Assimilating Remotely Sensed Surface Soil Moisture into SWAT Using Ensemble Kalman Filter

Abstract

In this study, a 1-D Ensemble Kalman Filter (EnKF) has been used to update the soil moisture states of the Soil and Water Assessment Tool (SWAT) model predictions. Experiments were conducted for the Cobb Creek Watershed in southeastern Oklahoma for 2006–2008. Assimilation of in situ data proved limited success in the top layers only, due to the weak surface and root-zone coupling of SWAT. Negative impact on stream flow and especially storm-scale runoff result from degraded deep-layer soil moisture prediction. This indicates that although the model is calibrated for stream flow, the parameters are not tuned to represent the realistic physical coupling of the system. Therefore the improved surface soil moisture failed to produce better runoff and flow estimates. Synthetic twin study proved the potential of assimilating remotely sensed surface soil moisture in improving SWAT's hydrologic predictions (root-zone soil moisture, evapotranpiration, surface runoff and stream flow) with EnKF. However, perturbations used to generate ensemble trajectories could result in systematic biases in model states and degradation of certain variables that cannot be corrected with EnKF. The mechanisms causing such biases and the methods to correct them need to be further investigated.