Wildfires in Canada, the United States, and globally are becoming more frequent and costlier, in terms of lives lost and the cost of combating the fires. And as the Earth continues to warm, we will be seeing more extremes of weather that in many cases have increased the frequency and magnitude of the blazes.
In a study published in the Canadian Journal of Forest Research in August, researchers described how using real-time meteorological data inside a neural network, a machine-learning system that functions like the human brain, could be used in predicting wildfires.
The research team created a “self-organizing map,” or SOM that relies on raw meteorological data to generate predictions. Giving the system artificial intelligence, over time it learns, without direction or outside intervention, using the raw data, to predict extreme fire weather in real-time and for days in advance.
Predicting extreme fire weather
Canada has about 8,000 wildfires every year, burning about 2 million hectares. Most of the wildfires are not large, accounting for 97 percent of all the fires. Most of the fires are associated with rain-free conditions and high winds, ideal weather for sparking a wildfire and allowing it to spread.
A large percentage of the area is burned in just a few days with such weather conditions. These days are called “spread-event days” or “spread days.” Remember these terms because they can be used in estimating the potential fire danger. The study authors say “predicting the spatial and temporal occurrence of spread days over a large area would be a valuable tool for fire management and prediction.”
OK, now it’s time to think about the weather patterns usually associated with extreme fire weather. This kind of weather most often is seen in large-scale weather patterns, or “synoptic patterns” found in the mid-troposphere 5-6 kilometers (3 to 3.7 miles) above sea level. Specifically, extreme fire weather is common under a strong long-wave ridge system.
SOMs are well-suited to handle gridded meteorological data, as well as predicting changes in the occurrence of synoptic patterns under climate change. The researchers hypothesized that by applying a SOM to forecast pressure fields, they could produce good fire-weather forecasts.
Using 53 years of historical data
To see how well the AI robot did on predicting extreme fire weather, the SOM compared the most current atmospheric data with what it had learned from 53 years worth of historical data on wildfires in Northern Alberta, already in its “brain.” The SOM produced a self-organizing map that identified patterns in ridges and troughs to predict extreme fire weather for the next day, next month and even further in time.
Mike Flannigan, co-author of the study and professor at the University of Alberta’s department of renewable resources said that after three years of research and testing in the “diagnostic mode, they are ready to test it in the field.
“It will take a couple years of hard work on both sides. I’m quite prepared to do that and I’m hoping that Alberta and other organizations across the province are interested as well,: Flannigan said, adding, “If we can get a better handle on those extremes, then we can be better prepared … We can look ahead to the next where and when.”
With the current wildfire situation in British Columbia today, the extreme fire weather prediction model could be put to use as another tool in fire management. “Fire management does a good job, but having an improved early warning system can be a great benefit,” Flannigan said.