Camp Fire shows need for early detection: How tech and artificial intelligence can help


The Mercury News



The fast-moving Camp Fire caused traffic gridlock as Butte Country residents tried to evacuate last week, leading some of them to abandon their vehicles as the fire grew closer to the roadways.

Had the now-deadly fire been detected earlier and officials been able to give more notice, some residents might have been able to avoid gridlock — or maybe even death.

How far can technology go to help automatically detect and even predict wildfires? From coast to coast, computer scientists, researchers and others are hoping to make it go further.

Early detection is key as wildfires have gotten worse in recent years, according to Jim Crawford, assistant chief of South Bay operations for Cal Fire’s Santa Clara unit, who has been a firefighter for 26 years.

“The job I signed up for 26 years ago is not the same job today,” he said. “In the 1990s, firefighters battled a normal large fire for a couple of weeks. Any more than that was considered a ‘career fire.’ Now we’re having career fires multiple times a year.”

The Camp Fire — which as of Monday before noon had burned more than 113,000 acres and destroyed thousands of structures — is one of more than 5,600 wildfires that have burned more than 829,500 acres in California so far this year.

Fire agencies mostly find out about fires from 911 calls — which is what happened with the Camp Fire, according to Cal Fire Capt. Scott McLean. That fire was first reported at 6:29 a.m. Thursday, but McLean said Friday that Cal Fire did not know how long it had been burning by then because it started in a hard-to-find area.

Fire agencies also monitor cameras placed throughout the state, including on undisclosed Bay Area mountaintops. Some of the cameras are aided by detection algorithms and satellites, but the systems are not quite real-time and the resolution could be better, experts say.

“Lots of people are looking into using artificial intelligence to figure out fire location, spread and behavior,” said Craig Clements, associate professor at the Fire Weather Research Laboratory for the Department of Meteorology and Climate Science at San Jose State University.

For example, there’s the Geostationary Operational Environmental Satellite (GOES) Early Fire Detection System, a collaboration between the Center for Spatial Technologies and Remote Sensing (CSTARS) at UC Davis and the U.S. Forest Service. It uses National Oceanic and Atmospheric Administration weather satellites and fire-detection algorithms.

CSTARS scientist Alex Koltunov, who has a Ph.D. in remote sensing, has been working on and off on the system’s algorithms — sets of rules programmed to perform calculations or carry out certain tasks — for the past several years, with the goal of optimizing data processing to help detect ignitions as early as possible.

He and his colleagues have been using GOES-15, a satellite that was launched in 2010, to conduct testing using past images and wildfire-incident reports over many states to help fine-tune the algorithm, which he hopes to one day apply to GOES-16 and GOES-17. Those satellites, launched in the past couple of years, can provide higher-resolution images “every 5 minutes or sometimes every minute,” Koltunov said. Higher-resolution images should allow firefighters and others to detect wildfires more quickly and accurately.

But he pointed out that no single tool can be a “silver bullet” when it comes to detecting wildfires.

“Each method has its limitations,” Koltunov said. When it comes to algorithm-based detection, which can detect something as small as an individual structural fire, “how good is the data? Is the fire a barbecue in a backyard, or a campfire? Some clouds can be confused with wildfires, too.”

(For more…)