Effect of Surface Soil Moisture Assimilation on SWAT Model Output

The Ensemble Kalman Filter(EnKF) was coupled with a watershed scale distributed hydrologic model, the Soil and Water Assessment Tool (SWAT) to improve predictions of soil water content and thus enhance overall model performance of the hydrologic system by assimilating spatially distributed measured surface soil moisture. The SWAT model is based on the concept of Hydrologic Response Units (HRU) and has been widely applied to many different areas of watershed scale modeling. However, there is a lack of investigative research as to how the spatial variability of inputs, especially from remotely sensed surface soil moisture data, affects the potential capability of data assimilation techniques with SWAT. In this study, a synthetic experiment is performed to better understand how soil moisture data assimilation affects various hydrologic processes in the model at the watershed scale. The study area for this work is the Upper Cedar Creek Watershed (UCCW) which is located in northeastern Indiana. There are two sources of rainfall data available for the UCCW: the National Climatic Data Center (NCDC) and hydrometerological network maintained by the USDA, Agricultural Research Service National Soil Erosion Research Laboratory (NSERL) to measure precipitation and soil moisture data. First, the “true” state is implemented by running the model with all available rainfall data from the NCDC and the NSERL raingauge network. To represent our imperfect knowledge of the true hydrologic processes, subsequently the model is run with an intentionally poor set of initial conditions and “limited” forcing data from only NCDC raingauges for the same time period. By limiting precipitation input, which is the driving force of soil moisture and streamflow, while keeping other model parameters unchanged, we determine how the updated soil water condition with surface measured soil moisture influences model predictions of profile soil water content, runoff and streamflow. Results show that daily assimilation of surface soil moisture for each HRU improves model predictions especially by reducing the overestimated streamflow due to the errors associated with insufficient spatially distributed rainfall input. Improved daily streamflow exceedance curve and statistical measures (the coefficient of determination and the Nash-Sutcliffe efficiency) demonstrate that better representation of soil water content through surface soil moisture assimilation can enhance the rainfall-runoff processes in the SWAT model. Distributed errors of the soil water content are also illustrated to show the influences of spatial variability.