Monitoringthedynamicsofwetlandinundationbyrandomse

Monitoringthedynamicsofwetlandinundationbyrandomsetsonmulti-temporalimagesXiZhaoa,b,⁎,AlfredSteina,Xiao-LingChenb,caFacultyofGeo-InformationScienceandEarthObservationoftheUniversityofTwente,Hengelosestraat99,Enschede,TheNetherlandsbStateKeyLaboratoryofInformationEngineeringinSurveying,MappingandRemoteSensing,WuhanUniversity,LuoYuroad129,Wuhan,HuBei,ChinacTheKeyLaboratoryofPoyanglakeWetlandandWatershedResearch(MinistryofEducation),JiangxiNormalUniversity,Ziyangroad99,NanchangJiangXi,ChinaabstractarticleinfoArticlehistory:Received26December2010Receivedinrevisedform9March2011Accepted1May2011Availableonline28May2011Keywords:UncertaintyDynamicTransitionzoneRandomsetMixedGaussianWetlandMulti-temporalimageHJimagePoyangLakeGeneralmodelsarerequiredforbetterunderstandingthedynamicsofwetlandinundationsandtheirwaterregimes.Extractionbyimagerysnapshotsanduseofacrispdatamodeldonotvaluetheinherentuncertaintiesinspaceandtime.ThisstudyaddressesparameterizationofamixedGaussianrandomsetmodelinamulti-temporalanalysis.ThemodelisappliedtomonitorannualvariationofwetlandinundationextentsfromaseriesofLandsatTMimagesin2004andHJimagesin2009onthePoyangLakenationalnaturereserve(PLNNR)inChina.Weuserelatedindicestorepresentspatialuncertaintiesofinundatedareasandtodelineatethetransitionzonebetweenwetlandvegetationandopenwater.ThePLNNRwaterregimeisinvestigatedbyaccumulatingaseriesofrandomsetsduringoneyearanddeterminingthewatercoveringdays(WCD)atthepixellevel.RandomsetsprovidedetailedspatialconfigurationsoftheWCDwhichhasastrongnegativecorrelationwiththeunderwaterDEM.Comparing2004and2009,thestudyshowsthatalmosthalfofthePLNNRareaexperienceddrought.WeconcludethatthemixedGaussianrandomsetmodelwiththreecomponentspresentedinthisstudyservesasageneralmethodtoparameterizetherandomsetmodelforlargedatasets.Moreover,itiswellsuitedtocapturedetailedinformationonspatialtemporaldynamicofwetlandinundationandcontributestoourunderstandingofwetlandswaterregimesfrommulti-temporalimages.©2011ElsevierInc.Allrightsreserved.1.IntroductionWetlandsareconsideredasadiverseecosystemforprovidingwildlifehabitats.Theyhaveimportantfunctionsinminimizingtheeffectsoffloodinganderosionaswellasinrechargingofgroundwater.Awetlandisusuallycharacterizedbyadominanceofwetlandplants,soilssubjectstowaterlogandindicatorsofwetlandwaterregime(Lyon,2001).Remotesensingimageshavebeenusedtoobtaininformationaboutthesecriteriaforidentifyingwetlands,detectingtheirchangesovertimeandmonitoringtheeffectsofhumaninfluences(e.g.,Harris&Bryant,2009;Jensenetal.,1993;Kindscheretal.,1998;Martinez&Toan,2007;Poulinetal.,2010).Wetlandsexperiencingperiodicfloodingexhibitchangesinthespatialdistributionandtemporaldurationofinundationareas.Abetterunderstandingofwetlanddynamicsisimportantforthewetlandecosystemmonitoringandhabitatassessment(Lyon,2001).Observingwetlandsfromsatelliteimagesislimitedbyspatialresolutionsandrevisitingintervalsofsensors(Ozesmi&Bauer,2002).Moreover,conventionaldataprocessingsuchashardclassificationanduseofthecrispdatamodelsimplifiestheextractionandrepresentationofwetlands.Exploringmoregeneralmodelsthatcanbetterrepresentspatiallycontinuousandtemporallydynamicphenomenaandtheiruncertainties,therefore,becomesanattractivetopicinspatialtemporalmodelingandchangeanalysis(Steinetal.,2009).ExamplesarefoundinmodelingsalinewetlandinSpain(Schmidetal.,2005),savanna-forestintergradeinBolivia(Fisher,2010),coastallandscapeintheNetherlands(vandeVlag&Stein,2006),glaciersintheTibetanplateau(Yeetal.,2009)andtundra–taigatransitionareasinRussia(Ransonetal.,2004).Mappingtheextentsofwetlandsisabasicpracticeforwetlandinventoryandchangetracking.Itcouldsuffer,however,fromproblemsinidentifyingtheuncertainboundarythatappearsasasetofmixedpixelsinimages(Ozesmi&Bauer,2002).Duetoregularinteractionsbetweenwaterandvegetationaswellasgradualelevationgradients,thetransitionchainfromdensevegetationthroughsparselydistributedvegetation,saturatedsoil,shallowandturbidstandingwatertoopenwateriscommonandtypicalinawetlandthatisperiodicallyflooded.Toextractinundatedareasinsuchanaturallandscapefromimages,wethusneedtoconsiderwidetransitionzonesfromopenwatertovegetation.ThespectralsignaturesofwaterarecondensedinthespectralfeaturespaceandthereforeopenwaterisfoundeasilyseparatedfromvegetationRemoteSensingofEnvironment115(2011)2390–2401⁎Correspondingauthorat:Hengelosestraat99,7500AAEnschede,TheNetherlands.Tel.:[phone];fax:[phone].E-mailaddresses:[e-mail],[e-mail](X.Zhao).0034-4257/$–seefrontmatter©2011ElsevierInc.Allrightsreserved.doi:10.1016/j.rse.2011.05.002ContentslistsavailableatScienceDirectRemoteSensingofEnvironmentjournalhomepage:(Ozesmi&Bauer,2002).Inundatedwetlands,however,havelesscondensedspectralsignatures,duetothetransitionzonesfromopenwatertovegetationwhichcontainthemixtureofshallowwater,wetsoil,flooddebris,deadflattenedalgaeandsparselydistributedyoungflushsedge.Thesetransitionzonescauseawidespreadofspectralsignaturesofinundatedwetlandthatmayoverlapwithsignaturesofwetlandvegetation.Acri