AI Framework for Faster Wildfire Hazard Mapping in Canada

Paper: Accelerating Wildfire Hazard Mapping Across Canada with an AI Surrogate Modelling Framework

Authors: Hager Radi Abdelwahed, Toluwaleke Olutayo, Chirag Goel, Jana Osea, Aldo Zaimi, Tom Swystun, Sandy Erni, Francisco Emiliano Lopez Saavedra, Yan Boulanger, Ga'etan Marceau Caron

Executive Summary

In Canada, the rising frequency of wildfires necessitates efficient national hazard mapping to inform proactive strategies such as fuel management, land-use planning, infrastructure protection, and evacuation routes. Traditional wildfire simulation methods, like BurnP3 and BurnP3+, utilize complex Monte Carlo simulations to assess burn probabilities, fire intensity, and spread rates. These simulations are computationally heavy, often requiring days to weeks to complete for a single regional modeling unit.

To address these challenges, researchers have developed a new approach that combines artificial intelligence (AI) with wildfire modeling. This collaboration has led to three key advancements.

First, a standardized framework has been established for training and evaluating AI models against traditional wildfire simulators. This framework aims to ensure consistency and reliability in the performance of AI-driven tools.

Second, a multi-task deep learning model has been created, trained on national-scale data. This model emulates key wildfire behavior metrics—burn probability, mean fire intensity, and mean rate of spread—within seconds, significantly reducing the computational burden compared to traditional methods.

Lastly, the research introduces domain-specific metrics and evaluation protocols designed to test the model's ability to generalize to unseen data. By integrating various geospatial factors such as fuel types, terrain information, and ignition likelihood along with regional weather data, the AI model approximates the outputs of BurnP3+ well. The results demonstrate strong performance when evaluated against metrics of agreement, rank correlation, error, and spatial overlap.

On a practical level, the AI surrogate model has shown remarkable efficiency, achieving an 806-fold acceleration in processing time and utilizing just 13% of the memory required by BurnP3+. This significant improvement allows for rapid scenario analysis, potentially transforming how wildfire risks are assessed and managed.

These advancements could lead to faster and more effective wildfire hazard mapping, which is critical for timely decision-making in public safety and resource allocation. The findings, derived from simulations rather than real-world applications, suggest that integrating AI into wildfire modeling can enhance the capacity for proactive mitigation efforts, ultimately contributing to better management of wildfire risks in Canada.

Academic Abstract

Canada's increasing wildfire activity requires frequent national hazard mapping to guide proactive mitigation, including fuel management, land-use planning, infrastructure protection, and evacuation routing. However, process-based wildfire simulation frameworks (e.g., BurnP3 and BurnP3+) rely on computationally intensive Monte Carlo simulations to model burn probability, fire intensity, and rate of spread. These workflows can take days to weeks to process a single regional modelling unit. To overcome these computational bottlenecks, we present three contributions developed through collaboration between AI researchers and wildfire modelling experts: (a) a standardized framework for training, fine-tuning, and evaluating AI surrogate models against process-based wildfire simulators; (b) a multi-task deep learning surrogate trained on national-scale data to emulate burn probability, mean fire intensity, and mean rate of spread at a fraction of the computational cost (seconds); and (c) domain-specific metrics and counterfactual evaluation protocols to probe out-of-distribution generalization. Integrating geospatial data (fuel, terrain, ignition likelihood) and regional weather, our model approximates BurnP3+ simulation outputs, achieving strong performance across agreement, rank-correlation, error, and spatial-overlap metrics. It preserves downstream hazard classifications and produces physically plausible responses to expert-informed counterfactual interventions. On a representative region, the surrogate achieves 806-fold end-to-end CPU acceleration and uses 13% of BurnP3+'s host memory, facilitating rapid scenario analysis.