Identification of transient contaminant sources in aquifers through a surrogate model based on a modified self-organizing-maps algorithm

The identification of transient groundwater contaminant sources in terms of source locations, contaminant magnitudes, and active durations remains a challenge. The problem becomes more complex due to spatial heterogeneity, sparse observation data, concentration measurement errors, and unexpected uncertainty. This study addresses this challenge by proposing a modified self-organizing maps (SOM) algorithm; this algorithm can improve the physically-based models by reducing the computational burden more efficiently. The method sufficiently increases the accuracy and efficiency for identifying the contaminant source, because the trained SOM-based surrogate models can identify the source characteristics independently without necessarily operating a formal linked simulation-optimization model. The performance of the proposed method was assessed on a hypothetical heterogeneous aquifer model; the assessment considered unknown observation data, concentration measurement errors, and an unknown pumping well. The proposed SOM-based surrogate model can not only approximate the results from the groundwater flow and transport simulation models, but it can also be used in lieu of the optimization model in a more efficient way for identifying the unknown transient contaminant sources in groundwater systems.