Abstract/Description

Accurate modeling of Snow Water Equivalent (SWE) and discharge in alpine catchments remains a critical challenge for hydrological forecasting and water resource management. The Val Venosta (South Tyrol, Italy), a dry inner-alpine catchment with strong snow seasonality and high water demand, provides an ideal test case. Within the SnowTinel project, we compare three approaches for SWE and discharge modeling, evaluating their predictive accuracy, operational complexity, and computational efficiency.

The first approach employs the fully distributed GEOtop model to simulate SWE and snowmelt processes. These outputs are used as input into the semi-distributed GEOframe model, allowing for a hybrid modeling strategy that leverages GEOtop’s spatial resolution and GEOframe’s hydrological robustness and computational efficiency. The second approach extends the first by incorporating data assimilation of snow depth observations from automatic stations as “virtual meteorological stations” and MODIS-derived Snow Covered Area (SCA) maps. This enhanced assimilation improves the accuracy of snow accumulation and melt estimates, and better captures the spatial extent of snow cover. The third approach employs random forest regression to combine topographic parameters with daily SCA maps derived from multi-source optical data, in order to downscale the GEOframe model SWE at high spatial resolution. Building on previous work, this method demonstrates the potential of integrating hydrological modeling with spatially enhanced snow cover information. Particular emphasis is placed on the 2022–2023 snow drought, which resulted in markedly reduced snow accumulation and anomalously low spring runoff across the basin. Comparing model behavior during this period allows us to assess the reliability and transferability of each approach when hydrological processes deviate from average conditions.

Preliminary results show that the machine learning–enhanced GEOframe approach provides a simple solution with good spatial accuracy but depends on snow depth and discharge observations. Physical models perform reasonably well with limited ground data, while assimilation-based methods achieve the highest spatial accuracy, at the cost of greater computational demand and operational complexity. This comparative study highlights the trade-offs between model sophistication and the capability of capturing correctly winter SWE in snow-drought years, which is key information for early drought detection in Alpine catchments.