MITRE used artificial intelligence, modeling, and remote aerial sensing to create two unique wildfire mitigation tools. Our researchers collaborated with firefighters on the ground to develop complementary data-driven capabilities to help augment human efforts to battle disastrous blazes.
The summer’s smoke from Canada blotted out the sun over New York City—drifting as far south as Florida. The haze was a startling reminder of the increasing dangers of uncontrolled wildfires and their climate impact, as the world’s forests grow hotter, become drier, and burn faster.
As these infernos spread closer to population centers in places like Colorado and California, they bring often devastating results to lives and property. Firefighters use an array of human-managed and automated measures to combat the blazes, from controlled burns and fireline barriers, to satellite imagery and drone reconnaissance.
MITRE researchers have added two potential capabilities to the arsenal. Developed under our independent R&D program, these systems use AI, data analytics and modeling, and remote sensing to help mitigate the effects of wildfires—and stop the burn.
FiReLine (the “RL” stands for reinforcement learning) blends projection modeling and machine learning to provide a scalable, open-source decision-support tool. It includes the unique feature of integrating historical data on human mitigations implemented during real-world fires, stored in a data set called BurnMD.
“As climate change intensifies and fire behavior evolves, we expect a lot of institutional knowledge from human experience fighting fires will not hold into the future,” says MITRE AI engineer Tim Welsh. “But if we can model how the environment will change over time, we can train officials to fight fires in ways not typically practiced under today’s operational conditions.”
The scale and frequency of wildfires today are so large that the human brain struggles to process all the data out there. AI can boil it down to advise decision-support strategy on a larger scale.
Our complementary tool, ART3MIS (short for Augmented Real-Time 3D Mapping with Intelligent Sensing AI), integrates deep-learning algorithms with remote-sensing data, such as from satellites and drones. This fusion offers a quick, tailored forestry mapping solution. It can also serve as a state-of-the-art data source for FiReLine.
“Wildfire managers indicate that existing data is often too low in resolution and recency and too expansive to efficiently support mitigation planning,” says Dhanuj Gandikota, who leads environmental AI/machine learning for MITRE’s climate effort. “ART3MIS fuses data from cost-effective sources across wide areas to provide a much better picture of what’s happening.”
The two spread-suppression technologies could offer widespread benefits, including more-effective and faster decision making, both before and during a fire. This could improve resource management and reduce acreage burned—and ultimately aid national, state, and local firefighting services in protecting property and lives.
Getting Closer to Ground Truth to Prevent Disaster
Last year, nearly 70,000 fires burned roughly 7.5 million acres of U.S. land. Both the amount of burned acreage and the severity of the damage have increased significantly since the 1980s.
This reflects an unsettling reality: traditional fire projection modeling tools can’t keep pace with the rapidly multiplying effects of climate change. As fires become more frequent and more unpredictable—and data less current and less reliable—firefighters must depend on technologies that don’t necessarily represent the ground truth.
FiReLine aims to get fire officials closer to reality. It operates based on BurnMD’s dataset composed of 308 medium-sized fires from the years 2018-2021, the first large dataset with specific mitigation information.
Much like a game of chess, the tool simulates various behaviors, anticipating actions and reactions to help predict outcomes. What happens when we move one piece (e.g., bulldozers clearing swaths of forest) this way, and another piece (e.g., constructing a fireline) a different way? What does that do the board (the terrain), and how do we get to checkmate (preventing wildfire spread)?
“The simulator plays this game a million times, providing the ‘experience’ of fighting millions of fires,” explains Welsh. “Eventually it learns the most-effective suppression strategies.”
FiReLine could serve as an “AI assistant,” enabling officials to build various scenarios and give options for how to combat a particular fire. Additionally, the capability could assist in integrating emerging technologies as they roll out. A simulator that’s “fought” 10-million fires, combined with human operational experience, could shape novel techniques for battling these disasters.
Welsh acknowledges deploying new tools is only part of the solution. He represents MITRE at a series of National Forest Foundation wildfire community roundtables that explore innovative mitigation strategies. His team also interviewed dozens of wildfire experts and attended conferences with thousands of firefighters at the state, local, and tribal levels in California, from incident commanders to smoke jumpers.
“We know that new firefighting strategies are possible as you introduce new tools. It's up to the people on the ground to determine how best to use that technology,” Welsh says.
Since the project began in 2020, FiReLine has demonstrated the ability to perform 100 million training steps in 12 hours. Its AI model tends to preserve large undamaged areas—meaning the simulator holds promising potential for the real world.
As our cross-disciplinary team refines the prototype, they hope BurnMD’s high-quality dataset will catalyze the research community toward collaborative, impactful wildfire mitigation solutions for resource-limited environments.
Human Action + Machine Learning = Stop the Spread
While FiReLine operates in real time, ART3MIS aims to provide strategic support to mitigation-management professionals. Our researchers collaborated with wildfire and forestry decision makers from Angeles National Forest and staff at University of Texas at Austin to develop a tool that can process and enhance existing imagery. The resulting fused data transitions smoothly into existing mitigation workflows. It’s also adaptable for emerging AI capabilities.
ART3MIS’ 3D dashboard with visualized mission metrics provides more-current, higher-resolution data to enhance wildfire professionals’ situational awareness. The prototype demonstrated capabilities to support fuel mapping (simulating fire growth and intensity), risk analysis, and virtual planning to an individual tree scale (considering how canopy height/width/density affect fire spread). Additionally, the team designed it to evolve for broader forestry, disaster, and even defense applications.
As we continue down the path of increasing climate change, our researchers strive to expand these systems into viable operational options for the people doing the real work.
“The scale and frequency of wildfires today are so large that the human brain struggles to process all the data out there,” Gandikota says. “AI can boil it all down to advise decision-support strategy on a larger scale.”
Welsh adds not only on a larger scale, but with a longer lead time.
“In the future, these fires will burn in locations we aren’t used to seeing them, and they’ll become more and more dangerous,” he says. “We need to learn faster and be more adaptable than ever before. We hope these techniques will help us do that.”
Join our community of innovators, learners, knowledge-sharers, and risk takers. View our Job Openings.