Contaminant plume reconstruction through joint geophysical and concentration data assimilation

Pollutants represent a significant environmental threat, impacting soil and groundwater quality and posing risks to human health. The inherent complexity of subsurface processes raises challenges to the precise location and definition of the extent of contaminant plumes. Hydrogeophysics emerges as a powerful tool by enabling the simultaneous combination of non-invasive geophysical techniques ( i.e., electrical resistivity, seismic survey…) and hydrological variables. This field contributes to a comprehensive approach to understanding and managing contaminant plumes, ensuring effective responses to environmental problems. While previous research has demonstrated the ability of hydrogeophysics to estimate aquifer characteristics, only a few studies have focused on its use to predict groundwater contamination. This work aims to accurately estimate contaminant spread and distribution within the subsurface system. To this aim, a novel approach that combines Ensemble Smoother with Multiple Data Assimilation (ES-MDA) with Convolutional Neural Network (CNN) is introduced. ES-MDA is applied to solve the inverse problem by assimilating Electrical Resistivity Tomography (ERT) data and sparse concentrations observed at monitoring wells, while a CNN replaces the forward model to reduce the computational effort. ES-MDA integrates observational data into numerical models to increase model accuracy and reduce uncertainties, whereas CNNs use spatial data to detect and characterize contamination plumes, particularly in imagery datasets. A 2D synthetic case study simulating a tracer test in a fully saturated unconfined aquifer is conducted to test the methodology. The study examined five different datasets to assess the performance of the proposed approach, allowing for a thorough examination of the advantages of combining data from multiple sources, as well as the effects of various observation datasets on the accuracy of plume distribution assessments. The first scenario interpolated 15 concentration values using a kriging-based approach, while subsequent scenarios tested the suggested inverse hydrogeophysical approach's capabilities. The second scenario used only apparent resistivity data as observations into the ES-MDA; while the third to fifth scenarios combined apparent resistivity data with different subsets of concentration values: 15, 9, and 3, respectively. The third scenario, which combines apparent resistivity with 15 concentration values, proved to be the most accurate and precise setup. The least accurate estimations were obtained using kriging interpolation (Scenario 1) and ES-MDA with only apparent resistivity data (Scenario 2). These findings suggest that a limited dataset may not provide sufficient information to capture the spatial variability of subsurface concentration maps accurately. Furthermore, they emphasize the importance of a more comprehensive approach and the need to combine multiple data sources when precise mapping of tracer plumes is desired. The results underscore the effectiveness of the proposed integrated approach in accurately estimating the spatial distribution of a concentration plume, offering a valuable and cost-effective tool for supporting optimal strategies in contaminated site remediation and contributing to the sustainable management of groundwater resources.