Testing treecbh in Central European forests: an R package for crown base height detection using high-resolution aerial laser-scanned data

Accurate information regarding tree canopy characteristics is crucial for forest management, but it is often difficult to assess. This study presents an innovative framework designed for crown base height (CBH) detection using high-resolution laser-scanned data, with a specific focus on individual trees within forests. The framework comprises three key steps: (i) segmenting the input tree point cloud to identify the tree trunk and its branches using the treesio software; (ii) applying vertical cross-sectional K-means clustering to cluster the identified tree and to define the elevation threshold for removing low-lying understory vegetation; (iii) employing a novel 2D kernel method for detecting CBH after eliminating low-lying understory vegetation. The 2D kernel method, developed for broadleaf forests using leaf-off airborne laser scanning (ALS) data, underpins the treecbh tool. This tool features a visual CBH adjustment component that shows a 2D profile plot of the tree point cloud, and suggests a CBH value for user approval or adjustment. To evaluate accuracy, in situ measured CBH data from five forest plots in Germany and Hungary with varied species compositions were used. ALS data were collected during leaf-off conditions for the two Hungarian plots and during leaf-on conditions for the three German plots. Leaf-off terrestrial laser-scanned data from individual trees were also used in the accuracy assessment. A sensitivity analysis using random point decimation was conducted on the terrestrial laser-scanned data to assess treecbh’s sensitivity to point density. The initial results exhibited matching rates of 45% and 60% for leaf-off ALS plots, which significantly improved to 71% and 77%, respectively, when using the visual CBH adjustment feature of the tool. The leaf-on ALS results demonstrated matching rates between 24% and 33%, whereas the CBHs of individual terrestrial laser-scanned trees could be detected with 93% accuracy in visual mode. It was observed that treecbh operates effectively when the input ALS data have a minimum point density of 20 pts/${\text{m}}^2$, with its optimal performance achieved at 110 pts/${\text{m}}^2$. These findings indicated treecbh’s sensitivity to ALS data quality, scanning season (leaf-on and leaf-off), and point density. This sensitivity can be effectively mitigated in the case of leaf-off ALS data by utilizing the visual CBH adjustment feature of the tool.