NMF-based ensemble clustering for GNSS velocity field: tectonic insights

Abstract

Block modeling is a frequently employed technique to yield the block motions such as block
rotations and internal strains as well as to estimate fault slip rates via GNSS velocity data. The
accuracy of the modeling outcomes heavily relies on the precise determination of the block
boundaries. Typically, surface traces of mapped faults, seismological data, and/or GNSS
velocity fields are utilized for this purpose. Nonetheless, the identification of suitable and
accurate block boundaries for the velocity field is subject to interpretation and subjectivity.
One method that can be used to determine block boundaries is the clustering method.
Clustering analysis assigns data to similar groups based on similarities and differences in the
data subject to clustering. Since the 2000s, this method has been used in the clustering of GNSS
velocities and helps to determine block boundaries before block modeling. In this study, we
utilized the Non-negative Matrix Factorization (NMF) method to cluster the current GNSS
horizontal velocity field of Türkiye, and we compared the results with the literature. First, we
clustered velocities from k = 2 to 9 using five distinct single clustering methods to determine
the block boundaries. Then, we used the optimum number k = 5 to cluster the data using NMFbased ensemble clustering