DOC

Statistically Downscaled Temperature Scenarios over China

By Calvin Cook,2014-02-18 23:17
10 views 0
Statistically Downscaled Temperature Scenarios over Chinaover,China

    Statistically Downscaled Temperature

    Scenarios over China ATMOSPHERICANDOCEANICSCIENCELETTERS,2009,VOL.2,NO.4,2082l3

    StatisticallyDownscaledTemperatureScenariosoverChina FANLi.Jun,

    KeyLaboratoryofRegionalClimate

    EnvironmentResearchUprTemperateEastAsia(RCE-TEA),InstituteofAtmosphericPhys

    ics

    ChineseAcademyofScience,Beo'ing100029,China CollegeofAtmosphericSciences,LanzhouUniversity,Lanzhou730000,China

    Received5May2009;revised15June2009;accepted21June2009;published16July2009

    AbstractMonthlymeantemperaturesat562stationsin Chinaareestimatedusingastatisticaldownscalingtech

    nique.Thetechniqueusedismultiplelinearregressions (MLRs)ofprincipalcomponents(PCs).Astepwise screeningprocedureisusedforselectingtheski1fulPCs aspredictorsusedintheregressionequation.Thepredic

    totsincludetemperatureat850hPa(7),thecombination

    ofsea.1evelpressureandtemperatureat850hPafP+n andthecombinationofgeo.potentialheightandtempera. tureat850hPa(+n.Thedownscalingprocedureis

    testedwiththethreepredictorsoverthreepredictordo

    mains.Theoptimumstatisticalmodelisobtainedforeach stationandmonthbyfindingthepredictorandpredictor domaincorrespondingtothehighestcorrelation.Finally, theoptimumstatisticaldownscalingmodelsareappliedto theHadleyCentreCoupledModel,version3(HadCM3)

    ouWutsundertheSpecialReportonEmissionScenarios (SRES,A2andB2scenariostoconstructlocalfuture temperaturechangescenariosforeachstationandmonth. Theresultsshowthatf11statisticaldownscalingproduces 1esswarmingthantheHadCM3outputitself;f21the downscaledannualcyclesoftemperaturedifferfromthe HadCM3outputbutaresimilartotheobservation;(3)the downscaledtemperaturescenariosshowmorewarmingin thenorththaninthesouth;f4,thedownscaledtempera

    turescenariosvarywithemissionscenarios,andtheA2 scenarioproducesmorewarmingthantheB2.especially inthenorthofChina.

    Keywords:statisticaldownscaling,temperaturescenarios, annualcycles,China

    Citation:Fan,L.J.,2009:Statisticallydownscaledtem

    peraturescenariosoverChina,Atrnos.OceanicSci.Lett., 2,208-213.

    1IntrOducti0n

    Generalcirculationmodels(GCMs1doareasonable jobinsimulatinglarge.scaleupper-layerfeaturesbutfail toreproducesurfacevariablesonregiona1andlpea1scales whichareessentialinassessmentsofclimatechangeim

    pactsfWilbyandWigley,l997;Faneta1..2005).There areseveralmethodologiestobridgethegapbetween GCMsimulationsandwhatisneededinclimateimpact studies(yonStorch.1995).Statisticaldownscalingmeth

    odsareperhapstheonesmostwidelyused.Theiressence istoseekstatistica1relationshipsbetweenthevariables simulatedwellbyGCMs.whicharetreatedaspredictors Correspondingauthor:FANLi-Jun,fanlj@teaac.cn

andthoserequiredbyimpactresearchers.treatedaspre

    dictands.

    Onlyafewstudiesonstatisticaldownscalinghave beencarriedoutinChina.ForinstanceFaneta1.(20071 hassuccessfullyappliedstepwiselinearregressionwith Drincipalcomponents(PCs)toconstructfuturelpealtem

    peraturescenariosinnorthernChina;ZhapandXu(20081 hasappliedtheStatisticalDownscalingModel(SDSM,to statisticallydownscaletemperatureinthesourceofthe YellowRiverbasin.However.noresearchhasbeencon. ductedinapplyingstatisticaldownscalingtechniqueto thewholeChina.Therefore,theobjectivesofthisstudy are1,toestablishtheoptimumstatisticaldownscaling modeloftemperatureforeachstationandmonthinChina and21tousetheoptimummodeltocreatefuturetern

    peraturescenariosforeachstationandmonth.Thedata usedareintroducedinSection2.Section3describesthe methodsemployed.Section4showstheresults.followed bySection5thatdiscussesandsummarisesthestudy. 2Data

    Thepredictandsarethemonthlymeansurfaceairtern. peraturedataat562stationsinChinafroml96lto2000 (Fig.1a,.ThisdatasetisderivedfromChinaMeteoro. 1ogicalDataSharingServiceSystem.Onlythestation recordsthatarecompleteforthewholetimeperiodare keptfortheanalysis.AccordingtoChineseadministrative divisionsandclimatologica1distributioncharacteristic. thestudyregionisdividedintosevensubregions.namely

    northwesternChina(I),northernChina(II),northeastern China(III1,southwesternChina(IV1,centralChina(V1.

    easternChina(VI),andsouthernChina(VII).Down. SCalingmodelsaredevelopedforeachsubregion.sepa.

    rately.Theannualcyclesofthemean1inearwarming trendsofthesevensubregionsareshowninFi2.1b.

    Huth(20021demonstratedthatlarge.scalefreeatmos pheretemperaturevariablesaremoreinformativepredic

    torsoflpealsurfacedailymeantemperaturethan large.scalecirculationfieldsandthatthebestresultsare achievedifonetemperatureandonecirculationfieldare used.Thus,temperatureat850hPaf,thecombination ofsealevelpressureandtemperatureat850hPa+711 andthecombinationofgeopotentialheightandtempera.

    tureat850hPaf?+arecheckedfortheseven

    sub.regionsandal112months.The1argescaleclimate

    predictorsarederivedfromtheNationalCenterforEnvi

    ronmentalPrediction/NationalCenterforAtmospheric Research(NCEP/NCAR_1reanalysisdatawitharesolution N0.4FAN:STATISTICALLYD0WNSCALEDTEMPERATURESCENAR10SOVERCHINA

    70.E80.E90.El00.Ell0.E12O.E130.El40.E

    l4

    l_2

    10

    08

    0.6

    04

    0.2

    Figure1(a)Stations(dots)andthesevensubregions(Ivii)usedinthe

    10.1.U~mqi,2.Hotan,3.Beijing,4.Harbin,5.Shiquanhe,6.Chengdu,

meanlinearwarmingtrendsforeachsubregion.

    of2.5.inlatitudeandlongitudefor12monthsfrom 1961to2000.Thethreepredictordomainsareselected foreachsub.region(Table1,.

    TheGCMoutputsaretheHadleyCentreCoupled Mode1.version3fHadCM31outputsundertheSpecial

     ReportonEmissionScenariosrSRES)A2andB2sce

    nariosfrom1950to2099.obtainedfromtheIntergov. ernmentalPanelonCliraateChange(IPCC)website http:ffipcc.ddc.cru.uea.ac.uk.Theatmosphericcompo

    nentofthemodelhas19levelswithahorizontalresolu. tionof2.5.in1atitudeby3.75.inlongitude.Bothore. dictorsandpredictandsarenormalizedusingtheirre

    spectivel961-90meansandstandarddeviationsfor furtheranalysis.

    3Methodology

    3.1Statisticaldownscalingmodels

    Multiplelinearregression(MLR,ofpredictorPCsin stepwisescreeningisusedtolinkthemonthlymeantem

    peraturetothe1argescaleclimatepredictorsbasedon theirhistoricalobservations(Faneta1..2007).MLRis appliedtoestimatethetemperaturesfromJanuaryto Decemberatthe562targetstationsindependently.PC analysisisperformedasthefirststepbeforeusingMLR. ForthecombinationsoftwopredictorssuchasfP+ and(n,twofieldcombinedPCanalysisisused

    (Brethertoneta1..1992).

    Stepwiseregressionmethodwithasignificancelevel O

    -

0.2

    123456789lO?l2

    Month

    209

    statisticaldownscaling.Stationsmentionedinthetextarelabeledby1to

    7.Nanyue,8.Shahghai,9.Fuzhou,10.Sanya.(b)Annualcyclesofthe

    of0.05iSusedtoobtaintheskillfu1PCs.Thestepwise regressionprocedurecombinedforwardselectionwith backwardelimination(vonStorchandZwiers.1999). Thefirst20PCsofeachpredictorvariableareputinto thestepwiseregressionequationtoselecttheskillful PCs.

    Thedownscalingmethodisevaluatedwithinthecross. validationframework(vonStorchandZwiers.1999). Correlationcoefficientsbetweenthedownscaledesti

    matesandtheobservationsareusedasaskillscoreto validatetheskillsofthestatisticalmodels. Statisticaldownscalingmodelskillsofthethreepre. dictorsetstT,P+T,andH+T,andthethreepredictor domainsareexaminedforeachstationandmonth. 3.2ApplicationtoGCMoutput

    Intermsofthehighestcorrelationcoefficient.theop. timumpredictorandpredictordomainareselectedto projectfuturetemperaturescenarioforeachstationand month.Firstly,theHadCM3outputislinearlyregridded to2.5.x2.5.latitudelongitude.andstandardizedbythe meanandstandarddeviationwithrespecttothel961-90 oftheHadCM3run.Thisprocedureensuresthattheva1. uesdownscaledfromtheGCMrunsarefreefromthe GCM'sbias.Secondly,thestandardizeddataarepro.

    iectedontotheobservedEOFstoobtainHadCM3corre. spondingPCs,andthentheprojectedPCsentertheop. timumdownscalingmodels.Finally,thedownscaled outputsarefirstinflatedbytheinverseoftheirstandard Table1Thethreepredictordomainsforthesevensub

    regionsusedforthestatisticaldownscaling 21OATMOSPHERCAND0CEANICSCIENCELETTERS

    deviationderivedfrom1961to1990andthenaredestan

    dardizedbytheobservedmeanandstandarddeviationin orderforthedownscalingoutputstobeadjustedtothe observedmeanandstandarddeviation(Winklereta1., 1997)

    4ResuRs

    4.1Downscalingofobserveddata

    Wehaveexaminedtheskillsofthestatisticaldown. scalingmodelsofthethreepredictorsandthethreepre- dictordomains,andthenobtainedtheoptimummodelby findingthepredictorandthepredictordomaincorre

    o9

    5

    0.8

    0

    o7

    .

    0.6

    S0.5

    04

    

    2

    g

    0

    g

    

    

    B

    U

    Month

    1.0

    0.9

    '6

    0.8

    8

    o-7

    0.6

    

    .0.

    5

    O.4

    V0L.2

    spondingtothehighestcorrelationforeachmonthand station.Theannua1cyclesoftheregionalmeancorrela

    tioncoefficientsaresimilarforallthemodelswithdif- ferentpredictoranddifferentpredictordomain(Fig.2,. Themodelsofor(+performalwaysbeRerthan

    theT-onlymodelsinmostregionsandmonths,especially insouthernChina.Thereisalittledifferenceintermsof correlationsbetweendifferentpredictordomainsforeach sub.region.Themodelskillsvarywithseason.Theper-

    formanceoftl1emodelsDeakSinspringandautumnandiS theweakestinsummerinmostregions,exceptineastern China.wherethemodelskillsinspringandautumnare poorerthanthoseintheothertwoseasons.ItcanalSObe l23456789l0l1l2

    Month

    l23456789101112

    Month

    1)

    3H2)

    H3)

    ?

    P+1)

    ×P2)

    JD+(3)

    ?肿I)

    ?T(21

    +3)

    ——

    Ma)(

    Figure2Cross.validationcorrelationsofthethreepredictorsetsandthethreepredictordomai

    nsf0rtemperatureatthesevensubregions.Max

    indicatestheregionalmeanofthehighestcorrelationateachstationandmonth.Panelsfrom(a

    )to(showtheresultsofsub-regionsI-VII,respec

    tively.

    O987654O

    lOOOOOO

    luI.08tl0焉一譬f0U

    N0.4FAN:STLTISTICAL

    YD0WNSCALEDTEMPERArURESCENARIOSOVERCHINA

    seenfromFig.2thattheoptimummodelskillforeach subregioncanbeobtained,iftheoptimumpredictorand predictordomaincanbeselectedforeachstationand month.Theregionalmeanofthehighestcorrelationat eachstationalsovarieswithseason.

    4.2DownscalingofGCMoutput

    Thestatisticallydownscaledtemperaturesaredevel

    opedbytheoptimalstatistica1downscalingmodelfor

     eachstationandmonthseparately.Thedownscaledsea

    sonalcyclesareconsistentwiththeobservedonesat10 stations,whereastherearesubstantialdifferencebetween thetemperaturesfromtheHadCM3andtheobservedones MOnth

    2l1

    atmostotherstations.Forinstance,thetemperaturesfrom theHadCM3aremuchlowerthanthoseobservedinHO tan,Beijing,Chengdu,Nanyue,Fuzhou,Shanghai,and Sanya.However.thetemperaturesinHarbinfromthe HadCM3arehigherduringwinterandlowerinthewarm halfoftheyearthantheobserved;theyarelowerduring thewaITnmonthsandhigherduringthewintermonths thantheobservationsatShiquanhe(Figuresomitted). Theregiona1meanoriginalHadCM3anddownscaled temperaturechangescenarioshavebeencomparedby subtractingthemeanvaluesof1961-90fromthoseof 2070-99inthesevensub.regions(Fig.3,.ItiSfoundthat downscalingprocedureproducedlowerwarmingrates 6.0

    5+5

    5.0

Report this document

For any questions or suggestions please email
cust-service@docsford.com