Multi-Objective Optimization of Coastal Aquifer Management Based on Temporal GALDIT and Explainable Deep Learning Models

This study introduces a novel framework that utilizes spatiotemporal groundwater susceptibility (STGS) maps to guide advanced surrogate-based optimization techniques for controlling seawater intrusion (SWI). To determine optimal strategies, this study employed a state-of-the-art multi-objective optimization approach based on the reference vector evolutionary algorithm (RVEA) to maximize groundwater levels, minimize artificial recharge, maximize pumping rates, and minimize recharge well costs. This work replaces traditional numerical simulation models with deep explainable neural network (DENN) models for improved efficiency and transparency in simulation-optimization. Additionally, the contribution of input features to the model's predictions is interpreted using shapley additive explanation (SHAP) analysis. Results indicate that strategically placing recharge wells in high-risk areas significantly reduces SWI and improves groundwater levels. By using DENN surrogates informed by STGS into advanced optimization techniques, this study offers forward-looking and practical solutions for sustainable coastal aquifer management.