In this study, we conduct a statistical verification and analysis on the performance of near-surface temperature, wind speed and precipitation forecasts during the warm season in Zhejiang Province using CMA regional ensemble prediction systems at 3 km and 10 km horizontal resolutions (hereinafter referred to as CMA-REPS 3 km and CMA-REPS 10 km). The results show that both CMA-REPS 3 km and CMA-REPS 10 km can effectively capture the diurnal variations of near-surface meteorological elements, but their forecasts exhibit a negative bias in temperature, a general overestimation of wind speed, and a general overestimation of precipitation from afternoon to early morning. Compared to CMA-REPS 10 km, CMA-REPS 3 km can effectively reduce forecast errors of temperature and wind speed, with a maximum decrease of 23.1% in continuous ranked probability score and an improvement in the area of relative operation characteristic of precipitation forecast by up to 19%. Meanwhile, CMA-REPS 3 km demonstrates more superior fraction skill score for heavy precipitation neighborhood space and provides more accurate forecasts in diurnal variations of near-surface elements. However, the early ensemble spread is smaller in the forecast. In complex terrains, CMA-REPS 3 km shows significant improvements in the probability forecast errors of temperature and wind speed and also in the spread-skill relationships for wind speed in hilly and mountainous and for temperature in plain and basin areas. Not only that, CMA-REPS 3 km has better capability for the ensemble spread of heavy precipitation and the development of rain bands in steep terrain areas. As such, CMA-REPS 3 km has superior forecasting capability for the near-surface elements in complex terrains, particularly for 2 m temperature, 10 m wind speed and precipitation. These findings could serve as a scientific basis for objectively evaluating the two CMA regional ensemble prediction systems under different topographic conditions and their subsequent improvements in ensemble forecasting methods.