denroclim2002指导手册

Computers&Geosciences30(2004)303–311DENDROCLIM2002:AC++programforstatisticalcalibrationofclimatesignalsintree-ringchronologies$FrancoBiondia,*,KishorWaikulbaDepartmentofGeography,UniversityofNevada,MailStop154,Reno,NV89557-0048,USAbDepartmentofComputerScience,UniversityofNevada,Reno,NV89557,USAReceived11March2003;receivedinrevisedform4November2003;accepted4November2003AbstractTree-ringchronologiesareoftencalibratedagainstinstrumentalclimaterecordsusingcorrelationandresponsefunctions.DENDROCLIM2002usesbootstrappedconfidenceintervalstoestimatethesignificanceofbothcorrelationandresponsefunctioncoefficients.Inputandoutputfileselection,aswellasanalyticaloptions,arechosenfromauser-friendlyGUI.FinalresultsaresavedinASCIIformat,andareplottedonscreenusingcolor-codedsymbols.DENDROCLIM2002isanextensionofexistingtask-specificsoftware,whichismostlyMS-DOSbased,andofavailableuser-suppliedcodeforstatisticalpackages,suchasSAS.Inaddition,DENDROCLIM2002incorporatestheabilitytotestfortemporalchangesofdendroclimaticrelationshipsbymeansofevolutionaryandmovingintervals.Thissimpleapproachallowsforacomplete,dynamicalrepresentationofstatisticalrelationshipsbetweenclimateandtreegrowth.Anexampleusingpublisheddendroclimaticdataisusedtoillustratetheanalyticalandgraphicalcapabilitiesofthesoftware.r2004ElsevierLtd.Allrightsreserved.Keywords:Dendrochronology;Dendroclimatology;Correlation;PrincipalComponentRegression;Bootstrap1.IntroductionNumericalcalibrationofproxyclimaterecordswithinstrumentalobservationsisanessentialrequirementofmodernpaleoscience(Alverson,2002).Becauseoftheirannualtoseasonalresolution,tree-ringchronologiesareusuallyregressedagainstmonthlyclimatevariables,especiallyprecipitationandtemperature(e.g.,Fritts,1976).Themostcommonstatisticalmodelsusedbydendrochronologistsarecalled‘correlationfunctions’and‘responsefunctions’(Blasingetal.,1984;Frittsetal.,1971).Theterm‘function’indicatesasequenceofcoefficientscomputedbetweenthetree-ringchronologyandthemonthlyclimaticvariables,whichareorderedintimefromtheprevious-yeargrowingseasontothecurrent-yearone.In‘correlation’functionsthecoeffi-cientsareunivariateestimatesofPearson’sproductmomentcorrelation(e.g.,Morrison,1983),whilein‘response’functionsthecoefficientsaremultivariateestimatesfromaprincipalcomponentregressionmodel(BriffaandCook,1990;MorzukhandRuark,1991).Interpretationofcorrelationandresponsefunctionsisfavoredbyanaccurateassessmentofstatisticalsig-nificance,sothatappropriateecophysiologicalhypoth-eses(e.g.Biondi,1993;Biondietal.,1997)andpaleoclimaticreconstructions(e.g.Biondi,2000;Biondietal.,1999)canbegenerated.Inresponsefunctions,normalsignificancelevelsofcoefficientsaremisleadingbecauseerrorestimatesareunderestimated(Cropper,1985;MorzukhandRuark,1991),hencesomeARTICLEINPRESS$ExecutableprogramforWindows98,ME,NT,2000,XPavailablefromserverat*Correspondingauthor.Tel.:[phone];fax:[phone].E-mailaddress:[e-mail](F.Biondi).URL::10.1016/j.cageo.2003.11.004coefficientscanerroneouslypassthesignificancetest.Thisusuallycausesagreaternumberofsignificantcoefficientsinresponsefunctionsthanincorrelationfunctions(e.g.Villalbaetal.,1994).Asasolution,bootstrappederrorestimatescanbeusedtoobtainmoreaccurateresults(Efron,1979;EfronandTibshirani,1986;Guiot,1990,1991).Correlationfunctionscanalsobeincorrectlytestedforsignificance,asexplainedbyBiondi(1997),anditisthereforedesirabletocomputebootstrappedconfidenceintervalsforcorrelationfunc-tionsaswell.Softwareprogramscurrentlyavailablefordendrocli-maticcalibrationaregearedtowardsnon-state-of-the-artoperatingsystems,suchasMS-DOS.TwosuchprogramsareRESPO(HolmesandLough,1999)andPRECON(Fritts,1999).Theformerdoesnotusebootstrappedconfidenceintervals,whereasthelatterusesthemforresponsefunctionsonly.SpecificcodefortheSASpackage(SASInstituteInc.,2000)wasrecentlyprovidedbyFekedulegnetal.(2002),butwithoutconsideringthefundamentalissuesstudiedbyCropper(1985).Furthermore,asclimate-treegrowthrelation-shipsmaychangeovertime,statisticalcalibrationbetweenatree-ringchronologyandmonthlyclimateparametersneedstobetestedfortemporalchanges.Asimple,buteffectivewaytoexplorethestabilityofcalibrationmodelsistocomputethemformultipleperiods,orintervals,whichideallyshouldbesystem-aticallyselectedtominimizebias.WereporthereonthedevelopmentofDENDRO-CLIM2002,asoftwarepackagefor32-bitWindowsoperatingsystemsthatcomputesbootstrappedresponseandcorrelationfunctionsforsingleandmultipleintervals.Thelatteraredefinedusingeitheraconstantlengthprogressivelyslidbyoneyear(movingintervals)oralengththatisincrementedbyonestartingfromthemostrecentyear(backwardevolutionaryintervals)orfromtheleastrecentyear(forwardevolutionaryintervals;Fig.1;Biondi,1997,2000).Inputandoutputfileselection,aswellasprogramoptions,arechosenfromauser-friendlygraphicaluserinterface(GUI).FinalresultsaresavedinASCIIformat,andareplottedonscreenusingcolor-codedsymbo