Statistical Downscaling Of Precipitation Based On The Multimodel Ensemble Forecast

Using1-7day precipitation forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), the Japan Meteorological Agency (JMA), the National Centers for Environmental Prediction (NCEP) and the UK Met Office (UKMO) in the TIGGE datasets, and the TRMM/3B42RT rainfall product as "observed value", the statistical downscaling forecasts of the precipitation over East Asian area (90°E-140°E,15°N-45°N) have been conducted. By the assesses on the precipitation forecasting capabilities of the four centers from the TIGGE data found that there exist some problems of the rainfall forecasting, such as trace amount rainfall forecasts without observed rain and underestimate of the extreme precipitate. First, Use the method of bilinear interpolation, the rainfall forecast data of the four centers was interpolated from1.25°×1.25°to0.25°×0.25°grid, consisted with the resolution of TRMM data. Then the weather phenomenon forecast was divided into rain and no rain by means of the logistic regression. The statistical downscaling based on linear regression was subsequently used to improve the interpolated precipitation forecast for rainy weather. The results show that the logistic regression can effectively eliminate false alarms of light rain, and after statistical downscaling, the ACC between the precipitation and the "observation" is larger with the downscaling method and the RMSE is smaller, which means the downscaling forecasts are more accurate than interpolated ones in each forecast leading time.The bias-removed ensemble mean (BREM) was applied to conduct the multimodel ensemble precipitation forecasts. For the problem of the underestimate of the extreme precipitation caused by the downscaling and integrated multi-mode methods, the classification downscaling methods was proposed, and the secondary revisions was used to correct the heavy rains. The results show that the multimodel ensemble forecasts are superior to those of individual models in terms of the root-mean-square errors (RMSE) and the anomaly correlation coefficients (ACC) of the precipitation forecasts. The improvement of the statistical downscaling multimodel forecast upon the original forecast decreases as forecast leading time increases. As to the problem of the underestimate of the extreme precipitation, classification downscaling and the secondary revisions methods was effective. The ETS scores that above the threshold of heavy rain was improved, and after the correction, the overall ACC was also lager than downscaling multimodel forecast Since the numerical models own problems and the application of statistical methods, the intersite correlations of precipitation forecast were smaller than the actual results, The approach of Schaake Shuffle presented for reordering the ensemble output in order to recover the space-time variability in precipitation and fields., the ensemble members for a given forecast day are ranked and matched with the rank of precipitation data from days randomly selected from similar dates in the historical record The ensembles are then reordered to correspond to the original order of the selection of historical data. Using this approach, the intersite correlations of each ensemble member forecasting could be recovered, which are more close to the observed ones.