Abstract: Improving numerical weather prediction depends on enhancing the model’s physical processes, initial fields, and spatial resolution. Among these, improving model resolution has proven to be an effective method for boosting prediction accuracy. Using the global forecast system, this study compares two models with different resolutions, T1534 (13 km) and T254 (55 km), to forecast temperature, pressure, wind speed, and precipitation across China. The results show that in simulating daily air pressure over China, the models perform best in North China. Across seven subregions, the root mean square error decreases significantly as the resolution increases. For daily temperature simulations, the eastern region shows better accuracy compared to the western region. However, when simulating wind speed in the northwest region, the root mean square error increases slightly with higher resolution. Overall, model variables with strong periodic changes, such as air pressure and temperature, are better simulated than those with weaker periodic changes, like wind speed. This difference is primarily attributed to the localized and terrain-sensitive nature of wind speed. A heavy precipitation event in Shandong on 10 August 2019, demonstrated that the two models with different resolutions effectively simulated the precipitation characteristics and matched the observed precipitation falling area. However, the high-resolution model exhibited greater accuracy, with lower deviation scores across various precipitation grades. In this heavy precipitation process, relative humidity was close to saturation, creating favorable conditions for condensation. The high-resolution model captured this more precisely, showing enhanced relative humidity and finer structural detail. Meanwhile, the central pressure of the low-level cyclone simulated by the high-resolution model was lower than that of the low-resolution model. This resulted in a stronger cyclone and a more intense convective precipitation process.