Abstract: A series of forecast experiments on a heavy Mei-yu precipitation process in July 1991 are performed to reduce the spin-up error of the limited-area numerical weather prediction model by assimilating observation data in four dimensions. Two schemes of assimilation are designed to better initial states for the improvement of the model forecast in the early stage; one is Newtonian relaxation by nudging which is used for assimilating the observations of temperature and wind fields; the other is humidity-modifying scheme for assimilating precipitation observations. The results show: 1) the former is characterized by the initial state with large-scale weather situation and the initial field gains meso-scale information; 2) the latter develops low-pressure system and the initial field gains more and better mesoscale information; 3) As a result, the incorporation of two schemes improves the early 6h precipitation prediction. Newtonian relaxation by nudging is more important to the Mei-yu weather process by forming the large-scale environmental conditions.

Key words: data assimilation, precipitation prediction, spin-up problem

摘要： 选取1991年7月3日至4日出现在我国江淮流域的梅雨暴雨天气过程，针对有限区模式降水预报中“旋转加强”问题，试验2种初值化方法即牛顿连续松弛逼近技术和降水-湿度场调整方案来改进模式的初期降水预报。试验结果表明：(1) 用湿度场调整方案同化观测降水资料，使同化时段降水区的中低层低压系统明显加强，所形成的初值具有较好的中尺度信息；但该方案会出现观测值与模式不协调引起的虚假小扰动。(2) 用牛顿连续松弛逼近技术同化业务观测系统的速度场和温度场资料，所形成的初值场形成了大尺度特征，同时对应江淮雨带的中低层低压系统加强，所形成的初值能够具有一定的中尺度信息。(3) 因此两者结合得到了最好的预报效果。对于梅雨天气过程，牛顿连续松弛逼近技术更有效。

关键词: 资料同化, 降水预报, “旋转加强”问题