A framework to characterize WUI firebrand shower exposure using an integrated approach combining 3D particle tracking and machine learning

Firebrand showers are known for their devastating effects throughout Wildland-Urban Interface (WUI) communities threatened by wildfires. In this work, we propose a framework to better characterize firebrand flows and facilitate exposure comparisons across experimental cases. This framework leverages the National Institute of Standards and Technology (NIST) Emberometer, a measurement device that allows time-resolved motion tracking of burning particles in full 3D space. An improved version of the Emberometer, geared towards field use with enhanced firebrand detection capability and data processing pipelines, is presented. The device was used to investigate, in outdoor settings, a firebrand shower artificially generated to produce mixed amounts of smoldering and flaming particles. The ability to perform meaningful quantitative exposure comparisons, via metrics such as Cumulative Particle Count (CPC) and Particle Number Flux (PNF), is demonstrated. A sub-set of 3Dtracked firebrand images was used to train several Convolutional Neural Networks (CNN) to recognize firebrand combustion state. The best performing model was selected to process the entire tracking dataset (over 70,000 firebrand images), and time-resolved volumetric number densities of both smoldering and flaming particles were derived, a first for complex airborne firebrand flows.