一种基于进化策略的气象学反问题求解算法研究

A Study on Meteorological Inverse Problems Solution Based on Evolutionary Strategy

数值天气预报准确性直接取决于好的预报模式和初始场。在预报业务中,依赖观测数据调整初始场和模式参数属气象上的反问题。通过对模式参数识别和初始场调整问题进行等价转化,提出了一种基于进化策略的气象学反问题求解算法。在一维扩散方程和Lorenz-96简单预报模式进行了两类理想数值试验,试验结果表明经过优化后的预报误差均控制在非常小的范围内且预报稳定,从而验证了方法的有效性。

High accuracy of numerical weather prediction （NWP） directly depends on good forecast models and initial conditions. In NWP practice, the adjustments of initial conditions or model parameters according to observational data are meteorological inverse problems. Through the equivalent transformation of the model parameter identification and the adjustment of initial conditions, a new meteorological inverse problems solution based on evolutionary strategy is presented in this paper. Two ideal numerical experiments are implemented by using the simple prediction models of one-dimensional diffusion equation and Lorenz-96. The results show that after the optimization, prediction errors could be controlled in a minor range and the prediction remains stable, which verifies the validity of the method.