California County Hopes Artificial Intelligence Can Mitigate Wildfire Risk

May 10, 2021 by

At this time of year, periodic rain showers on the north coast of California give way to months of daily sunshine and a wildfire risk that grows in severity until the next fall rains arrive.

In Sonoma County, a new set of eyes is watching over the forest. Those eyes will be able to tap into an artificial intelligence program to make sure emergency dispatchers are alerted to actual fires instead of mist rising off the forest floor or steam from the region’s numerous natural geysers.

The county has entered into a $300,000 contract with South Korea technology firm Alchera to provide artificial intelligence software that can alert fire dispatchers to the precise location of flames or smoke.

The two-year pilot project is funded through $3 million in hazard mitigation grants that the Federal Emergency Management Agency awarded to the county. The money will also be used to add 27 cameras to the state’s AlertWildfire system, said county Emergency Management Director Chris Godley. The new cameras will be erected on 17 towers to increase coverage in heavily forested Sonoma County and parts of neighboring Mendicino and Lake counties.

Godley said he expects that fire dispatchers will have to spend some time “teaching” the AI system how to discern smoke from the county’s numerous geysers, which are used to produce geothermal energy. He said he expects the learning period to end by this fall, at which point the alert system will be fully operational.

He said the Alchera system sends an alert to dispatcher when it detects smoke. The dispatcher teaches the AI how to better discern images by entering a simple “yes” or “no.”

The system also transmits an image of the fire through the dispatcher’s camera. A blue box appears over the suspected wildfire smoke.

Godley said he expects the system cut at 20 to 30 minutes from the usual time fires are detected and reported, and up to an hour at night time.

“That extra 20 minutes can make all the difference,” he said.

The Tubbs fire in October 2017 destroyed more than 5,000 homes in Sonoma and Napa counties and killed 43 people. From 1964 to 2015, Sonoma County experienced 18 major wildfires that destroyed 2,000 structures, according to the county.

Insurance losses caused by wildfires have increased dramatically in the past decade. According to a report by Swiss Re, wildfires have caused a cumulative $56.3 billion in insured losses from 2011 to 2020. By comparison, wildfires caused a cumulative $8.7 billion in losses from 2001 to 2010, $5 billion from 1991 to 2000 and $1.4 billion from 1981 to 1990. All of those figures are converted to 2020 values.

California is investing in wildfire resilience projects to stem the destruction. On April 13, Gov. Gavin Newsom signed into law a bill that provides $536 million in funding to support wildfire suppression, improve forest health and build community resilience. FEMA said in February that is provided $103 million in disaster assistance grants to California after the state experienced another disastrous fire season in 2020.

Advanced fire detection systems can address the problem by getting firefighters to the scene quicker. But false alarms are a nagging problem.

Bow Rodgers, US vice president of operations for Alchera, said despite Godley’s concerns, the AI system will have no problem telling wildfire smoke from geyser steam. He said Alchera has gained significant experience over the past two years through a proof-of-concept trial usingAlertWildfire’s network of cameras. The system, operated by the University of Nevada, Reno; the University of California, San Diego; and the University of Oregon, operates about 650 cameras mounted on towers scattered across California, Nevada, Oregon, Idaho and Utah.

While Sonoma County is so far the only jurisdiction that has contracted with Alchera to send alerts to dispatchers, Rodgers said the company is in discussions with the state’s investor-owned utilities with hopes of expanding the alert system throughout California.

Rodgers said Alchera got its start by creating facial-recognition software. He said the ability to detect visual anomalies can be used for a variety of purposes. Alchera’s first major deployment was through a contract with Korea Electric Corp. that is used to spot worn power transmission lines.

A 2020 research paper by the Center for Research and Technology in Greece says that scientific interest in remote sensing systems to detect forest fires has exploded in recent years. While fewer than 10 academic articles on the subject were published each year for most of the 1990s, the number of papers started increasing in 2000 and surpassed 200 in 2019.

Several commercial applications, in addition to Alchera, have emerged.

In Germany, 105 cameras deployed by IQ Fire Watch are trained on the Brandenburg National Forest, a vast stretch of pines that has the highest risk of wildfire in Northern Europe. The system sends alerts to a command center when it detects smoke, creates a map showing the location and transmits information about nearby water sources to firefighters, according to the company.

“Our system can be used for any kind of forest,” IQ Fire Watch spokeswoman Victoria Boehm said in an email. “It detects smoke and smoke-like occurrences with a multi-spectral sensor at an early stage and processes the information in real-time with a feature-based algorithm in combination with artificial intelligence.”

Insight Robotics, based in Hong Kong, has installed networks of thermal cameras to detect wildfires in China, Mexico, Spain and Portugal, the company says. Insight says it uses a patented software algorithm that extends the range of thermal sensors beyond the typical limit of a few hundred yards and minimizes false alarms.

Omdena, a global technology firm based in Palo Alto, Calif., boasts that a partnership it formed with Brazilian agriculture tech provider Sintecsys developed an artificial intelligence program that was able to correctly identify from images 95% to 97% of real fire outbreaks with a false positive rate of 20% to 33%.

Scientists at the University of Michigan built a fire detection system that is powered by harvesting the kinetic energy of moving tree branches. The self-powered sensing system designed by MSU’s Laboratory of Soft Machines and Electronics are equipped with carbon monoxide and temperature sensors to reduce the likelihood of a false positive, according to an article in MSU Today.

The machines — officially called triboelectric nanogenerators — produce enough electricity to send a new message every three minutes, head researcher Changyong Cao wrote for a paper published in Advanced Functional Materials.

Cao said sensing technologies to detect wildfires are becoming more common, but often they are depend on batteries. Solar power isn’t an option in a dense forest. He said a network of sensors spaced about 100 meters apart or more could provide full coverage to a forest. The nanogenerators provide enough power to send a report every three minutes.

The new fire-detection systems augment monitoring that has been conducted using satellites operated by the National Aeronautics and Space Administration since 2007. NASA’s Terra and Aqua satellites orbit the earth in a north to south pattern at alternative times during the day. Together, they can view the entire earth’s surface every one to two days. The systems can serve the same feature on the ground two four times every 24 hours.

About the photo: A firefighter from San Matteo helps fight the Kincade Fire in Sonoma County, Calif., on Sunday, Oct. 27, 2019. (AP Photo/Ethan Swope)