TLS point cloud classification into stem and non-stem
First, plot-level TLS point clouds
were segmented to identify points from individual trees. Local maxima
from canopy height models (CHMs) with a 20-cm resolution were identified
using the Variable Window Filter approach (Popescu & Wynne 2004) and
the Marker-Controlled Watershed Segmentation (Meyer & Beucher 1990) was
applied to delineate crown segments. A point-in-polygon approach was
applied for identifying all points belonging to each crown segment. To
identify points that originated from stem and crown within each crown
segment, a point cloud classification procedure by Yrttimaa et al (2020)
was used. The classification of stem and non-stem points assumed that
stem points have more planar, vertical, and cylindrical characteristics
compared to non-stem points representing branches and foliage (Liang et
al. 2012, Yrttimaa et al. 2020). The method by Yrttimaa et al. (2019,
2020) is an iterative procedure beginning from the base of a tree and
proceeding towards treetop. More detailed description of the point cloud
classification workflow can be found in Yrttimaa et al. (2019, 2020).
The result of this step was 3D point clouds for each individual tree
(n=741) within the 9 sample plots (Figure 3A).