Catching wildfires earlier: AI gives firefighters a head start

Microsoft's AI for Good Lab and UC San Diego's ALERTCalifornia network, comprising nearly 1,300 cameras, now processes over 7 million images daily to detect wildfires as early as 2.5 hours before the first 911 call, with the system spotting 77 incidents in its first two months. The AI tool, enhanced with wildfire technology experts and fire agencies, alerts first responders within minutes, giving them a critical head start in a year when nearly 78,000 wildfires were reported across the U.S., a 20 percent increase over the previous year.

August 27, 2026 Firefighters held the line. Radios crackled with updates as dry brush reignited. Kids counted “three, two, one” as planes dropped ribbons of red slurry, slowing the blaze. Then the wind shifted. Neighbors packed their cars. The Sorrento Fires burned more than 220 acres of parched coastal shrub in San Diego this summer. Once you watch a wildfire climb a canyon toward your home, you start noticing what might fuel the next. Fire has shaped California—long before neighborhoods sprawled across its hills. It renewed forests and grasslands, cleared old growth, and returned nutrients to the soil. Fire itself isn’t the enemy. What’s changed is how we live alongside it. Last year, nearly 78,000 wildfires were reported across the United States, a 20 percent increase https://www.nifc.gov/fire-information/statistics/wildfires over the previous year. Hotter temperatures, prolonged drought, and expanding development have transformed fire into a year-round threat. This urgency is bringing together scientists, firefighters, and Microsoft’s AI for Good Lab with the shared goal: to detect fires sooner. Where there’s smoke For Dr. Neal Driscoll, geophysicist and professor at the University of California San Diego, everything changed during the Cedar Fire of 2003. “The fire skirted up my neighborhood. People lost their lives. We didn’t know where to go. We called 911 and said, ‘We can’t see.’ The whole sky was orange.” At the time, Driscoll’s research focused on tectonics, work that even landed him on “Bill Nye the Science Guy.” But the Cedar Fire shifted his focus. Since then, he’s dedicated his career to understanding wildfires. He asked: What if smoke could be detected before the flames took hold? The question led him to found ALERTCalifornia at UC San Diego. The growing network of nearly 1,300 cameras watches over the state’s most fire-prone landscapes. In its first two months of operation alone, ALERTCalifornia detected 77 incidents before they were reported. Together with wildfire technology experts, fire agencies, and Microsoft’s AI for Good Lab, the team enhanced ALERTCalifornia’s award-winning https://alertcalifornia.org/awards/ AI tool that continuously analyzes live camera feeds https://cameras.alertcalifornia.org/?pos=37.2382 -119.0000 6 for the earliest signs of smoke. When the system detects a fire, it compares camera angles, verifies the location, and sends an alert within minutes. The system now processes more than 7 million images daily, spotting a wildfire as early as 2.5 hours before the first 911 call. It gives first responders something they never have enough of: time. Changing the timeline For decades, wildfire detection depended on someone seeing smoke—a passing driver, a lookout, or a neighbor placing a call. Increasingly, it begins with AI, according to Zachary Wells, deputy chief of Kern County Fire Department and deputy director of operations at ALERTCalifornia. Before a firefighter ever climbs into an engine, the system begins separating signal from noise. Clouds drift. Fog rises. Morning light shifts across canyon walls. AI learns the difference, filtering false positives to focus on real threats. When smoke is detected, dispatchers send out a live camera feed. But AI doesn’t fight fires. Firefighters do. They’ve spent their careers learning how to read smoke. What AI provides is a head start—allowing them to arrive sooner and make better decisions before a small fire becomes a big crisis. By the time crews leave the station, they’ve already seen the fire. They know whether they’re responding to a prescribed burn or an emerging wildfire, how it’s likely to behave, and which route offers the safest approach. Multiple camera views show them what lies ahead. Every fire has a chance to get out of the box. And that’s what we’re trying to solve. Let’s not let the fire out of the box. Photo: Emily Zheng We gain so much knowledge in the first five minutes. That early understanding informs everything that follows. and Deputy Director of Operations, ALERTCalifornia Photo: Emily Zheng If you detect a fire early, you can stop it with a shovel. If you wait, you’ll need a bulldozer, air tanker, and maybe even a miracle. Microsoft AI for Good Lab Everybody goes home “Among firefighters, there’s a saying: Everybody goes home,” explains Wells. It reflects the first priority behind every decision—protecting the communities they serve. In 2023, ALERTCalifornia’s cameras and AI helped crews rapidly detect and contain Kern County’s Trotter Fire at 52 acres. Without AI-powered imaging, Driscoll estimates it may have spread to nearly 4,000 acres. During Sonoma County’s chaotic Kincade Fire in 2019, improved early warning systems helped officials evacuate more than 180,000 residents without a single loss of life. But the greatest success is the fire that’s stopped before it makes the evening news. As climate change drives more extreme wildfires, faster detection is more urgent. “We want this technology to be available to people across the world,” says Juan Lavista Ferres, vice president and chief data scientist at Microsoft’s AI for Good Lab. The goal isn’t to export a California model but to build a platform that communities everywhere can adapt to their needs. Scaling the tech means placing tools in the hands of regional partners—experts who know the land and how fire moves through it. To help realize that vision, Microsoft recently announced a $5M contribution that includes $2M for in-kind technology development and a $3M Azure grant to support ALERTCalifornia’s next phase, setting the foundation for greater impact across the United States and around the world. After the fires Once a fire’s out, it’s not over. Charred hillsides become vulnerable to erosion. Heavy rains trigger debris flows. Watersheds shift. “We call these cascading disasters,” says Driscoll. The same AI-powered imaging transforming fire detection is now contributing to long-term resilience. ALERTCalifornia is building one of the most detailed digital models of California’s landscape ever assembled. Using nearly 98,000 square miles of LiDAR data, researchers map terrain, monitor forest health, and predict landslides. “The data isn’t just telling us what the landscape looks like,” says Driscoll. “It helps us understand how it changes over time.” This helps communities make more informed decisions to preserve habitats and rebuild homes. Landscapes cascade, but they also recover. Days after the Sorrento Fires, small patches of sage and chaparral were already pushing through the ash. Fire has always shaped California. It destroys, but it also renews.