Extracting Hydrofacies Patterns from a Real Time-lapse Electrical Resistivity Dataset Using Time Series Clustering Approaches

Abstract:

Electrical Resistivity Tomography (ERT) offers the opportunity of the non-invasive detection of spatial features in the shallow subsurface. In particular, we define hydrofacies as geological or hydrological structures presenting similar resistivity dynamics, and which could be extracted using Time Series Clustering (TSC) on a time-lapse ERT dataset. The latter is obtained from a 48 electrodes profile installed on the top of the Rochefort cave in Southern Belgium. Data consists of resistivity time-series defined over 465 days and associated with the 1558 cells of the 2D ERT model derived from a time-lapse inversion. While using various representations of the inverted time-series (raw, standardized, differenced data, or principal components), three clustering algorithms are considered: k-means, hierarchical agglomerative clustering (HAC), and Gaussian Mixture Model (GMM). The clustering results are evaluated using clustering validation indices and confronted with the expert-based structural model of the site.

The k-means, HAC, and GMM clusters reveal the spatial pattern of correlated resistivity time-series on standardized data similarly. Some clusters are spatially split and include time-series with a wide range of mean resistivity, suggesting different geological units within these hydrofacies groups. In general, applying TSC to various time-series representation leads to different spatial patterns, but allows gaining confidence from shared redundancies. We also tested the appropriate duration of the measurements: the TSC patterns obtained from the full dataset cannot be reproduced from continuous sub-samples up to 100 days, but well from less than 20 samples picked randomly over the 465 days. This finding suggests monitoring the subsurface system long enough and in a wide range of environmental conditions. Accordingly, this study highlights the importance of time-variable parameters in the identification of structural facies and hydrofacies with ERT while demonstrating the strength of long-term monitoring. It also encourages the retrieval of hydrofacies by combining multiple TSC approaches or, multivariate geophysical dataset, to expect a faster convergence to stable hydrofacies patterns.