimalayanregions(Figure3)
33emsp|emspQuantification of the drivers
331emsp|emspForest to agricultural land
Herewe shownational averages of the standardized coefficientswith their standard deviations from the GWR analysis (Figure 4)Overall the impacts of various drivers in different countries arequiteheterogeneousAsingledriverhasdifferenteffectsondiffer-entcountriesForexamplethedistancetowaterbodies(variable7inFigure4a)canhaveapositivenegativeornearlyno(~zerovalues)effectonthechangefromforesttoagriculturallandIngeneralthedistancetowaterbodies(variable7)meanannualprecipitation(vari-able8)meanburnedareafraction(variable15)goatcount(variable28) andGDPper capita (variable31) are the importantdrivers inmostcountriesbuttheirimpactsdifferacrosscountries
InordertogeneralizetheresultsweusetherelativeimportancetocombinetheimpactsfromdifferentdrivercategoriesTerrainsoilandwater(categoryI)andlivestock(categoryV)arethemostdomi-nantdrivercategoriestheyarethemostimportantdrivercategoriesinsix (ieBhutanLaosMalaysiaMyanmarNepalSriLankaandVietnam)andthreecountries (ieCambodia Indiaand Indonesia)respectively(Figure5a)Ontheotherhandnaturaldisasters(cate-goryIII)havetheleastimpactsinmostcountries(11countries)
Theimportanceofbiophysicaldriversisrelativelylower(theim-portanceof socioeconomicdrivers is higher) in someSouthAsiancountries such as Bangladesh Nepal and Pakistan and maritime
Southeast Asian countries Philippines and Indonesia mainly be-cause of lower JRWs of the terrain soil andwater (I) and higherJRWs of the urbanization and population (IV) and economy (VI)Socioeconomic and biophysical drivers have approximately equalimportance insomecountries inMainlandSoutheastAsiasuchasMalaysiaLaosThailandandVietnam(Figure5a)
332emsp|emspAgricultural land to forest
For the conversion from agricultural land to forest mean burnedarea fraction (variable 15) migration (variable 23) and povertyindex (variable 33) are the most important drivers in most coun-tries(Figure4b)Migration(variable23)hasastrongnegativeeffectonthischange(standardizedcoefficientltminus04)inIndiabutsmaller(mostlypositive)effectsinothercountriesSimilarlypigcount(vari-able27)hasastronglypositiveeffectonthischangeinIndia(stand-ardizedcoefficientgt04)Thesetwodriversarethemostimportantfor changes from agricultural land to forest in IndiaWe observesmallerimpactsofthesedrivers(smallabsolutevalueofthestand-ardizedcoefficients)onthechangefromagricultural landtoforestthanthechangefromforesttoagriculturallandSomevariableshavenearlynegligibleimpactsinmostcountriessuchasrateofchangeofannualprecipitation standarddeviationofannual temperaturechickencountandmarketaccessibilityindex(variables91324and30)
Terrainsoilandwater(categoryI)isthemostimportantdrivercategoryinfivecountries(BhutanCambodiaIndonesiaLaosandSriLanka)whilelivestock(categoryV)isthemostimportantcategoryinfourcountries(ieBruneiIndiaMyanmarandNepal)(Figure5b)Forexampletherelativeimportanceofterrainsoilandwatercon-dition (I) in Indonesia isashighas4036highlightingthe impor-tanceofthefavorableenvironmentalconditionsforforestregrowth
F I G U R E 1 emspFrequencydistributionofthemajordriversidentifiedfromthesynthesisofcasestudiesinSouthandSouthEastAsia(SSEA)countriesThedriversmentionedmorethan10times(sumofthefrequenciesforbothforestarealossandgain)areplottedhereOtheridentifieddriverscouldbefoundinTable1
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
8emsp |emsp emspensp XU et al
Naturaldisasters(categoryIII)hastheleastinfluenceinmostcoun-tries(11countries)
For changes from agricultural land to forest biophysical driv-ers are less important than socioeconomic drivers in BangladeshBhutan India Pakistan andNepal(Figure 5b) The importance ofbiophysical and socioeconomic drivers are approximately equalin most mainland Southeast Asia countries including CambodiaMalaysiaMyanmarPhilippinesandThailandTheimportanceofcli-mate(II)ishigherinmaritimeSoutheastAsiasuchasIndonesiaandPhilippines IndonesiahasthehighestJRWforbiophysicaldriversmainlydue to terrain soil andwater (I)whilePhilippineshas thegreatestJRWofclimate(II)amongallcountries
4emsp |emspDISCUSSION
41emsp|emspResults comparison
HereweuseaquantitativestudyofLULCCdriversinIndia(Meiyappanetal2017)tovalidateourresultsTheIndiastudycompiledgt200socioeconomicvariablesin~630000villages(~2kmtimes2kmonav-erage)toidentifythedriversofLULCC(includingforestarealossesandgains)The studyuseda ldquofractionalrdquobinomial logitmodel andsynthesizedcasestudiestoevaluatetheresultsTheldquofractionalrdquobi-nomiallogitmodelgeneratessimilarresultsasourGWRanalysisbyusing regression coefficients to indicate the influenceofdifferentdrivers (bothstudiesusedthesamez-scorestandardization forall
drivervariables)Hereweincludethe65casestudiescompiled inMeiyappanetal(2017)toidentifythedriversofforestchangeThiswillnotdirectlyaffectourquantitativedriveranalysisbecauseour33biophysicalandsocioeconomicproxydataarecollectedregard-less of the frequencies of the driversmentioned in the literatureInadditionwedidnotusethevillagelevelsocioeconomicdatasetcompiledintheIndiastudyaswedidnothavethesamelevelofin-formationinothercountriesThereforeourquantitativedriveranal-ysis(Figures4and5)isindependentfromMeiyappanetal(2017)andtheresultsofthetwostudiesarecomparable
Wecompared thedrivers for the forest to agricultural landofthisstudywiththedriversforforestarealossinIndia(1995ndash2005)(Meiyappanetal2017)Mostofthedriverproxiesatthecountryscale used in this studydonot directlymatch the drivers used inMeiyappanet al (2017) Insteadwehavecompared theeffectofthedriversinthisstudywiththedriverswithcloselyrelatedproxyvariablesinMeiyappanetal(2017)
Inourresultsmeanannualprecipitationhasapositive impactmeaninganincreaseinmeanannualprecipitationcausesmorefor-esttobeconvertedtoagriculturallandTheMeiyappanetal(2017)study used precipitation ofwettestmonth or quarter rather thanmeanannualprecipitationbutalsofoundapositiveimpactonfor-estlosssuggestingthatforestsaremorelikelytobeconvertedtocroplandwithincreasesinprecipitation
InourstudythemarketaccessibilityindexhasanegativeimpactonthechangefromforesttoagriculturallandThenegativeimpact
F I G U R E 2 emspEstimatedchangesfromforesttoagriculturalland(right)andagriculturallandtoforest(left)over1992ndash2015basedontheEuropeanSpaceAgencyClimateChangeInitiativelandcoverproduct(Defournyetal2017)
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp9XU et al
ismainlybecauseforestsintheregionthataredifficulttoreachareunlikelytobeconvertedtocroplandTheIndianstudyusesaldquoverysteep(gt50)sloperdquo(oftheland)asanindicatorofaccessibilityandfindsanegativerelationshipwhichmatcheswiththisstudy
IntheIndianstudytheavailabilityofpowersupplyfordomesticpurposeisnegativelyassociatedwithforestlossourstudysuggeststhaturbanareafractionhasanegativeimpactonforestlossIfweas-sumethaturbanareasinIndiahaveahigheravailabilityofpowersup-plyfordomesticpurposetheresultsfromthetwostudiesaresimilar
Our results suggest that the area of forest converted to agri-cultural land isgreater inregionsthatareclosetominingfacilities(negative impact from the distance to mining facilities) while inMeiyappanetal (2017) theoccupation (buildingminingmaterials)hasapositiveimpactItishighlypossiblethatintheregionsclosetominingfacilities(lowervalueofthedistancetominingfacility)morepeopleworkfortheminingindustryandtheoccupationinminingishigherThereforethesetwodrivershaveoppositeimpactsbutthesameinterpretation
Thedistancetowaterbodiesinourstudyhasapositiveimpactonthechangefromforest toagricultural landandtheproportionof cropland irrigated has a negative impact onforest area loss inthe Indian studyThe regionsnear towaterbodies (lowervalueofthe distance to waterbodies) have better irrigation conditions for
cropland and greater proportions of cropland irrigated this canpromotecropyieldandthereforereducethepressureonadjoiningforests(Meiyappanetal2017)Thereforetheareaofforestcon-vertedtoagriculturallandissmallerinsuchregionsThetwostudiesmatchwitheachotherinthisdimension
Theabovediscussionshowsthatourresultsonthedriversforthe conversion from forest to agricultural land in India are similartoapreviousnationalscalestudyinIndia(Meiyappanetal2017)
42emsp|emspDrivers for LULCC
421emsp|emspForest to agricultural land
Terrain(variable1inFigure4a)hasapositiveimpactontheconver-sionfromforesttoagriculturallandinIndiawhichisdifferentfromothercountriesThis ismainlybecausethehotspotregionsofthischange in India concentrate in theOrissa andChhattisgarh statesincenter-eastofthecountry(Figure3)whereforestaswellasitschangesaremainly locatedinhigherandsteeperregionsThisre-gionalsohasthelargestareaofshiftingcultivationinIndia(SinghPurohitampBhaduri2016)Shiftingcultivationmainlyoccursinhighandsteepregionsthereforewecanobservemoreforestarealossinregionswithhighterrainindexvalues
F I G U R E 3 emspSpatialdistributionofcasestudiesandhotspotregionslanduseandlandcoverchange(LULCC)driversThelocationsandresearchlevelsofthecasestudiesaswellasthenumberofcasestudiesareextractedfromthe213publicationslistedinTablesS1andS2aswellasTablesS10andS11inMeiyappanetal2017Thehotspotregionsat99confidenceintervalarerecognizedfromtheHotspotanalysiswiththeGetis-OrdGivaluesge3
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
10emsp |emsp emspensp XU et al
Unlike most of the countries analyzed most soil variables(variable2ndash6 inFigure4a)havepositive influences inBangladeshThe hotspot regions of change from forest to agricultural land inBangladesh are concentrated in the southwest coastal regions(Figure 3) Our previous work in Bangladesh has shown that theforest in this regionwasmainlymangrove forest and agriculturallandismainlyusedforaquaculture(fishorshrimpponds)Thereforethe changes from forest to agricultural land aremainlymangrovetoaquaculture in theseregions (UddinHoqueampAbdullah2014)Althoughsatisfactorypondbottomsoilconditionsfavoraquacultureproduction(SalamKhatunampAli2005)ourresultsindicatethatsoilconditionsdonotdirectlymotivatechangesfrommangroveforesttoaquacultureInBangladeshthemeanurbanandruralpopulation
densities(variable17ndash19inFigure4a)havenegativeimpactsonthischangewhichisalsodifferentfrommostothercountriesThismaybebecause regionswithhigherpopulationhavemoreurban landwhichcrowdsoutotherlandtypessuchasforestthereforethereislessforestareathatcanbeconvertedtoagriculturalland(Shehzadetal2014)
IntwomostpopulouscountriesinSSEAIndiaandIndonesialive-stock(variables24to29)isthemostdominantdrivercategorywhilepopulation andurbanization is the least important driver category(Figure5a)Ourresultsagreewiththeconclusionofameta-analysisbyRudelDefriesAsnerandLaurance(2009)thatinrecentyearswell-capitalized ranchers farmers and loggers producing for con-sumersindistantmarketshavebecomemoreprominentintropical
F I G U R E 4 emspThenationalaverageofstandardizedcoefficientsofeachdriverindifferentcountriesfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestThestandardizedcoefficientreferstoanumberofstandarddeviationschangeinLULCCsperstandarddeviationchangeindivingfactorsThedescriptionofvariables1ndash33andcategoriesIndashVIlistcouldbefoundinTable2Thesizeoftheblackdotindicatedthestandarddeviationofthestandardizedcoefficientineachcountry
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp11XU et al
forestsandthisglobalizationhasweakenedthehistoricallystrongrelationshipbetweenlocalpopulationgrowthandforestcover
Insomecountriesbiophysicaldriversaremore importantForexampleinBhutantheconversionfromforesttoagriculturallandhas been controlledmainly by biophysical drivers rather than so-cioeconomic drivers especially terrain soil andwater conditionsBhutan has a traditional culture of protecting forest (BruggemanMeyfroidtampLambin2016)andlocalresidencesavoidaffectingfor-estwhendeveloping their social economyBiophysical conditionsthereforehaveagreaterimpact
422emsp|emspAgricultural land to forest
Climateconditions(temperatureandprecipitation)aremostlyposi-tivelyorneutrallyassociatedwiththechangefromagriculturallandtoforestindicatingthatwetterandwarmerconditionswithhighervariations are favorable for forest regrowth However the higherstandarddeviationsofthestandardizedcoefficientsinallcountriesindicatethattheimpactsoftemperaturevariationsandmeantem-perature arehighlyheterogeneous (Figure4b) In the tropical for-estwetter(whenmeanannualprecipitationislessthan2445mm)
F I G U R E 5 emspTherelativeimportanceofdifferentdriversfortheconversions(a)fromforesttoagriculturallandand(b)fromagriculturallandtoforestinSSEAcountries
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
12emsp |emsp emspensp XU et al
andwarmerconditionsarebeneficialfortreeproductivity(Schuur2003)Thereforeprecipitationand temperaturehavepositive im-pactsWeobservenegativeimpactsofmeanannualprecipitationinIndiaandMyanmar(variable8inFigure4b)showingthatthischangemainlytakesplaceinrelativelydrierregionsinthesetwocountries
Population and urbanization usually have smaller impacts ex-cept in India Laos andMyanmar In India lower population andless urbanized regions favor reforestation and afforestation (Monetal2012)Onthecontraryweobservemostlypositive impactsfrompopulationandurbanizationvariablesinLaosPhompilaLewisOstendorfandClarke(2017)suggestthatpopulationincreasescansimultaneouslyleadtoanincreaseinforestandexpansioninforestclearanceindifferentlocationsofLaos
Livestockvariablesplay importantroles inthischangeinIndiaThe cattle count has a negative impact (variable 25 in Figure4b)GiventhatcattleiscommonagriculturallaborinIndia(Basu2011)wecaninferthattheregionswithlowercattlecounthavelessculti-vatingactivitiesandhumandisturbanceAgriculturallandinregionswithlesshumandisturbanceismorelikelytobeconvertedtoforestthusweobserveanegativeimpactfromcattlecount
InBangladeshsocioeconomicdriversaremore importantthanbiophysical drivers especially the economy category particularlyin the southern coastal regions of Bangladeshwhere aquaculturepondsareconcentratedAquaculturehashigherprofits(Ali2006)thechangesfromaquaculturetoforest(mangrove)couldhavelargeimpactsontheeconomyThereforewenotegreaterimportanceoftheeconomyonlandconversionsinBangladeshWealsofindthatclimate is important because of its impacts on aquaculture (HuqHugeBoonampGain2015)
43emsp|emspImplications
OurresultshaveseveralpracticalimplicationsFirstbothbiophysi-cal and socioeconomic drivers strongly influence the inter-changebetweenforestandagriculturallandHoweveroursynthesisofcasestudies indicates that there aremore studiesdiscussing socioeco-nomicdriversthanbiophysicaldrivers(Figure1andFigureS5)Ourresultsemphasizethe importanceofthebiophysicaldriversespe-ciallyforthechangefromagricultural landtoforestWhenoptingtoafforestandreforestdecision-makersshouldconsiderthe localterrainsoilwaterclimateconditionsaswellastheimpactsofnatu-raldisasters
Secondthespatiallyexplicitresultsshowthehighheterogene-ity andcomplexityof thedriversof forest changeTounderstanddeforestationafforestationandreforestationonehastofirstgeta thorough understanding of local conditions from both biophys-ical and socioeconomic aspectsWe have reported and discussedthedominantdriversfor16countrieswithlargedifferencesinso-cioeconomicconditionsTheresults improveourunderstandingofthekey factors influencingdeforestationafforestationand refor-estationineachcountryForexamplelivestock(categoryV)isthemostdominantdrivercategoryforIndiaandIndonesiaforthecon-versionfromforesttoagriculturallandMeanwhileforthechanges
fromagriculturallandtoforestlivestockvariables(categoryV)playimportant roles in thischange in Indiabut terrain soil andwaterconditions aremore important in Indonesia (category I) Both de-forestationandafforestation (reforestation)arestrongly impactedby livestock in Indiasuggesting localstakeholderscould influencedeforestation afforestation or reforestation through changes inlivestockorgrazingactivitiesInIndonesiasimilarstrategiesmaybeeffective inpreventingdeforestationMeanwhile terrainsoilandwaterconditionsaremorecriticalforafforestationandreforestationinIndonesia
Finallythedetaileddriversofforestchangecanbeincorporatedinto land use downscalingmodels such as spatial dynamic alloca-tionmodel to improve their projections (Meiyappan et al 2014)Improvementoflandusedownscalingmodelscanhelpbridgescalesbetween human and earth systems providing better LULCC pro-jections in the future In addition these detailed drivers can alsobeincorporatedineconomicmodelsestimatingfutureLULCClikeGCAM (Wise Dooley Luckow Calvin amp Kyle 2014)Models likeGCAMtypically relyonprofit todetermineLULCCbut thisstudyhasshownthatmanyotherfactorsareimportantfordrivingthesechangesSuchintegrationischallenginginmanyaspectssuchastheuncertainties of the spatially explicit driver datasets are high thelack of future datasets and the non-linear relationships betweenthese drivers and other socioeconomic processes To overcomethese challengeswe could use the historical results (such as thisstudy)asreferencesforthefutureprojectionadoptsocioeconomicscenarios (such as the Shared Socioeconomic Pathways [SSPs]) togenerate consistent biophysical and socioeconomic variables andfurtherstudytherelationshipsbetweenLULCCdriversanddifferentsocioeconomicprocesses
44emsp|emspUncertainties and limitations
Thisstudyquantifiesthebiophysicalandsocioeconomicdriversofchangesfromforesttoagriculturallandandagriculturallandtofor-estinSSEAforthefirsttimeHowevertherearesomecaveatsandlimitationsinthisstudy
Firstalthoughwehavecollectedthespatiallyexplicitproxydatatorepresenttheimportantdriversidentifiedfromcasestudieshow-ever there are still some important driverswe do not have proxydataforsuchasshiftingcultivation(swidden)fuelwood(loggingandcharcoal)globalization(internationaltrade)andforestmanagement(policy)whichareeitherdifficulttoquantifyorlackingreliablegrid-deddataThesedriverscouldhavelargeimpactsoneitherthechangefrom forest to agricultural land or from agricultural land to forestForexampleshiftingcultivationhaslargeimpactsonbothchanges(BruunNeergaardLawrenceampZiegler2009)globalization(interna-tionaltrade)islinkedtodeforestation(LopezampGalinato2005)TheabsenceoftheproxydataforthesedriversmayinfluenceourresultsHoweverthecurrentavailabilityofdatadoesnotallowustoincludethemthusweleavetheirincorporationtofuturestudies
SecondweshouldnotethatspatialandtemporalinconsistenciesexistedbetweenspatialproxydataandLULCCdata(Table2)These
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp13XU et al
will introduceuncertainties in thedriver analysisAdditionallywehavecollectedtheproxydata for thedrivers fromglobaldatasetsratherthanregionaldatasetsSomeoftheglobaldatasetshavenotbeencalibratedagainstregionaldatainSSEAcountrieswhichmayhavearelativelylowerqualityinourstudyareaandthusinfluencequantitativedriverresultsWeusetheseglobaldatasetsforseveralreasonsFirstglobaldatasetscovertheentirestudyareaRegionaldatasets covering the entire SSEAare relatively rare Second ourGWR analysis needs gridded driver proxy data tomatchwith thegridded forest change data Regional datasets are usually basedon geopolitical units that may introduce additional uncertaintiesFinallythedatausedarefromthebestavailableglobaldatasetsinliteratureandareconsistentatbothspatialandtemporalscalesThismeans that thesedatasets indifferent countries and timeperiodsarecomparableIncontrastsomeregionaldatasetsmaynotbecon-sistentinspatialortemporalscaleduetothedifferentdatasourcesanddataqualityThereforewhiletherearestillinconsistenciesandmissingdetailsinourdatasetsourquantitativeanalysisisrobustatthecountryscaleinSSEAregion
Third as mentioned in the ldquoSynthesis of the Site Level CaseStudiesrdquosectionweassumethat thesynthesizeddriversof forestarealossandgainrepresentthedriversofthespecificchangesfromforest toagricultural landandviceversaThisassumptionmay in-troduce noise and additional uncertainties into the driver synthe-sis because the drivers of other LULCC activities (except for theinterchangesbetweenforestandagriculturalland)areincludedandcountedindiscriminately(sincewealsoassumeallmentioneddriversincasestudiesareequallyimportantunlessspecified)Howevertheimpactsoftheseassumptionsonourquantitativedriverresultsarelimitedbecausetheseassumptionsonly influencethefrequenciesofdifferentdrivers(Figure1)butarenotdirectlyrelatedtothespa-tial proxieswhich are collected regardless the frequencies of thedriversmentioned in the literatureThereforewebelieve thecol-lectedcasestudiesrepresentthedriversofthechangesfromforesttoagriculturallandandviceversa
Lastlyacountry-by-countryvalidationwillfurtherreinforceourfindingsHoweverthequantitativestudiesatcountrylevelinSSEAarerare(ninestudies)andthedriverproxiesusedinthesestudiesgreatlydifferInthiscasewehavevalidatedourresultbycomparingtoaquantitativestudyinIndiaandfoundagoodmatchbetweentwostudiesstrengtheningourresults
By synthesizing the local-scale driver information of forestchangeandusingquantitativemodelsthisstudyprovidesinsightinto the complexity of LULCC processes Unlike previous stud-ieswhich focus on socioeconomic drivers this study highlightsthe importance of both biophysical and socioeconomic driversGenerally socioeconomic development increases food and landdemands drivingpeople to convert forest to agricultural land ifbiophysical conditions are favorable Agricultural land with lessfavorable biophysical conditionsmay be abandoned for tree re-growth or converted to forest plantations reduced human dis-turbanceand livestockpressureresults inmoreforestregrowthThese driving processes vary across regions and countries
emphasizingtheneedsforregion-andcountry-specificstrategiesfor deforestation and afforestation The biophysical and socio-economicdrivers identifiedcanhelp to improve theaccuracyoftheLULCCmodelingandprojectionsthatareimportantinputstoearthsystemmodels
ACKNOWLEDG EMENTS
ThisworkissupportedbytheUnitedStatesDepartmentofEnergy(NoDE-SC0016323)andNASALand-CoverLand-UseChangeProgram (NoNNX14AD94G)
ORCID
Xiaoming Xu httpsorcidorg0000-0003-1405-8089
Atul K Jain httpsorcidorg0000-0002-4051-3228
Katherine V Calvin httpsorcidorg0000-0003-2191-4189
R E FE R E N C E S
AliAMS(2006)RicetoshrimpLanduselandcoverchangesandsoildegradationinSouthwesternBangladeshLand Use Policy23421ndash435httpsdoiorg101016jlandusepol200502001
BasuP(2011)EngineeringCattleforDairyDevelopmentinRuralIndiaInSDBrunn(Ed)Engineering Earth The Impacts of Megaengineering Projects(pp189ndash215)DordrechtNetherlandsSpringerhttpsdoiorg101007978-90-481-9920-4_13
BruggemanDMeyfroidtPampLambinEF(2016)ForestcoverchangesinBhutanRevisitingtheforesttransitionApplied Geography6749ndash66httpsdoiorg101016japgeog201511019
Bruun T BDeNeergaard A LawrenceD amp Ziegler AD (2009)EnvironmentalconsequencesofthedemiseinSwiddencultivationinSoutheastAsiaCarbonstorageandsoilqualityHuman Ecology37375ndash388httpsdoiorg101007s10745-009-9257-y
CervarichMShuSJainAKArnethACanadellJFriedlingsteinP hellip ZengN (2016) The terrestrial carbon budget of South andSoutheast Asia Environmental Research Letters 11 httpsdoiorg1010881748-93261110105006
ChaoYCEZhaoYKupperLLampNylander-FrenchLA (2008)Quantifying the relative importance of predictors in multiplelinear regression analyses for public health studies Journal of Occupational and Environmental Hygiene 5 519ndash529 httpsdoiorg10108015459620802225481
CharltonM Fotheringham S amp Brunsdon C (2009)Geographically weighted regression white paper Maynooth Ireland NationalUniversityofIrelandMaynooth
DeSherbininALevyMAdamoSMacManusKYetmanGMaraVhellipPistolesiL(2012)MigrationandrisknetmigrationinmarginalecosystemsandhazardousareasEnvironmental Research Letters7045602httpsdoiorg1010881748-932674045602
Defourny P Moreau I Bontemps S Lamarche C Brockmann CBoettcherMhellipSantoroM(2017)Land cover CCI Product user guide version 20 Retrieved from httpsmapselieuclacbeCCIviewerdownloadESACCI-LC-Ph2-PUGv2_20pdf
DengJSWangKDengYHampQiGJ(2008)PCA-basedland-usechangedetectionandanalysisusingmultitemporalandmultisensorsatellitedataInternational Journal of Remote Sensing294823ndash4838httpsdoiorg10108001431160801950162
FAO(2015)GlobalForestResourcesAssessment2015Retrievedfromhttpwwwfaoorg3a-i4808epdf
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
14emsp |emsp emspensp XU et al
Feddema J JOlesonKWBonanGBMearns LOBuja LEMeehlGAampWashingtonWM(2005)Theimportanceofland-coverchangeinsimulatingfutureclimatesScience3101674ndash1678httpsdoiorg101126science1118160
Fischer G Nachtergaele F O Prieler S Teixeira E Toacuteth G vanVelthuizen H hellip Wiberg D (2012) Global Agro‐ecological Zones (GAEZ v3 0)‐Model DocumentationLaxenburgAustriaIIASA
FoleyJA(2005)GlobalconsequencesoflanduseScience309570ndash574httpsdoiorg101126science1111772
GhoshTAndersonS JElvidgeCDampSuttonPC (2013)Usingnighttimesatelliteimageryasaproxymeasureofhumanwell-beingSustainability54988ndash5019httpsdoiorg103390su5124988
GiglioLRandersonJTampVanDerWerfGR(2013)Analysisofdailymonthlyandannualburnedareausingthefourth-generationglobalfire emissions database (GFED4) Journal of Geophysical Research‐Biogeosciences118317ndash328httpsdoiorg101002jgrg20042
HoughtonRAHouseJIPongratzJvanderWerfGRDeFriesR S HansenM C hellip Ramankutty N (2012) Carbon emissionsfromlanduseandland-coverchangeBiogeosciences95125ndash5142httpsdoiorg105194bg-9-5125-2012
HuqNHugeJBoonEampGainAK(2015)Climatechangeimpactsin agricultural communities in rural areas of coastal BangladeshA tale of many stories Sustainability 7 8437ndash8460 httpsdoiorg103390su7078437
IslamMRMiahMGamp InoueY (2016)Analysisof landuseandlandcoverchanges inthecoastalareaofBangladeshusing landsatimagery Land Degradation amp Development27899ndash909httpsdoiorg101002ldr2339
Johnson J W (2000) A heuristic method for estimating the rel-ative weight of predictor variables in multiple regressionMultivariate Behavioral Research35 1ndash19 httpsdoiorg101207S15327906MBR3501_1
KirschbaumDBAdlerRHongYHillSampLerner-LamA(2010)AgloballandslidecatalogforhazardapplicationsMethodresultsandlimitationsNatural Hazards52 561ndash575 httpsdoiorg101007s11069-009-9401-4
Klein Goldewijk K Beusen A Doelman J amp Stehfest E (2017)Anthropogenic land use estimates for the Holocene - HYDE32Earth System Science Data9927ndash953httpsdoiorg105194essd-9-927-2017
Lambin E F Turner B l Geist H J Agbola S B Angelsen ABruceJWhellipXuJ(2001)Thecausesofland-useandland-coverchange Moving beyond the myths Global Environmental Change‐Human and Policy Dimensions11261ndash269httpsdoiorg101016S0959-3780(01)00007-3
Le Queacutereacute C Andrew R M Friedlingstein P Sitch S Hauck JPongratz J hellip Zheng B o (2018) Global Carbon Budget 2018Earth System Science Data102141ndash2194httpsdoiorg105194essd-10-2141-2018
LehnerBampDollP(2004)Developmentandvalidationofaglobalda-tabaseof lakes reservoirsandwetlandsJournal of Hydrology2961ndash22httpsdoiorg101016jjhydrol200403028
LesageJampPaceRK(2009)Introduction to spatial econometrics Boca RatonFLChapmanandHallCRCPress
LopezRampGalinatoGL(2005)Tradepolicieseconomicgrowthandthe direct causes of deforestation Land Economics 81 145ndash169httpsdoiorg103368le812145
MeiyappanPDaltonMOrsquoNeillBCampJainAK(2014)Spatialmodel-ingofagriculturallandusechangeatglobalscaleEcological Modelling291152ndash174httpsdoiorg101016jecolmodel201407027
Meiyappan Pamp JainAK (2012) Three distinct global estimates ofhistoricalland-coverchangeandland-useconversionsforover200yearsFrontiers of Earth Science6122ndash139httpsdoiorg101007s11707-012-0314-2
MeiyappanPRoyPSSharmaYRamachandranRMJoshiPKDefries R Samp JainAK (2017)Dynamics anddeterminants oflandchangeinIndiaIntegratingsatellitedatawithvillagesocioeco-nomics Regional Environmental Change 17 753ndash766 httpsdoiorg101007s10113-016-1068-2
Mon M S Mizoue N Htun N Z Kajisa T amp Yoshida S (2012)Factors affecting deforestation and forest degradation in selec-tively logged production forest A case study in Myanmar Forest Ecology and Management 267 190ndash198 httpsdoiorg101016jforeco201111036
Murakami D amp Yamagata Y (2016) Estimation of gridded popula-tion and GDP scenarios with spatially explicit statistical down-scaling ArXiv 161009041 Retrieved from httpsarxivorgabs161009041
MustardJFDefriesRSFisherTampMoranE(2012)Land-useand land-cover change pathways and impacts In G Gutman AC Janetos CO Justice EF Moran JF Mustard RRRindfuss D Skole BL Turner II amp MA Cochrane (Eds) Land Change Science Remote Sensing and Digital Image Processing vol6 (pp 411ndash429) Dordrecht Netherlands Springer httpsdoiorg101007978-1-4020-2562-4_24
Ord J K ampGetis A (1995) Local spatial autocorrelation statistics -distributional issues and an application Geographical Analysis 27286ndash306
PhompilaCLewisMOstendorfBampClarkeK(2017)ForestcoverchangesinLaoTropicalForestsPhysicalandsocio-economicfactorsarethemostimportantdriversLand623
PresteleRAlexanderPRounsevellMDAArnethACalvinKDoelman J hellip Verburg P H (2016) Hotspots of uncertainty inland-use and land-cover change projections A global-scale modelcomparison Global Change Biology 22 3967ndash3983 httpsdoiorg101111gcb13337
RamankuttyNamp Foley J A (1999) Estimating historical changes ingloballandcoverCroplandsfrom1700to1992Global Biogeochemical Cycles13997ndash1027httpsdoiorg1010291999GB900046
RobinsonTPWintGRWConcheddaGVanBoeckelTPErcoliVPalamaraEhellipGilbertM(2014)Mappingtheglobaldistributionof livestockPLoS One9 e96084 httpsdoiorg101371journalpone0096084
Rudel T K Defries R Asner G P amp Laurance W F (2009)Changing drivers of deforestation and new opportunities forconservation Conservation Biology 23 1396ndash1405 httpsdoiorg101111j1523-1739200901332x
SalamMKhatunNampAliM(2005)CarpfarmingpotentialinBarhattaUpazillaBangladeshAGISmethodologicalperspectiveAquaculture24575ndash87httpsdoiorg101016jaquaculture200410030
Schuur E AG (2003) Productivity and global climate revisited Thesensitivity of tropical forest growth to precipitation Ecology 841165ndash1170 httpsdoiorg1018900012-9658(2003)084[1165PAGCRT]20CO2
ShehzadKQamerFMMurthyMSRAbbasSampBhattaLD(2014)Deforestation trendsand spatialmodellingof itsdrivers inthe dry temperate forests of northern Pakistan - A case study ofChitral Journal of Mountain Science 11 1192ndash1207 httpsdoiorg101007s11629-013-2932-x
Singh S Purohit J K amp Bhaduri A (2016) Shifting cultivation inOdishaandChhattisgarhRichagro-biodiversesystemsunderriskJJDMS Ranchi147023ndash7027
UddinSMMHoqueATMRampAbdullahSA(2014)Thechang-ing landscapeofmangroves inBangladeshcomparedtofourothercountries in tropical regionsJournal of Forestry Research25605ndash611httpsdoiorg101007s11676-014-0448-z
vanVlietJMaglioccaNRBuumlchnerBCookEReyBenayasJMEllisEChellipVerburgPH(2016)Meta-studiesinlandusescience
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611
emspensp emsp | emsp15XU et al
Current coverage and prospects Ambio 45 15ndash28 httpsdoiorg101007s13280-015-0699-8
VerburgPHEllisECampLetourneauA(2011)AglobalassessmentofmarketaccessibilityandmarketinfluenceforglobalenvironmentalchangestudiesEnvironmental Research Letters6
WiseMDooleyJLuckowPCalvinKampKyleP(2014)Agricultureland use energy and carbon emission impacts of global biofuelmandatestomid-centuryApplied Energy114763ndash773httpsdoiorg101016japenergy201308042
SUPPORTING INFORMATION
Additional supporting information may be found online in theSupportingInformationsectionattheendofthearticle
How to cite this articleXuXJainAKCalvinKVQuantifyingthebiophysicalandsocioeconomicdriversofchangesinforestandagriculturallandinSouthandSoutheastAsiaGlob Change Biol 2019001ndash15 httpsdoiorg101111gcb14611