Development of an efficient conjunctive meta-model-based decision-making framework for saltwater intrusion management in coastal aquifers

This paper presents an integrated framework for management of aquifers threatened by saltwater intrusion (SI). In this framework, SEAWAT model is used for simulating the density dependent groundwater flow. Three meta-models based on the artificial neural network (ANN), M5 tree and random subspaces model (RSM) are developed, as surrogate models for SEAWAT to accurately simulate the groundwater response to different pumping and recharge scenarios. Various patterns of recharge to and discharge from aquifer are used to generate a database for training the mentioned surrogate models. To decrease the number of training parameters, the aquifer area is divided into different zones using k-means clustering technique (KMC). Additionally, a conjunctive model (CM) using a combination of the three surrogate models is proposed to enhance the accuracy of the simulation. It is then integrated with the Non-dominated Sorting Genetic Algorithm-II (NSGA-II) with the objectives of maximizing pumping rates and minimizing SI length. Next, the socially optimal scenarios are selected from the obtained Pareto-front using the Nash bargaining theory. The performance of the proposed model is evaluated by applying it to the Kahak aquifer, Iran, which is subjected to SI. The results show that the conjunctive model using KMC technique predicts SI length with a comparable accuracy and results in 95% reduction in runtime compared to a simulation-optimization (SO) model.