Mila and Natural Resources Canada Build an AI Wildfire Model That Runs Up to 3,400 Times Faster
Montréal research institute Mila and the forestry division of Natural Resources Canada (NRCan) have built an artificial intelligence model that maps wildfire hazard across the country, estimating how likely an area is to burn, how fast a fire could spread and how intense it might become, BetaKit reported on October 23. The tool is meant to replace BurnP3+, the physics-based simulator NRCan uses today.
BurnP3+ relies on physical fire-behaviour data and has to be updated constantly as climate and vegetation change, so producing new hazard maps can take months. The AI emulator was trained on maps generated by physical fire simulations this year, giving it a similar foundation, but Mila says it runs between 800 and 3,400 times faster depending on the computing hardware, turning that work into hours. Tom Swystun, a forest systems modelling specialist at NRCan, said the emulator produces imperfect approximations but with high fidelity, and that the team's first reaction to its speed was simply “wow.”
The project lands after a punishing season. According to Public Safety Canada's October update, more than 5,400 wildfires burned about 4.5 million hectares in 2026, the country's fourth-worst year on record, and forced 33,750 people from their homes, including more than 8,800 people from First Nations communities. NRCan's fire division expects fires to keep growing in frequency and intensity, which is why it turned to Mila's applied research team.
What it means in practice
Speed changes how hazard maps can be used. Instead of waiting months for a fresh simulation run, fire managers, provinces and communities could rerun scenarios as fuel and weather conditions shift, and test far more what-if cases before a season begins. The catch is that an emulator only knows what its training simulations covered, so its outputs will still need to be checked against BurnP3+ and field experience, particularly in unusual fuel types or extreme weather.
The model is also an early deliverable for Ottawa's open-source AI push. Mila has been tapped to carry out the open-source part of the federal AI strategy, building models and software that anyone can use and adapt, and BetaKit reports it is negotiating new federal funding for that work. It builds on Mila's open-source AI package for Canadian businesses with Mozilla and gives a concrete public-safety example of the practical, risk-aware approach behind Canada's push for pragmatic G7 and G20 AI rules.
What to watch: whether NRCan and Mila publish the model and its validation results, how quickly provincial wildfire agencies fold it into planning for 2027, and whether the new federal funding for Mila's open-source work is confirmed.