Fire Susceptibility Assessment in the Carpathians Using an Interpretable Framework

Climate change has heightened fire risks, not only in traditionally fire-prone regions but also in Central and Eastern Europe, including the Carpathian region, posing significant ecological and socio-economic threats. This study introduces a comprehensive dataset with twenty-seven variables and an interpretable machine learning framework to assess fire susceptibility across the seven countries of the Carpathians. We implemented a two folded approach: first, refining our predictor set using various feature selection techniques, and second, applying SHAP (SHapley Additive exPlanations) to improve interpretability. The optimized machine learning models, developed using H2O, demonstrated high predictive accuracy, revealing that approximately one-third of the study area (98,511 km²) is at elevated fire risk. Additionally, seven major fire hotspots were identified, providing valuable insights for policymakers and local communities to enhance fire prevention and management efforts.