Comparison of Tree Species Classification from Individual Tree Features Based on ALS and ULS Data

Accurately identifying tree species is crucial for conducting forest inventories and assessing ecosystems. Light Detection and Ranging (LiDAR) has exhibited considerable capability in tree species classification using derived structural characteristics. However, the impact mechanism of different LiDAR platform on tree species classification accuracy is not clear. In this study, airborne laser scanning (ALS) and unmanned aircraft vehicle laser scanning (ULS) data were used to compare tree species classification and evaluated the individual tree features affection. The point cloud distribution features, intensity features and tree shape features in ULS and ALS were extracted respectively and applied the Random Forest to recognize seven tree species for a German forest dataset. The classification accuracy reached 82% for ULS data and 78% for ALS data, indicating that there is minor improvement in classification accuracy by ULS data under these three types of features. It was found that tree shape features contributed the most to classification, and feature importance ranking was no significantly different between ALS and ULS data.