Geospatial monitoring and assessment of coastal land subsidence

Subsidence, the downward movement of the land, presents risks in coastal areas such as shoreline erosion and coastal flooding. The accurate estimation of subsidence and the identification of its underlying causes holds significant values for comprehending subsidence processes and guiding decision-making. However, both the subsidence estimation and interpretation are challenging due to its spatio-temporal variability, limited observability, and the complexity caused by natural processes and anthropogenic activities. The contributions of this dissertation were to 1) estimate subsidence at locations of tide gauge (TG) stations along the coastlines; 2) investigate coastal subsidence by integrating measurements from a variety of geodetic techniques such as global navigation satellite systems (GNSS), interferometric synthetic aperture radar (InSAR), TGs, and satellite radar altimtery (SRA); and 3) model subsidence with features related to natural processes and anthropogenic activities and identify potential drivers with machine learning (ML) techniques. These contributions were exemplified through case studies at the Texas Gulf Coast areas. First, two sea-level difference methods, through leveraging TG and SRA measurements, were developed to reconstruct subsidence time series at tide gauge (TG) locations along the Texas coastlines with observation periods exceeding ten years. In addition, synthetic aperture radar (SAR) imagery, continuously operating GNSS (cGNSS) observations, and sea-level measurements were harnessed to estimate the spatio-temporal patterns of subsidence spanning around three decades since the 1990s at the Eagle Point TG station, a prominent hotspot of sea-level rise in the United States. The results obtained from multiple geodetic techniques provided strong and consistent evidence of subsidence processes in the vicinity of Eagle Point. Moreover, a large-scale subsidence map along the Texas coastlines post-2016 was generated with SAR images, revealing that the Texas Gulf Coast experienced an average subsidence rate of -1 mm/yr near the shoreline with an increasing trend in magnitude inland. Attribution analysis indicated that hydrocarbon extraction and groundwater withdrawal were the predominant factors responsible for identified subsidence hotspots in the Texas Gulf Coast. ML demonstrated an impressive performance (with an 𝑅2 of 0.56) in modeling the observed large-scale subsidence, by incorporating a range of features related to natural terrain variations and anthropogenic activities. Explainable artificial intelligence (XAI) methods provided quantitative estimates of feature contributions of the ML model, and the data-driven results revealed that the digital elevation model (DEM) and anthropogenic factors were contributing features in relation to subsidence.