A flood forecasting model for small and medium-sized rivers, based on the short-term nowcasting and fine ensemble gridded precipitation forecasts, is established for increasing the flood foresight period. The model adopts the percentile correction method to develop the mapping relationship between the precipitation forecast field of numerical model and the observed field. Based on the GRAPES-3KM model and the Time-Lag-Ensemble fusion technique, a short-term ensemble precipitation forecasting that is consist of three members(the optimal, maximum(95% quantile)and minimum(5% quantile))is developed with the Bayesian model. Taking the developed ensemble precipitation forecasts as the driving force of GMKHM, the hourly real-time rolling forecasting of flood for small to medium-sized basin is performed. The Tunxi Basin of the Xin’anjiang River is selected as the experimental basin to test the real-time flood forecasting in 2020 flood season. Results show that the developed model performed well, the peak discharge of the Tunxi hydrological station was forecasted with 7 hours in advance, the relative error was 5. 6%, and the peak time difference was -1 hour. Compared with that without considering the precipitation in lead-time period, the flood forecast lead-time can be increased by 4 hours with the optional precipitation forecasts and 13 hours with the maximum and minimum precipitation forecasts. The developed model has certain reference significance for the flood forecast of similar basin. The span between maximum and minimum forecasts presented the trend of decreasing gradually since 9: 00 on July 7th. It is of great significance to improve the flood risk prevention and control ability of small and medium-sized rivers with introducing the ensemble nowcasting and short-term precipitation forecasts.