NMRF/RR/12/2010

RESEARCHREPORT

NCMRWF&UKMOGlobalModel ForecastVerification:Monsoon2010

GopalIyengar,R.G.Ashrit,M.DasGupta,ManjushaChourasia, KuldeepSharma,V.S.Prasad,E.N.Rajagopal,A.K.Mitra, SajiMohandasandL.Harenduprakash

December2010
ThisisaninternalreportfromNCMRWF PermissionshouldbeobtainedfromNCMRWFtoquotefromthisreport.

NationalCentreforMediumRangeWeatherForecasting MinistryofEarthSciences A50,Sector62,NOIDA201307,INDIA

Abstract
This report summarizes the results of an evaluation of the model forecast errors of the NCMRWF Global Forecasting Systems (T254L64 & T382L64) and the U.K. Met Office (UKMO)forecastsoverIndia,duringtheSouthwestMonsoonseason(JJAS)of2010. Inthelowertroposphere(850hPa)themodelforecastsfeaturerelativelydeepermonsoontrough compared to the respective initial analysis. In the upper troposphere the 200 hPa ridge is prominentonlyintheUKMOanalysisandforecasts.TheT254modelforecastsfeatureweakand diffusedstructureofthe200hParidge.Inthelowertroposphere(850hPa)theforecastsshow relatively higher mean (JJAS) temperatures over northwest India and the Gangetic plains compared to analysis. The UKMO model initial analysis and forecasts feature higher temperatures over the peninsula and the neighboring seas by at least 3 K. In the upper troposphere(200hPa) onlyUKMOforecastsseemtocloselyagreewiththeanalysis (andis coolerthantheT382andT254models).Themeanrelativehumidityof850hPalevelinanalysis andforecastsbothindicatebroadpatternofrainfallactivity.At 200hPalevel,itisstrikingto notethattheUKMOanalysisaswellasforecastsareexcessivelydryoverlargepartsofIndian oceanandIndiaindicating UKMOmodelanalysis andforecastsfeature wetter(drier) lower (upper) troposphere compared to T254 and T382 models. Verification against the actual observationsoverIndiashowthatUKMOforecastsfeaturerelativelysmallerRMSEforwindsat both850and250hPalevels.HoweverRMSEfortemperaturearecomparableinboththemodel forecastsat850hPa,whereastheT382modelhasrelativelysmallerRMSEat250hPalevel. SimilarlyverificationagainsttheGPSsondeobservationsalsosuggestthatpredictionerrorsin NCMRWFsystemarelargerthanthatofUKMOsystem. TheobserveddistributionofmeanJJASrainfall indicateshighestrainfall ofupto2cm/day alongthewestcoastofIndiasurroundedbyrainfallintherangeof12cm/day.Similarrainfall amounts in the range of 12 cm/day are prominently seen over parts of Northeast India, GangeticplainsandalargeregioncoveringWestBengalandOrissa.Overthewestcoastand partsofnortheasternIndiathemodelforecasts(alldays)showmeanrainfallinexcessof2 cm/day at many locations surrounded by rainfall in the range on 12 cm/day. The forecasts clearlyoverestimatetheobservedrainfalloverthesetworegions.TherainfallovertheGangetic plains is over estimated in all three models particularly in Day5 forecasts. While the dry conditions of June are well captured in all forecasts of all models, the wet conditions (particularly Gangetic plains) of July, August and September are overestimated in all the forecasts. Over the peninsula the UKMO model forecasts underestimate the rainfall in all months, on the other hand the observed rainfall minima in the rain shadow region of the peninsulaareexaggerated.

Contents
GlobalModelForecastVerificationduringIndianSummerMonsoon 2010:MeanCharacteristicsandForecastErrors
1 Introduction 2 MeanAnalysis 2.1 MeanMonsoonCirculation:850and200hPaWinds 2.2 Meantemperatureat850and200hPa 2.3 MeanRelativeHumidityat850and200hPa 2.4 EvolutionoflowlevelMonsooncirculation 3 ForecastErrors 3.1 SystematicErrorsinWindsat850and200hPa 3.2 SystematicErrorsinTemperatureat850and200hPa 3.3 RootMeanSquaredErrors(RMSE) (i) ZonalWind (ii) MeridionalWind (iii) Temperature (iv) GeopotentailHeight (v) RelativeHumidity (vi) TimeseriesofRMSEinzonalwind (vii) TimeseriesofRMSEinmeridionalwind (viii) TimeseriesofRMSEintemperature (ix) TimeseriesofRMSEingeopotentialheight (x) TimeseriesofRMSEinrelativehumidity Figures131 2 3 4 5 7 8 9 10 10 11 12 13 13 14 14 15 15 15 16 16 1848

GlobalModelForecastVerificationduringIndianSummerMonsoon 2010:Verificationagainstupperairobservations 1 VerificationagainstradiosondeobservationsoverIndia 2 VerificationagainstGPSsondeobservationsoverIndia GlobalModelForecastVerificationduringIndianSummerMonsoon 2010:RainfallForecastsfromtheModels

1 2 2.1 2.2 2.3 MeanMonsoonRainfallduringJJAS2010 RainfallForecastVerification MeanError EquitableThreatScore FalseAlarmRatio Figures3644 62 64 64 64 65 6876 77 81 82 83107

Conclusions Acknowledgments References Appendix

GlobalModelForecastVerificationduring IndianSummerMonsoon2010: MeanCharacteristicsandForecastErrors

1.Introduction

This report summarizes the results of an evaluation of the model forecast errorsoftheNCMRWFGlobalForecastingSystems(T254L64&T382L64)andthe U.K.MetOffice(UKMO)forecastsoverIndia,duringtheSouthwestMonsoonseason (JJAS)of2010.ThetwomodelsatNCMRWFmainlydifferinthespatialresolution (T254~50KmandT382~35Km)andthedataassimilation.TheUKMOmodel featuresanintermediategridspacingof~40Kmandtheforecastoutputisobtained fromtheU.K.MetOffice,Exeter(UK).Theorographicfeaturesasresolvedbythe threemodelsareasshowninFigure1,overthemonsoonregion.Thepurposeofthe analysis is tocharacterize,describeandcomparethemodelforecasterrorsofthe abovesystemsusingaselectsetofmeasureswhicharewidelyusedandalsowell understood.Thesignificantpointspertainingtothiscomparisonaregivenbelow. Thecomparisonisdonefor24hr,48hr,72hr,96hrand120hrforecastsagainst the analysis from the respective forecastanalysis system (UKMO and NCMRWF)for00UTC.Theperiodoftheanalysisis1Juneto30September (122days)whichistheSouthwestMonsoonSeason. Analysis was carried out on a regular 1o latitudelongitude grid (which is coarserthanthemodelgrids)andonstandardpressurelevels(1000,925,850, 700,600,500,400,300,250,200,150and100hPalevels).Interpolationto1o grids is done using bilinear interpolation or conservative areaweighted smoothing.Gridpointslyingoverthetopographyabovethepressuresurfaces wereexcludedfromtheanalysis(masked). Noseasonaltrendremovalwas usedinthisevaluation.

TheresultsarepresentedforDay1,Day3andDay5forecastsinthisreport. The prepared data set is archived for future reference. All data format conversions(gribtonetCDF),editingofmetadata,interpolation,smoothing andthefinalcomputationswereallcarriedoutusingtheopensourcesoftware NCOandCDO.

The parameters considered are; geopotential height of the isobaric surface (GHT),airtemperature(TEMP),relativehumidity(RH),zonalwind(U)and themeridionalwind(V).

Verification procedures include intercomparison of forecast mean characteristicswiththemeananalysisandforecasterrorsareshowninterms ofsystematicerrors.

Thescoresconsideredare:RootMeanSquareError(RMSE),andTimeseries ofdailyspatialRMSE

In this analysis we have generally used the measures given by the'WWRP/WGNEJointWorkingGrouponVerification'(WWRP2009)regarding theverificationofcontinuousvariables.Thoughwehavecarriedoutacomprehensive analysis,itispracticallynotpossibleandisnotnecessarytodescribeeveryaspectof theresultsoftheanalysis.Ouraiminthisreportistoconciselypresentsuchaspects of the results that could be acted upon. We include only a limited but the most significantandusefulsubsetoftheanalysisresults.

2.MeanAnalysis.

Inthissectionwepresenttheseasonal(JJAS,2010)meananalysisofwind, temperature and relative humidity at 850 & 200 hPa level. Monthly mean characteristicsarepresentedasandwherenecessary.

2.1MeanMonsoonCirculation:850&200hPaWinds: Thecrossequatorialflow(CEF)isoneofthemaincharacteristicfeaturesof lowlevelmonsooncirculationthatstandsoutasthestrongestlowlevelflowonthe earthduringtheboreal(northern)summer.ThewindspeedinthecoreoftheSomali jetexceeds25m/s (Findlater,1969a),thejetcoreislocatedabout1.5Kmabovesea level, 200400 Km east of the east African highlands. This CEF, which is now referredtoastheFindlaterjetorSomalijet,isanessentialcomponentoftheAsian monsoonsystem.IttransportsmoisturefromthesouthernIndianOceantosouthAsia, connectstheMascarenehighandIndianmonsoontrough,andcompletesthelower branchoftheHadleycelloftheAsianmonsoon.

ThethreepanelsinthefirstcolumnofFigure2showseasonalmean(JJAS) winds(vectors)(m/s)andgeopotentialheight(contour)(gpm)intheinitialcondition of T382, T254 and UKMO models at 850 hPa. Similarly column 2, 3 and 4 correspond to the Day1, Day3 and Day5 forecasts respectively. Shading in the figure indicates geopotential height (gpm). The panels indicate that the basic circulationcharacteristics,i.e.,southwesterlyflowovertheArabianSeaat850hPa arefairlywellcapturedintheanalysisandtheforecasts.AscanbeseenfromFigure 2,themodelforecastsshowwelldevelopedmonsoontroughintheformofanarrow troughextendingfromnorthwestofIndiatotheheadBayofBengal. Theforecasts features a relatively deeper monsoon trough compared to the initial analysis particularlyinalltheforecastsof T254andT382modelsandDay3andDay5 forecastsofUKMOmodel.Thisfeatureofstrongermonsoontroughinthemodel forecastscomparedtothemodelanalysisisprominentlyseenineachofthemonths duringJuneSept2010 (FiguresinAppendix).Themeanseasonalandmonthly850 hPawindsandgeopotentialheightinthemodelforecastsclearlyindicatethatallthe 4

threemodelsconsistentlyfeatureacommontendencytoforecaststrongermonsoon circulationat850hPa.

Anextremelyimportantcomponentofthemonsooncirculationistheupper level(200hPa)monsoonridge,whichnormallyextendsfromtheMiddleEastregion tosoutheasternAsiaatapproximately27.5N.This200hParidgenormallydevelops during June and reaches full strength in July and August. Accompanying this evolutionisapronouncedshiftofthemidlatitudewesterlywindsfromsouthtonorth oftheTibetanPlateaubymidJune.Figure3showsthewindsandgeopotentialheight at200hPaintheanalysisandforecasts.TheJJASmeanforecastsseemtoclosely matchtherespectiveanalysisinallthreemodels.However,unlikeintheT382and UKMOforecasts,theT254modelforecastsfeatureweakeranddiffusedstructureof the200hParidge.ThisisprominentlyseenduringthemonthsofJune(Figure4)and September(notshown).Monthlymean200hPawindsandgeopotentialheightfor eachofthemonthsduringJunetoSeptember2010arepresentedintheAppendix. TheUKMOmodelanalysisandforecastsshowwelldefinedstructureofthe200hPa ridgeineachofthemonthsfromJunetoSeptember2010withclearchangesinthe arealextentandmigrationofpositionindicatingtheadvanceandretreatofmonsoon. IntheT382modeltheforecastsshowcloseagreementwiththeobservationsonly duringJulyandAugust.Duringadvance(June)andretreat(September)theT382 forecastsshowratherweakanddiffusedanticycloneat200hPa.IntheT254model the analysis features a weak anticyclone in all the four months and the forecasts indicateevenweakeranddiffusedpatterns(seeAppendix).

2.2MeanTemperature850&200hPa

Thelowlevel circulationoverSouthAsiaisthemostintenseduringJJAS, essentiallyduetotheextensiveHimalayanTibetanHighland,withinputofdiabatic 5

heating over a large area in the middle troposphere oriented in nearly eastwest direction.TothesouthofIndialiesextensivemassofIndianOceanwaters.During thenorthernsummerseason,thisconfigurationoflandmassandwatermasscreates strongmeridionalgradientsoftemperatureindicatingflowofairfromcooloceanarea to warm land area. Accurate representation of the observed land sea temperature contrastanditsevolutioninthemodelinitialanalysisiscrucialforpredictingthe onsetandadvanceofthemonsoonoverIndia.Figure5presentstheseasonal(JJAS, 2010)meantemperaturedistributionat850hPaintheinitialconditions(analysis)and theforecasts.Asearlierdiscussed,thefirstcolumnintheFigure4correspondstothe modelanalysisandthecolumns2,3and4correspondtotheDay1,Day3andDay5 forecasts.Similaranalysisiscarriedoutforeachofthemonthsandthefiguresare presentedinthe Appendix. Inthemodel initial analysis (column1),allthethree modelsindicateastrongnorthsouthtemperaturegradientovertheIndianlandregion withthehighesttemperatureexcessof295KovernorthwestIndia.However,over adjoiningPakistanregion,highertemperaturesofupto298Kpersist.Overcentral IndiaandGangeticplains,temperaturesareintherangeof292295Kintheanalysis. Themodelforecasts(column2,3and4)indicaterelativelyhighermeantemperatures overnorthwestIndiaandtheGangeticplainscomparedtoanalysis.Thisfeatureis mainlyseenduringJuneandisnotveryprominentinJuly,AugandSept(Figuresin Appendix).PartsofpeninsularIndia,ArabianSeaandBayofBengalfeaturelower temperaturesintherangeof290292KintheT254andT382models.TheUKMO model initial analysis features higher temperatures over the peninsula and the neighboringseasbyatleast3K.ThisisalsoprominentintheUKMOforecaststoo. ThehighertemperaturevaluesovertheentirenorthIndianOceanisprominentinthe UKMOanalysisandforecastsineachofthemonthsduringJJAS2010.

Themeantemperaturedistribution at200hPais showninFigure6.Both analysisandforecastsshowhighertemperaturesoverthecontinentswithdecreasing 6

temperaturestowardstheIndianOcean.OnlyUKMOforecastsseemtocloselyagree withtheanalysis.BothT382andT254modelsunderestimatethewarmtemperatures overcontinent.FurtherUKMOmodelanalysisandforecastsshow relativelycooler temperatures(comparedtoothertwomodels)overtheIndianOcean;particularly over eastern Indian Ocean. The above two aspects of 200 hPa temperature distributioninthemodelforecastsisseenduringallfourmonths(seeAppendix).

2.3MeanRelativeHumidityat850&200hPa

Similartothediscussionintheearliersubsection,meanconditionofmoisture distribution(relativehumidity)intheinitialanalysisandtheforecastsarepresented here.TheAsiansummermonsoonplaysacrucialroleinmoisturetransport.Themost remarkablemoisturechanneloriginatesinthesouthernIndianOcean,crossingthe equatorneartheSomalicoastalregion,flowingtotheArabianSeaandtheBayof Bengal.

Theseasonal(JJAS)meanrelativehumidityat850hPaforthethreemodelsis showninFigure7.Thisfieldissimilartothemeanseasonalrainfalldistributionover theregion.Thehighmoisturecontentinallthemodelscanbeindicatedbyrelative humidityexceeding80%coveringtheIndianlandregionandBayofBengal.Arabian Sea and Indian Ocean feature relatively lower relative humidity. High values of relativehumidity exceeding90% alongthewestcoastof Indiaandwestcoastof Myanmarsuggesttheimpactofsteeporography.HighhumidityseenovertheIndo Gangeticbasin(>80%)aremanifestationofthemonsoontroughandtheassociated convectioninthatregion.Thehighmoistureinthisregionseenintheseasonalmean canbeassociatedwiththemonsoondepressionthattravelsalongthemonsoontrough. ThereducedhumidityoverthenorthwestIndiaisbyandlargewellcapturedinthe forecasts.However,monthlymean850hParelativehumidityplots(appendix)show 7

relatively drier conditions in June and September mainly over central and northwesternIndia.During July andAugust UKMO model analysis andforecasts typicallyshowlargeareaoverIndiawithrelativehumidityabove90%(Figure8a,b). Similaranalysisforthe200hParelativehumidityshowsthattheforecastsclosely agreewiththeanalysis.However,itisstrikingtonotethattheUKMOanalysisaswell asforecastsareexcessivelydryoverlargepartsofIndianoceanandIndia(Figure9). Thisisseeninallthemonthsoftheseason(appendix).Basedonthemonthlyand seasonalmeanrelativehumiditydiscussedinthissection,itcanbeconcludedthat UKMOmodelanalysisandforecastsclearlyfeaturearelativelywetter(drier)lower (upper)tropospherecomparedtoT254andT382models.

2.4EvolutionofLowlevelMonsoonCirculation:

Thestrongcrossequatoriallowleveljetstreamwithitscorearound850hPa isfoundtohavelargeintraseasonalvariability.Figure10a,b(andFigure11a,b)show theHovmollerdiagramofzonalwind(U)of850hPaaveragedoverthelongitude band6070E(and7580E)andsmoothedbya5daymovingaveragefortheperiod 1June30September2010fortheT382andUKMOmodelsrespectively.Thetop panelineachfigureshowstheanalysisandthemiddleandthelowerpaneldepictthe Day3andDay5forecastsrespectively.Theactivemonsoonspellsarecharacterized bystrongcoresofzonalwind. ThemonsoonhadsetinoverKeralaon31st May.
Subsequentadvancementofthemonsoonacrosswestcoastwasdelayedbyaboutone weekduetotheformationofaveryseverecyclonicStorm(PHET,31stMay2ndJune). Thereafter,themonsooncoverednearlyhalfofthecountrybythemiddleofJune.There

wasaprolongedhiatusintheadvancementofmonsoontilltheendofJune dueto
weakeningofmonsooncurrent.Thesouthwestmonsooncoveredtheentirecountryby6
th

July. AsseenfromtheanalysispaneloftheT382modelthezonalwindflowwas

quiteweak duringmostpartsofJuneandinthefirstfortnightofJuly. Thelowlevel 8

westerlyflowpickedupstrengthwithacoreofzonalwindofabout20 m/s inthe secondfortnightofJulyandremainedsotilltheendofthemonth.Thiswasfollowed byaspellofweakcoreofzonalwindforaperiodoftwoweeks.Anotherspellof strongcoreofzonalwindofabout15m/swasseeninthefirstfortnightofSeptember. TheUKMOanalysesshowcomparativelystrongerzonalwinds.TheDay3andDay5 forecastsofboththemodelsagreereasonablywellwiththeanalysisandareableto depicttheactiveandweakspellsofthemonsoonflow.Howeverthewindstrengthis weakerduringtheactivespellintheDay5forecasts.

Figure 11 shows the Hovmoller diagram of zonal wind (U) of 850 hPa averagedoverthelongitudeband7580Eandsmoothedbya5daymovingaverage for the period 1 June30 September 2010 for the T382 and UKMO models respectively.AsinFigure10,thetoppanelshowstheanalysisandthemiddleandthe lowerpaneldepicttheDay3andDay5forecastsrespectively.BoththeT382and UKMOanalysisshowaprominentnorthwardmovementofthecoreofzonalwind duringthesecondfortnightofJuly.Twoweakspellsareseeninthesecondandthird weakofJuneandfromthefourthweekofAugusttothesecondweekofSeptember. TheDay3forecastscomparewellwiththeanalysis.However,theDay5forecastsof theT382modelarenotabletodepictthenorthwardmovementofthecoreofzonal windasseenintheanalysis.However,theUKMODay5forecastsdepictthisfeature comparativelybetterthantheT382model.

3.ForecastErrors

Herewepresentthemodelforecasterrorsexpressedintermsofsystematic error and Root Mean Squared Error (RMSE), with an aim to provide qualitative descriptionofthespatialdistributionoferrors. 9

3.1SystematicErrorsinWindat850&200hPa:

Figures12illustratesthesystematicerrorsinthemodelforecastwindfields (vectors)(m/s)forJJASat850hPainthethreemodels.Thepanelsinthefirstcolumn correspondtothemeananalysisandthecolumns2,3and4correspondtosystematic errorsinDay1,Day3andDay5forecast.Theshadinginthepanelsindicateserrors intheforecastzonalwind.Theforecastsofallthreemodelsshow(i)westerlybias overnorthArabianSea,CentralIndiaextendingtoSoutheastAsia.(ii)Southofthis eastwestregioneasterlybiasisseen.Thisbroadlysuggeststhattheforecastgenerally produceamonsooncirculationthatisslightlyshiftednorthwards.(iii)Centraland easternequatorialIndianOceanfeaturestrongeasterlybias,particularlyintheUKMO modelforecasts.Otherthanthesethreeprominentfeatures,Figure10showseasterly biasovertheGangeticplainsintheUKMOandT254forecasts.Theabovementioned biasesareseeninallthemonthsoftheseason(Appendix).Similarlythesystematic errorsinthewindsat200hPaarepresentedinFigure13.Strongwesterlybiasinthe easternIndianOceanintheDay3andDay5forecastsisprominentparticularlyinthe T254andT382models.

3.2SystematicErrorsinTemperatureat850&200hPa

Figure14showstheSeasonal(JJAS,2010)mean850hPatemperatureinthe analysisandthesystematicerrorsintheforecasts.TheT254andT382modelsshow slightcoldbiasoverthenorthArabianSea(about0.4)withstrongwarmbiasover thePakistanregion(>1.2).TheUKMOmodelforecastsshowstrongwarmbias overthenorthArabianSea(>1.4)andstrongcoldbiasovertheheatlowregion(< 1.2).SimilaranalysisforthemonthsofJulyandAugustisshowninFigure15.The featuresofwarm(cold)biasovertheheatlowregion(northArabianSea)intheT254 andT382modelsispersistentinallthemonths.ForecasterrorsintheUKMOover 10

thisregionarejusttheoppositewithcoldbiasovertheheatlowregionandwarm biasoverthenorthArabianSea. Theseaspectsareconsistentlyseenduringallthe monthsoftheseasonandareshownforJulyandAugustinFigure15.

At200hPa(Figure16),T254andT382modelsfeaturecomparable/similar patternofsystematicerrorswhileUKMOmodelfeaturescompletelydifferentpattern. In the T382 model large area covering Bay of Bengal (Arabian Sea) dominantly showswarm(cold)biasinDay1,Day3andDay5forecasts.T254modelalsoshows similarbiasesovertheentiredomainintheDay1andDay3forecasts,whilethe Day5forecastsshowlargepartoflandandSeaundercoldbias.UKMOforecasts feature warmbias over continents andcold bias over the Indianocean.Only the UKMO forecasts consistently show warm (cold) bias over the continent (Indian Ocean)duringallthemonthsoftheseason(seeAppendix).

3.3RootMeanSquaredError(RMSE)

TheRootMeanSquaredErrorisgivenby

where Fi and Oi respresent forecast and observed fields respectively. TheRMSE measuresthe"average"error,weightedaccordingtothesquareoftheerror.However, itdoesnotindicatethedirectionofthedeviations.Withvaluesrangingfrom0to, RMSEputsgreaterinfluenceonlargeerrorsthansmallererrors,whichmaybeagood thingiflargeerrorsareespeciallyundesirable,butmayalsoencourageconservative forecasting.

11

Thegeographicaldistributionsoftherootmeansquareerroroftheforecasts (foreachmonthandtheseasonasawhole)werecalculatedfromthedifferenceat eachgridpointoneachday.Gridpointslyingbelowgroundwereexcludedfromthe computation.Thesechartsaregivenforallthevariablesatthepressurelevels850and 200hPa. Thetimeseriesofthespatialrootmeansquareerrorforavariableover India (6895E,538N), as a time series of 1, 2, 3, 4 and 5 day forecasts were computedforallthevariablesat850,700,500and200hPalevels.Incombination withtheotherscoresthetimeseriesprovidesusefulinformationonconsistencyinthe errorcharacteristicsoftheforecastduringtheseason(daytodayvariationsinforecast errors). TheRMSEofwinds(zonalandmeridional),geopotentialheight,temperature andrelativehumidityforUKMOandNCMRWF(T254andT382)Day1,Day3and Day5forecastsarediscussedindetailbelow.ThetimeseriesofdailyspatialRMSE oftheabovevariablesovertheIndianregionarealsodiscussed.

(i)ZonalWind:

PanelsinFigure17showtheRMSEof850hPazonalwindforDay1,Day3 and Day5 forecasts of the T254, UKMO and T382 models respectively. The magnitudeofRMSEat850hPaisoftheorderof24m/sinDay1forecastinallthe modelforecasts.TheT382andT254modelsshowamarkedincreaseintheRMSE fromDay1toDay5forecastascomparedtoUKMO,especiallyoverthenorthwest andplainsofIndia.Figures18issameasFigure17butfor200hPalevel.At200hPa, the magnitude varies from 24 m/s in UKMO and 26 m/s in Day1 forecast of T382/T254 models over the Indian region. T382 and T254 model shows a considerableincreaseinRMSEfromDay1toDay5forecastespecially overthe southernpeninsulawheretheTropicalEasterlyJet(TEJ)isprominent.

12

 (ii)MeridionalWind:

PanelsinFigure19aresameasinFigure17butformeridionalwinds.At850 hPa,themagnitudeofRMSEinmeridionalcomponentofthewindisoftheorderof 24m/sinDay1forecastinallthemodelsovertheIndianandneighborhoodregion. AllthemodelsshowaconsistentincreaseinRMSEfromDay1toDay5forecast withthemagnitudeofabout46m/s. Figures20issameasFigure18butfor

meridionalwinds.ThemagnitudeofRMSEisoftheorderof24m/sand26m/sin Day1forecastsoftheUKMOandT382modelsrespectivelyovertheIndianregion. TheT382andT254modelsshowaconsistentincrease(morethan6m/s)inRMSE fromDay1toDay5forecast.

(iii)Temperature:

PanelsinFigure21showtheRMSEof850hPatemperatureforecasts.At850 hPa,themagnitudeofRMSEisoftheorderof11.5KinDay1forecastoverthe northwest India for T254 and T382 whereas the magnitude is slightly smaller in UKMO.ThereisincreaseinRMSEfromDay1toDay5forecastinallthemodel forecasts. The T254 and T382 models show consistently higher RMSEs over the northwestpartsofIndiaascomparedtotheUKMOforecasts.Figures22showthe RMSEof200hPatemperatureforecasts.Atupperlevel(200hPa),themagnitudeof theerrorsislessascomparedtothelowerlevel(850hPa).Theorderofthemagnitude at200hPais~0.51KovertheIndianmainlandanditsneighborhood.Themagnitude ofRMSEsinDay5forecastsofallthemodelforecastsisabout0.52K,overthe Indianregion,withlargererrorsoverthenorthernparts.

13

(iv)GeoPotentialheight:

PanelsinFigure23showtheRMSEof850hPageopotentialheightforecasts. ThemagnitudeofRMSEinUKMOandT382Day1forecastsoverIndiaissimilar andoftheorderof510gpm.TheincreaseinRMSEfromDay1toDay5inT254is verymuchhighascomparedwithUKMOandT382model.HighvaluesofRMSEin theT254forecastsareseenovertheIndianlandregionsaswellastheArabianSea andBayofBengal.PanelsinFigure24showtheRMSEof200hPageopotential heightforecasts.ThemagnitudeofRMSEinUKMOandT382forecastsoverIndiais similarandoftheorderof010gpm.ThemagnitudeofRMSEsinDay5forecastsof UKMOandT382forecastsisabout1040gpm,overtheIndianregion,withlarger errorsoverthenorthernparts.TheT254modelshowslargerRMSEsascomparedto theUKMOandT382model.

(v)RelativeHumidity:

PanelsinFigure25showtheRMSEof850hParelativehumidityforecasts. TheDay1forecastserrorsofallthreemodelsarerathersimilar. Themagnitudeof RMSEishigherinUKMOascomparedtoT254andT382modelforecasts. High valuesofRMSEareobservedoverthenorthwestandplainsofIndiaintheDay3and Day5forecastsoftheUKMOandT382models.PanelsinFigure26showtheRMSE of200hParelativehumidityforecasts.At200hPa,theRMSEinUKMOmodelis less as compared to T254 and T382 model. The magnitude of RMSE is almost constantupto5thdayforecastforUKMOmodelovertheIndianmainlandwhileitis consistentlyincreasingforT254andT382modelsandgoesmorethan40%inthe Day5forecasts.

14

(vi)TimeSeriesofRMSEofZonalWind:

TheRMSEofzonalwindat850,700,500,and200PalevelsofDay1,Day3 andDay5forecastsforeachdayhavebeencomputedfortheIndiandomainof5 38Nand6894E.Figure27showsthetimeseriesofRMSEofzonalwindatabove mentionedfourlevels.Theblack,redandgreenlinescorrespondtotheT254,UKMO andT382modelforecastsrespectively.TheaverageseasonalRMSEateachlevelare alsoshowninsimilarcolors. TheRMSEintheUKMOforecastsislowerthanthe T254andT382modelsinalltheforecasts.InthebeginningofJune,allthemodels showalargespatialRMSE,whichwasduetotheerrorsintheforecastofthetropical cyclone'Phet'whichdevelopedovertheArabiansea.UKMOmodelshowsahigh valueinthelastweekofSeptemberinDay1forecast.

(vii)TimeSeriesofRMSEofMeridionalWind:

PanelsinFigure28showsimilarfiguresforthemeridionalcomponentofthe wind.TheyalsoshowthelowermagnitudesofRMSEinUKMOforecastsatallthe levels.

(viii)TimeSeriesofRMSEofTemperature:

ThetimeseriesofdailyRMSEoftemperature(Figure29)isalmostsimilarto thatofwindsexcepttwopeaksinlasttwoweeksofSeptembershownbyUKMO modelinDay1forecast.Onepeakaround23Septemberatallthelevels canbe consistentlyseeninalldaysforecastofUKMO.HoweverthemagnitudeofRMSEin UKMOforecastsarelessascomparedtoT254andT382modelsatallthelevels.At

15

850hpalevel,themagnitudeofRMSEismoreinJuneascomparedtotheother monthsinDay3andDay5forecasts.

(ix)TimeSeriesofRMSEofGeoPotentialheight

ThetimeseriesofdailyRMSEofgeopotentialheight(Figure30)showsthat theUKMOmodelhavesmallerRMSEsascomparedtotheothertwomodels.There areverymuchlargevariationsinerrorinalldaysforecast.T254modelpredictsgeo potentialheightverybadlyatallthelevels.

(x)TimeSeriesofRMSEofRelativeHumidity: RMSEinrelativehumidityat850hPaand700hPaishigherinUKMOas comparedtoT254andT382modelforecastwhilethesameislowerattheupper levels(500and200hPa)asshowninFigure31.TheRMSEvaluesarehigherduring thefirstweekofJuneinalldaysforecastanderrorsarehighatallthelevelsin UKMOinthelastweekofSeptember.

TheRMSEcomputationsdiscussedaboveshowthemodelerrorsasafunction oftimewithintheseason.Table1givestheaverageRMSEvaluescorrespondingfor the season as a whole. UKMO forecasts feature relatively smaller RMSE for meridionalwind(v),geopotentialheight(HGT)forboth850and200hPa;andRHfor 200hPa.

16

Table1.Day1Day5RootMeanSquareError(RMSE)ofWind(Zonal,Meridional), Temperature,RelativeHumidityandGeopotentialheightovertheIndianregion(68 94E,538N)ofdifferentModels(UKMO,T254,T382)

UKMO u(m/s) v(m/s) T(0K) RH(%) HGT(m) T382 u(m/s) v(m/s) T(0K) RH(%) HGT(m) T254 u(m/s) v(m/s) T(0K) RH(%) HGT(m) Day1
850hPa 200hPa

Day2
850hPa 200hPa

Day3
850ha 200hPa

Day4
850hPa 200hPa

Day5
850hPa 200hPa

2.11 1.93 0.60 7.37 5.19

3.10 2.96 0.50 10.70 7.03

2.57 2.33 0.74 8.91 6.73

3.76 3.47 0.64 12.29 9.05

3.03 2.63 0.85 9.98 8.06

4.17 3.80 0.72 12.86 11.72

3.44 2.89 0.94 10.90 9.68

4.55 4.16 0.79 13.49 13.95

3.77 3.11 1.02 11.61 11.25

4.95 4.47 0.88 13.96 16.77

Day1
850hPa 200hPa

Day2
850hPa 200hPa

Day3
850hPa 200hPa

Day4
850hPa 200hPa

Day5
850hPa 200hPa

2.80 2.51 0.67 6.84 6.97

4.23 3.88 0.63 16.38 8.53

2.57 3.04 0.84 8.39 9.53

5.30 4.64 0.83 20.58 12.73

3.89 3.29 0.93 9.07 10.01

5.77 4.94 0.90 21.94 15.52

4.24 3.50 1.01 9.61 11.18

6.09 5.36 1.00 22.87 18.37

4.61 3.72 1.10 10.05 12.47

6.49 5.73 1.08 24.08 21.29

Day1
850hPa 200hPa

Day2
850hPa 200hPa

Day3
850hPa 200hPa

Day4
850hPa 200hPa

Day5
850hPa 200hPa

2.88 2.53 0.66 5.91 9.83

4.62 4.13 0.70 16.23 14.87

3.57 3.06 0.83 7.83 11.85

5.46 4.84 0.89 20.78 18.11

4.03 3.37 0.93 8.65 12.87

5.85 5.23 0.98 21.75 20.51

4.37 3.66 1.02 9.24 14.05

6.18 5.48 1.05 22.54 23.89

4.79 3.92 1.07 9.77 14.73

6.60 5.77 1.15 23.14 26.68

17

18 Figure1.Terrainheight(Km)inT254,T382andUKMOglobalmodels.

19

Figure2.Meananalysisof850hPawindsandgeopotentialsforJJAS2010fromT254,T382andUKMOglobalmodels.

20

Figure3.Meananalysisof200hPawindsandgeopotentialsforJJAS2010fromT254,T382andUKMOglobalmodels.

21

Figure4.Meananalysisof200hPawindsandgeopotentialsforJune2010fromT254,T382andUKMOglobalmodels.

22

Figure5.Meananalysisof850hPaTemperature(K)forJJAS2010inT254,T382andUKMOglobalmodels.

23

Figure6.Meananalysisof200hPaTemperature(K)forJJAS2010inT254,T382andUKMOglobalmodels.

24

Figure7.Meananalysisof850hPaRelativeHumidity(%)forJJAS2010inT254,T382andUKMOglobalmodels.

(a)

(b)

Figure8.Meananalysisof850hPaRelativeHumidity(%)for(a)Julyand(b)August 2010inT254,T382andUKMOglobalmodels.

25

26

Figure9.Meananalysisof200hPaRelativeHumidity(%)forJJAS2010inT254,T382andUKMOglobalmodels.

Figure10.850hPaZonalwind(m/s)averagedover6070EduringJJAS2010inthe analysisandforecastsof(a)T382and(b)UKMOmodels.

27

Figure11.850hPaZonalwind(m/s)averagedover7580EduringJJAS2010inthe analysisandforecastsof(a)T382and(b)UKMOmodels.

28

29

Figure12.Systematicerrorin850hPawindforJJAS2010inT254,T382andUKMOglobalmodels.

30

Figure13.Systematicerrorin200hPawindforJJAS2010inT254,T382andUKMOglobalmodels.

31

Figure14.Systematicerrorin850hPaTemparatureforJJAS2010inT254,T382andUKMOglobalmodels.

Figure15.Systematicerrorin850hPaTemparaturefor(a)Julyand(b)August2010 inT254,T382andUKMOglobalmodels.

32

33

Figure16.Systematicerrorin200hPaTemparatureforJJAS2010inT254,T382andUKMOglobalmodels.

34

Figure17.RMSEof850hPazonalwindforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

35

Figure18.RMSEof200hPazonalwindforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

36

Figure19.RMSEof850hPameridionalwindforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

37

Figure20.RMSEof200hPameridionalwindforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

38

Figure21.RMSEof850hPatemperatureforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

39

Figure22.RMSEof200hPatemperatureforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

40

Figure23.RMSEof850hPageopotentialheightforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

41

Figure24.RMSEof200hPageopotentialheightforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

42

Figure25.RMSEof850hParelativehumidityforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

43

Figure26.RMSEof200hParelativehumidityforDay1,Day3andday5forecastsoftheT382,UKMOandT254models.

44

Figure27.TimeseriesofRMSEofzonalwindat850,700,500,and200PalevelsforDay1,Day3andDay5forecastsofT382,UKMOandT254models.

45

Figure28.TimeseriesofRMSEofmeridionalwindat850,700,500,and200PalevelsforDay1,Day3andDay5forecastsofT382,UKMOandT254models.

46

Figure29.TimeseriesofRMSEoftemperatureat850,700,500,and200PalevelsforDay1,Day3andDay5forecastsofT382,UKMOandT254models.

47

Figure30.TimeseriesofRMSEofgeopotentialheightat850,700,500,and200PalevelsforDay1,Day3andDay5forecastsofT382,UKMOandT254models.

48

Figure31.TimeseriesofRMSEofrelativehumidityat850,700,500,and200PalevelsforDay1,Day3andDay5forecastsofT382,UKMOandT254models.

GlobalModelForecastVerificationduring IndianSummerMonsoon2010: VerificationagainstObservations

1. VerificationagainstradiosondeobservationsoverIndia

ObjectiveverificationscoresfortheNCMRWFT382andUKMOforecastsof windsandtemperatureagainsttheobservationsvalidfor00UTCatstandardpressure levels(850and250hPalevels)asrecommendedbytheWMOwerecomputedforthe Indian regionforthemonsoonseasonof2010. Table2givestheaverageRMSE valuescorrespondingfortheseasonasawhole.UKMOforecastsfeaturerelatively smaller RMSE for winds at both 850 and 250 hPa levels. However RMSE for temperaturearecomparableinboththemodelforecastsat850hPa,whereastheT382 modelhasrelativelysmallerRMSEat250hPalevel.

2.VerificationagainstobservationsatdifferentGPSsondestationsoverIndia

Anumberofchangesoccurinthecirculationoftheatmospherealongwiththe onsetofsouthwestmonsoonovertheIndiansubcontinent. Acompletereversalof wind takes place at many places over India. Temperature pattern also exhibits a drastic change with occurrence of rainfall. Though the large scale features of monsoonarecapturedwellbynumericalmodellingsystems,theaccurateprediction oftheseparameters,atanylocationhasremainedachallengingtask.Inthissection anattempthasbeenmadetoverifywindandtemperatureasanalysedandpredicted by UKMO and NCMRWF T382L64 global model against upperair observations (GPSsonde)atvariousIndianstationsduringmonsoonseasonof2010.

49

Since March 2009, India Meteorological Department has upgraded its 11 radiosonde stations by GPSsonde, resulting in improved quality of upperair observationsoverIndia.Thelocationsofthesestationsaredepictedinfigure32.

Anattempthasbeenmadetoverifytheanalysedandpredictedtemperature

andwindofboththemodelsagainstrespectiveobservationsatthese11stations.Itis worthytomentionherethat,duringthisperiodtherearelargedatagapsatmany stationsexceptDelhi.Figures33and34depictthedailyvariationsinanalyses(0000 UTC)andsubsequentpredictions(day1,3,5)ofzonalwind(u)at850and200hPa level respectively, along with observations at Delhi. The count in each figure indicatesthetotalnumberofdaysonwhichobservationisavailableoutofthe122 daysofmonsoonseason.Forthisverification,observationsaretakenfromNCMRWF operationalarchives.Therootmeansquareerror(RMSE)ofanalysesandpredictions computedagainstobservationsarealsoshownintheplot.

Asseenfromtheplot,thoughtheRMSEofanalysisofNCMRWFsystemis lesserthanthatofUKMOsystemtheRMSEofpredictionsofNCMRWFmodelare higher than that of UKMO model over Delhi. During 2010 JJAS, there are two episodesofstrongeasterlyoverDelhiat850hPa,oneon4th Julyandotheron27th July,bothofwhicharewellcapturedinUKMOandNCMRWFanalyses.Though boththemodelsmoreorlesspredictedthechangeoverofwindregimefromwesterly toeasterly,thepredictionsofUKMOmodelfortheseepisodesarebetterthanthatof NCMRWFmodel. HoweverUKMOmodelhasshowneasterlybiasoverDelhiin contrasttowesterlybiasseeninNCMRWFmodel,especiallyin120hr.predictions.

Intheupperlevel,zonalwindvariationoverDelhiisrelatedtotheposition andmovementofsubtropicalanticyclone.Theobservedvariationof200hPazonal

50

wind at Delhi is well captured by both UKMO and NCMWRF analyses and predictions.Boththemodelshaveshownwesterlybiasintheupperlevel.

Figure35depictsthedailyvariationsinanalyses(0000UTC)andsubsequent predictions(day1,3,5)oftemperature(t)at925hPalevel.In2010,theprogressof southwestmonsoonoverDelhi,andneighboringregiontookplaceon5th July,and there after this region received a good amount of rainfall throughout the season resultinginaseasonalrainfall21%abovethelongperiodaverage(LPA)rainfallover thisregion.Asseenfromtheplot,theobservedtemperatureofDelhiat925hpalevel also shows a slight fall on 3rd July onwards and after that it remains moderate throughouttheseason.Thissuddenfalloftemperatureassociatedwithprogressof monsoon was well captured by UKMO model, but apart from 24hr. prediction, NCMRWFmodelcouldnotpredictthesame.UKMOmodelhasshowncoldbiasin contrasttowarmbiasinNCMRWFmodel.

SeasonalRMSEandbias(Analysis/predictionobservation)oftemperature (t)andzonal(u)componentofwindforbothUKMOandNCMRWFT382L64model analyses and 24hr., 72hr. , 120hr. predictions are shown in Table 3 and 4 respectively.

Ingeneralitisseenthat,thoughRMSEsofNCMRWFanalysesarelower comparedtothatofUKMO,astheforecastprogresses,predictionfromNCMRWF modeldeterioratesveryrapidlyandinmanycasesDay5RMSEofNCMWRFis equaltothatofDay3ofUKMOmodel.BoththemodelshaveshownlargeRMSEfor temperatureandzonalwindatVisakhapatnamatalllevels,whichmaybeassociated withpoorpredictionofthetrackofcycloniccirculations(monsoonlow)overthis region.SinceobservationsofMinicoywereveryfewduringthisperiod,thestatistics forthesamearenotshowninthetable. 51

UKMOglobalanalysesandmodelpredictions(upto5days)forzonalwind and temperature at different levels of atmosphere for 0000 UTC 1st June 30th September2010havebeenverifiedagainstallavailableradiosondeobservations. UKMOforecastsfeaturerelativelysmallerRMSEforwindsatboth850and250hPa levels.HoweverRMSEfortemperaturearecomparableinboththemodelforecastsat 850hPa,whereastheT382modelhasrelativelysmallerRMSEat250hPalevel.The verificationagainstthe11GPSsondeobservationsoverIndianregionisalsocarried out. The scores computed also have been compared with that from NCMRWF (T382L64) analysisforecast system. Results of this verification study are summarisedasfollows:

(i)

Transitionobservedindifferentmeteorologicalparameterswithonsetof monsoon are captured well in daily variation of UKMO analysed and predictedfields.

(ii)

RMSE of UKMO analyses are larger compared to that of NCMRWF analyses,especiallyinthelowerlevels.

(iii)

HoweverthepredictionerrorsinNCMRWFsystemarelargerthanthatof UKMOsystem.

52

Figure32Locationof11upgradedIndianupperairsondestations

53

54

Figure33.Dailyvariationofobserved&analysed/predictedzonalwind(u)atDelhiat850hPalevelfor0000UTC1stJune30thSeptember2010 (a)UKMO(b)NCMRWFT382L64

55

Figure34.Dailyvariationofobserved&analysed/predictedzonalwind(u)atDelhiat200hPalevelfor0000UTC1stJune30thSeptember2010 (a)UKMO(b)NCMRWFT382L64

56

Figure35.Dailyvariationofobserved&analysed/predictedzonaltemperature(K)atDelhiat925hPalevelfor0000UTC1stJune30thSeptember2010 (a)UKMO(b)NCMRWFT382L64

Table2:RMSEoftheUKMOandNCMRWFforecastscomputedagainst Indian upperairradiosondeobservations,for0000UTCWind(m/s)andTemperature(K) at850,and200hPalevelsforJJAS2010

UKMO Wind(m/s) Temp(0K) T382 Wind(m/s) Temp(0K) Day1 850hPa
4.76 1.89

Day3 850hPa
5.44 1.95

Day5 850hPa
6.19 2.03

250hPa
5.77 3.32

250hPa
6.12 3.30

250hPa
6.58 3.31

Day1 850hPa
5.11 1.92

Day3 850hPa
5.88 1.97

Day5 850hPa
6.51 2.00

250hPa
6.06 3.14

250hPa
6.69 3.10

250hPa
7.22 3.10

57

Table 3: RMSE and Bias computed against Indian GPSsonde observations, averagedfor0000UTCTemperature(K)analysesandpredictionsat850,500and 200hPalevelsforJJAS2010
Station Level (obs count) 500 (68) 200 (66) 850 (114) 500 (114) 200 (110) 850 (81) 500 (77) 200 (76) 850 (64) 500 (62) 200 (56) 850 (74) 500 (76) 200 (73) 850 (64) 500 (64) 200 (60) 850 (65) 500 (64) 200 (64) 850 (70) 500 (66) 200 (66) Analysis UKMO NCMR 1.48 0.15 0.76 0.53 0.86 0.22 0.83 0.01 0.74 0.54 0.73 0.24 0.89 0.32 0.96 0.51 0.76 0.32 0.63 0.09 1.08 0.78 1.90 1.36 1.32 0.56 1.04 0.41 0.76 0.36 0.82 0.17 0.92 0.33 0.54 0.01 0.59 0.02 0.86 0.51 0.65 0.25 0.57 0.02 0.82 0.47 1.5 0.02 0.91 0.36 0.91 0.46 0.66 0.03 0.56 0.06 0.73 0.04 0.92 0.29 1.06 0.34 0.85 0.40 0.63 0.03 0.88 0.31 1.94 1.30 1.31 0.52 1.11 0.45 0.62 0.02 0.81 0.03 0.99 0.30 0.67 0.38 0.68 0.01 1.27 0.74 0.78 0.34 0.63 0.04 1.01 0.61 D1(24hr)FCST UKMO NCMR 1.8 0.22 1.23 0.61 1.20 0.53 1.09 0.19 0.79 0.45 0.96 0.24 1.09 0.15 1.09 0.32 1.29 0.84 1.06 0.23 1.36 0.89 2.11 1.48 1.39 0.61 1.14 0.39 1.17 0.61 1.02 0.19 1.08 0.48 0.77 0.11 0.88 0.07 1.08 0.52 1.03 0.11 0.91 0.01 0.93 0.48 1.7 0.12 1.28 0.12 1.80 0.79 1.19 0.27 0.76 0.01 1.04 0.04 1.13 0.81 1.18 0.35 1.06 0.26 1.04 0.35 1.06 0.43 2.13 1.47 1.39 0.37 1.23 0.62 0.89 0.26 1.06 0.14 1.26 0.42 0.97 0.56 0.87 0.09 1.52 0.86 1.13 0.46 0.98 0.15 1.26 0.79 D3(72hr)FCST UKMO NCMR 2.03 0.67 1.12 0.56 1.53 0.55 1.17 0.08 0.87 0.44 1.01 0.05 1.18 0.27 1.14 0.34 1.37 0.99 0.98 0.12 1.33 0.90 2.04 1.32 1.52 0.68 1.17 0.18 1.22 0.66 1.21 0.35 1.07 0.45 0.75 0.13 1.06 0.23 1.02 0.40 1.17 0.02 1.02 0.31 0.94 0.42 1.95 0.71 1.51 0.10 2.25 1.15 1.30 0.28 0.87 0.01 1.12 0.05 1.59 1.08 1.11 0.55 1.12 0.14 1.10 0.26 1.27 0.75 2.11 1.33 1.36 0.39 1.20 0.48 0.99 0.37 1.08 0.11 1.27 0.25 1.09 0.68 0.96 1.17 1.22 0.68 1.25 0.56 1.21 0.14 1.21 0.69 D5(120hr)FCST UKMO NCMR 2.23 0.55 1.41 0.55 1.74 0.23 1.23 0.03 0.94 0.46 1.06 0.15 1.17 0.26 1.15 0.49 1.45 0.97 1.12 0.11 1.51 1.12 2.01 1.07 1.58 0.70 1.07 0.05 1.27 0.41 1.26 0.42 1.05 0.27 0.74 0.17 1.19 0.28 1.01 0.17 1.13 0.01 1.03 0.38 0.92 0.29 2.03 0.67 1.83 0.36 2.73 1.46 1.44 0.27 0.88 0.08 1.15 0.20 1.43 0.91 1.15 0.52 1.24 0.53 1.16 0.25 1.24 0.72 1.99 1.23 1.52 0.27 1.33 0.53 1.03 0.30 1.28 0.11 1.32 0.35 1.06 0.62 1.09 0.02 1.27 0.66 1.28 0.63 1.18 0.15 1.14 0.54

Srinagar

Delhi

Mohanbar i

Patna

Vishaka Patnam

Hyderaba d

Goa

Chennai

58

PortBlair

Trivandru m

850 (78) 500 (75) 200 (69) 850 (95) 500 (95) 200 (89)

1.02 0.30 0.83 0.36 1.88 1.38 0.84 0.60 2.44 0.23 1.28 0.81

1.22 0.66 0.71 0.13 2.02 1.55 0.79 0.30 2.41 0.38 1.43 0.99

1.33 0.60 0.93 0.24 1.90 1.34 1.09 0.77 2.51 0.35 1.31 0.62

1.40 0.74 0.88 0.18 2.19 1.80 1.04 0.47 2.47 0.32 1.82 1.35

1.34 0.55 0.89 0.12 1.83 1.31 1.11 0.68 2.58 0.56 1.21 0.38

1.51 0.86 1.00 0.11 2.22 1.78 1.08 0.49 2.57 0.41 1.65 1.25

1.38 0.55 0.93 0.03 1.81 1.20 1.15 0.74 2.66 0.60 1.18 0.14

1.44 0.83 0.99 0.00 2.14 1.67 1.09 0.40 2.59 0.51 1.67 1.15

59

Table4:RMSEandBiascomputedagainstIndianGPSsondeobservations,averaged for0000UTCzonalwind(m/s)analysesandpredictionsat850,500and200hPa levelsforJJAS2010

Station Srinagar
Level (obs count) 500 (69) 200 (67) 850 (118) 500 (113) 200 (111) 850 (84) 500 (82) 200 (76) 850 (71) 500 (62) 200 (53) 850 (81) 500 (76) 200 (75) 850 (64) 500 (62) 200 (61) 850 (68) 500 (66) 200 (64) 850 (68) 500 (68) 200 (65) Analysis UKMO NCMR 2.34 0.65 2.37 0.49 3.60 0.69 2.42 0.63 2.86 1.73 3.38 0.94 2.72 1.12 3.79 1.41 1.98 0.62 1.75 0.67 2.33 0.39 6.02 2.53 4.40 0.90 4.92 1.16 2.14 0.33 2.15 0.67 2.10 0.14 1.58 0.02 2.47 0.02 3.10 0.41 2.41 0.67 1.82 0.04 2.02 0.28 2.30 0.4 2.65 0.57 1.50 0.19 1.75 0.23 1.95 0.19 2.70 0.72 1.66 0.00 2.65 0.78 2.34 0.53 2.12 0.20 2.53 1.00 6.27 3.63 4.42 1.20 4.91 0.62 1.77 0.75 2.34 0.23 3.14 1.24 1.67 0.35 2.65 0.67 2.88 0.66 2.40 0.70 2.18 0.16 2.60 1.43 D1(24hr)FCST UKMO NCMR 3.87 0.39 4.18 0.20 3.66 2.03 3.15 0.70 4.26 2.28 4.06 1.51 3.10 0.28 4.13 1.02 4.29 2.87 3.59 1.35 4.34 2.58 6.82 3.15 4.81 1.02 5.25 1.66 2.99 0.63 3.26 0.69 5.09 0.59 2.65 0.61 3.63 0.43 4.86 0.67 3.52 0.65 3.93 1.23 5.07 0.47 3.48 0.66 5.72 0.78 3.68 0.33 3.40 1.02 4.14 0.02 4.46 1.51 3.76 1.05 4.71 0.42 4.16 1.23 3.98 0.72 4.74 2.01 5.77 3.31 5.61 1.23 5.64 0.81 3.31 0.70 4.51 0.51 5.86 1.84 3.21 0.37 4.61 0.74 6.92 2.35 4.42 1.09 4.37 0.01 6.35 2.27 D3(72hr)FCST UKMO NCMR 4.17 0.67 4.71 0.73 4.88 2.48 3.20 0.38 4.96 2.79 4.80 2.27 4.15 0.27 4.13 0.72 4.65 1.92 4.29 2.33 5.16 3.23 6.92 2.86 5.12 1.19 5.39 1.87 3.21 0.74 4.02 0.93 5.59 1.76 2.62 0.21 4.21 1.10 5.74 0.98 4.40 1.37 4.67 1.61 6.31 1.57 3.92 1.54 5.62 0.04 5.09 1.39 3.72 0.37 5.19 0.59 4.51 0.62 4.31 0.48 5.52 0.61 4.41 1.08 4.68 1.45 5.39 2.15 5.85 2.30 5.71 0.79 5.67 0.12 4.24 1.27 5.73 1.74 6.29 2.10 3.48 0.34 5.85 1.52 5.99 1.72 4.35 1.10 4.93 0.30 6.56 2.68 D5(120hr)FCST UKMO NCMR 4.12 0.90 6.25 1.06 5.8 2.14 4.12 0.89 5.59 3.47 4.61 1.80 4.96 0.94 5.70 1.87 4.43 0.80 4.54 2.42 5.80 3.61 6.81 2.34 5.26 1.30 6.06 2.28 3.85 0.16 4.71 0.77 5.59 1.38 3.07 0.21 4.91 1.35 5.90 1.03 4.90 1.81 5.38 1.62 6.47 0.76 4.04 1.35 7.44 0.69 5.85 2.13 4.20 0.07 5.60 0.35 5.24 1.16 5.20 1.02 6.67 0.44 5.59 0.64 5.36 2.38 5.67 2.20 6.63 1.75 6.21 1.13 6.06 0.66 5.44 2.03 7.01 2.06 6.49 0.88 4.10 0.86 7.18 2.64 6.07 0.05 5.90 1.98 6.75 1.71 6.42 1.81

Delhi

Mohanb ari

Patna

Visakha Patnam

Hyderab ad

Goa

Chennai

60

Port Blair

850 (79) 500 (76) 200 (77) 850 (96) 500 (89) 200 (84)

1.60 0.02 1.91 0.34 2.78 0.79 2.76 0.58 2.63 0.56 3.43 1.40

1.80 0.47 1.83 0.15 3.74 2.31 2.24 0.50 2.38 0.23 3.36 1.52

3.32 0.78 3.38 0.52 4.36 1.91 4.15 1.11 3.88 0.81 5.35 1.87

3.78 1.42 3.84 0.18 7.74 3.94 3.86 0.41 3.54 0.54 6.67 3.30

3.65 1.34 3.83 0.68 4.95 1.28 4.20 0.63 4.62 0.60 6.36 3.11

4.92 1.78 4.32 0.22 8.26 3.33 3.74 0.72 4.51 0.28 8.26 5.37

4.13 1.93 3.90 1.10 5.34 0.99 4.23 0.19 4.51 0.11 6.24 2.63

4.78 0.90 4.53 1.47 7.91 1.78 4.28 0.08 5.98 0.39 8.10 4.01

Trivandr um

61

GlobalModelForecastVerificationduring IndianSummerMonsoon2010: PrecipitationForecastsfromtheModels

ForIndiaasawhole,nearly78%oftheannualrainfallisproducedinthe summer monsoon season. However, the rainfall in the monsoon season over the homogenoussouthernpeninsularofIndiacontributesabout60%oftheannualmean, andasignificantamount(nearly40%oftheannual)alsooccursinthepostmonsoon seasonorthenortheastmonsoonrainyseason.Forannualaswellasmonsoonseason rainfall,thetwoprominenthighrainfallbeltsduetoorographiceffectsare:(i)offthe west coast of India and (ii) along northeast India and the foothills of the sub Himalayanranges.Thereisageneraldecreaseofrainfallfromeasttowestincentral India and along the Gangetic plains. The rainfall over the arid regions of west Rajasthan, Saurashtra, and Kutch is less than onethird of its magnitude over the Gangetic west Bengal in the east. The monsoon season features intraseasonal variationsinrainfallamountanddistribution.Thesearemainlydictatedbytheactive andweakcyclesinthemonsoonandtheBayofBengallowpressuresystemsthat moveinlandcausingheavyrainfalloverlandregions.

1MeanMonsoonRainfallduringJJAS2010

Themodelswithhighspatialresolutionareexpectedtoresolvethemesoscale processesinthestormsandimpactofhighresolutionorographytogivebetterrainfall prediction compared to the coarse resolution global models. In this section the performanceofthethreemodels(T382,T254andUKMO)formediumrangerainfall forecasting has been examined during monsoon (JJAS) 2010. For a detailed and quantitative rainfall forecast verification, the IMD's 0.5 daily rainfall analysis (RajeevanandBhate2008,Rajeevanetal2005)isused.Thisisthehighresolution

62

dailygriddedrainfalldatasetsuitableforthehighresolutionregionalanalysis.The dailyrainfalldatafromthefourmodelsisgriddedontotheobservedrainfallgrids overIndianlandregionsforthe122daysfrom1stJunethrough30thSeptember2010. Table2showsthecontingencytableforcategoricalforecastsofabinaryeventandthe followingstatisticsarecomputed. Thestatisticsarecomputedtakingintoaccount onlytherainydaysi.e.,dayswithrainfall>=0.5cmateachgridoverlandregions.

ThepanelsFigure36presentsobservedandforecastsofrainfall(cm/day)for JJASobtainedfromthethreemodels.Theobserveddistributionofrainfallindicates themaximumrainfallofupto2cm/dayalongthewestcoastofIndiasurroundedby rainfall in the range of 12 cm/day. Similar rainfall amounts in the range of 12 cm/daycanbeprominentlyseenoverpartsofNortheastIndia,Gangeticplainsanda largeregioncoveringWestBengalandOrissa.Overthewestcoastandpartsofnorth easternIndiathemodelforecasts(alldays)showmeanrainfallinexcessof2cm/day atmanylocationssurroundedbyrainfallintherangeon12cm/day.Theforecasts clearlyoverestimatetheobservedrainfalloverthesetworegions.Clearlytherainfall overtheGangeticplainsisoverestimatedinallthreemodelsparticularlyinDay5 forecasts.DuringthemonthofJune(Figure37)themonsoonisyettocompletely covertheIndiansubcontinent.Asseenintheobservations,alargepartofcentraland northernIndiaiscoveredwithverylittlerainfallinthemodelforecaststoo. Except fortheDay5forecasts ofUKMO,itcanbesaidthattheotherforecastsclosely matchtheobservednatureoftheadvanceofmonsoonduringJune2010.Duringthe month of July(Figure38) theregionis fullyunderthegripofmonsoonandthe modelscapturethisaspectverywellwithrainfallalloverIndia.However,allmodels overestimate(inDay5forecast)theobservedrainfalloverGangeticplains.TheT254 andT382modelsunderestimatetherainfallovernorthwestIndiaonalldays.Rainfall overestimationoverGangeticplainsisseeninthemonthofAugust(Figure39)as wellasinSeptember(Figure40).DuringeachofthemonthsUKMOmodelforecasts 63

underestimatetherainfalloverthepeninsula.Theobservedrainfallminimaintherain shadowregionofthepeninsulaisexaggeratedintheUKMOmodelforecasts.

2RainfallForecastVerification

Adetailedandquantitativerainfallforecastverificationispresentedinthis sectionusingtheIMD's0.5 dailyrainfalldata(RajeevanandBhate2008)forthe entire period of JJAS 2010. Table 5 shows the contingency table for categorical forecastsofabinaryeventandthefollowingstatisticsarecomputed.Thestatisticsare computedtakingintoaccountonlytherainydaysi.e.,dayswithrainfall>=0.5cmat eachgridpointoverlandregions.Therainfallforecastverificationisexpressedin termsofthreedifferentscoresdiscussedbelow. 2.1 Mean Error: The difference between the observed and forecast mean rainfall (Figure41)ispresentedtobringouttheareasofoverestimatedandunderestimated rainfalloverIndia. ModelsconsistentlyoverestimatetherainfallovertheGangetic plains. Rainfall over the dry regions of NW India is under predicted in all the forecasts.Rainfalloverthepeninsulaisunderpredictedandthisisprominentlyseen intheUKMOforecasts.

2.2Equitablethreatscore(Gilbertskillscore)

where

64

Thisisastandardskillscorethatisbeingusedbyvariousweatherservicesto evaluatetheirprecipitationforecasts.Itisfrequentlyusedtoassessskillofrainfall forecastsabovecertainpredefinedthresholdsofintensityofrain. ETStellsus how welldidtheforecast"yes"eventscorrespondtotheobserved"yes"events(accounting forhitsduetochance)?ETSrangesfrom1/3to1,0indicatesnoskilland1meaning perfectscore.ETSmeasuresthefractionofobservedand/orforecasteventsthatwere correctlypredicted,adjustedforhitsassociatedwithrandomchance(forexample,itis easiertocorrectlyforecastrainoccurrenceinawetclimatethaninadryclimate).Itis most suited for verification of rainfall in NWP models because its "equitability" allows scores to be compared more fairly across different regimes. This score is sensitivetohits.Becauseitpenalizesbothmissesandfalsealarmsinthesameway,it doesnotdistinguishthesourceofforecasterror.Figure42showstheETScomputed ontheforecastrainfallfromallmodels.Thegrayshadingintheplotsindicateno skill. Largepartsofpeninsulashowsnoskillandthisistrueinalltheforecasts. ForecastsoverthecentralIndiaincludingNWIndiashowsomeskillinpredictingthe rainyday.ETScomputationsfordifferentrainfallthresholdisshowninFigure43. Forlowerthresholds(0.0,0.1and0.6)thescoresarehighinallthreemodelsand there is not clear and consistent higher skill for any model. For higher rainfall amounts,thescoresarelowandUKMOconsistentlyshowsmarginallyhigherETS values than the ETS of T254 and T382 models. For higher rainfall thresholds (>9cm/day)theETSvaluesareverysmallandthenumberofoccurrencesarealso verylow.

2.3FalseAlarmRatio

65

FalseAlarmratio(FAR)isameasureof fractionofthepredicted"yes"eventsthat actuallydidnotoccur(i.e.,werefalsealarms). Thisscorerangesfrom 0to1anda scoreof0impliesperfectforecast.Thisscoreissensitivetofalsealarms,butignores misses.Itisverysensitivetotheclimatologicalfrequencyoftheevent.Figure44 shows the FAR computed for the forecast rainfall for all models. All the models indicatehigherforecastskillalongthewestcoast,northeasternstatesandalongthe foothillsofHimalayas.Allthemodelsshowverysimilarpatternsoverdryregions withhigherFARvaluesoverthenorthwesternregionandsoutheasterntip ofthe peninsula.

66

Table5.Contingencytableforcategoricalforecastsofabinaryevent.Herea, b,canddarethenumberofeventsobservedtooccurineachcategory. Event EventObserved Forecasts Yes No Total Yes a(hit) c(miss) a+c No b(falsealarm) d(correct rejection) b+d Total a+b a+d a+b+c+d=n

67

Figure36.ObservedandforecastmeanrainfallduringJJAS2010.

68

Figure37.ObservedandforecastmeanrainfallduringJune2010.

69

Figure38.ObservedandforecastmeanrainfallduringJuly2010.

70

Figure39.ObservedandforecastmeanrainfallduringAugust2010.

71

Figure40.ObservedandforecastmeanrainfallduringSeptember2010.

72

Figure41.MeanerrorintheforecastrainfallduringJJAS2010.

73

Figure42.EquitableThreatScoreforforecastofrainydayduringJJAS2010.

74

Figure 43. Equitable Threat Score for predicted rainfall exceeding different thresholds.

75

Figure44.FalseAlarmratioforforecastofrainydayduringJJAS2010

76

Conclusions
ForecastMeanCharacteristics:

The 850 hPa mean circulation suggests that model forecasts feature relatively deepermonsoontroughcomparedtotheinitialanalysis.(allforecastsofT254and T382 models and Day3 and Day5 forecasts of UKMO). The 200 hPa mean circulationsuggests thattheT254model forecasts featureweakeranddiffused structureof the200hParidge.The200hParidgeintheUKMOanalysis and forecastsiswelldevelopedandprominentduringallthemonths.

Theforecastsshowrelativelyhighermean(JJAS)temperaturesat850hPaover northwestIndiaandtheGangeticplainscomparedtoanalysis.Partsofpeninsular India,ArabianSeaandBayofBengalfeaturelowertemperaturesintherangeof 290292KintheT254andT382models.TheUKMOmodelinitialanalysisand forecastsfeaturehighertemperaturesoverthepeninsulaandtheneighboringseas byatleast3K.At200hPalevelonlyUKMOforecastsseemtocloselyagreewith theanalysis(andiscoolerthantheT382andT254models).BothT382andT254 modelsunderestimatethewarmtemperaturesovercontinent.

The850hPameanrelativehumidityinanalysisandforecastsbothindicatebroad patternofrainfallactivity.However,dryconditionsinJuneandSeptember(mainly overcentralandnorthwesternIndiaareprominentinT254andT382.At200hPa level, it is striking to note that the UKMO analysis as well as forecasts are excessivelydryoverlargepartsofIndianoceanandIndiaindicatingUKMOmodel analysisandforecastsfeaturewetter(drier)lower(upper)tropospherecomparedto T254andT382models.

To studytheoccurrenceofactiveandweakspellsofmonsoon,timelongitude sectionsareconstructedoverArabianSeaandoverPeninsula.WhiletheDay3 and Day5 forecasts agree with analysis, (i) the UKMO analysis show

77

comparativelystrongerzonalwinds.(ii) thewindstrengthisweakerduringthe activespellintheDay5forecasts.

ForecastErrors:

Theforecastsofallthreemodelsshowat850hPa,(i)westerlybiasovernorth ArabianSea,CentralIndiaextendingtoSoutheastAsia.(ii)tothesouthofthis eastwestregioneasterlybiasisfoundwhichimpliesnorthwardshiftofcirculation in forecasts) (iii) Central and eastern equatorial Indian Ocean features strong easterly bias, particularly in the UKMO model forecasts. The Gangetic plains showseasterlybiasintheUKMOandT254forecasts.

TheT382andT254modelsshowamarkedincreaseintheRMSE(U,V)fromDay 1toDay5forecastascomparedtoUKMO,especiallyoverthenorthwestand plainsofIndia.At850hPa,themagnitudeofRMSEinmeridionalcomponentof thewindisoftheorderof24m/sinDay1forecastinallthemodelsoverthe Indian and neighborhood region. All the models show a consistent increase in RMSEfromDay1toDay5forecastwiththemagnitudeofabout46m/s.

Systematicerrorsinthewindsat200hPalevelsuggeststrongwesterlybiasinthe eastern Indian Ocean in the Day3 and Day5 forecasts which is prominent particularlyintheT254andT382models.TheRMSEmagnitudevariesfrom24 m/sinUKMOand26m/sinDay1forecastofUKMOandT382/T254models overtheIndianregion.T382andT254modelshowaconsiderableincreasein RMSEfromDay1toDay5forecastespeciallyoverthesouthernpeninsulawhere theTropicalEasterlyJet(TEJ)isprominent.

At850hPaleveltheT254andT382modelsshowslightcoldbiasoverthenorth Arabian Sea (about 0.4) with strong warm bias over the Pakistan region (>1.2). The UKMO model forecasts show strong warm bias over the north ArabianSea(>1.4)andstrongcoldbiasovertheheatlowregion(<1.2).

78

ThemagnitudeofRMSEisoftheorderof11.5 KinDay1forecastoverthe northwestIndiaforT254andT382whereasthemagnitudeisslightlysmallerin UKMO.TheT254andT382modelsshowconsistentlyhigherRMSEsoverthe northwestpartsofIndiaascomparedtotheUKMOforecasts.

At200hPa,allforecastsofT382(andDay1andDay3forecastsofT254)show largeareacoveringBayofBengal(ArabianSea)havingwarm(cold)bias.The UKMOforecastsfeaturewarmbiasovercontinentsandcoldbiasovertheIndian ocean.ThemagnitudeofRMSEsinDay5forecastsofallthemodelforecastsis about0.52K,overtheIndianregion,withlargererrorsoverthenorthernparts.

Forecastverificationagainsttheupperairobservations(bothradiosondeandGPS sonde)clearlyindicatethattheUKMOforecastsfeaturesmallerRMSEcompared totheNCMRWFforecasts.

RainfallForecast.

Theobserveddistributionofrainfallindicatesthemaximumrainfallofupto2 cm/dayalongthewestcoastofIndiasurroundedbyrainfallintherangeof12 cm/day.Similarrainfallamountsintherangeof12cm/daycanbeprominently seenoverpartsofNortheastIndia,Gangeticplainsandalargeregioncovering WestBengalandOrissa.OverthewestcoastandpartsofnortheasternIndiathe model forecasts (all days) show mean rainfall in excess of 2 cm/day at many locationssurroundedbyrainfallintherangeon12cm/day.Theforecastsclearly overestimatetheobservedrainfalloverthesetworegions.Clearlytherainfallover the Gangeticplainsisoverestimatedinallthreemodels particularlyinDay5 forecasts.

WhilethedryconditionsofJunearewellcapturedinallforecastsofallmodels, thewetconditions(particularlyGangeticplains)ofJuly,AugustandSeptember areoverestimatedinalltheforecasts.

79

OverthepeninsulatheUKMOmodelforecastsunderestimatetherainfallinall months.Theobservedrainfallminimaintherainshadowregionofthepeninsulais exaggeratedintheUKMOmodelforecasts.

80

Acknowledgments:TheauthorsaregratefultoU.K.MetOffice,UKformakingthe model forecast available to NCMRWF in real time. Thanks are due to the India MeteorologicalDepartment(IMD)forprovidingthehighresolutionrainfallanalysis data used in this study. The authors express gratitude towards colleagues at NCMRWF for fruitful discussions and feedbacks during the preparation of this manuscript.

81

References
FindlaterJ.1969:AmajorlowlevelaircurrentneartheIndianoceanduringnorthern summer;Quart.J.Roy.Meteor.Soc.,95,362380. Rajeevan,MandJyotiBhate2008:Ahighresolutiondailygriddedrainfalldataset (19712005)formesoscalemeteorologicalstudies.NCC(9)ResearchReport.Indian MeteorologicalDepartment,Pune411005 Rajeevan, M., J. Bhate, J. D. Kale, and B. Lal (2005), Development of a high resolution daily gridded rainfall data for the Indian region. (Met Monograph Climatology,22/2005,26pp,availablefromIndiaMeteorologicalDepartment,Pune, India) WWRP200912009:Recommendationsfromtheverificationandintercomparisonof QPFsandPQPFsfromtheoperationalNWPmodels.WMOTDNo.1485,p37

82

Appendix

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107