Integrating Remote Sensing and Machine Learning for Monitoring and Managing Grasslands in North-East Italy.

Grasslands are key ecosystems supporting livestock production, biodiversity, and carbon storage, yet their productivity, botanical composition, and phenology are highly variable in space and time. This variability, driven by complex interactions among climate, soil, and management, makes their monitoring challenging through traditional approaches. This Ph.D. research aimed to assess the potential of remote sensing and machine learning for improving the monitoring and management of permanent meadows and pastures in North-East Italy. The main objective was to develop scalable and integrative methods to quantify productivity and phenological dynamics while accounting for environmental and management constraints. First, long-term satellite observations were combined with climatic, topographic, and soil data to identify the main drivers of grasslands and pastures dynamics across Alpine and pre-Alpine landscapes. Then, probabilistic machine learning models were applied to Sentinel-2 and PlanetScope imagery to estimate above-ground biomass (AGB) at high spatial and temporal resolution. The results demonstrated that the proposed models achieve high predictive accuracy and provide interpretable uncertainty estimates, supporting their operational use in precision management. Overall, this work demonstrates that integrating remote sensing and probabilistic modelling can effectively support data-driven, climate-smart grasslands management.