Global Burn Severity in Forest Ecoregions: Trends, Drivers, and Predictive Insights

Forest fires play an important role in shaping ecosystems but pose escalating threats to biodiversity, carbon dynamics and human safety. Advancements in remote sensing and machine learning techniques provide new opportunities for monitoring and predicting forest fire burn severity at large scale, yet a comprehensive global assessment of burn severity trends and climate interactions is missing. In this study, we analyzed trends of burn severity across ecoregions from 2003 to 2023 using a newly developed 30-m resolution Global Burn Severity Dataset. Using the Mann-Kendall test and Sen’s slope estimator we revealed significant increases in forest fire burn severity over tropical and subtropical regions, while boreal zones exhibit declines. For ecoregions showing significant trends, we developed predictive models using XGBoost and 14 climate variables from the TerraClimate dataset, which were used to identify key drivers, including vegetation water stress and atmospheric dynamics. SHAP analysis highlighted regional climate variables importance to wildfire dynamics. This integrated approach enhances our understanding of wildfire drivers and offers a new tool for projecting future scenarios under climate change. Through its combination of trend analysis, predictive modeling and climate applications, this research supports management strategies to mitigate wildfire impacts on carbon cycles and biodiversity.