Based on the high resolution integrated forecast system of RISE, using the meso-scale CMA-BJ model with 3km resolution and the global-scale ECMWF model with 0.125 degree resolution, adopting the hourly observation data of conventional and dense automatic weather stations, taking the outdoor mountainous competition area in Beijing Winter Olympics as the research area, this study compares the effects of different numerical weather prediction (NWP) model background on the accuracy of surface 2-m temperature and 10-m wind analysis and hourly forecast for the future 1-24h with 100m grid resolution. The results show that: (1) Both the regional and global models can be used as the background of RISE system to effectively form 100-meter fine integrated products in complex terrain, but the impacts of different NWP model backgrounds on the analysis and forecast performance for different meteorological elements are obviously different; (2) For temperature analysis, the model background has the least influence. The spatial distribution of RISE temperature analysis with CAM-BJ and ECMWF as the NWP model background is basically the same, and the MAE error is both less than 0.2oC; (3) For wind analysis, adopting high-resolution regional model can improve the refinement level of RISE high-precision wind integrated products better than the global model with coarse resolution; (4) For temperature forecast, the performance of RISE 100m grid forecasts with ECMWF model background is significantly better than that with CMA-BJ model background, and the average forecast errors for Winter Olympic alpine stations and all stations are reduced by 10.5% and 7.0%, respectively. (5) For wind speed forecast, the MAE errors of 1-6h forecast for RISE Winter Olympics alpine stations with CAM-BJ and ECMWF model backgrounds are 1.42m/s and 1.30m/s, respectively, and the MAE errors of 7-24h forecast are 1.52m/s and 1.54m/s, respectively. Besides, the average 1-24h MAE errors for all stations in RISE region are 1.38m/s and 1.24m/s, respectively. The results in this study are helpful to further understand the role of model background in the 100-meter-level integrated forecast, and have important scientific significance and practical value for improving the accuracy of weather forecast in complex terrain.