Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey

Crustal Anisotropy in the Eastern Sea of Marmara Region

in Northwestern Turkey

by Tuna Eken, Marco Bohnhoff, Fatih Bulut, Birsen Can, and Mustafa Aktar

Abstract The North Anatolia Fault Zone (NAFZ) is a transform zone 1600 km inlength representing the plate boundary between the westward moving Anatolian Plateand stable Eurasia. Almost the entire fault zone has failed during the last centuryexcept for the Sea of Marmara section, which is located in direct vicinity to the cityof Istanbul. In this study, we investigate the crustal anisotropy along the eastern Mar-mara section of the NAFZ based on shear-wave splitting. We measure seismicanisotropy parameters, namely, the fast polarization direction (PD) and time delay(TD), by analyzing local seismicity recorded at selected seismographs operatedthroughout the eastern Sea of Marmara region. Our shear-wave splitting (SWS)observations indicate a predominant northwest–southeast-oriented PD, which is sub-parallel to both the orientation of the regional SHmax in northwest Turkey and the localNAFZ strike along the Princes’ Islands segment. Toward the south, at the ArmutluPeninsula, we find a different PD pattern reflecting local fault strikes, SHmax as wellas strain asymmetry between different crustal blocks across the fault zone. Applyingstrict quality criteria enables us to identify robust, preferred fast PDs, which suggeststhat initially observed PD heterogeneities are sometimes caused by second-order ef-fects in the data rather than by varying PDs. Comparing TD and hypocentral depthdistribution, we find the depth extent of the anisotropy is confined to the uppermost10-km depth of crust. We combine our SWS results with those from previous studiesconducted along the San Andreas fault (SAF) and NAFZ in order to investigate therelation of angular deviations of the PDs from regional SHmax and local fault strikeswith fault-zone distance. We find that fast PDs are mainly controlled by the local faultstructure in close proximity to a fault zone (5 and 10 km) while they are controlled bycrustal stress at off-fault locations (5 and 10 km).

Introduction

Shear waves split into approximately orthogonal fastand slow directions within an anisotropic medium. Themechanisms for this phenomenon can be divided into twomajor groups (Boness and Zoback, 2006): stress-inducedand structure-induced. Stress-induced anisotropy is con-trolled by extensive dilatancy of fluid-filled microcracks(Crampin, 1987) or the preferential closure of fractures bythe in situ stress field (Boness and Zoback, 2006) and causesfast polarization directions (PD) of vertically propagatingshear waves to be parallel to the maximum horizontal stress(SHmax). Structure-induced anisotropy is related to the align-ment of macroscopic features near active faults (Zhang andSchwartz, 1994; Zinke and Zoback, 2000; Tadokoro et al.,2002), sedimentary bedding planes (Kern and Wenk, 1990),and preferred mineral alignments (Sayers, 1994), and causesfast PD of vertically propagating shear waves to be parallel tothe strike of the structural fabrics.

The 1600-km-long North Anatolian Fault Zone (NAFZ)is an intercontinental dextral strike-slip fault representing aboundary between the Eurasian plate in the north and theAnatolian plate in the south. It provides a natural laboratoryto investigate stress- and structure-induced anisotropy mech-anisms. Collision between the Arabian and Eurasian plates(in the east) as well as the southwest-trending rollback of theHellenic subduction zone in the south Aegean Sea (in thewest) are the main elements driving the westward movementof Anatolian plate (e.g., McKenzie, 1972; McClusky et al.,2000; Fig. 1). During the twentieth century, more than900 km of the NAFZ progressively failed as a series of west-ward propagating large earthquakes (e.g., Barka, 1999; Pinaret al., 2010).

Recent multichannel and wide-angle seismic profiles, aswell as seismological and geoelectric observations, havesignificantly contributed to our understanding of the seis-motectonic setting and crustal structure of the study area

911

Bulletin of the Seismological Society of America, Vol. 103, No. 2A, pp. 911–924, April 2013, doi: 10.1785/0120120156

(e.g., Carton et al., 2007; Bulut et al., 2009, 2011; Örgülü,2011; Kaya, 2012). Bulut et al. (2009) presented the spatio-temporal distribution of microseismic activity and focalmechanisms along the Princes’ Islands fault segment belowthe eastern Sea of Marmara to investigate the kinematics ofthe down-dip extension of fault structures imaged by Cartonet al. (2007) (Fig. 1b). Magnetotelluric measurements sug-gest that the fault zone itself represents a clear boundary be-tween conductive and resistant crustal blocks in the Sea ofMarmara region (Kaya, 2012). Recent seismological studieshave shown that a dextral strike-slip faulting regime ispredominantly hosted along the major strands of the NAFZ

bounding the Çınarcık Basin (Fig. 2; Bulut et al., 2009;Örgülü, 2011). Similarly, fault plane solutions of Izmit after-shocks indicate a predominant strike-slip motion along thePrinces’ Islands segment (Örgülü and Aktar, 2001; Karabulutet al., 2002; Pinar et al., 2003). Furthermore, the dominant

deformation pattern changes westward toward the bendingpoint of the NAFZ offshore of Istanbul where a substantialthrust component is present in the mechanisms (Bulut et al.,2009; Fig. 2). This change in deformation regime suggests anadded component of fault normal compression along theeast–west-trending NAFZ segment as opposed to the purestrike-slip motion observed along the northwest–southeast-trending Princes’ Island segment. Normal faulting activitycurrently plays a minor role in this region but is predomi-nantly observed further to the south below the ArmutluPeninsula (Bohnhoff et al., 2006; Örgülü, 2011).

Shear-wave splitting (SWS) measurements can provideadditional constraint on stress orientations for a specific re-gion, especially in cases of insufficient stress field measure-ments. Stress inversions of focal mechanisms from largeearthquakes in northwest Turkey suggest that maximumhorizontal stress is oriented ∼35° (N125°E) clockwise withrespect to the east–west-trending NAFZ (Kiratzi, 2002;Bohnhoff et al., 2006). This was also observed from theinversion of focal mechanisms in the Sea of Marmara region(Örgülü, 2011). Stress measurements from the World StressMap (Heidbach et al., 2008) indicate that SHmax generallystrikes northwest–southeast in the area of investigation rang-ing betweenN120°E andN160°E. Several previous SWS stud-ies along major strike-slip fault zones (e.g., Zinke andZoback, 2000; Peng and Ben-Zion, 2004; Boness andZoback, 2006; Liu et al., 2008; Hurd and Bohnhoff, 2012)have shown that the anisotropic properties within the faultzone and surrounding crust are influenced by both structure-and stress-induced mechanisms depending on the stationlocation with respect to the fault zone. A recent systematicanalysis of SWS performed at the San Andreas fault (SAF)shows that observed PDs tend to be nearly parallel to both themain strike of the SAF, as well as to the direction of SHmax (Liuet al., 2008). Boness and Zoback (2006) and Liu et al. (2008)reported that the use of seismic data from dense seismic

(a)

(b) (c)°

°

°

°

° ° °

20°

40°

30°

30° 40° 50°

Figure 1. (a) Tectonic map of Turkey and adjacent regions in-cluding the North Anatolian Fault Zone (NAFZ). The major plateboundaries are after Bird (2003). The white rectangle indicatesthe study area enlarged in (b). (b) Spatial distribution of seismicstations throughout the study area (eastern Sea of Marmara). Blackstars, triangles, and diamonds represent stations that belong to thePIRES, ARNET, and KOERI networks, respectively. Stations outlinedby a dark gray circle were selected for the SWS analysis. Sivri andYassi represent the islands of Sivriada and Yassiada, respectively. Thefive stations in each island are enlarged in (c). (c) Subarrays of PIRESnetwork on Sivriada and Yassıada. The two subarrays of five stationseach (dark gray triangles) are located at a lateral distance of less than5 km to the surface trace of the main fault branch (Princes’ Islandssegment). Both subarrays include a cross-shaped distribution of sta-tions with an aperture of ∼300 m. The color version of this figure isavailable only in the electronic edition.

28° 30 29° 00 29° 30

40° 45

41° 00

0 5 10

kmM-scale

321

Sea of Marmara

Çınarcık Basin

ISTANBUL

Princes Islands

Armutlu Peninsula

Figure 2. Seismotectonic map of the eastern Sea of Marmara.Shown is microseismic activity throughout the Çınarcık Basin aslocated by seismic networks (Fig. 1b) between October 2006and October 2010. Fault traces are taken from Armijo et al.(2005). Black focal mechanisms indicate composite fault plane sol-utions for spatial earthquake clusters derived from first motionpolarities (Bulut et al., 2009). Gray focal mechanisms are fromthe regional moment tensor (RMT) inversion catalog publishedby Örgülü (2011). The color version of this figure is available onlyin the electronic edition.

912 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar

networks, precise earthquake locations, and pre-existingknowledge regarding tectonic structures are crucial elementsin discriminating between structure- and stress-induced aniso-tropy mechanisms. Crustal anisotropy studies along the 1999Izmit rupture show a spatial correlation between the fast PDsand station distance from the fault (Tadokoro et al., 2002;Peng and Ben-Zion, 2004). However, some local variationsof the fast PDs observed at a number of on- and off-faultstations in these studies could not be fully explained bystation–fault distances. Recent results from SWS analysis ofIzmit aftershocks indicate the presence of both stress- andstructure-induced mechanisms causing seismic anisotropy(Hurd and Bohnhoff, 2012). Along the western segments ofthe NAFZ beneath the Sea of Marmara, the complex geologicand tectonic environment possibly results in spatial variationof fast PDs.

In the present study, we investigate seismic anisotropythroughout the eastern region of the Sea of Marmara. Weperform an SWS approach on direct shear-wave observationsfrom microseismic activity recorded throughout the studyarea. Our primary focus is to delineate the spatial distributionof shear-wave anisotropy. Discrimination between structure-and stress-induced crustal anisotropy is difficult to determinebecause the local NAFZ strike along the Princes’ Islandssegment is consistent with the regional trend of SHmax forthis part of the study area. We combine our splitting meas-urement results with that of other SWS studies observedalong the NAFZ and SAF in order to make a case for a pos-sible discrimination between structure- and stress-inducedmechanisms in a general sense. Additionally, we applyincreasingly strict quality control criteria to exclude low-quality results and second-order effects from further interpre-tation. The limited amount of data during the observationtime period considered does not allow us to look for potentialtemporal variations of SWS parameters.

Data

A permanent seismic array, the Princes’ Islands Real-time Earthquake monitoring System (PIRES), was installed in2006 on the two outermost Princes’ Islands, Sivriada andYassıada, in order to monitor the microseismic activity onthe main branch of the NAFZ offshore Istanbul, the Princes’Islands fault segment (Bulut et al., 2009, 2011; Fig. 1b).PIRES contains two subarrays of five stations on each islandthat are located at a lateral distance of less than 5 km to themain fault branch. Both subarrays are arranged in a cross-shaped distribution of stations with an aperture of ∼300 m(Fig. 1c). The average station spacing within each PIRES sub-array is ∼190 m. Recently, PIRES has been enlarged towardthe other Princes’ Islands in order to improve azimuthal cov-erage over the entire Princes’ Islands segment of the NAFZ.For this study, PIRES recordings are combined with data fromlocal permanent land stations of the Turkish KandilliObservatory and Earthquake Research Institute (KOERI,personal comm., 2010) and the Armutlu seismic network

(ARNET; Tunç et al., 2011) on the Armutlu Peninsula inorder to obtain the best available azimuthal coverage for seis-micity within the target area (Fig. 1b). We obtain a well-resolved hypocenter catalog of microseismicity allowingus to identify the seismically active structures and their rolein the local seismotectonic setting. The seismic data usedin this study spans the time period between 2006 and 2009.A total of 2694 shear-wave recordings were analyzed using773 out of a total of 913 earthquakes that have been locatedwith an average precision of ∼2 km (Fig. 2). The samplingrates vary between 100 Hz (9 stations) and 200 Hz (15 sta-tions). The average station spacing is ∼25 km for the com-bined network.

Data Processing and Shear-Wave Splitting Analysis

Prior to SWS analysis, we employed two main criteriafor event preselection as follows: (1) For each station, weonly consider earthquakes with ray paths arriving at astraight-line incidence angle of less than 45° from the verticalas suggested by Nuttli (1961). This is to eliminate free-surface effects that can result in an inhomogeneous P wavewith an associated phase change on the radial and verticalcomponents when S waves approach at incidence angleslarger than the critical angle (Paulsen, 2004; Peng andBen-Zion, 2004; Hurd and Bohnhoff, 2012). (2) We usedonly waveforms for which the signal-to-noise ratio (SNR) ex-ceeds 6 and 2.5 for P and S waves, respectively. After eventpreselection, a total of nine stations (ARM, BRG, BYZ,EAS, ISK, PIE, SCR, TSV, and YLV; Fig. 1b) containedmore than 50 events and thus were selected for SWS analysis.Stations with a smaller number of events are excluded inorder to achieve only statistically significant results. The pre-selection procedure results in a total of 1638 waveformsrecorded from 773 microseismic events.

Our data analyses procedure includes both visual andautomated methods. Considering the difficulty in analyzinglarge amounts of datasets, the use of automated processing isadvantageous because it provides an efficient method of con-ducting SWS analyses in an unbiased and systematic manner(Liu et al., 2008). Three main approaches are commonlyused for automated SWS analyses: the cross-correlation (CC)method (e.g., Fukao, 1984), the covariance matrix (CM)method (e.g., Silver and Chan, 1991), and the aspect ratio(AR) method (Shih et al., 1989). Essentially, these methodsperform a search for maximum similarities between fastshear-wave (FSW) and slow shear-wave (SSW) componentsand the time delay (TD) between them. Theoretically, thesemethods are considered to result in identical anisotropicparameters because they are all developed based on eigen-value decomposition in measuring the linearity of particlemotion (e.g., Silver and Chan, 1991). However, results showa variation when the three methods are applied separately onthe same recordings (Liu et al., 2008). Liu et al. (2008) statedthat the autoregressive (AR) method can be successfullyapplied for PD estimation in case of adequate SNR and large

Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey 913

TD of SSW (i.e., >0:1 s) irrespective of similarity betweenfast and slow components. Considering observed small TDbetween slow and fast components and assuming that the Sphases usually are contaminated with a significant coda ofthe P waves, we decided to employ the CC method in ouranalysis. Additionally, we perform a visual check (e.g., hori-zontal particle motion analysis) in order to confirm the qual-ity of splitting results and exclude inadequate results.

The first step of the CC approach is to rotate horizontalseismogram components into radial and transverse directionssince most of the S-wave energy appears on these compo-nents for high angles of incidence. We then performed acoherency analysis to investigate consistently recorded sig-nals through the closely spaced stations of the PIRES array(Fig. 3a). The results of this analysis indicate that the fre-

quency content is predominantly coherent for the frequencyrange 1–20 Hz. We applied a third-order band-pass Butter-worth filter for this frequency range. The length of the timewindow starts from 0.07 s prior to the S-wave onset andmoves in increments of 0.02 s; eventually, 10 steps altogetheris fixed to 0.4 s to frame the S wave. At each step of themoving window, radial and transverse components arerotated from 0° (west) to 180° (east) in increments of 1° andshifted for delay times ranging from 0 to 0.3 s within incre-ments of sampling rate of respective station (i.e., 0.005 s forPIRES and 0.01 s for KOERI and ARNET stations). Eachazimuth and TD pair results in a cross-correlation coefficient(CCC) which gives a measure of similarity between twoorthogonal seismograms (0, no similarity; 1, identical). Thefinal PD and TD were adopted from the largest CCC value of

0 20 40 60 80 100 120 140 160 180

0.05

0.1

0.15

0.2

0.25

TD

(s)

PD (deg)

Frequency (Hz)

Co

he

ren

ce

pie (SV)

scr (SV)

(b)(a)

(d)

(c)

(e)

N-S

E-W

-3 -2 -1 0 1 2 3

x 104

-3

-2

-1

0

1

2

3x 10

4

x 104

PM

(f)

E-W

N-S

Time (s)

Time (s)

Slow

Fa

st

Time (s)

0.2 -0.2

0.2

0.2

0-0.2

0

-0.4

-0.4

-0.2

00.20.4

0.60.8

0.40.20-0.2

-0.4-0.4

-0.200

0.2

-0.6-0.4

-0.2 0

Figure 3. (a) An example of the coherency analysis between radial (SV) components recorded at stations PIE and SCR. (b)–(f) show anexample of the SWS analysis using the cross-correlation (CC) approach. (b) Particle motions of the horizontal component seismograms beforecorrection. (c) Overlaid plot of horizontal component. The solid gray rectangle indicates the length of the time window of analyzed signal.(d) Particle motions of seismograms rotated to the estimated fast and slow PDs. (e) Overlaid plots of seismograms rotated to the estimated fastand slow PDs. The solid gray rectangle indicates the length of the time window of analyzed signal. (f) Misfit function and maximum CC valuemarked with a black cross corresponding to splitting parameters PD � N106°E and TD � 0:09 s from the CC method (Fukao, 1984). Thecolor version of this figure is available only in the electronic edition.

914 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar

the 10 moving windows. An example of the SWS evaluationprocedure is shown in Figure 3.

A quality check of the SWS measurements was per-formed using some of the criteria proposed by Peng andBen-Zion (2004) and Liu et al. (2008): (1) The CCC thresh-old value must exceed 0.75; (2) the SNR measured from theaverage amplitude of the SWS analysis window to that of theaverage of 0.2 s of time window prior S-wave arrival isgreater than 2.5; (3) the maximum difference between mea-sured fast PDs is less than 20° when the window size is variedby �0:02 s; (4) the maximum difference between measuredTD is less than 0.02 s when the window size is varied by�0:02 s; and (5) a visual check for linearity of particlemotions was also performed. The criteria (1) and (5) wereapplied to ensure sufficient similarity between FSW and SSW.However, a very high CCC threshold can cause the loss ofsome useful measurements when the fast and slow shearwaves are not sufficiently similar. In this study, we fixed thethresholds as compatible with previous SWS studies (Pengand Ben-Zion, 2004; Liu et al., 2008; Hurd and Bohnhoff,2012). Criterion (2) eliminates the results which might beartificially affected by contaminated P-wave coda. Criteria(3) and (4) ensure the robustness of the results which mustbe stable in case of varying length of analysis window. Atotal of 420 shear-wave recordings from 266 earthquakes re-main satisfactory for further interpretation after restrictingthe results with respect to our quality criteria.

Results

Fast Polarization Directions and Time Delays

At the northern part of the study area, the PIRES stationsBRG, BYZ, EAS, PIE, and SCR are located on the Princes’Islands at close proximity (3–10 km) to themain branch of theNAFZ (Fig. 1c). The islands are covered with a Devonian for-mation representing a relatively stiff and, therefore, conven-ient material for passive seismic monitoring. This side ofthe fault represents a rather uniform elevated crustal blockand is less influenced by tectonic deformation. Splays ofthe NAFZ cross the Princes’ Islands and thus might introducestructure-induced anisotropy into shear-wave splitting observa-tions. Station ISK is located at ∼30-km north of the main faultbranch. There are no known secondary faults nearby ISK.

At the southern part of the study area (Armutlu Penin-sula), the geological formation as well as the structural setupis much more complicated. In this region, we include threestations (ARM, TSV, and YLV) (Fig. 1b). Station ARM islocated on a Paleozoic-quartzite formation, which is dis-placed by a network of northwest–southeast- and northeast–southwest-oriented conjugate strike-slip faults. Station TSVis located on a basalt formation and just to the south of aneast–west-oriented, northward-dipping normal fault. StationYLV is located on a basalt formation, which is surrounded bynorthwest–southeast- and northeast–southwest-oriented con-jugate strike-slip faults (Eisenlohr, 1997).

The most common way of representing the spatial distri-bution of anisotropy is to employ rose diagram statistics.Figures 4 and 5 present such rose diagrams summarizingmeasured fast PDs for all nine stations. We analyze the con-sistency of themeasured fast PDs at each station quantitativelyby employing the Von Mises method (Cochran et al., 2003).We calculate themean fast PD (PDmean) andmean length of thefast PDs, r, which represents the measure of the scattering offast PDs at each station (Cochran et al., 2003; Liu et al., 2008).PDmean and r were calculated as follows:

A �X

N

i�1

sin�PD�; (1)

B �X

N

i�1

cos�PD�; (2)

PDmean � tan�A=B�−1; (3)

r �������������������

A2 � B2

p

=N; (4)

where PD and PDmean indicate individual and average PDmea-surements, respectively. The coefficient r represents the meanresultant length of PD azimuths and is equal to the inverseanalog of the variance. It ranges between 0 (uniformly scat-tered) and 1 (aligned).

Table 1 summarizes the calculated parameters as well asthe mean fast PD and the mean resultant length values of fastPDs for all stations. The six northern stations (with an aver-age of r ∼ 0:37) exhibit relatively higher r values withrespect to those obtained at three southern stations (with anaverage of r ∼ 0:24) implying a more consistent fast PD inthe north (northwest–southeast). The average fast PD for allstations is N110°E (Table 1). Rose diagrams for the individ-ual stations in the north indicate two very similar prominentfast PDs of ∼N135°E (for the stations BYZ, EAS, PIE, SCR,and ISK) and∼N150°E (for station BRG). These orientationsof the PDs are in good accordance with both the strike of thePrinces’ Islands segment of the NAFZ and the regional maxi-mum horizontal stress (Fig. 4a,b). The average split TDs ateach station range from ∼36 to 64 ms with a standard devi-ation of �22 to 48 ms. The average TD measured at the sixstations north of the fault is 46� 37 ms (Table 1).

The average resultant length (r) calculated for the threesouthern stations on the Armutlu Peninsula is comparativelysmaller (average ∼0:24, see Table 1) than those observed inthe north. Station ARM is located at the western tip of thepeninsula where two main fast polarization orientations of∼N150°E and N60°E are observed (Fig. 4b, Table 1). Thelocal fault trend is ∼N60°E and, thus, is in good accordancewith one of the two measured PDs while N150°E is in goodaccordance with SHmax. A comparison between the localstress regime and observed PDs is almost impossible due

Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey 915

to the very low resolution of local SHmax observations closeto station ARM (World Stress Map; Heidbach et al., 2008).At station TSV, located at the central north shore of thepeninsula, the main fast PDs is ∼N90–120°E. This stationindicates a secondary preferred orientation at ∼N150°E(Fig. 4b, Table 1). These orientations correlated well withlocal SHmax orientations estimated by Örgülü (2011). StationYLV is located at the center of the Armutlu Peninsula andreflects a relatively diffuse pattern (r value of ∼0:136, seeTable 1). The rose diagram for this station shows a predomi-nant orientation at ∼N13°E, which is more consistent withthe local distribution of SHmax (Örgülü, 2011) in the regionclose proximity to station YLV (Fig. 4c, Table 1). This pre-ferred orientation is not as prominent compared to other sta-tions and almost the entire range of azimuths from 0°–180° isrepresented in the results. The average TD for the southernstations (ARM, TSV, and YLV) is 56 ms and individual sta-tion averages range from ∼47 to 68 ms.

In summary, our observation is that the stations along thenorthern side of the Princes’ Island segment (BRG, BYZ,EAS, ISK, PIE, and SCR) indicate stable trends for the fastPDs, which range between N135°E and N150°E (Fig. 4a,b).The orientation of the PDs is clearly parallel both to the direc-tion of maximum horizontal stress SHmax, as well as to the

strike of the main fault. In contrast, the three stations locatedin the south on the Armutlu Peninsula do not show a com-mon preferred orientation, although the observed fast PDsdo partly correlate with local principal fault trends andSHmax (Fig. 4b,c). In Figure 5, the rose diagrams of all ninestations analyzed here are shown together with local seismic-ity, selected focal mechanisms, and maximum shorteningdirections.

Sensitivity to Quality-Control Criteria

As previously described, several quality criteria wereapplied on the waveform data prior to the SWS analysis toensure robust estimates of the anisotropy parameters. Tofurther analyze the impact of the selected quality criteria onthe robustness of the obtained anisotropy parameters, weincreased the quality thresholds in stepwise fashion. Figure 6summarizes this procedure showing the resulting rose dia-grams after four steps of increased quality criteria (four col-umns from left to right). The first column shows the rosediagrams of all fast PDs from all available waveforms withinthe database for each station. The second column shows therose diagrams obtained after selecting only results for eventswith an SNR of the shear-wave recordings larger than 2.5 and

N=61

N=45

PIE

N=45

SCR

0 30 60 90 120 150 180

Fast Polarization Direction (deg)

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00 ISK

N=420 5 10

km

SHmax

Armutlu Peninsula

Çınarcık Basin

NAFZPrinces Island Segment

NAFZ

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00 SHmax

0 5 10

km

BRG

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00 SHmax

0 5 10

km

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00 SHmax

0 5 10

km

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00

Princes Island Segment

BYZ

N=470 5 10

km

SHmax

Armutlu Peninsula

Çınarcık Basin

NAFZ

NAFZ

28° 30 29° 00 29° 30

29° 30

40° 30

40° 45

41° 00

EAS

N=470 5 10

km

SHmax

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00

N=51

ARM

0 5 10

km

SHmax

40° 30

40° 45

41° 00

TSV

N=400 5 10

km

SHmax

0 30 60 90 120 150 180

Fast Polarization Direction (deg)

28° 30 29° 00 29° 30

40° 30

40° 45

41° 00

0 30 60 90 120 150 180

Fast Polarization Direction (deg)

Princes Island Segment

N=41

YLV

0 5 10

km

SHmax

Armutlu Peninsula

Çınarcık Basin

NAFZ

NAFZ

(a) (b)

(c)

Figure 4. Shear-wave splitting (SWS) results for the selected stations showing the analyzed earthquakes (color-encoded by fast polari-zation directions, PDs). The rose diagram in the upper right summarizes the results for each respective station. The red triangle shows thestation location. (a) Fast shear-wave orientations for northern stations (ISK, BRG, PIE, and SCR). (b) Fast shear-wave orientations for two ofnorthern stations (BYZ, EAS) and two of southern stations (ARM, TSV). (c) Fast shear-wave orientations for southeast station, YLV. Allthree subplots are aligned so as to show results at individual stations from north to south. The local distribution of maximum horizontalshortening directions (SHmax) based by (Örgülü, 2011) is shown by solid bars in light red. Solid gray lines represent active faults (Armijoet al., 2005). The color version of this figure is available only in the electronic edition.

916 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar

a CCC of>0:65. The third column then shows rose diagramsfor SNR > 2:5 and CCC > 0:75. Finally, the fourth columnincludes only events whose maximum difference betweenmeasured PDs is less than 20° and the maximum differencebetween measured fast TDs is less than 0.02 when the win-

dow size is varied by �0:02 s. The columns typically reflectincreasingly more sharply preferred orientations of the PDs

from left to right while the number of results for other thanthe predominant PDs is significantly reduced (Fig. 6). Weconsider this to be a relevant observation not only for ourstudy but for previous SWS-derived crustal anisotropy inwhich quality control criteria were performed in a more gen-eral sense (e.g., Peng and Ben-Zion, 2004; Liu et al., 2008).Our results indicate that it is essential to apply ambitiousselection criteria to the input waveform data, although thisdrastically reduces the number of analyzed earthquakes.Events passing such strict criteria tend to show a muchclearer preferred fast PD orientation allow drawing well-constrained conclusions on the nature of crustal anisotropy.

Depth of Anisotropy

The depth range of the events analyzed in this studyextends throughout the entire seismogenic part of the crust(from near surface to the depth of 18.6 km), although hypo-centers typically fall between depths of 5 and 15 km. TheSWS analysis lacks the vertical resolution due to the pathintegrated effect of velocity anisotropy along the ray path.Therefore, mapping shear-wave anisotropy parameters is dif-ficult in 3D space. Figure 7 presents the obtained TDs versusthe event depths for all nine stations from north to south. Forstation ISK and the island-based stations EAS, SCR, andBRG, we observe a depth dependence of the TDs betweenhypocentral depths of 5 and 11 km. Compared with ISK theisland-based stations display a smaller effect of anisotropy atshallower depths. We note that island stations correspondto steeper incidence angles since they are closer to the faultzone. For this geometry, ray paths propagate mostly invertical axis and are therefore less affected by the laterallydeveloped anisotropic domain. In the south, this patternexists only for station ARM. Stations TSV and YLV do notoutline any clear pattern within the upper crustal depths sinceshallow-depth events are not properly recorded at these twostations. Time delays averaged for all stations versus theevent depth indicate an increasing strength of anisotropywith depth for the depth range of 3–10 km (Fig. 7). Withinthis depth range, average TDs gradually increase from 10 to

41°

40° 15

40° 30

40° 45

0 5 10

km

iskbrgpiescrbyzeas

ylvtsvarm

isk

brg

arm

tsvylv

yassiada

sivriada

SHmax

Istanbul

Çınarcık Basin

Armutlu Peninsula

Figure 5. Summary of fast polarization directions (PDs) foreach station. Rose diagrams show results of shear-wave splitting(SWS) analysis as plotted in Figure 4 in the study area. Gray circlesindicate the epicentral distributions of all events analyzed in thisstudy. Fault traces in the Sea of Marmara are taken from Armijoet al. (2005). Some examples from the composite fault plane sol-utions derived from the first polarity approach (Bulut et al., 2009)and are from the regional moment tensor (RMT) inversion catalog(Örgülü, 2011) are represented by black balls and gray beach balls,respectively. Light gray solid lines show local distribution ofmaximum horizontal shortening directions (SHmax) estimated by(Örgülü, 2011). The color version of this figure is available onlyin the electronic edition.

Table 1Statistics of the PDs and TDs Measured for Each Station

Station Mean PDs (°) Dominant PDs (°) r Mean TDs (s) Standard TDs

BYZ 117.674 135 NW–SE 0.462 0.0364 0.0288EAS 116.898 135 NW–SE 0.453 0.0461 0.0428PIE 106.133 135 NW–SE 0.403 0.0397 0.0378SCR 108.244 135 NW–SE 0.392 0.0560 0.0462ISK 101.928 135 NW–SE 0.270 0.0390 0.0221BRG 102.065 150 NW–SE 0.254 0.0640 0.0481ARM 104.296 150 NW–SE and 60 NE–SW 0.187 0.0541 0.0347YLV 82.891 13 NE–SW 0.136 0.0686 0.0367TSV 112.315 100 NW–SE 0.412 0.0476 0.0417

NW, northwest; SE, southeast; NE, northeast; SW, southwest.

Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey 917

65 ms. For greater depths below 10 km, we observe a ratherconstant distribution of the TDs with only a few minor fluc-tuations. Interestingly, standard deviations are as large ascorresponding averages, which could be possibly due to

averaging of the TDs over a large region possibly accommo-dating strong lateral variations of seismic anisotropy. Thestrength of the seismic anisotropy was calculated to bearound ∼1–3 percent throughout the area of investigationconsidering an average TD of 50 ms, 5–15-km range ofhypocentral depth and 3:0 km=s shear-wave velocity.

Discussion

We performed an SWS analysis to examine the aniso-tropic properties of the crustal volume beneath the easternSea of Marmara region using local seismicity recorded atnine selected stations of the PIRES, KOERI, and ARNET seis-mic networks surrounding the submarine Princes’ Islandssegment of the NAFZ (Figs. 1b and 5). The results obtainedfrom the six northern stations show consistent fast PDs pre-dominantly oriented in a northeast–southwest axis rangingbetween N135°E and N150°E. This range is in good accor-dance with both the local trend of the Princes’ Islands seg-ment (∼N120°E) and the regional trend of SHmax determinedfrom inversion of focal mechanisms (Kiratzi, 2002; Pinaret al., 2003; Bohnhoff et al., 2006; Örgülü, 2011). Similarly,stress measurements from theWorld Stress Map for the studyarea indicate a northwest–southeast-oriented regional SHmax

between ∼N120°E and N160°E (Heidbach et al., 2008).While the regional trend of SHmax is well-resolved, there areno reliable results regarding potential local variation of thestress field as observed along individual segments of theIzmit 1999 rupture (Bohnhoff et al., 2006). Örgülü (2011)has mapped the stress directions and associated deformationregimes along the Çınarcık Basin and on the ArmutluPeninsula. She has observed a gradual rotation of SHmax intowest-northwest–east-southeast stress orientation in thenorthern part of the study area and interpreted such rotationin relation to the extensional Aegean regime (i.e., Mulleret al., 1992; Heidbach et al., 2004). This correlates well withour PD measurements determined for the northern stations(BRG, BYZ, EAS, ISK, PIE, and SCR). In summary, theresults obtained from the six northern stations clearly indi-cate a northwest-trending preferred orientation. Becausethis orientation is consistent with both the local fault trendof the NAFZ, as well as with the regional trend of SHmax,it is difficult to discriminate between structure- and stress-induced crustal anisotropy for this part of the study area.

The three southern stations on the Armutlu peninsulashow a more diffuse fast PD pattern possibly reflecting bothlocal fault trends, as well as local SHmax. The westernmostArmutlu station (ARM) yields two preferred PDs that are inaccordance with the regional trend of SHmax (∼N150°E)and the strike of nearby offshore faults (∼N70°E), thus sug-gesting stress- as well as structure-induced anisotropy. Theabsence of local SHmax measurements does not allow a com-parisonwith fast PDs for this station (Figs. 4b, 5). Station TSVshows a predominant east–west trend of the fast PD distribu-tions in the rose diagram (Fig. 4b). This shows a good agree-ment with the local distribution of maximum horizontal stress

ISK (N=400) ISK (N=305) ISK (N=184) ISK (N=42)

BRG (N=316) BRG (N=252) BRG (N=191) BRG (N=61)

BYZ (N=336) BYZ (N=244) BYZ (N=137) BYZ (N=47)

EAS (N=336) EAS (N=244) EAS (N=137) EAS (N=47)

PIE (N=283) PIE (N=211) PIE (N=130) PIE (N=45)

TSV (N=128) TSV (N=113) TSV (N=83) TSV (N=40)

YLV (N=269) YLV (N=174) YLV (N=41)

SCR (N=283) SCR (N=211) SCR (N=130) SCR (N=45)

ARM (N=293) ARM (N=253) ARM (N=202) ARM (N=51)

YLV (N=319)

Figure 6. Four rose diagrams for each analyzed station display-ing how fast polarization directions (PDs) change after applying in-creasingly strict quality-control criteria. The first column representsall PDs estimated after a shear-wave splitting (SWS) analysis on allwaveforms within the database. The second column of rose dia-grams is obtained after taking only PDs of the events with a sig-nal-to-noise ratio (SNR) of larger than 2.5 (for shear waves) anda cross-correlation coefficient (CCC) of 0.65 into account. The thirdcolumn of rose diagrams represents PDs filtered by the same criteriaused for the second group, but, this time, the minimum CCC is in-creased to 0.75. In the fourth column, we filter the third group ofresults by only taking those PDs in which the maximum differencebetween measured fast PDs is less than 20° when the window size isvaried by �0:02 s. Predominant PDs become clearer with stricterquality criteria (see text for details). The color version of this figureis available only in the electronic edition.

918 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar

Figure 7. Event depth versus time-delay (TD) plots obtained from all stations. Here, SIVRI and YASSI represents average TDs of EASand BYZ, and PIE and SCR stations, respectively. Solid black squares and solid black bars represent average TDs calculated in 1-km depthbins and the corresponding standard deviation, respectively. Gray diamonds show individual TDmeasurements. The graph on the lower rightshows the average TDs with event depth calculated for all measurements. Histogram plot in the same graph indicates total number of eventswithin each 1-km depth bin.

Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey 919

directions (Figs. 4b, 5), thus representing stress-inducedanisotropy. Observed northwest–southeast components ofthe fast PDs seen in the rose diagram may imply the structuraleffect placed just at the northern side of this station in a com-plex faulting network. However, this effect is rather small. Al-most no clear northwest–southeast trend might be well-explained by the proximity of the station to the NAFZ faultbranch extending only several kilometers offshore. StationYLV located at the southeast of the network gives a north–south-oriented, predominant fast PD that does easily matchneitherwith known local structures norwithmapped local dis-tribution of SHmax observations in the region. The only cor-relation with such orientation of fast PDs at this station isobserved as clockwise rotation of local stress orientationsfrom northwest–southeast into north–south direction whenmoving eastward at the southern end of the eastern Sea ofMarmara (Örgülü, 2011). In contrast with six northern sta-tions, splitting parameters measured from common eventsused for the southern stations does show a back azimuthalvariation indicating that the heterogeneity of the seismicanisotropy is significant in this part of the study area. In sum-mary,we conclude that the shear-wave splitting analysis of thenine selected stations throughout the eastern Sea of Marmararegion suggest both structure- aswell as stress-induced crustalanisotropy. All six stations located at the north of the mainNAFZ fault branch (Princes’ Island segment) show a cleardominant northwest–southeast trend. The southern stationsidentify both structure- and stress-controlled mechanisms,however, reflecting a somewhat more complicated pattern,which is well explained by the presence of locally varyingfault trends in that area.

Clear lateral variation of fast PDs as observed from northto south of the northern margin of the Sea of Marmara can beattributed to the similarity between local fault trend andregional trend of SHmax. To the south at Armutlu Peninsula,however, complexity in structural properties and large vari-ability in the distribution of maximum horizontal stress direc-tions are reflected as a less consistency among the observedanisotropic parameters. Considering a possible contribution ofboth stress and structural mechanisms to crustal anisotropy inthis region, observed diffusive pattern of the fast PDs is notsurprising. Moreover, clear lateral variation of fast PDs is con-sistent with differences in material properties. Recent findingsfrom a geodetic study published by Le Pichon et al. (2005)introduce significant asymmetric elastic loading by a ratioof 10 along the northern margin of the Sea of Marmara. Ac-cording to these authors, a large asymmetric ratio is mainlydue to seismic velocity contrast, thus producing differencesin dynamic rigidity within the upper brittle section of the crustand contrasts in rheology in the deeper creeping sections.

The seismic activity we observe mostly occurs south ofthe main fault branch of the NAFZ in the eastern Sea ofMarmara. This is probably due to strain asymmetry acrossthe fault, as well as difference in crack density between thesides of the fault. The seismicity suggests that the southernside of the main fault represents a more fractured and inter-

nally deformed block while the northern side is relativelystiff and less deformed. All fast PDs tend to be parallel orsubparallel to the direction of maximum horizontal stressin the north, and they are more scattered in the relativelycompliant block in the south. It seems that stiffer crustalblocks better reflects the effect of the regional stress onanisotropy compared with more fractured southern blocks.

Clear lateral variations of fast PDs might also serve as anindicator for a clear structural boundary between different rockbodies, referred to as a bimaterial interface. These interfacesare formed progressively by the long-term relative motion ofcrustal blocks along the faults, as well as the cumulative pro-duction of rock damage associated with the faulting process(e.g., Ben-Zion and Sammis, 2003, and references therein).Lithology contrasts across the NAFZ have been previously dis-cussed by Şengör et al. (2005) and Le Pichon et al. (2005).More recently, Özeren and Holt (2010) have shown that a con-trast of lithology across faults can also significantly affect thegeodetic fields during the interseismic periods. The existenceof bimaterial interfaces in the structure of faults at depth wasevident from the analysis on head waves observed from highfrequency microseismic data across the Mudurnu segment ofthe North Anatolian Fault Zone (Bulut et al., 2012) and fromthe analysis of teleseismic arrival times across the 1944 Ger-ede rupture zone (Ozakin et al., 2012).

Our data reveals that the strengths of the crustalanisotropy measured from TDs is of the same order as thosefound from the previous shear-wave splitting studies per-formed along the NAFZ (e.g., Peng and Ben-Zion, 2004,2005; Hurd and Bohnhoff, 2012). There, the majority of TDscompiled from previous studies are ≤100 ms. The time delaymeasurements are also consistent with SWS studies per-formed along the San Andreas Fault Zone (e.g., Liu et al.,2008). According to Barruol and Mainprice (1993), TDsbetween split shear waves caused by a heterogeneous crustwith small-scale anisotropic bodies could be as high as200–300 ms depending on the strength of the anisotropy.

In general, rose diagrams of the PDs obtained both in thisstudy and previous studies exhibit a somewhat scattered pat-tern, which is most likely attributed to the presence of stronglateral variations of crustal anisotropy for a specific depthrange. For stations located outside the San Andreas FaultZone, Liu et al. (2008) suggests that ray paths sampling themedium within or close to the fault zone can be affected byboth structural- and stress-induced anisotropy, which con-trols the areas in and outside the fault zone, respectively.Furthermore, the sedimentary bedding planes and alignedminerals/grains in the region can also be responsible for adiffusive pattern of the shear-wave anisotropy parameters. Inaddition to the complexity of shear-wave anisotropy withinthe crust, possibly due to a mixture of different sources, weemphasize that employing various quality criteria in the SWS

analysis is crucial to exclude low-quality data and insensitiveresults from the interpretation (Fig. 6).

Employing a strict quality-control criteria procedure inSWS studies is, in fact, necessary to ensure the reliability for

920 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar

the estimates of anisotropic parameters. Figure 6 shows astepwise procedure in which we increased the quality thresh-old leading to a considerable decrease in the number ofanalyzed events. However, general patterns of the PDs onrose diagrams for each station usually remain similar (Fig. 6).This is mainly because of the events displaying smaller CCCvalues (<0:75) that were filtered out during the qualityassessment although they indicated similar direction of fastpolarization. The threshold value for the CCC was chosen as0.75 that is well consistent with the previous application ofthe SWS analysis. To obtain a much higher sensitivity inestimating TDs, conversely, we used a frequency range be-tween 1–20 Hz, including higher frequencies, and representthe S-wave energy better than the previous studies in which Swaves are usually filtered between 1–10 Hz. The only draw-back, however, is the increased amount of noise being intro-duced when high frequency content is used, which causes areduction in the degree of similarity between two horizontalcomponents. Thus, setting strictly higher values for that ofCCCs is useful in eliminating potential bias from the noisecontamination, unless it changes the general pattern of thedistribution of the fast PDs. By testing the measurementsfor varying window length, we ensure the reliability ofthe splitting parameters, which may be strongly influencedby high P-coda energy or manmade marine noise observedparticularly on vertical components used for the rotation ofseismograms.

Because of the path integration effect of shear-wavesplitting analysis, this technique is not capable of quantifyingand discriminating the seismic anisotropy with depth.However, a first-order approximation of the dependenceof the average TDs upon the event depths enables us to locatethe potential depth range of the major source of the aniso-tropy. The average of all TDs measurements, which is takenat 1-km intervals across the entire region, suggests that∼25% of all analyzed events display clear depth dependencewithin a depth range of 3–10 km. The amount of data pro-vides a reasonably sufficient resolution to explain observedincreases in split TDs within this depth range (Fig. 7). Anincrease in split TDs within the uppermost crust (3–10 km)may be required by the data especially considering that wedo not have observation of any systematic pattern for deeperparts (i.e., between 15 and 19 km) where the analyzed num-ber of events and the data resolution is relatively lower. Thisdepth range considered for the depth extent for seismicanisotropy and similar to those previously observed at differ-ent segments of the western NAFZ (Peng and Ben-Zion,2004, 2005; Hurd and Bohnhoff, 2012).

Results from the analysis of local earthquakes inCalifornia have shown that the maximum horizontal com-pressive stress is semi-perpendicular to the strike of the fault(Boness and Zoback, 2006). Similarly, Zinke and Zoback(2000) found that two zones of different but internally con-sistent fast PDs correlate with the fault strike and maximum-and minimum-stress directions. These authors associated theanisotropy parameters measured from a cluster of micro-

earthquakes located ∼4 km to the east from the main faultsystem to the state of stress in the crust. In contrast, severalstudies (i.e., Cochran et al., 2006; Zhang et al., 2007; Liuet al., 2008) also reported rather ambiguous correlationsbetween observed fast directions and fault-station distances.Existing SWS studies along major strike-slip transformfaults, namely the SAF in California and the NAFZ in north-west Turkey, did not always clearly distinguish betweenstructure- or stress-induced mechanisms as the driving forcefor crustal anisotropy. In particular, it has not yet beenresolved what the critical lateral distance to a nearby faultsegment would be in order to clearly discriminate the influ-ence of either mechanism. This might be due to existingcomplex networks of secondary or splay faults near thestation sites, an insufficient amount of observation points,the use of different techniques/quality criteria, or the effectof source-site anisotropy (e.g., Peng and Ben-Zion, 2004;Liu et al., 2008). To address the issue of the critical lateraldistance to a nearby fault segment, which can discriminatethe influence of either mechanism, we compile all availablecrustal anisotropy studies from previous SWS studies alongthe SAF and NAFZ, respectively (Peng and Ben-Zion; 2004;Boness and Zoback, 2006; Liu et al., 2008; Hurd and Bohnh-off, 2012). We systematically investigate the deviation of thefast PDs from the strike of the closest major fault near thestation, as well as the strike of regional maximum horizontalstress (SHmax; Fig. 8). We classified the observed PDs fromthe listed studies, including our own results, into major PDs,directions representing the best-defined preferred orientationfor each station, and secondary PDs, directions representingother but still clearly preferred orientations for PDs at a par-ticular station. The cumulative number of observations fromall the studies is 94 major PDs and 59 secondary PDs, respec-tively. The high-density region of observations in this analy-sis asserts that the deviations of major PDs with respect to thestructure rapidly increases with fault zone distance up to∼40°, particularly, within the first 5–10 km lateral distanceto the fault (Fig. 8a). In contrast, the angular deviation be-tween observed PD and regional SHmax within the high-density region of observations on average is substantiallyhigher close to the fault (50–90° up to ∼5 km lateral dis-tance) decreasing to 0–20° at >5 km lateral distance to thefault (Fig. 8b). A relatively large scattered pattern observedfor angular deviations of both major PDs and secondary PDs

with respect to structure or stress, however, always appears inthe low-density regions of observations. In this respect, theycan be considered as outliers, and the low resolution of thesplitting parameters could be proposed as the primary reasonfor these outliers. However, we can assume that the averageaccuracy for the representative fast PDs must be reasonablygood due to the fact that we take either primary or secondaryfast PDs as reference. In this case, an unknown local struc-tural effect (e.g., small-scale faulting, etc.) can cause theseoutliers. Exponential type of curve fits seem to be better rep-resenting the distribution of two different angular deviations(Fig. 8a,b). However, drawing a generalized conclusion

Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey 921

based on only these fits could mean an oversimplification ofthe effect of the heterogeneity of seismic anisotropy, which isevidenced, in fact, by a general scattered pattern of angulardeviations for the same range of fault zone distances.However, when we ignore the outliers and only focus on theoverall trend, then it is possible to say that the structural setupseems more effective at close distances (<5–10 km) to thefault while the state of stress (SHmax) seems to control theanisotropy at larger (>5–10 km) distances from the fault.For secondary PDs, we observe an entirely different relationin that where there is no clear preferred angular deviation

between PD and fault structure or SHmax, respectively(Fig. 8c,d). The results obtained for the secondary PDs mighteither indicate that observed secondary PDs are reflectingerroneous preferred orientations (e.g., due to preselection cri-teria not being strict enough, see Discussion above) or thatthese relations are due to more complex and not fully under-stood mechanisms. Our main conclusion from this compila-tion of existing SWS splitting studies along the SAF andNAFZ is that 5–10 km lateral offset of a station from a majorfault branch might be seen as a critical distance for the dis-tinction between structure- and stress-controlled anisotropy.

0 5 10 15 20 25 30 35 40

0

10

20

30

40

50

60

70

80

90

0 5 10 15 20 25 30 35 400

10

20

30

40

50

60

70

80

90

De

via

tio

na

ng

le(d

eg

)

(a)

(c) (d)

(b)

Distance to FZ (km)

De

via

tio

na

ng

le(d

eg

)

Distance to FZ (km)

deviation of the major from the fault structure

PDs

deviation of the secondary from the fault structure

PDs

Distance to FZ (km) Distance to FZ (km)

0

1

2

3

4

5

6

7

0

1

2

3

4

5

6

7

De

ns

i tyo

ft h

eo

bs

erv

atio

ns

c

0

0.5

1

1.5

2

2.5

3

0 5 10 15 20 25 30 35 40

0

10

20

30

40

50

60

70

80

90

0 5 10 15 20 25 30 35 400

10

20

30

40

50

60

70

80

90

Peng & Ben-Zion (2004)

Boness & Zoback (2006)

Liu et al. (2008)

Hurd & Bohnhoff (2012)

This study

deviation of the major PDs from S maxH

deviation of the secondary PDs from S maxH

0

0.5

1

1.5

2

2.5

3

De

ns

it yo

fth

eo

bs

erv

at io

ns

Figure 8. Angular deviation of polarization directions (PDs) with distance to the fault zone for various stations near the San Andreas (SA)and North Anatolian (NA) fault zones. (a) and (b) The angular deviation between the prominent direction of the PDs and corresponding strikeof the main fault structure and from the orientation of regional maximum horizontal stress, respectively. (c) and (d) The angular deviationbetween the secondary direction of the PDs from the corresponding strike of the main fault structure and from the orientation of regionalmaximum horizontal stress, respectively. A map of observation density and associated color bar aside are placed with each subplot. Anobservation density map is calculated within 1 km interval of the fault-normal distance and 5° of the angular deviation. Contours (in solidblack line) display an upper boundary limit of deviation distribution at each plot. Best-fitting exponential curve to the deviation of major PDdatasets are indicated by a dark gray solid curve. PD data are compiled from Peng and Ben-Zion (2004), Boness and Zoback (2006), Liu et al.(2008), Hurd and Bohnhoff (2012), and this study. The color version of this figure is available only in the electronic edition.

922 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar

Conclusions

We present results from an SWS analysis performedusing microseismic earthquakes recorded at nine selectedseismic stations operated in the eastern Sea of Marmararegion in northwest Turkey. Well-defined but variable pre-ferred fast PDs are observed indicating a strong lateralvariation of anisotropy throughout the target area. At thenorthern part of the study area along the Princes’ Islandssegment of the NAFZ, the PDs are predominantly orientednorthwest–southeast (N135°E–N150°E) and thus alignedwith the orientation of the maximum regional horizontalstress (SHmax) and the strike of the Princes’ Islands segmentof the NAFZ. At the southern part of the study area, weidentify both structure- and stress-controlled anisotropymechanisms depending on the location of the individual sta-tion with respect to the trend of nearby local fault branches.Applying stricter selection criteria to the SWS results allowfor identification of increasingly sharp preferred PDs at eachparticular station, indicating that careful quality-controlcriteria are crucial to isolating high-quality SWS results.Clear lateral variation in the distribution of fast shear-wavePDs from north to south of the basin is confirmed by a pre-existing structural boundary, as well as an asymmetric dis-tribution of strain between different crustal blocks across thefault zone. A depth range of 3–10 km is estimated for thedepth extent for seismic anisotropy. Combined fast PDs

from various published SWS studies along the SA and NA

fault zones suggest that prominent PDs beneath stations atvery close distances (<5–10 km) to a major fault branchseems to be controlled mainly by structural anisotropymechanisms whereas stress-controlled mechanisms tend toinfluence the PDs beneath stations located >5–10 km froma major fault trace.

Data and Resources

Waveform data used in the present study were extractedfrom continuous recordings of the stations within the PIRES

network, KOERI (national), and ARNET networks. The datafrom PIRES and ARNET networks are open to the public ac-cess. The data from KOERI can be obtained from the KandilliObservatory and Earthquake Research Institute-NationalEarthquake Monitoring Center (KOERI-NEMC) the followingat http://www.koeri.boun.edu.tr/sismo (last accessed June2011). The data regarding maximum shortening directionsfor the study area were provided by Gonca Örgülü.

Acknowledgments

We thank Associate Editor Samik Sil and two anonymous reviewersfor their valuable comments and recommendations during the revision proc-ess of this paper. We are indebted to Owen Hurd for his careful reading andconstructive comments on the manuscript. We also thank Gonca Örgülü forsharing her database regarding maximum shortening directions for the studyarea. We would like to thank the Helmholtz Foundation for funding in theframe of the Young Investigators Group (from microseismicity to largeearthquakes). Şerif Barış and Heiko Woith are thanked for providing

waveform data from the ARNET network. Furthermore, we thank KOERI

National Earthquake Monitoring Center for sharing digital recordings.

References

Armijo, R., N. Pondard, B. Meyer, G. Ucarkus, B. M. de Lepinay,J. Malevieille, S. Dominguez, M.-A. Gustcher, S. Schmidt, C. Beck,N. Cagatay, Z. Cakir, C. Imren, K. Eris, B. Natalin, S. Ozalaybey,L. Tolun, I. Lefevre, L. Seeber, L. Gasperini, C. Rangin, O. Emre,and K. Sarikavak (2005). Submarine fault scarps in the Sea ofMarmara pull-apart (North Anatolian fault): Implications for seismichazard in Istanbul, Geochem. Geophys. Geosyst. 6, no. Q06009, 1–29,doi: 10.1029/2004GC000896.

Barka, A. (1999). The 17 August Izmit earthquake, Science 285, 1858–1859,doi: 10.1126/science.285.5435.1858.

Barruol, G., and D. Mainprice (1993). A quantitative evaluation of thecontribution of crustal rocks to shear-wave splitting of teleseismicSKS waves, Phys. Earth Planet. Int. 78, 281–300.

Ben-Zion, Y., and C. G. Sammis (2003). Characterization of fault zones,Pure Appl. Geophys. 160, 677–715.

Bird, P. (2003). An updated digital model of plate boundaries, Geochem.Geophys. Geosyst. 4, no. 3, 1027, doi: 10.1029/2001GC000252.

Bohnhoff, M., H. Grosser, and G. Dresen (2006). Strain partitioning andstress rotation at the North Anatolian fault zone from aftershock focalmechanisms of the 1999 Izmit Mw 7.4 earthquake, Geophys. J. Int.166, 373–385, doi: 10.1111/j.1365-246X.2006.03027.x.

Boness, N., and M. Zoback (2006). A multiscale study of the mechanismcontrolling shear velocity anisotropy in the San Andreas FaultObservatory at depth, Geophysics 71, no. 5, F131–F146.

Bulut, F., Y. Ben-Zion, and M. Bohnhoff (2012). Evidence for a bimaterialinterface along the Mudurnu segment of the North Anatolian FaultZone from P-wave arrival times and polarization analysis, Earth

Planet. Sci. Lett. 327–328, 17–22.Bulut, F., M. Bohnhoff, W. L. Ellsworth, M. Aktar, and G. Dresen (2009).

Microseismicity at the North Anatolian Fault in the Sea of MarmaraOffshore Istanbul, J. Geophys. Res. 114, no. B09302, doi: 10.1029/2008JB006244.

Bulut, F., W. L. Ellsworth, M. Bohnhoff, M. Aktar, and G. Dresen (2011).Spatiotemporal earthquake clusters along the North Anatolian faultzone offshore İstanbul, Bull. Seismol. Soc. Am. 101, no. 4, 1759–1768, doi: 10.1785/0120100215.

Carton, H., S. C. Singh, A. Hirn, S. Bazin, B. de Voogd, A. Vigner, A. Ric-olleau, S. Cetin, N. Ocakoglu, F. Karakoc, and V. Sevilgen (2007).Seismic imaging of the three-dimensional architecture of the CınarcıkBasin along the North Anatolian Fault, J. Geophys. Res. 112,

no. B06101, doi: 10.1029/2006JB004548.Cochran, E., Y.-G. Li, and J. Vidale (2006). Anisotropy in the shallow crust

observed around the San Andreas fault before and after the 2004M 6.0Parkfield earthquake, Bull. Seismol. Soc. Am. 96, S364–S375.

Cochran, E., J. Vidale, and Y.-G. Li (2003). Near-fault anisotropy followingthe Hector Mine earthquake, J. Geophys. Res. 10, 2436.

Crampin, S. (1987). The geological and industrial implication of extensivedilatancy anisotropy, Nature 328, 491–496.

Eisenlohr, T. (1997). The thermal springs of the Armutlu Peninsula(NW Turkey) and their relocationship geology and tectonic, in Active

Tectonics of Northwestern Anatolia—The Marmara Poly-project,C. Schindler and M. Pfister (Editors), ETH Zürich, 197–228.

Fukao, Y. (1984). Evidence from core-reflected shear waves anisotropy inthe earth mantle, Nature 309, 695–698.

Heidbach, O., A. Barth, P. Connolly, F. Fuchs, B. Müller, J. Reinecker,B. Sperner, M. Tingay, and F. Wenzel (2004). Stress Maps in a Minute:The 2004 World Stress Map Release, EOS Trans. AGU 85, no. 49,521–529.

Heidbach, O., M. Tingay, A. Barth, J. Reinecker, D. Kurfeß, and B. Müller(2008). The World Stress Map database release 2008, doi: 10.1594/GFZ.WSM.Rel2008.

Crustal Anisotropy in the Eastern Sea of Marmara Region in Northwestern Turkey 923

Hurd, O., and M. Bohnhoff (2012). Stress and structural-induced shear-waveanisotropy along the 1999 Izmit Rupture, Northwest Turkey, Bull.Seismol. Soc. Am. 102, no. 5, 2177–2188, doi: 10.1785/0120110270.

Karabulut, H., M.-P. Bouin, M. Bouchon, M. Dietrich, C. Cornou, andM. Aktar (2002). The seismicity in the eastern Marmara Sea afterthe 17 August 1999 Izmit earthquake, Bull. Seismol. Soc. Am. 92,387–393, doi: 10.1785/0120000820.

Kaya, T. (2012). Imaging of the North Anatolian Fault Zone underneath theMarmara Sea, Turkey, by ocean bottom magnetotellurics, Ph.D Thesis,Tokyo Institute of Technology.

Kern, H., and H.-R. Wenk (1990). Fabric-related velocity anisotropy andshear-wave splitting in rocks from the Santa Rosa Mylonite Zone,California, J. Geophys. Res. 95, 11,213–11,224.

Kiratzi, A. (2002). Stress tensor inversions along the westernmost NorthAnatolian Fault Zone and its continuation into the North AegeanSea, Geophys. J. Int. 151, no. 2, 360–376.

Le Pichon, X., C. Kreemer, and N. Chamot-Rooke (2005). Asymmetryin elastic properties and the evolution of large continental strike-slip faults, J. Geophys. Res. 110, no. B03405, doi: 10.1029/2004JB003343.

Liu, Y., H. Zhang, C. Thurber, and S. Roecker (2008). Shear-waveanisotropy in the crust around the San Andreas fault near Parkfield:Spatial and temporal analysis, Geophys. J. Int. 172, 957–970, doi:10.1111/j.1365-246X.2007.03618.x.

McClusky, S., S. Balassanian, A. Barka, C. Demir, S. Ergintav, I.Georgiev, O. Gurkan, M. Hamburger, K. Hurst, H. Kahle, K. Kastens,G. Kekelidze, R. King, V. Kotzev, O. Lenk, S. Mahmoud, A. Mishin,M. Nadariya, A. Ouzounis, D. Paradissis, Y. Peter, M. Prilepin,R. Reilinger, I. Sanli, H. Seeger, A. Tealeb, M. N. Toksoz, and G. Veis(2000). GPS contraints on crustal movements and deformations in theeastern Mediterranean (1988–1997): Implications for plate dynamics,J. Geophys. Res. 105, no. B3, 5695–5719.

McKenzie, D. P. (1972). Active tectonics of the Mediterranean region,Geophys. J. Roy. Astron. Soc. 30, 109–185.

Müller, B., M. L. Zoback, K. Fuchs, L. Mastin, S. Gregersen, N. Pavoni,O. Stephansson, and C. Ljunggren (1992). Regional patterns oftectonic stress in Europe, J. Geophys. Res. 97, 11,783–11,803.

Nuttli, O. (1961). The effect of the earth’s surface on the S-wave particlemotion, Bull. Seismol. Soc. Am. 51, no. 2, 237–246.

Örgülü, G. (2011). Seismicity and source parameters for small-scaleearthquakes along the splays of the North Anatolian fault (NAF) inthe Marmara Sea, Geophys. J. Int. 184, no. 1, 385–404, doi:10.1111/j.1365-246X.2010.04844.x.

Örgülü, G., andM. Aktar (2001). Regional moment tensor inversion for strongaftershocks of the August 17, 1999 Izmit earthquake (Mw 7.4), Geo-phys. Res. Lett. 28, 371–374, doi: 10.1029/2000GL011991.

Ozakin, Y., Y. Ben-Zion, M. Aktar, H. Karabulut, and Z. Peng (2012). Veloc-ity contrast across the 1944 rupture of the North Anatolian fault east ofIsmetpasa from analysis of teleseismic arrivals, Geophys. Res. Lett. 39,no. L08307, doi: 10.1029/2012GL051426.

Özeren, S., and W. E. Holt (2010). The dynamics of the eastern Mediterra-nean and eastern Turkey, Geophys. J. Int. 183, 1165–1184.

Paulsen, H. (2004). Crustal anisotropy in southern California from localearthquake data, Geophys. Res. Lett. 31, no. L01601, doi: 10.1029/2003GL018654.

Peng, Z., and Y. Ben-Zion (2004). Systematic analysis of crustal anisotropyalong the Karadere-Duzce branch of the North Anatolian fault,Geophys. J. Int. 159, 253–272.

Peng, Z., and Y. Ben-Zion (2005). Spatiotemporal variations of crustalanisotropy from similar events in aftershocks of the 1999 M 7.4 IzmitandM 7.1 Düzce, Turkey, earthquake sequences, Geophys. J. Int. 160,1027–1043.

Pinar, A., K. Kuge, and Y. Honkura (2003). Moment tensor inversion ofrecent small to moderate sized earthquakes: Implications for seismichazard and active tectonics beneath the Sea of Marmara, Geophys.J. Int. 153, 133–145, doi: 10.1046/j.1365-246X.2003.01897.x.

Pinar, A., S. B. Üçer, Y. Honkura, N. Sezgin, A. Ito, Ş. Barış, D. Kalafat,M. Matsushima, and S. Horiuchi (2010). Spatial variation of the stressfield along the fault rupture zone of the 1999 Izmit earthquake, EarthPlanets Space 62, 237–256.

Sayers, C. M. (1994). The elastic anisotropy of shales, J. Geophys. Res. 99,767–774.

Şengör, A. M. C., O. Tuysuz, C. Imren, M. Sakinc, H. Eyidogan, N. Gorur,X. Le Pichon, and C. Rangin (2005). The North Anatolian fault: A newlook, Annu. Rev. Earth Planet. Sci. 33, 37–112, doi: 10.1146/annur-ev.earth.32.101802.120415.

Shih, X. R., R. P. Meyer, and J. F. Schneider (1989). An automated andanalytical method to determine shear-wave splitting, Tectonophysics165, 271–278.

Silver, P. G., and W. W. Chan (1991). Shear-wave splitting and subcontinen-tal mantle deformation, J. Geophys. Res. 96, 16,429–16,454.

Tadokoro, K., M. Ando, S. Baris, K. Nishigami, M. Nakamura,S. B. Ucer, A. Ito, Y. Honkura, and A. M. Isikara (2002). Monitoringof fault healing after the 1999 Kocaeli, Turkey, earthquake, J. Seiso-mol. 6, 411–417.

Tunç, B., D. Caka, T. S. Irmak, H. Woith, S. Tunc, S. Baris, M. F. Ozer,B. G. Lühr, E. Günther, H. Grosser, and J. Zschau (2011). The ArmutluNetwork: An investigation into the seismotectonic setting ofArmutlu–Yalova–Gemlik and the surrounding regions, Ann. Geophys.54, no. 1, doi: 10.4401/ag-4877.

Zhang, Z., and S. Y. Schwartz (1994). Seismic anisotropy in the shallowcrust of the Loma Prieta segment of the San Andreas Fault System,J. Geophys. Res. 99, 9651–9661.

Zhang, H., Y. Liu, C. Thurber, and S. Roecker (2007). Three-dimensionalshear-wave splitting tomography in the Parkfield, California, region,Geophys. Res. Lett. 34, no. L24308, doi: 10.1029/2007GL031951.

Zinke, J., and M. D. Zoback (2000). Structure-related and stress-inducedshear-wave velocity anisotropy: Observations from microearthquakesnear the Calaveras fault in central California, Bull. Seismol. Soc. Am.90, no. 5, 1305–1312.

Helmholtz Centre PotsdamGFZ German Research Centre for GeosciencesPotsdam D-14473, Germany

(T.E., M.B., F.B.)

Bogazici UniversityKandilli Observatory and Earthquake Research InstituteIstanbul, Turkey 34684

(B.C., M.A.)

Manuscript received 26 April 2012

924 T. Eken, M. Bohnhoff, F. Bulut, B. Can, and M. Aktar