Item

Abstract

The goal of this contribution is to present the “fully-automated” clustering algorithm developed by the authors. This is augmented by a case study where the authors carried out various cluster analyses of the focal mechanism solutions (FMS) estimated from local and teleseismic measurements and then performed stress inversions using the resulting clusters. This algorithm does not call for the setting of hyper-parameters by the users, thus greatly reducing the level of subjectivity introduced by user choice; the time required to finish the clustering can also be decreased. There is, however, an optional hyper-parameter which can be used for outlier detection, i.e. identify the data points considered to be noise in the input dataset, increasing the performance of the stress estimations.For probing the efficiency of their algorithm, systematic series of tests were executed on synthetic FMS data featuring clusters of various characteristics. During these tests, CluStress has been compared to other established clustering methods, namely, hierarchical density-based clustering for applications with noise (HDBSCAN) and agglomerative hierarchical analysis. Quantitative metrics, mainly the characteristic curves produced using the so-called silhouette-coefficient indicate that CluStress generates the most robust clusters for the stress inversions on synthetic data. A similar conclusion could be drawn from the same test performed using real FMS data of the investigated study area. As for this latter case, the resulting stress tensors show good agreement with the earlier published results with the notable exception of traces of extension, identified in some parts of the region after FMS clustering.