Unsupervised clustering to jointly interpret geophysical datasets across the Alhama de Murcia active fault, Spain.

Joint interpretation of multidisciplinary geophysical data is the best way to reduce ambiguity in subsurface exploration. The combination of seismic velocity and electrical
resistivity has proven to be an excellent geological characterization strategy, however, the integration of these geophysical parameters is a complex process. In this work, we use unsupervised clustering to jointly interpret three geophysical datasets (P wave velocity, S wave velocity, and electrical resistivity). The target is a cross-section across the Alhama de Murcia Fault (FAM), which is one of the main active faults in the Iberian Peninsula. In our approach, we first join the three datasets into a common multiparametric grid. Then, in order to find data clusters that can be correlated with known lithologies in the area, we investigated the performance of three unsupervised machine learning algorithms: one hierarchical, one centroid-based, and one model-based. The latter proved to be the most efficient for clustering our highly mixed data and providing geological meaning. The three classes obtained correlated well with the lithological units present in the area and, from their relationship, it was possible to deduce structural elements not yet well understood, providing new perspectives in the characterization of the Alhama de Murcia fault zone. Research supported by grants: VECTOR EU project ID 101058483, and SIT4ME -EITRawMaterials.