WildfireDB: An Open-Source Dataset ConnectingWildfire Spread with Relevant Determinants

Modeling wildfire spread is critical for emergency management and fire risk assessment, but comprehensive datasets relating fires to relevant environmental covariates have been lacking. Traditional fire spread models rely on physics-based approaches that require detailed parameter specification, while data-driven approaches are limited by insufficient data. This work is innovative because it provides the research community with a large-scale, multi-source dataset that enables machine learning approaches to improve fire spread forecasting and supports more sophisticated understanding of fire dynamics.

WildFireDB is a comprehensive open-source dataset connecting wildfire spread with relevant environmental determinants including weather, vegetation, and topography. The dataset comprises over 17.8 million data points covering wildfire occurrences in the continental United States from 2012-2017, integrating fire detection data from satellite imagery with spatial vegetation and topographic information. The work presents algorithmic approaches for merging large-scale raster and vector data to create a spatially and temporally coherent dataset for modeling fire spread.

The dataset successfully integrates fire occurrence data from satellite sensors with vegetation, topographic, and weather information across a standardized spatial grid. The publicly available dataset enables development of data-driven models for fire spread forecasting. The work demonstrates the feasibility of creating large-scale integrated datasets that combine multiple data types at different spatial resolutions, providing a foundation for advancing fire risk management through improved predictive models.