集合空间检验技术在东北夏季强降水预报中的应用

Application of Ensemble Spatial Verification Technique in Forecasting Heavy Precipitation in Northeast China during Summer

摘要:

将2018年东北夏季强降水天气过程分为西风槽型、切变线型和低涡型，筛选各类天气过程的典型降水个例，对过程分析及常规技巧评分与所研发的集合空间降水检验方法进行对比分析，结果表明：常规技巧评分在降水范围较广、强度较强时具有一定的评估能力，但当强降水分布零散、点降水占主导时，其评估劣势显示出来，即很难在点对点的检验中完美匹配预报与观测，由于严重的“双重惩罚”问题影响评估结论的可靠性及辨识度，而集合降水空间检验方法能够有效弥补，显著提高评估辨识度。

Abstract:

The summer heavy precipitation in Northeast China is characterized by its localized nature, short duration, and high intensity. Such extreme weather events have always been a challenge for numerical forecasting. The traditional precipitation skill scores are often ineffective for verifying heavy precipitation processes in Northeast China due to the "double penalty" effect caused by minor spatial and temporal discrepancies. The ensemble precipitation spatial verification technique can effectively improve the validity of precipitation assessment. In this paper, the heavy precipitation events in Northeast China during the summer of 2018 were categorized into westward trough, shear line, and low vortex types,with typical precipitation cases selected for each weather process.A comparative analysis involving process analysis, ensemble equitable threat score (EETS), and the developed ensemble precipitation spatial verification method was conducted.The results indicate that EETS have certain evaluative capabilities when the precipitation area is extensive and the intensity is high. However, when heavy precipitation is scattered and point rainfall dominates, the inherent evaluation disadvantages become evident. Specifically, it becomes challenging to achieve a perfect match between forecasts and observations in point-to-point verification, leading to significant reliability and discernibility issues in the evaluation conclusions due to the serious "double penalty" problem. However, the ensemble precipitation spatial verification method can effectively compensate for these shortcomings, significantly enhancing the discernibility of the assessment.