乌鲁木齐区域数值预报业务系统降水预报检验与评估分析

基于乌鲁木齐区域数值预报业务系统,运用Ts和Bias评分方法,对2012年9月1日—2015年8月31日逐日2个起报时次的逐6 h累积降水量的年与季节预报性能进行检验,并从空间上分析了2015年全疆站点逐6 h累积降水量在4个预报时段的评分特征。结果表明:(1)2个起报时次的降水评分相差较小,00 UTC起报略优于12 UTC起报,2015年系统改进了白天大量级降水的空报现象。(2)系统对晴雨预报较为准确,Bias接近1,空报、漏报率很小;随着降水阈值的升高,Ts评分减小,Bias变幅增大,空、漏报率也随之增加。系统对强降水过程以漏报为主。(3)系统的降水预报能力存在季节差异,夏季Ts评分最高,秋季次之,冬季最小;随时间模式对四季降水预报能力均有提高,降低了冬季大量级降水的漏报率和夏季大量级降水的空报率。(4)在新疆地区,08—14 BT(Beijing Time)、14—20 BT、20—次日02 BT空报站点数多于漏报,14—20 BT空报率最高;在02—08 BT整体呈漏报。(5)各站点整体来看,白天Ts评分高于夜间,山区及邻近地区评分高于平原地区;西天山评分略优于东天山,夜间晴雨预报有天山北坡漏报、南坡空报的趋势。

Evaluation and Analysis of Precipitation predictability by Desert Oasis GobiRegional Assimilation and Forecast System

Based on Desert Oasis Gobi Regional Assimilation and Forecast System (DOGRAFS), this paper evaluate the annual and seasonal predictability of both predicted and observed 6 h rainfall forecasts products at two initialized times daily during 1 November 2012 to 31 August 2015 by means of Threat Score and Bias Score. Additionally, monthly score of precipitation in the past four years is analyzed and the capability to precipitation forecast of DOGRAFS and NMC-GRAPES in 2014 and 2015 is compared. The results indicate that: (1) The prediction performance between two initialized times is quite similar, although the result at 12 UTC is slightly better than 00 UTC. (2) Threat Score decreases and variation amplitude of Bias Score increases with the increase of threshold of rainfall, i.e., the greater the rainfall level is, and the lower the credibility of forecast will be. And meanwhile, the false alarm ratio and missed event ratio will increase. Overall DOGRAFS overestimates daytime precipitation but underestimates nighttime rainfall. What''s more, most heavy rainfall events are missed. (3) The credibility of rainfall forecast varies with season, being the greatest in summer, secondary in autumn, and least in winter. (4) Rainfall predictability was gradually enhanced since 2012. The forecast accuracies for moderate rainfall event, rainfall in autumn and winter are improved significantly. Moreover, the missed event ratio of large threshold in winter and the false alarmed ratio of large threshold in summer are both reduced. (5) The precipitation predictability of DOGRAFS is generally better than NMC-GRAPES.