As wildfires burn throughout the West, officials are turning to AI

As of Monday morning, 55 large active wildfires were blazing throughout the West, burning more than 2 million acres and displacing tens of thousands of people. Wildfires have ravaged more than 7 million acres so far this summer—the largest acreage to have burned through early September since 2018, according to the National Interagency Fire Center. Now, to get a handle on the growing problem, some governments are turning to AI.

In California, where half a million acres are currently burning, the state is hoping it can use artificial intelligence to prevent fires from burning out of control in the first place. Last year, the California Department of Forestry and Fire Protection, or Cal Fire, began using ALERTCalifornia’s AI-equipped network of about 1,200 cameras to scan for smoke. ALERTCalifornia, which is managed by the University of California, San Diego, is a public safety program working to understand wildfires and other natural hazards and determine short- and long-term impacts on people and the environment.

When AI spots a potential fire on ALERTCalifornia’s network of cameras, it draws a red bounding box around the affected area and provides a percentage of how certain it is that it found smoke. Trained personnel monitoring the footage vet and confirm the alerts and initiate the appropriate action. Time magazine recognized the collaborative effort as one of the top innovations of 2023.

Early results indicate the technology is working. Used in all 21 Cal Fire dispatch centers statewide for the past year, AI has alerted emergency managers to fires before 911 calls did more than 30% of the time, said Caitlin Scully, ALERTCalifornia’s communications program manager.

Between Sept. 1 and Dec. 31 of last year, responders were dispatched to wildland fires 278 times, with ALERTCalifornia noting 190 of them before or at the same time a 911 call was received, wrote Isaac Sanchez, deputy chief of communications at Cal Fire, in an email to Route Fifty. For at least six fires, no 911 call was ever received and, because of the early detection, all of the fires were kept to less than 1 acre.

“The faster [emergency managers] can get out there to either start immediately fighting the fire or get trucks out there or get personnel or drop people out of a helicopter or get planes on it,” Scully said, “the more likely they are able to reach their goal of keeping those fires [to] within 10 acres before they explode into anything out of control.”

The artificial intelligence was trained on datasets that ALERTCalifornia has collected since it started implementing sensor-equipped cameras more than 20 years ago to monitor for wildfires in San Diego’s Laguna Mountains.

“It’s really twofold—the camera network and then also this great data collection that we have,” Scully said. “It has everything we need in order to learn how to spot smoke or other incidents. We also have a longstanding relationship with Cal Fire, and so by working collaboratively between the UC San Diego scientists and then the Cal Fire experts, we were able to come together and develop this AI tool that is really, really useful to them because they were part of it from the very inception.”

California also is using the program Wildfire Analyst, AI and machine learning algorithms to evaluate fire behavior and risk.

“By leveraging the Wildfire Analyst solution, we can enhance our ability to forecast fire spread, intensity and impact, ultimately improving response strategies and minimizing damage,” said Cal Fire’s Sanchez. “The AI and machine learning algorithms can process and integrate diverse datasets, including weather conditions, topography, fuel characteristics and satellite imagery. These algorithms can identify patterns and correlations that may not be apparent through conventional analysis. For instance, machine learning models can be trained to recognize the influence of specific variables on fire behavior, such as the effect of wind speed on fire spread or the impact of humidity levels on fuel moisture.”

In June at the Fire Weather Testbed in Boulder, Colorado, the National Oceanic and Atmospheric Administration’s National Environmental Satellite, Data and Information Service tested its next generation fire system, which can spot fires as small as 1 acre—much like ALERTCalifornia—using AI that analyzes imagery from geostationary satellites in orbit 22,000 miles above Earth.

The alerts are posted to a web dashboard, where a human can confirm the blaze and decide if any action is required.

“It’s really combining the automated satellite detection with data layers that allow decision-makers to sort those detections in ways that are meaningful to the job they have to do, what region they’re working in, what fire weather conditions they’re concerned about, those types of things,” said Mike Pavolonis, manager of the Wildland Fire Program at NOAA. “It will track that fire over time … [to] tell you its intensity, how it’s evolving, a little bit of information about how it's spreading. The whole idea here is to enable more efficient and effective decision-making.”

The system is still in testing, but the agency plans to make it fully operational in one to two years. The Integrated Warning Team paradigm, which was also part of the testbed, will be the best way to reach state and local partners, Pavolonis added.

The paradigm speeds the exchange of information among National Weather Service meteorologists, state and local land managers and emergency managers, enabling them to issue fire warnings using the same dissemination channels the agency uses to issue tornado warnings. Tests showed that it can take 60 minutes to get a fire warning out without the protocol and an average of nine minutes with it, he said.

Overall, AI is a force-multiplier for wildfire detection and management, Pavolonis said.

“The number of different sources and the volume and environmental data and information that decision-makers are confronted with constantly is enormous, and it’s growing, and humans can’t examine every bit of data as it comes in,” he said. “They need some automation to help them extract the most relevant pieces of data at the right time to make decisions they have to make. That’s really where AI comes in.”