A time lag exists in the response of observed variables to state variables in surface hydrological process. To improve the accuracy of streamflow data assimilation, a streamflow data assimilation scheme was constructed based on the ensemble Kalman smoother (EnKS) and SWAT model for the Minjiang River Basin, which was then compared with the ensemble Kalman filter (EnKF) method, so as to evaluate the accuracy of different assimilation methods and to analyze the effect of data assimilation on different streamflow components. The results indicate that the optimal time window length of EnKS differs in each hydrological period and basin, and the consideration of the time lag of the hydrological model can effectively improve the accuracy of model assimilation. Compared with the EnKF method, the NSE of the EnKS method increased by 0.03,0.12,0.03 at the Qilijie station, Shaxian station and Zhuqi station, respectively; whereas the RMSE of the three stations decreased by 7.43%, 26.81% and 4.25%, respectively. There is spatial and temporal heterogeneity in the improvement effect of data assimilation method on different streamflow components. The EnKS method can significantly improve the accuracy of the lateral flow in regions with high-permeability soil and steep slopes, and it has a better performance in wet season than dry season for the improvement of surface flow prediction.