Sentinel-2 Based Validation of Snow Covered Area of Alpine3D Simulations Applying Snow Depth Pattern Redistribution across Two High-Alpine Karst Catchments

Modelling high-alpine hydrology poses significant challenges due to terrain heterogeneity and complex topography. In snow-dominated karst catchments, accurate representation of spatiotemporal snow distribution is essential for simulating aquifer recharge and spring discharge dynamics. However, differences in data availability and landscape complexity can be a limit. Here, we assess the ability of Alpine3D applying snow pattern redistribution to capture the spatiotemporal snow cover variability in two adjacent alpine karst catchments with different spatial snow distribution characteristics.

We present an 11-year (2015-2025) validation of Alpine3D simulations in two adjacent high-alpine karst catchments in the Zugspitze region in Germany (European Alps): the Partnach Spring catchment (15.4 km², 1430-2962 m a.s.l.) and the Hammersbach catchment (17.8 km², 768-2951 m a.s.l.). While both catchments share similar karstified alpine geomorphology, Partnach Spring is characterized by higher elevations on average, limited vegetation, and more persistent snow cover, whereas Hammersbach exhibits stronger elevation gradients, greater forest cover, and higher radiation exposure, leading to more heterogeneous snow accumulation and melt dynamics.

Precipitation and snow were redistributed in order to correct snow water equivalent quantitatively and spatially. Therefore, we used a snow depth map derived by Pléiades stereo satellite images taken on the 9^th of April 2021, near peak snow accumulation. Data gaps, e.g. due to shaded areas and very steep terrain were filled using Random Forest trained on terrain attributes, topographic indices, and energy balance parameters. Alpine3D was run at 16 m × 16 m resolution with spatially interpolated meteorological station data on an hourly base and was validated against Sentinel-2 snow cover area (SCA) maps during the melt season (May-August). Snow classification employed dual thresholds (red band reflectance and NDSI) with manual cloud masking and DEM-based shadow removal. Modelled performance was evaluated using pixel-based confusion matrices across multiple dates per year, whereof we will present preliminary results for both catchments.

This multi-catchment approach with different characteristics, but similar meteorological conditions aim to demonstrate the transferability of this snow redistribution method across different alpine environments. The results are valuable insights for improving hydrological predictions in ungauged basins with limited spatially distributed snow observations.