In this paper, we use a nested grid model under complex topography and ECMWF global data with resolution of 2.5°×2.5° grid size to perform 48 hour forecasting experiment of coarsemesh and fine-mesh for the case of "81.7" heavy rain in Sichuan province. According to the principle of filtering we separate the fields of wind, geopotential height and temperature using optimum high-pass filter. The results show that the model can fairly predict the generation and development of meso-scale system causing this heavy rain. We have also roughly analyzed the dynamic and thermal effects. This model can fairly described the weather of heavy rain. In the other hand, it is also showed that the fine-mesh forecasting has been made further improvement in some aspect. The scale separative method can clearly separate the vortex in the southwest of China and other meso-scale system from large scale foundational field and make us further to recognize the effect of the meso-scale system during the early stage of disturbance on this heavy rain.