Spatial-temporal patterns and drivers of coastal water quality dynamics: Insights from explainable machine learning...

Effective management of coastal water quality is challenging in urbanized regions, where dynamic land-sea interaction and intensifying human pressures drive ecological change. This study proposes an integrated approach that combines Self-Organizing Maps (SOM) to identify spatial-temporal water quality patterns, machine learning algorithms coupled with explainable machine learning (SHapley Additive exPlanations, SHAP) to uncover key predictive drivers, and Partial Least Squares Structural Equation Modeling (PLS-SEM) to evaluate causal pathways. SOM analysis revealed two distinct clusters with turning point in water quality cluster between 2012 and 2018, corresponding closely with intensified coastal reclamation and socioeconomic activities. Among eleven machine learning algorithms tested, LightGBM outperformed others with over 94% accuracy and F1 score 0.88. SHAP and PLS-SEM analysis further identified land cover alterations, industrial emission, and socioeconomic variables as key drivers. Integrating these findings with historical policy contexts demonstrated the positive impacts of proactive management frameworks such as Integrated Coastal Management (ICM) and Marine Functional Zoning (MFZ). By bridging spatial-temporal pattern recognition with causal interpretation, this research provides data-driven insights into sustainable coastal water management in a rapidly urbanized coastal bay.