University of Nebraska - LincolnDigitalCommons@University of Nebraska - LincolnPublications, Agencies and Staff of the U.S.Department of Commerce U.S. Department of Commerce
2012
Estimating Precipitation from WSR-88DObservations and Rain Gauge Data: Potential forDrought MonitoringGregory J. StoryNational Weather Service
Follow this and additional works at: http://digitalcommons.unl.edu/usdeptcommercepub
This Article is brought to you for free and open access by the U.S. Department of Commerce at DigitalCommons@University of Nebraska - Lincoln. Ithas been accepted for inclusion in Publications, Agencies and Staff of the U.S. Department of Commerce by an authorized administrator ofDigitalCommons@University of Nebraska - Lincoln.
Story, Gregory J., "Estimating Precipitation from WSR-88D Observations and Rain Gauge Data: Potential for Drought Monitoring"(2012). Publications, Agencies and Staff of the U.S. Department of Commerce. 556.http://digitalcommons.unl.edu/usdeptcommercepub/556
Published in Remote Sensing of Drought: Innovative Monitoring Approaches, edited by Brian D. Wardlow, Martha C. Anderson, & James P. Verdin (CRC Press/Taylor & Francis, 2012).
This chapter is a U.S. government work and is not subject to copyright in the United States.
Author:
Gregory J. StoryWest Gulf River Forecast CenterNational Weather ServiceFort Worth, Texas
281
12 Estimating Precipitation from WSR-88D Observations and Rain Gauge DataPotential for Drought Monitoring
Gregory J. Story
CONTENTS
12.1 Introduction.................................................................................................. 28212.2 PastEffortsinDeterminingMeanArealPrecipitation................................ 283
12.2.1 RainGauge–OnlyEstimation.......................................................... 28312.2.2 RadarRainfallEstimationbeforetheWSR-88D............................. 283
12.3 CurrentEstimationofPrecipitation..............................................................28412.3.1 Radar:TheWSR-88DPrecipitationEstimationAlgorithm.............284
12.3.1.1 ProblemswithRadar-BasedPrecipitationEstimates........28512.3.1.2 BenefitsofRadar-BasedPrecipitationEstimates.............. 291
12.3.2 RainGaugeNetworks.......................................................................29212.3.2.1 Near-Real-TimeGauges.....................................................29212.3.2.2 DailyReportingGauges.....................................................292
12.4 Radar-BasedMultisensorPrecipitationEstimatorPrecipitationProcessingSystem........................................................................................ 29312.4.1 ThreeStagesofMPEPrecipitationProcessing................................294
12.4.1.1 StageIoftheMPEPPS.....................................................29412.4.1.2 StageIIoftheMPEPPS....................................................29412.4.1.3 StageIIIoftheMPEPPS.................................................. 295
12.4.2 Q2,theNext-GenerationQPE..........................................................29612.4.3 SatellitePrecipitationEstimates.......................................................29612.4.4 FinalPostanalysisQualityControlTechnique.................................297
12.5 DroughtMonitoring:HowTheseEstimatesCanBeUsedto Determine CurrentLocationsofDrought................................................297
282 Remote Sensing of Drought: Innovative Monitoring Approaches
12.1 INTRODUCTION
Since its deployment, the precipitation estimates from the network of NationalWeather Service (NWS) Weather Surveillance Radars-1988 Doppler (WSR-88D)havebecomewidelyused.TheseprecipitationestimatesareusedfortheflashfloodwarningprogramatNWSWeatherForecastOffices(WFOs)andthehydrologicpro-gramatNWSRiverForecastCenters(RFCs),and theyalsoshowpotentialasaninputdatasetfordroughtmonitoring.However,radar-basedprecipitationestimatescan contain considerable error because of radar limitations such as range degra-dation and radar beam blockage or false precipitation estimates from anomalouspropagation(AP)oftheradarbeamitself.Becauseoftheseerrors,foroperationalapplications,theRFCsadjusttheWSR-88Dprecipitationestimatesusingamultisen-sorapproach.Theprimarygoalofthisapproachistoreducebothareal-meanandlocalbiaserrorsinradar-derivedprecipitationbyusingraingaugedatasothatthefinalestimateofrainfallisbetterthananestimatefromasinglesensor.
Thischapterbrieflydiscussesthepasteffortsforestimatingmeanarealprecipita-tion(MAP).Althoughtherearecurrentlyseveralradarandraingaugeestimationtechniques,suchasProcess3,MountainMapper,andDailyQualityControl(QC),thischapterwillemphasizetheMultisensorPrecipitationEstimator(MPE)PrecipitationProcessingSystem(PPS).ThechallengesfacedbytheHydrometeorologicalAnalysisandSupport(HAS)forecastersatRFCstoqualitycontrolallsourcesofprecipita-tiondataintheMPEprogram,includingtheWSR-88Destimates,willbediscussed.TheHASforecastermustdetermine in real time ifaparticular radar iscorrectlyestimating,overestimating,orunderestimatingprecipitationandmakeadjustmentswithintheMPEprogramsotheproperamountofprecipitationisdetermined.Inthischapter,wediscussproceduresusedbytheHASforecasterstoimproveinitialbestestimatesofprecipitationusing24hraingaugedata,achievingcorrelationcoeffi-cientsgreaterthan0.85.Finally,sinceseveralorganizationsarenowusingtheoutputofMPEforderivingshort-andlong-termStandardizedPrecipitationIndices(SPIs),thischapterwilldiscusshowspatiallydistributedestimatesofprecipitationcanbeusedfordroughtmonitoring.
The U.S. Drought Monitor (USDM), which is considered the current state-of-the-artdroughtmonitoringtoolfor theUnitedStates, ispresentlynotdesignedforcounty-scalerepresentations,yetitsoutputisusedbycustomersforcriticaldecisionmakingatthisspatialscale.Thus,droughtindicatorsareneededatthecountyandsubcountyscale.TheMPEestimatescanbeusedasa“goldstandard”precipitationproducttocomparewithorvalidateotherremote-sensingdroughtproducts,aslongas theuserunderstands theweaknessesofMPE.In thehandsofaknowledgeableuser,MPEprovidesinformationthatnootherexistingdroughttoolcanprovide.Withtheseproducts,wecanlookatdetailedrainfallpatternsandseehowtheycorrelate
12.5.1 NWSSouthernRegionPrecipitationAnalysisProject.....................29712.5.2 AdvancedHydrologicPredictionService.........................................299
12.6 Conclusions...................................................................................................302Acknowledgment...................................................................................................303References..............................................................................................................304
283Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
withevapotranspiration(ET)productsacrosslargeareas,aswellasidentifylocalizedareasofrainfalldeficitsovertime.Thesedatacouldalsoprovidehigher-resolutioninputsforremote-sensingdroughtindexformulationssuchastheVegetationDroughtResponse Index (VegDRI) (Brown et al., 2008). VegDRI currently integrates SPIgridsspatiallyinterpolatedfromAppliedClimateInformationSystem(ACIS)gaugedata,whichcharacterizebroadscaleprecipitationpatternsbutareoftenunrepresenta-tiveofcounty-scalelevelprecipitationvariations.Higher-spatial-resolution4kmMPEobservationsarenowavailabletoenhancethesetypesoftoolsandsupportlocal-scaledroughtmonitoringandearlywarningactivitiesthathavebeenidentifiedasaprioritybytherecentlyestablishedNationalIntegratedDroughtInformationSystem(NIDIS).
12.2 PAST EFFORTS IN DETERMINING MEAN AREAL PRECIPITATION
ThischapterbrieflydiscussessomeofthereasonswhytheWSR-88DdoesnotalwaysestimateprecipitationaccuratelyandexplainhowHASforecastersusetheMPEPPStodeterminetheaccuracyofradarprecipitationestimates,aswellashighlightsomeknownissueswithtraditionalraingaugedata.Butbeforewelookatthecurrentstateofground-basedradarrainfallestimation,anexaminationofpastestimationtech-niqueswillbepresentedtogainanappreciationofthecurrentalgorithms.
12.2.1 Rain GauGe–Only estimatiOn
BeforeMPE,theRFCsonlyusedraingaugedatatocalculatebasin-averagedMAP,which is the average depth of precipitation over a specific area for a given timeperiod.Thisledtotimingandlocationerrorsintheidentificationofheavyrainfallevents,especiallyinahighlyconvectiveenvironmentwhereintenserainfalloftenoccursoversmallcoreareas.PrecipitationestimatesweregeneratedfromdiscreteraingaugeobservationsusingtheThiessenpolygonmethod.ThismethodattemptedtocalculateMAP,allowingforanonuniformdistributionofgaugesbyprovidingaweightingfactorforeachgauge.Inbasinswherenoraingaugesexisted,thismethodwasforcedtouseraingaugesthatwereoutsidethebasininquestionforitscalcula-tion.Althoughgauge-onlyanalysesexistfordroughtmonitoringintheUnitedStatesattheclimatedivisionscale(e.g.,the1monthaccumulatedprecipitationproductathttp://www.wrcc.dri.edu/spi/spi.html), these products are noisy, particularly in thewesternUnitedStateswheregaugedensityissparsewithonlyafewobservationsperclimatedivision.AndsinceolderradarsystemsdescribedinthenextsectiondidnothavethecomputeralgorithmsnecessarytoproduceMAP,RFCshadnochoicebuttousearaingauge–onlymethodology.
12.2.2 RadaR Rainfall estimatiOn befORe the WsR-88d
Earlyradarsystems(WSR-57,WSR-74S,andWSR-74C)cameonlinein1973andwereusedthrough1993,butmeteorologistsatthattimeusedaverycrudetechniquefordeterminingrainfallrates.Theseearlyradarnetworkswouldshowrainfalland
284 Remote Sensing of Drought: Innovative Monitoring Approaches
storm intensities using digital video integrator and processor (D/VIP) levels. TheD/VIP levels were based on a predetermined value of returned power called theequivalentreflectivity,Z.AlookuptablewasusedtoestablishrainfallratesforeachD/VIP level.Radaroperatorswouldplaceadigitalgridover theplannedpositionindicator(PPI)radarscopeandmanuallywriteinavaluerangingfrom0to6thatrepresentedthemaximumD/VIPlevelineachgridcell.Therectangulargridcellsareknownasmanuallydigitizedradar(MDR)boxes,whicharebasedonasubgridoftheLimitedFineMesh(LFM)model.ThespatialresolutionoftheMDRgridcellwasapproximately40km.Bycontrast,theHydrologicRainfallAnalysisProject(HRAP)gridnowusedbytheWSR-88Dhasfurtherimprovedthespatialresolutionto∼4km.
After theradaroperatorsdeterminedthemaximumD/VIPlevel ineachMDRbox, they would transfer these values onto a paper overlay, which was usually acountyboundarymap.Asanexample,aD/VIPlevelof5meantthereturnedpowerfromtheechohadanequivalentreflectivityZofbetween50and57decibels(dBZ).Next theoperatorswould attempt todeterminehowmuch rainhad accumulated.Usingareflectivityrainfallratetable,thehourlyrainfallrateforthisvaluewouldbefoundtobe4.5–7.1in./hinaconvectiveenvironment.TheywouldthenvisuallyinspecttheD/VIPlevelsoverthepastfewhoursandaddtheD/VIPlevelstogetherforlonger-termrainfallestimatesforspecificcounties.Usingtheseearlymethods,considerableguessworkandmanualanalysiswasinvolvedinusingradartodeter-minetheamountofrainfall.
12.3 CURRENT ESTIMATION OF PRECIPITATION
12.3.1 RadaR: the WsR-88d PReciPitatiOn estimatiOn alGORithm
Estimatesfromradarhavebecomethebaseproductforderivingmeanareal,basin-averagedprecipitationwithintheNWS.AphotographofatypicalWSR-88DstationisshowninFigure12.1.TheprecipitationalgorithmintheWSR-88Dradarproductgenerator(RPG)iscomplex,andgivenallthefactorsinvolvedinradarsamplingandperformance,suchasproperradarcalibrationandassumptionsregardingradiowavepropagation through the atmosphere, errors in radar precipitation estimates oftenoccur.Theprecipitationalgorithmcontainsdozensofadaptableparametersthatcon-trolitsperformance(Fultonetal.,1998),improvingaccuracyoverearlierradaresti-mationmethods(PereiraFoetal.,1988).Thealgorithmitselfconsistsoffivemainscientific processing components (or subalgorithms) and an external independentsupportfunctioncalledtheprecipitationdetectionfunction(NWS/ROC,1999).Thefivescientificsubalgorithmsare(1)preprocessing,(2)determinationofrainfallrate,(3)determinationofrainfallaccumulation,(4)rainfalladjustment,and(5)generationofprecipitationproducts.Thefivesubalgorithmsareexecutedinsequenceaslongastheprecipitationdetectionfunctiondeterminesthatrainisoccurringanywherewithina230kmradiusoftheradar,whichisreferredtoastheradarumbrella.
Onceprecipitationisdetected,thefirstsubalgorithmisexecuted:Thebasereflec-tivitydatagothroughthepreprocessingstage,whichincludesaqualitycontrolstepthatcorrectsforbeamblockageusingaterrain-basedhybridscan(O’Bannon,1997)andchecksforAPandbiscanmaximization(seeFultonetal.,1998formoredetails).
285Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
The reflectedpower returned to the radar (Z) is thenassigneda rainfall rate (R)usingaconversionknownasaZ/R relationship.As thevalueZ increases, theRestimate in inches per hour increases exponentially based on the Z/R equationemployed.Withinthisprecipitationratesubalgorithm,morequalitycontrolisper-formedusingatimecontinuitytest,aswellascorrectionsforhailandrangedegra-dation.Next,precipitationaccumulationsaredetermined through interpolationofscan-to-scanrainaccumulationwhilesimultaneouslyrunningclock-houraccumula-tions.Precipitationproductsarethengeneratedandupdatedwitheachvolumescan(NWS/ROC,1999).An important endproduct is thehourlyDigitalPrecipitationArray(DPA)productthatprovides1hestimatesofrainfallonthe4kmHRAPgriddiscussedearlier.TheseDPAsaretheoneoffourprimaryinputstotheMPEPPSprogram, a tool primarily used east of the Rocky Mountains, which will be dis-cussedlaterinSection12.4.
12.3.1.1 Problems with Radar-Based Precipitation EstimatesTheWSR-88Dprecipitationalgorithm isnotwithoutdeficienciesand limitations,whichalloperationalradarsexperiencewhenattemptingtoestimaterainfall.Manyfactorsthatmakeaccurateradarprecipitationestimatesdifficulthavebeenwelldoc-umented (WilsonandBrandes,1979;Hunter,1996).The following text isabriefdescriptionofsomeofthesefactorsandhowtheyaffectprecipitationestimates.
12.3.1.1.1 Radar Reflectivity CalibrationPrecipitationestimatescanexperiencesignificanterrorifthereflectivity(i.e.,valueofreturnedpower)fromarainfall target is toolargeor toosmall(ChrismanandChrisman,1999).TheWSR-88Dcalibrates reflectivitybefore everyvolume scanusinginternallygeneratedtestsignals.Thesecalibrationchecksshouldmaintainanaccuracyof1dBZ,whichtranslatestoanaccuracyof17%inrainfallrateswhenthedefaultZ/Rrelationship (Z=300R1.4) isemployed.However,hardwareproblems
FIGURE 12.1 AWSR-88Dradar.(PhotocourtesyofNOAA,Washington,DC.)
286 Remote Sensing of Drought: Innovative Monitoring Approaches
(suchasachangeinactualtransmittedpower,orpathlossofthereturnedpowerbeforereachingthereceiversignalprocessorsincethelastoff-linecalibration)cancause significant changes in absolute calibration over time. Absolute calibrationneeds tobemaintainedbecauseachange inZof±4dBZwill result indoubling(orhalving)theestimatedRwhenthedefaultZ/Rrelationshipisused.Therefore,theWSR-88DRadarOperationsCenter(ROC)hasdevelopedabsolutecalibrationproceduresthataredesignedtoensurethatreflectivitydataareaccuratetowithin±1dBZ.
12.3.1.1.2 Proper Use of Adaptable ParametersAsmentionedearlier,severaladaptableparametershaveabearingontheprecipita-tionalgorithm, includingparametersdefining theZ/R relationshipand themaxi-mumprecipitation rate (MXPRA). In theWSR-88D, thedefaultZ/R relationshipis the convective Z = 300R1.4, and the default MXPRA is established at 53dBZ,whichequates toamaximumrainfall rateof∼104mm/h (4 in./h)when thecon-vectiveZ/Risemployed.ThisvalueofMXPRAwasestablishedtoeliminatetheeffects of hail contamination on rainfall estimates, as water-coated ice in cloudsreturnslargerreflectivityvaluesthanliquidwateralonewouldproduce.However,extremerainfallratesabovethedefaultMXPRAhavebeenshowntooccurwhena deep warm cloud layer exists and warm rain processes prevail, which is mostprevalentintropicalrainfallregimeswherelargerwaterdropsizediametersexist(BaeckandSmith,1998)andhailisabsent.Tocompensateforthis,radaroperatorshavetheoptionofusingadifferentZ/RrelationshipcalledtheRosenfeldtropicalZ/R (Z = 250R1.2). When the tropical Z/R relationship is employed, significantlymorerainfallisestimatedforreflectivitieshigherthan35dBZ(VieuxandBedient,1998).Forexample,theconvectiveZ/Rrelationshipyieldsarainfallrateof28mm/h(1.10in./h)whenZ=45dBZ,whilethetropicalZ/Ryieldsdoubletherainfallrateof56mm/h(2.22in./h).ThreeadditionalZ/RrelationshipshavebeenapprovedforusebytheROC:theMarshall–Palmerrelationship(Z=200R1.6)forwarmoraridclimateswhererainfalleventsaremostlystratiforminnatureandtwocool-seasonstratiform relationships (EastZ=200R2.0 andWestZ=75R2.0).RadaroperatorsmayalsochangetheMXPRAparametersothatahigherrainfallratewillbeusedintheprecipitationaccumulationfunctiontoamaximumof152mm/h(6.00in./h).Ingeneral,changesintheZ/Rrelationshiphavebeenshowntobeextremelyimportantinradarprecipitationestimation(Fournier,1999),whilechangesinMXPRAhavefarlessimpact.
Twootherimportantadaptableparameters(RAINAandRAINZ)controlwhenrainfall accumulations start and stop (Boettcher, 2006). Rainfall underestimationcanoccuriftheseparametersaresetsuchthataccumulationsbegintoolateand/orend too early. RAINA is the minimum areal coverage of significant rain with adefault settingof80km2.RAINZ is thedBZ threshold that represents significantrain (i.e., the levelof returnedpower forwhichyoudesire tobegin radar rainfallaccumulation)withadefaultsettingof20dBZ.WhenthereflectivitiesofechoesareatoraboveRAINZandthetotalarealcoverageofreturnsmeetsorexceedsRAINA,the precipitation algorithm will accumulate rainfall. If these parameters are notadjustedfortherainfalltypenotedonanygivenday,thiswouldhaveimplications
287Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
fordroughtmonitoring.Ifaraineventisisolated(coveringlessthan80km2)orifthedBZdetectedislessthantheminimumdefinedlevel,thenrainfallwillnotbeaccu-mulated.Thiscouldintroducea“drybias”suchthat,ifitisconsistentoveraperiodoftime,itwouldindicateasignaldrierthantherainfallthatisactuallyreceived.
12.3.1.1.3 Hail Contamination, Bright Band, Snow, and Subcloud EvaporationThepresenceoffrozenorwetfrozenprecipitationcancausesignificantlyenhancedreflectivity values (Wilson and Brandes, 1979). As hail stones grow in size, theybecome coated with water and reflect high amounts of power back to the radar,whichcanbesignificantlyhigherthanthepowerreturnedfromliquidprecipitationpresentwithin thestorm.Thehail-contaminatedhigherpowervalueresults inanoverestimationoftheprecipitationreachingtheground.Similarly,whenicecrystalsfallthroughthefreezinglevel,theiroutersurfacesbegintomelt.Thesewater-coatedicecrystalsalsoproduceabnormallyhighreflectivities,whichleadto“brightband”enhancement(thelayeroftheatmospherewheresnowmeltstorain)andanoveres-timationoftheprecipitation.
Snowflakesaresampledfairlywellbyradar,butimproperZ/RrelationshipscanleadtoanunderestimationofthesnowfallbytheWSR-88D.Asnowaccumulationalgorithm(SAA)hasbeenaddedusingamorerepresentativerelationshipbetweenreflectivityandfrozenprecipitation(Z/Srelationship,identicaltotheEastorWestcoolseasonstratiformZ/Rrelationship) to improve thewaterequivalentsnowfallestimates.Vasiloff(2001)andBarkeretal.(2000)providemoredetailedreviewoftheSAA.
Subcloudevaporationbelowtheradarbeamwillalsocauseoverestimation.Thisoccurswhentherainfallsintoadrysubcloudlayerandismostlikelytooccurinlocationswherecloudsfrequentlyhaveveryhighbases.Inthissituation,therainfallestimateinthecloudmayberelativelyaccurate,buttheestimatewillbetoohighiflittleornorainfallreachestheground.Aprimeexampleofthisisvirga(ordrymicrobursts).
12.3.1.1.4 Range DegradationAtfarranges, rainfall ratesmaybereducedbecauseofsignaldegradationfrompartialbeamfillingthatoccurswhentheradarbeamwidenswithdistancefromtheantennaandprecipitationfillsonlypartofthebeam’sfieldofview.Althoughthecapabilityexists forrangecorrection, it iscurrentlynot implementedon theWSR-88Dpendingscientificdatatosupportaccurateparameterization.Twootherrangedegradationproblemsaremore significant compared topartial beamfill-ing.Certainrainfall types,suchasstratiformrains(e.g., rainfall fromcloudsofextensivehorizontaldevelopment asopposed toverticallydevelopedconvectiveclouds),showstrongverticalreflectivitygradients.Thestratiformgradientisposi-tiveuntilyougetpastthe“brightband,”andthenitdecreasessharply,leadingtoanoverestimationofprecipitationclosetotheradarandanunderestimationwithgreater range.Orographicraineventsalsohavesharpvertical reflectivitygradi-entsascancertainrainfalleventsassociatedwithdistinctmeteorologicalliftingsurfacessuchasawarmfront.Arainfalleventwithasharpverticalreflectivitygradientwillshowfairlystrongrangedegradation.Thereflectivityvaluesdecrease
288 Remote Sensing of Drought: Innovative Monitoring Approaches
sorapidlywithheightwithinacloudthattheradarwillhaveahigherdegreeofunderestimationastheradarbeamincreasesinaltitude.Insuchrainfallevents,thebeamheightbecomesthelargestsinglecontributortoradarrainfallunderestima-tions.Last,instratiformraineventsandwithrainsfromthunderstormsthathavesmallverticalheight(usually20,000ftorless),arainfallunderestimationoccursduetotheradarbeamovershootingtheprecipitationatfarranges,whichisalackofdetectionproblem.Tocompensateforthis,theNWSsetuptheNEXRADradarnetworkwithaspatialdistributionofroughly300kmapart.Figure12.2showstheWSR-88D radar coverage area for theUnitedStates.Notice thatmany sectionsof the western United States are without adequate radar coverage, which leadstounrepresentativeprecipitationestimates.Thus,radar-andrange-dependentlowprecipitationbiasescanaccumulateover time, leading toanunderestimationofprecipitation and a depiction of drier conditions. Users should understand thisissuebeforeusingtheseestimatestoevaluatedroughtconditionsandotherinfor-mationalproducts.
12.3.1.1.5 Anomalous Propagation and Clutter SuppressionThe WSR-88D displays reflectivity returns at locations assuming the beam isrefractingnormallyinastandardatmosphere.Attimes,severedeviationsfromthestandardatmosphereoccurinlayerswithlargeverticalgradientsoftemperatureand/or water vapor. When these deviations occur, super-refraction of the radarbeam can result, and inaccurate calculations of actual beam height are made.Thesechangesinrefractionusuallyoccurinthelowertroposphereandcanleadtopersistentandquasi-stationaryreturnsofhighreflectivityeitherfromductingof the radar beam (where radio waves traveling through the lower atmospherearecurvedtoavaluegreater thanthecurvatureof theearth)orfromthebeamcomingincontactwiththeground(Chrismanetal.,1995).ThisAPcanleadtoextremeprecipitationaccumulationestimatesfromfalseechoes.TheWSR-88Ddoesemployacluttermitigationdecisionalgorithm,whichallowstheradaropera-tor to filter undesirable reflectivity returns, often from permanent targets neartheradar(Maddox,2010).However,thiscapabilitydependsontheradaropera-tor’sabilitytorecognizetheAPandinvokethealgorithm.Improperorexcessiveuseofclutterfilteringmaycauserealmeteorologicalechoestobeunnecessarilyremoved,leadingtorainfallunderestimation.ThisoccursmostfrequentlywhenrealrainfalltargetsareembeddedinornearareasofAP,whichiscommonbehinda lineof strong thunderstorms.Also, precipitation estimates fromnonmeteoro-logicaltargets(suchaswindfarms)arestillobservedonprecipitationproducts,ascertaintargetsthatexhibitmotionarenotremovedusingcurrentclutterfilteringtechniques.Figure12.3showsanexampleofAPacrossthesouth-centralUnitedStatescausedbysuperrefractionofthebeamsofseveralradars.Notethewide-spreadlightrainfallindicatedoverOklahomaandcentralanddeepsouthTexasandheavyrainovertheGulfofMexico.Norainfallwasactuallyoccurringatthistime.Forhydrologicapplications,thisfalserainfalliseliminatedbyconductingfurtherdataqualitycontrolexternaltotheWSR-88DandisperformedwithintheMPEPPSatRFCs.
289Estim
ating Precip
itation
from
WSR
-88D O
bservatio
ns an
d R
ain G
auge D
ata
Seattle Tacoma
Portland
Pendleton
Medford
Eureka
Beale AFB RenoElko
Boise
Sacramento
San Francisco Bay Area
San Joaquin Valley Cedar City
Grand Junction
Salt Lake City
Riverton
Cheyenne
Denver
Pueblo
Pocatello/Idaho Falls
VandenbergAFBLos angeles
EdwardsAFB Las vegas
Santa Ana Mountains
San Diego
Yuma
Flagsta�
Phoenix
Tucson
WSR-88D
4,000 ft above ground level
6,000 ft above ground level
10,000 ft above ground level
200 0 200 400 600 Miles
El Paso
Holloman AFB
AlbuquerqueCannon AFB
Amarillo
Lubbock
Midland/Odessa
San angelo
Spokane
MissoulaGreat Falls Glasgow
Billings
Minot AFB
BismarckFargo/Grand Forks
Aberdeen
Rapid CitySioux Falls
La crosse
Minneapolis/St. Paul
Duluth
Green bay
Marquette
North CentralLower Michigan
Grand rapids/muskegonMilwaukee
Detroit
ChicagoQuad citiesDes MoinesOmaha
Central Illinois
Northernindiana
Cleveland
Loring/Caribou
PortlandBurlington
Ft. drum
BostonAlbany
New York CityBinghamton
Bu�alo
Centralpennsylvania
Philadelphia
Baltimore/Washington
DoverAFB
Norfolk/RichmondRoanokeJackson
Raleigh/DurhamMorehead City
WilmingtonGreer
LouisvilleFt.
Campbell
NashvilleNortheastern Alabama
Knoxyville/Tri-Cities
Columbia
CharlestonAtlantaBirmingham
EasternAlabama
RobinsAFB
Ft. Rucker
NW FloridaMobile
Jackson
New Orleans/Baton Rouge
Moody AFB
TallahasseeJacksonville
MelbourneTampa Bay Area
Miami
Key West
Cincinnati
Pittsburgh
CharlestonIndianopolisSt. Louis
Evansville
Paducah
Topeka
Kansas City
Spring�eld
Grand Island
North Platte
Goodland
Dodge City Wichita
Vance AFB
OklahomaCity
Tulsa
Frederick
DyessAFB
Dallas/Ft. Worth
CentralTexas
LaughlinAFB Austin/
San Antonio
Corpus Christi
Brownsville
Houston/GalvestonLake charles
Ft. Polk
Shreveport
WesternArkansas
Little rockMemphis
Columbus AFB
FIGURE 12.2 GeographiccoverageofWSR-88DradarsovertheCONUS.(CourtesyofNOAA,Washington,DC.)
290 Remote Sensing of Drought: Innovative Monitoring Approaches
12.3.1.1.6 Beam BlockageBeamblockage isamajorproblemwhereradarsaresituatednearmountainsandisunavoidableinmanywesternU.S.locations.Forradials(portionsofthecircularscanoftheradaratasetelevationangle)withablockageofnomorethan60%intheverticaland2°orlessinazimuth,correctionsaremadetothereflectivitiesandareincreasedby1–4dBZintherangebinsbeyondtheobstacle,dependingonthepercentageoftheblockage.Manysiteshavebeamblockagesofmorethan60%andgreaterthan2°inazimuth,andthiscorrectioncannotbeapplied.Instead,theWSR-88Demploysaterrain-basedhybridscan(O’Bannon,1997),soradialsthatexperi-encethishighdegreeofbeamblockageusethenexthigherelevationslice(completescanoftheradaratasetelevationangle)forthePPSforthatradial(uptoamaximumelevationangleof3.4°,whichisthefourthelevationsliceaboveground).However,ifahigherelevationsliceisemployed,rangedegradationismorelikely,leadingtounderestimation of the precipitation. As a result, precipitation underestimation iscommonfromradars locatednearmountains.Theproblemhasbeenmitigatedatsomesitesbyinstallingradarsonapeak.However,inthissituation,thelowesteleva-tionslicesaresohighabovevalleysthatnear-surfaceprecipitationisnotdetected,which leads to theunderestimationof rainfall fromcloudsof lowverticalextent.Figure12.2alsoillustratesthegapsinradarcoverageoverthewesternUnitedStatesduetothemountainousterrain.
FIGURE 12.3 (See color insert.)Widespreadfalseprecipitation,orAP,shownontheMPEradarmosaic.(PhotoscourtesyNOAA/NWS,SilverSpring,MD.)
291Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
12.3.1.1.7 AttenuationTheradarcorrectsforgaseousattenuationofthemicrowaveradarsignal,leavingawetradardomecoveringtheantennaandinterveningprecipitationastheprincipalattenuatorsofenergytoandfromthetarget.AlthoughthisattenuationforS-bandradars(10cmwavelength)isconsideredtohaveminimalimpactsonrainfallestima-tion,RyzhkovandZrnic(1995)showresults indicatingthatattenuationmayhaveagreater impacton rainfall estimates thanpreviously thought.Signal attenuationcouldbeonereasonwhyrainfall isoftenunderestimatedduringextremelyheavyraineventsduetoreducedreflectivityreturns,butitisdifficulttoquantifyexactlyhowmuchtherainfallratesarereduced.
12.3.1.1.8 PolarizationThecurrentWSR-88Disasinglehorizontallinearpolarizedradar.Dualpolarizationradarmeasurementsofaspecificdifferentialphaseattwoorthogonalpolarizations(horizontalandvertical)haveshownimprovedskillinrainfallestimationcomparedto single polarization radars using Z/R relationships (Zrnic and Ryzhkov, 1999).Additionalhydrometeormicrophysicalinformationcanbeinferredfromtheaddi-tionofverticalpolarizationmeasurementstoobtaindifferentialreflectivity,whichaidsindeterminingthesizeandtypeofliquidorfrozenwaterparticles(e.g.,precipi-tationsuchasrain,sleet,hail,orsnow),whichwouldleadtoimprovedprecipitationestimation.AretrofitfortheWSR-88Dtoimplementdualpolarizationonanationalscaleisslatedfor2011–2013.IthasbeendeterminedthataddingdualpolarizationcapabilitytotheWSR-88Dwillprovideimprovedrainfallestimationforfloodsanddroughtandadditionalbenefitsthatincludeimprovedhaildetectionfordiscriminat-ingbetween liquidand frozenhydrometeors, rain/snowdiscrimination forwinterweather,dataretrievalfromareasofpartialbeamblockagetoimproveservicesinmountainousterrain,andremovalofnonweatherartifactssuchasbirdsandgroundcluttertoimproveoveralldataqualityfortheprecipitationalgorithm.
12.3.1.2 Benefits of Radar-Based Precipitation EstimatesInspiteof the limitationsandsomeof the issuesrelated toradar-basedprecipita-tionestimates,therearevalidreasonsforusingthem.ArecentstudybyKrajewskiet al. (2010) summarized the operational capability of radar to provide quantita-tiverainfallestimateswithpotentialapplicationsnotonlyinhydrologybutalsoindroughtmonitoringbyimprovinggriddedstandardprecipitationindices.Radarhastheabilitytoshowthespatialandtemporaldistributionofrainfallmoreaccuratelythanothertraditionalsensorssuchasraingauges.Thetimingandintensityoftherainfallismoreeasilydeterminedbecauseoftheavailabilityofhourlyandsubhourlyestimates.Radaralsoprovidesamoreaccuratedeterminationofrainfall location,which is critical for providing more local-scale information to the drought com-munity about spatial variations in rainfall patterns and the identificationofmorelocalizedareasexperiencingprecipitationdeficits.Thisisfarsuperiortowaitingfor24hraingaugedatatobereportedandperformingonlyasinglecalculationofMAPoverapredefinedgeographicarea(e.g.,ariverbasin),aswasthestandardoperatingprocedureinthepast.
292 Remote Sensing of Drought: Innovative Monitoring Approaches
12.3.2 Rain GauGe netWORks
Raingaugenetworksformasupplementalsourceofprecipitationdataconcurrenttogriddedprecipitationestimates.TwobasictypesofraingaugenetworkssupportNWShydrologicoperations.Onenetworkhastheabilitytotransmitrainfalldatainnearrealtime,whiletheotherstationsreport24hdataonceaday.Thesetwotypesofnetworkswillbediscussedseparatelyasfollows.
12.3.2.1 Near-Real-Time GaugesSeveral near-real-time rain gauge networks with the ability to report precipita-tionhourlyorevenat15minintervalsexist.TheseincludetheAutomatedSurfaceObservingSystem(ASOS)raingaugesatairports,datacollectionplatformsoper-atedbytheU.S.GeologicalSurvey,andmesonetalertsystemsmaintainedbyvari-ouscities,states,andriverauthorities.Althoughthesegaugesarepartofdifferentnetworks,theyallusetippingbucketgauges(Figure12.4a)toautomatethequantifi-cationofprecipitationamounts.
Unfortunately, although these data are important, they are not without error,which can be introduced by wind, tipping bucket losses, poor siting (e.g., block-agefrombuildings,trees,andothertallvegetation),frozenprecipitation,electronicsignalmalfunctions,mechanicalproblems,andtiming/codingissuesrelatedtothetransmission of rainfall data. Linsley et al. (1982) showed that strong winds willcauseallraingauges,regardlessoftype,toundercatchtheprecipitation.Forexam-ple,approximatelya10%lossisestimatedata10mphwindspeed,withlossesoftenexceeding50%atwindspeedsover39mph.Tohelpcompensateforlosses,ASOStipping bucket gauges have a shield around them to disrupt the air flow over thetopofthegauge(seeFigure12.4b).Tippingbucketgaugesalsotendtounderreportintenserainfallwhentherainfallrateexceedsthebucket’sratetodiscardthecap-turedrain(∼1.5s).Thus,theycannotbecalibratedfor0.01ofaninchprecisionorwellcalibratedforhighrainfallrates.Maintenanceisalsoanissuebecausemanygaugesarelocatedinremotelocationsandfrequentsitevisitsbytechniciansmaynotbepossible.Ingeneral,automatedgaugesprovidegoodqualityrainfalldataifthe gauges have good exposure, are well maintained, are recording when the airtemperatureisabovefreezing,whenwindconditionsarerelativelylight(15mphorless),andtherainfallrateisnotinexcessof4in/h.
12.3.2.2 Daily Reporting GaugesGaugenetworksthatreportdaily,24hrainfalltotalsareusuallysubmittedbyhumanobserverswhotypicallyuseanontippingbuckettypeofgauge.Datareceivedfromthesenetworksareconsideredtobeofhigherqualitythanthedatareceivedfromthe hourly automated networks partially because of the standard 4 in rain gaugeoraweighinggaugeusedbytheobservers,whicharetypicallyfreefromsomeoftheerrorscommonlyencounteredwithtippingbucketgauges.ThetwobestknowndailygaugenetworksaretheNWSCooperativeObserver(COOP)networkandtheCommunityCollaborativeRain,HailandSnow(CoCoRaHS)network.Wewilldis-cusshow thesedata areused to improveprecipitation estimates producedby theRFClaterinSection12.4.5.
293Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
12.4 RADAR-BASED MULTISENSOR PRECIPITATION ESTIMATOR PRECIPITATION PROCESSING SYSTEM
ThemainpurposeoftheMPEPPSistotaketherawhourlyDPAsfromtheWSR-88Dsandperformadditionalqualitycontroltoachievethebestradar-basedprecipitationesti-matespossibleforinclusionintotheNWSRiverForecastSystem(NWSRFS)fortheprimarypurposeofriverstreamflowprediction.Theseestimatesalsoholdconsiderablepotentialforprovidingbothspatiallyandtemporallyexplicitinformationaboutprecipita-tionpatternsanddeficitsoveranextendedperiodoftime,whichwouldgreatlyenhancethedroughtcommunity’smonitoringcapabilitiesbeyondthespatiallyinterpolatedpre-cipitationgridsgeneratedfromstationobservationsthatarecurrentlyusedinoperationalmonitoringsystems.ThefollowingsectionsareabriefoverviewofthethreePPSstages.
(a)
(b)
FIGURE 12.4 Tippingbucketraingauge(a)andASOStippingbucketraingaugewithwindshield(b).(PhotoscourtesyofNOAA/NWS,SilverSpring,MD.)
294 Remote Sensing of Drought: Innovative Monitoring Approaches
12.4.1 thRee staGes Of mPe PReciPitatiOn PROcessinG
12.4.1.1 Stage I of the MPE PPSThefirstPPSstageingeststhehourly4kmDPAdatathataregeneratedbytheWSR-88D,selectingtheDPAthatistimedclosesttothetopofeachhour.TheonlyqualitycontrolappliedtotheDPAdataisfeaturesassociatedwiththeWSR-88Dprecipita-tionalgorithmitself.SomeofthesefeatureswerediscussedinSection12.3.1,butforamoredetaileddiscussion,seeStory(1996).
12.4.1.2 Stage II of the MPE PPSThesecondPPSstagecalculatesandappliesabiasadjustmentfactorbasedonacom-parisonofraingaugereadingsandradarprecipitationestimates(Seoet al.,1999).TwobiasingtechniquesarederivedinthePPS:amean-fieldbiasandalocalbias.Themean-fieldbiasrepresentstheratioofthesumofallpositive(nonzero)raingaugedataovertheradarumbrellafromthepreviousxnumberofhourstothesumofallnonzeroDPArainfallestimatesatthecorrespondinggaugelocationsoverthesametemporalsamplingwindow.Thesizeofthetemporalwindowxisspecifiedbytheadaptableparameter“mem-span”(memoryspaninhours,determinedasafunctionofhowwidespreadtherainfallis,howmanygaugesareavailableforsampling,andhowlongagosinceitlastrained).TheMPEprogramcalculatesamean-fieldbiasfor10memoryspans,rangingfromthecurrenthour(instantaneousbias)to10,000,000h(climatologicalbias).TheprogramalsohasanadaptableparameterthattellsMPEwhichbiascalculatedfromthe10memoryspanstoapplytotheDPAfile.Thedefaultforthisadaptableparameterisaminimumof10radar-rainpairs(calledN-Pairs)foramean-fieldbiastobeappliedtothe“raw”radarrainfallestimate.Ifthereare10ormoreN-Pairsformem-span1,theprogramusesthebiascalculatedfromtheradar-gaugepairsfromthecurrenthour.Ifthereareno10N-Pairsforthecurrenthour,theprogramgoesbackintimeuntilamem-spanisfoundwhere10radar-gaugepairsareachieved.Atime-weightingfactorisappliedtoolderN-Pairssothatthemostrecentdatacarrythemostweightinthesecalculations.Forexample,ifthebiascalculatedfrommem-span720isused,theprogramhadtogobackbetween168(themaximumnumberofhoursfromthepreviousmem-span)and720htofindenoughraineventsthathadatleast10N-Pairs,whichwouldincludeallnonzeroradar-gaugepairsfromthepast30days. Ingeneral, thedenser theraingaugenetwork is, theshorter themem-span, unless a drought is in progress or the radar samples an area in a dryclimate.Intimesofdrought,themem-spancontinuestoincreaseovertimeasfewN-Pairsareachieved,leadingtothepossibilitythatwhenitdoesrainagain,thebiascalculationwillbeinappropriate.ThegoalofMPEistocapturethetemporalvari-abilityofthebiasfordifferentrainfallregimestoallowforthevariabilityofradarprecipitationestimates.AdetaileddescriptionofallMPEfunctionalitycanbefoundintheMPEEditorUser’sGuide(NWS/OHD/HL,2010).
Inshort, the larger thenumberof raingauges locatedundera radarumbrella,thebetterchancetheprogramhasofobtainingnonzeroradar/raingaugepairsandcalculatingamean-fieldbias.Under radarumbrellas thathavea largenumberof
295Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
hourlyraingaugesavailable,thecalculatedMPEmean-fieldbiasadjustmentfactorisagoodindicatorofwhetheraradar isover-orunderestimatingrainfall.Abiasof1.00meansthattheMPEprogramhasacceptedtheradarestimatesascorrect.Ifthemean-fieldbiasadjustmentisgreaterthan1.00,theradarisunderestimatingcomparedtoitsassociatedgauges,andifabiasislessthan1.00,theradarisoveres-timatingrainfall.ThisfactorisusedtoeitherincreaseordecreasetheprecipitationestimatesintheMPEmean-fieldbiasadjustedanalysis.
In addition to themean-fieldbias (onebias for each radar), a local bias tech-nique is also calculated in the MPE program, assigning a bias correction factorforeachHRAPgridbox(orcell)intheMPEarea.Likethemean-fieldbias,localbiasvaluesarecomputedbycomparinggaugevaluestorawradarestimates.Theyarealsoprocessedover10memoryspans,selectingthememoryspanwhosebiasvaluehasat least10contributinggauge/radarpairsfallingwithina40kmradiuscirclearoundeachHRAPgridboxforwhichabiasfactorisbeingcomputed.Theresultinggridoflocalbiasvaluesisthenappliedtotherawradarmosaic(similarto how the mean-field bias is applied) to produce the local bias–corrected radarmosaic.Bycomputing thebias foreachHRAPgridbox, localgeographicalandmicroclimatologicaleffectsonrainfallcanbeaccountedfor(SeoandBreidenbach,2002).Becauseofthisaccounting,thechosendefaultMPEfieldatmanyRFCsisthelocal bias multisensor field(i.e.,thecombinationofthelocalbiasradarmosaicandagauge-onlyanalysis).
In addition to the biased radar mosaics, a gauge-only gridded field is derivedusinghourlyraingaugeobservations,whichmustbequalitycontrolledatthisstage(Fultonetal.,1998).ToolsexistwithinMPE(suchasagaugetable)thatallowHASforecasterstodetectraingaugereadingsthatsubjectivelyappeartobeinaccurate.Although raingaugedata areoften referred to as “ground truth,” thesedata alsohaveknowndeficiencies,asmentionedintheprevioussection.However,theWestGulfRFC(WGRFC)HASforecastershavefoundthatmostraingaugedatareceivedareofacceptablequalityandcanbeused(withsomecaution)tomakeaccuratebiasadjustmentsduringmostevents.Ifanygaugereadingappearsincorrect(e.g.,whenradarfieldsarenonzeroandagaugereadszero),itisremovedbytheHASforecaster,andalltheMPEfieldsareregenerated.Thismaycauseachangeinthebiasadjust-ment factors for one or more radars and in the gauge-only fields. The end resultofthissecondstageisanadjustedradarprecipitationestimateforeachWSR-88DdefinedintheMPEprogram.
12.4.1.3 Stage III of the MPE PPSInstagethreeofthePPS,theadjustedradarfields(thosederivedinStageII,whichwerediscussedintheprevioussection)aremergedwiththederivedgauge-onlyfieldtocalculatethefinalmultisensorfields.Themultisensorfieldofthespecificradarsiteisthenmosaickedwiththemultisensorfieldsofotherradarsitestoobtainthefinalmultiradarprecipitationmap.TwoprimarymultisensorfieldsarecreatedinMPE,one for eachbiasing techniquedescribed in theprevious section.TheHAS fore-castermakesadeterminationofwhichmultisensorfieldisestimatingcorrectlyeachhour(touseasourbestestimatefield,discussedfurtherinthenexttwosections).
296 Remote Sensing of Drought: Innovative Monitoring Approaches
ThesemultisensorfieldsarecreatedforthearealextentcoveredbyeachRFCandareuseddailybytheNationalCentersforEnvironmentalPrediction(NCEP)togen-erateanationalStageIVquantitativeprecipitationestimation(QPE)product.TheHASforecasterhasotherqualitycontroloptionswithintheMPEprogram,suchastheremovalofAP.Foramoredetaileddiscussionofprecipitationprocessing,seeStory(2000).
12.4.2 Q2, the next-GeneRatiOn QPe
TheWGRFChasbeenexperimentingwithanewprecipitationestimationtechniquecalledQ2,whichisthesecondtechniquederivedbyresearchmeteorologistsattheNationalSevereStormsLaboratory(NSSL).TheNationalMosaicandMultisensorQPE(NMQ)projectisajointinitiativebetweentheNSSLandotherentities(suchas the Federal Aviation Administration [FAA] and the University of Oklahoma).TheNationalMosaicandQ2systemisanexperimentalsystemdesignedtoimproveQPEandeventuallyveryshort-termQuantitativePrecipitationForecasts(QPF).Fordetailed information on the system, readers are referred to the NMQ web site athttp://nmq.ou.edu.TheNMQingestsdatafrom128WSR-88Dstationsevery5min,qualitycontrolstheradardata,andderivesaverticalprofileofreflectivityfromeachradar.AnalysesaredoneoneighttilesofradardatathatarestitchedtogethertoformacontinentalU.S.(CONUS)three-dimensional(3-D)grid.Hybridscanreflectivityandotherproducts(suchasacompositereflectivitymapandprecipitationflagprod-uct)arethenderivedtoproducetheexperimentalQ2products.Theproducts(suchasQPEaccumulationsforthecurrenthourorseveralhoursofupto72h)arethentranslatedover to the4kmHRAPgrid.TheQ2productsholdseveraladvantagesovertraditionalradar-basedestimates,withtwoprimaryadvantagesincludinganAPremovaltechniqueandrainfallestimatesbeyondthenominal230kmrangeoftheDPAfilesthatareusedinregionswhereradarumbrellasdonotoverlap.Becauseoftheseadvantages,WGRFCHASforecastershavetheoptionofimplementingQ2asourfinalbestestimatefield.
12.4.3 satellite PReciPitatiOn estimates
The MPE also ingests satellite-derived precipitation estimates from the NationalEnvironmental Satellite, Data, and Information Service (NESDIS). TheHydroestimatorisanautomatedtechnique,initiallydesignedforlarge,moistthun-derstorm systems, which uses Geostationary Operational Environmental Satellite(GOES) infrared (IR) imagery cloud top brightness temperatures (Scofield andKuligowski,2003).PixelswiththecoldestIRtemperaturesareassignedtheheavi-est rainfall rates at the surface. Numerous other factors, including the cloud-topgeometry, the available atmospheric moisture (precipitation efficiency), stabilityparametersfromweathermodels,radar,andlocal topography,areusedtofurtheradjusttherainrates.Althoughcautionshouldbeusedindrawingconclusionsaboutradarperformancebasedonsatellite-derivedprecipitationestimates,HASforecast-erscanconfirmradarperformanceiftheprecipitationestimatesfrombothsourcesare in closeagreement.However, correlationcoefficients comparing24h satellite
297Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
precipitationestimates(SPEs)to24hraingaugesshowthelowestcorrelationofanyoftheremote-sensingfields(biasedradarestimatesorQ2)usedbyHASforecasters.Therefore,satellite-basedestimateshavethemostbenefitoverlandareaswherenoorlimitedobservationsofprecipitation(e.g.,radar,Q2,orraingauges)areavailable(e.g.,borderareaofMexico).SPEscanbeusedwithoutbiascorrection,orcanbecorrectedforlocalbiasesusingthetechniquesdescribedearlierforradar.Andliketheotherfieldspreviouslydiscussed,theoptionexiststointegrateSPEsintoourfinalbestestimateprecipitationfieldbyperformingpolygonedits.OneexampleofwhenSPEswouldbeintegratediswhenlightningdataindicatethunderstormactivityandSPEsaretheonlyfieldestimatingrainfallinthislocation.Asaresult,thefinalbestestimatefieldisbasedonacombinationofradar-basedmultisensorfieldsfromDPAfiles,Q2,andSPEs.
12.4.4 final POstanalysis Quality cOntROl techniQue
Hundredsof24hCOOPrainfallreportsandCoCoRaHSobservationsareavailableforpostanalysisoftheMPEresults.DirectcomparisonsoftheMPEandobserverrainfall totals shortly after 12 Coordinated Universal Time (UTC) each morningallow HAS forecasters to determine areas where the MPE estimates may be toolowortoohigh.Forecasterscanraiseorlowerestimatesinspecifichoursinordertoproducea24hestimatethatismoreconsistentwith24hgaugereports.Thegoalis to achieve a “general” level of acceptable error in the estimates.ProgramsarerunthatshowthecorrelationcoefficientandpercentbiasofMPEestimates,whichvarybytimeandlocation.Thegoalistomodifytheestimatestoachievecorrelationcoefficientsofgreater than0.85.Most initialestimatesare low(meaningthe24hgaugereportsarehigher thanMPE)andhavecorrelationcoefficientsof less than0.85.WheninitialMPEestimatesareraisedorlowered,theinherenterrorofmostestimatesisimprovedtothedesiredcorrelation.Sincethesedataaretobeusedforimproveddroughtmonitoring,removalofthetraditionalunderestimationiscrucial.Ifthesebiasesarenotmitigated,afalseidentificationoftheonsetofdroughtmightoccurovertime.
12.5 DROUGHT MONITORING: HOW THESE ESTIMATES CAN BE USED TO DETERMINE CURRENT LOCATIONS OF DROUGHT
12.5.1 nWs sOutheRn ReGiOn PReciPitatiOn analysis PROject
In the early and mid-2000s, NWS Southern Region offices began to displaythe gridded MPE output maps on the Internet, and the data became avail-able fordownloada short time later. Initially, thesepagesgraphically showedthe short-term observed and climatic trends of precipitation across the south-ern region (from New Mexico eastward to Tennessee, Georgia, and Florida).In 2009, this project was expanded to include the entire CONUS and PuertoRico. The national-level products can be found on the Advanced HydrologicPrediction Service (AHPS) web site (http://water.weather.gov). Tools are also
298 Remote Sensing of Drought: Innovative Monitoring Approaches
available to compare MPE estimates to normal rainfall over different times-cales(http://water.weather.gov/precip/),whichcanprovidevaluableinsightintodetailed spatiotemporal patterns of precipitation deficits to characterize bothshort-andlong-termdroughtconditions.
“DeparturefromNormal”and“PercentageofNormal”productsaregeneratedbyusingsimplegridmathematics,wherethe“Normal”datasetisrespectivelysub-tractedfromordividedintothe“Observed”dataset.“Observed”dataarederivedfromoutput(e.g.,fromMPEorsimilarPPSs)from12NWSRFCs.“Normal”pre-cipitation isderivedfromParameter-elevationRegressionson IndependentSlopesModel(PRISM)climatedata(Gibsonetal.,2002),whichrepresenta30yearperiodof record (1971–2000). The data sets were created as a unique knowledge-basedsystem that uses point measurements of precipitation, temperature, andother cli-maticfactorstoproducecontinuous,digitalgridestimatesofmonthly,yearly,andevent-based climatic parameters. This unique analytical tool incorporates pointdata,adigitalelevationmodel,andexpertknowledgeofcomplexclimaticextremes,includingrainshadows,coastaleffects,andtemperatureinversions.Inordertofillinareas thathaveradar-coveragegaps in themountainouswesternUnitedStates,gaugereportsareplottedagainstlong-termclimaticPRISMprecipitationdata,andamountsbetweengaugelocationsarespatiallyinterpolated(moreinformationaboutthis method is available at http://www.cnrfc.noaa.gov/products/rfcprismuse.pdf).The derived precipitation products (specifically, “Departure from Normal” and“PercentageofNormal”products)canprovideusefulcontextualinformationtoiden-tifytheamountandmagnitudeofprecipitationdeficitsthatcanbeusedfordroughtmonitoring.
Figure 12.5 shows an example of a percent of normal rainfall graphic fromDecember 2010 across the southern United States. This month was exception-allydry,andthisgraphicdepictsfewareaswherepercentofnormalprecipitation
FIGURE 12.5 (See color insert.)PercentofnormalrainfallforthesouthernUnitedStatesfromtheAHPSprecipitationanalysispageforDecember2010.(ImagecourtesyofNOAA/NWS,SilverSpring,MD.)
299Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
exceeded100%(theupperTexasGulfcoastnearHoustonwasonearea).Ofnotearethelargeregionswherethepercentofnormalprecipitationwaslessthan50%ofnormal,specificallyfromnorthernLouisianaintoeast-centralTexas,andacrosstheTexas/MexicanbordertowesternTexas.Thisindicatesastrongdrysignal,col-locatedwithanextremedroughtcategorydesignatedontheUSDM(notshown)fortheselocations.
12.5.2 advanced hydROlOGic PRedictiOn seRvice
Before 2009, all radar-based product data displayed by the Southern RegionPrecipitation Analysis Project were considered to be “experimental.” To makethesedata“operational,”thedatapageswerepackagedintoanationwideprogramknownastheAHPS,anewandessentialcomponentoftheNWSClimate,Water,andWeatherServices.AHPSisaweb-basedsuiteofproductsthatdisplaydroughtmagnitudeanduncertaintyofoccurrence,basedontherangeofpotentialoutcomescomputedfromhistoricalhydrometeorologicaldataandcurrentconditionsusinganensemblestreamflowpredictionmodel.ThesenewproductsareenablingtheUSDM,NationalDroughtMitigationCenter(NDMC),governmentagencies,privateinstitu-tions, and individuals tomakemore informeddecisionsabout risk-basedpoliciesandactionstomitigatethedangersposedbydroughts.Althoughtheseproductswerenotdesignedspecificallyfordroughtmonitoring,thehigh-spatial-resolutionprecip-itation information they provide has substantial potential to support this applica-tion.Forexample,theofficeoftheTexasStateClimatologistcreatesagridded4kmresolutionandacounty-scaleresolutionSPIfromtheAHPSprecipitationanalysesdata(http://atmo.tamu.edu/osc/drought/).AmoredetaileddescriptionoftheSPIgridgenerationusingtheAHPSisprovidedbyNielsen-GammonandMcRoberts(2009).
Traditionally, coarse resolutionSPImapsderived fromspatial interpolationsof point-basedgaugedatahavebeenused for droughtmonitoring, as shown inFigure12.6a.InFigure12.6b,the4kmSPImapsgeneratedfromradar-basedpre-cipitationdatadepictconsiderablymorespatiallydetailedprecipitationvariations,whichprovideconsiderablymorelocal-scaleinformationaboutprecipitationdefi-citsthatismoreappropriateforcountytosubcountydecisionmakingrelatedtodrought. In brief, the SPI map generated from AHPS precipitation analyses iscreatedusing the followingprocess. Initially,aclusteranalysis isperformed todetermineTexasprecipitationnormalsbylocationandseason.Afrequencydistri-butionisthencalculatedforeachlocationandseason,fromwhichhigh-resolutiongriddedfrequencydistributionsareproduced(usingPRISMdataoverhigherter-rain of west Texas and roughly 1500 COOP stations in Texas and surroundingstates).Finally,accumulationsofprecipitationarecomputed,creating4kmandcounty-aggregatedSPIforvarioustimeperiodsfrom2to24months,andrelatedproductssuchasanSPIblend,anSPIblend1weekchangemap,andapercentofnormalprecipitationmap.
TheprimarymotivationforusingAHPSprecipitationdatainthisprojectwastofacilitatelocal-scaledroughtmonitoringforTexas.Climatedivision-scaledroughtmonitoringtoolsarewhollyinadequateforthestate,andevenACISgaugedataaretoocoarseandunrepresentativeinmanyareas.Forexample,theUSDAappliesthe
300 Remote Sensing of Drought: Innovative Monitoring Approaches
Exceptional droughtExtreme drought
NormalExceptionally wetStations
(a)
Abnormally wetModerately wetSeverely wetExtremely wet
Abnormally dryModerate droughtSevere drought
Exceptional wetness
Abnormal wetnessModerate wetnessSevere wetnessExtreme wetness
Exceptional droughtExtreme drought
Normal
(b)
Abnormal drynessModerate droughtSevere drought
FIGURE 12.6 (See color insert.) An 8 week SPI map interpolated from station-basedprecipitation data (a) and an8weekSPImapderived from4kmprecipitation fromMPE(b) (Image courtesy of Dr. John Nielsen-Gammon) for early September 2009 during theseveredrought insouthernTexas,asshownby theUSDMmaponSeptember7,2009(c).ThecirclehighlightsanareaofexceptionaldroughtintheUSDMthatisshowntohavenear-normalconditionsintheinterpolatedSPImap(a)butclearlyhadlocalizedareasofseveredroughtconditionsthatweredetectedintheSPImapbasedonhigher-resolution,radar-basedprecipitationobservations(b).
301Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
USDMmaponacountyscalefor itsdrought reliefdecisions,yet theUSDMandother existing drought index tools do not have the sufficient spatial resolution toenableestimationofdroughtatthisspatialscalewithinTexas.AnexampleoftheSPIblendforTexasduringthe2009droughtcanbeseeninFigure12.6b.Duringthisdrought,theMPE-basedSPIblendwasabletoaccuratelyhighlightthelocationsofmostseveredroughtinTexas.Gaugeswithinthesehardest-hitareas,asindicatedbyourMPEproducts,wereindeedexperiencinghistoricdroughtseveritybasedonananalysisoftheperiod-of-recorddata,whilestationsadjacenttotheseareaswerenot.Ninecounties(Nueces,SanPatricio,Aransas,Refugio,Jackson,Calhoun,Bee,Brazoria,andGoliad)experiencingunprecedenteddroughtseveritywereidentifiedinsouthernTexasalongtheGulfcoastusingMPEdata,eventhoughmostofthosecountiesdidnothavelong-termprecipitationrecordsbecauseofthesparsenumberofCOOPstationsthathadalonghistoryinthatregion(Nielsen-Gammon,August2010, personal communication). Without the long-term precipitation records, SPIblendsbasedonMPEdataprovidedinformationthatimprovedtheassessmentoftheseverityofthelocaldroughtsituation.
Incontrast,Figure12.6ashowsastation-basedSPImapwithdatatakenfromtheNDMC’sDroughtAtlasforthesametimeperiodasFigure12.6b.Eight-weekSPIdatafrom475weatherstations(226stationsinTexasand249stationsfromthesurroundingstates tominimizeedgeeffectsduring thespatial interpolationoftheSPIpointdata)wereusedtogeneratethemapinFigure12.6a.Thestation-basedSPI shows theoverall locationof exceptionallywet conditions (northeastTexasand theeasternTexaspanhandle) andexceptionaldrought (middleTexasGulfCoastalregion).Butitisclearthatthestation-basedSPImissedhowwide-spreadtheextremeandexceptionaldroughtconditionswereacrosssouth-centralTexas.Forexample,inseveralcountiesinsouth-centralTexaswherethestation-basedSPIshowednormalconditionstoabnormallydryconditions(ComalCounty
D0 abnormally dry(c)
D1 drought-moderateD2 drought-severeD3 drought-extremeD4 drought-exceptional
FIGURE 12.6 (continued) (See color insert.)
302 Remote Sensing of Drought: Innovative Monitoring Approaches
eastward to Gonzales County), the MPE-based blended SPI showed severe toextreme drought conditions. Since hourly gauge data are incorporated into thefinalmultisensorMPE,thedroughtfeaturesthatappearoverthisareainFigure12.6bshouldberepresentativeofrelativeprecipitationpatterns(anddeficits)atalocal subcountyscalebecauseground-basedprecipitationobservationsarecon-sideredintheadjusted,radar-basedprecipitationfields.AcrossthisareaofTexas,notablerainfalldiscrepanciesamongstationsduringthedefinedSPIintervalwerelikelyduetotheconvectivenatureoftherainfallinthisregion,withtheintersta-tion variations being relatively consistent with the drought/nondrought patternsdepictedinFigure12.6b.TheUSDMmapforSeptember8,2009(Figure12.6c),reaffirmstheseveredroughtconditionsoverthisarea,classifyingthesecountiesinthemostseveredroughtclass(D4,anexceptionaldroughtthatisdefinedasaonein50yearevent).FurthervisualanalysisoftheMPE-derivedSPImapofthearea reveals many subtle subcounty variations in dryness that are not detectedin thestation-basedSPImap.Manycounties insouthernTexashavepocketsofbothdroughtandnondroughtconditionsintheradar-generatedSPImapthatcan-notbe spatially resolvedusing traditional interpolatedmaps fromstation-basedobservations.
The use of 4km precipitation data provides a more accurate depiction of thebreadthandscopeoftheTexasdroughtconditionsin2009.Thisresultsuggeststhattheimprovedspatialresolutionofthisinformationwillbeatremendousbenefitforlocal-scaledroughtmonitoringactivitiesbycharacterizingdetailedsubcountyspa-tialvariationsinprecipitationdeficits.The4kmprecipitationandotherderivativeproductssuchastheSPIwillalsobeextremelyvaluableinareaswithsparseweatherstationnetworksandforcountieswithlargeareasthatcommonlyexperienceconsid-erablewithin-countyclimatevariations.
12.6 CONCLUSIONS
Over the past several years, advancements have been made in both radar-basedprecipitation sensing and multisensor estimation processing techniques. Furtherimprovementswillbemadeinradarprecipitationestimationwiththeimplementa-tionofdualpolarizationinthenextfewyears.NewrainfallratealgorithmssuchasQ2havealsobeenimplementedwithintheMPEPPS.Thischapterhasdiscussedthebenefitthatimproved,quality-controlled,andfiner-scaleprecipitationdatacanhaveindroughtmonitoringbydetailingdeficitsinrainfallwithgreaterspatialresolutionthatisnotavailableusinggauge-basedSPIdataalone.
EastoftheContinentalDivide,RFCsderiveestimatesofprecipitationusingamultisensorapproach.HourlyprecipitationestimatesfromWSR-88Dradarsarecomparedtogroundrainfallgaugereports,andabias(correctionfactor) iscal-culatedandapplied to the radarfield.Theradarandgaugefieldsarecombinedintoa“multisensorfield,”whichisqualitycontrolledonanhourlybasis.Inareaswithlimitedornoradarcoverage,SPEcanbeincorporatedintothismultisensorfield,andtheSPEcanalsobebiasedagainstraingaugereports.Inmountainousareas west of the Continental Divide, a different method is used to derive the
303Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
estimatesofprecipitation.Gauge reports areplottedagainst long-termclimato-logicalprecipitation(PRISMdata),andderivedamountsareinterpolatedbetweengaugelocations.
Studieshaveshown(Seo,1999;SeoandBreidenbach,2002)thatalgorithmsthatcombinesensorinputs—radar,gauge,andsatellite—yieldmoreaccurateprecipita-tionestimates than those that relyonasinglesensor (i.e., radaronly,gaugeonly,andsatelliteonly).Althoughit isnotperfect, theMPEdataset isoneof thebestsources of timely, high-resolution precipitation information available. Still, usersshouldunderstandtheinherentweaknessesofthisdatasetbeforeusingitindroughtmonitoringapplications,especiallythosethatrequireahighdegreeofaccuracy.
ManyquantitativemeasuresofdroughthavebeendevelopedintheUnitedStates,dependingonthesectorimpacted,theregionbeingconsidered,andtheparticularapplication.Althoughdifferentdefinitionsandmeasuresofdroughtexist, theyalloriginatefromadeficiencyofprecipitationresultingfromanunusualweatherpat-tern.Therefore,usinganimprovedsourceofprecipitationdatasuchasMPE4kmproductswouldleadtoabetterdeterminationoftheonset,intensity,andgeographicandtemporalevolutionofdrought.
SeveralofthePalmerindicesandtheSPIareusefulfordescribingdroughtonvaryingtemporalscales(i.e.,weeks,months,oryears).Onaclimate-divisionscale,astandardsuiteofproductsincludingtheNCDC’sSPI,theCPC’ssoilmoisture–relateddroughtseverityindex,andtheWesternRegionClimateCenter’sSPIexist.Onasta-tionscale,theU.S.GeologicalSurveyprovidesgauge-basedstreamflowdata,andtheHighPlainsRegionalClimateCenterproducesa30daySPIusingdailydatafromACISthatincorporatesCOOPobserverandautomatedweatherdata.Satellite-basedtoolssuchasVegDRI(Brownetal.,2008)thatassistinagricultural-relateddroughtmonitoringalsorelyonprecipitationdataasaprimaryinput.Collectively,thesedroughtindiceshavereliedongauge-baseddataandhavenotprovidedindi-ces representative of county- to subcounty-scale drought information because ofthe coarse spatial resolution inputs. The higher-resolution 4km precipitation dataproducedbyMPEcanbeusedtoreplacethetraditionalpointorinterpolatedpre-cipitationproductsinthedevelopmentoftheseindicestoprovideamoredetailedcharacterizationofdroughtpatterns.Thisholdsthepotentialtoadvancelocal-scaledroughtmonitoringactivitiesasprioritizedbyNIDIS,aswellas improvecurrentstate-of-the-artmonitoring toolssuchas theUSDM,whichwas initiallydesignedtoclassifybroadscale,nationaldroughtpatternsbutisincreasedbeingrelieduponforcountyandsubcountydroughtinformation.Withthegoalofimproveddroughtmonitoring,TexasA&MUniversity,NorthCarolinaStateUniversity, andPurdueUniversityreceivedaUSDAawardtoimprovethelong-termcalibrationoftheAHPSMPEanalyses,andtaketheSPIproductsbeyondTexastoincludeatleasttheeasternpartsoftheUnitedStates(i.e.,south-centralandeasternsections).TheprojectbeganinJanuary2011,withtangibleresultsexpectedafewmonthsafterthat.
ACKNOWLEDGMENT
TheviewsexpressedarethoseoftheauthorofthischapteranddonotnecessarilyrepresentthoseofNOAAortheNWS.
304 Remote Sensing of Drought: Innovative Monitoring Approaches
REFERENCES
Baeck,M.andJ.Smith.1998.RainfallestimationbytheWSR-88Dforheavyrainfallevents.Weather Forecasting13:416–436.
Barker,T.,P.Felsch,T.Mathewson,C.Sullivan,andM.Zenner.2000.TestoftheWSR-88DsnowaccumulationalgorithmatWeatherFieldOfficeMissoula.Technical Attachment,NWS Western Region00–13:1–6.
Boettcher, J. 2006. PPS parameters & their impact on rainfall estimates. NEXRAD Now Newsletter16:9–11.
Brown, J.F., B.D.Wardlow,T.Tadesse, M.J. Hayes, and B.C. Reed. 2008.The vegetationdroughtresponseindex(VegDRI):Anewintegratedapproachformonitoringdroughtstressinvegetation.GIScience and Remote Sensing45(1):16–46.
Chrisman, J. and C. Chrisman. 1999.An operational guide to WSR-88D reflectivity dataquality assurance. WSR-88D Radar Operations Center paper. http://www.roc.noaa.gov/WSR88D/PublicDocs/Publications/Reflectivity_Quality_Assurance.pdf (accessedDecember13,2011).
Chrisman, J.N., D. Rinderknecht, and R. Hamilton. 1995. WSR-88D clutter suppres-sion and its impact on meteorological data interpretation. Preprints, First WSR-88D User’s Conference. WSR-88D Radar Operations Center, Norman, OK.http://www.roc.noaa.gov/WSR88D/PublicDocs/Publications/Legacy_Clutter_paper.pdf(accessedJuly20,2011).
Fournier,J.1999.Reflectivity-rainfallraterelationshipsinoperationalmeteorology.NationalWeatherServiceTechnicalMemo,NationalWeatherService,Tallahassee,FL.
Fulton,R.,J.Breidenbach,D.-J.Seo,D.Miller,andT.O’Bannon.1998.TheWSR-88Drain-fallalgorithm.Weather Forecasting13:377–395.
Gibson, W.P., C. Daly, T. Kittel, D. Nychka, C. Johns, N. Rosenbloom, A. McNab, andG.Taylor. 2002.Development of a 103-year high-resolution climate data set for theconterminousUnitedStates.Reprints, 13th AMS Conference on Applied Climatology,AmericanMeteorologicalSociety,Portland,OR,pp.181–183.
Hunter,S.1996.WSR-88Dradarrainfallestimation:Capabilities, limitationsandpotentialimprovements.National Weather Association Digest20(4):26–36.
Krajewski,W.,G.Villarini,andJ.Smith.2010.Radar-rainfalluncertainties:Whereareweafterthirtyyearsofeffort?Bulletin of the American Meteorological Society91:87–94.
Linsley,R.,M.Kohler,andJ.Paulhus.1982.Hydrology for Engineers.3rdedn,pp.55–61.NewYork,McGraw-Hill.
Maddox, A. 2010. The importance of proper clutter filtering. NEXRAD Now Newsletter19:8–12.
Nielsen-Gammon, J. and B. McRoberts. 2009. Tracking drought in Texas:Kicking it down a notch. Sixth US Drought Monitor Forum, Austin, TX.http://atmo.tamu.edu/osc/library/osc-pubs/drought_products.pdf(accessedDecember 13,2011).
NWS/OHD/HL. 2010. MPE Editor User’s Guide, Build 9.2. NWS/OHD HydrologicLaboratory,SilverSpring,MD.
NWS/ROC. 1999. WSR-88D Interactive Training Modules: Volume 5; Build 10 WSR-88D Products.NWS/ROCOperationsTrainingBranch,Norman,OK,CD-ROM.
O’Bannon,T.1997.Usinga‘terrain-based’hybridscantoimproveWSR-88Dprecipitationestimates.Preprints, 28th Conference on Radar Meteorology,Austin,TX,pp.506–507.AmericanMeteorologicalSociety,Boston,MA.
PereiraFo,A.,K.Crawford,andC.Hartzell.1988.ImprovingWSR-88Dhourlyrainfallesti-mates.Weather Forecasting13:1016–1028.
Ryzhkov,A.andD.Zrnic.1995.Precipitationandattenuationmeasurementsata10cmwave-length.Journal of Applied Meteorology35:2121–2134.
305Estimating Precipitation from WSR-88D Observations and Rain Gauge Data
Scofield,R.A.andR.J.Kuligowski.2003.Statusandoutlookofoperationalsatelliteprecipita-tionalgorithmsforextreme-precipitationevents.Weather Forecasting18:1037–1051.
Seo,D.-J.1999.Real-timeestimationofrainfallfieldsusingradarrainfallandraingaugedata.Journal of Hydrology208:37–52.
Seo,D.-J.andJ.Breidenbach.2002.Real-timecorrectionofspatiallynonuniformbiasinradarrainfalldatausingraingaugemeasurements.Journal of Hydrometeorology3:93–111.
Seo,D.-J.,J.Breidenbach,andE.Johnson.1999.Real-timeestimationofmeanfieldbiasinradarrainfalldata.Journal of Hydrology223:131–147.
Story,G.1996.TheuseofthehourlydigitalprecipitationarrayattheWestGulfRiverForecastCenter.NWS/WGRFC,FortWorth,TX,p.15.
Story,G.2000.DeterminingWSR-88DprecipitationalgorithmperformanceusingtheStageIIIprecipitationprocessingsystem.NWA Electronic Journal of Operational Meteorology,FTT2.http://www.nwas.org/ej/2000/2000.php(accessedJuly20,2011).
Vasiloff,S.2001.WSR-88DperformanceinnorthernUtahduringthewinterof1998–1999.Part 1: Adjustments to precipitation estimates. National Weather Service Technical Attachment,National Weather Service Western Region01–02:1–7.
Vieux,B.andP.Bedient.1998.Estimationof rainfall forfloodpredictionfromWSR-88Dreflectivity:Acasestudy,17–18October1994.Weather Forecasting13:407–415.
Wilson,J.andE.Brandes.1979.Radarmeasurementofrainfall:Asummary.Bulletin of the American Meteorological Society60:1048–1058.
Zrnic,D.andA.Ryzhkov.1999.Polarimetryforweathersurveillanceradars.Bulletin of the American Meteorological Society80:389–406.
FIGURE 12.3 Widespread false precipitation, or AP, shown on the MPE radar mosaic. (Photos courtesy NOAA/NWS, Silver Spring, MD.)
FIGURE 12.5 Percent of normal rainfall for the southern United States from the AHPS precipitation analysis page for December 2010. (Image courtesy of NOAA/NWS, Silver Spring, MD.)
Exceptional droughtExtreme drought
NormalExceptionally wetStations
(a)
Abnormally wetModerately wetSeverely wetExtremely wet
Abnormally dryModerate droughtSevere drought
Exceptional wetness
Abnormal wetnessModerate wetnessSevere wetnessExtreme wetness
Exceptional droughtExtreme drought
Normal
(b)
Abnorma drynessModerate droughtSevere drought
D0 abnormally dry(c)D1 drought-moderateD2 drought-severeD3 drought-extremeD4 drought-exceptional
FIGURE 12.6 An 8 week SPI map interpolated from station-based precipitation data (a) and an 8 week SPI map derived from 4 km precipitation from MPE (b) (Image courtesy of Dr. John Nielsen-Gammon) for early September 2009 during the severe drought in southern Texas, as shown by the USDM map on September 7, 2009 (c). The circle highlights an area of excep-tional drought in the USDM that is shown to have near-normal conditions in the interpolated SPI map (a) but clearly had localized areas of severe drought conditions that were detected in the SPI map based on higher-resolution, radar-based precipitation observations (b).