In 2024, a brush fire broke out near Cherry Grove, Oregon, a rural community southwest of Portland with more than 300 residents and sometimes spotty cellular coverage. The fire might have burned unnoticed until a passerby saw the smoke—or, worse, flames—and then found cellular service to report it. By then, the fire would be established, and a potential threat to Cherry Grove.
Fortunately, that didn’t happen. An AI-powered camera saw the smoke first.
“The camera had already notified Portland General Electric (PGE) and all the emergency management agencies a half an hour before the first 911 call came through,” says John Farmer, a spokesperson for PGE. “I’ve been out there. Your phone literally doesn’t work.”
PGE has deployed 39 AI-enabled wildfire detection cameras across its service area since 2021 (all of which are accessible online). Most of the cameras are concentrated in high-risk terrain spanning from Mount Hood in the east to Newberg in the west, with larger cities like Portland, Salem, and McMinnville in between.
Born Under an Orange Sky
The cameras are from Pano AI, a San Francisco company that monitors more than 70 million acres across 18 U.S. states. The company also has cameras deployed inAustralia and Canada.
Pano AI closed its first round of seed funding during San Francisco’s “Orange Skies Day” in September 2020. Smoke from roughly a half-dozen wildfires poured in from the north and east, creating a choking, orange overcast. The smoke was so thick that many light-sensitive streetlights stayed on throughout the day.
“Our founder, Sonia [Kastner], had got together with one of her good friends, Arvind [Satyam], and they talked about why there’s not a better way to early detect and treat wildfires,” says Bandar El-Eita, vice president of marketing at Pano AI.
El-Eita says the co-founders pondered concepts such as drones, blimps, satellites, and lasers before deciding on a rotating panoramic camera as a practical, reliable, and affordable approach. . The first prototype was installed in San Francisco. “It sort of finagled its way onto the roof of Sonia’s apartment building,” says El-Eita.
PGE’s investment in Pano AI’s cameras followed the 2020 Labor Day fires near Mount Hood, which struck simultaneously with fires in California. In Oregon, the fires caused substantial damage to rural communities and placed approximately 100,000 residents under a Level 3 “Go Now” evacuation order.
“That was a big event for Portland, for Oregon in general,” Farmer said. “It catalyzed decisions on future innovation and investments.” PGE became one of Pano AI’s two pilot customers in 2021, alongside Pacific Gas & Electric Company (PG&E), through an innovation program run by the Electric Power Research Institute.
Two Cameras, One Minute, Ten Miles
Pano AI’s camera stations deploy cameras in pairs. “Each camera is capturing 180 degrees every minute,” explained El-Eita. “So it does one rotation per minute with a stated range of ten miles,” though El-Eita adds that the cameras often detect fires farther out.
The dual-camera deployment is used for two reasons.
El-Eita says this provides fire spotters with options if they need to manually review an AI-reported wildfire. One camera can be focused on the suspected fire, while the other continues scanning the terrain. Using two cameras helps Pano AI deduce the exact location of a suspected fire by comparing the slightly different images captured from each camera.
Pano AI’s technology analyzes every frame of every rotation with an AI image model trained on more than 4 billion images captured by the company’s own cameras. Pano AI actually has two distinct models—one for day and one for night. The daylight model is trained to detect smoke rising higher than 100 feet. Near-infrared imaging is used at night, and it’s supplemented by commercial satellite data from third-party partners.

Importantly, El-Eita says the AI image model is trained to look for smoke, not flames. That’s necessary for early detection, as smoke is visible at a distance far before flames can be seen. However, it also makes false positives more likely.
During an early deployment of Pano AI’s cameras in Aspen, Colorado, the system flagged a rash of apparent fires across a cold November landscape. “When we looked at that, we realized it was snow blowers that were getting the mountain ready for ski season,” El-Eita said. The experience taught Pano AI to provide the option to map exclusion areas around known false-positive sources, like factory smokestacks.
Farmer says PGE has experienced the same challenge in its own deployments. While Oregon authorities put burn bans in place during peak fire season, rural burns are otherwise common and, unlike a factory smokestack, they don’t always occur in predictable areas.
Despite that, Farmer says Pano AI’s models have become better over time, learning that “a plume of smoke at such a size, at this range, at this location, is not likely wildfire smoke.”
AI Detection is Effective, But One Tool Among Many
Pano AI has built a history of fires that the company cites as proof for its technology.
The Hearst Fire, which a Pano AI’s camera caught early in an uninhabited area, would have reached an estimated 340 acres by the end of day one and more than 5,000 acres by day two, threatening 44 structures, according to fire-spread modeling by Pano AI partner Technosylva. Pano AI’s camera provided an alert before 911 calls and the fire was held to two acres.
911 calls also tend to be vague in ways that a camera is not. The first humans to lay eyes on a fire may not be trained to spot them and, in rural areas, may not have the cellular coverage necessary to call in the fire while it’s in sight. This can lead to imprecise reports, forcing dispatchers to work out the exact location before committing resources. A camera alert with coordinates removes several in-between steps, so firefighting agencies can skip straight to response.

With that said, both PGE and Pano AI were quick to stress that AI wildfire cameras aren’t a replacement for human experience.
Farmer says that PGE’s wildfire operations team includes former firefighters, a fire behavior analyst, and two meteorologists, and he’s careful to frame the cameras as one tool among many. The cameras are “incredibly helpful,” he says, but “the other critical piece is human eyes on the situation, and the human interpretation and judgment.”