Variations in the initial structure of tropical cyclones (TCs) inevitably affect prediction results; however, the bogus model cannot accurately describe the structure of a weak tropical cyclone with increased initial field resolution. This study aims to construct a model to improve the prediction of weak TC in southern China. Based on the ECMWF 0.1° analysis data, several vortices were filtered out from tropical depressions and tropical storms in 2018 and 2019 to rep-resent a weak TC reservoir in the South China Sea. For different simulation objects, filtered vortices were combined with the TC environmental field to form ensemble members. The observed TC information was assimilated for simu-lating TCs Bebinca, Mun, and Ewiniar to verify the feasibility of the proposed model, based on the Global/Regional Assimilation and Prediction Enhanced System (GRAPES) 9-km model developed by the Guangzhou Institute of Tropical and Marine Meteorology. The results show that the initialization scheme of the weak tropical cyclone model improved the intensity prediction of the TC by 26.81％ (Bebinca), 18.65％ (Mun), and 47.00％ (Ewiniar), compared with the control experiment. Because typhoon intensity forecasting has not notably improved for many years, this scheme has certain scientific and operational significance.