Integrating Multidisciplinary Observations and AI-Driven Modeling for Land Subsidence Management in Taiwan

Land subsidence driven by intensive groundwater abstraction remains a major concern in Taiwan, particularly in Yunlin County. This study integrates multidisciplinary observations with artificial intelligence (AI)-driven modeling to support land-subsidence management. A hydrogeological conceptual model was developed to simulate groundwater-level dynamics and aquifer-system compaction, and an AI approach was used to capture the nonlinear relationship between groundwater fluctuations and soil-layer compression. Results indicate that subsidence is influenced by climate extremes and pumping intensity. The strong positive correlation and synchronized temporal variations between groundwater level and soil compression suggest a coupled hydro-mechanical response. To identify mitigation measures, five scenarios were evaluated, focusing on crop conversion and pumping regulation. Compared with current pumping conditions, both crop conversion and rotational pumping reduce groundwater drawdown and associated compression. Among the alternatives, conversion to sweet potato combined with rotational pumping yields the smallest drawdown, indicating a practical pathway for sustainable groundwater management and land-subsidence mitigation.