变分四维同化方法若干问题的数值试验

资料误差和模式误差都可能影响变分四维同化的结果，针对这一问题利用浅水方程模式进行了变分四维同化的数值模拟试验。试验分三种情况进行：（１）仅修正初始场，（２）仅修正模式；（３）二者同时修正。试验结果表明，当模式有误差时，传统的变分四维同化方法（仅修正初始场）可能将模式的误差混淆到初始场中去，尽管在同化期间可得到较好的拟合，但却不一定能保证同化后有好的预报。如果不修正初始场而修正模式，当模式误差不大时

THE RESEARCH OF SEVERAL ASPECTS OF FOUR-DIMENSIONAL VARIATIONAL DATA ASSIMILATION

It is possible that both observational data error and model error will influence results through four dimensional variational data assimilation. According to this, some numerical experiments are performed on a shallow water equation model. In these experiments initial data and model error parameter are adjusted simultaneously or sequentially as control variables. It shows when taking initial data as control variables, model error will be aliased to the initial fields if model equations exist error. Although it can be well adjusted during the periods of assimilation, it may not have good forecast after assimilation. But if taking model error parameter as control variables, although model error will varies with time, we can substitute it with constant model error control if the errors in the model have small effect, otherwise, assimilation will fail. It shows when initial data and model correction term adjusted simultaneous as the control variables, perhaps it will be the best, but the choice of scale will influence the results.