Skip to main content
eScholarship
Open Access Publications from the University of California

UCLA

UCLA Electronic Theses and Dissertations bannerUCLA

From discrete LiDAR footprints to landscape-scale insights: tropical forest structure, biomass, and mapping potential

Abstract

Three-dimensional tropical forest structure underpins estimates of aboveground biomass, yet reliably mapping it across broad scales remains difficult. Remote sensing through LiDAR platforms such as Global Ecosystem Dynamics Investigation (GEDI) and airborne laser scanning (ALS) has expanded our capacity to measure forest structure, but how accurately these sparse data can support continuous, high-resolution mapping across the tropics remains uncertain. This dissertation evaluates the relationship between forest structural diversity and biomass using non-continuous LiDAR, and assesses the feasibility and limits of scaling these derived measurements into wall-to-wall datasets. Chapter 1 examines structural diversity-biomass relationships across the Congo Basin using two GEDI-derived structural metrics: richness and evenness. These relationships varied by land cover class, with richness and evenness operating largely independently, indicating that they encode distinct ecological signals and may complement traditional species-based diversity metrics. However, because these metrics were derived from discrete GEDI measurements, analyses were spatially discontinuous, whereas carbon monitoring requires spatially continuous information. To assess the scalability of GEDI-derived structural metrics, Chapter 2 evaluates the predictability of three GEDI-derived structural attributes: relative canopy height (RH), foliage height diversity (FHD), and plant area index (PAI). I assess their suitability for continuous landscape mapping while accounting for spatial resolution, extreme error, and spatial autocorrelation. Predictive performance followed a consistent hierarchy (RH, FHD, then PAI), reflecting differences in attribute derivation and alignment with available predictors. Chapter 3 characterizes the error structure of such wall-to-wall maps, evaluating a 1 m global canopy height product against ALS data across the Congo Basin, Amazon, and Borneo. Error was predominantly systematic, with the magnitude of mean error reaching 58-75% of mean absolute error. A correction model trained in the Congo Basin transferred to the other two regions without retraining and achieved lower mean absolute error than the pixelwise height-and-slope corrections. Improvement held in canopies taller than roughly 16-19 m and reversed in shorter canopies, including within the Congo Basin. Much of this error is therefore shared across structurally distinct forests and reducible from single-region reference data. These findings show where LiDAR-derived structure can reliably inform tropical carbon assessments.