Statistical Downscaling And Calibration Of Precipitation,Wind Speed And Surface Temperature Forecast In China

Based on the global ensemble forecasting data from ECMWF,JMA,UKMO and NCEP in TIGGE dataset,the hourly precipitation data merged by China Automatic Station(CAS)and CMORPH precipitation products and ERA-Interim reanalysis data,the study of statistical downscaling and calibration of precipitation,wind speed and surface temperature forecasts in China is carried out.Firstly,the low-resolution model forecast data is integrated into the same high-resolution fine grid as the observation data by bilinear interpolation method.The statistical downscaling of the categorized regression using the historical observation data is carried out to improve precipitation forecast and compared with the statistical downscaling of unary linear regression.Because of the serious wetness and dryness of precipitation forecast and little improvement of forecast skill after statistical downscaling by the categorized regression,the zonal frequency matching method improved by Kalman filter is used to calibrate the area deviation of each precipitation level,and the threshold method is used to reduce the false alarm of light rain.Meanwhile,the temperature and wind speed forecasts are improved by the unary linear regression statistical downscaling.On the basis of linear regression,the temperature and wind speed are further calibrated by decaying average error correction method of Kalman filter type.The results show that: compared with the simple bilinear interpolation,the unary linear regression can reduce the errors of precipitation,temperature and wind speed forecasts.The categorized regression can further reduce the errors of precipitation forecast and improve the correlation between precipitation and observation.After frequency matching,the wet deviation of light rain and the dry deviation of heavy rain are significantly reduced.The area of precipitation of each threshold is closer to the observed value.The ETS score and the accuracy of precipitation forecast are improved to a certain extent,and the false alarm rate of light rain and the missing rate of heavy rain are reduced.The threshold method can greatly reduce the wet deviation and the false alarm rate of light rain,and improve ETS score of light rain.On the basis of unary linear regression statistical downscaling,decaying mean error correction can reduce the bias of temperature and wind speed forecasts,improve the accuracy of them and the correlation between temperature forecast and observation.