PyroStack

Wildfires are an increasing threat to ecosystems, air quality, and human communities. Rising temperatures, changing climate conditions, and expanding development in fire-prone areas are contributing to more frequent and severe fires. Accurate prediction of wildfire spread depends on datasets that combine meteorology, fuels, vegetation, topography, and fire observations at high spatial and temporal resolution. PyroStack was developed to meet this need by providing a comprehensive, event-based spatiotemporal wildfire dataset spanning the contiguous United States and Alaska. PyroStack integrates topography, vegetation, fuels, meteorological variables, and fire observations across nested spatial resolutions ranging from 30 m to 9 km, with meteorological data available at hourly intervals and fire progression observations provided every 12 hours. The dataset contains 6,994 wildfire events representing a wide range of ecosystems and climatic conditions, enabling systematic evaluation of both physics-based and machine learning fire spread models. By offering standardized, high-resolution data across diverse fire environments, PyroStack supports model initialization, benchmarking, validation, and comparative studies aimed at improving wildfire prediction capabilities.