Learning Optimal Wildfire Evacuation Routes for Flin Flon 2025 Wildfire Case Study

This project develops a dynamic, self-learning wildfire evacuation routing system designed to improve public safety during fast-moving wildfire events. Traditional routing methods can identify effective paths but often fail to adapt quickly to rapidly changing fire behaviour and real-world driver responses. To address these limitations, the project introduces a Deep Reinforcement Learning (RL) framework in which an intelligent agent continuously learns and updates optimal evacuation routes within a simulated environment.
The 2025 Flin Flon wildfire, spanning Saskatchewan and Manitoba, serves as the primary case study. This incident provides a complex, cross-border dataset reflecting real traffic conditions, fire progression, and environmental impacts. Using Esri’s ArcGIS Pro and ArcGIS API for Python, the project will construct a high-fidelity simulation environment where the RL agent balances evacuation time with safety considerations such as fire proximity and smoke density. High-resolution GPS trajectory data will be incorporated to better understand evacuee movement patterns and improve model realism.
The final system will be delivered through an interactive ArcGIS Experience Builder interface and ArcGIS Dashboard, enabling emergency managers to access real-time, AI-optimized route guidance. This research advances GeoAI and demonstrates how Esri technologies and RL can transform wildfire evacuation planning across Canada.

Faculty Supervisor:

Xin Wang

Student:

Partner:

Esri Canada Ltd

Discipline:

Engineering

Sector:

Information and cultural industries; Professional, scientific and technical services

University:

University of Calgary

Program: