Multiscale Modeling and Assessment of Forest Carbon Stocks and Drivers of Carbon Dynamics in The Southeastern United States

Abstract

This dissertation investigates forest structure and biomass using an integrated framework that combines functional ecology, spatial statistics, and remote sensing. Its main objective is to understand how environmental heterogeneity, structural inversion, and the scale of observation collectively determine biomass patterns. The research is structured across three hierarchical levels: (i) plot-level variation in the structural traits of tree wood, (ii) the occurrence of extreme biomass at the landscape-level, and (iii) regional estimation of biomass under spatial uncertainty using L-band synthetic aperture radar (SAR). At the plot level, analysis of wood specific gravity (SG) in Quercus species demonstrates that SG reflects the long-term integration of environmental constraints, including topography, atmospheric demand, wind exposure, and hydrological conditions. These results support the interpretation of SG as a structural memory of site conditions, rather than as a transient physiological response. At the landscape-level, the study focuses on the upper tail of biomass distributions. Using a multi-scale geographically weighted regression with a negative binomial framework (MGWR-NB), it shows that extreme biomass is neither random nor driven solely by productivity or species composition. Instead, it arises from spatially heterogeneous combinations of climatic, geomorphic, hydrological, and edaphic factors, reflecting environmental filtering, disturbance regimes, and competitive dynamics.

At the regional level, the research evaluates L-band SAR data from the SAOCOM mission under conditions of substantial spatial mismatch with reference plots. Despite uncertainty at the kilometer scale, polarimetric SAR observables capture broad gradients of forest structural variability, confirming that radar backscatter responds to cumulative forest structure shaped by long-term ecological processes. Together, these findings establish a conceptual progression from structural features to emergent biomass extremes and structural signals detected by remote sensing. This work provides a unified multi-scale framework for understanding and predicting forest biomass, emphasizing the ecological importance of extreme biomass as an emergent property and highlighting the robustness of SAR-based approaches for large-area biomass monitoring (MAE=47.8 Mg ha-1, RMSE = 59.3 Mg ha-1, CCC = 0.50) under realistic data constraints.