Explainable AI for Wildfire Risk Mapping in Mediterranean Regions Using Climate and IoT Data

¹ SISA Lab, Faculty of Science, Abdelmalek Essaadi University, Tetouan, Morocco
² ELITT-Lab, Higher School of Technology, Abdelmalek Essaadi University, Tetouan, Morocco
³ National school of business and management, Abdelmalek Essaadi University, Tetouan, Morocco

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Wildfires in Mediterranean regions have increased in frequency and severity over the past two decades, driven by climate change, prolonged drought cycles, and growing landscape vulnerability. While machine learning approaches have significantly improved wildfire risk modeling, most existing systems operate as black-box predictors, producing risk scores without interpretable explanations. This lack of transparency limits their practical usability for emergency managers, land-use planners, and policymakers, who require not only accurate forecasts but also clear and actionable reasoning to support decision-making under uncertainty. This paper proposes a conceptual framework for Explainable AI (XAI)-driven wildfire risk mapping, specifically tailored to Mediterranean climatic and ecological conditions and based on the integration of climate reanalysis data and Internet of Things (IoT) sensor networks. The central contribution is the design of an Explainability Engine a dedicated architectural component that translates model outputs into physically meaningful and human-interpretable insights across local, regional, and temporal scales. Additional contributions include an explainability-first design paradigm, a domain-aware data integration pipeline, and a multi-level explanation strategy aligned with operational decision processes.

The proposed framework is illustrated through scenario-based analyses reflecting representative Mediterranean wildfire conditions, including Foehn wind events, post-drought vegetation stress, and seasonal fuel accumulation dynamics. A preliminary validation conducted on a proof-of-concept dataset covering northern Morocco over the period 2019-2024 yields encouraging results, with an accuracy of 82.4% and a ROC-AUC of 0.898, supporting the conceptual coherence of the framework.

This work advances the development of transparent and decision-oriented wildfire risk modeling systems, with direct implications for early warning strategies, resource allocation, and climate adaptation planning. Keywords: Explainable AI; Wildfire Risk Mapping; Mediterranean Climate; IoT Sensing; SHAP; Interpretable Machine Learning; Environmental Decision Support.