An Explainable Comparative Spatial Decision-Support Framework for Forest Fire Risk Assessment

Forest fires are becoming increasingly complex due to climate change and growing anthropogenic pressures, increasing the need for reliable, transparent, and interpretable spatial risk assessment approaches. This study develops a comparative spatial decision-support framework for forest fire risk assessment and landscape planning in Antalya Province, Türkiye. Forest fire risk factors were identified through the Delphi technique, and their relative importance was determined using the Analytic Hierarchy Process (AHP) and Best–Worst Method (BWM), with criterion priorities integrated through BORDA aggregation. Expert-based risk maps were generated using Geographic Information System (GIS)-based Weighted Overlay and Map Algebra methods. In parallel, Random Forest (RF) and Extreme Gradient Boosting (XGBoost) models were applied for data-driven risk prediction, while SHapley Additive exPlanations (SHAP) analysis was used to interpret model outputs. The expert-driven and data-driven approaches were developed separately and compared in terms of variable importance and spatial risk patterns. The results revealed partial agreement between the two approaches in identifying key risk factors, alongside differences in variable rankings and spatial distributions. Under nested spatial block cross-validation, RF and XGBoost achieved mean ROC-AUC values of 0.9066 and 0.9129, respectively. The proposed framework supports evidence-based forest fire risk management and landscape planning.