Characterizing an unknown pollution source in groundwater resources systems using PSVM and PNN

This paper presents a new methodology for estimating location and amount of leakage from an unknown pollution source using groundwater quality monitoring data. The proposed methodology includes a multi-objective optimization model, namely Non-dominated Sorting Genetic Algorithm-II (NSGA-II) which is linked with MODFLOW and MT3D groundwater quantity and quality simulation models. The main characteristics of an unknown groundwater pollution source are estimated using two probabilistic simulation models, namely Probabilistic Support Vector Machines (PSVMs) and Probabilistic Neural Networks (PNNs). In real-time groundwater monitoring, these trained probabilistic simulation models can present the probability mass function of an unknown pollution source location and the relative error in estimating the amount of leakage based on the observed concentrations of water quality indicator at the monitoring wells. The efficiency of the proposed methodology is demonstrated through a real-world case study.