Improving Physically-Based Snow Melt Modeling at the Sub-Grid Scale: Optimizing Terrain-Related Roughness and albedo decay time Using Satellite-Derived Snow Cover Fraction as a Model Constraint

Water stored in mountain snowpacks, like those in the Alps, is a vital resource. However, monitoring it in remote areas is challenging. While in-situ measurements are often difficult to acquire and unrepresentative, physically-based snow models provide high-resolution (HR) information but commonly neglect sub-pixel processes influencing snow accumulation, redistribution and melt dynamics. An alternative approach is using remote sensing. By combining multi-source satellite data, HR Sentinel-2 and Landsat with daily low-resolution MODIS data, it is possible to derive daily, HR Binary Snow Cover (BSC) time-series (Premier et al., 2021). This study aims to improve snowmelt estimation by using satellite-derived Snow-Cover-Fraction (SCF) to constrain snowpack modeling. We evaluated how a numerical model i.e., FSM2 (Essery et al., 2024), typically calibrated at a single-point, can be adapted to better represent an entire area and improve the modeled Snow Water Equivalent (SWE). The study was carried out on a 100-meter resolution pixel in the Dischma Catchment (CH), where HR spatially distributed hydro-meteorological-snow data are available (Magnusson et al., 2024). The core of our method is to use SCF as a model input. SCF implicitly reflects the effect of topography on snow redistribution and was integrated into the parameterization of albedo and surface roughness. Although FSM2 can already incorporate SCF, the calibration of these parameters has often been ignored due to a lack of spatialized data, with models relying on settings calibrated at single point. We obtained our SCF dataset from BSC information at 20m aggregated to 100m. Focusing on the melting phase, the model FSM2 was initialized with snow depth measurements acquired at the SWE peak by a drone (Bührle et al., 2023). This allows to evaluate the melting model performance without considering the potential errors introduced by precipitation spatialization and snow redistribution. Unlike ensemble approaches that vary temperature or radiation, our method controls the time decay of albedo during the melting phase using surface temperature, SCF, and surface roughness. Additionally, the roughness of the snow-free portion of each grid-cell was optimized to ensure that simulations end precisely when observed SCF reaches zero. Our results show consistency between elevation and the flatness-index, supporting the effectiveness of this approach. Overall, this method provides a more realistic representation of numerical snow model simulations by accounting for the entire grid cell and its sub-grid variability, rather than treating the cell as a single point. Acknowledgements: This work was co-funded by the European Union - Next Generation EU. References: Essery, R., et., al, 2024. A Flexible Snow Model (FSM 2.1.0) including a forest canopy. Magnusson, J., et., al, 2024. High-resolution hydrometeorological and snow data for the Dischma catchment in Switzerland. Premier, V., et., al, (2021). A novel approach based on a hierarchical multiresolution analysis of optical time series to reconstruct the daily high-resolution snow cover area. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 14, 9223-9240. Bührle, L. J., et., al, (2023). Spatially continuous snow depth mapping by aeroplane photogrammetry for annual peak of winter from 2017 to 2021 in open areas. Cryosphere, 17(8), 3383-3408.