Ai-driven spatiotemporal modelling of agricultural biomass residues in Northern Italy: A data-fusion and explainable-AI framework

• Data fusion of MODIS NPP, TerraClimate and crop maps for province-year 2002–2031. • FNN/XGBoost predict residues accurately (R 2 > 0.89) • Explainable AI (SHAP, LIME) reveals climatic water deficit and soil moisture. • Maize and wheat supply > 100 Mt/year residues; Po Valley peaks near 30 MJ/m 2. • Post-2026 declines in biomass potential linked to climate and land-use change. Agricultural biomass residues offer a promising pathway for sustainable energy transitions and climate resilience. However, long-term assessments of biomass availability under changing climatic and land-use conditions remain limited, especially at fine spatiotemporal scales. This study presents an AI-based framework to assess crop residue potential across 47 provinces in Northern Italy from 2002 to 2031, addressing the need for high-resolution, long-term modelling of biomass availability under evolving climate and land-use conditions. The framework integrates Feedforward Neural Networks, XGBoost, remote sensing data, and explainable AI techniques (SHAP and LIME) to predict Net Primary Productivity (NPP) and estimate dry biomass from crop residues. The models demonstrated strong predictive performance (R 2 > 0.89), capturing spatio-temporal dynamics in response to variables like climate water deficit, soil moisture, evapotranspiration, crop area, and fertilizer input. Results indicate that maize and wheat dominate the residue supply, contributing over 100 million tonnes of dry biomass annually, supporting approximately 1,100 billion MJ of energy or 100 billion kWh of electricity, with energy density peaking at 30 MJ/m 2 in the Po Valley. However, forecasts beyond 2026 suggest a decline in biomass productivity driven by rising temperatures, increased water deficits, and land-use change. High-yield provinces like Bologna, Verona, and Padova may experience reductions up to 49 %, while moderate and low-productivity areas remain more stable or constrained. These findings highlight the value of AI-based models in supporting climate-resilient agriculture and regional resource planning. The proposed framework advances smart agriculture by integrating machine learning, remote sensing, and model interpretability to enhance biomass monitoring and long-term sustainability.