Prometheus: A Data-Driven Framework for Modeling Air Quality Impacts of Wildfires

Abstract

Wildfires have been increasing in frequency, particularly in the western United States. These fires threaten ecosystems and infrastructure while also generating widespread air pollution, which poses significant public health risks. Modeling the air quality impact of wildfires is challenging due to the complexity of fire behavior and the heterogeneity of environmental data. This thesis introduces Prometheus, a data-driven framework that integrates satellite imagery, meteorological records, vegetation and land cover data, and EPA air quality sensor readings to assess the spatiotemporal effects of wildfires on air quality. We estimate daily burn areas using a convex hull method applied to satellite-derived fire pixel detections. This produces consistent and interpretable fire boundaries that serve as the foundation for downstream analysis. We also construct a 30-meter resolution fuel grid for the continental U.S., combining vegetation types, land use, and seasonal weather indicators to estimate burn potential. These inputs, along with dynamic meteorological variables are used in a deep neural network that forecasts air quality degradation at varying distances from the fire. Prometheus enables both retrospective assessment and short-term prediction that can inform decision making.