Abstract: Based on the real-time observation data and the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble forecast data, this study analyzes the heavy precipitation in Guangxi associated with typhoon Mangkhut (1822) and the key systems of different precipitation stages influence on prediction bias by using the ensemble sensitivity method. Results show that: (1) the precipitation associated with typhoon Mangkhut can be divided into three stages, corresponding to different rain areas, the key system is typhoon noumenon, interaction between typhoon periphery and cold air and the southerly flow of typhoon, respectively. (2) ensemble sensitivity analysis indicates that the most sensitive system to the first precipitation stage of the process is typhoon’s location. The more northwestward typhoon is, the stronger the precipitation. In the second stage, the sensitive system is typhoon’s location, subtropical high ridge area and ground cold high pressure, while the most sensitive system is the subtropical high ridge point. The third stage is sensitive to tropical depression’s location and intensity, subtropical high ridge area and ground cold high pressure, which is most sensitive to ground cold high pressure. (3) Comparing best ensemble members and ensemble average, we can find that the difference in the first precipitation stage mainly depends on the typhoon’s intensity and its location. The difference in the second stage is related to typhoon’s location and intensity of the ground cold high pressure that located in the northern of the typhoon. The difference in the third stage mainly corresponding to the tropical depression and the intensity of ground cold high ridge that located in the southern of subtropical high.