Semester of Graduation
Spring 2026
Degree
Master of Science in Civil Engineering (MSCE)
Department
Civil and Environmental Engineering
Document Type
Thesis
Abstract
Land subsidence represents an intensifying geohazard across rapidly developing urban centers, including East Baton Rouge Parish (EBRP), Louisiana. Groundwater withdrawal, structural loading from urban development, and geological heterogeneity collectively drive surface deformation that threatens transportation infrastructure, utility networks, and the built environment. This research develops an integrated analytical framework based on Interferometric Synthetic Aperture Radar (InSAR), a satellite-based geodetic technique for measuring millimeter-scale ground deformation. Sentinel-1 imagery is processed using the Small Baseline Subset (SBAS) time-series approach to derive deformation measurements. The resulting dataset is combined with ensemble machine learning models, physics-constrained deep learning, and interpretable artificial-intelligence methods. Processing 246 ascending-track acquisitions using the SBAS approach revealed spa- tially concentrated subsidence near fault structures and in areas of intensive groundwater withdrawal. These patterns are particularly evident in fault-adjacent zones and alluvial depositional settings. Multicollinearity testing using the variance inflation factor removed three redundant predictors and retained twelve independent conditioning variables. Ex- tremely Randomized Trees (ERT) and Random forest (RF) regressors were trained using these geological, hydrological, topographic, and anthropogenic factors, with Gaussian noise added to improve generalization. ERT attained coefficient of regression (R2 ) = 0.92 compared to R2 = 0.88 for RF; binarized ROC analysis further confirmed excellent discriminative ability with AUC values of 0.97 (ERT) and 0.94 (RF). A physics-constrained LSTM network achieved R2 = 0.83 and Pearson r = 0.92 when applied to the SBAS displacement time- series, effectively capturing complex temporal evolution and non-linear system behavior. The
model accurately reconstructed observed cumulative displacement and generated projections extending to 2030. Explainable AI analysis revealed land use/land cover, fault proximity, groundwater well density, and river–elevation interactions as principal controlling factors, providing both aggregate feature rankings and location-specific interpretations. Projections through 2030 indicate persistent deformation at a mean velocity of −0.53 mm/year, accumulating to a total displacement of −6.42 mm over the 5-year forecast period. The integrated satellite-based and AI framework improves predictive capability in areas with reduced interferometric coherence. It also produces susceptibility maps that support urban planning, infrastructure resilience, and groundwater-resource management in subsidence-affected regions. The methodology is robust, transferable, and reproducible, and can be applied to other urban environments experiencing similar geohazard conditions.
Date
3-27-2026
Recommended Citation
Kangah, Desmond, "LAND SUBSIDENCE SUSCEPTIBILITY MAPPING AND SPATIO-TEMPORAL FORECASTING BY INTEGRATING SENTINEL-1 INSAR IMAGERY WITH MACHINE AND DEEP LEARNING MODELS" (2026). LSU Master's Theses. 6366.
https://repository.lsu.edu/gradschool_theses/6366
Committee Chair
Abdalla, Ahmed
LSU Acknowledgement
1
LSU Accessibility Acknowledgment
1