Wildfires scorched nearly nine million acres of land across the United States in 2024—an area nearly twice the size of New Jersey. The flames consumed homes, destroyed habitats, and blanketed communities in smoke, leaving behind billions of dollars in damages. As environmental conditions change and fire seasons grow longer, the question looms: how can we get ahead of the next inferno?
Heng Huang, the Brendan Iribe Endowed Professor in Computer Science at the University of Maryland, believes that AI may hold the key. Backed by a $1.86 million award from the National Science Foundation, he is partnering with researchers at American University to combine advanced AI techniques with massive streams of environmental data to improve how wildfires are forecasted, tracked and managed.
“This is about creating tools that are not only accurate, but also useful for people on the ground,” says Huang, who holds an appointment in the University of Maryland Institute for Advanced Computer Studies (UMIACS) and is active in two of the centers UMIACS supports: the Center for Bioinformatics and Computational Biology and the University of Maryland Center for Machine Learning.
Huang is collaborating on the project with Leah Ding, an associate professor of computer science at American University. Building on a long history of collaboration on projects involving machine learning and remote sensing, the two aim to tackle gaps in existing wildfire forecasting systems.
While scientists currently rely on satellite data, local surveys, and geographic information, Huang says, these sources are rarely brought together in a systematic way—and few efforts have used AI to combine them at scale.
Unlike traditional fire models that rely on narrow inputs, the NSF-funded project will draw on a wide range of information: satellite images from different orbits, ground sensors, weather records, vegetation moisture, historical fire maps, and surface characteristics. Huang says that by layering these variables together, the research team can build a more dynamic picture of when and where wildfires are most likely to ignite and spread.
Managing this flood of information presents enormous challenges. Records arrive in different formats, at different times, and at overwhelming scale. Huang says his team will use UMIACS’ high-performance computing resources to train AI models capable of working with incomplete or uneven data, a common problem when satellites miss a cloudy day or a ground sensor goes offline.
One of the project’s biggest innovations is its ability to combine different kinds of data in real time. For example, some satellites provide quick updates with low detail, while others offer sharp images less frequently. By blending these strengths, the team hopes to generate predictions that capture both speed and precision. They are also using a method that allows models to learn from sensitive data at agencies like NASA or the U.S. Forest Service without that information ever leaving secure systems.
Equally important is ensuring the technology is understandable to the people who use it. Huang and his collaborators are designing systems that not only make predictions but also explain why they raised an alarm. Their models also track fire as an ongoing event rather than a single moment—showing how risks grow, shift and spread depending on conditions.
By working closely with partners like NASA, the U.S. Forest Service, and the National Park Service, the research team hopes to align their tools with real-world needs. For instance, after models predict fire risks in specific areas, the team plans to collaborate with park services on field validation, potentially using drones equipped with cameras to confirm local fire activity.
The project also carries a strong educational and outreach mission. Huang plans to involve undergraduate and graduate students directly in the research, preparing them at the cutting edge of AI and environmental science, with additional opportunities for local high school students to contribute through drone testing and data validation.
Beyond campus, the team will release an open-source integrated dataset for the broader research community, share findings through tutorials and workshops at major conferences, and introduce a new course that connects data science with applications like remote sensing.
“This work is about more than algorithms—it’s about building a bridge between science and practice to protect lives, communities, and ecosystems,” Huang says. “With the support of NSF and our partners, we hope to move wildfire response from reacting to disasters to anticipating and preventing them.”
—Story by Melissa Brachfeld, UMIACS communications group
***
“FIRE-MODEL: Advanced AI Framework to Improve Understanding and Prediction of Wildland Fire” is supported by NSF grant #2536663 from the NSF’s Division of Research, Innovation, Synergies and Education (GEO/RISE)
PI: Heng Huang, the Brendan Iribe Endowed Professor in Computer Science