Fig 1.
Demonstration of the 3D Forest workflow on a small sub-sample of the VPFDP (20m x 40m transect).
(a) TLS data imported into 3D Forest (i.e. the Base cloud) prior to any segmentation; (b) automatically segmented Terrain cloud (brown) and Vegetation cloud (green) using the octree search method, refined by manual adjustment; (c) individual trees segmented into Tree clouds displayed in random colors; (d) DBH and tree height displayed for each tree; (e) concave hulls of tree crowns; (f) crowns represented by 3D convex hulls and their mutual intersections (in yellow).
Fig 2.
Extraction and visualization of tree parameters from a single tree cloud.
(a) visualization of tree parameters: CBH–crown base height, CH–crown height, CTH–crown total height, CL–Crown length, CW–crown width, CC–crown centroid, DBH–diameter at breast height, TH–tree height, white sphere–tree position; (b) tree with computed basic parameters: position (blue sphere), DBH (60.8 cm), TH (green line; 35.6 m) and stem profile (yellow cylinders); (c) tree crown (black cloud) represented by CTH (24.9 m), CH (green line; 24.3 m), CL (green line; 14.6 m), CBH (11.3 m), crown centroid (orange sphere) and its planar projection (green sphere) with distance and azimuth from the tree position; (d) 3D convex hull of the crown with volume (2009 m3) and surface (866 m2) and orthogonal projection into plane with appropriate surface area (133.5 m2); (e) concave hull of the crown with volume (803 m3) and surface (1617 m2) and orthogonal projection into plane with its surface area (113.4 m2).
Fig 3.
Accuracy of automated segmentation with different settings of S and N, as compared to manually segmented trees used as a reference.
Four basic input distances (S) and two minimal numbers of points (N) were tested: (a) overall segmentation accuracy (blue dashed line for N of 5 points and red dashed line for N of 10 points in the cluster) and mapping accuracies of individual trees (boxplots and solid lines); (b) commission errors and (c) omission errors. In all charts square symbols connected by solid line represent medians, boxex upper and lower quartiles and whiskers represent upper and lower minimum and maximum; yellow color represents N of 5 points and green color N of 10 points in the cluster. Asterisks above boxplot mark settings that are statistically comparable to the best achieved result of the respective accuracy indicator according to the Nemenyi test.
Table 1.
Results of paired t-tests comparing automated methods of estimating DBH (Least Square Regression and Randomized Hough Transform) and tree height with conventional measurements using calipers and a digital inclinometer; computed for the significance level α = 0.05.
Significant test is marked by asterisk in the last column.
Table 2.
Results of logistic regression fit for all factors of DBH computation by both methods.
Significant tests (at significance level α = 0.05) are marked by asterisks in the last column.
Fig 4.
The effect of different factors on DBH computation analyses for both methods by logistic regression: RHT in red, LSR in blue.
Factors: (a) percentage of noise points in the DBH ring (N); (b) diameter of the ring (D); (c) missing part of the ring (M); (d) number of points forming the ring (P); (e) number of iterations (I); (f) time required for DBH computation in relation to the number of trees and iterations (Y axis in logarithmic scale).