Regional Mountain Conference

Evaposublimation of snow – the direct transition of water from solid or liquid phases to vapor through turbulent heat flux exchange with the atmosphere – can significantly influence snowpack energy balance. In low- and mid-latitude mountain regions, sublimation losses can account for 10–90 % of winter snowfall, with important implications for spring meltwater availability. Accurate modeling of turbulent latent heat fluxes is challenging due to violations of the law of the wall and Monin-Obukhov similarity theory, as well as uncertainties in aerodynamic roughness length that vary spatially and temporally. Remote sensing provides a valuable tool to monitor snow surface properties – including snow cover fraction, albedo, grain size, and land surface temperature – which, together with meteorological conditions, control turbulent latent heat fluxes between the snow surface and the atmosphere.

We present a machine learning regression framework to predict turbulent latent heat fluxes over snow-covered areas in the European Alps, combining satellite-derived snow products and meteorological data. The model is trained and validated using eddy-covariance measurements from alpine stations, providing high-quality flux observations for evaluation. Our approach captures spatio-temporal variability in evaposublimation rates, integrating satellite and meteorological data to estimate turbulent latent heat fluxes across heterogeneous mountain terrain.