Abstract: Taking the landfall typhoons “Nepartak”, “Nida”, “Meranti ”, and “Megi” in mainland China in 2016 as examples, this study evaluates the ensemble forecasts of typhoon track, landfall time, and landfall location using products from the China Meteorological Administration–Regional Ensemble Prediction System (CMA-REPS). Based on this analysis, a new real-time optimal selection scheme for typhoon track ensemble forecasts, the 0–12 h Minimum Cumulative Track Error Scheme (MCTES), is proposed by integrating operational real-time typhoon positioning data. This scheme is compared to the 12 h Minimum Track Error Scheme (MTES) to provide a reference for more effective operational applications of CMA-REPS typhoon track ensemble forecasts. The results are summarized as follows: (1) Overall, for typhoon track forecasting of landfalling typhoons, the ensemble mean forecast of CMA-REPS is inferior to the control forecast. (2) The MCTES real-time optimal typhoon track ensemble forecast selection scheme significantly improves the typhoon track predictions of both the CMA-REPS ensemble mean and control forecasts, outperforming the MTES scheme, CMA-REPS control forecast, and ensemble mean in sequence. Compared with the control forecast, the MCTES scheme reduced the average absolute errors in typhoon landing time, landing point, and the 12-hourly track movement within the 24–72 h forecast validity time by 0.4 h, 34.3 km, and 21.1 km, respectively. Compared with the ensemble mean, these errors were reduced by 1.1 h, 36.4 km, and 31.7 km, respectively. (3) Integrating the optimal members selected by the MCTES scheme with the control forecast members further reduces the typhoon track forecast errors. Compared with the ensemble mean, the average distance error of the control forecast was reduced by −5.21%–11.51%; the MCTES scheme increased the reduction to 13.42%–19.83%; and the integrated scheme of MCTES and control forecast further increased the reduction to 16.28%–20.83%.