ZhangHJ,HanXY,DaiS.2013.Fireoccurrenceprobability
mappingofNortheastChinawithbinarylogisticregres-sionmodel.IEEEJournalofSelectedTopicsinAp-pliedEarthObservationsandRemoteSensing,6(1):121-127.
968地理科学进展第33卷
Spatialanalysisoffire-influencingfactorsinHenanProvince
ZHANGHaijun
(SchoolofEnvironmentalScienceandTourism,NanyangNormalUniversity,Nanyang473061,Henan,China)
Abstract:Thespatialrelationshipsbetweenfireeventsandfire-influencingfactorshaveimportantimplicationsforfiremanagersandscientificallyrevealingtheserelationshipsisthereforesignificantformanagementpurpos-es.Thespatialstationaryrelationshipsbetweenfireeventsandfire-influencingfactorsconsideredbypreviousfireriskstudiescontradictthefactthatfireeventsandtheirquantifiableinfluencingfactorsarealwayscharacter-izedbyspatialheterogeneity.Iftheintrinsicallynon-stationaryrelationshipsbetweenfireeventsandfire-influ-encingfactorsaremodeledbysomestationarymodels,misleadingandevenerroneousconclusionscanbedrawn,whichhamperfirepreventionoperations.Inthisstudy,logisticgeographicallyweightedregression(LG-WR)thatcanaccountforlocalvariationsofspatialrelationshipsbetweenfireeventsandfire-influencingfactorswasemployedtoanalyzetheinfluencesofdifferentfire-influencingfactorsonfireeventsinthehighriskseason(SeptemberandOctober)from2002to2012inHenanProvince.Theindependentvariablesofthemodelincludealtitude(Al),slope(Sl),distancetothenearestvillage(Dv),distancetothenearestpath(Dp),landsurfacetem-perature(LST),NormalizedDifferenceVegetationIndex(NDVI),andglobalvegetationmoistureindex(GVMI);andthebinarydependentvariableismonthlyfirepresence,with1representingpresenceand0representingab-sence.Atrainingsubsetderivedfromspatialrandomsamplingwascreatedandpotentialmulticollinearityamongtheindependentvariableswasexcluded,andthenaLGWRfireprobabilitymodelwasdevelopedusingtheGWR4.0software.Thereliabilityanddiscriminationcapacityofthedevelopedfireprobabilityspatialmod-elwasevaluatedusingatestingsubsetandanindependentvalidationsubsetandtheresultsshowgoodmodelperformance.Themodelwasusedforfire-influencingfactoranalysisinthenextstep.Afterdelineatingandover-layingthesignificantareasofthenon-stationaryfire-influencingfactors,sevenfirepreventionregionswereiden-tifiedinHenanProvince.Therelativeimportanceofthenon-stationaryfire-influencingfactorswasevaluatedbycomparingtheabsolutevaluesoftheirestimatedcoefficientsspatially.Theresultsindicatethat:I)Theinfluenc-esofSl,Dv,LST,NDVIandGVMIonfireeventspresentsignificantspatialvariability,whereastheinfluencesofAlandDpexhibitinsignificantspatialvariabilityinHenanProvince.II)TheinfluencesofLSTandNDVIonfireeventsaresignificantgloballyinHenanProvince,whereastheinfluencesofSl,DvandGVMIareonlysignifi-cantlocally.ThesitesmoststronglyinfluencedbyLSTaremainlyNanyang,Zhumadian,Xinyangandtheircon-tiguousareas.Thesitesmoststronglyinfluencedbyvegetationcover(NDVI)areprimarilyZhoukou,Xinyang,Luohe,Xuchang,ZhumadianandShangqiu.InXinyangandsoutheastZhumadian,fireeventsaremoststronglyinfluencedbySl,whileinLuoyangthisfactorisDv,andinZhoukouandtheadjacentareaofLuoyang,NanyangandPingdingshan,itisGVMI.III)ThisstudydemonstratestheusefulnessofLGWRforexploringlocalvaria-tionsoffire-influencingfactorsandforexaminingthevalidityofaglobalfireprobabilitymodel.Thepracticalimplicationofspatialanalysisoffire-influencingfactorsresultedfromLGWRisthatdifferentfirepreventionpoliciesandemphasesshouldbeformulatedforeachofthesevenfirepreventionregions.Becauseofsuchhet-erogeneity,firepreventionpoliciesneedtotakeintoconsiderationlocalconditions.
Keywords:fire-influencingfactors;GeographicallyWeightedRegression(GWR);logisticregression;
localmodel;HenanProvince