Abstract: The current weather forecasting (WRF) applications on various domestic processor platforms have varying degrees of efficiency problems. In this paper, we took the domestic Sunway processor platform as an example, and analyzed the resource occupation and hotspot functions of the WRF application, optimized it from different dimensions, and validated it on the Sunway 128 core server. The experimental results show that the optimized computation time is four times higher than before the optimization, and the real-time performance is significantly improved. Further analysis indicates that such optimization methods are important references for other scientific and numerical computing performance problems on the same type of platform.