Forest Fire Hazard Assessment using Remote Sensing Data and Machine Learning, Case Study of Jijel, Algeria

Abstract

Climate change, particularly in vulnerable areas such as the Mediterranean hotspot, exacerbates the risk of wildfires, turning these regions into true danger zones. In Algeria, where climatic conditions are rapidly evolving, these risks are especially pronounced in the northern coastal regions. Wild-fires, which are already frequent, are becoming increasingly difficult to control due to rising temperatures and prolonged periods of drought. In this context, wildfire hazards in Algeria, particularly in the northern coastal regions, present significant challenges. Despite frequent forest fires, effective management strategies are lacking. This paper assesses wildfire hazards in Jijel, Algeria, using advanced machine learning techniques combined with spatial analysis of environmental and anthropogenic factors. Two algorithms, K-Nearest Neighbours (KNN) and Histogram-Based Gradient Boosting (HGB), were employed. Evaluation metrics such as Receiver Operating Characteristic Area Under the Curve, F1 score and accuracy, supported by repeated cross-validation, were used to gauge performance. Both models performed well (with an average area under the curve of 0.977 and 0.984, respectively), with HGB maintaining a marginal but consistent advantage over KNN, with relatively low standard deviations, suggesting stability in its predictive capability. The study offers valuable insights for wildfire management and land-use planning in Algeria, highlighting the need for tailored risk mitigation strategies.

Keywords

• wildfire susceptibility mapping
• remote sensing
• machine learning
• jijel
• algeria
• shap