Data Science Investment Counter —Funding raised by UK data science companies in 2018.
£ 5.640 Million

Researchers and forest managers use AI to predict wildfires

A base firefighter communicates with other firemen during a wildfire sparked by lightning near the Grandview Gate in 2007. Credit: U.S. Air Force photo/Airman 1st Class Ryan Crane.

Forest managers and scientists are testing Artificial Intelligence (AI) programs to predict wildfire risks across the US.

According to Mike Flannigan, director of the Western Partnership for Wildland Fire Science at the University of Alberta, research on this has been extensive lately, with over 150 papers being produced on the matter.

“It is definitely front and centre in terms of the research agendas in terms of wildland fire and will continue to be for the next years,” he said.

With the aid of AI technologies, he added, fires are tackled before they get too large, by getting equipment and crews to the right place to fight them in the early stages. Once truly ablaze, they are difficult to contain or stop.

“Once the fire gets to be a crown fire and it’s two football fields or larger, it’s nearly impossible to put it out until the weather changes,” Flannigan said. “You’re spitting on a campfire.”

Last year, Flannigan worked with Ryan Lagerquist, a PhD student at the University of Oklahoma’s School of Meteorology, to create an AI program able to go through high-intensity wildfires data and predict where extreme weather is most likely to create the conditions for a forest fire.

Commenting on current efforts in predicting wildfires, Balz Grollimund of insurance firm Swiss Re said that occurrence and severity of fires are hard to foresee. Droughts or forest conditions can be easily monitored, but ignition factors are often entirely random events such as lightning strikes.

All these things are very tricky with wildfires,” Grollimund said. “We’re trying to anticipate where wildfires will occur.”

According to him, AI would be well-suited to find patterns in that apparently unrelated mass of data.

“You start with your observations. What have you seen in the past decades in terms of where wildfires have occurred and how big they got? And you look for correlations with any factor that might have any impact.

“The question is which data really does have any correlation. That’s where the AI comes in play. It automatically figures those correlations out.”

Grollimund said he’s helped scientists at the Massachusetts Institute of Technology develop an AI intelligence program that can predict fire risk as far as six months out.

They tested the program by feeding data from November 2015 from across Canada and concluded that the great majority of fires that did occur in April and May of 2016 happened in the high-risk zones identified by the program.

However, both Flannigan and Grollimund agree that the efficiency of those methods largely depends on climate change’s future developments.

“With climate change, we’re seeing conditions and situations that have no real analogue in the recent past,” Flannigan said.

 “A lot of the factors that foster wildfire risk seem to increase — longer, hotter, dryer summers; wetter winters; more vegetation; more lightning.

“There’s a lot of reasons why we think, if anything, (fire risk) is going to be increasing.”


Co-working space and blog dedicated to all things data science.

Subscribe to our newsletter