Abstract: This article simulated a weather developing process in the agro-pastoral transitional zone of Inner Mongolia from 1800 UTC 23 July to 1800 UTC 27 July, 2001 with mesoscale atmospheric numerical model ARPS (Advanced Regional Prediction System). To understand the change of mesoscale numerical models ability in atmospheric multi-scale movement simulation, especially for smaller space-time scale motion at different model resolutions, the authors analyzed simulation results of u, v, w, θ, p for different model resolutions. The rank sum test and Analysis of variance (ANOVA) show that the model resolution improvement can affect the five variables simulation to different extents. The impact on p, θ and w is the greatest, the impact on u and v takes second place. The lower troposphere and the boundary layer are the most sensitive to the resolution, followed by the upper troposphere. The spectrum analysis and wavelet analysis show that enhancing resolution can improve the model's ability in the atmospheric multi-scale characteristics simulation and the improvement is different for different atmospheric layers and meteorological elements. In general, the simulation effect in upper layers is better than in lower layers, and the simulation effect of variables with strong cycle signals is better than that with weak cycle signals.