三种数值模式在黔东南州面雨量预报的检验评估

Verification and Evaluation of Three Numerical Models for Area Rainfall Forecast in Qiandongnan Prefecture

摘要: 本文通过双线性插值将ECMWF、GRAPES_GFS、NCEP_GFS三种数值模式输出的空间格点数据精确匹配到各个观测站点位置，并使用平均绝对误差、平均相对误差、TS评分、漏报率、空报率、正确率等检验方法对其在黔东南州㵲阳河、清水江、都柳江三个流域2022年7~9月的面雨量预报效果进行了检验。结果如下：1) 随着预报时效的延长，各数值模式的面雨量预报效果逐步降低。其中NCEP_GFS模式稳定性最差，虽在部分流域24 h预报效果较好，但48、72 h其预报效果下降趋势更为显著。ECMWF模式和GRAPES_GFS模式稳定性较好。2) 综合所有检验结果来看，ECMWF模式的预报效果总体优于另外两种模式。分流域分预报时效来看，㵲阳河流域中的72 h预报可以参考GRAPES_GFS模式，清水江、都柳江流域中NCEP_GFS模式的24 h预报表现较好。3) 三种数值模式所有预报时效的漏报率均远高于空报率，因此在面雨量预报中需注意消空处理；各数值模式在都柳江流域的预报效果最好，在㵲阳河流域的预报效果较差，需注意进行人工订正。

Abstract: In this paper, the spatial grid data outputted by ECMWF, GRAPES_GFS, and NCEP_GFS numerical models are precisely matched to the locations of each observation station using bilinear interpolation. The forecasting performance of surface rainfall for the Wuyang, Qingshui, and Duliu river basins from July to September 2022 is then evaluated using metrics such as mean absolute error, mean relative error, TS score, false negative rate, false positive rate, and accuracy. The results are as follows: 1) As the forecast time extends, the effectiveness of surface rainfall predictions from each numerical model gradually diminishes. The NCEP_GFS model exhibits the least stability. While it performs well for 24-hour forecasts in some basins, the decline in performance becomes more pronounced at 48 and 72 hours. Both the ECMWF and GRAPESGFS models demonstrate greater stability. 2) Considering all inspection results, the ECMWF model’s forecasting performance generally surpasses that of the other two models. For specific basins and forecast time periods, the GRAPESGFS model should be consulted for 72-hour forecasts in the Wuyang River basin, while the NCEPGFS model shows stronger performance for 24-hour forecasts in the Qingshui and Duliu river basins. 3) The omission rate for all three numerical models significantly exceeds the false alarm rate, necessitating attention to mitigating false positives in surface rainfall forecasts. The numerical models perform best in the Duliu River Basin and worst in the Wuyang River Basin, highlighting the need for manual correction.