Denis Blinov, G. Rivin, M. Nikitin, A. Bundel, A. Kirsanov, I. Rozinkina Data assimilation system based on nudging for COSMO-Ru7/So2 1COSMO GM Sibiu, Sept. 2013
Outline Initial conditions (IC) in operative Cosmo-Ru system Errors IC Data Assimilation System (DAS) based on nudging Estimation experiments with DAS Conclusions and outlook 2COSMO GM Sibiu, Sept. 2013
Initial conditions in Cosmo-Ru In the current operational scheme initial data are prepared from the interpolated data GME (h=20 km); Initial conditions are concluded atmosphere fields (U, V, P, W, T, QV), surface fields (T_S, T_SNOW, T_ICE, H_ICE, T_SO, W_SO) external parameters. 3COSMO GM Sibiu, Sept. 2013
Errors of initial conditions For the entire domain COSMO there are significant errors in the field T_2m at the beginning of the forecast; It is supposed that errors in the initial data associated with all 3 types of fields: external parameters, atmospheric and surface fields. 4COSMO GM Sibiu, Sept. 2013
Errors T_2m in COSMO-So2 (Adler-cluster) for winter 2011/2012 (left) and 2012/2013 (right) for 00, 06, 12, 18 runs 5 RMSE ME Errors of initial conditions
DA system Through the assimilation, we were trying to reduce the errors in the atmospheric fields DAS uses observations SYNOP, TEMP, SHIP, BUOY, AIREP The scheme Nudging in Cosmo is correcting atmosphere fields T, PS, H, QV, U, V On first step DAS is based on 6 hour cycle - Main Run (MR) On the next step DAS will included on continuous cycle - Recalculate Run (RR) 6COSMO GM Sibiu, Sept. 2013
Scheme of assimilation system Namelist variableDA-M 07 RU 07 kmDA-R RU02DA-M RU02 RU 02 km data_iniGME — 6 hDA-M 07 DA-R — 6h DA-M + 0h data_bdGME — 6 hGME + 0 h RU07 — 6 h hstop number of runs cut-off time02:45:0002:50:00 06:5001:1001:10:00 hnudgend104 4 DA nudgcast DA nudgcast 7COSMO GM Sibiu, Sept. 2013
The old(left) and new parallel(right) operational scheme COSMO-RU07 и COSMO-RU02 download GME 00UTC06UTC12UTC18UTC start02:5008:5014:5020:50 end03:2509:1015:2521:10 Time work00:3500:2000:3500:20 00UTC06UTC12UTC18UTCRU07 00UTC06UTC12UTC18UTC 02:5008:5014:5020:50start02:3002:5008:3008:5014:3014:5020:3020:50 03:3009:1015:3021:10end02:5003:3008:5009:1014:5015:3020:5021:10 00:4000:2000:4000:20Time work00:2000:4000:20 00:4000:20 00UTC06UTC12UTC18UTCRU02 00UTC06UTC12UTC18UTC 03:3009:1015:3021:10start 02:4003:0008:4009:0014:4015:0020:4021:00 05:1010:5017:1022:50end 03:0005:0009:0011:0015:0017:0021:0023:00 01:40 Time work 00:2002:0000:2002:0000:2002:0000:2002:00 COSMO GM Sibiu, Sept
Model grids and used observations gridOBSERVATIONS domain model T_2m (TEMP) wind_10m, pmsl, Td_2m (TEMP + SYNOP) ETR CFO SFO VFO ENA SIB The number of grid points of the model ETR7 more than stations in ~1600 for the temperature and ~32000 for SFO2. Not enough of upper-air stations for successfully correcting temperature. You must use t_2m observation from SYNOP. 9COSMO GM Sibiu, Sept. 2013
Experiments Experiments for COSMO-Ru7 with: IC from DAS IC from GME List verification stations: EPR887 – 887 stations on entire domain COSMO-Ru7 EPR293 – 293 stations on entire domain COSMO-Ru7 SFO – 145 stations on South Federal District CFO – 220 stations on Central Federal District … Period testing: February 2012 July 2012 December 2012 COSMO GM Sibiu, Sept
Mean error (ME) T_2m for EPR December COSMO GM Sibiu, Sept. 2013
Mean error T_2m for SFO 1-10 December COSMO GM Sibiu, Sept. 2013
Errors T_2m for initial time and forecast on 6 hours Found the understatement of the amplitude for all terms and for all seasons; The daytime temperature is underestimated, night overstated, morning and evening errors are minimal; On the chart of errors at the beginning of the forecast is observed local extremum. We believe that this effect is connected with interpolation from coarse grid (spinup); Very significant differences between the terms. On a 12-h forecast error temperature is minimal. 13COSMO GM Sibiu, Sept. 2013
SFO 00h_forecast03h_forecast06h_forecast refdasrefdasrefdas 00UTC 0,79 0,950,92 0,73 0,37 0,11 06UTC0,35 0,19-0,24 -0,41 -0,50 -0,68 12UTC 0,02 -0,64 0,14 -0,180,41 0,08 18UTC0,45 0,23 0,47 0,24 0,76 0,44 total0,40 0,18 0,32 0,10 0,26 -0,01 Krasnaya Polyana 00h_forecast03h_forecast06h_forecast refdasrefdasrefdas 00UTC2,53 1,70 1,98 1,74 2,44 2,19 06UTC3,25 2,88 0,37 -0,21-1,92 -2,37 12UTC -0,75 -1,45 -0,33 -0,501,07 0,89 18UTC2,82 1,370,87 1,531,55 1,14 total1,96 1,12 0,72 0,64 0,79 0,46 * Blue value indicate better results Mean Error for T_2m ref - experiment with initial conditions from GME das - experiment with initial conditions from Data Assimilation System 14COSMO GM Sibiu, Sept. 2013
15COSMO GM Sibiu, Sept Experiments with DWD assimilation for COSMO-Ru domain, February 2012, EPR293
Experiments with DWD assimilation for COSMO-Ru domain, February 2012, SFO 16COSMO GM Sibiu, Sept ME RMSE
Results testing DAS for field T_2m Assimilation helps to improve the temperature field during the early stages in the case of large errors in GME fields; When errors GME are minimal assimilation brings errors from 6-hour lead time. In some cases, it worsens the initial field - forecasts from 12 UTC; Mainly ME better on 0.3 degree for DAS; It is necessary to correct/change the temperature of the soil, snow, sea for Continuous Cycle. 17COSMO GM Sibiu, Sept. 2013
ME, RMSE T_2m for EPR December COSMO GM Sibiu, Sept. 2013
ME, RMSE Td_2m for EPR December COSMO GM Sibiu, Sept. 2013
Mean error, RMSE PMSL for EPR December COSMO GM Sibiu, Sept. 2013
Mean error, RMSE wind speed 10m for EPR December COSMO GM Sibiu, Sept. 2013
Conclusions Mainly effect of “spinup” at the beginning of the forecast disappears; Fields T_2m, Td_2m, PSML, Wind Speed 10m is better with DAS in 0-3 hours forecast time; When errors GME are minimal assimilation brings errors from 6-hour lead time. In some cases, it worsens the initial field - for example forecasts from 12 UTC; It is necessary to correct/change the temperature of the soil, snow, sea for Continuous Cycle; It is necessary to use more observations. 22COSMO GM Sibiu, Sept. 2013
Outlook Correct surface field: o T_Soil, T_air (correction module by Sapuntsova) o T_S via Flake Increase count of data in DAS: o feedback-file o AMS o Radar data (Latent Heat Nudging) Testing continuous cycle for COSMO-So2 km 23COSMO GM Sibiu, Sept. 2013
24COSMO GM Sibiu, Sept Experiments with module correction T_SO, T_air by Sapuntsova Verification for 18 February 2012
25COSMO GM Sibiu, Sept Experiments with module correction T_SO, T_air by Sapuntsova RMSE t_2m for February 2012
Thank you for attention! Questions? Remark? COSMO GM Sibiu, Sept
COSMO GM Sibiu, Sept Soil temperature influence Рис. 5.7 Сравнение профилей температуры почвы по наблюдениям в Метеорологической обсерватории МГУ и по прогнозу COSMO-RUот 26 апреля за 10:00 26 апреля.
Интеллектуал, Балчуг COSMO GM Sibiu, Sept