Combined assimilation of Sentinel-2 snow cover fraction and snow station data improves fully distributed snow simulation across multiple spatial scales

Abstract. Seasonal snow cover and meltwater runoff have wide-ranging ecological, hydrological, and socioeconomic effects on regions within and downstream of mountainous areas, underscoring the need for accurate monitoring. Snow data assimilation improves estimates of snowpack evolution by combining numerical simulations with observational data, as neither source alone can adequately capture the strong spatial and temporal variability of mountain snowpacks. However, the benefit of assimilating spatially sparse or temporally infrequent observations is maximized only when these observations are representative of unobserved areas and unmonitored periods. In this study, we present a novel two-step framework for the combined assimilation of in situ snow depth observations and spaceborne snow cover fraction (SCF) observations. First, a particle filter–based assimilation of point-scale snow depth observations accounts for spatiotemporal uncertainties in the meteorological forcing at subregional scales over consecutive three-day assimilation windows. Second, the remaining small-scale errors are addressed by assimilating SCF maps using a particle batch smoother, targeting grid-cell-specific uncertainties in model parameters that control the representation of albedo decay and gravitational redistribution within the model. These parameters explicitly address deviations in snowpack evolution related to slope and aspect relative to the flat-field locations, and are therefore independent of the forcing corrections inferred during the first assimilation step. The proposed approach reduces the RMSE and bias of SCF estimates during the ablation season by approximately 50% at observed locations. At unobserved locations, estimates are updated by interpolating the inferred parameters based on horizontal distance and terrain differences, yielding considerable albeit less pronounced improvements. Overall, this study demonstrates that assimilating complementary snow observations can substantially improve near-real-time snowpack simulations across multiple spatial scales over complex mountainous terrain.