Facies discrimination with electrical resistivity tomography using a probabilistic methodology: Effect of sensitivity and regularization

Electrical resistivity tomography (ERT) has become a standard geophysical method in the field of hydrogeology, as it has the potential to provide important information regarding the spatial distribution of facies. However, inverted ERT images tend to be grossly smoothed versions of reality because of the regularization of the inverse problem. In this study, we use a probabilistic methodology based upon co-located measurements to assess the utility of ERT to identify hydrofacies in alluvial aquifers. With this methodology, ERT images are interpreted in terms of the probability of belonging to pre-defined hydrofacies. We first analyze through a synthetic study the ability of ERT to discriminate between different facies. As ERT data suffer from a loss of sensitivity with depth, we find that low sensitivity regions are more affected by misclassification. To counteract this effect, we adapt the probabilistic framework to include the spatially varying data sensitivity. We then apply our learning to a field case. For the latter, we consider two different regularization procedures. In contrast to the data sensitivity which affects the facies probability to a limited amount, the regularization can affect the probability maps more considerably because it has a strong influence on the spatial distribution of inverted resistivity. We find that a regularization strategy based on the most realistic prior information tends to offer the most reliable discrimination of facies. Our results confirm the ability of ERT surveys, when properly designed, to detect facies variations in alluvial aquifers. The method can be easily extended to other contexts.