基于多模式数值预报产品的黔东南州面雨量检验

Performance Evaluation of Areal Precipitation Forecasting for Qiandongnan Prefecture Based on Multimodal Numerical Forecasting

摘要: 本文使用黔东南州2022年7月~2022年12月15个国家基本气象站、1个国家基准气候站及448个气象观测站逐日降雨资料以及ECMWF、JAPAN_MR、GRAPES_GFS、NCEP_GFS 4种数值模式20时起报的24小时降水预报产品，采用平均绝对误差、正确率、模糊评分、TS评分、漏报率、空报率等方法，对该时间段黔东南州面雨量预报效果进行检验评估。结果表明：1) 从4种数值预报模式预报效果分析得出JAPAN_MR模式在3个流域预报效果均较好，ECMWF模式在㵲阳河流域的预报效果优于其它模式，NCEP_GFS模式在清水江上游预报效果优于其它模式。对各模式进行预报误差分析得出GRAPES_GFS模式在黔东南州面雨量预报误差最小。2) 随着降水等级的增大，TS评分逐渐降低，漏报率、空报率逐渐增大，所有模式的预报效果逐渐降低。NCEP_GFS模式在小雨、中雨等级面雨量预报中优于其余3种模式。4种数值模式的空报率均大于漏报率，小雨等级最为明显，因此在面雨量预报中需注意消空处理。3) 从分流域预报总体效果来看，ECMWF、JAPAN_MR、GRAPES_GFS、NCEP_GFS 4种数值模式在㵲阳河流域表现较差，在都柳江流域表现最好，在清水江上游的预报效果优于清水江下游的预报效果。

Abstract: This paper uses the daily rainfall data of 15 national basic meteorological stations, 1 national reference climate station and 448 meteorological observation stations in Qiandongnan Prefecture from July to December 2022, as well as the 24-hour precipitation forecast products reported by the ECMWF, JAPAN_MR, GRAPES_GFS and NCEP_GFS numerical models starting at 20:00, and adopts mean absolute deviation, accuracy, fuzzy grading, threat score, false negative rate and false alarm rate, to evaluate the performance of areal precipitation forecasting for Qiandongnan Prefecture during this time period. The results show that: 1) From the analysis of the prediction effects of four numerical prediction models, it can be concluded that the JAPAN_MR model has good prediction effects in all three river basins. The ECMWF model has better prediction performance than other models in the Wuyang River Basin, and the NCEP_GFS model has better prediction performance than other models in the upstream of the Qingshui River. Analyzing the prediction error of each model, it is found that the GRAPES_GFS model has the smallest prediction error of surface rainfall in Qiandongnan Prefecture. 2) With the rising level of precipitation, the threat score gradually decreases, the false negative rate and false alarm rate increase, and the forecast ability reduced. The NCEP_GFS model performs better than the other three models in predicting surface rainfall with light and moderate rain levels. The false alarm rate of all four numerical models is greater than the false negative rate, and it is the most obvious at light rain level. Therefore, attention should be paid to void elimination in surface rainfall forecasting. 3) From the overall effect of watershed forecasting, the ECMWF, JAPAN_MR, GRAPES_GFS and NCEP_GFS numerical models perform poorly in the Wuyang River Basin, perform best in the Duliu River Basin, and have better prediction results in the upstream of the Qingshui River than in the downstream of the Qingshui River.