Decaying Average Statistical Downscaling Forecasting Technique Improvement Research Of Surface Temperature In China

In order to improve forecasting accuracy, a fine forecast test is designed by using self-adapting Kalman Filter-typed decaying average statistical downscaling forecasting technique of surface temperature in China. Decaying weight w which was defined as an only spatially variant function w(i) in decaying average statistical downscaling forecasting technique, has been redefined as spatially and synoptic process variant function w(i,p) in this research which i represents station information and p represents synoptic process information. Picking up similar weather information from history to find the law of cooling day is defined as similarity decaying average statistical downscaling technique; while collecting systemic bias from interpolation method to modifed forecasting result is defined as statistics decaying average statistical downscaling technique.The improved techiques are applied on observation data and numerical forecasting productions to get the fine forecast productions.1) The improved w(i,p) decaying average statistical downscaling forecasting technique performances better than the original method. Ideal experiment shows that, it is reasonable to redefined the decaying weight as spatially and synoptic process variant function w(i,p) which i represents station information and p represents synoptic process information, for the RMS error decreases by 0.11 ℃-0.28 ℃ averagely in 1-3 d forecast after intense cooling process day compared with the original method. The RMS error after mild cooling process day decreases about 0.1 ℃ averagely in 1-3d forecast.While the averaging RMS error of intense cooling process day 24h forecast decreases by 1.12℃ and 1.54℃ using w(i,p) similarity decaying average statistical downscaling tevhnique and w(i,p) statistics decaying average statistical downscaling tevhnique respectively. Forecast results has been significantly increased in spring and winter, which reduces the seasonal difference of prediction result.2) Based on the T639 numerical prediction and the hourly surface temperature data of observation stations, a forecasting test is designed. It shows that the model systematic errors has been greatly reduced using the decaying average algorithm used as bias correction tevhnique, the RMS error decreased by 2℃ and the correlation coefficient increased by 0.06. Compared with the original w(i) decaying average method, the RMS error decreases in nationwide using w(i,p) decaying average statistical downscaling forecasting technique. The RMS error decreases by 0.16 ℃-0.41 ℃ averagely in 24-168h forecast of intense cooling process day compared with the original method. The w(i,p) statistics decaying average tevhnique performance better for the RMS error decreases about 2℃ in 24h forecast.3) A prediction test is designed based on T213 ensemble forecast outcomes and the hourly surface temperature data of observation stations. It shows that the correlation coefficient increased by 0.09 averagely of each prediction member. The RMS error decreases in nationwide by w(i,p) decaying average statistical downscaling forecasting technique, which decreases by 0.15℃ averagely of 15 prediction members in 24h forecast compared with the original w(i) decaying average method. Better prediction results are calculated by similarity decaying average technique and statistics decaying average technique whose improved capacity on 15 prediction members are equivalent.