California deploys AI-based wildfire detection systems

Enlarge Getty Images

California's leading firefighting agency, Cal Fire, is training AI models to detect visual signs of wildfires using a network of 1,039 high-definition cameras , report the New York Times and the Los Angeles Times. When it detects signs of smoke, it quickly alerts firefighters to emerging threats. During the pilot program, the system has already detected 77 wildfires before dispatch centers received 911 calls, a success rate of about 40%, according to The New York Times.

Traditionally, Cal Fire detects emerging wildfires by relying on the same network of more than 1,000 human-monitored mountaintop cameras, which watch for signs of smoke. But it is tedious and tiring work. Phillip SeLegue, Cal Fire's intelligence chief of staff, told The New York Times that the new AI system not only improved response times, but also identified some fires earlier, helping to tackle these fires while remaining of a manageable size.

However, the technology is not without its limits. It can only detect fires visible to its network of cameras, and human intervention is still required to confirm alerts from the AI ​​model. Engineers from DigitalPath, the California-based company responsible for creating the software, manually checked each fire identified by the AI. The process was difficult, with many false positives due to fog, mist, dust kicked up by tractors and steam from geothermal power plants, according to Ethan Higgins, chief software architect. "You can't imagine how many things look like smoke," Higgins told The New York Times.

Some experienced firefighters, like Cal Fire North Region Chief Andrew Emerick, remain skeptical of AI's ability to understand the nuanced context in which some fires occur, such as those deliberately started at agricultural purposes. "I don't think this robot will ever take my job away from me," he said.

The AI ​​system processes "billions of megapixels" of images every minute from cameras that cover approximately 90% of California's fire-prone areas. The AI ​​program began in June and was initially rolled out to six of Cal Fire's command centers, but will expand to all 21 Cal Fire command centers in September.

Cal Fire also uses other methods of fire detection, including 911 calls from residents and a partnership with the US military, Fireguard, which uses classified spy satellites, drones and other aircraft to detect fires. Even with early downsides like false positives, Cal Fire AI project manager Neal Driscoll told The New York Times that the ultimate success of the AI ​​program will be measured by the fires the public won't hear from. never speak, for they will soon be extinguished. still small.

This year has been relatively less destructive in terms of wildfires than previous years: Cal Fire has reported 4,792 wildfires so far, which is below the five-year average of 5,422 for the months of summer. Still, that doesn't make wildfires any less devastating when they get out of hand.