On the origin of upper crustal shear-wave anisotropy at Samos Island, Greece

Abstract

Shear-wave splitting is associated to different sources in the upper crust. Preferentially oriented minerals, stress-aligned microcracks and tectonic structures have all been identified as causes of seismic anisotropy in the upper crust. However, distinguishing between them and discovering the actual origin of the splitting effect has important implications; changes in the anisotropic properties of the medium related to the behavior of fluid-filled microcracks could have potential connections to the occurrence of an impending significant earthquake. The recent 2020 Samos M_w = 6.9 event and its associated sequence was a great opportunity to study shear-wave splitting in the area. The spatial constrains in such studies, i.e., the requirement of events located very close to the receivers, did not permit exploring local anisotropy in the past, due to a severe lack of suitable data. To establish a background of splitting, we searched for any appropriate earthquake in a five-year period preceding the mainshock. We performed an automatic analysis on over 200 event-station pairs and obtained 164 high-quality splitting observations between January 2015 and November 2020. Results indicated a strong connection to local structures; S_fast polarization axes seem to align with faults in the area. However, we also observed a period of increasing and decreasing time-delays, associated with an M_w = 6.3 earthquake that occurred on June 2017 near Lesvos Island. The latter behavior implies the possibility of stress-induced anisotropy in the area. Thus, the Samos Island could be represented by two different sources of splitting; structures to the NW and microcracks to the SE.

Appendices

Appendix 1–Shear-wave splitting processing

To process as many station-event pairs as possible and remove user bias during the analysis, we employed a fully automatic process using the Pytheas software (Spingos et al. 2020). As a first preprocessing scheme, a series of user-predefined band-pass filters were applied to the initial waveforms and the one yielding the highest SNR was selected (Savage et al. 2010). The Eigenvalues (EV) method of Silver and Chan (1991) performs a series of shear-wave splitting corrections based on different combinations of φ and t_d. For each set of parameters, the covariance matrix of the two horizontal components, after correction, is obtained and the second, minimum, eigenvalue (λ₂) is extracted. The φ and t_d pair that yielded the lowest λ₂ is considered as the optimal measurement. In Fig. 7 we present an example of the analysis with the EV method. In the selected signal window (see next paragraph for its automated selection) the particle motion was linearized after correcting for anisotropy. Moreover, Silver and Chan (1991) offered a comprehensive error estimation system. However, Walsh et al. (2013) identified an underestimation in the original system and proposed new formulations which resolved the issue. We followed the formulations of the latter.

Fig. 7

Summary report of the Eigenvalue method, showcasing the waveforms (a) and particle motion diagrams (b) before and after the removal of the splitting effect. The optimal signal window selected through cluster analysis is represented by the shaded area. The contour plot (c) displays the variation of the minimum eigenvalue (λ₂) of the radial and transverse components covariance matrix after the removal of anisotropy, with the crosshair indicating the minimum value and the 95% confidence interval outlined (bold contour). Note the linearization of the particle motion in the NE plane, after the removal of splitting (panel b, bottom)

To automatically select the signal window analyzed by the EV method, we adopted the Teanby et al. (2004) approach, which utilizes cluster analysis (Fig. 8). In brief, EV is first applied to a prefixed range of candidate signal windows. Then, clusters are hierarchically formed in the initial space of φ and t_d observations. Then, the number of optimal clusters is estimated and, consequently, the most constrained cluster is identified. Out of the latter, the observation pair with the minimum errors corresponds to the optimal signal window.

Fig. 8

Summary of the results of cluster analysis for 37 candidate signal windows. Top: the initial space of φ and t_d observations used in clustering. Middle: clusters of measurements (left) as obtained by the algorithm and selection of the final cluster and measurement, as indicated by the crosshair (right). Bottom: variation of φ (left) and t_d (right) per index number of candidate signal window. The bottom plots essentially showcase the stability of either parameter with differing windows

Appendix 2–Null measurements

In the following, we present rose diagrams for all measurements graded as “null” (Fig. 9). SAM2 and SMG stations exhibit only a few null measurements. KRL1 showcases a dominant NW–SE direction of null measurements. This is similar to the direction of possibly flipped microcracks near the causative fault of the 2020 Samos earthquake, as discussed in the main text.

Fig. 9

Rose diagrams displaying the distribution of φ for each station, for observations characterized as “null”. N is the total number of observations and F the count of measurements per grid line