Bayesian Chance-constrained Saltwater Scavenging Design for Deep Aquifers under Surrogate Uncertainty

The Baton Rouge aquifer system in the southeastern Louisiana supplies high-quality groundwater and is imperative to the social-economic development in the greater Baton Rouge area. However, prolonged over pumping and lacking management plans, many freshwater aquifers are being threatened by saltwater intrusion. The goal of this study is to numerically test a horizontal scavenger well in order to hydraulically control the movement of a salty plume. This study targets the chloride-elevated "2,000-foot" sand and investigates the effectiveness using a horizontal scavenger well. Firstly, a high-fidelity groundwater flow and solute transport model is developed to simulate chloride transport in the Baton Rouge multi-aquifer system. Secondly, optimized saltwater scavenging strategies are developed, which take into account modeling uncertainty. Developing optimal saltwater scavenging strategies using the physically-based simulation model is almost impossible due to massive model runs in optimization. Instead, in what follows, this study presents an ensemble surrogate modeling approach to address uncertainty in surrogate model predictions and improve computational efficiency in the saltwater intrusion mitigation design. Surrogate models are trained and verified by the output data from the physically-based solute transport model. The optimal saltwater scavenging strategies obtained from the surrogate models are placed back to the solute transport model to make sure all solutions satisfy mitigation performance. The Bayesian model averaging method is introduced in conjunction with chance-constrained programming to quantify uncertainties in surrogate model predictions. Results show that using a horizontal scavenger well can be an effective mitigation approach. The developed surrogate models perform well and prove strong predictive capabilities while considerably reducing computation time. The ensemble-surrogate-assisted simulation-optimization technique can identify reliable optimal saltwater scavenging strategies. This study provides a management framework to serve as a scientific tool that can assists policy makers to mitigate the urgent saltwater encroachment issue.