Towards a Generalized Method for Tree Species Classification Using Multispectral Airborne Laser Scanning in Ontario, Canada

The aim of this paper was to develop a generalized classification model using multispectral airborne laser scanning (ALS) data to provide species or genus level tree identification. We tested the robustness, transferability, and generalizability of the developed method across two sites in Ontario, Canada. We focused on the generalization of the approaches to consider the various common species as well as their variable characteristics along the latitudinal gradient (e.g., the shape variations of pine), and variations of multispectral ALS features. The crown sample for training and validation was composed of 984 and 762 crowns for each site respectively. The generalized model for nine individual tree species was developed using a random forest machine learning algorithm with k-fold-cross validation for each study site accuracy assessment. We have extracted both 3D and intensity features from multispectral ALS data to identify trees. Our preliminary analysis revealed that intensity feature varied across two study sites and among the species, while 3D features were comparatively less variable. In addition, both 3D and intensity features are influenced by tree height. We also found that the three most useful features in tree species classification were multispectral vegetation indices (based on CI-1550 nm and C3-532 nm channels) and intensity features derived from CI-1550 nm. Our preliminary analysis shows that a generalized method can identify nine species with a 73% overall accuracy, whereas the site-specific overall accuracies were 75% and 66% respectively. Our work demonstrated the potential of a multispectral ALS sensor to develop a generalized classification model for the identification of diverse tree species.