A novel BMA-PMOSO model for optimal design of groundwater remediation systems under multi-source uncertainties

• Multi-source uncertainties of model structure and parameters are considered • BMA theory is used to quantify uncertainties of objective functions • BMA-PMOSO can quantify and distinguish uncertainties of model parameters or structure • A new OR-IDS method is proposed to help the final decision making The predictive accuracy of groundwater flow and contaminant transport simulation models is often limited by multi-source uncertainties associated with model structure and model parameters. Previous studies on optimizing groundwater contaminant remediation systems (GCRSs) have not adequately accounted for these multi-source uncertainties, thereby compromising the reliability of remediation schemes and increasing the risk of remediation failure. In this study, a novel Bayesian model averaging-probabilistic multi-objective simulation-optimization (BMA-PMOSO) model is developed to optimally design GCRSs under multi-source uncertainties. In the new model, the BMA theory is employed to calculate the mean and variance values of objective functions. The Arithmetic Mean Estimator (AME) method is used to determine the marginal likelihoods of each conceptual models and subsequently calculate their posterior weights. The Differential Evolution Adaptive Metropolis (DREAM) algorithm is applied to analyze the parameter uncertainty for each conceptual model, thereby obtaining the posterior distribution of parameters. The PMOFHS-GA is adopted to identify Pareto solutions. Furthermore, a new final decision-making strategy, the objective reliability based improved decision strategy (OR-IDS), is developed to assist decision-makers in selecting the final best remediation scheme. The proposed model is applied to a 2-dimensional synthetic problem and a 3-dimensional field problem. Results demonstrate that uncertainties in both model structure and parameters will cause significant impacts on optimization results. The BMA-PMOSO model can effectively optimize GCRSs under complex uncertainties, offering decision-makers highly reliable remediation schemes.