A Deep Learning Algorithm for Locating Contaminant Plumes From Self-Potential: A Laboratory Perspective

Leachate leakages from municipal landfills are significant environmental problems that threaten groundwater and soil resources. Geophysical techniques, such as the self-potential (SP) method, are commonly used to detect and delineate underground contaminated plumes. However, traditional inversion techniques for SP source information require precise knowledge of subsurface conductivity, which can be challenging to obtain. In this study, we proposed an inversion algorithm, called SP-Net, based on a convolutional neural network (CNN) that can directly train the intrinsic relationship between SP signals and the location of SP sources, while being a great performance within a comparative heterogeneous resistivity setting. In this work, we used the U-shaped network as the structure of the SP-Net and treated the problem of locating SP sources as an image segmentation problem. We designed a sandbox experiment model by adding humus and the microorganism called Shewanella oneidensis MR-1 to simulate the scenario of microbial-mediated SP generation, which typically exists at organic-rich contaminated sites. We used this situation to generate numerous 3-D SP datasets for SP-Net training. We tested the SP-Net on both synthetic testing datasets and the measured laboratory case, and the results show the effectiveness of SP-Net. We also developed a field-scale synthetic model for landfills as a preliminary attempt to test the SP-Net in practical applications, and the results reveal that our study has a valuable reference for potential future applications. Our work provides a promising tool to locate SP sources from both laboratory and field-scale SP data.