A comparison and development of methods for estimating shrub volume using drone‐imagery‐derived point clouds

Abstract

AbstractShrub volume is used to calculate numerous, essential ecological indicators in rangeland ecosystems such as biomass, fuel loading, wildlife habitat, site productivity, and ecosystem structure. Field techniques for biomass estimation, including destructive sampling, ocular estimates, and allometric techniques, use shrub height and canopy widths to estimate volume and translate it to biomass with species‐specific allometric equations. These techniques are time‐consuming and pose challenges, including removal of plant material and training of observers. We compared canopy volume estimates from field‐based measurements with drone‐collected canopy volume estimates for seven dominant shrub species within mountain big sagebrush (Artemisia tridentata subsp. vaseyana) plant communities in southern Idaho, USA. Canopy height and two perpendicular width measurements were taken from 103 shrubs of varying sizes, and volume was estimated using a traditional allometric equation. Overlapping aerial images captured with a DJI Mavic 2 Professional drone were used to create a 3D representation of the study area using structure‐from‐motion photogrammetry. Each shrub was extracted from the point cloud, and volume was estimated using allometric and volumetric methods. The volumetric method, which involved converting point clouds to raster canopy height models with 2.5‐ and 5‐cm grid cells, outperformed the allometric method (R2 > 0.7) and was more reproducible and robust to user‐related variability. Drone‐estimated volume best‐matched field‐estimated volume (R2 > 0.9) for three larger species: A. tridentata subsp. tridentata, A. tridentata subsp. vaseyana, and Purshia tridentata. The volume of smaller shrubs (canopy widths <1 m) was slightly overestimated from drone‐based models. We argue that drone‐based models provide a suitable alternative to field methods, while having the added benefit of being less time‐consuming, with fewer limitations, and more easily scaled to larger study areas than traditional field techniques. Finally, we demonstrate a proof of concept for automating canopy volume estimates using point‐cloud‐based automatic shrub detection algorithms. These findings demonstrate that drone‐collected images can be used to assess shrub canopy volume for at least five upland sagebrush steppe shrub species and support the integration of drone data collection into rangeland vegetation monitoring.