Spatial and temporal modelling of pastoral value in the Eastern Alps using Sentinel-2 and satellite embeddings

Semi-natural mountain pastures are multifunctional systems that provide key ecosystem services, including forage for livestock. Their long-term preservation depends on appropriate grazing management, a condition increasingly challenged in the European Alps by declining stocking rates and the resulting undergrazing of pastoral areas. In this context, we tested a remote sensing-based approach to spatially and temporally model pastoral value (PV), a comprehensive indicator of pasture quality and applicable for determining grasslands carrying capacity, in two alpages located in the Italian Eastern Alps. Using 111 botanical field observations, we applied Partial Least Squares regression to temporal composites of Sentinel-2 and Google Embeddings data to predict PV at 10 m spatial resolution. We then compared predictions for recent conditions (2023–2025 imagery) with those obtained by applying the same models to 2017–2018 imagery, examining their relationship with slope and distance from night barns as proxies of management intensity. Models achieved R 2 values between 0.53 and 0.75, with RMSE ranging from 9.11 to 12.08, and were able to spatially distinguish lower-quality pastures (PV around 15) from more productive grasslands (PV > 25). Temporal comparisons indicated a reduction in predicted PV at one of the two sites, detected only by the embeddings-based models, while Sentinel-2 predictions showed no change. A stronger decline was predicted in steeper areas, whereas no clear relationship emerged with distance from night barns. Overall, the results show that remote sensing is a promising approach for mapping PV and supporting grazing management in mountain grasslands, while detecting changes over time requires further improvements before operational use.