On June 17, early in the fire season in Washington state, a blaze ignited acres of undeveloped land in Mason County, where people are dispersed amongst the Olympic Peninsula’s dense forests and jagged peaks. No one called 911 to report the burn. The local fire department had no idea their jurisdiction was on fire.
But within hours, the fire was out, burning just under 20 acres in total.
The covert eyes of two AI-trained cameras on nearby vantage points relayed a warning to dispatch centers far afield. The cameras, developed by artificial intelligence startup Pano AI, are able to detect plumes of smoke, the earliest sign of a wildfire in the making.
After confirming the blaze, Pano AI called Washington’s Department of Natural Resources (DNR), which sent firefighters to the site. Coordinates from the cameras guided them to within 100 feet of the ignition source.
“That fire would have gotten a lot bigger before it attracted any kind of attention,” said Thomas Kyle-Milward, DNR’s wildfire communications manager.
And speed matters. “The faster you get on a fire,” he added, “the smaller you can keep it.”
DNR launched a pilot program with Pano AI last year, placing 21 cameras at sites around Washington where wildfire risk is high and the chances of a person reporting it are lower, based on historical data and models. DNR has extended the contract through 2029. The cameras cost around $45,000 each, for a total annual price tag of $948,000 after the pilot.
Washington DNR is not the only one benefiting from extra eyes. Pano AI has cameras in 12 states and provinces across the U.S., Canada and Australia covering 16 million acres, according to an April press release. Other customers include utility companies Xcel Energy and Austin Energy, private landowners and government fire agencies.
The San Francisco-based company raised $45 million from investors as of March 2023 and employs 45 people, reports PitchBook.
In Washington, Pano AI has picked up several ignitions this summer before they were visible on satellite imagery.
Matthew Dehr, DNR’s lead fire meteorologist, said Pano AI detected a now-contained fire near Yakima in Central Washington before he could see it, and the technology currently is allowing him to monitor other blazes firefighters are struggling to contain.
Currently 55 large fires are burning in Washington and Oregon, according to the Northwest Coordination Center. A fire near Chelan in Central Washington that began in early June has already consumed over 33,000 acres. With fire season far from over, Dehr said it’s difficult to compare this year to previous ones, but he predicts the next month or two will be “very challenging.”
Fighting fire with force
A recent report from the civic organization Challenge Seattle showed that the number of fires in Washington is not changing, but they’re burning more acres. In the past two decades, the average number of acres burned by wildfire each year has increased four-fold.
The same report found that humans are responsible for more than 80% of the wildland fires started in the state.
In the past decade, DNR has shifted its response strategy to prioritize a strong attack on fires at the outset, aiming to contain 95% of fires to fewer than 10 acres. In Washington, where the average number of annual wildfires hovers around 1,500, keeping fires small could be a promising strategy for dealing with the consequences of climate change.
Dehr said the cameras are a huge benefit to his job, which involves forecasting fires and tracking them to guide the response.
“The goal is really to help preposition resources,” said Dehr, who was formerly a weather officer with the U.S. Air Force. “My duty is to keep firefighters safe and to get firefighters to the right place ahead of a fire start.”
The Pano AI system is just one tool in an increasingly sophisticated suite of new firefighting technologies.
Less than a decade ago, DNR had fewer than 10 aircraft. Now, there are more than 40 fire-designated aviators in the fleet, including two new Kodiak planes equipped with military-grade surveying tools that use infrared technology to map heat, which helps personnel draw a footprint around active fires and identify new ones.
DNR also relies on remote, automated weather systems to assess fire danger. Data from hourly satellite reports on solar radiation, relative humidity, precipitation, temperature and wind from 30 stations statewide are routed through risk calculators that spit out information on threat levels. Fire crews also report their observations from the field.
And the University of Washington runs an ultra-high-resolution weather model that accounts for the state’s variable topography. Steep, jagged peaks and deep valleys contort the landscape. “If your resolution isn’t high enough,” Dehr said, “it just washes all that out.”
While the technology and data bolster firefighting responses, an effective effort still requires human input. The Pano AI feeds go through human quality control, DNR dispatchers coordinate statewide efforts, and the survey planes have sensor operators with special training to operate their cameras. One of the department’s challenges is finding enough people with the necessary skills to collect data around the clock.
Taking advantage of the tech boom means continuing to invest in training programs for employees at all levels, Dehr said.
“You still need people,” he added, “and you need smart people.”
