The utility of soil moisture estimates from a land surface model in seasonal streamflow forecasts

Wintertime soil moisture has long been known to be an important determinant of spring snowmelt runoff, but observations that could be used in operational streamflow forecasting are essentially nonexistent. Thus, water managers in the western United States typically rely on indices of more observable hydrometeorological data, such as snow water equivalent (SWE), accumulated winter precipitation, and water-year-to-date runoff, as predictors of seasonal streamflow in late winter and spring. While recent research in physically-based modeling suggests that the predictive skill of soil moisture may be comparable to that of more widely observed water balance terms such as SWE, such assessments are inherently contingent on the accuracy of the soil moisture estimates, which show considerable variability across models. We explore methods that address both the modeling of soil moisture and its utility in seasonal streamflow forecasts using simulations from the semi-distributed variable infiltration capacity (VIC) macroscale hydrology model, and a version of the model that is coupled to an explicit groundwater model. The groundwater model we use is the University of Texas SIMple Groundwater Model (SIMGM), which represents an unconfined aquifer underlying the LSM through a single integrated element. We evaluate streamflow forecast skill at a set of sites across the western U.S. using VIC both with and without coupling to SIMGM.