An unsupervised K-means based clustering method for geophysical post-earthquake diagnosis

After an earthquake takes place, the geophysical problem to solve has always been to determine the geodynamical behaviour of the rupture zone. Nonetheless, such geodynamical process is usually a result of the contribution of more than one enchained tectonic process. At this point, the geophysical clustering problem turns in a complex task from a pattern recognition approach, due to the dynamical behaviour of the problem to solve. In this regard, the use of unsupervised techniques is mandatory. Such techniques, however, present three important drawbacks for their application to the addressed problem: (1) the identification of the number of clusters and their spatial localization from a geophysical point of view, (2) the geodynamical sense of the performed clusters, and (3) the statistical dependence between features. To solve such limitations we developed the GEO K-means algorithm by incorporating precursor contextual information within the clustering process. In this respect, our study provides a novel approach to scientists for a more accurate post-earthquake diagnosis about the geodynamical behaviour of the affected area.