基于TIGGE资料的沂沭河流域6小时降水集合预报能力分析

全球多模式集合预报(TIGGE)资料为发展局地水文风险预报方法提供了新基础。对不同预报系统的集合预报资料进行评价与对比,可为综合应用多源资料实现超集合预报提供参考。本文以沂沭河流域内10个站点观测降水作为参照,对2007～2010年7、8、9月中BABJ(北京)、ECMF(欧洲)、EGRR(英国)、RJTD(日本)和KWBC(美国NCEP)五种预报模式的6h集合预报降水做了相关系数、均方根误差、Nash效率系数、TS评分(风险评分)和Brier评分等定量评估和对比。对于各模式集合平均预报,EGRR表现最好,4日预见期内的相关系数达0.48,Nash系数为0.21,BABJ最差,其他三模式预报能力相当。对于确定的控制性预报,4日预见期内RJTD表现最优,相关系数为0.19,Nash系数为0.13,其次为BABJ和EGRR。各模式集合平均与控制性预报相比,预报能力都占绝对优势,而多模式集合平均其预报能力又强于任何单模式集合平均。在4日预见期内,多模式平均的相关系数达0.49,Nash系数达0.24。在不同百分位阈值下TS评分和Brier评分也表明了类似的各模式评比结果,但多模式平均虽然在较低阈值下评分较优,但不占据绝对优势。各中心资料均具有一个随预见时长增加的稳定衰减期,其中EGRR衰减期最长(达9天)且最为稳定,而其他资料则存在不同稳定程度的衰减,稳定衰减期都能持续4天以上。各中心资料对较大降水的预报还存在各自不同的系统性偏差。

Predictability of 6-Hour Precipitation in the Yishu River Basin Based on TIGGE Data

TIGGE offers an opportunity to develop methods of applying global ensemble predictions systems (EPSs) from different models to improve the predictions of local hydrological risks. Comparative assessments of data from different EPSs improve the application of the grand ensemble forecasts from multiple sources. Based on the observed rainfall records from 10 stations in the Yishu River basin during August 2007, ensemble forecasts of 6-h precipitation from 5 EPSs, i.e., BABJ (Beijing), ECMF (ECWMF), EGRR (UKMO), RJTD (Japan), and KWBC (NCEP), are compared and assessed using several quantitative methods. Comparing the ensemble mean precipitation rates of all the EPS, EGRR scored highest with a correlation coefficient (R) of 0.48 and Nash-Sutcliffe efficiency (NE) of 0.21 in the lead time of four days; BABJ scored the lowest. For the control predictions of all the EPSs, RJTD scored highest with R = 0.19 and NE = 0.13 in a lead time of four days, followed by BABJ and EGRR. Compared with the control predictions, the ensemble mean of each EPS showed better performance. The multimodal ensemble mean showed a further improvement of the prediction skill. In the lead time of four days the multimodal ensemble mean had an R = 0.49 and NE=0.24. For different threshold values, the analysis of the threat score (TS) and Brier score (BS) showed similar comparative assessment. When the prediction lead time was increased, all the EPS showed a stable decline of prediction skills, with EGRR showing the longest and most stable decline period (9 days).