Spatial Modeling of Anthropogenic Forest Fire Risk from a Landscape Planning Perspective: Comparison of Expert-Based MCDM and Machine Learning Approaches in Kahramanmaraş, Türkiye
MAS Journal of Applied Sciences, cilt.11, sa.3, ss.664-686, 2026 (Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 11 Sayı: 3
- Basım Tarihi: 2026
- Doi Numarası: 10.5281/zenodo.21346163
- Dergi Adı: MAS Journal of Applied Sciences
- Derginin Tarandığı İndeksler: Scopus, Applied Science & Technology Source, Central & Eastern European Academic Source (CEEAS)
- Sayfa Sayıları: ss.664-686
- Anahtar Kelimeler: Forest fire risk assessment, Landscape planning, Machine learning, Multi-criteria decision-making
- Hatay Mustafa Kemal Üniversitesi Adresli: Evet
Özet
Forest fires cause significant environmental and socioeconomic losses, particularly in regions where human activities are concentrated. This study aims to spatially model anthropogenic forest fire risk in Kahramanmaraş Province using expert-based multi-criteria decision-making (MCDM) and machine learning approaches, and to evaluate the results from a landscape planning perspective. Anthropogenic factors identified through the Delphi technique based on expert opinions were used in the study. Within the scope of the MCDM approaches, AHP and SIMOS techniques were applied, and the resulting criterion weights were integrated using the BORDA method to generate risk maps through Weighted Overlay and Map Algebra approaches. Within the scope of the machine learning models, Random Forest (RF) and XGBoost algorithms were employed. SHAP analysis was conducted to enhance the interpretability of the models. The findings indicate that anthropogenic factors such as power lines, population density, and proximity to settlements play a determining role in the spatial distribution of forest fire risk. The RF and XGBoost models achieved accuracy values of 90% and 89%, respectively, and produced similar spatial risk distributions. Similarly, the risk maps obtained from the MCDM techniques also showed consistent results in terms of general spatial patterns. Although some local differences were observed between the MCDM and machine learning approaches, all modelling techniques revealed similar spatial trends, demonstrating the determining influence of anthropogenic factors on forest fire risk in the study area. This study provides an applicable methodological framework for developing risk-oriented landscape planning and land-use decisions.