Canada's wildfire hazard maps no longer take a year: an AI emulator reproduces BurnP3+ in hours
Synopsis
Canada's national wildfire hazard map has relied on BurnP3+ simulating millions of ignitions, taking over a year to produce. A Mila model trained with the Canadian Forest Service emulates BurnP3+ in seconds, running 806 to 3400 times faster depending on hardware, cutting a one-year task to 2 to 11 hours. It does not replace BurnP3+, but screens scenarios before precise runs.
Interpretation
Wildfire hazard mapping in Canada gains a fast path: an AI emulator returns burn-probability maps close to BurnP3+ in seconds, moving national mapping from years to hours. The national hazard map was previously generated by BurnP3+ simulating millions of individual ignitions from physical fire-behavior equations; the 2026 map took over a year, and the prior one dated to 2017. Depending on hardware, the emulator runs 806 to 3400 times faster than BurnP3+; a task that normally takes one year takes about 2 to 11 hours with the model.
The emulator learns the spatial relationship between BurnP3+ output maps and terrain and vegetation, giving approximate burn probabilities without re-solving fire-behavior equations. Training data came from BurnP3+ simulation maps produced during the Canadian Forest Service's national mapping that year, and the model approximates those maps. The model predicts lower burn probability where a lake is present, reflecting the lake's protective effect; an NRCan modeling specialist judges the risk of misinterpreting the model to be rather low because it reproduces what BurnP3+ does.
Perspective
The tool is aimed at the BurnP3+ user community and NRCan analysts, positioned as a complementary item in the analysts' toolkit rather than a replacement for BurnP3+ or for NRCan teams' expertise. The workflow is to test several parameters quickly with the AI model, then compute a likely scenario precisely with BurnP3+. The next objective is to distribute the tool to the BurnP3+ user community and continue tests to fully realize its capabilities and understand its operational limitations.
The emulator produces imperfect approximations, though with high fidelity, and its speed advantage ranges from 806 to 3400 times depending on hardware. The risk of misinterpreting the model is described as rather low because it reproduces BurnP3+ behavior. The tool's operational capabilities and limitations still require further testing to be fully understood.