Abstract: The Weather Research and Forecasting (WRF) model with the hybrid ETKF-3DVAR (ensemble transform Kalman filter-three-dimensional variational data assimilation) data assimilation system is used to investigate the impact of different ensemble generation schemes on a squall line forecast in the Huanghe-Huaihe region in the summer by assimilating the surface observations. Ensembles are created in three different ways-by using the different initial ensemble samples (RCV), by using different model physical process schemes (PPMP), and by combining the first two ensembles (BLE). Based on the increments of model states, assimilating the surface observations mainly adjusts the spatial structures of wind and water vapor mixing ratio below the level of 850 hPa; the RCV scheme mainly updates the wind distribution, the PPMP scheme updates the water vapor mixing ratio, and the BLE scheme has the characteristics of both RCV and PPMP. Assimilating surface observations can also improve 6-h precipitation forecasts, and the PPMP scheme can give a relatively better performance compared to PPMP and BLE, especially for the prediction of rainfall location and intensity. RCV and BLE schemes present a better simulation for the bow echo, and the performance with BLE is similar to that with RCV. PPMP and RCV schemes can adjust the position and intensity of the cold pool, and also influence the times of appearance and disappearance of the squall line. Generally, PPMP scheme has a greater impact on the squall line than RCV and BLE.