DNR pilot program places high-definition cameras around state
People are turning to artificial intelligence (A.I.) for everything from research papers to crime investigations — and now the technology is being tested for its ability to detect wildfires.
The Washington Department of Natural Resources (DNR) is running a pilot project using A.I. cameras and technology in a partnership with Pano, a company based in San Francisco. Pano has installed high-definition cameras at a dozen stations around the state, including one on Aeneas Mountain near Tonasket. There will be 21 stations by the end of June 2024, DNR Assistant Division Manager for Plans and Information Angie Lane said at a DNR press conference last week.
Pano representatives also made their pitch to the Okanogan County commissioners and staff. The DNR pilot project includes a second station on Little Buck Mountain, near the Loup Loup summit, which will be installed in coming months, Pano Director of Government Development Kat Williams told the commissioners in August. DNR selected locations that are considered high risk based on fuels and topography.
Each Pano station has two cameras that rotate 360 degrees. The A.I. technology works in concert with standard 911 dispatch, interagency coordination centers, and thermal imagery from satellite feeds to ground-truth findings by the cameras, Williams said.
The cameras notify a dispatch center when they pick something up, which is treated like any report of smoke or fire. The A.I. system is continually learning from interactions with human fire experts, which improves the system and teaches it about false positives, Lane said.
The cameras, which need to be within 10 miles to detect a smoke plume or fire, can distinguish “texture” even against an overcast or smoky sky. The stations rely on cell signals to transmit information and, in remote areas — like much of Okanogan County — on Starlink satellites, Williams said.
Mike Worden, chief deputy of communications with the Okanogan County Sheriff’s Office, was concerned about having adequate bandwidth to transmit imagery.
Williams, who worked as a wildland firefighter before joining Pano, stressed the importance of early detection — knowing from the first instant what a fire is doing, where it’s going and its rate of spread. A.I. can also provide key information about the potential need for evacuations and nearby water supplies available for bucket drops and fire engines, she said.
Multiple users can view the data from the Pano cameras at once and can zoom in on details like fuel type and water sources. They hope A.I. detection can help identify the types of firefighting resources that are necessary — such as bulldozers, smokejumpers or aircraft — and deploy them as soon as possible, Williams said.
At the DNR press conference, Commissioner of Public Lands Hilary Franz pointed to a situation in Okanogan County to highlight the importance of deploying new technology in fire detection. When smoke was spotted in difficult and treacherous terrain in the county, a firefighter had to hike 3 miles to investigate before they ordered smokejumpers. Ideally, using cameras would enable incident managers to understand the topography and trajectory of a blaze and to send appropriate resources as quickly as possible, she said.
Building more robust fire-detection systems is increasingly important as the reality shifts from a fire season to a “fire year,” Williams said.
The discussion with the county commissioners and staff illuminated some of the challenges of the emerging technology. Because the Pano cameras were 13 miles away and over a ridge, they didn’t pick up the Eagle Bluff Fire that started near Oroville in July until it had been burning for some time, Williams said in answer to a question from Okanogan County Emergency Manager Maurice Goodall.
Goodall asked how high smoke needs to rise above mountains before the cameras detect it, and how effective the cameras are in heavy smoke conditions. He also asked Pano for data that compares fire detection by the A.I. system and by human reports to 911 dispatch.
Pano is still refining its technology and learning from firefighters and analysts on the ground, Williams said. The cameras can’t always distinguish smoke from dust and, in mountainous regions, a smoke plume needs to be high enough to be visible.
In areas with active construction where heavy equipment could kick up a lot of dust, the A.I. technology can learn to identify suspicious plumes as false positives. The A.I. cameras can also be used during prescribed burns, where they can be taught to differentiate between smoke within the perimeter and a situation that would require intervention by a fire crew, she said.
Pano installs the camera stations on a variety of structures, including cell, water, and fire-lookout towers. Ideally, the stations connect to the electric grid, but the cameras can run off solar power. They’re equipped with heaters to melt snow and ice and a wiper to clear the camera screen, Williams said.
While the technology could add a potentially useful fire-detection tool, and sites could be available in the county, the cost — $50,000 per station per year — is prohibitive for Okanogan County, Worden said.
Pano presented a map of potential sites for stations in the county based on their risk analysis. None of the proposed stations is in the northern part of the Methow Valley, where a lot of wildfires start, Okanogan County Commissioner Andy Hover said. Since 83% of Okanogan County is public land, it would be worth exploring the potential for partnering with the state or federal agencies to pay for additional sites, he said.
Pano has stations in six western states, but DNR is the first government agency to test the technology, according to Pano CEO Sonia Kastner.
It’s rare for a fire to burn even for a day before it’s reported, but it can be a challenge to find the fire, and A.I. might help identify the location, Worden said. The key question is, how much time is saved by A.I. detection and reporting versus reports to 911 dispatch, Worden and Goodall said.