A Research On 4-D Variatonal Data Assimilation Based On Genetic Algorithm

The technique of variational data assimilation has been considered as an effective method in the aspect of improving the quality of the initial fields of numerical weather predication (NWP). However, it does make strong demands on mathematical property of the cost function, In the meantime, it needs higher demand by using local optimization method and depends on the computer resource. Therefore, it's necessary to bring forward a new method, which not only needs less computer resource and weak demands on the cost function, but also will not trap into the local minimum but search the global minimum when the cost function is a multimodal function.In this context a genetic algorithm (GA) was applied to the four-dimension (4-D) variational data assimilation, which provided a kind of new and effective method in the initial field optimization. The theoretical basis and detailed algorithm were introduced in this paper. At the same time, according to the property of the variational problem itself, we designed the rational genetic coding, operators and parameters. In the end, we built a model of the 4-D variational data assimilation for a barotropic primitive equation based on GA and compared with the adjoint assimilation model. The results of the numerical experiments showed that the new assimilation system had achieved the relatively satisfying performance. Consequently, this scheme is feasible and valid. The research enriches the contents of variational assimilation and makes it more extensive application, which increases the quality of initial fields in NWP.