Abstract: An extremely heavy rainfall occurred in the Pearl River Delta on 22 May 2020 with the sliding 1 h maximum precipitation of 201.8 mm, and the total rainfall of 351 mm in 3 h. In order to investigate the key forecasting factors and the predictability of this case, the evaluation and sensitivity analysis of precipitation forecasts by the mesoscale ensemble prediction system that is based on the Tropical Regional Atmosphere Model for the South China Sea (CMA-TRAMS(EPS)) are carried out. The results show that compared with the ECMWF Ensemble Prediction System (ECMWF-EPS), the good-performance members of CMA-TRAMS(EPS) have better ability to capture the intensity and spatial distribution of the heavy rainfall, but they still miss the extremity. The better prediction ability of these members comes from their effective prediction of the evolution characteristics of the low vortex and low-level jet (or boundary layer jet) as well as the intensity and location of the coupling between the two systems. The strong southerly wind component of the low-level jet (boundary layer jet) over the eastern part of the Pearl River Delta is conducive to the deceleration of the low vortex moving and the enhancement of cyclonic convergence, both of which are favorable for long duration and high efficiency of precipitation. In addition, the development of the low vortex itself has feedback on changes in the intensity of low-level jet (boundary layer jet). The low-level jet (boundary layer jet) is accelerated in a small area due to the feedback of the enhanced vortex circulation. This coupling mechanism is well described by the good-performance ensemble members. However, the coupling position of the low vortex and low-level jet (or boundary layer jet) predicted by other members is located to the east and south of its actual position, and the convergence intensity is underestimated. The above biases in the simulation lead to the deviation of rainfall intensity and rainfall area. In addition, heavy rainfall results in the formation of cold pool and intensifies the contrast between the surface warm ridge and cold pool (with horizontal temperature gradient of 0.23—0.76℃/km), which is conducive to the maintenance of mesoscale convergence line and strengthens the backward propagation of convection, leading to extreme rainfall. However, all the members of CMA-TRAMS(EPS) have obvious deficiencies in forecasting the organization and propagation characteristics of the mesoscale systems, which limits the ability of CMA-TRAMS(EPS) for predicting extremely heavy rainfall.