To investigate the effect of the ensemble transform Kalman filter with rescaling (ETKF_R) initial perturbation method on typhoon forecasting, we perform retrospective experiments from 18 to 29 July 2021 using China Meteorological Administration-Regional Ensemble Prediction System (CMA-REPS). The effect of the initial perturbation structure on the track and intensity forecasts of Typhoon In-Fa is analyzed and compared with the ECMWF and NCEP global ensemble forecasts. The results can be summarized as follows. The ETKF_R method improves the amplitude and structure of initial three-dimensional wind field perturbations, but the initial ensemble spreads of typhoon location and intensity are small. By reasonably reducing the ensemble spread of weather system which significantly influences the forecast of typhoon track, ETKF_R can constrain the excessive dispersion of typhoon translation speed and direction. This further improves the ensemble mean track forecast skill for the whole life of Typhoon In-Fa and the relationship between ensemble mean error and ensemble spread of typhoon track. In ETKF_R, the ensemble spreads of typhoon structure and intensity grow rapidly in the first 24 h, and the performance of ensemble mean intensity forecast after 24 h is comparable to that of the ETKF method without rescaling. Compared with the international advanced global ensemble forecasts, ETKF_R has the best landfalling forecast of Typhoon In-Fa. The statistically averaged 0-2 d track forecast error of ETKF_R is comparable to that of ECMWF ensemble. Although NCEP ensemble has the smallest 0-2 d track forecast error, its overdispersed feature is obvious. Meanwhile, ECMWF ensemble generally underestimates the intensity of Typhoon In-Fa, while NCEP ensemble has a high accuracy in predicting the maximum intensity of Typhoon In-Fa, with a slower intensification speed than ETKF_R. Our results suggest that the forecast of typhoon track and intensity by CMA-REPS has operational significance of reference.