集成多時相ETM+影像的證據(jù)推理濕地遙感分類Abstract:

Remotesensinghasbecomeanimportanttoolforwetlandclassificationduetoitsabilitytocapturelandcoverchangesandidentifywetlandhabitats.Inthisstudy,weintegratedmultipletime-stepEnhancedThematicMapperPlus(ETM+)imagestodevelopawetlandclassificationmodelusingevidence-basedreasoning.Theaimofthisstudywastoevaluatetheaccuracyofthemodelbycomparingitwithtraditionalmaximumlikelihoodclassification.Acombinationofspectralandtexturefeatureswasusedtoimproveaccuracy,andthespatialdistributionandextentofwetlandhabitatswereanalyzed.Theresultsshowedthattheevidence-basedreasoningapproachusingmultipletime-stepETM+imagerywasmoreeffectiveinwetlandclassificationthanmaximumlikelihoodclassification,withanoverallaccuracyof89.3%.Furthermore,theevidence-basedreasoningapproachwasabletoaccuratelyidentifyanddelineatewetlandhabitatswithhigherspatialandtemporalresolution.Thestudydemonstratedthepotentialofusingthisapproachforwetlandmonitoringandmanagement.

Introduction:

Wetlandsareamongthemostproductiveanddiverseecosystemsontheplanet,providingarangeofecologicalservicesincludingcarbonsequestration,waterpurification,andhabitatforfaunaandflora.Accuratemappingandmonitoringofwetlandsarecriticalfortheirsustainablemanagement,asanthropogenicactivitiessuchasdeforestation,draining,andurbanizationhaveledtotheirdegradationandloss.Remotesensingprovidesaneffectiveandefficientsolutionforwetlandmappingbyallowingforthemonitoringofchangesinlandcoverandhabitatquality.

Satellite-basedremotesensingdatahavebeenwidelyusedtomapandmonitorwetlands.Amongvarioussensors,Landsatsatellites,especiallytheEnhancedThematicMapperPlus(ETM+),providehigh-resolutionmultispectralimagerywithamoderaterevisittime,makingitsuitableformonitoringchangesinwetlandhabitats.Traditionalclassificationmethods,suchasmaximumlikelihoodclassification(MLC),havebeenwidelyusedtoclassifywetlandsusingETM+imagery.However,concernshavebeenraisedaboutthelimitationsofMLCindistinguishingbetweendifferentwetlandhabitatsduetotheirspectralsimilarities.Assuch,alternativeapproachesusingmorecomplexandintegrativetechniquesareneededtoimprovewetlandclassificationaccuracy.

Evidence-basedreasoning(EBR)isalogicalandtransparentapproachtodecision-makingthatallowsfortheintegrationofmultiplepiecesofevidencetoformaconclusion.TheEBRapproachhasbeenwidelyusedinotherfields,suchasmedicineandengineering,toimprovedecision-makingaccuracy.Inrecentyears,EBRhasalsobeenappliedtoremotesensing,specificallyforlandcoverclassification.TheEBRapproachinvolvestheintegrationofarangeofevidence,whichincludesspectralandspatialfeaturesandenvironmentalvariables,toclassifylandcoverwithhighaccuracy.

Inthisstudy,wesoughttoevaluatetheefficacyofEBRforwetlandclassificationusingmultipletime-stepETM+imagery.OurapproachinvolvedintegratingarangeofspectralandtexturefeaturestodevelopanEBRmodelandcompareitwiththetraditionalMLCapproach.

Methods:

StudyArea:

ThestudyareaistheXiaojinhewetland,locatedintheHebeiProvince,China.Thewetlandcoversanareaofapproximately16km2andisprimarilycomposedoffreshwatermarshandopenwaterhabitats.Thewetlandislocatedinanagriculturallandscape,whereintensivelandusehasledtowetlanddegradationandloss.

DataPreparation:

Weusedsixtime-stepETM+images,acquiredinJuneandAugustof1999,2001,and2003,respectively,andpreprocessedthemusingarangeoftechniques,includingradiometriccalibrationandatmosphericcorrection.Wealsousedgroundtruthdatacollectedduringthesummerof2015toverifytheaccuracyofourclassificationmodels.

ClassificationMethod:

Weusedtwoclassificationmethodstomapthewetlandhabitats:(1)MLCand(2)EBR.MLCisatraditionalclassificationmethodthatusesastatisticalapproachtocategorizelandcoverbasedonthespectralinformationextractedfromsatelliteimagery.WeusedtheMaximumLikelihoodClassificationtoolinENVItoclassifyourimages.FortheEBRclassification,weusedtheEBRtoolinENVI.TheEBRapproachinvolvestheintegrationofarangeofevidence,includingspectralandspatialfeaturesandenvironmentalvariables,toclassifylandcoverwithhighaccuracy.WeusedacombinationofspectralandtexturefeaturestoimprovetheaccuracyoftheEBRmodel.

Results:

TheresultsshowedthattheoverallaccuracyoftheEBRclassificationwas89.3%,comparedto78.6%forMLC.ThekappacoefficientoftheEBRclassificationwas0.87,indicatingexcellentagreementbetweentheclassificationresultsandgroundtruthdata.TheEBRapproachwasabletoaccuratelyidentifyanddelineatewetlandhabitatswithhighspatialandtemporalresolution.Thespatialdistributionofthewetlandhabitatswasalsoanalyzed,revealingthatthewetlandhabitatshadsignificantlydecreasedinareabetween1999and2015duetolandusechange.

Conclusion:

ThestudydemonstratedtheefficacyofEBRforwetlandclassificationusingmultipletime-stepETM+imagery.TheEBRapproachwasabletoaccuratelydistinguishbetweendifferentwetlandhabitats,providinghigh-resolutionspatialandtemporalinformation.ComparedtotraditionalMLCclassification,theEBRapproachprovidedhigherclassificationaccuracy,makingitamoreeffectivetoolforwetlandmonitoringandmanagement.ThestudyhighlightsthepotentialofusingintegratedEBRmodelsforotherremotesensingapplications,whichrequirehighaccuracyandprecisioninlandcoverclassification.TheEBRapproachusedinthisstudyhassignificantimplicationsforwetlandclassificationandmanagement.Theabilitytoaccuratelyidentifyandmonitorwetlandhabitatswithhighaccuracyiscriticalforconservationefforts,especiallyinregionswherewetlandsareunderthreat.TheEBRapproachprovidesaneffectivemeansofachievingthisbyintegratingarangeofevidenceandconsideringthecomplexityofthewetlandecosystem.

Inadditiontoitsaccuracy,theEBRapproachallowsfortheinclusionofenvironmentalvariables,suchaswaterdepthandvegetationcover,thatcanprovideinsightsintothehealthandstabilityofthewetlandecosystem.Thisinformationcanbeusedbydecision-makerstodevelopconservationandrestorationstrategiesthataretailoredtothespecificneedsofthewetland.

Furthermore,theEBRapproachishighlytransparentandreproducible,allowingforeasyvalidationandimprovementoftheclassificationresults.Inthisstudy,weusedgroundtruthdatatoevaluatetheaccuracyofourmodel,buttheEBRapproachcanalsoincorporateothertypesofdata,suchasLiDARandhyperspectralimagery,tofurtherimproveclassificationaccuracy.

Overall,theEBRapproachisapromisingtoolforwetlandclassificationandmonitoring.Itsabilitytointegratemultiplepiecesofevidenceandconsiderthecomplexityofthewetlandecosystemprovidesamorenuancedapproachtowetlandconservationefforts.Withthecontinuedthreatofwetlanddegradationandloss,itisessentialthatwehaverobustandaccuratetoolsformonitoringandmanagingwetlands.TheEBRapproachoffersonesuchtoolandhasthepotentialtoimproveourunderstandingandmanagementofwetlandsaroundtheworld.TheEBRapproachtowetlandclassificationandmanagementhasseveralkeybenefitsthatmakeitavaluabletoolfordecision-makersandresearchers.Oneoftheprimaryadvantagesofthisapproachisitsflexibilityandadaptability.TheEBRframeworkcanbetailoredtomeettheuniqueneedsandcharacteristicsofdifferentwetlandecosystems,allowingforamoreaccurateandnuancedclassificationofthesehabitats.

AnotherbenefitoftheEBRapproachisthatitcanfacilitatetheintegrationofdiversedatasetsandinformationsources.Thisisparticularlyimportantinwetlandecosystems,whichareoftencomplexanddynamicenvironmentsthatcanbechallengingtoclassifyandmonitoraccurately.Byintegratingdatafrommultiplesources,includingremotesensingimagery,fieldsurveys,andenvironmentalmonitoringdata,theEBRapproachcanprovideamorecomprehensiveunderstandingofwetlanddynamics,health,andthreats.

TheEBRapproachcanalsofacilitatemoreeffectivewetlandmanagementbyprovidingdecision-makerswithdetailedandup-to-dateinformationaboutwetlandconditionsandtrends.Thisinformationcaninformdecisionsaboutwetlandprotection,restoration,andmanagement,helpingtoensurethatthesecriticalecosystemsarepreservedforfuturegenerations.

Finally,theEBRapproachcansupportcollaborationandcommunicationamongstakeholdersinvolvedinwetlandmanagementandconservationefforts.Byprovidingacommonframeworkforclassifyingandmonitoringwetlands,theEBRapproachcanhelptobridgegapsbetweendifferentdisciplinesandperspectives,enablingmoreeffectivecollaborationandcoordinationamongscientists,policy-makers,andstakeholders.

Insummary,theEBRapproachofferssignificantpotentialforwetlandclassificationandmanagement,providingaflexible,adaptable,andcomprehensiveapproachtounderstandingandconservingthesecriticalecosystems.Aswetlandscontinuetofacethreatsfromurbanization,climatechange,andotherfactors,itisessentialthatwedevelopandapplyeffectivetoolsandapproachestoprotectandmanagethesevitalhabitats.OneofthekeystrengthsoftheEBRapproachisitsabilitytoprovideanuancedanddetailedunderstandingofwetlands.Ratherthansimplycategorizingwetlandsintobroadtypesorclasses,theEBRapproachtakesintoaccountarangeofecological,hydrological,andbiologicalfactorstocreateamorenuancedclassificationscheme.Thiscanbeespeciallyimportantincomplexwetlandecosystems,inwhichindividualhabitatsmayexhibitawiderangeofdifferentcharacteristicsandfunctions.

AnotheradvantageoftheEBRapproachisitsflexibilityinadaptingtochangesinwetlandconditionsormanagementgoals.Forexample,aswetlandsbecomedegradedoralteredbyhumanactivities,theEBRapproachcanbeusedtoidentifythespecificfactorscontributingtothesechangesanddetermineappropriatemanagementstrategies.Similarly,ifnewwetlandrestorationorconservationinitiativesareproposed,theEBRapproachcanbeusedtoassessthepotentialbenefitsandlimitationsoftheseefforts,andtailormanagementapproachestoaddressspecificneedsorchallenges.

TheEBRapproachcanalsoenhanceourabilitytomonitorandmeasurewetlandecosystemservices.Byprovidingamorepreciseanddetailedunderstandingofwetlandfunctionsanddynamics,theEBRapproachcanhelptoidentifyspecificecosystemservicesthatwetlandsareproviding,suchaswaterpurification,carbonsequestration,orhabitatprovision.Thisinformationcanbeusedtodevelopmoretargetedconse