Individual tree identification using a new cluster-based approach with discrete-return airborne LiDAR data

Individual tree identification is a key step for forest surveying and monitoring. To identify individual trees with airborne LiDAR data, a local maximum (LM) filter technique is typically performed. With LM, the highest point in a filtering window is generally considered to represent the tree position. This assumption, however, has great limitations, especially for mixed forests. To address this problem, we developed a new approach, the cluster center of higher points (CCHP), for tree position detection with LiDAR data. CCHP assumes that a tree position is located at the clustering center of higher points within a spatial neighborhood, and the center can be detected by a location-based recursive algorithm. The developed CCHP method was applied to a simulated forest and then verified in two real urban forests. In comparison with the variable window-sized LM filter method and layer stacking method, CCHP successfully identified 97% of trees in the simulated forest, while only 78% and 81% of the trees were recognized by LM and layer stacking methods respectively. The average absolute and relative offsets of CCHP are 0.33 m and 6.59%, respectively, and over 80% of the detected trees have an offset of less than 10% of the tree crown radius. CCHP also correctly detected 93.80% and 88.74% of individual trees in the first and second real forests, respectively, but the detection rates from the variable window-sized LM approach and layer stacking were less than 80%. In addition, the tree positions located by CCHP are considerably more accurate than the other two methods. Therefore, CCHP is proven to be promising for detecting individual tree positions from airborne LiDAR data for both simulated and real forests.

Automated summary

The study developed a new method CCHP for detecting individual tree positions using LiDAR data, demonstrating advantages in identifying trees in mixed forests. CCHP successfully identified 97% of trees in the simulated forest and showed higher accuracy compared to other methods.