The Utility of Satellite Snow Depth Observations for Downscaling Hydrologic Variables over the Indus Basin Mountain Ranges

Abstract Accurate, high-resolution, and continuous estimates of snow characteristics and related hydrologic variables are critical to understanding freshwater availability in regions where snowmelt dominates the hydrologic cycle. However, such estimates are difficult to obtain, especially in mountainous regions. Recent advances in data assimilation and radar satellite technologies can be leveraged to produce snow depth estimates in these data-sparse regions. In paired synthetic and real-data experiments, we use a particle batch smoother to assimilate Sentinel-1 satellite snow depth observations into a coarse-resolution land surface model. We produce continuous, downscaled estimates of snow depth, snow-covered fraction, snow water equivalent, precipitation, runoff, and evaporation over the Indus basin in High Mountain Asia (HMA). In the synthetic experiment, data assimilation reduces the initial open-loop root-mean-square error (RMSE) for snow depth and snow water equivalent (SWE) estimates by ∼50% and decreases the bias in other model variables. We show that assimilating only a single peak snow depth satellite observation improves snow depth and SWE estimates by reducing bias and RMSE by 40% or more. In the real-data experiment, assimilating all available Sentinel-1 snow depth observations reduces the bias in snow disappearance date by over 30 days and improves the temporal evolution, spatial representation, and magnitude of SWE. Ultimately, we demonstrate that downscaling model variables through a data assimilation update step reduces model errors and improves the spatial representation of peak snow.