Abstract:The core competition zone of the Winter Olympic Games is mostly located in the complex terrain area, and the surface meteorological field is greatly affected by the terrain which shows high heterogeneity. Therefore, the range represented by the observation data in the complex terrain area is highly limited. In order to understand the structure characteristics of the low-level surface meteorological field in the competition area in detail, the high-resolution numerical simulation method becomes very important. In this study, to meet the Fine meteorological service requirements in 100 m/minutes level which proposed by the Organizing Committee for the Beijing Winter Olympics, a real-time forecast system with 100 meters horizontal grid spacing "RMAPS-LES" which coupled with mesoscale meteorological model and large eddy simulation was especially built for Yanqing and Zhangjiakou competition zones. Based on the observation data of Winter Olympics zones and the 3×3 km horizontal grid spacing from "CMA-BJ" mesoscale forecast system, the “RMAPS-LES” results of the event months (February to March) from 2020 to 2022 were compared and analyzed at different space-time scales. The results show that the high-resolution microscale “RMAPS-LES” has a higher prediction skill than the mesoscale model. The Mean absolute error(MAE) of average hourly 2 m temperature of the whole winter Olympic site is 1.85 ℃, which is 28% less than the mesoscale model. The MAE of the hourly average 10 m wind speed is 2.11 m ? s-1, which is 23.6% less. The instantaneous wind direction MAE is 44.43 degree, which is 30.33% lower. Compared with the mesoscale model, “RMAPS-LES” has the advantage of obtaining high-frequency data that can describe the atmospheric micro scale turbulent movement in detail. The results of evaluation show that the probability density distribution of wind speed fluctuation is basically close to the normal distribution, and there are relatively small differences between the average value and the standard deviation between the observations. It is also found that when dynamic downscaling is used, inflow turbulence generation method needs to be used between mesoscale and microscale to accelerate the generation of atmospheric turbulence in the large eddy simulation region. The "RMAPS-LES" used the generalized cell perturbation method, which employs a novel stochastic approach based upon finite amplitude perturbations of the potential temperature field applied within a region near the inflow boundaries of the LES domain, which substantially reduced the fetch to achieve fully developed turbulence. After proper improvement, the technical route of “RMAPS-LES” can also be widely used in the Wind Power Generators Selection，forest fires，outdoor activities，and other scenes that have high requirements for fine local meteorological forecast.