Author ORCID Identifier

https://orcid.org/0009-0000-7050-9249

Semester

Spring

Date of Graduation

2026

Document Type

Dissertation

Degree Type

PhD

College

Eberly College of Arts and Sciences

Department

Geology and Geography

Committee Chair

Aaron Maxwell

Committee Member

Brenden McNeil

Committee Member

Michael Gallagher

Committee Member

Nicholas Skowronski

Committee Member

Michael Harman

Abstract

The aim of this dissertation was to investigate the potential for analyzing tree and forest structure across multiple spatial scales using lidar-derived point cloud data and topological data analysis. This work focuses on how structural information captured by terrestrial laser scanning (TLS) and airborne laser scanning (ALS) can be represented and interpreted to support structural segmentation, species classification, and landscape-scale mapping. The following research themes were explored: decomposition of tree architecture into structural components, application of deep learning models to topological representations of individual trees, and delineation of spatially contiguous structural classes for scaling fine spatial resolution measurements. This research is presented as three interrelated manuscripts.

Topology-guided approaches were found to be effective for decomposing individual tree point clouds into trunk, branch, and foliage components. Methods that incorporated global structural context alongside local geometric information achieved high classification accuracy, particularly for trunk and foliage identification. Misclassification occurred primarily between branches and foliage in terminal canopy regions, where structural complexity is greatest. These results suggest that representing tree architecture using structural relationships improves segmentation relative to approaches based solely on local geometric descriptors.

Representations derived from one set of topologic relationships, persistent homology, provided a consistent basis for characterizing tree structure across a wide range of sizes and sampling conditions. When used for species classification, these representations achieved high accuracy and demonstrated strong performance across trees of varying height. This indicates that the extracted features capture structural characteristics that are not solely dependent on absolute size. However, some confusion was observed between species with similar architectural forms, particularly at intermediate size classes, suggesting that structural similarity across developmental stages remains a limiting factor.

At broader spatial scales, topological summaries derived from ALS data were effective for identifying regions of similar forest structure. Structural classes defined in this way exhibited low within-class variability and strong separation between classes for several TLS-derived metrics, particularly those related to canopy volume and voxel occupancy. In contrast, traditional height-based metrics showed weaker differentiation between classes. These findings indicate that the identified structural units provide a meaningful basis for extending fine-scale measurements across landscapes.

Collectively, this research suggests that lidar-derived point clouds can be used to characterize forest structure in a way that supports analysis across multiple levels of organization. Approaches that account for spatial relationships and structural organization provide improved interpretability compared to methods based on aggregated geometric descriptors. The integration of topological analysis with machine learning further enables the extraction of structural patterns that can be applied to segmentation, classification, and landscape-scale mapping.

Download

Available for download on Wednesday, April 28, 2027

Share

COinS