Clustering analysis of ERT/IP data for leachate mapping in urban waste landfills

In this work, we present a quantitative integration of electrical resistivity and induced polarization tomographic data for imaging leachate in urban waste landfills. The main goal is to reduce the residual ambiguities, often arising from a speculative interpretation of the geophysical models in such complex scenarios, providing a more effective image of the hazardous zones linked to the leachate accumulation. Field data, acquired in a municipal waste site, are firstly inverted for resistivity and chargeability. Then, we use log-transformed inverted parameters (resistivity and normalized chargeability) as the input for a clustering analysis performed by the K-means unsupervised machine learning-based algorithm. We found in the log-transformed space a clear decreasing trend from the hazardous to the non-hazardous areas, which eases the clustering labelling in terms of different hazard levels. The most hazardous areas, likely due to the leachate accumulation zones, are found not only at the bottom of the landfill, as expected from a standalone data inversion, but also in the shallower part, accordingly to well data. Our findings show the potential of machine learning-based techniques for data integration, offering new perspectives for the characterization of landfills.