Downscaling Spatial Rainfall Field from Global Sca

47B164AnnualsofDisas.Prev.Res.Inst.,KyotoUniv.,No.47B,2004DownscalingSpatialRainfallFieldfromGlobalScaletoLocalScaleUsingImprovedMultiplicativeRandomCascadeMethodRoshanK.SHRESTHA*,YasutoTACHIKAWA,KaoruTAKARA*GraduateSchoolofUrbanandEnvironmentalEngineering,KyotoUniversitySynopsisNon-homogenousmultiplicativerandomcascademethoddownscalesspatialrainfallfieldfromacoarsescaleintoafinerone.Currently,thiskindofdownscalingislessreliableeventhoughitcorrectlyproducesalongtermaveragespatialpattern.Itfailsreproducingthepatternsinrepeatedtrials;andthereisahigherchanceofmagnitudefluctuation.Thesedrawbacksareneededtoovercome.Inthisstudy,anewmethod,namedasrandomcascadeHierarchicalandStatisticalAdjustment(HSA)method,isintroducedandtestedtodownscale1.25degreeGAMERe-analysisdatainto10-minutespatialresolution.Theobtainedresultsarehighlyimproved,quiterobustandreliablethanthepreviousmethod.Keywords:randomcascademethod,downscaling,GAMERe-analysisdata,HSAmethod1.INTRODUCTIONAccuratesimulationofspacetimerainfallfieldisanimportanttaskinhydrology.Itisanimportantbindingforcingtounderstandthespacetimevariabilityofhydrologicfactors,andtodrivesmalltolarge-scale,shorttolong-termsimulationsofrunoffquantityandquality.Therearenumerousattemptstouseproductsofglobalscalespacetimerainfallmodels,e.g.GeneralCirculationModels(GCM),inlocalscalehydrologicalanalysisforanumberofreasons.ThisdemandsareliabledisaggregationofacoarseGCMscalerainfallfieldtoasmallerscaleoflocalcatchments(BurlandoandRosso,2002).Thispaperpresentsanimprovedmethodtodisaggregatethespatialrainfallfieldusinganon-homogenousmultiplicativerandomcascademethod.Thereisalargescaledifferencebetweenglobalscale(climateoratmospheric)modelsandregionalorlocalhydrologicalmodels.Stillthesemodelsarenecessarytobecoupledinordertounderstandandpredictaclearscenariooflocalandregionalimpactsonhydrologicalcycleduetoglobalchanges.CoarsescaleproductsofGCMsareaninadequatebasisforassessinglocal/regionalscaleimpactsasitishardlyabletoresolvemanyimportantsub-gridscaleprocesses(Hostetler,1994;Wilbyetal.,1999).Itisnecessarytoidentifythesub-gridscalefeaturesforlocalorregionalhydrologicalanalysis,whichisnotseeninacoarserscaleframe.Currentemphasisonthephysicalbasisofrainfallrepresentations(Eagleson,1984;GuptaandWaymire,1979;SmithandKarr,1984)aredevelopedafterunderstandingaclearpictureofrainfallfieldstructurethatarainedareaofagivenscalehasoneorseveralsmaller-scaleareasofmoreintenserainzonesembeddedwithinit(Waymireetal.,1984).Investigationonthestatisticalfluctuationsinspaceandtimerainfallintensityandtheirmathematicalrepresentationshasyieldedtwomajorstochasticspace-timerainfall-modelingapproaches.Thefirstapproachthatfocusesonclusterpointprocesstoreproducethehierarchicalspatialandtemporalorganizationexhibitedbyobservationsofspace-timerainfall(AustinandHouze,1972;GuptaandWaymire,1979;Waymireetal.,1984;Kavvasetal.,1987)hasbeencriticizedforitsdifficultyandunambiguityinparameterestimation(SivapalanandWood,1987)andinabilitytofullydescribetherainfallstructureoveralargerangeofscales(Foufoula-GeorgiouandKrajewski,1995).Thesecondapproachisbasedonthescalinginvariancefeaturesofobservedspatialrainfallfields(SchertzerandLovejoy,1987;LovejoyandSchertzer,1990;GuptaandWaymire,1990)withextremevariabilityandstrongintermittence(GeorgakakosandKrajewski,1996),whichhasyieldedamultiplicativerandomcascadetheory(LovejoyandSchertzer,1990;GuptaandWaymire,1993).Duetothescalinginvarianceorself-similarityconceptinthisapproachofspace-timerainfallmodeling,theparameterizationisparsimoniousandvalidoverawiderangeofscales(LovejoyandSchertzer,1990;GuptaandWaymire,1993;OverandGupta,1994;Foufoula-GeorgiouandKrajewski,1995;Olsson,1996).Inalargeorglobalscale,thespace-timerainfallfieldsobtainedfromthere-analysisofGCMoutputsarenowabundantlyavailablewithconsiderablyacceptableaccuracy(Prudhommeetal.,2002).Theadvancementofglobalscaleclimate/atmosphericmodelshasalreadyachievedmuchhigherresolutionintemporalscaleasshortas15minutesunlikethespatialscale.Currently,GCMoutputsat6hour,12hourordailyintervalsareoftenbeingtestedinhydrologicalsimulationsoflargescalecatchments.Thisrangeofdatafrequencyapproximatelyfulfillsthegeneralneedoftemporaldatainhydrologicalmodeling.Ontheotherhand,thereisawidegapinthespatialscalebetweentheavailabilityofGCMorsimilarlarge-scalemodeloutputsandtheneedofhydrologicalmodels(BurlandoandRosso,2002).Thisisoneofthemajorobstaclestoapplytheglobalscaleobservationintheassessmentsoflocalscalehydrologicalbehavior.TheneedofreliableandaccuratespatialdisaggregationisprettyhightoanalyzerealworldproblemsbyusingcurrentGCMscaleoutputsasthespatialrainfallstructureinducessignificanteffectinhydrologicalanalysisofsmalltolarge-scalecatchments(Shresthaetal.,2002).Spatialrainfallfield,whichplaysasignificantroleinanysubsequentanalysesinvolvingtherainfallfieldasprimaryorsecondaryinformation,containsahigherdegreeofspatialvariabilitythathastobemodeledatthelocalscale.Amultiplicativecascadetreatmentbasedonthestatisticaltheoryofturbulence(Mandelbrot,1974)offersaconcretewayofmode