A Multi-Criteria Forest Fire Danger Assessment System on GIS Using Literature-Based Model and Analytical Hierarchy Process Model for Mediterranean Coast of Manavgat, Türkiye
İzzet Ersoy, Emre Ünsal, Önder Gürsoy
Abstract
Forest fires pose significant environmental and economic risks, particularly in fire-prone regions like the Mediterranean coast of Türkiye. This study presents a comprehensive Forest Fire Danger Assessment System (FoFiDAS), by integrating Geographic Information Systems (GIS), a literature-based model, the Analytical Hierarchy Process (AHP), and machine learning (ML) to improve forest fire danger classification. Both models integrate 13 key parameters identified through the literature. A comparison of these models revealed 53% overlap in fire danger classifications. While the AHP model, based on expert-weighted assessment, provided a more structured and localized classification, the literature-based model relied on broader scientific data but lacked adaptability. Pearson correlation analysis demonstrated a strong correlation between fire danger classifications and historical fire occurrences, with correlation scores of 0.927 (AHP) and 0.939 (literature-based). Further ROC analysis confirmed the predictive performance of both models, yielding AUC values of 0.91 and 0.9121 for the literature-based and AHP models, respectively. Five ML algorithms were used to validate classification performances, with Artificial Neural Network (ANN) achieving the highest accuracy (86.5%). The accuracy of the ANN algorithm exceeded 0.93 for each danger class, and the F1-Score was above 0.85. FoFiDAS offers a reliable tool for fire danger assessment, supporting early intervention and decision making.