INSTITUTE OF METEOROLOGY AND WATER MANAGEMENT Hydrological applications of COSMO model Andrzej Mazur Institute of Meteorology and Water Management Centre of Numerical Weather Forecasts 61 Podleśna str., PL-01673 Warsaw, Poland
Hydrological applications of COSMO model Contents 1. Goals 2. Methods 3. Results 4. Conclusions
Hydrological applications of COSMO model Goals What is needed of DMO for precipitation-runoff models? 1. Forecast of precipitation (of course)… 2. … together with air temperature While precipitation is obviously the main driving factor for a hydrological model, the temperature data provides information on the state of the precipitation and the available potential for evaporation Moreover,…
Hydrological applications of COSMO model Methods
Hydrological applications of COSMO model where: y - measurement vector b - multiple regression coefficients (time dependent) h - predictors - model forecast values Q - error covariance r - observational error P - forecast covariance e - forecast error w - temporary scalar k - Kalman gain
Hydrological applications of COSMO model Results Results of experiments for selected days/periods for temperature and precipitation: - June 30, 2007 – change from COSMO version 3.05 to 4.0 - January 01, 2008 – six months of COSMO version 4.0 runs - August 04, 2008 – heavy storm over Poland (part I) - August 15, 2008 – heavy storm over Poland (part II) and only for precipitation – increased resolution computations, heavy precipitation (mainly convective type): - May 04, 2005 - June 10, 2005 - August 09, 2005
Hydrological applications of COSMO model End of COSMO version 3.05 runs, June 30, 2007 Results of June 30, 2007 – precipitation (morning and afternoon model runs, DMO and corrected results)
Hydrological applications of COSMO model End of COSMO version 3.05 runs, June 30, 2007 Bias of June 30, 2007 – precipitation (morning and afternoon model runs, DMO and corrected results)
Hydrological applications of COSMO model First six months of COSMO version 4.0 runs Results of January 01, 2008 – precipitation (morning and afternoon model runs, DMO and corrected results)
Hydrological applications of COSMO model First six months of COSMO version 4.0 runs Bias of January 01, 2008 – precipitation (morning and afternoon model runs, DMO and corrected results)
Hydrological applications of COSMO model Bias and RMSE values of DMO and AR-corrected results for selected hour of forecast June 30, 2007 January 01, 2008 Output, hour bias RMSE AR, 06:00 0.318 1.526 DMO, 06:00 0.328 1.566 AR, 12:00 0.092 0.662 DMO, 12:00 0.105 0.674 AR, 18:00 0.153 0.901 DMO, 18:00 0.176 0.915 Output, hour bias RMSE AR, 06:00 0.311 1.154 DMO, 06:00 0.774 1.307 AR, 12:00 0.223 1.248 DMO, 12:00 0.894 1.349 AR, 18:00 1.450 2.118 DMO, 18:00 1.774 2.358
Hydrological applications of COSMO model Heavy storms over Poland, August 04, 2008 Results of Aug. 04, 2008 – precipitation. Morning model run, AR results (left), DMO (right) for 06:00 (upper) and 18:00 UTC (lower). Measured sum of precipitation marked with crosses with size proportional to amount.
Hydrological applications of COSMO model Heavy storms over Poland, August 15, 2008 Results of Aug. 15, 2008 – precipitation. Morning model run, AR results (left), DMO (right) for 06:00 (upper) and 18:00 UTC (lower). Measured sum of precipitation marked with crosses with size proportional to amount.
Hydrological applications of COSMO model Bias and RMSE values of DMO and AR-corrected results for selected hour of forecast August 04, 2008 August 15, 2008 Output, hour bias RMSE AR, 06:00 1.010 4.146 DMO, 06:00 1.727 4.504 AR, 12:00 0.215 3.079 DMO, 12:00 0.473 3.221 AR, 18:00 1.437 5.264 DMO, 18:00 2.089 5.648 Output, hour bias RMSE AR, 06:00 0.998 6.251 DMO, 06:00 1.874 6.563 AR, 12:00 0.258 4.854 DMO, 12:00 -0.945 6.472 AR, 18:00 2.332 9.983 DMO, 18:00 3.473 10.148
Hydrological applications of COSMO model Increased resolution model runs Results of May 04, 2005 – precipitation. Morning model run, AR results (left), DMO (right) for 06:00 (upper) and 18:00 UTC (lower). Measured sum of precipitation marked with crosses with size proportional to amount.
Hydrological applications of COSMO model Increased resolution model runs Results of June 10, 2005 – precipitation. Morning model run, AR results (left), DMO (right) for 06:00 (upper) and 18:00 UTC (lower). Measured sum of precipitation marked with crosses with size proportional to amount.
Hydrological applications of COSMO model Increased resolution model runs Results of Aug. 09, 2005 – precipitation. Morning model run, AR results (left), DMO (right) for 06:00 (upper) and 18:00 UTC (lower). Measured sum of precipitation marked with crosses with size proportional to amount.
Hydrological applications of COSMO model Bias and RMSE values of DMO and AR-corrected results for selected hour of forecast, May 04, 2005, June 10, 2005 August 09, 2005. Output, hour bias RMSE AR, 06:00 3.225 7.452 DMO, 06:00 3.779 7.862 AR, 12:00 0.461 7.298 DMO, 12:00 1.727 9.049 AR, 18:00 3.295 7.701 DMO, 18:00 5.165 8.427 Output, hour bias RMSE AR, 06:00 2.970 8.503 DMO, 06:00 3.986 11.517 AR, 12:00 2.134 6.775 DMO, 12:00 2.413 6.953 AR, 18:00 3.972 11.681 DMO, 18:00 4.659 13.720 Output, hour bias RMSE AR, 06:00 -0.186 3.324 DMO, 06:00 -1.190 5.858 AR, 12:00 -1.472 5.316 DMO, 12:00 -3.636 9.077 AR, 18:00 0.352 7.676 DMO, 18:00 -1.686 9.964
Hydrological applications of COSMO model Air temperature RMSE changes during period August 01, 2008 to August 21, 2008 (DMO and AR results shown). Precipitation RMSE changes during period August 01, 2008 to August 21, 2008 (DMO and AR results shown).
Hydrological applications of COSMO model Conclusions Temperature forecast corrections (even based on redundant predictors) is easier to develop, seem also to be more stable during a learning process (no sudden/drastic changes of coefficients over the entire period). Method – even in this simple approach – is able to “detect” and correct not only any factor „out” of the model, but also systematic errors in its results. The change of COSMO model version (from 3.05 to reference version 4.0) has not significant influence on time-evolvement of coefficients. Precipitation seems to be well-posed as far as the predictors are concerned. Geographical coordinates, elevation, time of measurement and previous measured and forecasted values seem to be fairly set for the purpose. Of course, artificial, but obvious constrain has always to be applied – corrected forecast value of precipitation must not be less than zero. Additional problem with precipitation forecast correction - correction process is hardly able to “create” any amount of precipitation from “nothing”. If DMO forecast predicts no rain at a certain point, it’s not possible to obtain a non-zero (or significant amount of) precipitation using AR scheme.
Thank you for your attention. IMGW 01-673 Warszawa, ul.: Podleśna 61 tel.: (022) 56 94 134 fax: (022) 56 94 356 mobile: 0 503 122 134 andrzej.mazur@imgw.pl www.imgw.pl