Near real-time subsidence monitoring and AI forecasting with multi-depth extensometers

Land subsidence induced by excessive groundwater withdrawal poses a growing threat to infrastructure, water resources, and environmental sustainability. Developing robust early warning systems and accurate subsidence forecasts remains a significant key challenge in engineering geology. This study integrates deep, high-frequency extensometer data with AI-based prediction to improve short-term land subsidence forecasting—a novel framework particularly suited for subsidence-prone, groundwater-stressed regions. The AI-driven model enhances predictive accuracy by 35 %, effectively capturing both long-term subsidence trends and seasonal variability. We compare extensometer observations with multilayer compaction well (MLCW) data to evaluate their respective advantages. MLCWs provide millimeter-resolution compaction measurements at up to 20 discrete depths, revealing depth-dependent aquifer-system deformation. Extensometers, by contrast, offer continuous measurements at selected depths (10-min intervals), enabling near-real-time detection of vertical displacement. When combined, the two systems form a hybrid monitoring framework with enhanced spatial and temporal resolution. This integrated approach supports more accurate subsidence assessment and forecasting, informing groundwater management strategies and infrastructure risk mitigation. Our results demonstrate that coupled monitoring and AI modeling are essential tools for sustainable groundwater development in subsidence-prone regions.