Understanding the benefit of Sentinel 1 snow-depth retrievals on runoff estimation over Danube River

Abstract:

Snow is an important component of water storage at high latitudes and over mountainous regions.

Particularly, snow melt water – which is mainly determined by the total precipitated snow during the winter – is a significant component of the annual water budget impacting both soil moisture and runoff in snow-dominated basins. Therefore, when modeling hydrological processes over these basins, the quality of runoff predictions not only depends upon the ability of models to predict snow dynamics but also on the quality of the forcing driving the snow accumulation and melt.

Over mountainous regions the estimation of precipitation is very challenging. Classical estimation methods rely on rain gauge networks and radar, satellite observations or models. The first suffer from significant interpolation errors. Radars also are subject to ground echoes and errors caused by shielding of the radar beam by mountain range. Satellite estimates are sub-optimal over mountains regions while the physical parameterization of mountainous orography is very demanding and challenging for numerical models. Moreover, the contrast between snow modelling spatial requirement (generally less 1 km over mountains) and the coarse spatial resolution of the majority of the precipitation products make the snow modelling exercise very challenging.

Recently, the ESA and Copernicus Sentinel-1 constellation have been used to map snow depth across the Northern Hemisphere mountains with 1 km spatial resolution by exploiting C-band cross-polarized backscatter radar measurements. These estimates can offer a viable way to improve snow accumulation errors due to precipitation and other factors. In this study, we assimilated Sentinel-1 snow-depth estimates on a modified version of the snow-17 model (which includes snow depth calculations) over a large-scale domain covering the mountainous regions of the Danube River Basin. Snow-17 was driven by gridded ground temperature and different large-scale (i.e., 25 km) rainfall sources. The assimilation was carried out at 1 km resolution by using Particle Filter. Resulting 1 km snowmelt estimates were then used to force the Sacramento Hydrological model over the Danube river. We show the advantages and problems in using Sentinel-1 snow-depth to improve snowmelt estimates and consequent runoff predictions.