Development of a Framework for Forest Aboveground Biomass Estimation Using Airborne Lidar and High-resolution Satellite Imagery in Disturbance-prone Forests

Abstract

Accurate estimation of forest aboveground biomass (AGB) is essential for understanding ecosystem health, carbon storage, and supporting climate-related policies. The increasing availability of free and open airborne light detection and ranging (lidar) data has advanced AGB modeling. However, no standardized framework exists, particularly in the southeastern United States, where species diversity and frequent disturbances pose challenges for accurate estimation. To address this gap, the following three objectives were pursued: (1) to conduct a systematic review and meta-analysis of airborne lidar-based AGB studies to identify methodological trends globally, (2) to investigate an AGB modeling framework in five disturbance-prone, mixed forest sites in the southeastern US, and (3) to evaluate the feasibility of high temporal and spatial resolution PlanetScope imagery for characterizing canopy heights and assess its potential for capturing forest structure and thematic changes following wind disturbance. Following PRISMA guidelines, 52 peer-reviewed studies (2013-2023) focused on airborne lidar-based AGB modeling were selected and synthesized to conduct a comprehensive review and analysis of the literature. For AGB modeling, airborne lidar from the United States Geological Survey 3D Elevation Program and 3 m PlanetScope imagery through Planet’s Education and Research Program were leveraged. To develop a framework, Random Forest (RF) and Bayesian modeling techniques and five variable selection methods were examined: all predictors, top 5 and top 10 RF-derived predictors, lasso, and Recursive Feature Elimination (RFE). To address Objective 3, multi-stereo PlanetScope-derived photogrammetric canopy heights were extracted and validated using airborne lidar. Subsequently, RF regression models were used to generate pre- and post-disturbance canopy height maps, while RF classification models were trained to produce forest/non-forest maps to assess vertical structure and thematic changes following wind disturbance. The systematic review highlighted the limited use of Bayesian methods for AGB estimation despite their effectiveness with small training samples. Methodologically consistent studies showed slightly reduced heterogeneity (I²=91.67%) compared to the overall analysis (I²=96.38%), emphasizing consistency. Modeling results revealed that lasso and RF-based selection outperformed RFE, while Bayesian-based Gaussian Process Regression outperformed RF. Model accuracy (R²=0.29-0.73; Root Mean Squared Error (RMSE)=16.29-75.14 Mg/ha) was highest in the undisturbed forests and lowest in the disturbance-prone sites, indicating sensitivity to disturbance-driven heterogeneity. When compared with airborne lidar-derived estimates, PlanetScope-derived canopy heights showed moderate agreement with airborne lidar (correlation coefficient=0.68-0.79, RMSE=3.70-3.86 m). Multi-temporal 20 m maps revealed substantial canopy height reductions (36.52%-60.92%) after disturbance. Overall, this dissertation advances critical gaps in airborne lidar-based AGB estimation by establishing standardized methodologies, demonstrating the feasibility of Bayesian methods, and highlighting a promising monitoring framework using PlanetScope imagery for fine-scale AGB mapping and rapid disturbance assessment.