We made a study in the backyard of an agrochemical plant using a small-loop, frequency-domain electromagnetic induction (EMI) system. Such systems are very sensitive to conductive structures buried at shallow depths. Frequently, they are only used to locate and delimit these structures by direct observation of data. However, much more information can be obtained by applying numerical modeling techniques to the data. First we mapped an anomalous zone that indicates the possible presence of buried waste or some other underground contamination by visualizing data. Then we applied a 1D inversion method to the data from this zone. By joining 1D inversion results, this method builds 2D images of the subsoil structure below survey lines. Because the code applies smoothness constraints to the 1D inversions, the subsoil properties in these 2D images change gradually with depth. The code does not impose any correlation between the data or 1D models corresponding to neighboring points, so sharp lateral changes can appear. Several of them do not represent real features of the subsoil. We designed and applied two spatial filters to smooth the spurious lateral variations in our models. One correlates the data acquired at adjacent points prior to inversions. The other applies an analogous correlation to the inverse models obtained from the original data. Both filters greatly improve the quality of the 2D images. Compiling these results, we obtained a 3D model of the subsoil that characterizes the anomalous structure. Excavations made later at the site confirmed the results.
Laterally filtered 1D inversions of small-loop, frequency-domain EMI data from a chemical waste site