Using Discrete-Point LiDAR to Classify Tree Species in the Riparian Pacific Northwest, USA

Abstract

Practical methods for tree species identification are important for both land management and scientific inquiry. LiDAR has been widely used for species mapping due to its ability to characterize 3D structure, but in structurally complex Pacific Northwest forests, additional research is needed. To address this need and to determine the feasibility of species modeling in such forests, we compared six approaches using five algorithms available in R’s lidR package and Trimble’s eCognition software to determine which approach most consistently identified individual trees across a heterogenous riparian landscape. We then classified segments into Douglas fir (Pseudotsuga menziesii), black cottonwood (Populus balsamifera ssp. trichocarpa), and red alder (Alnus rubra). Classification accuracies based on the best-performing segmentation method were 91%, 92%, and 84%, respectively. To our knowledge, this is the first study to investigate tree species modeling from LiDAR in a natural Pacific Northwest forest, and the first to model Pacific Northwest species at the landscape scale. Our results suggest that LiDAR alone may provide enough information on tree species to be useful to land managers in limited applications, even under structurally challenging conditions. With slight changes to the modeling approach, even higher accuracies may be possible.

Keywords

• LiDAR
• forest inventory
• individual tree position
• image segmentation
• tree species