您好,欢迎访问三七文档
1、Notes_7,GEOS585A,Spring201517DetrendingTrendinatimeseriesisaslow,gradualchangeinsomepropertyoftheseriesoverthewholeintervalunderinvestigation.Trendissometimeslooselydefinedasalongtermchangeinthemean(Figure7.1),butcanalsorefertochangeinotherstatisticalproperties.Forexample,tree-ringseriesofmeasuredringwidthfrequentlyhaveatrendinvarianceaswellasmean(Figure7.2).Traditionally,seasonalorperiodiccomponents,andirregularfluctuations,andthevariouspartswerestudiedseparately.Modernanalysistechniquesfreque。
2、ntlytreattheserieswithoutsuchroutinedecomposition,butseparateconsiderationoftrendisstilloftenrequired.Detrendingisthestatisticalormathematicaloperationofremovingtrendfromtheseries.Detrendingisoftenappliedtoremoveafeaturethoughttodistortorobscuretherelationshipsofinterest.Inclimatology,forexample,atemperaturetrendduetourbanwarmingmightobscurearelationshipbetweencloudinessandairtemperature.Detrendingisalsosometimesusedasapreprocessingsteptopreparetimeseriesforanalysisbymethodsthatassumestationarit。
3、y.Manyalternativemethodsareavailablefordetrending.Simplelineartrendinmeancanberemovedbysubtractingaleast-squares-fitstraightline.Morecomplicatedtrendsmightrequiredifferentprocedures.Forexample,thecubicsmoothingsplineiscommonlyusedindendrochronologytofitandremovering-widthtrendthatmightnotbelinear,ornotevenmonotonicallyincreasingordecreasingovertime.Instudyingandremovingtrend,itisimportanttounderstandtheeffectofdetrendingonthespectralpropertiesofthetimeseries.Thiseffectcanbesummarizedbythefrequen。
4、cyresponseofthedetrendingfunction.7.1IdentifyingtrendIdentificationoftrendinatimeseriesissubjectivebecausetrendcannotbeunequivocallydistinguishedfromlowfrequencyfluctuations.Whatlooksliketrendinashortsegmentofatimeseriessegmentoftenprovestobealow-frequencyfluctuation–perhapspartofacycle--inthelongerseries.Byextension,wecanviewtheentireobservedtimeseriesasasegmentofanunknowninfinitelylongseries,andcannotbesurethatanobservedchangeinmeanoverthatsegmentisnotpartofsomelow-frequencyfluctuationimparted。
5、byastationaryprocess.Sometimesknowledgeofthephysicalsystemhelpsinidentifyingtrend.Forexample,adecreaseofringwidthofatreewithtimeisexpectedpartlyongeometricalgrounds:theannualincrementofwoodisbeinglaiddownonanever-increasingcircumference.Ifthevolumeofwoodproducedannuallylevelsoffasthetreeages,ringwidthwouldstillbeexpectedtodecline.Figure7.2.Trendinmeanandvariance.RingwidthsfromaDouglas-firtreeinJemezMountains,NewMexico,1785-2007.Bothmeanringwidthandvarianceofringwidthdeclinewithageoftree.Figure7.。
6、1.Trendinmean.AstrongtrenddominatestheDecemberatmosphericCO2concentrationatMaunaLoa,Hawaii,1958-2007.Source:“agecurve”inringwidthcanbecomputedassumingthecross-sectionalareaofwoodaddedeachyearisconstant(Figure7.3).Suchaconceptualmodelwasusedindendrochronologyasjustificationfor“modifiednegativeexponential”detrending(Fritts1976).Ifaphysicalbasisislacking,weneedtorelyonstatisticalmethodstoquantifytrend.Statisticalmethodscanhelpdistinguishtrendfromothervariations.Asimplestatisticaltechniqueofidentify。
7、inglineartrendistoregresstheobservedtimeseriesagainsttimeandtesttheestimatedslopecoefficientoftheregressionequationforsignificance(Haan2002).Thenullhypothesisisthattheslopecoefficientiszero(nolineartrend)andthealternativehypothesisisthattheslopediffersfromzero.At-testappliedtotheestimatedslopecoefficientwillindicaterejectionoracceptanceofthenullhypothesis.Thisapproachcanbeextendedtomultiplelinearregressionfortrendsinmeanmorecomplexthansimplelineartrend(Haan2002).Nonparametrictestsarealsoavailabl。
8、eforidentifyingtrend.TheMann-Kendalltestisonesuchtestcommonlyusedinclimatologyandhydrology(Salas1993).Thefrequencydomainisparticularlyusefulhere.GrangerandHatanaka(1964p.130)givesomeinsightintospectralinterpretationoftrend.Theyconcludethatweareunabletodifferentiatebetweenatruetrendandaverylowfrequencyfluctuation,andgivethefollowingadvice:Ithasbeenfoundusefulbytheauthortoconsideras“trend”inasampleofsizenallfrequencieslessthan1(2)nasthesewillallbemonotonicincreasingifthephaseiszero,butitmustbeemph。
9、asizedthatthisisanarbitraryrule.Itmayalsobenotedthatitisimpossibletotestwhetheraseriesisstationaryornot,givenonlyafinitesampleasanyapparenttrendinmeancouldarisefromanextremelylowfrequencyfluctuation.Ifweapplytheabovereasoningtoa500-yeartree-ringseries,wewouldsaythatvariationswithperiodlongerthantwicethesamplesize,or1000years,shouldberegardedastrend.Inanotherpaper,Granger(1966)defines‘trendinmean’ascomprisingallfrequencycomponentswhosewavelengthexceedsthelengthoftheobservedtimeseries.Cooketal.(19。
10、90)refertoGranger’s(1966)“trendinmean”conceptingivingsuggestionsfordetrendingtree-ringdata:Giventheabovedefinitionoftrendinmean,anotherobjectivecriterionforselectingtheoptimalfrequencyresponseofadigitalfilterisasfollows.Selecta50%frequency-responsecutoffinyearsforthefilterthatequalssomelargepercentageoftheserieslength,n.Thisisthe%ncriteriondescribedinCook(1985).TheresultsofCook(1985)suggestthatthepercentagei。
本文标题:时间序列去趋势
链接地址:https://www.777doc.com/doc-4922363 .html