Date of Award

Winter 2024

Project Type

Thesis

Program or Major

Earth Sciences

Degree Name

Master of Science

First Advisor

Michael W Palace

Second Advisor

Ruth Varner

Third Advisor

Franklin B Sullivan

Abstract

The study of terrestrial carbon stocks is essential to understand global carbon cycles and the magnitude of anthropogenic carbon emissions. Forests are significant terrestrial carbon stores, with the above ground biomass of the Brazilian rainforest alone conservatively estimated to contain ten percent of the terrestrial carbon stock (Asner et al., 2001; Drake et al., 2002; Keller et al., 2004; Lloyd et al., 2007; Truehaft et al., 2014; Cushman et al., 2021). The Amazon Basin primarily consists of closed-canopy tropical forest, covering 5.8 x 106 km2 interspersed with substantial portions of cerrado surrounding the Amazon River and its tributaries in the southernmost parts of Brazil (Keller et al., 2004), the majority of which is challenging to access directly. Light Detection and Ranging (LiDAR) inventory plays a crucial role in estimating carbon stocks across the globe and is specifically well-suited to study large, remote regions, like that of the Amazon (Kent et al., 2015; Palace et al., 2016; Taubert et al., 2021). In this study, an algorithm utilizing semivariance, an autocorrelation technique, was developed to examine horizontal tropical forest features through a Canopy Height Model, and a previously developed crown delineation program, SpaceForester, which were then related to field-based measurements from the Sustainable Landscapes Brazil project database. LiDAR data from airborne surveys along with field-measured forest attributes collected from five Brazilian sites were used to estimate the three attributes at each location: average above ground biomass (AGB), basal area (BA), and diameter breast height (DBH). Artificial neural networks (ANNs) were constructed to model the relationships between these three attributes and four input attributes of horizontal structure: the maximum canopy height, tree density, mean tree crown radius, and product of the semivariance algorithm at each site. The ANNs served as a framework to assess the impact of semivariance, which proved to be effective in improving the estimation of the model output variables examined.

Download

Share

COinS