搜索
PredictionandcomparisonofurbangrowthbylandsuitabilityindexmappingusingGISandRSinSouthKoreaSoyoungParka,,SeongwooJeonb,,,ShinyupKimc,andChuluongChoia,aDepartmentofGeoinformaticEngineering,PukyungNationalUniversity,599-1Daeyeon3-Dong,Nam-Gu,Busan608-737,SouthKoreabKoreaAdaptationCenterforClimateChange,KoreaEnvironmentInstitute,290Jinheung-Ro,Eunpyong-Gu,Seoul122-706,SouthKoreacDepartmentofEnvironmentalDataandInformationOffice,MinistryofEnvironmentRepublicofKorea,88Gwanmoon-Ro,Gwacheon-Si,Gyeonggi-Do427-729,SouthKoreaReceived6April2010;revised17August2010;accepted19September2010.Availableonline10November2010.AbstractThisstudycompareslandsuitabilityindex(LSI)mapscreatedusingageographicinformationsystem(GIS)withfrequencyratio(FR),analyticalhierarchyprocess(AHP),logisticregression(LR),andartificialneuralnetwork(ANN)approachestoforecastingurbanland-usechanges.Varioussocial,political,topographic,andgeographicfactorswereusedaspredictorsofland-usechange,includingelevation,slope,aspect,distancefromroadsandurbanareas,roadratio,landuse,environmentalscore,andlegalrestrictions.Then,LSImapswerecreatedusingFR,AHP,LR,andANNapproaches,andsignificanceandcorrelationwereexaminedamongthemodelsusingrelativeoperatingcharacteristic(ROC),overallaccuracy,andkappaanalyses.TheROCanalysesgaveresultsof0.940,0.937,0.922,and0.891fortheLR,FR,AHP,andANNLSImaps,respectively.ThehighestcorrelationwasfoundbetweentheLRandAHPLSImaps(0.816911),andthelowestcorrelationwasbetweentheANNandFRLSImaps(0.759701).TheANNapproachproducedthehighestoverallaccuracyat92.3%,followedby91.74%forFR,89.12%forAHP,and88.93%forLR.Inthekappaanalysis,thehigheststatisticwas45.38%forFR,followedby40.84%forANN,30representingthecityarea,theANNmethodhadarelativelyhighvalueof71.71%,andtheFR,LR,andAHPmethodshadsimilaraccuraciesof57.68,55.05,and54.31%,respectively.TheseresultsindicatethattheFR,AHP,LR,andANNapproachesproducedsimilarLSImapsforKorea.GraphicalabstractResearchhighlightsAcomparativeanalysisonthemethodologicalapproachestoforecastingurbanland-usechanges.TheusingoftheFR,AHP,LR,andANNmethodsforanalysisoflandsuitabilityindex.Eachlandsuitabilityindexmapshowingthedifferentspatialdistribution.Inthecaseofaccuracyanalysis,eachlandsuitabilityindexmapshowingasimilarresults.Rangesof0.891–0.939forROC,88.33–92.93%foroverallaccuracy,and24.78–45.38%for.Keywords:Landsuitabilityindexmap;Geographicinformationsystem;Frequencyratio;Analyticalhierarchyprocess;Logisticregression;ArtificialneuralnetworkArticleOutline1.Introduction2.Studyareaandmaterials2.1.Studyarea2.2.Materials3.Methods3.1.Mappingurbangrowthsuitability3.1.1.Frequencyratio(FR)model3.1.2.Analyticalhierarchyprocess(AHP)model3.1.3.Logisticregression(LR)model3.1.4.Artificialneuralnetwork(ANN)model3.2.Validationinurbangrowthsuitabilitymaps3.2.1.Analysisoftherelativeoperatingcharacteristic(ROC)3.2.2.AnalysisofaccuracyusingtheLULCmap4.Results4.1.Calculationofthelandsuitability4.1.1.Frequencyratio(FR)4.1.2.Analyticalhierarchyprocess(AHP)4.1.3.Logisticregression(LR)4.1.4.Artificialneuralnetwork(ANN)4.2.Comparisonofthelandsuitabilityindex(LSI)maps4.3.Validation4.3.1.Relativeoperatingcharacteristic(ROC)4.3.2.Accuracyanalysis5.Discussion6.ConclusionsAcknowledgementsReferences1.IntroductionCitieshavedevelopedrapidlysincetheIndustrialRevolution,expandingworldwideinconjunctionwithsocioeconomicdevelopment.However,therapidgrowthofurbanareashasledtocomplexproblems,includingtrafficcongestion,environmentalpollution,reducedopenspace,thedeteriorationofold,downtowncenters,andunplannedorpoorlyplannedlanddevelopment(Lee,2008).Toaddresstheseurbanproblemsandtoidentifyapproachesforsustainabledevelopment,manyresearchershavefocusedondevelopingurbangrowth-predictionmodels.Establishedmodelsincludethecellularautomata(CA)-basedUrbanGrowthModel(UGM),LandTransformationModel(LTM),ConversionofLandUseEffects(CLUE)model,andSlope,Landuse,Exclusion,Urbanextent,roadTransportation,andHillshade(SLEUTH)model,whichcanincorporatedatafromremotesensing(RS)andgeographicinformationsystems(GIS)([Clarkeetal.,1997],[Tobler,1970],[VeldkampandFresco,1996],[Pijanowskietal.,2002],[Verburgetal.,2002]and[SilvaandClarke,2002]).Theseurbangrowthmodelsidentifythebestcoefficientforpredictinggrowthuntilthepresentusingdatafromthepasttothepresentasinputforpredictingurbangrowth(Jeongetal.,2002).Acriticalissueinsuchsimulationistheprovisionofproperparametervaluesorweightssothatrealisticresultsaregenerated(Wu,2000).Empiricaldatacanbeusedtocalibrateland-useandland-cover(LULC)changemodelstofindsuitableparametervalues(LombardoandRabino,1986).Empiricaldata,suchasslope,elevation,distancefromroads,protectionstatus,anddistancefromurbanareas,canbecombinedwithoneormoremapsofhistoricallandcovertocompletethecalibration.Additionally,eachmodelrunusesamapofsuitabilitytogenerateamapofsimulatedfuturechange,placingsimulatedchangeincellsthathavethelargestsuitabilityvalues(PontiusandSchneider,2001).Themapofsuitabilityisgeneratedusingstatisticalmethods,whichcanbequalitativeorquantitative.Qualitativeorheuristicmethodsareb