Hierarchically scaled remote sensing and field datasets for three-dimensional wildland fuel characterization

Background: Next-generation models of fire behavior and smoke production rely on gridded, 3D inputs of wildland fuel complexes. We used a hierarchically scaled sampling design to characterize canopy and surface fuels that are common to prescribed burning programs in the southeastern and western US. Sampling included airborne laser scanning, terrestrial laser scanning, close-range photogrammetry, and destructive field sampling. The objective of this study was to use a combination of airborne laser scanning (ALS), terrestrial laser scanning (TLS), structure-from-motion photogrammetry (SfM), and field observations to create co-located 3D datasets of live and dead understory fuels for use in wildland fuel mapping and prescribed burn decision support Results: Using our integrated, co-located methods, we produced hierarchically-scaled datasets detailing the structure and composition of canopy and surface fuels across 9 southeastern pine sites, 5 western pine sites, and 4 western grassland sites. These are now publicly available at within the Wildland Fire Science Initiative data repository (https://doi.org/10.60594/W4859C). In this paper, we detail methods and the repository structure. Conclusions: The study was designed to evaluate and advance methods for 3D fuel characterization and to provide consistently scaled and labelled datasets for model training and evaluation. More specifically, machine learning models can be used to parse 3D point clouds collected from ALS, TLS, and structure-from-motion photogrammetry into fuel objects and metrics. Calibration with field plots will allow our hierarchically-scaled datasets to be used as the foundation for synthetic fuelbed mapping, starting with fine-scale objects such as individual shrubs or downed wood and scaling to vegetation patches and operational burn units.