From clustering to ensemble clustering in GNSS velocities: A Meta CLustering-based approach

Abstract

Although there are different approaches and models to understand and interpret the structures in crustal deformations, one of them is the block modeling method. Using block modeling, one can determine plate movements, parameters such as slip rates, locking depths or Euler poles on faults. However, the accuracy of the block modeling results is related to how well the block boundaries are determined. One of the most important steps of block modeling is the detection of block boundaries and clustering can be used as a tool for this. Cluster analysis assigns data to similar groups based on similarities and differences in the data subject to clustering. In this study, Türkiye was determined as the study area. In this context, we utilized the ensemble clustering algorithm to cluster recent Global Navigation Satellite Systems (GNSS) velocity field in Türkiye and determine block boundaries. Current GNSS velocity field, which consists of 78% survey and 22% continuous type GNSS data processed together, used for clustering analysis for the first time in this study. Before clustering, we employed three different methods-Davies-Bouldin, Gap statistics, and Silhouette-to determine the optimal number (cluster number that best fit to GNSS velocity field) of clusters. Then, k-means, HAC, and spectral clustering techniques were then applied to cluster current GNSS velocities. Finally, we utilized the Meta-Clustering Algorithm (MCLA) as an ensemble clustering technique to cluster the horizontal components of the current velocity domain and present our findings. © 2023, Gumushane University. All rights reserved.