An Operational Spatio-Temporal and Explainable Machine Learning Framework for Daily Wildfire Risk Prediction in Morocco

This chapter introduces AI4Fire, an operational and explainable spatio-temporal machine learning framework for daily wildfire risk prediction in Northern Morocco. The system integrates large-scale geospatial data, advanced spatio-temporal feature engineering, high-performance computing, and explainable artificial intelligence to generate high-resolution risk and interpretability maps over more than one million grid points. AI4Fire combines meteorological dynamics, heatwave indicators, vegetation status, topography, socio-environmental factors, and historical wildfire pressure within a scalable Big Data architecture enabling near real-time inference. Validation over the 2021-2023 fire seasons demonstrates strong and stable predictive performance, with more than 75% of wildfire events occurring in areas classified as very high risk. By coupling accuracy, scalability, and transparency, AI4Fire provides an effective decision-support tool for proactive wildfire prevention and sustainable forest management in Mediterranean ecosystems.