三维变分系统在我国夏季降雨预测中的应用试验

The 3DVAR Application Experiments on Rainfall Prediction in the Summer in China

采用MM5模式及其三维变分系统(MM5/3DVAR)对我国夏季降雨进行了一个月的连续预测试验,并对试验结果进行评估。试验中首先采用"National Meteorological Center(NMC)"方法,将2005年8月的MM5模式的预测结果形成与试验区域和水平分辨率相匹配的背景误差场,并将其与全球背景误差场进行了对比分析,结果表明,采用2005年8月MM5模式预报结果生成的背景误差场的基本特征与系统提供的全球背景误差场相似,且长度尺度随着水平分辨率的提高而减小。之后,分别利用NCEP再分析资料(NCEP试验)、NCEP再分析资料基础上采用CRESSMAN方法分析观测资料(LITT试验)和NCEP再分析资料基础上采用3DVAR系统同化观测资料(3DVAR试验)形成模式预报初始场,再次对2005年8月降雨进行逐日连续预报。逐日降雨预报结果表明,相对NCEP试验,LITT试验中1和10 mm的预报评分有明显提高,但25和50 mm的预报评分却有所下降,而3DVAR试验的预报评分在10 mm以上均有明显提高。对于降雨期间的形势场预报,3个试验中,除温度场和湿度场外,其他变量场的均方根误差随高度增加而增加,但相比而言,3DVAR试验的均方根误差小于其他2个试验。3DVAR试验对降雨的明显改进,可能是因为其对与背景场信息差别比较大的反应中尺度系统的观测资料的分析结果比较靠近观测资料。

The MM5 modeling system and its three dimensional variation assimilation system （MM5/3DVAR） are employed to prediction rainfall during August 2005 in China, and then, the prediction results are analyzed. The background error used by 3DVAR system was reproduced by using ＂National Meteorological Center （NMC）＂ meth- od and the prediction results of August 2005, which adapt to the prediction domains and horizontal, vertical resolution in the following experiments. The characteristics of reproduced background error are very similar to the global background error, and the length-scale of background error decreases with horizontal resolution. Three experiments （NCEP, LITT, 3DVAR Experiments） are designed to analyze the impacts of 3DVAR system. NCEP Exp. only uses NCEP data to form initial field. The initial field of LITT and 3DVAR Experiments utilize CRESSMAN and 3DVAR method to analyze surface and sounding data on the base of NCEP data. Compare to the results of NCEP Exp. , the LITT Exp. improves the 1 mm and 10 mm rainfall prediction, but 25 mm and 50 mm rainfall prediction is not improved. 3DVAR Exp. improves 10 mm, 25 mm and 50 mm rainfall prediction significantly. The RMSE （Root Mean Square Error） of synoptic field prediction, except for temperature and relative humidity, in three experiments increase with height generally. All of the prediction of 3DVAR Exp. is the best among the three experiments. The significant improvement of 3DVAR Exp. maybe caused by that the analysis of mesoscale information by 3DVAR is more similar to observational data.