Sentinel-1 snow depth assimilation to improve river discharge estimates in the western European Alps

Abstract

Seasonal snow is an important water source and contributor to river discharge in mountainous regions. Therefore the amount of snow and its distribution are necessary inputs for hydrological modeling. However, the distribution of seasonal snow in mountains has long been uncertain, for lack of consistent, high resolution satellite retrievals over mountains. Recent research has shown the potential of the Sentinel-1 radar satellite to map snow depth at sub-kilometer resolution in mountainous regions. In this study we assimilate these new snow depth retrievals into the Noah-Multiparameterization land surface model using an ensemble Kalman filter for the western European Alps. The land surface model was coupled to the Hydrological Modeling and Analysis Platform to provide simulations of routed river discharge. The results show a reduction in the systematic underestimation of snow depth, going from 38 cm for the open loop (OL) to 11 cm for the data assimilation (DA) experiment. The mean absolute error similarly improves from 44 cm to 37 cm with DA, with an improvement at 59% of the in situ sites. The DA updates in snow depth results in enhanced snow water equivalent and discharge simulations. The systematic negative bias in the OL is mostly resolved, and the median temporal correlation between discharge simulations and measurements increases from 0.61 to 0.73 for the DA. Therefore, our study demonstrates the utility of the S1 snow depth retrievals to improve not only snow depth amounts, but also the snow melt contribution to river discharge, and hydrological modeling in general.