Generalizing Human-Driven Wildfire Ignition Models Across Mediterranean Regions Using Harmonized Remote-Sensing and Machine-Learning Data

Wildfires represent a growing environmental and socio-economic threat across Mediterranean landscapes, where prolonged summer droughts and human activity increasingly shape ignition susceptibility. This study presents an open and reproducible modelling framework for comparing the relative influence of anthropogenic and biophysical drivers of wildfire ignition susceptibility across selected Mediterranean regions. Using harmonized 500 m predictors derived from global remote-sensing datasets, we integrate vegetation condition, topography, climatic context, and human pressure indicators within a cloud-based Google Earth Engine workflow. Two tree-based machine-learning models (Random Forest and Extreme Gradient Boosting) are trained and evaluated using spatial cross-validation and cross-region transfer experiments. Results consistently highlight the dominant role of anthropogenic pressure in shaping ignition susceptibility across all study areas, with night-time lights and human modification indices contributing to the largest share of model importance. Both models achieve high predictive performance (AUC > 0.90) and retain stable accuracy under cross-region transfer (mean transfer AUC ≈ 0.85), indicating partial generalization of human-driven ignition patterns across Mediterranean landscapes. Beyond predictive performance, the principal contribution of this work lies in its harmonized cross-regional comparison and explicit evaluation of model transferability using open data and scalable cloud processing. The resulting susceptibility maps provide a transparent and operational basis for comparative wildfire risk assessment and prevention planning within comparable Mediterranean contexts.