Introduction

Seismic anisotropy is a well-known property of propagation media in the Earth, observed in multiple scales; from the mantle (Montagner and Tanimoto 1991; Salimbeni et al. 2008) to the crust (Boness and Zoback 2006a; Crampin et al. 1980; Crampin and Peacock 2005). It refers to changes in seismic velocity in respect to varying directions. In the case of Shear-wave Splitting (SwS), secondary S waves are polarized orthogonally when entering an anisotropic medium. This leads to the distinction of two waves. One travels with the higher velocity (Sfast) and the other has a lesser velocity (Sslow). The Sfast is polarized in line with the dominant anisotropic feature of the medium, orienting its polarization direction (φ) accordingly. A measure of the strength of anisotropy is the time-delay (td) between the arrivals of the two shear-waves at the station. Oftentimes, to reduce the effect of the ray path, this quantity is normalized according to the source-receiver distance (tn). Removing the effect of anisotropy can yield the polarization direction of the non-split shear-wave (p), as if the propagation medium were isotropic.

There is a variety of features that could act as the source of anisotropy. Mainly, a characteristic of the rock volume that the shear-waves traverse must be strongly oriented and homogenously distributed. The lattice preferred orientation of minerals, either olivine for the upper mantle (Ismaïl and Mainprice 1998) or mica and other phyllosilicates in sedimentary and metamorphic lithologies for the crust (Paulssen 2004; Valcke et al. 2006), is commonly considered. There is a plethora of reports of mantle flow-related olivine layers affecting φ and causing splitting in teleseismic waves (Evangelidis, 2017; Kaviris et al. 2018a; Kreemer, 2009). At a local scale, shear-wave splitting is frequently interpreted as induced by aligned microcracks which permeate the upper crust and fluids contained in them. Specifically, the Anisotropic Poro-Elasticity model (APE) provides a connection between regional stress and fluid processes, such as diffusion and migration (Crampin and Zatsepin 1997; Zatsepin and Crampin 1997). APE in tectonic regimes predicts a Sfast polarization axis oriented according to the maximum horizontal stress (SHmax). Moreover, APE has significant implications for earthquake forecasting. Since td works as a proxy for changes in the stress balance of a given rock volume, it can be used to extrapolate accumulation and release of stress associated with an impending strong seismic event (Gao and Crampin 2008). Such connections have been reported in literature (Crampin et al. 1999; Gao et al. 1998; Kaviris et al. 2018b), even though there are also cases debating this claim (Aster et al. 1990; Peng and Ben-Zion 2004). In any case, such variations have been better constrained in volcanic environments, preceding eruptions (Bianco and Zaccarelli 2008; Liu et al. 2010). There have also been observations of changes in the polarization of surface waves obtained from ambient noise, caused by crack-induced anisotropy, before an earthquake (Durand et al. 2011). Finally, shear-wave splitting has been attributed to the existence of dominant tectonic structures, such as faults, which dictate the Sfast polarization direction (Boness and Zoback 2006b; Zinke and Zoback 2000). According to the above, we can identify two possible origins of shear-wave splitting in the upper crust; (a) microcracks controlled by regional stress (stress-induced) and (b) tectonic structures (structurally-controlled).

There have been several shear-wave splitting studies in Greece. Polarization directions in the Western Gulf of Corinth (WGoC), where there is a rich archive of data, are usually oriented according to the SHmax, with a few exceptions that alignment to local faults is present (Bernard et al. 1997; Bouin et al. 1996; Giannopoulos et al. 2015; Kaviris et al. 2017, 2018b). Equivalently, splitting results in the Eastern Gulf of Corinth (Papadimitriou et al. 1999) and Eastern Attica seem to also agree with the APE model; there is even some indication of time-delay variations associated with a Mw = 4.2 event (Kaviris et al. 2018c). The Florina basin featured a more complex state of anisotropy, with a variety of polarization directions observed. However, this was probably affected by the local emission of CO2 and its related processes (Kaviris et al. 2020). All three above-mentioned areas are characterized by normal faulting, as in the case of the earthquakes observed at Samos.

Samos is a Greek island located at the eastern Aegean Sea, about 2 km off the opposite coast of Turkey. The broader region’s tectonics is mainly constituted by the SW motion of the Anatolia block, with a rate of about 30 mm/yr relative to a stable Eurasia (McClusky et al. 2000), mostly accommodated along the North Anatolian Fault, a dextral strike-slip major structure (Papadimitriou and Sykes 2001), while the Hellenic subduction in the south is responsible for the extensional deformation of the Aegean microplate due to the subducting slab’s rollback (Brun et al. 2016). In the vicinity of the Samos Island, the geodetic strain-rate field has a second invariant of 20–30 nstrain/yr (Kreemer 2009), mostly exhibiting elongation in a N-S to SSW-NNE direction, but also including a minor E-W shortening component and counter-clockwise rotation (Floyd et al. 2010). The stress-field that is derived from the inversion of focal mechanisms of crustal earthquakes indicates that the north Aegean is characterized by uniaxial extension in a SSW-NNE direction, while the maximum horizontal stress (SHmax) is mainly directed WNW-ESE (Kapetanidis and Kassaras 2019).

Given their geometry and kinematics (Fig. 1), the mapped active faults on Samos Island are capable of generating earthquakes with moment magnitude of the order of 6.3–6.8 (Chatzipetros et al. 2013). Historically, the broader region has been struck by strong earthquakes. In particular, several significant events of estimated Mw > 5 have reportedly occurred at the eastern part of the island in the eighteenth and nineteenth century (Fig. 1), with the more notable ones being the Mw = 6.7 event of 1751/06/18, the Mw = 6.2 event of 1873/02/01 and the Mw = 6.2 event of 1877/10/13 (Kouskouna and Sakkas 2013; Stucchi et al. 2013). Strong earthquakes near Samos Island have also been reported around 201–197 BC and 46–47 AD (Ambraseys 2015). During the instrumental era, the more recent significant earthquakes have been the Mw = 6.0 event of 1904/08/11, at the south-eastern part of the island, and the Mw = 6.7 event of 1955/07/16 near the Turkish coast ESE of the Samos Island (Makropoulos et al. 2012).

Fig. 1
figure 1

Seismotectonic map of the broader Samos area. Seismicity (circles and stars, with size proportional to magnitude) are from SHEEC (years 1000–1899, gray; Stucchi et al. 2013), from the compilation of the Seismotectonic Atlas of Greece (1900—June 2020, M ≥ 4, blue; Kassaras et al. 2020) and from a combined catalogue of routine analysis of seismicity at SL-NKUA and GI-NOA for the 2020 Samos aftershock sequence (2020/10/30–2020/11/30, orange). Local seismographs and accelerographs (triangles) are also shown. Faults after Ganas et al. (2013). The beachball represents the focal mechanism of the 2020 Samos mainshock. The inset map shows the location of the study area (red rectangle) in the context of major tectonic structures, such as the subduction zone (Hellenic Arc) and the North Anatolian Fault (NAF). The location of Lesvos (L) and Kos (K) islands are also marked on the inset map

Full size image

On 2020/10/30 11:51:28 UTC, a Mw = 6.9 earthquake occurred approximately 10 km offshore Samos Island (Ganas et al. 2020; Papadimitriou et al. 2020), generating the largest tsunami in the Aegean Sea since 1956 (Triantafyllou et al. 2021). The earthquake produced an aftershock sequence that extends 60 km in an E-W direction, including several events of moderate magnitude, the largest of which has been a Mw = 5.0 event, hours after the mainshock (Papadimitriou et al. 2020). The sequence was recorded by permanent seismological and accelerometric stations of the regional Hellenic Unified Seismological Network (HUSN; Evangelidis et al. 2021) in Greece, along with stations of the Boğaziçi University Kandilli Observatory and Earthquake Research Institute—Regional Earthquake-Tsunami Monitoring Center’s (KOERI-RETMC) network in Turkey. To monitor the sequence, the Geodynamic Institute of the National Observatory of Athens (GI-NOA) installed two additional temporary stations on the island (SAM1 and SAM2; Fig. 1). In this study we take advantage of the recent sequence to study the anisotropic properties of the crust in the immediate area of the mainshock.

Data and methodology

The recent seismic sequence of the Mw = 6.9 mainshock provided a substantial amount of recorded data. However, since this dataset is temporally constrained (between 2020/10/30 and 2020/11/30), we retrieved further events and waveforms from the previous five years (i.e., since 2015/01/01), to establish a background of anisotropy. Event and phase arrival information were acquired from the European and Mediterranean Seismological Centre (EMSC) for the period between 2015/01/01 and 2018/07/31, the public catalogue of the Seismological Laboratory of the National and Kapodistrian University of Athens (SL-NKUA) between 2018/06/01 and 2020/10/29 and a seismic catalogue of the first month of the 2020 Samos aftershock sequence, compiled from the bulletins of SL-NKUA and GI-NOA, part of which was presented in Papadimitriou et al. (2020). Our final catalogue consisted of 1,376 seismic events in the vicinity of the island. Seismicity in the area is generally constrained in the crust, with the majority of the foci located shallower than 15 km. Four stations that were operational before the mainshock (GMLD, KRL1, KUSD and SMG) and two temporary stations installed after the event (SAM1 and SAM2) were available. As expected, almost half of the considered events (647) were part of the 2020 sequence.

It is noted that we used recordings by both seismographs (SAM1, SAM2 and SMG) and accelerographs (GMLD, KRL1 and KUSD). Thus, raw recordings were first very broadly filtered between 0.005 and 50.0 Hz, the instrument response was removed and, in the cases of accelerographs, were integrated to velocity. Preprocessing was carried out using the ObsPy package (Krischer et al. 2015). All waveform data were acquired through the international Federation of Digital Seismograph Networks’ (FDSN) service from the European Integrated Data Archive (EIDA) nodes at KOERI-RETMC (https://www.orfeus-eu.org/data/eida/nodes/KOERI/) for stations GMLD and KUSD and GI-NOA (http://eida.gein.noa.gr/webdc3/) for KRL1, SAM1, SAM2 and SMG.

To eliminate unsuitable arrivals at candidate stations, the “shear-wave window” criterion (Evans 1984) was enforced. The interaction of shear-waves with the free surface can generate secondary arrivals (e.g., converted pS phases) which need to be avoided during analysis. As such, a maximum incidence angle of the seismic ray is used to exclude such incidents. This threshold (i.e., the “shear-wave window”) was set to 45.0°, per previous studies (e.g., Gao et al. 2011; Kaviris et al. 2017). The incidence angle for each candidate ray was estimated during the analysis with the TauP algorithm (Crotwell et al. 1999). This necessary step, to ensure the quality of the final dataset, severely reduced the number of available data to 262 event-station pairs. However, it accurately represents the scarcity of events in the close vicinity of the stations (Fig. 1). Moreover, low Signal-to-Noise Ratio (SNR) recordings were also rejected. The SNR limit was set to 1.5, after manually reviewing waveforms with the splitting software described in the next paragraph.

The analysis for shear-wave splitting was carried out with the Pytheas software (Spingos et al. 2020). The processing algorithm of choice was the Eigenvalue method (Silver and Chan 1991) with the automation introduced by Teanby et al. (2004). These methods are frequently used in SwS analysis and have been integrated in multiple software tools in the past (Illsley-Kemp et al. 2017; Johnson and Savage 2012; Nolte et al. 2017). To improve on the quality of the measurements, two additional algorithms were used through Pytheas. First, for each waveform set, a bandpass filter had to be applied to remove the noise content. The boundaries of the filter were automatically selected with the process described in Savage et al. (2010). Second, a quality grade had to be assigned to measurements, to identify spurious and ambiguous cases. The grading algorithm (Spingos et al. 2020) uses a combination of three criteria; (a) the error of φ, (b) the error of td and (c) the correlation coefficient between shear-waves in the two horizontal components, after the removal of the anisotropy’s effect. Given grades range from “A” (best) to “E” (worst). Furthermore, in cases where φ was either (sub)parallel or (sub)perpendicular to p, a special grade (“N”) denoting null (Wüstefeld and Bokelmann 2007) measurements was provided by the algorithm. The overall processing scheme is detailed in Appendix 1.

Results

In total, 262 splitting measurements were produced. However, after considering observations of sufficient quality (only those graded “C” or better) and excluding null measurements, we obtained a final catalogue of 164 anisotropy results. Out of the six initially considered stations, two of them yielded no results of satisfying quality (i.e., KUSD and SAM1). A summary of the average splitting parameters for the stations with reliable results (i.e., GMLD, KRL1, SAM2 and SMG) is presented below (Table 1). Since φ observations are directional, we used circular statistics (Berens 2009) to account for wrapping of polarization directions.

Table 1 Summary of statistics for the shear-wave splitting analysis
Full size table

Station SMG displays a clear WNW-ESE polarization direction of the Sfast, perpendicular to the ones observed in the other three stations (Fig. 2a). KRL1 has a dominant NE-SW average φ, which is intersected by a secondary perpendicular direction. However, in contrast to SMG, φ observations at KRL1 are more scattered. GMLD has a very small count of observations (4), while SAM2 has the second smallest number of results (15). In general, as errors in Table 1 suggest, the behavior of φ is better constrained in KRL1 and SMG, i.e., in the stations where the vast majority (88.4%) of the SwS results was obtained. Average time-delays generally deviate around the 90.0 ms mark, with the exception of SAM2. Normalized time-delays, i.e., time-delays normalized per the hypocentral distance, follow a similar pattern Table 1, being around 6.0 ms/km in GMLD, KRL1 and SMG. Errors for the latter were calculated using the formulation proposed by Del Pezzo et al. (2004), to integrate uncertainties of the source-receiver distance estimation. Regarding the azimuthal distribution, as the majority of the analyzed events belong to the 2020 sequence, results are mostly constrained to the N-NE for the three stations on the island. Only SMG offers some insight, where odd measurements of φ are present all around the station, suggesting no correlation of deviating polarization directions with specific ray paths (Fig. 2b).

Fig. 2
figure 2

a Rose diagrams displaying the distribution of φ for each station. N is the total number of observations and F the count of measurements per grid line. b Equal-area projections of individual splitting observations per station (triangle at the center). The circle’s radius indicates an angle of incidence equal to the shear-wave window (45°). The length of each vector is proportionate to the respective td

Full size image

The processing of the data yielded a total of 27 null observations, among three of the four stations (KRL1, SAM2, SMG), with the majority observed at KRL1 (21 measurements). Polarization directions of null measurements follow a general NW–SE trend, which does not agree with either the regional stress regime or the local tectonic structures. Moreover, nulls at KRL1 are perpendicular to Sfast, indicating some coherence with expected Sslow polarization directions. We cannot comment on the other two stations due to the very small number of observations in either SAM2 (2) or SMG (4). Summary plots of the null results are presented in Appendix 2.

Discussion

The analysis revealed, for the first time, that the upper crust at the Samos Island is anisotropic. However, attributing the origin of anisotropy to either stress or structures is paramount for understanding the mechanics behind the phenomenon and determining the implications of the measured splitting parameters. In the following, we attempt to distinguish between the two possible causes of anisotropy, constrained by the limited available data.

Source of anisotropy at Samos Island

As mentioned above, shear-wave splitting is a phenomenon commonly associated with crustal microcracks (Crampin 1994; Crampin et al. 1980; Crampin and Peacock 2005; Margheriti et al. 2006). This association is generally supported by observed polarization directions of the Sfast being approximately parallel to the regional maximum horizontal stress component SHmax. Evidence of a microcrack-filled upper crust have significant connotations; if we can reliably sample changes in microcrack features from shear-wave splitting, we can associate them with an impending earthquake (Crampin et al. 1999, 2015). However, SwS phenomena can also originate from local tectonic structures. This case can be identified by φ directions parallel to the axes of rock characteristics. Structurally-controlled anisotropy has been observed globally in various environments (Cochran et al. 2003, 2006; Graham et al. 2020; Hiramatsu et al. 2010; Shi et al. 2020; Zinke and Zoback 2000). Differentiating between a microcrack-dominant and a structurally-controlled anisotropy state can be an arduous task; if the SHmax is parallel to the fault lines (as in the case of tectonic environments with pure normal dip-slip focal mechanisms) it is almost impossible to make such a distinction (Pastori et al. 2019).

The regional deformation characteristics compose a trans-tensional stress regime in the eastern Aegean Sea. The dominant active structures in the region of Samos Island, such as the Samos Basin fault (e.g., Ganas et al. 2013), just offshore to the north, are E-W trending normal faults, shaping the main morphology of Samos Island, the basins, but also the mountains at the opposite coast of Turkey. However, strike-slip faulting is also present in the region, evident from the focal mechanisms of recent seismicity (Benetatos et al. 2006; Tan et al. 2014; Yolsal-Çevikbilen et al. 2014). Along the north-western coast of the island, the NE-SW trending, north-dipping Carlovasi fault exhibits dextral strike-slip faulting, whereas the roughly WNW-ESE-trending, south-dipping Pythagorion fault, further to the SE, is characterized by normal kinematics (Chatzipetros et al. 2013). The NE-SW faulting trend is further extended westwards, emphasized on a regional scale by the 250 km-long Samos Basin normal fault feature, spanning from western Turkey to the middle of Central Aegean Sea (Ganas et al. 2013; Kreemer and Chamot-Rooke 2004). Samos Island also hosts a stack of several nappes and thrusts, the latter mainly striking NW–SE to N-S (Ring et al. 2007), with a strong presence of metamorphic rocks (Stamatakis et al. 2009). The nearby coast of Turkey is dominated by complex structures (east of the island) and NNE-SSW trending faults, north of Samos (Ocakoǧlu et al. 2005).

Our results (Fig. 3) seem to follow the strike of faults in the area. KRL1 and SAM2, located to the NW, present similar orientations to the Carlovasi fault. The average φ value in SMG is parallel to the Pythagorion fault’s strike. GMLD presents a different polarization direction than the strike in Seferihisar and Tuzla faults, but the small number of observations in the station cannot offer a reliable picture of splitting there. According to the above, two stations (KRL1 and SAM2) present a φ value that is very different from the orientation of the SHmax, which offers strong evidence for a state of structurally-controlled anisotropy in Samos. Additionally, co-seismic deformation measured at nearby GNSS stations showed similarly oriented horizontal ground displacement vectors, with a direction of approximately N190°E (V. Sakkas, personal communication). The mean anisotropy direction in station SMG is also parallel to the local fault line, but parallel to SHmax as well, which does not allow for a clear interpretation.

Fig. 3
figure 3

Average Sfast polarization directions (red lines) for each station (triangles). The length of each vector is proportionate to the average time-delay at the station. Red circles show events with at least one shear-wave splitting result, whereas gray circles represent events with all-rejected measurements. Cyan arrows indicate the orientation of SHmax, after Kapetanidis and Kassaras (2019). The focal mechanism of the 2020 Mw = 6.9 event is also shown. Rest of notation as in Fig. 1

Full size image

Temporal evolution of splitting parameters

To identify whether changes of polarization directions or time-delays with time exist in our case, we explored variations of the estimated splitting parameters, i.e., φ and normalized time-delays (tn), for the two stations that offered an adequate number of observations, i.e., KRL1 and SMG. Temporal variations are a common point of discussion in shear-wave splitting studies (e.g., Piccinini et al. 2006). According to literature, rays that travel within a solid angle between 15° and 45° to the microcrack plane are more sensitive to stress variations; these rays are classified in “Band-1”, while the ones that propagate through a solid angle narrower than 15° belong to “Band-2” (Crampin et al. 1999). In the following, we present the five-year temporal evolution of the splitting parameters in all three categories (“all bands”, “Band-1” and “Band-2”) for each station.

Polarization directions in KRL1 (Fig. 4a, c) are generally affected by the extent of their scattering (Fig. 2a). In the five-year interval between 2015 and 2020 (Fig. 4a), the observed changes in φ are not correlated with the occurrence of a strong seismic event. A few outliers, which exist between N120°E and N180°E, appear only circumstantially. Abrupt changes by 90° in φ (the so-called 90°-flips) have been attributed to increased pore pressure in rocks and have been observed primarily in volcanic areas due to rays passing either through or near magmatic bodies, such as dykes (e.g., Johnson et al. 2011; Unglert et al. 2011). High pressure of pore-fluids which is not related to volcanism can also result in this phenomenon; increased pore pressure on the fault causes flips to microcracks adjacent to it, while the polarization direction recorded at the station is affected gradually by the propagation medium away from the source (Crampin et al. 2002). Consequently, as the resulting φ is affected by different causes, its per station distribution is rendered complicated (Gao et al. 2011, 2019; Crampin and Gao 2014). Such a phenomenon could be the cause of the significant scattering we observe in KRL1, especially during the aftershock sequence (Fig. 4c). KRL1 is located almost next to the causative fault (Papadimitriou et al. 2020) of the 2020 event (Samos basin f., Fig. 3) and, consequently, the observation of flipped φ due to high pore pressure on the fault is more likely. SMG showcases a temporal distribution of φ generally constrained around its average value (Fig. 4b), before 2020, with the occasional outlier. On the other hand, observations related to the aftershock sequence (Fig. 4d) indicate an initially flipped φ until 2020/11/03 which then reverts to its prior state. A possible explanation is the release of tectonic stress due to the mainshock. As time progresses and the release of seismic energy is expressed through aftershocks, the stress is reduced. Thus, the zone (and as such, the microcracks) affected by the pressure buildup on the fault is reduced. As SMG is located further away from the aftershocks zone, seismic rays are travelling at a greater distance away from the source (as opposed to KRL1) and are affected more by the intermediate propagation medium.

Fig. 4
figure 4

Variations of φ between 2015 and 2020 (a, b) and for the aftershock sequence between 2020/10/30 and 2020/11/30 (c, d) for stations KRL1 (a, c) and SMG (b, d). In each panel, observations classified as “all-bands” (top), “band-1” (middle) and “band-2” (bottom) are presented, respectively. Each data point is represented by a circle with its corresponding error. The bold red dashed line indicates the three-point moving average. The horizontal black solid line denotes the average φ, as exhibited in (Table 1. The vertical green dotted lines note the origin times of earthquakes with M ≥ 5.5 in epicentral distances of approximately up to 150 km. The shaded area in (b) is detailed in Fig. 6b

Full size image

Concerning the variations of tn, a decrease is observed in KRL1 before 2016/04, which follows a suggested period of increasing time-delays (Fig. 5a). Examining measurements that belong exclusively to “Band-1” pronounces this change. Concerning the recent Mw = 6.9 event, the few data points between 2016/12 and 2020/10 do not permit the extrapolation of solid conclusions. An increase is observed in the running mean after 2019/08, but it is strongly biased by the scattered measurements associated with the aftershock sequence and cannot be considered indicative of any physical process. Nevertheless, it is noteworthy that individual tn values are less than 20 ms/km, in the five-year period preceding the mainshock. In SMG station (Fig. 5b), changes are similarly ambiguous. Examining all data, a clear increase and decrease can be observed between 2016/08 and 2017/04. However, this variation is much weaker in “Band-1”; this is a result of the individual high (over 20 ms/km) tn being classified to “Band-2”. A mean increase, affected by the scattering of time-delays of the aftershock sequence, is then observed. However, in this case, time-delay values are generally constrained below 20 ms/km, even for measurements from the 2020 sequence events. In contrast, KRL1, which is closer to the epicenter of the mainshock, exhibited a significant number of high tn values in the same period.

Fig. 5
figure 5

Variations of normalized time-delays between 2015 and 2020 for stations KRL1 (a) and SMG (b). The shaded area in “band-1” of panel (b) denotes the time period shown in Fig. 6. Other notation as in Fig. 4

Full size image

At first glance, the number of observations in either case is not adequate to draw reliable conclusions. We searched the seismic catalogue of NKUA-SL for strong earthquakes (Mw ≥ 5.5) to find possibly associated events, in a radius of 150 km from Samos; shorter distances have been reported in the literature, for weaker events (e.g., Gao and Crampin 2006). Other than the recent 2020 Mw = 6.9 event, there are two large shocks in the broader area; (a) the 2017/06 Lesvos earthquake of Mw = 6.3 (Kiratzi, 2018; Papadimitriou et al. 2018) at an approximate distance of 130 km and (b) the Kos Mw = 6.6 event, which occurred a month later on 2017/07 (Ganas et al. 2019), 100 km away from the island (see Fig. 1 for the locations of Lesvos and Kos islands). Variations in KRL1 (Fig. 5a) seem to be unrelated to the above events. The increase and drop of time-delays is recorded over half a year before. However, there is evidence for a connection between the occurrence of the 2017 Aegean earthquakes and variations in SMG (Fig. 6a). The observation of a tn increase period (Ti), indicating stress accumulation, starting from 2015/11 and lasting for 556 days is documented, with a regression correlation coefficient (R) equal to 0.93. It is followed by a decrease period (Tr), suggesting stress relaxation (Gao and Crampin 2004) of 90 days, with a worse R = 0.41. The time periods selected for the regression analysis were obtained from the proposed relations of Crampin et al. (2013), for a Mw = 6.3 event. We do not have enough data to investigate changes associated with a Mw = 6.6 earthquake, which would correspond to Ti = 766 days and Tr = 164 days. If this is the case, then SMG might indeed sample an anisotropic space where stress-sensitive microcracks are prevalent. Temporal changes of φ at SMG seem to be concentrated before the 2017/06 event (Fig. 6b). A group of three ~ N60°E is followed by two ~ N140°E measurements. However, we do not believe these can be associated with the event, which is located 130 km away; rays analyzed for anisotropy are not affected by the distant fault (and its adjacent microcracks of varying orientation) at Lesvos. Further work is required to comprehend their relation to local tectonic processes.

Fig. 6
figure 6

Variations of tn (a) and φ (b) of “band-1” observations, between 2015/11 and 2017/09, for station SMG. In (b), the suggested stress accumulation (green dashed line) and relaxation (red dashed-dotted line) are shown. The linear models have been determined by regression analysis using the least squares method with tn data points exhibiting increase (green squares) or decrease (red triangles). A single high-tn point (tn > 12 ms) greatly affects the relaxation model, as highlighted by the much smaller correlation coefficient (R). Other notation as in Figs. 4 and 5

Full size image

Conclusions

The recent Mw = 6.9 Samos event and its seismic sequence was a unique opportunity to study the anisotropic properties of the upper crust in the area, due to the provision of a high number of earthquakes located close to local seismographs and accelerographs. Automatic analysis of data from the island and its neighboring mainland offered new perspectives on the state of crustal anisotropy. Average polarization directions of the Sfast are generally aligned according to local faults, irrespective of the regional SHmax. This is strong evidence supporting a mechanism of anisotropy dominated by local structures. However, measurements at station SMG, located in the southern part of the island, where the local Pythagorion fault is oriented according to the SHmax, perplex the situation, as it is not possible to distinguish between the two anisotropy sources, from φ alone.

Changes of φ were observed during the aftershock sequence, possibly due to the pressure buildup near the causative fault, with the scattering of φ being more pronounced in the station closer to the fault, i.e., KRL1. Variations of normalized time-delays exhibited some evidence of association with seismic events at larger distances. Namely, an Mw = 6.3 which occurred at 130 km away from Samos on 2017/06, followed by an Mw = 6.6 on 2017/07, at a distance of 100 km. These changes, related to stress accumulation and relaxation, are observed in SMG. If we accept that the variations of time-delays are indeed a prelude to the 2017/06 event, which is statistically (although not robustly) supported as shown in Fig. 6, then this station seems to sample a stress-controlled anisotropic space and its average φ is aligned according to SHmax. However, this behavior is not well-documented, as the associated data points are few. The Samos event would be a great opportunity to support this, since the epicenter is located in the vicinity of the station, but there were not enough data before its occurrence. Future work on the area should focus on the period preceding this earthquake, by conducting a more detailed microseismic analysis to enrich the catalogue of suitable events. In any case, the absence of significant time-delay temporal variations in KRL1 and its existence in SMG, along with their difference in polarization direction, might suggest a structurally-controlled origin of anisotropy to the NW and a dominant effect of stress to the SE.

As the area of the Eastern Aegean has shown from the activity in the past two years, a densification of the local networks would greatly enhance our ability to study seismic anisotropy in the region, identify the mechanism of SwS and recognize potential time-delay variations as precursors.