UsingMarkovchainstoanalyzechangesinwetlandtrendsin

MathematicalandComputerModelling54(2011)924–930ContentslistsavailableatScienceDirectMathematicalandComputerModellingjournalhomepage:∗,ChengjieTangb,SuhuaMab,HuiYuana,LinglingGaoa,WenyuFanaaInstituteofAgriculturalRemoteSensingandGIS,CollegeofInformationandElectricalEngineering,ChinaAgriculturalUniversity,TsinghuaEastRoadNo.17,Beijing,100083,ChinabChinaLandSurveyingandPlanningInstitute,XichengDistrict,GuanyingWestern,No.37,Beijing100035,ChinaarticleinfoArticlehistory:Received11August2010Accepted4November2010Keywords:WetlandChangestrendMarkovchainsTransitionprobabilitymatrixRemotesensingYinchuanPlainabstractThreewetlanddistributionmapsweredrawnusingTwoLand-sat5ThematicMapper(TM)imagesfrom1991and1999,andaChina–BrazilEarthResourcesSatellite(CBERS)-02Bimagefrom2006.Atransitionprobabilitymatrixwasconstructedusingtwowetlanddistributionmaps,onefrom1991andone1999andGIS(GeographicInformationSystem).ThetrendsinchangesinwetlandtypesandthedistributionareawerepredictedusingaMarkovmodel.Thepredictionmodelwasthentestedforrelativeaccuracyandthefeasibilityofax2test.Thepredictionmodel’srelativeaccuracywas98.5%.Thex2testresultsshowedthatboththesimulatedresultsandtheactualwetlanddistributionareawereingoodagreement.Therefore,itisfeasibletousethewetlandsareatransfermatrixtoestablishatransitionprobabilitymatrixbasedontheMarkovmodelandtopredictthedistributionpatternofthewetlandinYinchuanPlain.Theresultsofthisstudymaybehelpfulforlocalgovernmentstodevelopwetlandmanagementpolicies.©2010ElsevierLtd.Allrightsreserved.1.IntroductionWetlandsareamongthemostimportantecosystemsonearthandaresometimesdescribedas‘‘thekidneysofthelandscape’’becausetheyfunctionasthedownstreamreceiversofwaterandwastefrombothnaturalandhumansources[1].Wetlandsarevaluedfortheirabilitytostorefloodwaters,protectshorelines,improvewaterquality,andrechargegroundwateraquifers[2].Wetlandsprovideahabitatforfishandwildlife,supportingarichbiodiversity,includingmanythreatenedandendangeredspecies[3].Theareaofwetlandsinaridandsemi-aridareasisthebasisofthemaintenanceofagriculturalandoasisecosystems[4].Inthelastfewdecades,climatechangeandanthropogenicactivityhaveresultedinarapidreductionintheareaofwetlandsinYinchuanplain.Thedegreeandextentofwetlandchangeoverthelastfewdecadeshasgivenanewurgencyandrelevancetothedetectionandunderstandingofwetlandchange[5].Geographicinformationsystems(GIS)andremotesensing(RS)areessentialtoolsformonitoringthepresentwetlanddistributionareaandspatial-temporaldynamicvariety[6,7].GISalsoservesasatoolforpredictingandanalyzingwetlandchangeanditsunderlyingcauses[8].OtherusesofGISincludedeterminingmethodsofefficientstorage,management,andanalysisofspatialandnon-spatialdata[9].Landscapeecologyisaninterdisciplinarystudythatincludesgeographyandecology.Someofthemostimportantaspectsofwetlandchangeatthebroadspatiallevelcanbedistinguishedbasedonlandscapepatternandstructure[10].However,itisnecessarytomonitorandassesswetlandchangeswithaviewtowardtheconservationandwisemanagementofwetlandresources[11].Theuseofremotesensingtechnologyfortheidentification,inventory,mapping,andclassificationoflandwetlandshasbeenacommonapplicationofsatelliteimagery[12,13].Inrecentyears,manystudieshavebeenconductedto∗Correspondingauthor.Tel.:[phone].E-mailaddress:[e-mail](R.Zhang).0895-7177/$–seefrontmatter©2010ElsevierLtd.Allrightsreserved.doi:10.1016/j.mcm.2010.11.017R.Zhangetal./MathematicalandComputerModelling54(2011)924–930925Fig.1.Locationmapofthestudyarea.evaluatewetlandlandscapepatternandstructurechange[8,3],butfewerstudieshavebeenconductedtoevaluatewetlandchangetrendsimulation.AMarkovprocess,likeaMarkovchain,canbethoughtofasadirectedgraphofstatesofthesystem.Thedifferenceisthat,ratherthantransitioningtoanew(possiblythesame)stateateachtimestep,thesystemwillremaininthecurrentstateforsomerandom(inparticular,exponentiallydistributed)amountoftimeandthentransitiontoadifferentstate[14].Inprinciple,whenasequenceofchanceexperimentsisobserved,allofthepastoutcomescouldinfluenceourpredictionsforthenextexperiment[15,16].Thus,acontinuous-timeMarkovprocessbeingsuitableforwetlandsimulationofdynamicchange,wecanusetwotime-statesshowingthewetlanddistributionareatopredictfuturewetlanddistributionareas.Inthisstudy,weusedthreeimages(TMimagefromSeptember1991,TMimagefromSeptember1999,andCBERS-02BimagefromSeptember2006)forremotesensingwetlandmapping.AMarkovtransitionprobabilitymatrixwasthenbuiltusingthewetlanddistributiondatesfrom1991and1999topredictwetlandchangetrendsinthefutureinYinchuanPlain.TheobjectiveofthisstudywastoapplytheMarkovmodelpredictiontochangesinwetlandtrendsandtoanalyzetheunderlyingcausesofwetlandchangeinthearidYinchuanPlain,China.2.Materialsandmethods2.1.StudyareasYinchuanPlainislocatedinthearidportionofwesternChina(Fig.1).Theplainisapproximately165kmfromsouthtonorthand50kmfromeasttowest,coveringanareaofabout7793km2.Theelevationofthestudyarearangesfromapproximately1000–