Modeling wildfires burned area in the continental US using the extreme boosting machine learning model

Abstract:

Wildfires are becoming more frequent and intense in the continental US, which causes property damage and poor air quality. Wildfire is a complex process that intermingles multiple factors, including ignition, fuel, weather, topography, and climate. Thus, developing prediction models that encompass the complex and non-linear relationships between fires and their drivers, estimate the burned area, and reveal the explanatory power of the drivers, is increasingly important and demanding. In this study, we develop an extreme boosting model incorporating instantaneous meteorological, land-fuel, and socioeconomic factors, to estimate monthly burned area in regions across the continental US during 2000-2017. In addition, a predictor that characterizes the synoptic meteorological variability is incorporated in the machine learning model to account for the influence of large-scale meteorology on wildfires. The model reproduces the monthly spatial-temporal patterns of burned area reasonably, with an index of agreement of 0.7. Furthermore, we use the Shapley Additive explanations (SHAP), a game theory approach, to interpret the machine learning model and examine the relative importance of the predictor variables to wildfire prediction. The major contributing factors of burned area are analyzed and contrasted among different regions, specifically for the large burn events. The results demonstrate the use of the machine learning technique for wildfire predictions and provide a better elucidation of the complex processes contributing to wildfires.