A Comprehensive Data-Driven Approach for Detecting Regime Shifts in Coastal Groundwater: Towards an Integrated Machine Learning Framework

Escalating climate crises, characterized by rising sea levels, alongside excessive groundwater pumping, have severely exacerbated saltwater intrusion, posing a critical threat to coastal aquifers. These combined environmental stressors induce complex, non-linear dynamics in groundwater systems, making the exact prediction of regime shifts driven by tipping points increasingly challenging. To address these uncertainties, this study proposes a comprehensive data-driven approach designed to identify early warning signals (EWS) for approaching tipping points using Electrical Conductivity (EC) time-series data. The primary objective is to investigate the feasibility of utilizing complementary statistical indicators—Variance and Fisher Information (FI)—to assess system instability. We analyzed monitoring data from Incheon and Jeju, South Korea, to validate whether these metrics can effectively filter noise and detect genuine precursor signals. Empirical results demonstrate that our approach achieves significantly enhanced performance in distinguishing critical transitions compared to single-indicator methods. Ultimately, this study serves as a foundational step towards establishing an "Integrated Machine Learning" framework. By validating these statistical metrics as key features, we aim to incorporate them into advanced learning algorithms to further improve the robustness and predictive accuracy of coastal groundwater management systems against climate-induced risks.

Acknowledgement
This work was supported by National Research Foundation of Korea(NRF) grant funded by the Ministry of Science and Technology (RS-2024-00356786).