A Research Of MM4 Genetic Algorithm 4-D Variatonal Data Assimilation System And Its Numerical Experiments

The technique of adjoint 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 same time, by use of descend algorithm to get the gradient of cost function, it is easy to fall into the trap of local optimization. Genetic algorithm (GA), as a rising theory, which demands none or little on mathematical property of the cost function, has better ability of finding the global minimum, the character of universal application and easier expansion. Therefore, it's necessary to make the technique of variational data assimilation have more comprehensive application in research and practical operation, by merging GA into variational data assimilation, applying it during the course of gaining the global optimized initial fields of NWP model to enhance the accuracy of NWP model.In this paper, GA is combined with the mesoscale NWP model MM4 to solve the puzzle of adjoint variational data assimilation, and four-dimension (4-D) variational data assimilation based on GA is established. The theoretical basis and detail algorithm are also introduced in this paper. At the mean time, according to the property of the variational problem itself, rational genetic operators and genetic parameters are designed. In the end, numerical experiments of ideal field, real observed data are carried out on the GA 4-D variational data assimilation system. Results show GA 4-D variational data assimilation system has achieved the relatively satisfying performance. It also has strong ability of filtering and smoothing on initialfields random errors, and it is able to release information of observed data, which is absorbed during assimilation window, when forecast model is running. Consequently, it is pressing feasible and effective to merge GA into 4-D variational data assimilation of NWP model. This scheme expends the content of GA variational data assimilation system. The GA 4-D variational data assimilation system can provide more precise quality initial fields, thus enhance the forecast accuracy of NWP model, eventually make the technique of GA receive further application.