Date of Award

Summer 8-7-2015

Level of Access Assigned by Author

Open-Access Thesis

Degree Name

Master of Science (MS)


Forest Resources


Shawn Fraver

Second Committee Member

Brian Roth

Third Committee Member

Aaron Weiskittel


This study develops and tests novel methodologies for measuring the attributes of individual trees from three-dimensional point clouds generated from an aerial platform. Recently, advancements in technology have allowed for the acquisition of very high resolution three-dimensional point clouds that can be used to map the forest in a virtual environment. These point clouds can be interpreted to produce valuable forest attributes across entire landscapes with minimal field labor, which can then aid forest managers in their planning and decision making.

Biometrics derived from point clouds are often generated on a plot level, with estimates spanning many meters (rather than at the scale of individual the individual tree), a process known as area-based estimation. As the resolution of point clouds has increased however, the structural attributes of individual trees can now be distinguished and measured, which allows for tree lists including species and size metrics for individual trees. This information can be of great use to forester managers; thus, it is essential that proper methods be developed for measuring these trees.

To this end, an algorithm called layer stacking, was developed to isolate points representing the shapes of individual trees from a Light Detection and Ranging (LiDAR) derived point cloud, a process called segmentation. The validity of this algorithm was assessed in a variety of forest stand types, and comparisons were made to another popular tree segmentation algorithm (i.e., watershed delineation). Results indicated that when compared to watershed delineation, layer stacking produced similar or improved detection rates in almost all forest stands, and excelled in deciduous forests, which have traditionally been challenging to segment.

The algorithm was then implemented on a large scale, for individual measurements on over 200,000 trees. The species and diameter of each tree was predicted via modeling from structural and reflectance characteristics, and allometric equations were used to obtain volume and carbon content of each tree. These estimates were then compared to measurements taken in the field, and to area-based estimates. Results indicated improved accuracy of plot level basal area, volume, and carbon estimation over traditional area-based estimation, as well as moderately reliable individual tree estimates, and highly reliable species identification.

Finally, because LiDAR point clouds can be expensive to acquire, point clouds generated from aerial photos via structure-from-motion (SfM) reconstruction were evaluated for their accuracy at a tree level. An analysis between tree height measurements obtained by SfM, SfM in conjunction with LiDAR, LiDAR alone, digital stereo-photo interpretation, and field measurements was conducted. Results indicated no difference between SfM in conjunction with LiDAR and LiDAR alone. We concluded that SfM represents a valid low cost means of producing a point cloud dense enough to measure individual trees.

Thus, high resolution point clouds can be used to generate forest inventories containing a number of valuable biometrics, such as tree height, species, volume, biomass, and carbon mass. Such estimates may allow for the automatic development of large-scale, detailed, and precise forest inventories without the cost, effort, and safety concerns associated with extensive field inventories.