Synergistic approaches in forest fire risk mapping using fuzzy AHP and machine learning models in the Chure Tarai Madhesh Landscape (CTML) of Nepal

Abstract

Forest fires are recurrent natural hazards threatening ecosystems, biodiversity, and nearby communities. The Chure Tarai Madhesh Landscape (CTML) is a biodiversity hotspot harboring endangered species of flora and fauna. The increasing severity of forest fires in this region has raised immediate concerns, yet research remains limited. This study explores synergistic approaches for forest fire risk mapping using a knowledge-based model (Fuzzy Analytical Hierarchy Process (FAHP)) and data-driven models (Random Forest (RF) and Boosted Regression Tree (BRT)). This study utilized eleven conditioning factors and assessed model accuracy using the ROC curve and multiclass error matrix. The results demonstrate low multicollinearity among factors and a robust FAHP consistency ratio of 0.03. The RF model outperformed with an AUC of 0.95 and an overall accuracy of 0.91. The study revealed an increasing seasonal trend in fire incidents, with the western region showing heightened vulnerability. The RF, BRT, and FAHP models classified landscape forest areas as highly susceptible to fires; 47.85%, 33.25%, and 50%, respectively, with fourteen out of thirty-six districts of CTML were at heightened risk of wildfires. This holistic approach to fire risk assessment aids in creating more impactful fire risk management plans and provides a foundation for automated fire risk assessment.

Keywords

• Forest fire
• Chure Tarai Madhesh Landscape
• fuzzy AHP
• machine learning
• forest fire risk map