Abstract

Forests play a vital role in global carbon cycling, biodiversity conservation, and the provision of ecosystem services. Accurate quantification of forest carbon sequestration rates – or forest production – therefore is essential for understanding ecosystem dynamics and informing sustainable forest management practices. Light detection and ranging (LiDAR) systems, as an active remote sensing modality, have emerged as a powerful tool for capturing detailed structural information of forests. Studies employing terrestrial LiDAR demonstrated that canopy structural complexity (CSC) metrics are strongly correlated with forest production at stand to landscape scales. While terrestrial LiDAR systems provide highly detailed maps of internal canopy structure, their limited spatial coverage restricts large-area applications. Airborne laser scanning (ALS), on the other hand, offers a method for rapid, spatially exhaustive vegetation structural measurements over large areas, yet its capability to capture CSC metrics that are closely linked to forest production remains underexplored. Spaceborne laser scanning (SLS) also promises to provide continental to global-scale mapping of vegetation structure with higher temporal resolution than ALS, but its potential for extracting CSC metrics linked to forest productivity requires further investigation. Furthermore, measurements of vegetation structure from existing ALS and SLS instruments are often limited by occlusion, lack of sampling density, and other quality issues arising from platform constraints. In the context of these knowledge gaps and challenges, this dissertation investigates how airborne and spaceborne light detection and ranging (LiDAR) systems can be used to characterize and map canopy structural complexity (CSC) metrics that are linked to productivity and other ecosystem functions.   First, we investigated the utility of ALS for deriving key terrestrial-based CSC metrics by simulating flights of the National Ecological Observatory Network’s (NEON) Airborne Observation Platform (AOP) over a virtual scene of a section of Harvard Forest (HF), Massachusetts, USA, in the Digital Imaging and Remote Sensing Image Generation (DIRSIG) simulation environment. Additionally, we varied key collection parameters across flights to determine the optimal collection settings for deriving the target structural metrics. Our results showed that ALS-derived estimates of the CSC metrics representing outer canopy complexity were strongly correlated with ground truth (R = 0.78 – 0.86). In contrast, ALS estimates of whole canopy complexity exhibited poor correlation (R < 0.5), likely due to limited laser penetration in nadir-oriented scans, which resulted in insufficient sampling of the subcanopy. These findings suggest that utilizing dense point clouds from cross-track scanning ALS systems could improve subcanopy sampling and lead to more accurate CSC measurements that incorporate inner canopy structure.   Next, we derived a novel suite of three-dimensional (3D) CSC metrics from small-footprint, high-density ALS data, acquired by NEON’s AOP. We evaluated relationships between these metrics and net primary production (NPP) separately for deciduous and evergreen forests, followed by an analysis for combined forest types across seven temperate forest sites in the continental US. We found that ALS-derived CSC metrics explained a significant amount of variance in NPP using regression-based methods and statistical significance tests; in fact, three metrics accounted for 77% of the variance in deciduous forests (RMSE = 11%) and another three accounted for 76% of variance in evergreen forests (RMSE = 13%). However, CSC metrics failed to explain a significant amount of NPP variation when forest types were combined, suggesting that distinct CSC-NPP relationships exist between deciduous and evergreen forests. Our findings demonstrate that high-density ALS data can generate strong, biome-wide CSC predictors of primary production when forest types are modeled separately. Finally, this dissertation developed deep learning methods to extend fine-scale CSC retrieval to high-altitude and future spaceborne waveform LiDAR observations. An encoder–decoder transformer was developed to reconstruct high-resolution canopy profiles from waveforms, acquired by NASA’s Land Vegetation and Ice Sensor (LVIS) at the Smithsonian Environmental Research Center, using co-located ALS data as reference. This approach improved correlations with ALS-derived CSC metrics from R = 0.62 to 0.84 and from R = 0.76 to 0.90, thereby demonstrating that transformer-based learning can recover sufficient vertical detail from high-altitude waveform observations to reliably quantify ecologically-relevant, ALS-derived CSC metrics.  A generative diffusion framework next was developed to build on this profile-level reconstruction and extend the task into three dimensions. This was achieved by integrating LVIS waveforms with optical imagery to reconstruct high-resolution, voxelized 3D forest plots against ALS-derived voxel volumes. Reconstruction fidelity improved markedly at coarser horizontal resolutions, and the framework preserved entropy-based CSC metrics beyond the information present in the raw waveforms, whereas surface-roughness and standard-deviation-based metrics did not exceed the raw-waveform baselines. This shortfall was attributed to the modest sample size of coincident waveform-ALS plots and the irregular LVIS sampling geometry.  These contributions advance LiDAR-based approaches for large-area monitoring of forest structure, productivity, and ecosystem function, and point toward denser, more uniform waveform acquisitions from forthcoming missions as the path to operational, spaceborne 3D structural mapping. We recommend that future efforts apply these reconstruction methods to the larger, more uniformly sampled datasets expected from forthcoming missions. Such efforts should concentrate on the entropy-based metrics most amenable to recovery, and on aggregating reconstructed observations into continuous, wall-to-wall structural maps. Linking the CSC metrics derived from these reconstructions directly to field-measured productivity, biomass, canopy fuels, and biodiversity indicators will further establish their value for operational forest monitoring. Future work should also examine whether the relationships observed between ALS-derived CSC metrics and NPP generalize to tropical, boreal, and other non-temperate forests.

Publication Date

8-2026

Document Type

Dissertation

Student Type

Graduate

Degree Name

Imaging Science (Ph.D.)

Department, Program, or Center

Chester F. Carlson Center for Imaging Science

College

College of Science

Advisor

Jan van Aardt

Advisor/Committee Member

Callie Babbitt

Advisor/Committee Member

Carl Salvaggio

Campus

RIT – Main Campus

Download

Share

COinS