摘要
Background:Inthispaper,aregressionmodelforpredictingthespatialdistributionofforestcockchaferlarvaeintheHessianRiedregion(Germany)ispresented.Theforestcockchafer,anativebioticpest,isamajorcauseofdamageinforestsinthisregionparticularlyduringtheregenerationphase.ThemodeldevelopedinthisstudyisbasedonasystematicsampleinventoryofforestcockchaferlarvaebyexcavationacrosstheHessianRied.Theseforestcockchaferlarvaedatawerecharacterizedbyexcesszerosandoverdispersion.Methods:Usingspecificgeneralizedadditiveregressionmodels,differentdiscretedistributions,includingthePoisson,negativebinomialandzero-inflatedPoissondistributions,werecompared.Themethodologyemployedallowedthesimultaneousestimationofnon-linearmodeleffectsofcausalcovariatesand,toaccountforspatialautocorrelation,ofa2-dimensionalspatialtrendfunction.Inthevalidationofthemodels,boththeAkaikeinformationcriterion(AIC)andmoredetailedgraphicalproceduresbasedonrandomizedquantileresidualswereused.Results:ThenegativebinomialdistributionwassuperiortothePoissonandthezero-inflatedPoissondistributions,providinganearperfectfittothedata,whichwasproveninanextensivevalidationprocess.Thecausalpredictorsfoundtoaffectthedensityoflarvaesignificantlyweredistancetowatertableandpercentageofpureclaylayerinthesoiltoadepthof1m.Modelpredictionsshowedthatlarvadensityincreasedwithanincreaseindistancetothewatertableuptoalmost4m,afterwhichitremainedconstant,andwithareductioninthepercentageofpureclaylayer.Howeverthislattercorrelationwasweakandrequiresfurtherinvestigation.The2-dimensionaltrendfunctionindicatedastrongspatialeffect,andthusexplainedbyfarthehighestproportionofvariationinlarvadensity.Conclusions:Assuchthemodelcanbeusedtosupportforestpractitionersintheirdecisionmakingforregenerationandforestprotecti
出版日期
2014年04月14日（中国期刊网平台首次上网日期，不代表论文的发表时间）
