Robotics Institute Seminar: Steven Waslander, “TankerVision: Turning Firefighting Flights into a Wildfire Behaviour Dataset”

Models of wildfire rate of spread, intensity and spotting still rely mostly on small experimental burns from decades ago. Fires are hardest to observe exactly when it matters most, during rapid, intense spread. Satellites are too coarse in space and time, and drones and research aircraft are kept out of fire airspace. The only platforms reliably overhead during a major run are the aircraft fighting the fire.

TankerVision turns these operational flights into a continuous, automated source of fire behaviour data. Working with Conair Aerial Firefighting, the BC Wildfire Service and Alberta Wildfire, we developed a self-contained imaging unit for the bird-dog aircraft that direct airtanker operations. It records georeferenced colour and thermal infrared imagery whenever the aircraft is near a fire, with no action required from the crew. A prototype flew roughly 50 missions over 30 fires from Kamloops in 2025.

With support from DRDC and NRCan, three upgraded aircraft flew fires across five provinces and territories in 2026. An automated pipeline segments fire and smoke, reconstructs georeferenced 3D fire fronts and derives rate-of-spread observations from successive passes. The result is the first large multi-pass, multi-fire dataset of active wildfire behaviour, a foundation for data-driven fire prediction models, with an initial public release planned for December 2026.