Improving Soil Salinity Simulation by Assimilating Electromagnetic Induction Data into HYDRUS Model Using Ensemble Kalman Filter

Assimilation of proximally and remotely sensed information on soil salinization-related attributes into a hydrological model is essential to improve the forecast performance of the profiled soil salinity dynamics for developing appropriate soil amendment practices. Although the family of ensemble Kalman filters (EnKF) is widely used in data assimilation, their applicability and reliability for soil salinization estimation requires further experimental validation. Here, we evaluated the assimilation performance of apparent electrical conductivity (ECa) data obtained from an electromagnetic induction meter (EM38) into the HYDRUS hydrological model. Re-sults showed that the EnKF method improved the simulation accuracy of soil salinity at 0 ~ 100 cm soil depths, as indicated by the de-creased root-mean-square error of 32.6 ~ 76.7% and increased Nash–Sutcliffe efficiency of 9.6 ~ 71.2%. The HYDRUS-simulated values with EnKF were closer to the measured values than the values simulated by the HYDRUS model, and this benefitted from updating the running trajectory of the HYDRUS model. The EnKF values derived from measured ECa data were better than HYDRUS-simulated val-ues with EnKF. Soil salinity simulation was sensitive to ensemble size, error level, and ECa data depth. Considering the ensemble repre-sentativeness and computational efficiency, the optimal ensemble size was judged to be 50. The maximum acceptable observation error was 10%, and observation data to a depth of 100 cm was suggested in EnKF assimilation to minimize the root-mean-square error. It was concluded that proximally sensed EM38 data coupled with the EnKF algorithm is promising for improving the simulation performance and providing a prospective method for simulating large-scale ecological and hydrological processes by coupling multi-source data and hydrological models.