Effective susceptibility mapping of land subsidence in Louisiana’s Capital Area using Data-Driven GIS and InSAR technologies

Land subsidence represents a critical geohazard, significantly impacting regions such as Louisiana's Capital Area, where a combination of natural processes and anthropogenic activities exacerbates land deformation. This study develops a high-resolution susceptibility mapping framework by integrating Interferometric Synthetic Aperture Radar (InSAR) data, geostatistical methods, and advanced machine learning algorithms. The research explicitly addresses deformation across East Baton Rouge, West Baton Rouge, East Feliciana, West Feliciana, and Pointe Coupee parishes.

The framework utilizes multi-source datasets, incorporating Landsat images, SRTM-DEM, Synthetic Aperture Radar (SAR) from Sentinel-1 imagery (2017–2020) and Global Navigation Satellite System (GNSS). The SAR data were processed via the PyGMTSAR package to generate precise displacement velocity fields and corrected for atmospheric effects and phase unwrapping errors. On the other hand, the QGIS open-source software was used to analyze and classify the Landsat images into several land cover categories. These outputs form the foundation for subsequent geostatistical analyses integrating geophysical, geological, and anthropogenic variables to model subsidence susceptibility.

Key drivers of subsidence are ranked and weighted through Frequency Ratio (FR) and Analytical Hierarchy Process (AHP) methodologies to quantify the influence of causative factors. Temporal deformation trends are modeled using Long Short-Term Memory (LSTM) neural networks, capturing non-linear relationships and dynamic interactions in the temporal domain. A Weighted Linear Combination (WLC) approach synthesizes weighted spatial layers, with the Optimum Index Factor (OIF) applied to reduce multicollinearity and enhance model robustness. Validation incorporates observed deformation data and Receiver Operating Characteristic (ROC) curve analysis, providing quantitative metrics such as the Area Under the Curve (AUC) for assessing predictive accuracy. The model outputs were classified into five sustainable categories representing areas at risk from this phenomenon using the Natural Break method.

This integrated approach advances the precision and reliability of subsidence susceptibility mapping, enabling detailed spatial resolution and enhanced predictive capability. The findings facilitate targeted risk assessments, support disaster mitigation strategies, and optimize resource allocation for land use planning and critical infrastructure protection. By addressing Louisiana's Capital Area's unique geophysical and socio-environmental characteristics, the framework provides a scalable solution applicable to subsidence-prone regions worldwide, contributing to the broader discourse on geohazard resilience and sustainable development.