Flood Forecasting via the Ensemble Kalman Filter Method Using Merged Satellite and Measured Soil Moisture Data

Flood monitoring in the Chaohe River Basin is crucial for the timely and accurate forecasting of flood flow. Hydrological models used for the simulation of hydrological processes are affected by soil moisture (SM) data and uncertain model parameters. Hence, in this study, measured satellite-based SM data obtained from different spatial scales were merged, and the model state and parameters were updated in real time via the data assimilation method named ensemble Kalman filter. Four different assimilation settings were used for the forecasting of different floods at three hydrological stations in the Chaohe River Basin: flood forecasting without data assimilation (NA case), assimilation of runoff data (AF case), assimilation of runoff and satellite-based soil moisture data (AFWR case), and assimilation of runoff and merged soil moisture data (AFWM case). Compared with NA, the relative error (RE) of small, medium, and large floods decreased from 0.53 to 0.23, 0.35 to 0.16, and 0.34 to 0.12 in the AF case, respectively, indicating that the runoff prediction was significantly improved by the assimilation of runoff data. In the AFWR and AFWM cases, the REs of the small, medium, and large floods also decreased, indicating that the soil moisture data play important roles in the assimilation of medium and small floods. To study the factors affecting the assimilation, the changes in the parameter mean and variance and the number of set samples were analyzed. Our results have important implications for the prediction of different levels of floods and related assimilation processes.