As Los Angeles residents battle wildfires, communities elsewhere are deploying artificial intelligence-based systems to detect blazes and forecast their paths.
Fire agencies are exploring a suite of AI innovations to combat blazes such as machine learning algorithms that analyze satellite data to forecast fire paths. At the same time, networks of smart sensors scan for heat signatures and filter out false alarms, potentially giving firefighters crucial early warnings. Though these tools have yet to be deployed in the current California crisis, they point to a future where technology could dramatically speed up fire detection and response.
"Satellite and weather station data can be used to produce a real-time map showing the most wildfire-prone areas," explains Supratik Mukhopadhyay, a Louisiana State University professor studying AI and fire prediction. "Idle resources can be preemptively deployed to these high-risk areas."
The race to predict a fire's path
Identifying high-risk fire zones is just the first step. As fires become potentially more frequent and intense, researchers are racing to develop AI systems to predict where fires might start and how they will spread.
University of Southern California researchers have developed an AI model known as a conditional Wasserstein Generative Adversarial Network (cWGAN), initially trained on simulated data under ideal conditions. The system was then tested on California wildfires between 2020 and 2022, analyzing patterns influenced by weather, fuel and terrain.
The research comes as agencies are already deploying AI in the field. Austin Energy has deployed an AI-powered network of cameras across central Texas that automatically scans for signs of wildfire, aiming to spot blazes before they spread. The system uses 13 high-definition cameras that continuously monitor a 437-square-mile area, alerting firefighters with location data and live images when smoke is detected.
Other companies are racing to develop similar detection systems with extended capabilities. Brad Listerman, Founder & CEO of Los Angeles-based PriviNet, is developing AI-powered sensor networks that could run on battery power for a year or longer with solar power. His company is finalizing prototypes and gearing up to launch its debut product.
"Unless you're going to have the bandwidth of wifi or 5G spreading throughout, you need to have low-powered sensors on the ground," Listerman says. His system would incorporate multiple detection methods to identify fire risks, including infrared sensors, heat detection devices and an array of Internet of Things (IoT) sensors that would alert users to suspicious activity. Cameras would then provide visual verification of the situation through photographs.
The technology relies on LoRaWAN, a free network that can work over distances up to 10 kilometers. "The really important thing is, how do you keep low-power devices on there, and then how do you get that information out?" Listerman says.
The challenge of collecting and analyzing environmental data is one that major tech companies are also tackling. IBM and NASA have developed a geospatial foundation model, available on Hugging Face, that can help scientists estimate the extent of past wildfires. The model is part of a broader collaboration to make climate and weather applications more accessible, with future developments aimed at identifying conditions that could lead to wildfires.
The latest AI News + Insights
Discover expertly curated insights and news on AI, cloud and more in the weekly Think Newsletter.
Satellite challenges
While AI technology races ahead, satellite resolution remains a barrier to wildfire tracking.
"Today's satellites depict fire locations with pixels, which have a size range between 300 x 300 M to 2 x 2 KM, which is often too coarse to explicitly draw out a discrete fire perimeter," says Derek Mallia, Research Assistant Professor in the Department of Atmospheric Sciences at the University of Utah. He noted that methods are being developed to incorporate data from different satellites to more accurately define these fire perimeters using AI, which could also help identify false detections triggered by hot smoke or warm surfaces.
Mallia notes that the urgency of these technological developments is highlighted by the conditions driving current fires. He explains that the Los Angeles wildfires share characteristics with other destructive fires, including the 2023 Lahaina in Hawaii, and 2018 Camp fires in California: for example, they all occurred during significant downslope windstorms that created exceptional fire weather conditions.
"In these conditions, the window to get these fires under control is small, since a fire will grow explosively," Mallia says. Once fires reach a specific size, trees and shrubs become fully engulfed and emit "firebrands"—burning leaves and twigs that can be carried miles ahead by the wind, starting new spot fires.
Fire departments can complement these preventive approaches with sophisticated prediction tools that can help communities manage fire risks and response. The Canadian Forest Fire Weather Index System, currently used by many fire departments in the US and Canada, combines weather features with scientific data to assess fire risk. After fires start, spread models predict how the flames will advance.
"Machine learning or AI-based fire spread models run in real time and could be used by fire managers to identify the best strategies for fire containment," Mallia says.
While weather prediction models can accurately forecast dangerous conditions days in advance, the fires can still prove devastating. The National Weather Service uses terms like "catastrophic downslope windstorm" and "extreme fire risk" to communicate the danger. Still, Mallia suggests more focus is needed on preventing fires in the first place.
"Generally speaking, the LA fires were not ignited in an area that is practical for prescribed burning," Mallia explains. He suggests several preventive measures, including "power shutoffs, better power infrastructure maintenance and better education on activities not to do during fire weather."
AI Academy
Choose the right AI model for your use case
Bigger isn’t always better when it comes to AI models. Learn how to find the right fit for your business needs. Then get the guide to help you take action.
