Meteorologists can reasonably forecast hot weather and lightning storms, but exactly where lightning will strike — and whether it will spark a wildfire — is nearly impossible to predict.
That’s why researchers are turning to machine learning in a bid to get ahead of catastrophic wildfires that they say are becoming more common and more severe.
“These machine learning approaches are even better than we are at seeing patterns that we might not see because they have much more processing power than the human brain,” said Mike Flannigan, a professor of wildland fires at the University of Alberta.
Changing weather patterns are making Canada’s forests more vulnerable by bringing warmer temperatures, drier conditions and more lightning storms that spark wildfires. Meanwhile, a reduction in controlled burns have left forests full of fuel.
As a result, wildfire seasons across the country are starting earlier and the blazes are more intense.
“In Canada, we have about 2.5 million hectares of forest burn every year. That’s about half the size of Nova Scotia,” Flannigan said, adding approximately 1 million hectares burned each year in the 1970s.
On Saturday, more than 100,000 hectares outside of High Level, Alta., had been charred due to an out-of-control fire that started early last week.
Firefighters are focusing efforts on containing the wildfire outside High Level. Yesterday a break in conditions allowed a controlled burn of some of the area between the wildfire and the town to consume materials that might have become fuel for the wildfire. <a href=”https://twitter.com/hashtag/ABFire?src=hash&ref_src=twsrc%5Etfw”>#ABFire</a> <a href=”https://t.co/90YrlylOa7″>pic.twitter.com/90YrlylOa7</a>
Using machine learning, Flannigan hopes to provide wildfire agencies with data about where fires, like those in Alberta, might start in hopes of mitigating them.
‘It’s not a panacea’
There are more than 150 academic papers focused on using machine learning to predict fires, Flannigan says.
“This is an emerging field. Machine learning, AI, is in vogue, and we’re just one more field that’s applying [it].”
Though his technology hasn’t been tested yet in the field, the U of A professor says it shows promise.
The software is a neural network — a system that works similarly to the human brain — that incorporates historical weather conditions which led to wildfires with traditional meteorological data like precipitation, temperature, wind speeds and humidity.
By also examining the pressure systems that determine wet and dry weather, the artificial intelligence system can predict areas which might face weather conditions that could spark a blaze.
“It’s not a panacea,” Flannigan warned. “It’s not going to solve all our problems, not the silver bullet, but in some areas it can be very beneficial.”
Fire-resistant, not fireproof
Fires are more frequently breaking records, particularly in the western provinces.
Last year, wildfires across B.C. scorched so much land, it broke a provincial record set only the year before — a record “heads and shoulders” above the previous record set in 1958, Flannigan says.
He attributes the larger blazes to climate change and Canadians, he says, should get used to a “new reality.”
“We’re on a downward spiral and trajectory, and so it’s dynamic and changing,” he said.
Communities can take precautionary measures by building fire breaks, like green spaces, to keep high-intensity flames at bay. When it comes to safeguarding homes, Flannigan says that property owners can make a few small changes.
Clearing debris from eavestroughs and storing firewood away from your home is key, he says.
If mulch and peat moss used in landscaping are dry, they can be a hazard as well.
“We can’t make our communities fireproof, but we can make them more fire resistant.”
With files from Samantha Lui