Effects of voxel size, scan angle and crown structure on the accuracy of tree species classification using airborne full-waveform LiDAR

Abstract

Accurate classification of tree species provides key information for mapping species diversity, managing forest ecosystems and modeling individual tree growth. While airborne Light Detection and Ranging (LiDAR) technology offers significant potential to estimate forest structural attributes, the capacity of this new tool to classify species is less well known. In this research, full-waveform metrics were extracted by a voxel-based composite waveform approach to classify four subtropical tree species. As part of the analysis, the effect of voxel size and scan angle on the accuracy of classification was investigated, and the crown structural variables were compared to the waveform metrics, to develop an understanding of the relationship between them. Results demonstrate that all tree species were classified with relatively high accuracy. The “Medium” resolution of voxel size has the highest classification accuracies, followed by “High”, and the “Low” resolution voxel case had the lowest classification accuracies. Tree crowns with small scan angles have the highest classification accuracies, which is slightly higher than trees crowns with large scan angles, whereas, the classification accuracy of all correctly detected trees irrespective of scan angle fall between the two. In addition, most of the crown structure variables have a significant correlation with the average of the height of median energy and waveform distance) for all tree species.