Thai J. For. 33 (3) : 56-74 (2014) วารสารวนศาสตร์ 31 (1) : 1-8 (2556)

Original article

Predicting Land-use and Land-cover Patterns Driven by Different Scenarios in the Emerald Triangle Protected Forests Complex

Yongyut Trisurat1* Naris Bhumpakphan1 Suchat Kalyawongsa2 Sapol Boonsermsuk2

Kamol Wisupakan3 Utai Dechyosdee3

1Department of Forest Biology, Faculty of Forestry, Kasetsart University, Bangkok 10900, Thailand2OfficeofForestryInternationalCooperation,RoyalForestDepartment,Bangkok10900,Thailand3ManagementoftheEmeraldTriangleProtectedForestsComplextoPromoteCooperationforTrans-boundaryBiodiversityConservationbetweenThailand,CambodiaandLaoPDRProject(PhaseIII),Royal Forest Department, Bangkok 10900, Thailand

*Corresponding Author, Email: fforyyt@ku.ac.th

Received: Sep 17, 2014 Accepted: Oct 1, 2014

ABSTRACT

TheEmeraldTriangle is the largest, extensive intact block in theGreaterMekongSub-region,whichstillremainsalongthetri-nationalbordersbetweenThailand,LaoPDRandCambodia.Theremaininghabitatsareveryimportantforthesurvivalofwide-rangingspeciesinthis area. This research aimed to assess land-use change in recent years and to predict land-use patternsacrosstheEmeraldTrianglelandscape.Thecurrentland-usemapwasvisuallyinterpretedfrom satellite images and the rate of reductionwas calculated accordingly. Future land-usepatternswerepredictedusingtheDyna-CLUEmodelbasedondifferentland-usescenariosin2030definedbymulti-stakeholdersfromthethreecountries.ThemodelresultsindicatedthatdrydipterocarpforestinthenorthofDongKhanthungproposedNationalBiodiversityConservationAreainLaoPDRandtothewestofPhaTaemNationalParkinThailandwouldbethreatenedbyencroachmentforagricultureandrubberplantation.Ifnorestrictionpolicy,partsofthePreahVihearprotectedforestinCambodiaandPhouXiangThongNationalBiodiversityConservationAreainLaoPDRwouldbeconvertedtoarablelandin2030.Evergreenforestswerepredictedtoberelativelyintact,asatthecurrentstage,becausetheyarefoundeitherinsideprotectedareasorinsteepterrains,thusbecomenaturalbarriersforhuman-intervention.

Keywords: CLUE-smodel,EmeraldTriangle,Landusescenarios,Trans-boundarybiodiversityconservation

Thai J. For. 33 (3) : 56-74 (2014) 57

INTRODUCTION

The Southeastern Indochina Dry Evergreen Forests ecoregion is situated across northern and central Thailand, Lao PDR, CambodiaandVietnam.However,abouttwo-thirds of the original forest of this eco-region has been converted or seriously degraded(Wikramanayake et al., 2000). The largest, extensiveintactblockstillremainsalongthetri-national borders betweenThailand,LaoPDRandCambodia,theso-calledtheEmeraldTriangle ForestsComplex (ETFC). It isrecognized globally as outstanding forbiodiversityconservationandimportanthabitatsforthelargevertebratesintheGreaterMekongSub-region(OfficeofEnvironmentalCenter,2005).According toBhumpakphan (2003),therearemorethan50specieslistedasIUCNthreatened status, and more than 10 species are categorized as critically endangeredorendangered species found in the ETFC, such as, Asian elephant (Elephans maximus),banteng(Bos javanicus), Eld’s deer (Cervus eldii), Clouded leopard (Neofelis nebulosa) and Siamese crocodile (Crocodylus siamensis). The ETFC landscape contains heterogenuous landscape patterns and hydrological conditions. The general topography in Thailand is mountaineous and slopes gently towardsthesoutheast.Incontrast,theareasinLaoPDRandCambodiaaregenerallyflatand part of the forested areas are inundated duringthewetseason.Asaresultofthehighlyheterogenuouslandscape,wide-rangingspeciesseasonally migrate across the tri-national boundariesandaredependentonverylimitedresources,includingpermanentwaterbodiesandlowlandforestpatchesthatarescatteredinprotected areas in the dry season (Bhumpakphan,

2003, 2006).Residual forests outside thereservesarevulnerabletodisturbance. Besides the diverse natural feaures, the disparities of conservation efforts and human capacity are clearly noted among the threecountries.CambodiaandLaoPDRhavesomeof themost extensive intact naturalforests, but they lack sufficient capacity toeffectively maintain the remaining forest cover andconservebiodiversityatalllevels(Galtet al., 2000). In contrast, the Government of Thailand has deployed many park rangers and facilitiestomanageandtoprotectbiologicalresources, butThailand’s protected areascontainrelativelylessbiodiversitythaninLaoPDRandCambodia(Trisurat,2003).Inaddition,a recent study on land-use change in Thailand between 2002 and 2008 indicated thatdeforestationwas continuing in the bufferzones(Trisurat,2009). Therefore, long-term persistence of trans-boundarybiodiversity in theETFC islargelydependentonthecooperationbetweenthe three countries to safeguard the remaining habitatsandtoreduceanthropogenicpressuresboth inside and in the buffer zones of theprotected forests. To address some of these issues, theGovernment ofThailandwithtechnical suport from the International Tropical TimberOrganization (ITTO) andfinancialsupportfromJapan,SwitzerlandandUSAhasinitiated the frameworkof trans-boundarybiodiversityconservationincooperationwithCambodiaandLaoPDRsince2001(Kalyawongsaand Hort, 2010). The current ITTO project phaseIII(2012-2015)aimsatstrengtheningtheprotectionoftrans-boundaryhabitatsofprotectedwide-rangingwildlifespeciesintheETFC landscape.

Thai J. For. 33 (3) : 56-74 (2014)58

In this context, it is important tounderstand present and future land-use/land-coverpatternsbecausedeforestationisconsidered as having an important effect on wildlifedistributionandbiodiversity(Sodhiet al.2004;Corlett2012).Thisisduetothefact that deforestation does not only cause habitat loss, but also it results in habitatfragmentation, diminishing patch size andcore area, and isolationof suitablehabitats(MacDonald,2003).TrisuratandDuengkae(2011) indicated that the predicted occurrence of Black-crestedBulbul(Pycnonotus melanicterus) in the Sakaerat Man and Biosphere Reserve inNakhonRatchasimaProvince,Thailandwouldsignificantlydecrease if forestcoverslightly declined from 45.3%to42%ofthereserveandintacthabitatswouldbeseverelyfragmented. Variousmodelshavebeendevelopedto forecast future land-use patterns,whichrange from simple system representations includingafewdrivingforcestosimulationsystemsbasedonaprofoundunderstandingofsituation-specificinteractionsamongalargenumberof factors (Verburg andVeldkamp,2004;Pontiuset al., 2008). The Markov Chain Model is a simple land use model that uses previouslandusetrendstopredictwhatwillhappeninthefuture.However,itisnotcapable ofaddressinglandsuitability,landdemandsand government policies in the model (Pontius et al., 2008). A Cellular Automata incorporates spatial component in the traditional Markov ChainModelanditcanaddressdynamicswithsimplerules(Baker,1989),andhasbeenappliedinawiderangeofland-usechangeapplications(Houet andHubert-Moy, 2006;Ballestores

and Qiu, 2012). Recently,agent-basedmodelwasdevelopedtoallowtheinfluenceofhumandecision-makingon the environment to beincorporated in a mechanistic and spatially explicitway,alsotakingintoaccountsocialinteraction, adaptation and decision-making atdifferentlevels(Matthewset al., 2007). The current research used the Dyna-CLUE (Conversion of Land Use and its Effects)model(VerburgandOvermars,2009)to assess future land-use at the ETFC. The Dyna-CLUEmodelwaschosenforthisstudybecauseitexplicitlyaddressesthedynamicsofthe different future land demands. In addition, ithasbeenusedatbothlocallevel(Trisuratet al.,2010)andregionallevel(VerburgandVeldkamp,2004)andhasbeenproventobeascapableasotherpopularland-usechangemodels (Pontius et al.,2008).Specificobjectivesofthisresearchwere(1)toquantifyrateofland-use/land-cover change in recent years, and (2) to allocate land-use change and land-use patternsacrosstheETFCbasedondifferentdemand scenarios of stakeholders from the three participating countries.

MATERIALS AND METHODS

Study area The ETFC comprises the Pha Taem ProtectedForestsComplex(PPFC)inThailand,the Preah Vihear Protected Forest for the Conservation of Genetic Resources of Plants andWildlife (PVPF) inCambodiaand twonationalbiodiversityconservationareas(NBCA)in Lao PDR.ThePPFC located inUbonRatchathaniprovinceincludesfiveprotectedareas,namely,PhaTaemNationalPark,KaengTanaNationalPark,PhuJong-NaYoiNational

Thai J. For. 33 (3) : 56-74 (2014) 59

Park, Yot Dom Wildlife Sanctuary and Bun Thrarik-Yot Mon Wildlife Sanctuary. The collectiveareaofthecomplexisapproximately1,736km2. The PVPF is located in Preah Vihear Province,Cambodia.Itadjoinsthesouthofthe Yot Dom Wildlife Sanctuary and to the westoftheMekongRiver,coveringanareaofapproximately1,900km2. To the northeast, itborderstoDongKhanthungproposedNBCA,whichcovers1,828km2. Another protected

area in Lao PDR situated in the ETFC is the PhouXiangThongNBCA,which islocatedeastofPhaTaemNationalPark,andhas an area of approximately 1,015 km2. Therefore, the entire protected areas encompass approximately6,500km2. The study area of this research also included surrounding protected areas locatedwithin a defined rectangularperimetercoveringaltogetherapproximately25,800km2 (Figure 1).

1

Figure 1 Location of the Emerald Triangle protected forests complex along the borders of Thailand, Lao PDR and Cambodia.

Figure 1Location of the Emerald Triangle protected forests complex along the borders ofThailand,LaoPDRandCambodia.

Thai J. For. 33 (3) : 56-74 (2014)60

Land-use/land-cover Change Detection Basically, there are three main steps to detect land-use/land-cover (LU/LC) change. Thesestepsaredescribedasfollows:

a) Gathering past land-use/land-cover map TherasterLU/LCmapin2003witharesolutionof250mcoveringtheentirestudyareawas obtained from theMekongRiverCommission Secretariat (MRC). The original LU/LC map comprised 14 classes, namely 1) moist evergreen forest, 2) dry evergreen forest, 3)hillevergreenforest,4)mixeddeciduousforest, 5) dry deciduous forest, 6) forestplantation,7)rubberplantation,8)oilpalmplantation,9)cashcrop,10)paddyfield,11)settlementandinfrastructure,12)baresoilandmiscellaneouslanduses,13)rockoutcrop,and14)waterbody.Theoriginal14LU/LCclassedweregeneralizedto9classesbycombiningsome classes and renaming them as evergreen forests (moist, dry and hill evergreen forests), arableland(paddy,cashcropandoilpalm),andbaresoilandrockoutcrop.ThisisduetothecurrentDyna-CLUEversionisapplicablenotmore than 11 classes (VerburgandOvermars,2009), for land-use transitions at landscape level. Duringfield reconnaissance, itwasfoundthatoilpalmwasrecentlyintroducedandmainlyplanted in paddyfield in smallpatch(lessthan250-mresolution).Thus,bothclassesshowsimilarimagesignaturesandtheyaredifficulttodiscriminate.

b) Preparing current land-use/land-cover map ThecurrentLU/LCmapin2013was interpreted from satellite images.Therelatively cloud-free Landsat-8 TM imageries

weredownloadedfromUSGS(UnitedStatesGeological Survey Department - http://earthexplorer.usgs.gov/).Asub-sceneofimagespath/row126/49and126/50dated8October2013and26October2013,respectively,wasextractedandgeometricallyregisteredtotheUTM coordinate system WGS Zone 48 using thetopographicmapatscale1:50,000.Then,thesetwosub-scenesweremadeintoamosaicusingERDASImaginesoftwareandfalsecolorcompositeimages(bandcombination453–RGB)wereproducedforvisualinterpretationbasedon tone, shape, size,pattern, texture,shadowandassociation(Lillesandetal.,2004).DuetounavailabilityofreferencedatafromtheNBCAsinLaoPDR, key image features ofLU/LCtypesweresampledfromthePPFCinThailandandthePVPFinCambodia. The current research used a contingency tableorclassificationmatrixtoquantifytheagreement between the interpreted classesandtheknownclassesofLU/LCmap(Foody,2002). Omission and commission errors for each LU/LC class, overall accuracy, and the kappastatisticwerecalculated(Jensen,1996).ThenumberofsamplesclassifiedinpreliminaryLU/LCclasseswereselectedusingstratifiedrandomsamplingscheme.Thetotalnumberof sample locations for arable land, rubberplantation, forest plantation, dry dipterocarp forest,mixeddeciduousforest,dryevergreenforest,settlementandothersclasseswere238,38,14,43,42,61,53and19,respectively.

c) Assessing annual change rate The annual rate of LU/LC change wasdeterminedbyusingthedeforestationrate(DR)equationfromyearP(start)untilyearN(endyear)(Trisurat,2009),asbelow:

Thai J. For. 33 (3) : 56-74 (2014) 61

DR(%)= − [1− (NP)1/t]×100 where DR = Annual rate of change N = Land-useofendyear P = Land-use of start year t = Timeperiod;t2–t1 (10 years)

Land-use Change Modeling The Dyna-CLUE model requiresfour inputs to allocate a set of conditions and possibilitiesofLU/LCpatterns:(1)land-userequirements,(2)locationcharacteristics,(3)spatial policies and restrictions and (4) land-usetype-specificconversionsettings(Verburgand Overmars, 2009).Landuserequirementswerecalculatedattheaggregateleveljointlydefinedby50multi-stakeholdersofthethree

participating countries attending the Joint Training Workshop on GIS Modeling for Forest LandUsePlanningduring10-15March2014inCambodia.Thereweresuperintendentsofprotectedareas,governmentofficials,NGOsrepresentatives and lecturers from universities. Using a two-dimensionalmatrix todevelop the LU/LC scenarios (Van der Heijden, 1996), theworkshopparticipants identifiedpopulation growth as an important factor, and economic growth as a result of the ASEAN Economic Community (AEC) in 2015 scheme as a critical uncertainty, to drive four LU/LC. Inaddition,theyjointlydefinedfourLU/LCscenariosfor theperiodfrom2013to2030(Figure2).Thedefinitionsof the scenariosaredetailedasbelow.

2

Figure 2 Land allocations for four scenarios in 2030.

Figure 3 Land use/land cover in 2003 and 2013 in the Emerald Triangle Protected Forest Complex.

Figure 2Landallocationsforfourscenariosin2030.

a) Low economic decline and localized resource degradation (business as usual): a continuation of land transformation of recent

years (2003-2013) is foreseen. The recent land-use change detection revealed that only limitedencroachmentwasobservedinsidethe

Thai J. For. 33 (3) : 56-74 (2014)62

PPFC (Protected Areas Region 9, personal communication). Therefore, this scenario definesthePPFCasrestrictionareas.

b) Unsustainable economic development and serious resource degradation scenario: it ispredictedthatthecontinuoushighrubberprices andhighpopulation growthmake itprofitabletotransformlargeforestlandandbaresoiltopaddyfieldandeconomiccrops.

c) Sustainable poverty and stable resources scenario: a lower rate of landconversionisassumedduetolowpopulationgrowth and the delayof theAEC scheme,meaning limited deforestation. In addition, this scenario anticipates effective protection of remainingforestinallexistingandproposed

protected areas. d) Sustainable development and limited resources degradation scenario: a relative land conversionrateappliesforrubberplantation.Limited forest encroachment for agriculture outside protected areas and along inner and outer buffer zones ofDongKhanthung isassumedduetolowpopulationgrowth. The Dyna-CLUE model determines the location preferences of the different LU/LCclassesbasedonlogisticregressionmodels(VerburgandVeldkamp,2004),whichdefinetherelationbetweenoccurrenceofaparticularLU/LC type and the physical and socio-economic conditions of a specific location(location factors):

Log(pi) = ln (pi)/(1-pi)=β0+β1X1+β2X2+…+βnXn

where,piistheprobabilityofagridcell for the occurrence of the considered LU/LC type and theXi parameters are the explanatoryfactors.Thecoefficients(βi) are estimated through logistic regression using theoccurrenceoftheLU/LCin2013asthedependentvariable. The physical factors to indicate the preferenceforaspecifictypeoflandusewerealtitude,slope,aspect,distance toavailablewater, annual rainfall, rainfall inwettestquarter,rainfallinthedriestquarterandsoilcharacteristics. In addition, the socio-economic factorsinfluencingland-usechangeweredistanceto district, population density and distance to mainroad.Topographicvariablesof100-mresolutionweregatheredfromAdvancedSpaceThermalEmissionandReflectionRadiometer(ASTER)archive(http://asterweb.jpl.nasa.gov/data.asp).Roadandstreamnetworksanddistrict

locationwereupdatedfromtopographicmapsatscale1:50,000andLandsat-8TM,andlatertheywereinterpolatedtoobtainproximitytoroad,proximity to streamandproximity tocity. Climaticvariablesofapproximately1-kmresolutionweredownloadedfromtheWorldclimatedatabase(http://www.worldclim.org/download),whilesoilmapandpopulationdensitywereobtained from theMRC.Thepixelresolutionof250mwasselectedforthisresearchbecauseitwassuitableforlandscapescale and relevant to original LU/LC map in 2003andthedrivingfactors.Inaddition,thegoodness-of-fitofalogisticregressionmodelwas evaluatedusing the receiver operatingcharacteristic(HosmerandLemeshow,2000).The value of the area under curve (AUC) ranged between0.0(completelyunfit)and1.0(perfectfit),whereAUC0.5iscompletelyrandom.

Thai J. For. 33 (3) : 56-74 (2014) 63

Generally, elasticity is estimated based on capital investment and expertjudgment, ranging from 0 (easy conversion) to 1(irreversiblechange)(VerburgandVeldkamp,2004).Inthisstudy,theelasticityvalueswereobtainedfromtheprobabilitytransitionmatrixofland-usebetween2003and2013.In addition, waterbodyandsettlementandinfrastructureclasseswere assigned “not possible to betransformed”. In addition, a minimum of 10 yearswas assigned in the transition-settingas a requirement for thenatural successionof reforestation to forest class and 20 years wasspecifiedforsuccessionfromabandonedagriculturebacktoforestcover,basedonaprevious study (Sahunaru et al., 1993) andimage signature. Whenallinputswereprovided,theDyna-CLUEmodelcalculatedthetotalprobabilityforeachgridcellofeachlandusetypebasedonthesuitabilityoflocationderivedfromthelogit model, the conversion elasticity and the competitive advantage of the location. The totalallocatedareaofeach landuseequalsthe total land requirements specified in thescenario(VerburgandVeldkamp,2004).

Assessing Landscape Configuration Features TheFRAGSTATSver.3.0software(McGarigal andMarks, 1995)wasused toassess landscape structure and fragmentation indicesofforestclasses,arablelandandrubberplantationin2013and2030intermsoftotal

area, number of patches,meanpatch size,largestpatchindex,shapeindex,totalcoreareaandmeancorearea.Aneight-neighborrulewasusedtoidentifypatchinthelandscape.Inaddition,acoreareawasassignedfrom1kmfrom the perimeter. These landscape indices imply direct and indirect impacts of forest fragmentationonbiodiversity(Turneret al., 2001;TrisuratandDuengkae,2011).

RESULTS AND DISCUSSION

Land-use/Land-cover Change between 2003 and 2013 The classificationmatrix (Table 1)showsthatfivesamplelocationsofarablelandclassweremisclassified as drydipterocarpforest.Thisisbecausesomesmallwoodlotsstillremaininnewpaddyfields,particularlyinthenorthernplainsofCambodia.Inaddition,seven locations of dry dipterocarp forest were interpretedasmixeddeciduous forestduetosimilarimagesignaturesofthesetwoclasses.This effectwas also foundduringfieldsurveyofyoungrubberplantationthatwasincorrectlyinterpretedasarableland.Allsamplelocationsofevergreenforest,waterbodyandhumansettlementandinfrastructurewerecorrectlyclassified.TheoverallaccuracyfortheclassifiedLU/LCmapin2013was94%andkappastatisticvaluewas0.91,whichwashighlyreliableandacceptableformostremotesensingscientists(Jensen,1996).

Thai J. For. 33 (3) : 56-74 (2014)64

Table 1 Contingencytableresultingfromfieldvalidation

LU/LC class

Interpreted class

Ara

ble

land

Rub

ber

plan

tatio

n

Plan

tatio

n

Bar

e so

il

Dry

ipte

roca

rp fo

rest

Mix

ed d

ipte

roca

p fo

rest

Dry

Eve

rgre

en fo

rest

Wat

er

Sett

lem

ent &

infr

a.

Tota

l

Prod

ucer

’s A

ccur

acy (

%)

Om

issi

on e

rror

(%)

Kno

wn

clas

s

Arableland 224 5 229 98 2

Rubberplantation 7 35 1 43 81 19

Plantation 3 13 16 81 19

Bare soil 8 1 9 89 11

Dry diptercarp forest 7 38 2 47 81 19

Mixeddeciduousforest 40 7 47 85 15

Evergreen forest 54 54 100 0

Water 10 10 100 0

Settlement & Infra. 53 53 100 0Total 238 38 14 8 43 42 61 10 54 508

User’saccuracy(%) 94 92 93 100 88 95 89 100 98 94

Commissionerror(%) 6 8 7 0 12 5 11 0 2

The spatial analysis and change detectionindicatedtheextentofLU/LCclassesasshowninTable2andtheirdistri-butionpatternsareasfollows: Evergreen forest: In the ETFC, moist evergreenforestisfoundalongstreamnetworkwhere soilmoisture is high all year round(Marod, 2003). In addition, dry evergreenforest is dominant among evergreen forest classesmainlyfoundinPhuJong-NaYoi,YotDom and the core area of Dong Khanthung (Figure3).In2003,evergreenforestcovered21.36%oftheETFC,anditslightlyincreasedto 21.71%, or approximately 9,000 ha, in2013.Thismaybeduetomorerainfallaswas

recordedin2013(DepartmentofMeteorology)or,maybebecausetheLU/LCmapin2013wasinterpretedfromLandsat-8TMtakeninOctober2013,whichwasearlydryseason.Therefore,therewerecommissionerrorsfromthesignatureofdensemixeddeciduousforest(Table1)andfloodeddrydipterocarpforestsituatedinDongKhanthung.Approximately60%ofthetotalevergreenforestwasfoundinside protected areas. Mixed deciduous forest: This forest typeexistsinPhouXiangThong,PhaTaem,KaengTanaandalongtheescarpmentbetweenThailandandLaoPDR.Smallpatcheswerescattered in the southern part of the ETFC

Thai J. For. 33 (3) : 56-74 (2014) 65

landscape.Thisforesttypecovered245,412hain2003andsubstantiallydeclinedto226,573hain2013.Thereductionratewas7.68%in10yearsor0.80%annually.Table2showsthatthedeforestationrateofmixeddeciduousforestinprotectedareaswasgreatlylowerthanthe entire ETFC landscape due to protection measures. Dry dipterocarp forest: This forest type usuallyoccursondryshallowandlateriticsoils.ItisnowdominantinPhaTaem,KaengTana,PVPF and Dong Khanthung. In Thailand, it used tobeabundantalongthebufferzoneofPPFC(Trisurat,2003).In2003,thedrydipterocarpforestcoveredanareaof540,678haor20.92%oftheETFClandscape.Approximately30%ofdrydipterocarpforestshrunkduring2003-2013andlarge-scaleconversionwasobservedbothinsidethePVPFandthePhouXiangThongNBCAandinthebufferzones.Thisfindingwasconsistentwiththeresultofforestcoverassessment in PVPF that indicated that the total forestcoverdeclinedfrom97.62%in2002to95.33%in2010,equivalentto4,353ha(Sobonet al., 2014). The primary manifestation of that changewasconcentratedindeciduousforest. Forest plantation: EstablishmentofforestplantationshasbeenmainlyconductedinThailandbytheForestIndustrialOrganizationandDepartmentofNationalParks,WildlifeandPlant Conservation, for economic purposes and naturalrehabilitation,respectively.Eucalyptus sp. is a common plantation species and most plantationsweresituatedalongthebufferzoneof Bun Thrarik-Yot Mon Wildlife Sanctuary. Although the annual increment rate of forest plantationsduring2003-2013wasgreaterthan3%intheentireETFClandscapeandgreaterthan6% in protected areas, it covered lessthan1%ofthetotalstudyarea.

Para rubber plantation: Thisisanewcultivation practice in the ETFC landscape. However,ithasincreasedrapidlyinthelastdecadeinThailandandnowexpandedtoLaoPDRandCambodia.In2003rubberplantationcovered85,456ha,butin2013itincreasedto164,225haor92%duringthisperiod.Inaddition,theannualincrementrate(6.75%)wasthehighestamongnineLU/LCclasses.MostplantationsweresituatedinthebufferzoneofBunThrarik,PhuJong-NaYoiandYotMon.AfewpatcheswereobservedinsideBunThrarik-Yot Mon. The result of the LU/LC transformationmatrixrevealedthatsomepaddyfieldsandcashcropareaswereconvertedtorubberplantationduetotheincreasingrubberprices in the last decade. Arable land:Arableland,includingpaddyfield,cashcropandoilpalm,iswidespreadinthe ETFC landscape. Although oil palm wasintroducedasaneweconomiccroprecently,itisanticipatedthattheextentofoilpalmwillnotcoverlargeareaasrubberorothercashcropsdue to the constraints of soil characteristics andclimaticconditions.In2003,arablelandconstitutedabout37%andslightlyincreasedto41%in2013.Table2showsthatthearableland inside protected areas greatly increased from4%to12.7%duringthesameperiodasa result of forest destruction in Phrea Vihear (Figure 2). Bare soil and rock outcrop: This includes adrawdownzonealongtheriverbanksandrock outcrops in the marginal land or unfertile soil.Theareaofover2,000ofthisclasswaschangedtootherclasses,particularlyarableland,rubberplantationandsettlement.Thereare82villagessituatedwithina3-kmbufferof the PPFC and 4 villages are located inside the PPFC (Trisurat, 2007). In the last 10 years

Thai J. For. 33 (3) : 56-74 (2014)66

(2003–2013),human settlement areas had expandedandthenumberoflocalresidentsincreasedfrom49,324to65,016individuals.Water body class includes reservoirs, ponds andmajorrivers.Thetotalareaofwaterbodywasvirtuallystablebetween2003and2013.Afewhundredshas increasedin theETFClandscape and less than 100 ha decreased in protectedareasduetothefluctuationofwaterlevelduringwetanddryseasons.

Projected Land-use/Land-cover in 2030 Thesignificantfactorsandcoefficientsof the logistic regression models that determine the location suitability of the eightLU/LCclassesareshowninTable3.Waterbodywasexcludedbecauseitwasdeterminedasstablein the land demand scenarios (Figure 2). It is notedthateachdrivingfactorcontributedto

different LU/LC types. High altitude, steep slope, high annual rainfall and further distance fromcityandstream,aswellasdifficultytoaccess by roadwere positively correlatedwithremainingevergreenforest.Incontrast,areasthatwereclosetothestreamandmaincity,situatedonfertilesoil,accessiblefrommainroads,andatlowaltitude,wereaprimetargetforagriculture.Aspectisaconsiderablefactoronlyforrubberplantationinthelogisticregressionmodel.Ingeneral,rubbertreecangrowinallaspectdirections(Ranstet al.,1996)butthemodelresultsindicatedthatthegreaterthedegreesinaclock-wisedirection,themoresuitabletheaspectisforrubber.ThisisduetothefactthatmostexistingrubberplantationareasaresituatedinthebufferzonesofthePPFC(Figure3).TheareastotheeastofPPFCare mountainous landscape.

Table 2 LU/LCclasses in 2003 and 2013 in theEmeraldTriangle landscape and protectedareas (ha) and changes.

Type of land-use 2003 2013 Change Change in %ha % ha % ha (+/-) 10 yrs. Yearly

Evergreen forest 552,1121 21.36 561,104 21.71 8,992 1.63 0.16320,8132 49.39 325,719 50.16 4,906 1.53 0.15

Mixeddeciduousforest 245,412 9.49 226,573 8.77 -18,839 -7.68 -0.80106,563 16.41 103,856 15.99 -2,707 -2.54 -0.26

Dry dipterocarp forest 540,687 20.92 374,337 14.48 -166,350 -30.77 -3.61179,731 27.67 121,788 18.75 -57,943 -32.24 -3.82

Forest plantation 9,850 0.38 13,475 0.52 3,625 36.80 3.18119 0.02 231 0.04 112 94.12 6.86

Pararubber 85,456 3.31 164,225 6.35 78,769 92.18 6.754,381 0.67 3,856 0.59 -525 -11.98 -1.27

Arableland 965,087 37.34 1,058,836 40.96 93,749 9.71 0.9326,575 4.09 82,756 12.74 56,181 211.4 12.03

Settlement 77,700 3.01 80,365 3.11 2,665 3.43 0.34913 0.14 1,500 0.23 587 64.29 5.09

Bare soil & rock outcrop 25,549 0.99 23,293 0.90 -2,256 -8.83 -0.926,075 0.94 5,288 0.81 -525 -11.98 -1.27

Waterbody83,050 3.21 82,697 3.20 -353 -0.43 -0.044,356 0.67 4,419 0.68 63 1.45 0.14

Total: Emerald Triangle 2,584,903 100.00 2,584,903 100.00 0.00 0.00 0.00Protected areas 649,413 100.00 649,413 100.00

Remarks: 1EntireEmeraldTrianglelandscape;2withinprotectedareas.

Thai J. For. 33 (3) : 56-74 (2014) 67

Tabl

e 3

Betavaluesofsignificantlocationfactorsforregressionmodelsrelatedtoeachlandusetype.

Vari

able

sE

verg

reen

fo

rest

Mix

ed

deci

duou

s fo

rest

Dry

di

pter

ocar

p fo

rest

Plan

tatio

nR

ubbe

rA

rabl

e la

ndSe

ttle

men

tB

are

soil

DEM

(m)

0.00

2ns

-0.013

0.015

0.01

1-0.003

0.003

0.00

7Slope(%

)ns

0.103

ns-0.096

-0.1

18-0

.120

ns-0.035

Asp

ect

nsns

nsns

0.00

1ns

nsns

Popu

latio

n de

nsity

(per

son/

km2)

-0.039

-0.0

01<-

0.00

1-0

.001

-0.0

01<-

0.00

10.

002

nsA

nnua

l rai

nfal

l (m

m)

0.00

9-0.006

-0.003

0.01

9-0.006

-0.0

070.

008

-0.015

Rainfall inthewettestquarter(mm)

-0.0

070.

008

0.00

7-0.023

0.00

80.006

-0.0

110.016

Rainfallinthedriestquarter(mm)

-0.045

0.152

0.29

0-0.384

-0.1

79-0

.102

-0.1

22ns

Dis

tanc

e to

road

(m)

<0.0

018.9E-05

5.7E-05

-0.0

01-0

.001

-0.0003

-0.003

<-0.

001

Dis

tanc

e to

stre

am (m

)8E-05

0.6E-05

5.1E05

<-0.

001

-0.0

01<-

0.00

1<-

0.00

1ns

Dis

tanc

e to

city

(m)

3.8E05

-0.3.2E-05

1.1E-05

nsns

1.6E-05

-3.0E-05

-3.4E-05

Acr

isol

soil

3.222

0.437

2.332

0.763

0.667

1.590

-0.366

0.74

4A

reno

sol s

oil

1.503

ns1.311

0.735

ns2.

174

-0.513

nsCam

bisol/Plinthosolso

il3.503

0.572

2.446

ns-1.594

1.383

-0.628

nsFe

rral

sol s

oil

3.554

nsNs

-1.836

0.834

1.161

nsns

Gle

ysol

/Flu

viso

l soi

l2.168

0.92

91.916

nsns

1.836

-1.161

nsLe

ptos

ol so

il 4.786

ns2.

099

nsns

1.933

nsns

Lixixolsoil

2.893

nsns

ns1.664

1.673

nsns

Luvisol/Solonetzsoil

2.495

ns3.001

ns-2

.447

1.77

7ns

nsSlopecomplex

2.475

ns4.245

ns1.736

ns-0

.740

2.22

2R

ock

3.660

ns3.512

-3.355

-1.326

Ns

-1.7

180.

997

Con

stan

t-11.252

-0.232

-8.307

-5.176

4.12

07.532

1.48

98.

008

AU

C0.

902

0.758

0.767

0.837

0.80

20.815

0.903

0.79

7

Not

es:A

UC=areaundercurve;ns=

notstatisticallysignificant

Thai J. For. 33 (3) : 56-74 (2014)68

Table 4 Landscapeindicesofremainingforesttypes,rubberplantationandarablelandunderdifferent LU/LC scenarios.

ScenariosTotal area

(km2)

No. of patches

Mean patch

size (ha)

Largest patch index (%)

Shape index

Total core area

(ha)

Mean core area

(ha)

Baseline(2013)EvergreenMixed Dec. For.Dry Dip. For. Rubber plt.Arable land

5,4742,1693,5691,61710,308

5,6787,5887,54135605415

120294745

190

13.801.941.580.70

27.84

1.211.151.251.221.24

69,768119500

68,887

15<1<1<113

LoweconomicdeclineEvergreenMixed Dec. For.Dry Dip. For. Rubber plt.Arable land

5,2771,9382,6372,34810,733

3,4975,5844,69247135080

151355650

211

13.792.310.731.0328.55

1.251.171.281.221.24

70,1257550737

79,575

20<1<1<116

UnsustainableeconomicEvergreenMixed Dec. For.Dry Dip. For. Rubber plt.Arable land

5,2781,8811,9122,459

11,218

3,3005,4093,41447004664

160355652

240

14.32.240.631.0128.67

1.251.171.271.211.24

73,2067525743

86,500

22<1<1<118

SustainablepovertyEvergreenMixed Dec. For.Dry Dip. For. Rubber plt.Arable land

5,4732,0533,1011,969

10,449

4,2376,6945,4924,3505,311

130315645

197

13.951.951.570.9128.31

1.221.161.291.221.24

72,0681005012

75,856

17<1<1<014

SustainabledevelopmentEvergreenMixed Dec. For.Dry Dip. For. Rubber plt.Arable land

5,4622,0532,614,2404

10,448

3,6496,4384,6054,9115,208

146325749

201

14.241.960.741.0928.05

1.251.161.291.221.24

78,90610037

1,7817,290

22<1<1<114

According to HosmerandLemeshow(2000),thepredictedmodelswereoutstandingfor evergreen forest and settlement (AUC>0.9), excellentforforestplantation,rubberplantationandagriculture(0.8≤AUC<0.9),andacceptableformixeddeciduousforest,drydipterocarpforestandbaresoil&rockoutcrop(0.7≤AUC<0.8).

Thisisbecauseevergreenforestandhumansettlementweremainlyrestrictedandclusteredin certain areas,while other classeswerewidely distributed in theETFC landscape.ThegradientofAUCvaluesshowedsimilaragreementwith theaccuracyassessmentoftheinterpretedLU/LCclasses(Table2).

Thai J. For. 33 (3) : 56-74 (2014) 69

ThesimulatedLU/LCmapsin2030forthefourscenariosareshowninFigure4.The results of the low economic decline and localized resource degradation (business as usual)scenariowithrestrictionpolicyinthePPFCshowthatfuturedeforestationforagriculturewaspredictedintheremnantforestssituatedinthebufferzonesofthePPFCandareasclosetotheChong-Mekbordercheckpoint (Figure 4a). In addition, substantialamount of forest cover in thePhouXiangThong and to the north of Dong Khanthung were predicted to be converted to rubberplantations.InThailand,expansionofrubberplantationswaspredictedinthewestoutsidethe Pha Taem national park and close to road networkcurrentlycoveredbycashcropandmixeddeciduousforest. The unsustainable economic development and serious resource degradation scenario predicteda lotof landconversiontoarablelandandrubberplantation.Theareaofmixeddeciduous anddry dipterocarp forestswaspredictedtodeclinefrom22.9%oftheentire

ETFCin2013to15.1%in2030.Incontrast,rubber plantation areawas anticipated toincrease50%fromthecurrentstatus(Figure2).Figure4bshowsthatnewarablelandisalso predicted in thewest ofPVPF,whichis close to the Preah Vihear Temple cultural worldheritagesite. The sustainable development and limited resources degradation scenario (Figure 4d) predicted similar land-use patterns as the business as usual scenario (Figure 4a). High deforestationwasfoundinthenorthofDongKhanthungandtothewestofPhaTaem,butlimited areas in all protected areas. Finally, the sustainable poverty and stable resources scenarioshoweddifferentland-usepatternsthan other scenarios. This scenario assumed less demandforagricultureandrubberplantationsduetolowpopulationgrowthandthedelayof AEC implementation scheme, leading to limited deforestation. A small amount of land conversiontorubberplantationwasexpectedoutside all existing andproposedprotectedareas (Figure 4c).

2

Figure 2 Land allocations for four scenarios in 2030.

Figure 3 Land use/land cover in 2003 and 2013 in the Emerald Triangle Protected Forest Complex.

Figure 3Location of the Emerald Triangle protected forests complex along the borders ofThailand,LaoPDRandCambodia.

Thai J. For. 33 (3) : 56-74 (2014)70

3

Figure 4 Predicted new areas for arable land and rubber plantation in 2030.

Figure 4Predictednewareasforarablelandandrubberplantationin2030.

Implications for Trans-boundary Biodiversity Conservation Land-use conversion as a result of expansionofhumansettlementsandpermanentcropsdoesnotonlydiminishsuitablehabitats

forwildlifespecies,butalsoitcauseshabitatfragmentation, reducedpatch size and corearea,andisolationofsuitablehabitats(Turneret al.,2001;MacDonald,2003;TrisuratandDuengkae, 2011). Theseconsequentialeffects

Thai J. For. 33 (3) : 56-74 (2014) 71

wereveryclearfordrydipterocarpforestinwhichthetotalareawaspredictedtodecreasefrom3,569km2in2003to2,637km2 in2013.Thenumberofthreatenedpatcheswasover4,000andmostofthesewereremnantpatchesdistributedoutsideprotectedareasacrosstheETFClandscape(Table4). In addition, the results of landscape analyses revealed that the total core area of drydipterocarp forest declined from50hain2003to25hain2013.Incontrasttotheseresults,evergreenforestwaslessthreatenedunder all scenarios because the remainingareas are situated either in protected areas or inhigh steep slope,whichbecomebarriersfor encroachment for agriculture (Trisurat, 2007). Inaddition,Table4alsoshowsthattheextentandaveragepatchsizeofarablelandandrubberplantationaregreaterin2030forallmodeledscenarios,exceptthesustainable poverty scenariowhere small landholder subsistence farmers practice agriculture fortheir daily livelihoodnot drivenbymarketprices. ThedestructionoflowlanddrydeciduousforestsituatedinLaoPDRandCambodiawillcause significant impacts to iconicwildlifespecies (e.g., Eld’sdeer,giantibis,Saruscrane),large herbivores (Asian elephant, banteng,gaur) andmedium-to-large sizedmammalsin the ETFC landscape. Population densities anddiversityoflargeherbivoresaregreaterinseasonaldryforesthabitatsthaninrainforeststhatarecharacterizedbyclosedcanopiesandtallgrasses(Sukumar,2003;McSheaandBaker,2011).Round(1998)foundthatthelowlanddry diptercarp forest in the Dong Khanthung forestreservesupportsmorewildlifespeciesthan evergreen forest. The results of land-

usemodelingarebeingusedasexplanatoryvariables in the distributionmodeling ofwide-rangingspeciesandasimportantinputfor preparationof collaborative frameworkfortrans-boundarybiodiversityconservationin the ETFC landscape.

CONCLUSION

TheLU/LCchangeassessmentbetween2003and2013indicatedthatapproximately30%ofremainingdrydipterocarpforestbothinside protected areas and the entire ETFC landscapewas converted to other land-useclasses.RubberplantationexpandedsubstantiallyinthebufferzonesofthePPFC.FutureLU/LCintheETFClandscapewouldbedrivenbypopulationgrowthandtheimplementationof the AEC scheme. The sustainable poverty and stable resources scenario predicted a small amount of rubberandarablelandexpansion.Allprotectedareas are secured from future deforestation. The low economic decline and localized resource degradation scenario indicated future deforestation for agriculture in the remnant forestsinthebufferzonesofthePPFCandinLaoPDRclosetothebordercheckpoint(Figure4a).Thetworemainingscenarios,unsustainable economic development and serious resource degradation and sustainable development and limited resources degradation,showedsimilarland-usepatternsbut greater extentof newarablelandandrubberplantation,especiallyfor the unsustainable economic development. Large conversion of dry dipterocarp forest in thePVPFwas expected as the result ofland allocation program and infrastructure development for tourism activities.

Thai J. For. 33 (3) : 56-74 (2014)72

It is important to note that the projected LU/LCpatternsshouldnotbeinterpretedtobetheactualpredictionsofwherealand-usetypewillbe in the future,becauseofsomelimitations and uncertainties in the models used for the predictions (e.g., scale dependent, futuresocio-economicconditions).However,the general patterns emerging from the projections are very useful to inform policy makers and stakeholders of the three participating countriestoproactivelyformulatecollaborativeframeworktopreventfuturedeforestationinrisk areas and to put more efforts for conserving important habitats andmigratory routes oftrans-boundaryspecies.Themodeledoutputsarebeingemployedtopredictthedistributionsofwide-rangingspeciesintheETFClandscapeasoutlinedintheobjectiveofITTOprojectphase III.

ACKNOWLEDGEMENTS

ThisresearchwasconductedundertheITTOprojectPD577/10Rev.1(F),Managementof the Emerald Triangle Protected Forest ComplextoPromoteCooperationforTrans-boundaryBiodiversityConservationbetweenThailand,CambodiaandLaoPDR(PhaseIII)withfinancialsupportfromtheGovernmentofJapan.TheauthorsalsowishtothanktheSuperintendent and staff Thailand, Lao PDR andCambodiaparticipatinginthisprojectfortheir support. We are grateful to the Mekong River Commission Secretariat, Royal Forest DepartmentandDepartmentofNationalParks,Wildlife and Plant Conservation for providing thedata.TheauthorsthankDr.Hwan-okMa,Mr. Chheang Dany and Dr. Dennis J. Cengel fortheircontributions.

REFERENCES

Baker,W.L. 1989.A reviewofmodels oflandscape changes. Landscape Ecology 2(2):111-133.

Ballestores, F.Jr. and Z. Qiu. 2012. An integrated parcel-basedlandusechangemodelusing cellular automata and decision tree. Proceedings of the International Academy of Ecology and Environmental Sciences 2(2):53-69.

Bhumpakphan,N.2003.Management of the Pha Taem Protected Forest Complex to Promote Cooperation for Trans-boundary Biodiversity Conservation between Thailand, Cambodia and Laos (Phase I): Wildlife Ecology Final Report. Faculty of Forestry, Kasetsart University, Bangkok, Thailand.

Corlett, R.T. 2012. Climate change in the tropics: theendof theworldasweknow it.Biological Conservation 151:22-25.

Foody,G.2002.Statusoflandcoverclassificationof accuracy assessment. Remote Sensing of Environment80:185-201.

Galt, A., T. Sigaty and M. Vinton. 2000. The World Commission on Protected Areas, 2nd Southeast Asia Regional Forum, Pakse, Lao PDR: Volume I: Executive Summary. IUCN,Vientiane.

Hosmer,D.W. and S. Lemeshow. 2000.Applied Logistic Regression. 2nd ed. Wiley,ChichesterandNewYork.

Houet,T.andL.Hubert-Moy.2006.Modellingand projecting land-use and land-cover changeswithaCellularAutomaton

Thai J. For. 33 (3) : 56-74 (2014) 73

in considering landscape trajectories: An improvement for simulation of plausible future states.EARSeL eProceedings 5(1): 63-76. http://halshs.archives-ouvertes.fr/halshs-00195847,September9,2014.

Jensen, J.R. 1996. Introductory Digital Image Processing: A Remote Sensing Perspective (Second edition). Prentice Hall,Inc.,UpperSaddleRiver,NewJersey, USA.

Kalyawongsa, S. and S.Hort. 2010.Aconservationjewel.ITTO Tropical Forest Update 20(2): 20-21.

Lillesand, T.M., R.W. Kiefer and J.W. Chipman. 2004. Remote Sensing and Image Interpretation, 5th ed. Jonh Wiley andSons,NewYork..

MacDonald, G. 2003.Biogeography: Introduction to Space, Time and Life.JohnWileyandSons,NewYork.

Marod,D.2003.Management of the Pha Taem Protected Forest Complex to Promote Cooperation for Trans-boundary Biodiversity Conservation between Thailand, Cambodia and Laos (Phase I): Forest Ecology Final Report. Faculty of Forestry, Kasetsart University, Bangkok, Thailand.

Matthews,R.,N.Gilbert,A.Roach,J.G.PolhillandN.M.Gotts.2007.Agent-basedland-usemodels:areviewofapplications.Landscape Ecology22(10):1447–1459.

McGarigal,K.andB.Marks.1995.FRAGSTATS: spatial pattern analysis program for quantifyinglandscapestructure.Gen.Tech.Rep.PNW-GTR-351,Portland,Oregon.

McShea, W.J. and M.C. Baker. 2011. Tropical deer in the seasonally dry forests of Asia: ecology, concerns and potential forconservation.pp.165-178.In W.J. McShea,S.J.DaviesandN.Bhumpakphan(Eds.). The Ecology and Conservation of Seasonally Dry Forests in Asia. Smithsonian Institution Scholarly Press, Washington, D.C.

OfficeofEnvironmentalCenter.2005.Greater Mekong Subregion Biodiversity Conservation Corridors Initiative. Asian Development Bank. Manila, Philippines.

Pontius, R., W. Boersma, J.C. Castella, K. Clarke,T.deNijs,C.Dietzel,Z.Duan,E.Fotsing,N.Goldstein,K.Kok,E.Koomen, C. Lippitt, W. McConnell, A.MohdSood,B. Pijanowski, S.Pithadia,S.Sweeney,T.Trung,A.Veldkamp andP.Verburg. 2008.Comparing the input, output, and validation maps for several models of land change. Annals of Regional Science 42:11–37.

Ranst, E.V., H. Tang, R. Groenemam and S. Sinthurathat. 1996.Application offuzzy logic to land suitability forrubber production on peninsularThailand. Geoderma 70: 1-19.

Round, P.D. 1998. Wildlife, Habitats and Priorities for Conservation in Dong Khanthung Proposed National Biodiversity Conservation Area, Champasak Province, Lao PDR. Department of Forestry, Lao PDR.

Sahunalu, P., P. Dhanmamomda, M. JamroenpruksaandC.Khemnak.1993.

Thai J. For. 33 (3) : 56-74 (2014)74

Effects of Reforestation, Abandoned Areas and Natural Forests on Sakaerat Environment. Faculty of Forestry, Kasetsart University, Bangkok.

Sobon,K.,N.BunthanandS.Sinly.2014.Forest Cover Assessment of Preah Vihear Projected Forest. Forest Administration.PhnomPenh,Cambodia.

Sodhi,N.S.,L.P.Koh,B.W.BrookandP.K.L.Ng.2004.SoutheastAsian:animpendingdisaster. Trends in Ecology and Evolution19:654-660.

Sukumar,R. 2003.The Living Elephant. OxfordUniversityPress,Oxford.

Trisurat,Y.2003.DefusingtheTrans-boundaryMinefield. ITTO Tropical Forest Update 13:10-13.

Trisurat,Y.2006.Trans-boundarybiodiversityconservation of the Pha Taem Protected ForestComplex:Abioregionalapproach.Applied Geography26:260-275.

Trisurat, Y. 2007. The Emerald Triangle Protected Forests Complex:Anopportunityforregionalcollaborationon trans-boundary biodiversityconservation in Indochina. In S. Ali (Ed.). Peace Parks: Trans-boundary Issues and Conflict Resolution, pp. 141-162.MITPress,Washington,D.C.

Trisurat, Y. 2009. Application of geo-informatics for trans-boundary biodiversityconservation of the Pha Taem Protected Forest. International. Journal of Terrestrial Observation 1(2): 17-29.

Trisurat, Y., R. Alkemade andP.Verburg.2010. Projecting land use change and itsconsequencesforbiodiversity inNorthernThailand.Environmental Management45:626-639.

Trisurat,Y.andP.Duengkae.2011.Consequencesoflandusechangeonbirddistributionat Sakaerat Environmental Research Station. Journal of Ecology and Field Biology34(2):203-214.

Turner,M.G.,R.H.GardnerandR.V.O’Neill.2001. Landscape Ecology: In Theory and Practice.Springer,NewYork.

VanderHeijden,K.1996.Scenarios: the Art of Strategic Conversation. Wiley Press, Chichester.

Verburg, P.H. andK.Overmars. 2009.Combiningtop-downandbottom-updynamics in land use modeling: exploring the future of abandonedfarmlandsinEuropewiththeDyna-CLUE model. Landscape Ecology 24:1167–1181.

Verburg, P.H. andA.Veldkamp. 2004.Projecting land use transitions at forest fringesinthePhilippinesattwospatialscales. Landscape Ecology19(1):77–98

Wikramanayake, E., R. Boonratana, P. Rundel andN.Aggimarangsee.2000.Terrestrial Ecoregions of the Indo-Pacific: A Conservation Assessment. Island Press, Washington, D.C.