Statistical Modeling of Wildfire Extremes Based on Extreme Value Theory

Abstract

Wildfires have become an increasingly severe environmental hazard in the lower northern region of Thailand, causing substantial ecological degradation, economic loss, and public health impacts. Understanding the behavior of extreme wildfire events is essential for long-term risk management and mitigation planning. This study applies Extreme Value Theory (EVT) to model annual maximum burned areas in five provinces—Tak, Phitsanulok, Sukhothai, Phetchabun, and Uttaradit—using official wildfire records from the Department of National Parks, Wildlife and Plant Conservation covering the period 1998–2024. The Generalized Extreme Value (GEV) distribution was fitted using maximum likelihood estimation under the block maxima framework. Model adequacy was assessed using goodnessof-fit diagnostics and graphical tools. Return levels were estimated for multiple recurrence intervals, including 10, 20, 50, and 100 year horizons. Results reveal pronounced spatial heterogeneity in extreme wildfire behavior across provinces. Tak province exhibits the heaviest tail and the highest projected 100-year return level, followed by Phitsanulok, whereas Uttaradit demonstrates comparatively lower extreme magnitude. A normalized spatial risk index was constructed to facilitate comparative hazard ranking. The findings indicate that extreme wildfire risk is strongly localized and non-uniform across the region, emphasizing the necessity of province-specific mitigation strategies. The integration of EVT-based return level analysis with spatial risk ranking provides a quantitative decision-support framework for wildfire hazard assessment in climate-sensitive regions.