Comparing the accuracy of single trees delineation from ALS

Bc. Alejandro Rodríguez Vivancos

Master's thesis

Abstract:

This study analyses different Individual Tree Detection (ITD) techniques from airborne LiDAR data (Airborne Laser Scanning, ALS), to increase the knowledge of this potential methodology, since it offers detailed information at the stand scale. Two segmentation techniques of the Canopy Height Models (CHM) are evaluated: First, the watershed segmentation by an elaborating process in ArcMAP. Then, testing the “TreeSeg” algorithm developed in FUSION. The influence of some parameters is also analysed, such as smoothing (different smoothing radii or kernel) and the resolution of these raster layers (0.2 m and 0.5 m). Additionally, these results are compared with those obtained by classifying trees on the point cloud directly, using the LIDR tool in the RStudio program. The vegetation was classified into two typologies based on their height (low and high vegetation). Finally, the relationship between the Diameter at Breast Height (DBH) and other biophysical parameters of the tree is analysed, to obtain a regression model. The watershed segmentation results in ArcMAP were significantly better than those obtained with FUSION (improvement of 20%, p-value = 0.028 < 5%). Low vegetation trees were not detected by the CHM methods, while few of them were detected by the point cloud classification, which produced estimates as accurate as the CHM methods. Resolution and smoothing had a significant influence on the estimates. CHMs of 0.5 m resolution showed better results with 3-cells radius (K3) smoothing, while CHMs of 0.2 m required the highest smoothing (K7). However, a higher point clouds density allowed to reduce the smoothing of the layers in both resolutions. Finally, a good potential regression model (DBH ~ Height, R2 = 0.91) was obtained after finding strong correlations between all the variables.

Keywords

Airborne Laser Scanning, Individual Tree Delineation, Watershed Segmentation, Canopy Heigh Model, Remote sensing, Resolution, Smoothing, Kernel, point cloud classification, regression model, RMSE, ANOVA

Jazyk práce: English

Datum vytvoření / odevzdání či podání práce: 19. 4. 2021

Obhajoba závěrečné práce

• Vedúci: doc. Ing. Peter Surový, PhD.

Czech University of Life Sciences Prague

Faculty of Forestry and Wood Sciences

Master programme:
Forestry, Water and Landscape Management