Sonoma County turns to artificial intelligence to spot wildfires with lookout cameras

North Bay fire lookout cameras

For live images from lookout cameras in the region, visit www.alertwildfire.org/northbay

Sonoma County will bolster its nascent network of fire-lookout cameras with artificial intelligence that aims to automatically identify potential wildfire starts and provide alerts even when no one is watching.

County officials announced the program Wednesday after awarding a $300,000 contract to Alchera, Inc., a South Korea-based company that develops algorithms for visual artificial intelligence systems.

The technology, which is promising but still in development, is meant to automate Sonoma County’s alert-and-warning efforts to provide more of a heads-up in case a wildfire starts, said Chris Godley, the county’s emergency management director.

“This is really designed to help us catch those extremely early starts, so it gives us that much more time to investigate and, if need be, respond,” Godley said.

Most of the funding for the new technology comes from a $2.7 million grant the county received from the Federal Emergency Management Agency, with the county chipping in about $75,000.

The remaining $2.5 million in grant money will be used to improve the underlying camera network’s resilience and performance, Godley said. The grant funding also will pay to bring the number of fire cameras in the county up to 26 by adding six new cameras: a second camera at the Oak Ridge site in northwest Sonoma County and one each near Guerneville, Fort Ross, Sea Ranch, the Sonoma Raceway and Fish Rock (which is in Mendocino County but will show images in Sonoma County).

The local network has sprung up in the wake of the 2017 firestorm, when wind-whipped nighttime infernos stormed into Santa Rosa and Sonoma Valley with little to no warning while other fires overtook homes in Mendocino, Napa and Lake counties. Now, dozens of lookout cameras dot many of the region’s flame-torched ridges and mountaintops, giving authorities and residents alike a head start when fires break out in remote spots.

Statewide, more that 600 such lookout cameras are now in operation.

In this image from an AlertWildfire camera, the 13-4 fire (later renamed the Waldbridge fire) burns in northwest Sonoma County, Tuesday, Aug. 18, 2020. (alertwildfire.org)
In this image from an AlertWildfire camera, the 13-4 fire (later renamed the Waldbridge fire) burns in northwest Sonoma County, Tuesday, Aug. 18, 2020. (alertwildfire.org)

Alchera’s fire-spotting algorithm already has been integrated into the existing ALERTWildfire camera network spanning Sonoma County. It scans the images provided by the cameras for signs of smoke. Once a potential start is detected, key personnel can be notified via email or text to check out the situation and confirm or deny the existence of a fire.

Alchera said its technology has been trained on millions of images over the past few years and has detected “hundreds of early-stage wildfire ignitions” since June 2019.

“With our cutting-edge AI-based wildfire detection system, we are able to help reduce the damage of forest fires in California, which, as we have seen, suffers greatly from this phenomenon every year," said Young-Kyoo Hwang, Alchera’s CEO, in a statement.

However, the system is not expected to be able to automatically provide alerts until November due to the need for “training and modifications,” according to the county.

On Wednesday, small planned burns across the county had already generated 16 hits, Godley said. For one such event — a permitted burn northeast of Healdsburg spotted via a camera in the Pepperwood Preserve — the system created an image and a short video showing a blue rectangle around the area the algorithm believes a fire has started.

Because the system is not fully proven, county staff will be manually validating each automated alert over the summer to verify that a notification for a fire does in fact represent a fire — and not a false alarm, like steam from The Geysers geothermal power plants, Godley said.

Other California jurisdictions such as Palo Alto and Napa County have explored or implemented this technology already, Godley said, though the number of agencies that have added artificial intelligence to their fire-spotting efforts remain “relatively few and far between.”

Sonoma County expects to use Alchera’s algorithms through the 2021 and 2022 fire seasons before the county reevaluates the technology for possible future use.

The technology is no substitute for diligent brush management and other fire prevention efforts, but it does address “the basic principle that we get notified about fires as soon as possible,” said Marshall Turbeville, the Northern Sonoma County Fire District chief.

Turbeville said the automated system wouldn’t do much more to help officials spot fires near populated areas during the daytime. But it will come in handy in remote places where potential fire-spotters are few and far between and overnight when most people are asleep, he said.

“I see that being the biggest value,” Turbeville said. “Maybe at 3 in the morning, not too many people are awake and nobody’s going to see the fire.”

The artificial fire-spotting tool also could be useful during weather events like last August’s lightning storm, with numerous small fire starts and poor conditions limiting aerial reconnaissance opportunities, Turbeville said.

“I think it’s great,” he said. “It has a little niche that it’s filling.”

You can reach Staff Writer Will Schmitt at 707-521-5207 or will.schmitt@pressdemocrat.com. On Twitter @wsreports.

North Bay fire lookout cameras

For live images from lookout cameras in the region, visit www.alertwildfire.org/northbay

Show Comment