Decoding nutrient dynamics in coastal aquifers: Machine learning insights into submarine groundwater discharge and...

Coastal aquifers are vulnerable to natural and human-induced processes that impact their resilience and ecosystems. Submarine Groundwater Discharge (SGD) and Seawater Intrusion (SWI) play crucial roles in transporting nutrients and contaminants into coastal waters and threatening coastal aquifers, respectively. This study aims to characterize hydrogeochemical processes governing SGD and SWI using an integrated machine learning (ML) algorithm, overcoming limitations of traditional geochemical methods in analysing complex, nonlinear, and high-dimensional hydrogeochemical datasets. The ML framework, integrating statistical and geochemical analyses, was applied to Ramanathapuram and Rameswaram Island coastal aquifers. Spearman correlation analysis identified key indicators of seawater influence, anthropogenic inputs, redox reactions, dissolution, and ion exchange. Self-Organizing Maps (SOM) and Fuzzy C-Means (FCM) clustering revealed hydrogeochemical patterns in groundwater (GW) and porewater (PW). In GW, Group 1 (27 %) indicated pollution from agricultural NO₃^-, Group 2 (60 %) represented long-residence freshwater with high DSi and reduced NO₃^- under clay-layer redox conditions, and Group 3 (13 %) contained high-salinity GW impacted by SWI and saline traps. In PW, Group 1 (11 %) reflected fresh SGD with high DSi and NH₄⁺, Group 2 (68 %) showed SW dominance in intertidal zones, and Group 3 (21 %) represented recirculated SGD enriched in salinity and nutrients due to ion exchange and desorption reactions. Factor analysis clarified hydrogeochemical drivers such as anthropogenic inputs, silicate dissolution, redox reactions, and SW interactions, while ionic ratios (Na/Cl, NO₃^-/Cl) and delta-analysis geochemically supported these findings. This ML-based approach enhances SGD identification and SWI assessment, offering a novel methodology for coastal aquifer management and ecosystem protection.