A Study On Statistical Downscaling Forecast Of Precipitation In China Using Categorized Rainfall Regression

Based on the ensemble forecasts of 1-7 day daily accumulated precipitation of three summers 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 hourly merged precipitation product over China as the observed data, a refined weather forecast study on daily precipitation in China have been conducted.Firstly, the low-resolution forecast data was interpolated to the high-resolution grids consisted with the resolution of the observed data by the method of bilinear. Then the forecast data was categorized according to rainfall and the spatial sliding window was used to increase rainfall samples. The statistical downscaling by constructing different regression equations according to different levels of rainfall was used to improve the precipitation forecast and compared with the statistical downscaling by constructing the single equation. The results show that the forecasts after the categorized regression are more accurate than the ones after the direct regression, because the anomaly correlation coefficient is bigger and the root mean square error is smaller and the equitable threat score over different threshold values improves significantly. Also, the improvement of the forecast differs in models, leading time and rainfall levels and depends on the forecast of models.The forecasts of different levels of rainfall was improved by matching the frequency of the forecast data with the frequency of the observed data called Frequency Matching Method. The second correction was used to reduce the false alarm of the light rain forecast. Then the multi-model ensemble of four single centers was conducted. The results show that the equitable threat score of different rainfall levels improves significantly and the area deviation between the forecast data and the observed data decreases so that the accuracy rate of the forecast increases. The superfluous light rain and the missing heavy rain decreases obviously in different leading time. The second correction is applied to improve the forecast ability of light rain and the study demonstrates the significant effect of the method. A more accurate single forecast can be obtained by Multi-model Ensemble because the root mean square error is smaller and the anomaly correlation coefficient is bigger than all single centers.A method called Schaake Shuffle was applied to reconstruct the spatial variability and temporal persistence lost in the process of using statistical methods by putting the ensembles of the forecast data in the order of the observed data. The results show that the spatial and temporal correlations from the reconstruction methodology are closer to observed correlations than the raw result and the deviation of correlation between ensembles decreased to some extent.