Integration of LiDAR and Hyperspectral Imagery for Tree Species Identification at the Individual Tree Level

ABSTRACT Due to the varying roles of different tree species in biodiversity, disaster prevention, soil and water conservation, and carbon storage, species classification is an important research area in forest conservation and management. This study utilizes airborne LiDAR and hyperspectral imaging technology to perform species classification at the individual tree level in broadleaf mixed forests. First, to address the issue of insufficient detection accuracy for large natural broadleaf mixed forests using existing methods, a secondary detection strategy that combines canopy height models with raw point cloud data is proposed to improve detection precision. During the detection process, spectral clustering algorithms are used to integrate hyperspectral information, further resolving issues related to inadequate tree detection. Finally, based on the features of the segmented individual trees, a species classification study is conducted using the random forest algorithm. The results show that after incorporating hyperspectral information and secondary detection of individual trees, the detection rate of individual trees increased by an average of 9.8%, the R 2 of crown width increased by 0.13, and the root mean squared error (RMSE) decreased by an average of 1 m. The species classification, combining both hyperspectral and LiDAR data, achieved the best results, with an average user accuracy of 82.77%, where spectral features, crown width, and crown volume were identified as the key features for species classification.