Optimal pumping strategies for the management of coastal groundwater resources: application of Gaussian Process Regression metamodel-based simulation-optimization methodology

Abstract

ABSTRACT The present study utilizes a coupled simulation-optimization (S-O) methodology to develop a multi-objective management strategy for a coastal aquifer system. The aim of the multi-objective management model is to maximize pumping from freshwater wells (FWs) and minimize pumping from the barrier wells (BWs), while keeping salinity concentration in the aquifer within pre-specified limits (optimization constraint). To achieve computational feasibility of the management model, the numerical simulation model is substituted by the relatively new Gaussian Process Regression (GPR) metamodels. The GPR models are used to approximate coastal aquifer responses to variable transient pumping patterns from FWs and BWs. Prediction capabilities of the developed GPR metamodels are quantified using standard statistical parameters. Once trained and validated, the GPR metamodels are coupled to a multi-objective genetic algorithm optimization model and used to prescribe optimal groundwater pumping patterns. The outcomes of this study establishes the potential applicability of the GPR metamodel-based S-O model for developing sustainable coastal groundwater management strategies, which can utilize accurate and efficient prediction of management strategy impacts on the saltwater intrusion (SI) process when the optimal management policy development is based on the trained metamodel predictions. Once implemented, the developed strategy can help in controlling SI in coastal aquifer systems.