Mila and Natural Resources Canada develop rapid AI wildfire model

The Montreal-based AI institute Mila and Natural Resources Canada have developed an artificial intelligence model designed to map wildfire hazards significantly faster than current methods. The new AI emulator evaluates three critical dimensions: the probability of an area burning, the speed of fire propagation, and potential fire intensity.

This tool is intended to replace BurnP3+, a physical simulator used by Canadian authorities that requires frequent updates based on changing climate and vegetation data. While the existing simulator can take months to produce results, the new AI model—trained on physical simulation maps from 2026—can generate hazard maps within hours. Depending on the computing hardware used, the model is estimated to be between 800 and 3,400 times faster than previous methods.

The development comes amid a severe wildfire season in 2026, with over 5,400 wildfires burning approximately 4.5 million hectares and forcing the evacuation of more than 33,750 people. Researcher Tom Swystun noted that while the emulator provides approximations, it does so with “high fidelity and at speeds incomparably faster” than current tools.