Météo-France status report Hervé Roquet (DP/CMS) & Bruno Lacroix (DPrévi/COMPAS)

1. 17th North America / Europe Data Exchange Meeting Météo-France status report Hervé Roquet (DP/CMS) & Bruno Lacroix (DPrévi/COMPAS) • Models • Last modifications • Impact of Cut-off time on forecast • Use of data • Meteosat data (geowinds) • AMSU-A data • HIRS radiances added to AMSU-A • AMSU-B • AIRS • Plans Montréal 26/5/2004

2. Supercomputer configuration • FUJITSU VPP5000: • Second half-year 2003 a new configuration with 124 processors (instead of 31) = ex ECMWF computer. • 2 separate machines: • 60 processors for operations • 64 processors for research

3. Last modifications : ARPEGE model • June 2003 • increase of average model resolution TL199 to TL358 • reduction of stretching factor 3.5 to 2.4 • reduction of minimisation resolution TL161 to TL149 • December 2003 • Use of HIRS raw radiances in 4DVAR • Tuning of AMSU-A • + weekly run up to 120H based on 00UTC long cut-off (MFSTEP) • January 2004 • New solver for 4DVAR minimization • reduction of iteration numbers (45->40, 20->15) • March 2004 New very short cut-off (1H) • May 2004 • New radiation scheme (FMR15)

4. Models configuration • Global model up to 102H at 00UTC, • (72H at 06 and 12UTC, 60H at 18UTC) • ARPEGE global spectral model TL358 C2.4 L41 • 41 levels, up to 1hPa, resolution 23km to 133km • Linear grid with T240 C2.4 orography • 3processors for ARPEGE forecast (20’ for 24H forecast) 4DVAR assimilation : • 2 loops of minimization T107 C1 L41(40 it.), T149 (15 it.) Last loop with simplified physics • 16 processors for assimilation (30’ between cut-off and P0) data used: • SYNOP, SHIP, BUOY, AIREP, AMDAR, ACARS, TEMP, PILOT • SATOB GOES 9, 10, 12 + BUFR Meteosat 5, 7 • HIRS NOAA16 & 17, AMSU-A NOAA15 & 16 • SST 0.5 degree from NCEP/NESDIS + SSM/I sea ice mask

5. Previous configuration T298 C3.5 from 19 to 233km Since June 2003, T358 C2.4 from 23 to 133 km

6. Analysis Analysis Guess Guess Guess long cut-off 18UTC long cut-off 00UTC Analysis Analysis Analysis Short cut-off 18UTC Very Short cut-off 00UTC Short cut-off 00UTC Forecast 60H Forecast 36H Forecast 102H • 5 runs a day and 9 analyses on 6 hour windows [H-3h ; H+3h[ • 4 short cut-off analyses to run short range forecasts • 1h50 at 00UTC (P102H) and 12UTC (P72H) • 3h at 06UTC (P72H) and 18UTC (P60H) • 4 long cut-off analyses to produce first guess (7-8h) • 1 very short cut-off 1H at 00UTC (P36H) ARPEGE-Métropole, very short cut-off ( 1H at 00UTC )

9. = = Impact of 1H cut-off on the forecasts • experiment : • 46 forecast runs at 00h up to 36 H • with a 1-hour cut-off • and short cut-off first guess • compared to 1H50 operationnal run • geopotential, verified against radiosondes data •  No significant degradation over Europe

11. Soundings available for 1H50 and 8H cut-off period 1998-2004

12. Number of available soundings since 1993

13. = = Impact of 8H cut-off on the forecasts • experiment : • 23 forecast runs at 00h • (once a week from 1/12/2004) • with a 8-hour cut-off • compared to 1H50 operationnal run • (same assimilation) • geopotential, verified against radiosondes data •  Significant improvement over Europe

14. The improvement can be seen almost every day. It is mainly given by the late ATOVS data  Importance of the EARS data

16. Models configuration (follow up) • Regional model up to J+2 06UTC (at 00, 06, 12 and 12UTC) • ALADIN spectral limited area model • 9.5 km resolution on 2740kmx2740km domain. • No data assimilation: it is a dynamical adaptation model. • Many coupling files • Tropical model 72H range at 00 and 12 UTC • ARPEGE uniform model (TL358 C1 L41) ~55km • 4D-VAR analysis configuration TL107 L41 (40/15 it.) • Use of simple bogussing and more SATOB data • Seasonal Forecast 4 months range • ARPEGE T63 C1 L31 configuration ARPEGE-Climat • 9 runs a month from ECMWF analysis + SST forcing • Experimental Short Range Ensemble Prediction System 60H range • 10 runs ARPEGE TL358 C2.4 L41 (control = oper) • Based on singular vector perturbation

17. Area for verification Short range ensemble Forecast (ARPEGE) • Targetting over Atlantic Ocean and Western Europe • Optimization time window : • 0-12h • Total Energy norm • (initial and final) • 16 first SVs • No physics • SV computation with a T63 regular truncation

18. Analyse Controle Vignettes ARPEGE CEP Tropiques AVARC

19. Brier Skill Score on wind speed (ref. : ECMWF EPS)Tested over a sample of 85 cases (Western Europe)

20. Rank diagram – T199 - Z500 – 48h Comparison T199 C3.5 L 31 / T 358 C2.4 L41 Rank diagram – T199 – Z500 – 24h Rank diagram – T358 – Z500– 24h Rank diagram – T358 - Z500 – 48h

21. Telecom and data received (files) • links relevant to US/Europe data exchange: • Daily volume of satellite data files received from Exeter : • HIRS, AMSU-A, AMSU-B from NOAA15, 16 and 17: total 350Mb • SSM/I from DMSP F13 and F15: total 130 Mb • Seawind (240 Mbytes) from Quikscat • AIRS+AMSU-A (250 Mbytes) from Aqua • Lannion to suitland(CIR 128 kbit/s, MIR 256 kbit/s): • IODC data from Meteosat 5 all channels, fullresolution every 30 minutes, and AVHRR data received at Lannion • Suitland to Lannion(CIR 128 kbit/s, MIR 256 kbit/s): • GOES-W data for FSDS of Eumetsat

22. Use of BUFR winds produced by EUMETSAT with a quality index and disseminated every 90 minutes compared to Use of previously operational SATOB winds produced every 6 hours Assimilation of Meteosat data(Chistophe Payan)

23. Experiments with the uniform ARPEGE configuration, over the period 23 Dec 2002 - 5 Jan 2003 SATOB experiment, the operational use Blacklist north of the 30° North on land BUFR experiment with conditional use QI>0.85 (0.90 for HWW in tropics area) Satob Blacklist plus : - on land for p > 700hPa - in extra-tropics for 350 < p < 800 hPa (data too biased) Assimilation of Meteosat data

24. Meteosat 5&7 observation fit to first guess and analysisarea=50N/50S/113E/50W Used U component • BUFR versus SATOB • more data used • rms and bias reduction Used V component

25. Meteosat BUFR winds assimilated since december 2003 Use of BUFR winds of the other satellites on going Structure of the observation errors as a function of the quality indicator and spatial correlation Bias correction BUFR winds : present time, near/next future

26. Operational use of AMSUA raw radiances Raw radiances instead of preprocessed radiances: 22 October 2002 (+ European & American profilers) NOAA17 on top of NOAA15 & NOAA16 : 17 December 2002 Research on locally received data (EARS) Operational suite with HIRS data (+ revision of rain detection for AMSUA) : December 2003 Pre-operational suite with AMSUB data : Summer 2004 Status of ATOVS data at Météo-France

27. Ts in the control variable T extrapolation above the model top (1 hPa) up to 0.1 hPa by regression 250 km thinning Conditions for use  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 3<scan position<28         Open sea (Ts 271.45 K)         Sea ice (Ts < 271.45 K)       Land orog<500m/1500 m for channels 5/6         Clear |ob-fg| ch 4 1.5 K         0.7 K Cloudy |ob-fg| ch 4 > 1.5 K |lat|>30 for channel 8      0.7 K Assimilation of AMSUA raw radiances (Elisabeth GERARD and Florence Rabier) or CLWP(ch1; ch2) > 0.1 mm

28. Assimilation of AMSUA raw radiances Northern Hemisphere Time series of rms errors and biases 24 hour forecast 200 hPa geopotential scores over 1 month 22 Aug - 22 Sep 2002 rms difference Southern Hemisphere scores computed wrt own analysis Preprocessed radiances Raw radiances rms difference

29. On top of AMSUA data over 3 weeks (23 Dec 2002 – 12 Jan 2003) 250 km thinning (as for AMSUA) Conditions for use  1 2 3 4 5 6 7 8 10 11 12 13 14 15 3<scan position<54         Sea (Ts > 271.45 K)         Land (orog<1500 m)  x(lat)<(ob-fg) ch 8<y(lat)         (ob-fg) ch 11/12 > -3 K   Experiments with HIRS radiances(Elisabeth Gérard and Delphine Lacroix) water vapour channels

30. Forecast scores (rms & bias) over Europewith HIRS & without HIRS 24 hour 48 hour 72 hour forecast range  geopotential temperature wind rel. humidity

31. % Global Sea Land Globe -6.1 -7.4 -0.1 N. Hem -3.0 -4.3 0.8 Tropics -8.0 -9.6 -0.6 S. Hem -4.7 -3.2 0.5 % Global Sea Land Globe -3.9 -5.1 -0.3 N. Hem 1.1 1.9 -0.1 Tropics -7.1 -8.7 -0.7 S. Hem -3.3 -3.7 0.7 TCWV time series WINTER SUMMER

32. Time series of rms errors and biases 48 hour forecast - 500 hPa geopotential 2 weeks - 26 Dec 2002 – 10 Jan 2003 Scores (winter) Northern Hemisphere scores computed wrt own analysis without HIRS with HIRS Southern Hemisphere Tropics

33. Test suite with HIRS and Meteosat BUFR44 runs in Autumn 2003

34. Preparation for assimilation of OBS • AMSU-B • EARS • QuikSCAT • AIRS • MODIS winds

35. Conditions for use  1 2 3 4 5 9 < scan position < 82    Sea    Land orog<1000m/1500m for channels 4/5   Ts > 278 K and |ob-fg|ch 2< 5 K    Assimilation of AMSUB data • In a similar way as for AMSUA & HIRS data: • Scan and air-mass bias correction • 250 km horizontal thinning

36. Effect of assimilating AMSUB data on TCWV (Total Column Water Vapour) field Winter period Spring period

37. = = RMS, std dev. and bias errors (without minus with AMSUB data) for geopotential [m] as a function of forecast and vertical ranges Winter scoreswrt radiosondes (27 Nov – 7 Dec 2003) Red: degradation from AMSUB Green: improvement from AMSUB RMS Std dev. Bias Northern Hemisphere Southern Hemisphere Tropics

38. = = RMS, std dev. and bias errors (without minus with AMSUB data) for geopotential [m] as a function of forecast and vertical ranges Spring scoreswrt radiosondes (17 Mar – 3 Apr 2004) Red: degradation from AMSUB Green: improvement from AMSUB RMS Std dev. Bias Northern Hemisphere Southern Hemisphere Tropics

39. Increase of humidity over land in Tropics and Southern Hem. and slightly in Northern Hem. Positive impact on the forecast scores (for geopotential, temperature and slightly for wind) Experiments ongoing Spring period to be run again with adjusted bias correction Bias correction tuning necessary as assimilating AMSUB data modifies the background, i.e. the way AMSUA/AMSUB/HIRS data are being used Assimilation of AMSUB data - Conclusion

40. EARS data • Eumetsat ATOVS Retransmission Service • Data rapidly available • No blind orbit for NOAA17 Research experiments with locally received AMSUA data. (N. Fourrié) 13 March 2003 12 UTC Data available for the operational production, 1h50 cut-off time Nesdis/Bracknell data • Lannion data 45W/40E/70N/30N • Even more rapidly available, but smaller area & only NOAA16/NOAA17

41. « EARS-Lannion » hybrid product since February 2004 : Recalibrated level 1c radiances computed from level 1a ATOVS data received from EARS HRPT stations and locally at Lannion

42. Impact of EARS and Lannion data in addition to Bracknell data (rms/bias wrt radiosondes) Temperature @ 250 hPa forecast range (hour)  12 36 48 BracknellBracknell+EARS+Lannion First step: assimilation in operational model ARPEGE Next step: assimilation in regional model ALADIN (… AROME) in research mode AMSUA, HIRS, AMSUB (observation density, bias correction, …)

43. First test to be set up : - preprocessing (like ECMWF) of the raw data, (inversion at 50km of resolution). - choice of the most likely solution, produced by the inversion, and her opposite - final choice among the deviation against the first guess - quality control against rain, sea-ice and land contamination - thinning at 100km Towards an assimilation of the Quikscat data

44. AIRS : AtmosphericInfRaredSounderThomas Auligné 1 spot out of 18 324 out of 2378 channels

45. Flat bias correction for each channel calculated over all active data Basic definition for so: 0.6 K for upper temperature channels 1 K for lower temperature channels 2 K for water-vapor channels AIRS : assimilation suite Bias correction Observation error estimation Assimilation period of 19 days : 2003.08.01  2003.08.19 CTRL = latest ARPEGE suite (including HIRS) EXP = CTRL + AIRS (all data in 6h assim window) + more iterations in the 2nd 4D-Var minimisation

46. - Gross check : 150 < Tb < 350 & (obs - fg) < 20 - First-guess check : (obs - fg)² <  (o² + b²) - Channels in O3 and SW bands, over land, peaking above/near model cloud top (1hPa), at edges of scan are blacklisted 176 channels used Mitch Goldberg cloud detection scheme: based on thresholds recomputed for ARPEGE model • LW window channel: Tb(965.43cm-1) > 270 K • Model SST versus SW window channel (2616.095cm-1) (night only) • Model SST versus predicted SST (from channels 918.65, 965.32, 1228.09, 1236.40 cm-1) • VIS/NIR imager: less than 10 % cloud in pixel (day only) Information on a pixel basis AIRS : assimilation suite Channel selection Cloud detection

47. Forecast range Pressure Geopotential Temperature = = RMSCTRL - RMSEXP VERIF = ECMWF analysis

48. Geopotential Temperature Humidity VERIF = TEMP observations

49. Pre-operational by summer 2004 Conclusion for AIRS data “Conservative” assimilation (only 176 channels, over clear pixels, flat bias correction) is neutral/slightly positive for summer experiment • To be confirmed/improved with more extensive testing

50. Use of MODIS winds at Météo-France • First experiment with a 3-weeks data set of Terra Modis winds • Basic - Modis winds used "as is“ • Terra Modis only • 17 march – 5 april 2004 • Horizontal density thinning: average of ~250 km • No use of quality index