Development of a Single Tree Classification Method Using Airborne LiDAR

Abstract

In recent years, airborne LiDAR has been proposed as a means to acquire accurate information regarding forest structure. However, while measurements, such as tree height and crown size, can be obtained directly using airborne LiDAR, mathematical equations must be used to indirectly estimate stem volume and diameter at breast height. The equation parameters vary with tree species, so it is important to obtain accurate tree species information for each tree species to estimate accurate forest information. In this study, we propose a method for classifying tree species with high precision in single tree units using airborne LiDAR intensity data and LiDAR structure metrics. Five tree species were included: Japanese cedar, Japanese cypress, red pine, larch, and broad-leaf (deciduous and evergreen) species. We used LiDAR intensity data and tree crown shape metrics to calculate eight tree feature quantities and then compared the measurements among the five tree species to evaluate the importance of feature quantities. The comparison was performed using the random forest machine learning algorithm. We found that crown shape calculated for a grid unit of 10 m×10 m was more effective for classification than those of single trees. Classification accuracy of 93.7％ was recorded after classifying the above five tree species using the proposed eight feature quantities. Thus, it is possible to accurately classify single tree units using the feature measurements presented in this study.