Reconstructing the trajectories and combustion behaviour of individual untethered burning firebrands in a vertical wind tunnel

Wildfire spread is strongly influenced by the transport of burning firebrands, which can ignite spot fires well ahead of the main fire front and substantially increase the risk to communities and infrastructure. However, quantifying firebrand flight and combustion behaviour remains experimentally challenging because of their highly chaotic motion and rapid mass loss in turbulent flows. Here we present an automated machine-learning-based framework for reconstructing the three-dimensional trajectories and combustion phases of untethered burning firebrands in freefalls within a vertical wind tunnel. The approach combines deep neural network–based object detection with multiframe trajectory association to recover continuous 3D motion, enabling detailed characterisation of velocity components and combustion dynamics. This framework overcomes the key limitations of manual tracking and conventional image-based methods in complex flow environments, providing reproducible and scalable measurements of firebrand transport behaviour. By enabling high-fidelity experimental data on firebrand flight, the method supports improved understanding and modelling of spotting processes in wildfires, with direct relevance to fire behaviour prediction and hazard assessment.