基于邻域最优概率的GRAPES快速更新循环同化预报系统逐时降水订正方法

CORRECTION METHOD BASED ON NEIGHBORHOOD OPTIMAL PROBABILITY FOR HOURLY PRECIPITATION FORECAST FROM GRAPES RAPID UPDATING CYCLE ASSIMILATION AND FORECASTING SYSTEM

针对高分辨率数值天气预报的时空不确定性,利用邻域最优概率方法对华南区域GRAPES快速更新循环同化预报系统的24 h预报进行逐时降水订正和检验评估。结果表明:(1)邻域法能改善模式降水预报的空间不确定性,最优邻域半径随降水等级增加而减小,强降水的最优邻域半径约为60 km;(2)通过引入时间滞后因子,可进一步改善模式不同时间起报的不确定性,结合Brier评分确定了时间滞后窗为4 h;(3)提出基于邻域最优概率阈值的降雨进行分级订正方法,有效提升了降水客观预报能力,晴雨预报较模式全部为正技巧,TS评分达到0.89以上,总体提升幅度约5.3%;强降水预报同样均为正技巧,TS评分呈先降后升趋势,在12 h时效前后预报效果最优,进一步提升了GRAPES快速更新循环同化预报系统的业务预报水平。

To tackle the spatial-temporal uncertainty of high-resolution numerical models,in this paper,the neighborhood optimal probability method is developed to improve the hourly precipitation forecast in the24 h forecasts from the GRAPES rapid updating cycle assimilation and forecasting system(GRAPES_GZ_R)for southern China.Results show that:(1)The spatial neighborhood probability can help reduce the spatial uncertainty of precipitation forecast from the GRAPES_GZ_R.The optimal neighborhood radius decreases with the increase of precipitation grade,and it is about 60 km for heavy rainfall prediction.(2)The time-lagged factor can help further reduce the uncertainty of forecasting initialized at different time,and the time-lagged window is determined to be 4 h by using the Brier score.(3)The precipitation classification correction method based on the optimal probability threshold of neighborhood is proposed,which effectively improves the objective precipitation forecast.Compared with previous forecasts,the present clear-rainy forecast shows all positive skills with the TS score above 0.89 and the overall improvement being about 5.3%.The heavy rain forecast also shows all positive skills;corresponding TS score decreases first and then increases,and finally achieves the best effect at about 12 h,which further improves the operational application capacity of the GRAPES_GZ_R.