Abstract: ECMWF and GRAPES (Global/Regional Assimilation and Prediction System) forecast products are the main service products in China. To understand their performance and enable users to selectively apply these products according to their needs in practical application, this study evaluates the applicability of air temperature, ground temperature, wind speed, and relative humidity from ECMWF (C1D), GRAPES_MESO (Meso), and GRAPES_GFS (Gfs) in July 2017, November 2017, January 2018, and April 2018 and these models are compared with automatic observations from 2421 national stations and 8155 backbone stations reported by the Chinese Meteorological Administration. Results show that systematic errors are observed for the three numerical models compared with the in situ observations. The ground temperature prediction is easy to underestimate, and the wind speed forecast is easy to overestimate. There are obvious regional, seasonal, and diurnal variations in the forecasting capability of the three numerical models, which is evidently lower in the Tibet area than that in other areas. The forecasting capability for the air temperature and wind speed is the worst in spring, while that for humidity is the best in summer using the three models. For the analyzed meteorological variables, the correlation coefficient of the wind speed is the lowest, that of air temperature is the highest, and the accuracy of humidity prediction is the lowest. The accuracy of ground temperature prediction using Meso is the highest, and the accuracy of wind speed prediction using Gfs and C1D is the highest.