基于格点实况的海南岛网格晴雨预报检验

采用高分辨率的格点降水实况资料，对2019年5月1日—9月30日中国气象局广东快速更新同化数值预报系统(CMA-GD)、中国气象局上海数值预报模式系统(CMA-SH9)和欧洲中期天气预报中心全球模式(ECMWF)的降水预报产品作海南岛晴雨预报的检验评估，结果表明:(1) CMA-SH9具有较高的晴雨准确率及较小的雨区面积偏差。CMA-GD(ECMWF)评分偏低与较高漏(空)报比例有关，漏(空)报易出现在五指山以南和东部沿海一带。三个模式在海南岛东部沿海一带的晴雨准确率随预报时效减小而提高；(2) CMA-GD多漏报，CMA-SH9和ECMWF多空报。CMA-GD和CMA-SH9比ECMWF具有较快速的调整能力，预报时效缩短，雨区面积偏差减小，晴雨评分提高；(3) 降水面积百分比为0~20%、20%~40%、40%~60% 的局地降水事件中，CMAGD和CMA-SH9预报雨区面积偏大，预报时效增加，面积偏差由偏多转偏少；降水面积百分比为60%~80%、80%~100%的降水事件中，雨区面积随时效增加呈增加趋势。(4) CMA-GD对海南岛北部高频降水中心具有较强的预报能力，但易漏报南部高频降水中心。CMA-SH9在海南岛北部易出现高频降水面积偏大，质心偏东的误差，但对南部的高频降水预报能力优于CMA-GD。通过最优面积阈值择优方案迭代集成CMA-GD、CMASH9和ECMWF的降水预报，可有效提高海南岛高分辨率网格晴雨预报准确率。 

VERIFICATION OF GRID SUNNY/RAINY FORECAST IN HAINAN ISLAND BASED ON GRIDDED OBSERVATION DATA

Based on high-resolution gridded precipitation data, precipitation products of the ECMWF global model, and CMA-GD and CMA-SH9 mesoscale models are examined for the period from May 1 to September 30 in 2019. These precipitation products are compared in terms of respective sunny/rainy scores and distribution patterns, as well as precipitation areas and the location of high-frequency precipitation centers. Results are as follows: (1) CMA-SH9 model has a higher sunny/rainy accuracy with minor precipitation area bias. The lower score of CMA-GD (ECMWF) is related to the high rate of missing (false) alarm ratio which is easy to appear to the south of Wuzhi Mountain and the eastern coastal area of Hainan Island. Three models show an identical feature that their sunny/rainy accuracy increases in the eastern coastal area of Hainan Island when time range reduces; (2) CMA-GD has more missing alarms while the CMA-SH9 and ECMWF models show more false alarms. Compared with ECMWF, CMA-GD and CMA-SH9 both adjust fast. They reduce the precipitation area bias and get higher sunny/rainy score as time range becomes shorter; (3) For local precipitation events with precipitation area accounting for 0-20%, 20%-40%, and 40%-60% of Hainan, CMA-GD and CMA-SH9 models predict larger precipitation areas compared with observations and show a trend of decrease as time range becomes longer. However, for precipitation events with precipitation areas accounting for 60%-80% and 80%-100% of Hainan, they forecast smaller precipitation areas and show a trend of increase as time range becomes longer. (4) The forecast products of CMA-GD model match the high-frequency precipitation center in the north of Hainan Island better; however, they are likely to miss the centers in the south of Hainan Island. CMA-SH9 model shows a larger high-frequency precipitation area with eastward deviation of center of mass in the north of Hainan Island, but its forecast capacity for the south of Hainan Island is better than that of CMA-GD. By integrating CMA-GD, CMA-SH9 and ECMWF precipitation products according to the optimal area threshold selection scheme, we can effectively strengthen the accuracy of high-resolution grid sunny/rainy forecasting in Hainan Island.