One Scheme Which Maybe Improve The Forecasting Ability Of The Global (Regional) Assimilation And Prediction System

The application of the historical data in the numerical weather prediction is studied in this dissertation. Numerical weather prediction can be summarized as an initial value problem of differential equation. However, in operational weather prediction recent his-torical data is widely used. To solve the conflicts between these two predictive strategies, one new scheme to use historical data in the numerical weather prediction is put forward.For the mathematical scheme to this problem, the prediction equation may be linear differential equation or nonlinear differential equation. For the linear differential equa-tion, the analytical solution is gotten using the theory of eigenvalue problem, while for the nonlinear differential equation, numerical solution is gotten using the numerical method. Upon the variational principle and Euler equation, the existence of solution of the nonlin-ear differential equation is discussed.The skeleton of the new scheme is as following:The initial value problem of dif-ferential equation of the numerical weather prediction is firstly converted into inverse problem of differential equation. And the ill-posed inverse problem is then transformed to well-posed problem by applying mathematical skills. Finally, the improvement of the numerical model is integrated as the solution of a polynomial-form error modification function whose coefficients can be determined by applying historical data. The using of the method is simple, operation-friendly and good portability, and can be applied with the improvement of the numerical model synchronously.A series of ideal experiments on a simple oscillation equation are made to demon-strate the new scheme. Five specific aspects are tested:1) The modification effect of the error modification function;2) The relationship between the number of historical data and the modification effect;3) The predictable period in the modification prediction;4) The recognition of the method for the high and low frequency error signal;5) The relationship between intensifying dense observation and the improvement of the modification prediction.The results show that the new method can reduce numerical model error, especially for low frequency error signal. And the method is operation-friendly, for it only calls the subprogram of the tendency instead of involving the kernel of the numerical model. This method can also be used on the GRAPES, and may improve the model's predictive skill.