Abstract
H. Tamiminia. Forest Aboveground Biomass Estimation and Change Monitoring Using Multisource Remote Sensing Data and Machine Learning Techniques, 309 pages, 24 tables, 58 figures,
2023. APA style guide used.
Forest is one of the most valuable Earth’s resources which is required to be monitored in a timely
manner by environmental organizations, government agencies, and conservation groups. One of
the key tools that paves the road for long-term forest monitoring is aboveground Biomass (AGB)
estimation. A methodology to rapidly and accurately estimate AGB is essential in many
applications including carbon stock monitoring, sustainable forest management, deforestation and
forest degradation investigation. Importantly, forest AGB estimation is of paramount significance
for studying carbon cycle and climate change mitigation and adaptation. Accurate AGB estimation
has been an area of interest for many researchers. One method is to develop robust approaches
using remotely sensed data. Remote sensing techniques can provide a cost-effective way to
estimate forest AGB over large areas. Thus, the goal of this dissertation is to investigate multisource remotely sensed datasets for accurate AGB estimation of New York State using machine
learning models. First, the potential of different remote sensing sources, such as airborne LiDAR,
optical imagery, SAR data, and their combination, for AGB estimation was investigated. The
results suggested that integrating these data sources provides valuable insights into the vertical
structure of trees, spectral and physical characteristics, and overall improves the accuracy of AGB
estimation. Second, various machine learning models were employed, with random forest
exhibiting the highest degree of performance. Then, pixel-based and object-based image analysis
(OBIA) approaches were compared for AGB estimation, and the OBIA was found to outperform
the pixel-based approach. Furthermore, an RF model was developed for historical AGB mapping
of New York using Landsat imagery from 2001 to 2019, employing the OBIA method. Finally,
the integration of Global Ecosystem Dynamics Investigation (GEDI), Sentinel-1, and Sentinel-2
data was used to create 10 m state-wide canopy height and AGB maps. Overall, the research
highlights the potential of remote sensing data and machine learning techniques for forest AGB
estimation, which has implications for carbon cycle and climate change studies.