Intercomparison of Global Precipitation Products: The Third Precipitation Intercomparison Project (PIP-3) Robert F. Adler Laboratory for Atmospheres, NASA Goddard Space Flight Center, Greenbelt MD Christopher Kidd Department of Geography, University of Birmingham Grant Petty Department of Meteorology, University of Wisconsin Mark Morrissey College of Geosciences, University of Oklahoma H. Michael Goodman Global Hydrology and Climate Center, NASA Marshall Space Flight Center (Submitted to Bulletin American Meteorological Society/September 2000) Corresponding author: Robert F. Adler Laboratory for Atmospheres Mail Code 912 NASA Goddard Space Flight Center Greenbelt, MD 20771 Ph: 301-614-6290 Email: Robert.Adler @gsfc.nasa.gov Abstract A set of global, monthly rainfall products has been intercompared to understand the quality and utility of the estimates. The products include 25 observational (satellite-based), four model and two climatological products. The results of the intercomparison indicate a very large range (factor of two or three) of values when all products are considered. The range of values is reduced considerably when the set of observational products is limited to those considered quasi-standard. The model products do significantly poorer in the tropics, but are competitive with satellite-based fields in mid-latitudes over land. Over ocean, products are compared to frequency of precipitation from ship observations. The evaluation of the observational products point to merged data products (including rain gauge information) as providing the overall best results. 2 1.Introduction Overthepastsix yearsseveralintercomparisonosf satelliteprecipitationalgorithms,such asthefirstandsecondPrecipitationIntercomparisonProjects(PIP-1andPIP-2),andAIP (AlgorithmIntercomparisonProject)-1, -2 and-3 haveaidedthedevelopmenatnduseof globalsatelliteprecipitationproducts.AsummaryofresultsfromtheAlPprogramis given by Ebertetal. (1996). ThePIP-1project,whichis closestin form to thecurrentPIP-3 beingdiscussed,is discussedin Barrettetal. (1994). ThePIP-2intercomparisonwhich focused on instantaneousestimatesbased on passive microwave observationsis summarizedby Smith et al. (1998).The currentPIP-3 follows the successof these previousefforts, but putsincreasedemphasisonevaluationof quasi-standards,atellite- based,global,monthlyprecipitationfields. ThePIP-3projectwassponsoredbyNASAthroughtheWetNetProjectandwasendorsed bytheGlobalPrecipitationClimatologyProject(GPCP)oftheWCRP/GEWEXProgram. ThisarticlesummarizestheresultsofthePIP-3Workshop,heldatCollegePark,MD, and thepre-workshopandpost-workshopanalysiscarriedout with thesubmitteddatasets. Sixty scientistsattendedtheworkshoprepresentingnumerousorganizations.involvedin precipitationanalysisfrom both anobservationaal ndmodelingperspective. Additional informationontheprojectandtheworkshopanddetailedresultscanbefoundatthePIP-3 homepageaddressof.http://ghrc.msfc.nasa.gov/pip3A. compacdt isc(CD)of theresults, datasets,imi/_'ese, tc.isalsoavailable. 2. PIP-3ObjectiveandApproach Theobjectiveof PIP-3 isto determinetheutility of the currentquasi-standardglobal, monthlyprecipitationproductstotheclimatemodelinganddiagnosticcommunityandthe potentialimprovementexpectedwith thelatestsatellitealgorithms.Theusercommunity needsrecommendationosnaccuracyandusefulnessfor avarietyof applicationsincluding globalmodelvalidationandclimatemonitoringanddiagnostics.Thealgorithmcommunity needsinformationonthefuturerequirementosf theusercommunity. PIP-3wasdesigned toproduceanevaluationof thecurrentproductsandfacilitatetheexchangeof information onfuturedirections. Themonthly,globalraintotalsandrainfallfrequenciessubmittedby theparticipantswere evaluatedagainstsurfacevalidationdatasets,includingan atoll gaugedataset, ocean precipitationfrequency,andlandgaugedatasets. A full year(1992)wasanalyzedtotestannualcycleretrieval. JanuaryandJulyof 1991, 1992,and1993wereincludedtoallowevaluationof interannuavlariations. August1987 fromthe PIP-1periodwas alsoevaluatedto seekevidenceof algorithmimprovement duringthepastfive years. ProductsusingSpecialSensorMicrowave/ Imager (SSM/I), geosynchronous infrared, Microwave Sounding Unit (MSU), TIROS Operational Vertical Sounder (TOVS)data, merged analysis schemes and composite microwave algorithms were included as well as prototype Tropical Rainfall Measuring Mission (TRMM) and Earth Observing System (EOS) microwave algorithms. Precipitation fields calculated from General Circulation Models (GCMs) were also included in the comparison. The evaluation statistics were kept fairly simple and consist mainly of bias, root mean square error, and correlation versus the validation data. The satellite-based products were examinedwith regardto theiroverallreasonablenes(se.g., rainfallmaximain theright place andof reasonableintensity), freedomfrom artifacts(e.g., unnaturalcoastline precipitationfeatures)andthestatisticalcomparisontothevalidationdata. 3.DescriptionofProductsandValidationDataSets Table1summarizesthethirty-oneproducts. Thetwenty-fiveobservationaplroducts,four model-basedproductsandthetwo climatologieswereintercomparedwith eachotherand thevalidationdatasetsin termsof monthlyrainfall statisticsduring 1992,interannual variationsamongtheJanuary'sandJuly's of 1991,1992and1993andthefrequencyof precipitationovertheocean(monthlyandannualstatistics). Thetwenty-fiveobservationaplroductsweredividedintotwo groupsfor certainaspectsof theanalysis.TheQuasi-Standar(dQ-S)productswereidentifiedasthosealreadyin useby the modeling/diagnosticcommunity,availablefor long, multi-yearperiodsand readily availablefromarchives,etc. Theseproductstendedtobethemorematureproductsamong theobservationaslubmissions.Theremainingobservationaplroductswerecategorizedas Experimenta(lEXP). All butoneof theExperimentaplroductswerebasedonSSM/Idata aloneandtypicallywereproducedespeciallyfor thePIP-3activity. SevenSSM//-based productsandoneotherproductwereoceanonlyestimates.OfthenineQuasi-Standar(dQ- s) products,threewereSSM/I-based,oneeachbasedon MicrowaveSoundingUnit (MSU) data,TIROSOperationaVl erticalSounder(TOVS)data,andgeosynchronouIsR dataandthreeweremergedestimatesusing a combinationof satelliteobservationsor a combinationofsatelliteandsurfacegaugeobservations. 5 The four model productsincludedcalculatedfieldsfrom the reanalysisefforts at the EuropeanCentreforMedium-rangeWeatherForecasting(ECMWF),theNationalCenters for EnvironmentalPrediction(NCEP) andthe NASA GoddardSpaceFlight Center (GSFC)andaclimatologicalaverageof theAtmosphericModelIntercomparisonProject (AMIP)climatemodels(Laueta1.,1996). Thedescriptionsof thesubmittedproductscanbefoundonthenotedwebsite or project CD,orthroughthereferencesinTable1. ThevalidationdatasetusedinthePIP-3studywasaccumulatedfromanumberof sources. Landareas: i)theGlobalPrecipitationClimatologyCentre(GPCC)gaugeproduct(Rudolf, 1994)was taken as the base validationdata set. The raw data product (as opposedto the climatological-correctepdroduct)wasused.However,thereweresomenoteableareasthat havelittle data.OutsideWesternEuropethenumberofgaugesis sparse,evenin countries suchastheUSandAustraliawheregaugecoverageisknowntobegood. ii) theSurfaceReferenceDataCentre(SRDC)(Huffmanetal., 1997)datasetwasseenas themostaccurateofthedatasets.Each2.5degreeboxusednumerousgaugestogenerate the rainfall estimate. Unfortunately,it was alsothe leastcomprehensive,being only availableforafew2.5degreeboxes. iii) supplementarygaugedatawassoughtfor USA, AustraliaandSouthAfrica to fill in someofthevoidsintheGPCCdataset.Thegaugedatafromtheseareaswasinterpolated andmappedtothe2.5degreeresolution.Inaddition,toboostthenumberof gaugesin the tropicalregion,datafrom theAmazonregionwasincoiporatedfrom theAmazonRiver BasinPrecipitationdatasetattheOakRidgeNationalLaboratory.Thelandgaugedatasets weremergedon thebasisof the SRDCproducthavingtop priority, followed by the supplementarygaugedatawherethenumberof gaugesexceededthatof the GPCCdata, andlastly,theGPCCdataset. A subseot fthisdatabasewaschosenfor thevalidationof thealgorithmproductsin order toachievearepresentativgeeographicdataset.Theselectionwasbaseduponthenumber of gaugesavailableper2.5 degreebox, by the numberof boxeswithin eachclimatic region,andtheproximitytootherboxes:Figure1showsthedistributionof thevalidation boxeschosen. Theinterannuavlalidationdatasetwasbaseduponselectedareasoffour contiguous2.5 x 2.5degreeboxesinordertoreducethenoisefromboththevalidationandalgorithmdata. Areaswerechosenasrepresentativseamplesofthe.differentclimaticregimes.Theseareas canbeseeninFig.1asthethreegroupsoffouroutlinedboxesin theU.S.andAustralia. Ocean: Oceanicvalidationdata,especiallygaugedata,is verylimited.For thePIP-3 study,atoll raingaugedatafromtheComprehensivPeacificRainfallDataBase(CPRDB;Morrisseyet al.,1995)wereused,Datafromtheatollswerecollected,qualitycontrolledandmappedby the EnvironmentalVerificationand Analysis Center (EVAC) at the University of OklahomaT. hedataweregroupedintothreeregionsreflectingtheseasonaclharacteristics of therainfalldata,namelynorthof 5" N, 10° S- 5° N andsouthof 10° S. Figure1 7 showsthelocationsof theatollvalidationgridboxes.For theinterannuaclomparisonsthe sumofthe10° S- 5°N boxeswasused(asshownbytheoutlineinFig.1). FrequencyofPrecipitation: DatafromtheComprehensiveOcean-AtmospheDreataSet(COADS;seePetty1995)was usedasvalidationdataovertheoceans,andwaspreparedby oneof theauthors(Petty). TheCOADS data set, comprised of ship observations of present and past weather, were used to determine the occurrence, or frequency, of precipitation. Due to the sparse nature of the observations in certain parts of the globe, data from the period 1958 to 1991 were used to generate an average, and therefore should be treated as a climatological average of the frequency of precipitation. The fractional-time-precipitating was derived from ship reports falling within a latitude-longitude window centered on the grid box in question. The dimensions of the window were chosen so as to achieve an adequate statistical sample without unnecessarily smoothing real gradients in rainfall distributions. Two sets of validation data were generated, one using all the COADS data with observations reporting all precipitation, except drizzle, and another set reporting all precipitation except drizzle and snow. The latter data set was included on the basis that estimates of precipitation from the passive microwave sensor would not include drizzle and snowfall. 4. Intercomparison of Monthly Rainfall Totals The global, monthly rainfall total maps for 1992 were examined and intercompared in a number of ways and against the validation data sets over the Western Pacific Ocean atolls and over land. They were also examined for artifacts and for reasonableness over areas where no validation data sets exist, for example in the mid-latitude oceans. Four examples (of the different product types) of monthly maps for July 1992 are shown in Fig. 2. All four examples display the main features of a July precipitation map. The Inter-Tropical Convergence Zone 0TCZ) stretches across the Pacific and Atlantic Oceans and northern South America, The Asian summer monsoon is producing rainfall maxima over India, Indochina and adjoining water areas. Northern Australia is in its dry season. In the tropics, the four example maps show very similar patterns and similar magnitudes. In mid- latitudes oceanic maxima are evident, with varying intensities. For example, the three maxima (the top, right panel (model)) at approximately 40°S east of Africa, in the central Pacific and east of South America are evident in the experimental and quasi-standard examples, but with different magnitudes. The noisiness of the Experimental product is due to the limited sampling with the low-orbit satellite. A similar noisy pattern is evident in the quasi-standard example at ocean latitudes above 40". 4.1 Zonal annual totals over water and land Zonal averages of the annual (1992) total over the ocean of each of the 25 observational products indicate a wide variation among the products, both in the tropics and in middle and high latitudes. The mean value and range of values at each latitude are plotted in Fig. 3a. All the products generally capture the tropical maximum, the sub-tropical minima and the mid-latitude maxima. However, among all the observational products the peak value in the Inter-Tropical Convergence Zone (ITCZ) at 8°N varies from 1300 to 3200 mm for the annual total. This large variability among the estimates is also evident in mid-latitudes with values ranging from 900 to 1800 mm in the Northern Hemisphere maximum, with additional outliers above and below those values. In very high latitudes, for example at 60°S, the range becomes an order of magnitude, going from 100 to I000 ram. At first this largevariabilityamongtheestimatesis disconcerting.However,if thesetof productsis limitedto theQuasi-Standarddataproducts,the rangeof valuesdecreases significantly. This effectcanbeseenin Fig. 3b, which showsthe standarddeviation amongtheestimatesasafunctionoflatitudeforboththe25observationaelstimatesandthe subsetof eightQ-Sproducts. This decreasein thevariabilityaswe go from all to the Quasi-standarpdroductsmainlyreflectsthematurityof theproducts.In addition,thereis someinterdependencaemongtheQ-Sproductsbecauseof mergedproductsusingsomeof thesameinputfields.ManyoftheproductsintheExperimentaglroupwerebasedonearly versionsof retrievalalgorithmsandduetoerrors,or perhaps a lack of tuning, some of these products produced values outside the range of reasonableness. These facts point to the need for the user community to exercise caution in selecting products with which to work. Fig. 4a compares the average of all the observational products with the two climatologies. The tropical peak in the Legates/Wilmott climatology is significantly larger than the observational products or the Jaeger climatology. In fact, the zonal totals also indicate that the LegatesAVilmott climatology has higher values in the ITCZ as compared to all the Q-S products. This difference is mainly related to the large peak found in the climatolo_ in the east-central Pacific Ocean in the ITCZ during the Northern Hemisphere summer. None of the observational products support the existence of this feature, although they are looking at only one year of data. In the dry, subtropic zone in the Southern Hemisphere oceans the Legates/Wilmott (LAV) climatology also carries significantly higher values than all the observational products. The difference here is due mainly to the lesser westward extent .... -from the South American and African coasts of the subtropic minima in the climatology as compared to the satellite estimates. In mid-latitudes (poleward of 40 °) the mean of all the observational products is significantly less than the climatologies. The observational mean 10