Performance of Global Forecast System NCMRWF/IMD (INDIA) Presentation for Annual performance evaluation of NCEP production suite at NCEP, Maryland, USA during 6-8 December 2011

Care-takers NCMRWF Data/monitoring: Munmun Das Gupta, Indira Rani Analysis : V S Prasad Model/post : Saji Mohandas Verification : Gopal Iyengar GEFS : E N Rajagopal IMD Data/monitoring: S D Kotal Analysis/Model/Verfication : V R Durai Co-ordinator : S.K.Roy Bhowmik

Numerical Weather Prediction System of NCMRWF Global Observations Data Reception Global Data Assimilation Forecast Models Users SURFACE from land stations Upper Air RSRW/ PIBAL SHIP BUOY Aircraft Satellite High Resolution Satellite Obsn Internet (FTP) RTH, IMD GTS NCMRWF OBSERVATION PROCESSING NKN 24x7 NESDISEUMETSAT 45mbps proposed dedicated link Observation quality checks & monitoring 4 times a day for 00,06,12,18 UTC Global Analysis (GSI) Initial state Global Forecast Model ( 9hr Fcst – first guess ) Global Model T574L64 10day FCST Meso-scale Data Assimilation & Model Statistical Interpolation Model (location specific FCST) once in a day for 00 UTC Visuali- sation IMD INCOIS Other sectors NKN ISRO (MT)

ModelVersionHorizontal Resolution Forecast Length Performance GFS T382L64 GFS version ~35km168 Hrs (3hr data cutoff) 4 min. for 24 hr forecast (IBM- P6 16 nodes) GFS T574L64 GFS version ~23km240 Hrs (5hr data cutoff) 9 min. for 24 hr forecast (IBM- P6 16 nodes) GEFS T190L28 Latest version 20 members ~70km240 Hrs (Not operational) 6 min. For 24 hr forecast (IBM-P6 8 nodes) GFS Models (NCMRWF) – Current status High Performance Computing Systems HPCConnectivit y No of Processors available Per node memory Processor speed IBM Power 6 Infini Band38x32 (NCMRWF) 24x32 (IMD) 4x32 GB4.7 Gflops

Recent developments in NCMRWF GFS system Implementation of the T382L64 GFS from May 2010 (latest versions of upgraded model and GSI) Assimilation of additional data in T382L64 GFS The Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E) winds Rainfall rates (TRMM, SSMI) NOAA19 radiances Atmospheric Infrared Sounder (AIRS) radiances GPSRO (COSMIC) Implementation of the T574L64 GFS from mid-November, 2010 (July 2010 Version) Implementation of latest NCEP version of T574L64 (from June, 2011)

The T382L64 GFS was implemented in May The T574L64 GFS was first implemented in November T382L64 performance was evaluated during Monsoon 2010 and was found to marginally better than the T254L64 GFS. T254L64 stopped. T574L64 performance was evaluated during November- December 2010 and was found to better than the T382L64 GFS. T382L64 model was run parallel to T574L64 for Monsoon 2011 The latest version of the NCEP T574L64 GFS was implemented in May 2011 and found to be better than T382L64 system T382L64 run stopped from November 2011 End-to-end T574L64 GFS system was transferred to IMD on 15 November, 2011 Time line - NCMRWF GFS

PhysicsT382L64T574L64 Surface FluxesMonin-Obukhov similarity Turbulent DiffusionNon-local Closure scheme (Hong and Pan (1996)Non-local Closure scheme (Lock et al., 2000) SW RadiationBased on Hou et al –no aeroslos – invoked hourlyRapid Radiative Transfer Model (RRTM2) (Mlawer et al. 1997; Mlawer and Clough, 1998)- aerosols included– invoked hourly LW RadiationRapid Radiative Transfer Model (RRTM) (Mlawer et al. 1997). –no aerosols- invoked 3 hourly Rapid Radiative Transfer Model (RRTM1) (Mlawer and Clough 1997;1998). –aerosols included- invoked hourly Deep ConvectionSAS convection (Pan and Wu (1994)SAS convection (Han and Pan, 2006) Shallow ConvectionShallow convection Following Tiedtke (1983)Mass flux scheme (Han and Pan, 2010) Large Scale Condensation Large Scale Precipitation (Zhao and Carr,1997; Sundqvist et al., 1989) Cloud GenerationBased on Xu and Randall (1996) Rainfall EvaporationKessler (1969) Land Surface ProcessesNOAH LSM with 4 soil levels for temperature & moisture (Ek et al., 2003) Air-Sea InteractionRoughness length by Charnock (1955)Observed SST,Thermal roughness over the ocean is based on Zeng et al., (1998).3-layer Thermodynamic Sea-ice model (Winton, 2000) Roughness length by Charnock (1955), Observed SST, Thermal roughness over the ocean is based on Zeng et al., (1998). 3-layer Thermodynamic Sea-ice model (Winton, 2000) Gravity Wave Drag & mountain blocking Based on Alpert et al. (1988)Lott and Miller (1997), Kim and Arakawa (1995), Alpert et al., (1996) Vertical AdvectionExplicitFlux-Limited Positive-Definite Scheme (Yang et al., 2009) Physical Parameterization schemes in T382L64 and T574L64

New observations assimilatedImprovements in Data Assimilation system Inclusion of METOP IASI (Infrared Atmospheric Sounding Interferometer) data Use of variational qc Reduction of number of AIRS (Atmospheric Infrared Sounder) water vapor channels used Addition of background error covariance input file Assimilating tropical storm pseudo sea-level pressure observations, Flow dependent reweighting of background error variances NOAA-19 HIRS/4 (High Resolution Infrared Radiation Sounder) and AMSU-a (Advanced Microwave Sounding Unit) brightness temperature, Use of new version and coefficients for community radiative transfer model (CRTM ) NOAA-18 SBUV/2, (Solar Backscatter Ultraviolet Spectral Radiometer) Ozone, EUMETSAT-9 atmospheric motion vectors. Improved Tropical Cyclone Relocation Using uniform thinning mesh for brightness temperature data. Change in land/snow/ice skin temperature variance Improving assimilation of GPS radial occultation data. RE-tuned observation errors. ASCAT (Advanced Scatterometer) winds included Korean AMDAR data and more number of Aircraft Reports European Wind profiler data Differences of the T574L64 GSI Data Assimilation system compared to T382L64

Observation categoryName of Observation. SurfaceLand surface, Mobile, Ship, Buoy (SYNOPs) Upper airTEMP (land and marine), PILOT (land and marine), Dropsonde, Wind profiler AircraftAIREP, AMDAR, TAMDAR, ACARS Atmospheric Motion Vectors from Geo-Stationary Satellites AMV from Meteosat-7, Meteosat-9, GOES-11, GOES-13, MTSAT-1R, MODIS (TERRA and AQUA), Scatterometer windsASCAT winds from METOP-A satellite, NESDIS / POES ATOVS Sounding radiance data 1bamua, 1bamub, 1bmhs,1bhirs3, 1bhirs4 Satellite derived Ozone dataNESDIS/POES, METOP-2 and AURA orbital ozone data Precipitation RatesNASA/TRMM (Tropical Rainfall Measuring Mission) and SSM/I precip. rates Bending angles from GPSROAtmospheric profiles from radio occultation data using GPS satellites NASA/AQUA AIRS & METOP/ IASI brightness temperature data IASI,AIRS,AMSR-E brightness temperatures Types of observations Assimilated in GFS

Observation Type JuneJulyAugustSeptember SYNOP RS/RW PILOT AWS ARG BUOY Count of different types of observations over Indian Region (received at NCMRWF at 00 UTC from 1 to 25 of months June, July, August, and September 2011)

Data Reception: NCMRWF vs ECMWF (S-W Monsoon, 2011) (Average number of observations received in 24 hours ) RED COLOUR INDICATES LESS DATA ; BLUE COLOUR INDICATES COMPARABLE DATA JuneJulyAugust NCMRWFECMWFNCMRWFECMWFNCMRWFECMWF SYNOP/SHIP Pressure 46,11878,39056,73478,48157,60778,574 BUOY (Drifter)10,71513,95313,88213,87913,75213,531 TEMP 500 hPa Geopotential 1,0881,2861,2711,2861,3201,336 TEMP/PILOT 300 hPa Wind 1,0981,4341,3001,4291,3491,490 AIRCRAFT winds ( hPa) 58,4121,04,98772,8421,02,56272,6851,03,059 AMV winds ( hPa) 2,00,28310,06,2802,55,6909,78,2692,43,2829,45,691 AMV winds ( hPa) 1,28,7868,53,1931,54,7248,91,2041,52,9058,21,724

Vertical Profile of T574L64 (Dotted Line) and T382L64 (Bold Line) Analyses (Black) and First Guess (Red) Vector Wind Fits (Bias and RMSE) to RAOBS over Global for JJAS, The Right Panel graph gives the observation data counts over the region used for the comparison.

Vertical Profile of T574L64 (Dotted Line) and T382L64 (Bold Line) Analyses (Black) and First Guess (Red) Vector Wind Fits (Bias and RMSE) to RAOBS over Tropics for JJAS, The Right Panel graph gives the observation data counts over the region used for the comparison.

Vertical Profile of T574L64 (Dotted Line) and T382L64 (Bold Line) Analyses (Black) and First Guess (Red) Moisture Fits (Bias and RMSE) to RAOBS over Global for JJAS, The Right Panel graph gives the observation data counts over the region used for the comparison.

Vertical Profile of T574L64 (Dotted Line) and T382L64 (Bold Line) Analyses (Black) and First Guess (Red) Moisture Fits (Bias and RMSE) to RAOBS over Tropics for JJAS, The Right Panel graph gives the observation data counts over the region used for the comparison.

Vertical Profile of T574L64 (Dotted Line) and T382L64 (Bold Line) Analyses (Black) and First Guess (Red) Temperature Fits (Bias and RMSE) to RAOBS over Global for JJAS, The Right Panel graph gives the observation data counts over the region used for the comparison.

Vertical Profile of T574L64 (Dotted Line) and T382L64 (Bold Line) Analyses (Black) and First Guess (Red) Temperature Fits (Bias and RMSE) to RAOBS over Tropics for JJAS, The Right Panel graph gives the observation data counts over the region used for the comparison.

Global Circulation Features

NCEP NCMRWF Day 05 Forecast Errors 850 hPa Zonal Wind JJA 2011

Regional Circulation Features

T382 T574 ANA D03 ERR

ANA D05 ERR T382 T574

ANA D03 ERR T382 T574

D05 ERR ANA T382 T574

ANA D03 ERR T382 T574

T382 T574 ANA D05 ERR

T382 T574

T382 T574

T382 T574

T382 T574

VERIFICATION AGAINST ITS OWN ANALYSIS Models: T574, T382 and UKMO Parameters : Zonal & Meridional Wind, Geo-potential Height, Temperature, Relative Humidity  forecast and analysis fields used are valid for 00UTC and the forecasts are based on initial condition valid for 00UTC.  computed the scores using the data at 1 degree resolution from all the models.

T382 T574 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

T574 T382 UKMO

Time Series of RMSE of D01, D03 & D05 Forecasts of Zonal Wind D01D03D hPa 700hPa 500hPa 200hPa

Time Series of RMSE of D01, D03 & D05 Forecasts of Meridional Wind D01 D03D05 200hPa 500hPa 700hPa 850hPa

Time Series of RMSE of D01, D03 & D05 Forecasts of Temperature D01D03D05 200hPa 500hPa 700hPa 850hPa

Time Series of RMSE of D01, D03 & D05 Forecasts of Geop Height D01 D03 D05 200hPa 500hPa 700hPa 850hPa

Time Series of RMSE of D01, D03 & D05 Forecasts of RH D01 D03D05 200hPa 500hPa 700hPa 850hPa

200 hPa ModelDay1Day3Day5Day1Day3Day5 T T UKMO RMSE against own analysis 850 hPa Zonal Wind

850 hPa200 hPa ModelDay1Day3Day5Day1Day3Day5 T T UKMO RMSE against own analysis 850 hPa Meridonal Wind

Factors and methods used In standardized verificatlon of NWP products Verification agalnst analysis Area Northem hemisphere extratropics (90°N ‑ 20°N )(all inclusive) Tropics (20°N ‑ 20°S)(all inclusive) Southem hemisphere extratropics (20°S ‑ 90°S)(all inclusive ) Grid Verifying analysis is the centre's on a latitude ‑ longitude grid 2.5° x 2.5°; origin (0°,0°) Variables MSL pressure, geopotential height, temperature, winds Levels Extratropics: MSL, 500 hPa, 250 hPa Tropics: 850 hPa, 250 hPa Time 24h, 48h, 72h, 96h,120h,144h,168h,192h, 216h, 240h... Scores Mean error, root ‑ mean ‑ square error (rmse), anomaly correlation, S1 skill score, root ‑ mean ‑ square vector wind error (rmse V )

Verification against observations The seven networks used in verification against radiosondes consist of radiosondes stations Iying within the following geographical area: North America25°N ‑ 60°N 50°W ‑ 145°W Europe/North Africa25°N ‑ 70°N 10°W ‑ 28°E Asia 25°N ‑ 65°N 60°E ‑ 145°E Australia/New Zealand10°S ‑ 55°S 90°E ‑ 180°E Tropics20°S ‑ 20°N all longitudes N. Hemisphere Extratropics20°N ‑ 90°N all longitudes S.Hemisphere Extratropics 20°S ‑ 90°Sall longitudes

The anomaly correlation values are comparatively higher in the T574 GFS with a gain of 1 day in the skill of the forecasts. In the lower panel the line plot depicts the difference of the forecasts of Geopotential Height of the T574 GFS from the T382 GFS. The difference values outside the histograms are statistically significant at 95% level of confidence. Anomaly correlation of 10 day forecasts of 500 hPa Geopotential Height over the Northern Hemisphere from the T382 (black line) and T574 (red line) GFS

The RMSE values are comparatively lower in the T574 GFS with a gain of 1 day in the skill of the forecasts. In the lower panel the line plot depicts the difference of the forecasts of Zonal Wind of the T574 GFS from the T382 GFS. The difference values outside the histograms are statistically significant at 95% level of confidence. RMSE of 10 day forecasts of 850 hPa Zonal Wind over the Regional Specialized Meteorological Centre (RSMC) region from the T382 (black line) and T574 (red line) GFS

Verification of Day Forecast against Observations over Tropics Root Mean Square Error (RMSE) 850 hPa winds in m/s JUNE 2011 ModelDay 1Day 2Day 3Day 4Day 5 ECMWF UKMO NCEP NCMRWF T

T80 T254T382 T574 Verification of Day 03 Forecasts against Radiosondes over India ( ) Root Mean Square Error (RMSE) of 850 hPa winds in m/s

Monsoon Depressions

Track Errors

RAINFALL FORECAST VERIFICATION DURING MONSOON 2011:T382,T574 & UKMO A detailed and quantitative rainfall forecast verification has been made using the IMD's 0.5° daily rainfall data for the entire period of JJAS 2011.

Model Day-1Day-3Day-5 T T UKMO UKMO T382 T574

Seasonal (a), Monthly (b) and weekly (c) rainfall (mm) predicted by T574L64 model for Monsoon-2011 against observed and long period average (climatology). Weekly rainfall is accumulated 7-day forecast from single initial conditions of every week.

Forecasts of rain meeting or exceeding specified thresholds For binary (yes/no) events, an event ("yes") is defined by rainfall greater than or equal to the specified threshold; otherwise it is a non-event ("no"). The joint distribution of observed and forecasts events and non-events is shown by the categorical contingency table. hitsfalse alarms missescorrect rejections OBSERVED YESNO YES NO FORECAST FORECAST YES FORECAST NO OBSERVED YES OBSERVED NO

Bias The bias score is a measure of the agreement between the forecast frequency of "yes" events and the observed frequency of "yes" events. It is given by the ratio of the frequency of forecast events to the frequency of observed events. The score values range from 0 to infinity and the score of 1 implies a perfect forecast. >1 prediction <1 Underprediction

Bias score for the predicted rainfall during JJAS 2011

ETS ETS measures the fraction of observed and/or forecast events that were correctly predicted, adjusted for hits associated with random chance (for example, it is easier to correctly forecast rain occurrence in a wet climate than in a dry climate). ETS tells us how well did the forecast "yes" events correspond to the observed "yes" events (accounting for hits due to chance)? ETS ranges from -1/3 to 1, 0 indicates no skill and 1 meaning perfect score.

ETS

ETS for the predicted rainfall during JJAS 2011

IMD GFS Verification - Areas 1)ALL-INDIA (Lon: 68 E – 98E, Lat: 9N – 37N) 2)Central India (Lon: 75E – 80E, Lat: 19 – 24N) 3)East India (Lon: 83E -88E, Lat: 20N -25N) 4)North East India (Lon: 90E – 95E, Lat: 24N -29N) 5)North West India (Lon: 75E – 80E, Lat: 25N -30N) 6)South Peninsula India (Lon: 76E - 81E, Lat: 12N- 17N) 7)West Coast of India (Lon: 70E - 75E, Lat: 13N - 18N)

(Lon: 68 E – 98E, Lat: 9N – 37N) ALL India : Weekly Cumulative Rainfall

(Lon: 90E – 95E, Lat: 24N -29N) (Lon: 75E – 80E, Lat: 19 – 24N)

(Lon: 83E -88E, Lat: 20N -25N) (Lon: 75E – 80E, Lat: 25N -30N)

(Lon: 70E - 75E, Lat: 13N - 18N) (Lon: 76E - 81E, Lat: 12N- 17N)

CC :7 Day Cumulative Rainfall of GFS T382 &T574 vs. Observation

(Lon: 70E - 75E, Lat: 13N - 18N)(Lon: 76E - 81E, Lat: 12N- 17N) (Lon: 83E -88E, Lat: 20N -25N)((Lon: 75E – 80E, Lat: 19 – 24N) (Lon: 90E – 95E, Lat: 24N -29N) (Lon: 75E – 80E, Lat: 25N -30N)

GFS T574 : hr F/c shows a FALSE ALARM of Cyclonic Storm over Arabian sea on 6 June 2011 Analysis of 6 June 2011

(Lon: 90E – 95E, Lat: 24N -29N) (Lon: 75E – 80E, Lat: 25N -30N) Tmax (Top) and Tmin (bottom) over North-East India Both Tmax and Tmin mostly under predicts in all 4 months i.e. from 1June to 30 Sep 2011 GFS T574: Daily Error in Maximum and Minimum Temperature over North-East India

GFS T574: Seasonal Mean Error in Maximum and Minimum Temperature over different Homogeneous regions of India Mean Absolute Error (MAE) of Tmax (top) and Tmin (bottom) For 1 June to 30 September,2011

Conclusions The vertical profile of T574L64 analyses and first guesses fits to radio-sonde observation for JJAS 2011 shows improvement over T382L64 analyses. The RMSE values of fields of T574L64 forecasts against analyses and observations show improvements over T382L64 forecasts  Equitable Threat Square (ETS) computed for different rainfall thresholds shows that UKMO has higher skill score as compared with T382 and T574 for rainfall threshold >1.0 cm/day. For rainfall intensity of 0.01 cm/day all three models feature high ETS (>0.6) for all days forecast. T574 shows better skill score then T382 for all the rainfall intensities for all days.  The impact of more satellite data incorporated in T574L64, especially the AMVs over the tropics, is more evident in T574L64 analysis and forecasts when compared to the T382L64 system.

Future data plans (NCMRWF) VAD winds from Indian Doppler Weather Radar. Oscat winds (Oceansat-2 scatterometer) INSAT / Kalpana AMV Precipitation rates from MADRAS-MT GEOS Sounder data Radiances from INSAT-3D and MT Preparing plans for Indian Doppler Weather Radar

NCMRWF wish list Future NCEP upgrades in dynamics/physics? Higher resolutions? (T764L91? T1148L91/ Semi_Lagrangian?) Diversification ? (GEFS, Hybrid GSI-EnKF VA system) More diagnostics and verifications? Sensitivity studies and physics improvements? Participation in National/international compaigns/experiments? (MJOWG, MT)

IMD plans GFS T 574 EPS T 382 MME based on EPS Thrust Area: Probabilistic Forecast of high impact weather in short to medium range time scale

Hurricane WRF Model The HWRF model has been implemented at the India Meteorological Department (IMD) following the Implementation Agreement (IA) between India’s Ministry of Earth Sciences (MoES) and USA’s National Oceanic and Atmospheric Administration (NOAA) with an objective to provide improved tropical cyclone prediction capability for the Bay of Bengal and Arabian Sea regions. Under the program Dr Vijay Kumar and Dr Zhan Zhang, EMC, NCEP, USA were on deputation to IMD, New Delhi in July 2011 for technology transfer of HWRF model system and provide training on initial operating capability of HWRF model. The basic version of the model HWRFV(3.2+) which was operational at EMC, NCEP was ported on IBM P-6/575 machine, IMD with nested domain of 27 km and 9 km horizontal resolution and 42 vertical levels with outer domain covering the area of 80 0 x80 0 for NIO and inner domain 6 0 x6 0 with centre of the system adjusted to the centre of the observed cyclonic storm. HWRF model successfully tested for two Bay of Bengal TC cases JAL (4-8 Nov 2010), GIRI (21-22 Oct 2010) with vortex initialization and 6 hourly cyclic mode using the NCEP GFS data provided EMC team and also tested with IMD GFS spectral fields. The Atmospheric HWRF model was made operational (Experimental) to run real-time during the cyclone season The Ocean Model and Coupler is to be implement for Indian Ocean region (regional MOM) in collaboration with EMC, NCEP and Indian National Centre for Ocean Information Services (INCOIS), Ministry of Earth Sciences, Hyderabad, India by April Testing of the atmospheric HWRF model for the last 5 years Cyclonic Storm formed over Arabian Sea and Bay of Bengal for 6 to 8 cases with vortex initialization and 6 hourly cycling of forecast runs for each case with total 70 to 80 runs using the initial and boundary from NCEP GFS spectral fields are expected to be completed by the end of January 2012 and a joint report will be prepared by the end of February 2012.

THANKS