顾及分类与定量误差订正的数值预报降水统计后处理方法  

Integrated statistical post-processing methods for categorical and quantitative errors correction of numerical precipitation forecasts

作　　者：李伶杰 王银堂[1,2] 云兆得 刘勇 王磊之[1] 苏鑫 徐勇 LI Lingjie;WANG Yintang;YUN Zhaode;LIU Yong;WANG Leizhi;SU Xin;XU Yong(The National Key Laboratory of Water Disaster Prevention,Nanjing Hydraulic Research Institute,Nanjing 210029,China;Yangtze Institute for Conservation and Development,Nanjing 210098,China;State Key Laboratory of Water Resources Engineering and Management,Wuhan University,Wuhan 430072,China;The Suzhou Branch of Jiangsu Province Hydrology and Water Resources Investigation Bureau,Suzhou 215011,China)

机构地区：[1]南京水利科学研究院水灾害防御全国重点实验室,江苏南京210029 [2]长江保护与绿色发展研究院,江苏南京210098 [3]武汉大学水资源工程与调度全国重点实验室,湖北武汉430072 [4]江苏省水文水资源勘测局苏州分局,江苏苏州215011

出　　处：《水科学进展》2024年第1期48-61,共14页Advances in Water Science

基　　金：国家重点研发计划资助项目(2022YFC3202802);国家自然科学基金资助项目(52009081)。

摘　　要：采用统计后处理方法的数值预报降水是延长水文预报有效预见期的重要途径,已有统计后处理方法不能同时订正预报降水的分类和定量误差,且对预报降水有效预见期的影响关注不足。提出耦合经验分位数映射模型(EQM)和伯努利-元高斯模型(BMGD)的统计后处理方法EQM-BMGD,建立用于有效预见期评价的综合精度指标,应用于汉江流域。研究结果表明:EQM-BMGD集成了2种单一方法的优势,并输出了更高精度的预报降水;订正后面平均预报降水各预见期晴雨预报准确率(O P)和绝对平均误差(E MA)的增益均超过了10%,预见期222~228 h的O P仍接近0.7且E MA低于0.7 mm/(6 h),有效预见期延长18~66 h;在栅格尺度上,所有栅格在预见期96~102 h的O P和E MA增益分别超过10%和20%,除西南少数栅格外,O P超过0.8同时E MA控制在1.0 mm/(6 h)以下,北部部分栅格有效预见期延长了18~54 h。EQM-BMGD被证实能够兼顾分类和定量误差的削减,丰富了数值预报降水统计后处理方法的选择。The utilization of statistical post-processed numerical precipitation forecasts is a significant approach to extend the effective forecast period of hydrological forecasting.Existed statistical post-processing methods struggle to simultaneously correct dichotomous and quantitative errors,and their impact on the effective forecast lead time for precipitation forecasting is frequently overlooked.In this study,we introduce a novel post-processing scheme called EQM-BMGD,which combines the Empirical Quantile Mapping model(EQM)and the Bernoulli-meta-Gaussian Distribution(BMGD).Additionally,we establish a comprehensive accuracy metric for evaluating the effective forecast period.Using the Han River Basin as a case study,comparative outcomes showed that EQM-BMGD integrated the strengths of the two individual methods,achieving precipitation forecasts with superior accuracy.The forecast accuracy(O P)and mean absolute error(E MA)of the post-processed average-basin forecasts increased by more than 10%,the O P of the forecast period 222—228 h was still close to 0.7,and E MA was less than 0.7 mm/(6 h),and the EFPs were extended by 18—66 h.On a grid scale,the gains of O P and E MA for the 96—102 h forecast period exceeded 10%and 20%respectively for all grids.Except for a few grids in the southwest,the O P surpassed 0.8 while the E MA remained below 1.0 mm/(6 h).In addition,the EFPs of the grids in the northern part were lengthened by 18—54 h.It is demonstrated that the EQM-BMGD can effectively correct both categorical and quantitative errors,thereby enriching the available methodologies for statistical post-processing of numerical precipitation forecasts.

关 键 词：数值预报降水 统计后处理 经验分位数映射 伯努利-元高斯分布 有效预见期 

分 类 号：P456.8[天文地球—大气科学及气象学]