Abstract: The devastating rainstorm that hit Zhengzhou on July 20, 2021, resulted in substantial losses of life and property. This storm was extraordinarily intense and presented considerable challenges in numerical prediction owing to its inherent uncertainty. To improve the ability of the China Meteorological Administration’s (CMA) numerical model to predict such rainfall intensity, we used a convective-scale numerical forecast system which is called CMA-Meso with a 3-km horizontal resolution. We designed an ensemble assimilation observation disturbance scheme and adjusted the cloud analysis radar reflectivity filtering threshold. This allowed us to conduct convective-scale ensemble forecast experiments. Then, we evaluated the uncertainty characteristics of extreme rainfall intensity forecasts from the convective-scale ensemble forecast. These characteristics were compared to those obtained from the CMA global ensemble forecast and regional ensemble forecast. The following are the results obtained from this study. (1) While CMA-Meso 3km still exhibited some deviation in predicting extreme precipitation, there was a good spread in the extreme rainfall intensity values among different ensemble forecast members. Some values of individual members were close to the actual 24-hour cumulative precipitation of 624.1 mm, providing a better representation of the forecast uncertainty of extreme precipitation. (2) Further comparison of the CMA ensemble forecasts at different resolutions showed a strong correlation among the intensity of extreme rainfall, dispersion, and the skill scores derived from assessing the fraction of cumulative precipitation as well as the model’s horizontal resolution. The higher the resolution, the higher the ability to accurately predict extreme rainfall intensity. This indicates that the convective-scale ensemble forecast can effectively describe the uncertainty and extremeness of extreme rainfall intensity forecasts. (3) When the convective-scale ensemble forecast simultaneously disturbed the conventional observation and radar data, it affected the model’s water vapor field and circulation situation in a short forecast period. This promoted the development of precipitation forecast ensemble dispersion during the same period and effectively improved the probability forecast ability of extreme rainfall intensity. In conclusion, the convective-scale ensemble forecast that disturbs conventional observation and radar data could help enhance our ability to accurately predict the intensity of extreme rainfall.