WILDEFIRESAI-powered Tool Developed for Near Real-Time, Large-Scale Wildfire Fuel Mapping
Researchers have developed a new system that could help enhance nationwide wildfire preparedness by combining satellite imagery with artificial intelligence to rapidly and accurately identify wildfire fuel sources.
Researchers from the UCLA Samueli School of Engineering and their collaborators have developed FuelVision, a new system that could help enhance nationwide wildfire preparedness by combining satellite imagery with artificial intelligence to rapidly and accurately identify wildfire fuel sources.
In validation tests using data from two of California’s most intense recent wildfires — the Dixie and Caldor fires of 2021 — FuelVision’s predictions closely matched actual fuel maps, demonstrating the tool’s potential for real-world use. The system achieved 77% mapping accuracy in the tests. A study describing the new system was recently published in the International Journal of Applied Earth Observation and Geoinformation.
“We’ve built a tool that lets anyone — from local agencies to global researchers — generate wildfire fuel maps using satellite data,” said Riyaaz Shaik, lead author of the study and a research scientist at UCLA. “That helps make vital wildfire risk information accessible for faster, smarter response.”
Although some models have achieved higher accuracy on large scales, they are slower and rely on expert analysis. In contrast, FuelVision operates autonomously, utilizing commonly available data.
Because it draws data from global satellite inputs, FuelVision is readily adaptable to forested areas nationwide. The system does not require ground surveys to support fire-mitigation strategies or guide emergency responses.
To test and validate their model, the researchers trained the system using real data from the Forest Inventory and Analysis program of the U.S. Forest Service. The team also utilized generative adversarial networks, a type of machine learning that uses a generator to create data and a discriminator that evaluates the data’s accuracy, to produce reliable synthetic training data and help improve the system’s mapping accuracy.
“FuelVision can help anticipate where fires might spread and how to prepare,” said Ertugrul Taciroglu, study corresponding author and a professor of civil and environmental engineering at UCLA Samueli. “It’s versatile, easily adaptable and can help agencies globally with both organizing emergency response and developing long-term risk assessment and fire mitigation strategies.”
The researchers are making FuelVision accessible in two ways. They plan to release a Python-based interface that allows users with basic coding experience to generate their own fuel maps. They will also offer on-demand fuel-map production based on user needs.
