Abstract: Based on the radar data direct assimilation system established by the Chongqing Meteorological Bureau in cooperation with the Center for Analysis and Prediction of Storms of the University of Oklahoma, a set of radar data assimilation and forecasting experiments for a continuous month in the 2020 are carried out by using ensemble Kalman filter (EnKF) assimilation and hybrid ensemble three-dimensional variational (Hybrid Ensemble Three-dimensional Variational, Hybrid En3DV) assimilation with different hybrid weights (with 0%, 20%, 50%, 80% and 100% of static background error covariance). The performance of different direct assimilation methods of radar data on the short-term precipitation is explored. The results of one month continuous experiment indicate that the precipitation forecasting skills of different assimilation experiments show significant difference in the forecast with short leading-time (within 3 h), and then the difference gradually decreases with the increase of the forecast leading-time. In general, the short-term precipitation forecasting skills of EnKF are much better than the others. In the Hybrid En3DV assimilation experiments, the experiment with only the ensemble-based background error covariance performs the best, suggesting that the static background error covariance has no positive effect on radar data assimilation. The results of case study show that with the increase of ensemble-based background error covariance weights, the spin-up time of model decreases, suggesting that the ensemble-based background error covariance can make the variables in assimilation analysis filed more balanced compared with the static background error covariance.