Joint identification of contaminant source characteristics and hydraulic conductivity in a tide-influenced coastal aquifer

Coastal aquifers are a vital water source for the more than one billion people living in coastal regions around the globe. Due to the intensity of economic activities and density of population, these aquifers are highly susceptible not only to seawater intrusion, but also to anthropogenic contamination, which may contaminate the aquifer and submarine groundwater discharge. Identification and localization of contaminant source characteristics are needed to reduce contamination. The techniques of contaminant source identification are based on numerical models that require the knowledge of the hydrodynamic properties of aquifers. Thus, the challenging topic of contaminant source and aquifer characterization (CSAC) is widely developed in the literature. However, most of the existing studies are concerned with inland aquifers with relatively uniform groundwater flow. Coastal aquifers are influenced by density-driven seawater intrusion, tidal forces, and water injection and abstraction wells. These phenomena create complex flow and transport patterns, which render the CSAC especially challenging and may explain why CSAC has never been addressed in coastal settings. The presented study aims to provide an efficient methodology for the simultaneous identification of contaminant source characteristics and aquifer hydraulic conductivity in coastal aquifers. For this purpose, the study employs numerical modeling of density-dependent flow and multiple-species solute transport, to develop trained and validated artificial neural network metamodels, and then employs these metamodels in a version of the ensemble Kalman filter (EnKF) termed the 'constrained restart dual EnKF (CRD-EnKF)' algorithm. We show that this variant of the EnKF can be successfully applied to CSAC in the complex setting of coastal aquifers. Furthermore, the study analyzes the influence of common issues in CSAC monitoring, such as the effect of non-ideal monitoring network distributions, measurement errors, and multi-level vs. single level monitoring wells.