A Critical Review of Wildfire Risk Prediction Models in Data-Scarce Mediterranean Environments

Abstract

Wildfires are a growing threat in Mediterranean regions where climate variability and land-use practices increase vulnerability to fire risk. Developing effective prediction models is essential for robust wildfire management, particularly in such data-scarce environments. Focusing on data-scarce Mediterranean environments, with reference to environmental conditions observed in Morocco, this review presents prediction models across three methodological categories: spatial risk mapping, temporal forecasting, and fire spread simulation, alongside the satellite data products that support their deployment. Each category is assessed in terms of predictive performance, data requirements, and adaptability to low-resource environments. XGBoost showed strong applicability in data-scarce Mediterranean contexts, while ARIMA was validated for forecasting fire-relevant time series under limited data resources. Freely accessible MODIS-derived products represent a significant asset to the region. Based on this synthesis, a hybrid XGBoost-ARIMA framework incorporating MODIS-derived inputs and SHAP-based interpretability is proposed as a promising candidate architecture to be validated after further investigation. The findings aim to support researchers, land managers, and policymakers in strengthening local wildfire prevention and mitigation efforts by aligning model capabilities with regional data and environmental constraints.