Wildfires Illustration (Credit: Midjourney)
After almost 20 months of investigation since the Eaton Fire first spread, the Los Angeles Fire Department recently released a report identifying the source of the fire.
It’s not unusual for these investigations to take this long— identifying where a fire started quickly remains a major challenge for fire experts and scientists.
Traditional investigations rely on physical evidence, such as scorch marks, that requires investigators to be on site and can be time-consuming. Limited satellite data and a lack of tools grounded in scientific rigor can make it even more challenging to track wildfires in real time and determine where they began.
USC engineers are teaming up with the California Department of Forestry and Fire Protection (Cal Fire) to develop an artificial intelligence (AI) tool called GenFire that could track down a wildfire’s initial source and reconstruct how it spreads during the first 24 hours.
As California faces another wildfire season, the research team aims to speed up and standardize fire investigations, and more precisely identify a fire’s source, to help fire experts develop and implement prevention efforts.
Launching next month, the project will span two years, with the research team collaborating with the Wildland Fire Investigation Subcommittee (WFISC) to test GenFire and ensure the tool’s operational relevance. The team also plans to make GenFire open source, providing a resource that fire investigators and scientists can use, evaluate and build upon.
The research project behind GenFire is backed by a newly awarded grant from the U.S. Forest Service and the Department of the Interior’s Joint Fire Science Program (JFSP).
Led by Assad Oberai, chair of USC Viterbi’s Department of Aerospace and Mechanical Engineering and a Hughes professor of aerospace and mechanical engineering, the research team also includes USC electrical and computer engineering professor Peter Beerel, who also serves as associate chair of graduate programs at USC Viterbi School of Engineering and USC Mark and Mary Stevens School of Computing and AI’s Ming Hsieh Department of Electrical and Computer Engineering, and Cal Fire Deputy Chief and WFISC Chair Shawn Zimmermaker.
“We aren’t just modeling how a fire spreads,” said Oberai, “We are using forensic data reconstruction to look backward and determine precisely where that spread began.”
Current Gaps in Wildfire Tracking
The team explained that many complex factors make tracing a wildfire back to its ignition point difficult.
As a fire moves across a landscape, its behavior is constantly shaped by a changing combination of factors, including terrain, weather and fuel types. These environmental factors can influence the direction and speed of how a fire spreads, making it difficult to distinguish where the fire started from the areas it reached later. Read more about Oberai’s recent work on AI modeling for wilfire forecasts here.
For investigators, the challenge is to reconstruct those early moments from the evidence a fire leaves behind. The same forces that shape a fire’s behavior can make that reconstruction difficult, especially when conditions change rapidly as the fire spreads.
Scientists have tried using satellite data to infer a fire’s ignition location, but most satellite systems provide observations at intervals of 12 to 24 hours. This lack of real-time data means a fire may be “raging for six, 10, 12 hours” before a satellite captures it.
Other video evidence, such as ALERTCalifornia, a network of cameras installed on hilltops and mountaintops to capture wildfires, can provide high-resolution, visible-spectrum video in near real time. However, these cameras are still limited by their distance from a fire and terrain obstructions.
Beerel further pointed out that existing fire investigation approaches have not relied on a consistent scientific framework. Previous methods have achieved an average accuracy of about 2 kilometers from the initial fire spot, leaving too much room for investigators to accurately narrow down and pinpoint where the fire started.
Building GenFire AI Tool to Trace a Wildfire’s First 24 Hours
The team’s goal is to train the AI model on multiple sources of data that influence how a fire spreads as they develop and build GenFire.
GenFire will be trained on environmental data, such as wind, temperature, humidity, terrain and vegetation, along with observations of where and when a fire was detected.
The generative AI tool will use these different data sources to work backward from a fire’s observed spread and generate possible scenarios for how the fire developed during its first 24 hours.
By comparing these possible scenarios, GenFire will identify the most probable ignition area and reconstruct the sequence of the fire’s early growth.
The researchers also plan to train GenFire on thousands of virtual wildfires generated using a physics-based fire simulation model. These simulated fires will allow the AI to learn how environmental factors influence fire behavior before the model is tested on historical wildfires with known ignition locations.
Next, the team plans to validate the model against approximately 100 historical wildfires, comparing its predicted ignition locations with their known origins to measure its accuracy.
The team also plans to build on the FireLoc framework, a wildfire-specific geolocation framework that Beerel previously introduced in his work on wildfire crowdsensing, and adapt the framework to process video from the ALERTCalifornia and ALERTWest networks. This would turn continuous camera footage into time-stamped maps of fire activity, giving GenFire observations at intervals of minutes.
“There is a significant legal and scientific need for this work,” emphasized Oberai. “Identifying a fire’s origin provides the transparency and accountability required to distinguish between natural causes, accidents and criminal acts like arson.”
The researchers hope this new approach can help streamline fire investigations by providing investigators with a more scientifically rigorous method for determining a fire’s origin. GenFire could help investigators better understand ignition events, including potential human or infrastructure-related causes, and inform future wildfire prevention and mitigation efforts.
Published on September 30th, 2026
Last updated on September 30th, 2026