This article is over 5 years old and may contain outdated information.


How artificial intelligence could help predict major forest fires

Researchers at the University of Alberta have trained a "self-organizing map" to identify high-risk days for fires 

  • Sep 07, 2017
  • 738 words
  • 3 minutes
Expand Image

Last spring, Canadians watched, horrified, as residents of Fort McMurray, Alta. were forced to flee their community as a nearby forest fire exploded out of control. Now, as neighbouring British Columbia continues to grapple with its worst wildfire season in recorded history, forest science researchers at the Universities of Alberta and Oklahoma say artificial intelligence may be key to predicting where the next major fire will start.

A paper published last month in the Canadian Journal of Forest Research describes the researchers’ efforts to train a computational model to predict extreme “fire weather” — the combination of prolonged hot, dry and windy conditions that leads to the biggest and most destructive fires. The model, called a self-organizing map (SOM), uses neural networks that mimic the processes of the human brain to study data sets, and over time, learns to recognize patterns. In this case, the map learned to recognize large-scale atmospheric pressure variables associated with hot, dry and windy weather on the surface.

SOMs have been successfully used in other meteorological studies, including efforts to broadly predict the impacts of climate change, but according to the researchers, this is the first time anyone has developed a SOM that could be used to make critical fire management decisions in realtime.

“Weather is the biggest factor affecting fire risk,” says Mike Flannigan, study co-author and a professor with the Department of Renewable Resources at the U of A. “I see this as an early warning system — if you know there’s extreme weather coming, where it will occur and how extreme it will be, you can get the right resources to the right places at the right time.”

Pressure vs. precipitation

Canada typically records an average of 7,000 wildfires each year, which collectively burn about 2.5 million hectares. However, most of that destruction is caused by a relatively small number of fires — about three per cent of the total. These fires are usually found to have started or grown out of control on so-called “spread days,” or days of strong winds coming on the heels of a prolonged rain-free period.

Spread days are most often associated with a stalled ridge of high pressure in the mid-level atmosphere. “It just parks there for seven to 10 days and it’s dry,” explains Flannigan. Factor in high temperatures, which accelerate the drying of soil and plant material, and strong winds, and an unattended campfire or tossed cigarette butt can quickly become an inferno.

Currently, the fire weather index in Canada is calculated based on forecast values of surface temperature, relative humidity, wind speed, and 24-hour precipitation, but as anyone who’s driven through a thundershower only to encounter bright sunshine 10 kilometres down the highway knows, summertime precipitation can be highly localized and difficult to forecast more than a day or so in advance. By contrast, numerical models used to forecast large-scale weather systems are much more accurate over a longer period of time, meaning the SOM was able to zero in on potential spread days up to a week in advance. For a fire manager tasked with ensuring that resources are in place to combat a dangerous fire, that extra advance notice could make all the difference, says Flannigan.

“If you have a fire in British Columbia and request assistance from Ontario, it will be three days before personnel are on the fire line; if you’re requesting assistance from Mexico or Australia, it’s a week or more,” he says. “If you have five to seven days to make that call for additional resources, you have a chance of getting boots on the ground before a fire even starts.”

A tool for fire managers

Of course, hot, dry and windy weather doesn’t guarantee that a fire will start, and not every fire needs to be battled full-on, Flannigan notes. Unless a fire directly threatens human lives or infrastructure, fire managers will usually decide to let nature take its course.

However, as climate change threatens to increase the frequency of large, destructive fires and human settlements increasingly encroach upon wilderness, the fire manager’s job is becoming more difficult. Flannigan believes SOMs can help.

“Tools like this are needed more than ever, because the climate is changing, and the landscape is changing,” he says. “[Artificial Intelligence] is going to be driving our cars soon, so why not help us with fire management?” 


Are you passionate about Canadian geography?

You can support Canadian Geographic in 3 ways:

Related Content

People & Culture

The cowboy exclaims: The ballad of an ageing vaquero and his troubled horse, Bunny

The ultimate goal of vaquero horsemanship is to produce a “finished” horse: an exceptionally responsive animal that is a true partner to its rider

  • 2524 words
  • 11 minutes


A wildfire expert on the nuances of wildfire fighting

Fire detection, forecasting, management — and how decisions are made

  • 2139 words
  • 9 minutes
Ladybug from tribe Chrysomelini, as identified by iNaturalist

Science & Tech

Citizen science platform iNaturalist is testing artificial intelligence to help with species identification

Can a computer learn to identify species?

  • 684 words
  • 3 minutes


Digital map shows Alberta wildfires in near real-time

An interactive map released by Esri Canada allows users to track wildfires and see where they are most active

  • 584 words
  • 3 minutes