Integrating multi-source high-resolution satellite data to redistribute SWE from semi-distributed hydrological modeling using a random forest approach

Alpine regions are highly sensitive to climate change, and snow melt dynamics are crucial in their hydrological processes. A key variable is the snow water equivalent (SWE), but direct measurements are rare and limited, making spatial estimates difficult. Remote sensing (RS) provides well-established, high-spatial resolution snow cover area (SCA) observations, yet lacks accurate water mass balance information essential for hydrological modeling. To overcome these limitations and improve high-resolution SWE estimates, we developed a redistribution approach combining semi-distributed hydrological modeling and remote sensing information. SWE simulated by a semi-distributed model (GEOframe) was redistributed at high spatial resolution using a random forest regression trained on topographic features and dynamic snow cover duration (SCD) derived from optical remote sensing. The methodology was tested in an alpine catchment (Dischma, Switzerland) using high-resolution snow-hydrological datasets for evaluation. The semi-distributed hydrological model reproduces discharge dynamics, achieving Kling – Gupta efficiency (KGE) values of 0.90 and 0.73 during calibration and evaluation, but provides coarse SWE spatial patterns. The proposed redistribution approach effectively improves the spatial representation of SWE, achieving a mean bias error (MBE) of -28.24 mm and a Pearson correlation coefficient (CORR) of 0.75 at SWE peaks. In addition, redistributed SWE maps show high agreement with the Swiss operational snow-hydrological model system (OSHD), with a CORR of 0.90, while providing a finer spatial resolution (20 m). The catchment-average SWE is bound to the water mass balance estimated by the hydrological model, ensuring consistency with hydrological processes and improving the spatial and temporal representation of snow-melt dynamics.