# Mila and NRCan unveil AI model that maps Canadian wildfire risk in seconds

> The new AI surrogate speeds up BurnP3+ simulations up to 3,400 ×, letting scientists evaluate fire scenarios instantly.

Oossa · 2026-10-19 · https://oossa.com/en/mila-and-nrcan-unveil-ai-model-that-maps-canadian-wildfire-risk-in-seconds

Mila’s applied research team and Natural Resources Canada have released an AI surrogate model that copies the BurnP3+ wildfire simulator but runs in seconds instead of days. The model can map a 130,000 km² region in about 174 seconds on a standard CPU, or 41 seconds on a GPU. It delivers a speedup of 800‑to‑3,400 times while keeping accuracy acceptable to wildfire experts. The code and model are open‑source, so any BurnP3+ user across Canada can download and use it for planning fire prevention, fuel management, and land‑use decisions.

## The facts

- Speedup: 800 – 3,400 times faster than BurnP3+ (days to seconds).
- CPU run time: ~174 seconds; GPU run time: ~41 seconds for a 130,000 km² region.

## Why it matters

It turns wildfire risk mapping from a multi‑day batch job into a near‑real‑time tool for Canadian agencies and industry.

## Sources & references

1. [Accelerating Wildfire Hazard Mapping Across Canada with AI](https://mila.quebec/en/ai4humanity/applied-projects/accelerating-wildfire-hazard-mapping-across-canada-with-ai) – Mila

Last updated: 2026-10-19