Geophysics and Machine Learning Based Surveying of Groundwater Pollution: A Case Study of Southern Acidic Tar Lagoon, Inčukalns, Latvia

Acid tar lagoons (ATLs) are a common source of groundwater pollution in industrialized countries worldwide. Even after recultivation, the spatial extent of the residual environmental pollution from such lagoons is often unknown, as the monitoring of groundwater is mostly done in discrete locations - monitoring wells. Shallow surface geophysical methods, such as ground penetrating radar (GPR) and electrical resistivity tomography (ERT) are often used in monitoring groundwater and soil contamination. The greatest value from shallow surface exploration can be received by coupling multiple geophysical methods, which can be a challenge, considering the different geophysical properties measured. We propose the use of machine learning techniques to identify the different subsurface contamination zones. In this study we combine two-dimensional electrical resistivity measurements with GPR amplitude measurements using simple k-means clustering to map out residual soil contamination zones near the acid tar lagoon of Inčukalns, Latvia. We then compare the clustering results with more conventional quasi-Newton two-dimensional ERT inversion results of the same site. Information obtained from clustering shows promise in future geophysical exploration, as comparable results can be obtained with fewer assumptions of subsurface geology. Our study demonstrates the potential of clustering methods to integrate different shallow-surface geophysical exploration methods and points out to possible future improvements in machine-learning based soil contamination mapping.