Integrating Space Geodesy and Deep Learning for Mapping and Monitoring Land Subsidence Hazard: A Case Study from the Nile Valley, Egypt
Amir Ahmed is a PhD student in the Department of Applied Earth Sciences. (Co)Promotors are prof..dr. M. van der Meijde; dr.ir. I. Manzella and dr. I.E.O.M. Fadel from the faculty of Geo-Information Science and Earth Observation, (ITC) University of Twente.
Land subsidence is a widespread geohazard involving the gradual or sudden sinking of the Earth’s surface. It arises from natural processes such as soil compaction and geological conditions, as well as human activities including groundwater extraction, irrigation, and increased surface loading. Subsidence can cause significant social, economic, and environmental impacts, particularly in regions experiencing rapid population growth and intensive land and water use.
This thesis addresses the lack of a regional-scale assessment of land subsidence across Egypt’s Nile Valley and Delta by providing a comprehensive InSAR-based analysis that identifies spatial patterns of deformation and highlights the most vulnerable areas. SBAS processing was carried out using LiCSBAS for the Nile Delta and GMTSAR for the entire Nile Valley, and a new Python-based workflow (SNAPWF) was developed to enable flexible PSI and SBAS processing. A ConvLSTM deep learning model, combined with SHAP explainability, was applied to quantify the relationship between land subsidence and its main environmental and anthropogenic drivers.
Results from the Nile Delta show significant subsidence in coastal and reclaimed areas, with rates reaching –15 mm/year in major cities such as Alexandria, Damietta, and Rosetta. The deep learning model achieved high prediction accuracy and linked subsidence to key proxies, including land-use/land-cover change, urban load, TWS, and clay content.
At the regional scale, this thesis presents the first full InSAR deformation time series of the Nile Valley, generated from 14 Sentinel-1 ascending frames and approximately 15,000 interferograms. While most of the valley remains stable, localized subsidence exceeding -12 mm/year was observed in governorates such as Port Said, Assiut, Monufia, and Qaliubiya. Applying the deep learning model across five selected regions showed that the dominant drivers vary spatially: coastal areas are influenced by Sabkha deposits, clay-rich sediments, and water fluctuations, while southern regions are more affected by geological structures, soil composition, and vegetation-related proxies.
Overall, the integrated use of InSAR, deep learning, and explainable AI provides a scalable framework for identifying subsidence hotspots, understanding their key drivers, and supporting future monitoring and mitigation strategies. This approach can be adapted to other vulnerable or data-scarce regions worldwide.
