Estimating snow distribution using drones and machine learning in Swedish mountain catchments

The snowpack stores a substantial part of the seasonal freshwater in cold environments, impacting catchment runoff generation and timing. Alterations of the seasonal snowpack may affect the availability of water resources, with implications for energy production, relying on meltwater from mountain catchments. Spatial and temporal variability of snow processes at multiple scales challenges snowpack monitoring, snow volume estimations and runoff predictions. Drone acquisition techniques have emerged as a new methodology for snowpack monitoring to obtain dense and high spatial-resolution snow data. This study uses drone observations to estimate snow depth close to the accumulation peak in the Överuman catchment, Northern Sweden. We compared the snow depth average and distribution in the catchment areas where drone acquisitions occurred and in the whole catchment. We explored the use of topographic and wind shelter factors and different machine learning methods to obtain snow depth maps of the entire catchment. We further evaluated the impact of aggregating snow depth data at various spatial resolutions on snow spatial distribution and runoff. Results show high correlations of snow depth, especially with wind shelter factors, which are among the selected predictors in cross-validation, together with topographic roughness at a fine spatial scale. Drone observations provided valuable insights into the snow depth variability to improve process understanding and model development.