To improve the accuracy of runoff simulation, taking the Qinhuai River Basin as an example, data of remotely sensed soil moisture from the SMAP, SMOS and AMSR2 satellites is downscaled with the ensemble averaging method and merged with Kalman filter algorithm based on Grid-Xin'anjiang model. The results of data assimilation on 11 flood events recorded by 3 flow stations in the Qinhuaihe Watershed from 2016 to 2018 showed that, the qualified ratios of the relative errors of flood peak and runoff depth in the daily simulations are 71.43% in calibration period, while the two qualified ratios are 66.67% and 100% in validation period, respectively. After data assimilation, the runoff depth relative errors for the eight flood events are reduced by 29.01% in average. The deterministic coefficients of the eight flood events, ranging between 0.01 and 0.09, are improved with a maximum improvement of 11.84%. Data assimilation on remotely sensed soil moisture can effectively improve the accuracy of soil moisture and runoff simulations.