Improving streamflow simulations in the Western United States via ensemble snow data assimilation

The seasonal snowpack is a critical source of water in the western US. Past studies of snow data assimilation (DA) show that the better estimates of snow have the potential to enhance the precision of runoff prediction. In this study we select nine basins across the western United States, with a clear snow cover period and supporting snow water equivalent (SWE) measuring gauges, to test the ability of DA of SWE to improve streamflow simulations made with the coupled Snow17 and Sacramento Soil Moisture Accounting (SAC) models. We find that the relatively drier basins with little snow or runoff and basins with relatively complex snow runoff dynamics have less potential for improvement using SWE DA. For the higher potential basins, sensitivity analysis of the Ensemble Kalman Filter (EnKF) DA behavior shows that the correct estimation of SWE mean value is more important than accurately estimating of observed and forecasted error variance, which nonetheless can strongly influence SWE DA performance. This presentation describes key findings from the study, and also comments on different strategies for representing observed SWE, which typically differs from modeled SWE, in performing SWE DA.