University of Alberta Machine-Learning Model Helps Predict Wildfire Behaviour

Researchers at the University of Alberta have released a study that outlines the development of a machine-learning program that can forecast the behaviour of northern Alberta wildfires.

Each year, roughly 8000 wildfires burn throughout Canada, spreading across two million hectares. Most of the havoc wreaked by these fires is often attributed to few days of severe fire weather, called “spread days.”

In a joint effort with the University of Oklahoma, forest science researchers programmed self-organizing maps (SOMs) to predict spread days of extreme fire weather, mapping where and when wildfires could break out in the province.

SOMs are a type of neural network–technology that mimics a human mind–that can be trained to process large quantities of meteorological data, find patterns and produce fire-weather forecasts.

The researchers found the SOMs can learn from the raw data without external guidance, making automated predictions and flagging potential extreme fire-weather events.

“This is the first study to develop a machine-learning model for extreme fire weather that could be deployed in real time,” wrote researchers in the study published Tuesday in the Canadian Journal of Forest Research.

Researchers used two atmospheric pressure variables that affect weather conditions to train the SOMs, determining that weather is the best predictor for fire activity when creating monthly forecasts.

While the machine-learning study found that SOMs could be used to predict spread days, researched noted if trained for too long, the SOMs will overfit and generalize poorly to new data. Despite the shortcoming, researchers believe the implications of the predictive models are far reaching.

“Our product could be a useful and easily applied tool for fire-management agencies throughout Canada,” researchers wrote. “This could contribute to the development of an early warning system so that fire management agencies could ensure that resources are available and in position.”

The current SOMs do have limitations as they only apply to northern Alberta, but separate SOMs could be trained for other regions.

The University of Alberta has become recognized as a global leader in the artificial intelligence industry. In March, the government announced a pan-Canadian AI strategy with $125 million in funding that the university is sharing with other AI programs and centres.