Hierarchical cluster analysis and multiple event relocation of seismic event clusters in Hungary between 2000 and 2016

Abstract

The objective of our paper is to develop a workflow that allows us to calculate more accurate hypocenter locations in seismic event clusters of aftershock sequences or artificial events. Due to the increased sensitivity of the seismological instruments and density of the network, we are able to record small natural and artificial events. The discrimination of these events is necessary to investigate the recent tectonic movements in the Pannonian Basin. As a first step, we performed a hierarchical cluster analysis on the events in the Hungarian National Seismological Bulletin using the spatial distances between the events to obtain event clusters. We selected 5 different test clusters from the list of clusters where two clusters consist of quarry blasts, another two consist of earthquakes, and the last one is a mixture of earthquakes and anthropogenic events. In the second step, to prepare for the double-difference multiple event location analysis, we manually revised the arrival time picks in the Hungarian National Seismological Bulletin in order to increase the consistency and accuracy of the arrival times. We obtained improved single-event locations with the iLoc algorithm using the RSTT 3D global velocity model to provide initial locations for the double-difference relocation. We applied waveform cross-correlation at every station to obtain the differential times and correlation matrices. In order to discriminate the events in the mixed event cluster, we repeated the hierarchical cluster analysis, but this time, we used the correlation matrix as a distance metric. Examining the shape of the resulting dendrogram, it is clear that certain subclusters are well separated. In these subclusters, the coordinates of the events are close to the mines, where explosive quarrying takes place. With this technique, we are able to identify explosions that were listed as earthquakes in the catalogue.

An unsupervised machine learning approach for ground‐motion spectra clustering and selection

“…Hierarchical agglomerative clustering (Agg.) has also been used for clustering static earthquake features 26,36 . In hierarchical agglomerative clustering, a “bottom‐up” approach, each observation starts as a separate cluster and iteratively merges, using the pairwise distance or dissimilarity between observations and clusters and a given linkage criterion, to the most similar cluster until a single cluster containing all the data is formed 37,38 …”

“…has also been used for clustering static earthquake features. 26,36 In hierarchical agglomerative clustering, a "bottom-up" approach, each observation starts as a separate cluster and iteratively merges, using the pairwise distance or dissimilarity between observations and clusters and a given linkage criterion, to the most similar cluster until a single cluster containing all the data is formed. 37,38 A gaussian mixture model (GMM) was also utilized to perform clustering on the component space beginning with random initialization of Gaussian distribution parameters and iteratively updating them using the Expectation-Maximization (EM) algorithm until convergence.…”

Assessing similarity in continuous seismic cross-correlation functions using hierarchical clustering: application to Ruapehu and Piton de la Fournaise volcanoes

“…Popular algorithms include the single linkage (or nearest neighbor) method (e.g. Sibson, 1973;Czecze and Bondár, 2019), complete linkage (or furthest neighbor) method (e.g. Deyasi et al, 2017), the average linkage method (e.g.…”

The Hiccup of Villarrica volcano (Chile) during the 2015 eruption and its expression in LP activity and VLP ground motion

“…The primary purpose of the hierarchical cluster analysis is to divide the fingerprint similarity given by the symmetric matrix CC ij into smaller groups that are useful for locating similarity between different days. Following Czecze & Bondár (2019), we apply a hierarchical cluster analysis using a matrix of Euclidean distances between fingerprint similarities constructed from the CC ij matrix. The grouping method is agglomerative, which means that each fingerprint similarity is a cluster at the beginning of the algorithm.…”