Warm-sector rainfall event under the background of weak synoptic-scale forcing is difficult to predict accurately due to its abruptness and nonlinear, thus ensemble forecasts have become one of the crucial options considering uncertain factors. However, the core problem of convection-allowing ensemble forecast is that the spread is low after integration for a period, which will lead to prediction failure. Therefore, this paper compares the differences among the convection-allowing ensemble forecast schemes with different scale information for perturbations, and optimizes the initial perturbation scheme for the warm-sector rainstorm over the middle and lower reaches of the Yangtze River. Four convection-allowing ensemble forecast experiments with different initial perturbation schemes including dynamic downscaling (DOWN), breeding of the growing mode (BGM), Local BGM (LBGM), and BLEND, were carried out for a typical warm-sector rainfall on May 4-5, 2018. The aim is to explore the impact on spread and forecast. The results show that the ensemble forecast results of LBGM and BGM are better than that of DOWN in 0~6 h in the early stage of model integration, and LBGM has a certain degree of improvement compared with BGM, which indicates that accurate small and medium scale perturbations can obtain effective growth at this stage. After 12 hours of integration, the forecast effect of DOWN is better than that of BGM and LBGM instead, which indicates that after the initial error increases rapidly for a period, the large-scale perturbations begin to play a major role. However, BLEND possesses the advantages of both LBGM and DOWN, and has a good forecast effect in almost the whole forecast period, reflecting the superiority of multi-scale blend initial condition perturbations. Since the convection-allowing ensemble forecast for this type of events are sensitive to the scale characteristics of the initial condition perturbations, adjusting the scale of the initial condition perturbation has certain significance in operation for the development of high-quality convection-allowing ensemble forecast system. [ABSTRACT FROM AUTHOR]