The UN recently reported that the risk of devastating fires could increase by over 50% by the end of the century, mostly due to climate change. In the battle against these blazes, the most important time is the very beginning; it’s easier to put out a small fire than a large one. But it’s tough to spot a small wisp of smoke in a huge landscape under ever-changing conditions.
A number of tech startups are deploying AI-enabled cameras to bolster firefighting efforts, giving first responders a critical edge in snuffing the flames before they turn into all-out natural disasters. Many of them are based in California, which had its worst wildfire year in recorded history in 2020, with over 4% of the state burned.
In 2019, Illumination Technologies California (ITC) deployed two camera systems on hilltops near its headquarters in Calistoga, CA, in Napa Valley. The area is famous for wine, but recently it has been in the news for a different reason: in 2020, wildfires burned over 40 percent of Napa County.
The fire detection system, made the German company IQ Technologies, is already used in 350 locations around the world, but this was its first application in the U.S. IQ Firewatch uses multispectral optical imaging technology originally developed for the German space program.
The cameras rotate continuously, scanning the countryside with color, monochrome and near-infrared detectors. A feature-based AI algorithm uses an artificial neural network to scan the images for the telltale heat and smoke signature of wildfires. Under the right conditions, it can see as far as 40 miles away.
The algorithm processes the data in real time on dedicated servers on site. Using historic data stored in the cloud as a comparison, the system alerts a human operator to the most suspicious events, who then decides whether to contact the fire department. It’s like a 24/7 version of a human fire watcher in a forest tower, just with much better vision under a wider range of conditions – including after dark, thanks to the near-infrared detector.
“Fires are challenging to detect because there are so many intricate variables: topography, light conditions, weather,” said ITC’s CEO Chris Eldridge.
“We always want to ‘over-detect’ events and then apply the AI to limit the number of false alarms.”
He said the detection algorithm is constantly teaching itself to be more accurate.
The Napa IQ system spotted a fire that started at night in 2019, and in September 2020 it gave the first warnings of what grew into the Glass Fire Complex, which eventually burned 100,000 square miles over 23 days. A third camera was installed in 2021, doubling the coverage area. (The county pays a subscription of $6,000 per camera per month.) Since then, the system has processed over three dozen suspicious events and forwarded at least 13 to the Napa County Fire Department.
ITC is already in discussions with other counties as far away as Lake Tahoe, Eldridge said.
Another automated fire detection system designed by San Francisco-based startup Pano AI is already in use in over two dozen locations in California, Oregon, Montana and Colorado. The Pano system also employs rotating ultra-high-resolution cameras perched on mountaintops, with a 15-mile range and 45x optical zoom.
The system, based on open-source Performance Co-Pilot (PCP) software, uses a cloud-based deep learning AI to detect and verify wildfire events in real time, drawing from satellite imagery and historical data.
“Wildfire smoke behaves very differently from things like clouds, industrial plants and geysers,” says Pano CCO Arvind Satyam.
“It’s not hard to see on a bluebird day, but when it’s 15 miles out against a hazy backdrop, it’s a real challenge.”
The current Pano model can detect 70% of wildfires in under 5 minutes, he said.
Once it spots a potential fire, the system sends auto-centered images with enhanced zoom to the professionals in Pano’s dedicated intelligence center. The Pano system also provides firefighters with up-to-date time-lapse images and data on how the fire is progressing as they fight it.
Speed is key, Satyam said.
“Whenever there is an incident, the faster you can determine its location, and the more intelligence you can gather to determine if it’s a real threat or not, the better.”
Wildfires can blow up in minutes, but it can take hours for first responders to reach blazes reported by bystanders.
Meanwhile, San Francisco-based Firemaps is building a product to help individual households take steps to protect themselves from fires. Based on a customer’s home address, Firemaps analyzes satellite imagery and other fire risk data to automatically generate a personalized fire defense plan for the property, including things like brush removal, creating fuel breaks and fireproofing structures.
These efforts may not slow the effects of climate change, but they could help dampen the effects of wildfires which seem to burn with greater ferocity and frequency every year.
“We’re probably going to be in a year-round fire season before long,” Eldridge said. “You don’t want to miss a single event.”
Julian Smith is a contributing writer.
© 2022 Nutanix, Inc. All rights reserved. For additional legal information, please go here.