Abstract: The initial perturbation approach based on the subtropical singular vectors has been applied in the Global and Regional Assimilation PrEdiction System (GRAPES) global ensemble forecast system. However, the tropical cyclone track ensemble spread is not enough for this system. It is found that the initial perturbation near the tropical cyclone is not reasonable. As a result, the forecast ensemble could not well estiminate the forecast uncertainty of the tropical cyclone, which may be one reason for the small track ensemble spread. Through computing the tropical cyclone singular vectors (TCSVs) and linearly combining the TCSVs and subtropical singular vectors to generate initial perturbation, the problem of almost zero initial perturbation near the tropical cyclone is solved, and the tropical cyclone ensemble forecasts are improved. Based on the computation scheme for the GRAPES global singular vectors, the square area around the typhoon center with the side length of 10 degree is identified as the targeted area to construct the computation scheme for TCSVs. The ensemble forecasts experiments are conducted for the typical typhoon Banyan and other TC cases, and the best typhoon track at the National Meteorological Center of China Meteorological Administration (CMA) and precipitation observation data at National Meteorological Information Center of CMA are used to verify the forecasts. The structure characters of the TCSVs and the initial perturbations are analyzed, and the influence of TCSVs on the tropical cyclone track ensemble forecasts and 24 h accumulated precipitation probability forecast skill for the China area is analyzed. The experiments show that TCSVs are localized. When TCSVs are used to generate initial perturbations, the tropical cyclone track ensemble spread is increased; the relation between the track ensemble spread and the ensemble mean forecast error is improved; the track ensemble mean forecast error is decreased; the ensemble members can better describe the forecast uncertainty of the tropical cyclone track; the probability forecast skill of 24 h accumulated precipitation in China for the light rain, moderate rain, heavy rain and rainstorm is increased. In a word, the use of TCSVs is effective for improving the initial perturbations and may help improve the tropical cyclone track ensemble forecasts when there exists one or more TCs.