Researchers at the University of West Florida have introduced a new benchmark dataset and evaluation of artificial intelligence methods that could greatly improve how wildfire damage is mapped and monitored at scale.
The study, led by Valeria Martin, a Ph.D. student in the intelligent systems and robotics program, introduces CalFireSeg-50, a dataset built from satellite imagery and data from 50 of California’s largest wildfires between 2019 and 2023.
Martin collaborated with Dr. Brent Venable, director of the intelligent systems and robotics doctoral program, and Dr. Derek Morgan, associate professor of earth and environmental sciences at UWF, to teach AI systems how to recognize signs of wildfire damage.
To do this, the team created a large collection of satellite images — essentially showing the AI what wildfire damage looks like. By giving the AI many examples, it learns patterns and becomes better at identifying fire-damaged areas in new images.
This benchmark can also help models identify where fires burn the hottest, helping emergency responders understand how wildfires spread and quickly pinpoint high-priority zones.