Spatiotemporal Earthquake Clusters along the North Anatolian Fault Zone Offshore Istanbul

Spatiotemporal Earthquake Clusters along the North

Anatolian Fault Zone Offshore İstanbul

by Fatih Bulut, William L. Ellsworth, Marco Bohnhoff, Mustafa Aktar, and Georg Dresen

Abstract We investigate earthquakes with similar waveforms in order to charac-terize spatiotemporal microseismicity clusters within the North Anatolian fault zone(NAFZ) in northwest Turkey along the transition between the 1999 İzmit rupture zoneand the Marmara Sea seismic gap. Earthquakes within distinct activity clusters arerelocated with cross-correlation derived relative travel times using the double-difference method. The spatiotemporal distribution of microearthquakes within indi-vidual clusters is resolved with relative location accuracy comparable to or better thanthe source size. High-precision relative hypocenters define the geometry of individualfault patches, permitting a better understanding of fault kinematics and their role inlocal-scale seismotectonics along the region of interest. Temporal seismic sequencesobserved in the eastern Sea of Marmara region suggest progressive failure of mostlynonoverlapping areas on adjacent fault patches and systematic migration of micro-earthquakes within clusters during the progressive failure of neighboring fault patches.The temporal distributions of magnitudes as well as the number of events followswarmlike behavior rather than a mainshock/aftershock pattern.

Introduction

TheNorthAnatolian fault zone (referred to asNAFZ here-after) is one of the largest intracontinental transform zonesextending over ∼1200 km between Eastern Anatolia and theNorthern Aegean separating Euroasia in the north from theAnatolian plate in the south (e.g., Şengör et al., 2005). TheAnatolian plate is moving westward at a rate of ∼25 mm=yrin the framework of a rollback of theHellenic subduction zoneand the northward moving Arabian subplate (e.g., McCluskyet al., 2000).

The 1999 İzmit earthquake (Mw 7:4 of 17 August 1999)ruptured a segment over ∼110 km long of the NAFZ in north-west Turkey. The rupture extended into the easternmost Sea ofMarmara in the west and to the Karadere–Düzce area in theeast (e.g., Barka et al., 2002). While the 1999 Düzce earth-quake (Mw 7.2 of 12 November 1999) extended the İzmitrupture by ∼40 km to the east only 87 days after the İzmitevent, there is still a high probability for a destructiveM > 7

earthquake in the Sea of Marmara west of the 1999 İzmit rup-ture (e.g., Parsons et al., 2004). The Sea of Marmara segmentof the NAFZ remains the only part of the NAFZ that has notfailed during the last century, representing a seismic gapbetween the MS 7.3 1912 Ganos earthquake to the westand the 1999 İzmit rupture to the east. This segment hasnot been activated since 1766. If fully locked, it would haveaccumulated a slip deficit of 4–5 m (Fig. 1a). The fault struc-ture below the Sea ofMarmara is complex and still a matter ofdebate. Along this NAFZ segment, three pull-apart sub-basins

have been identified (LePichon et al., 2003; Armijo et al.,2005), of which the Çınarcık basin is the easternmost(Fig. 1b).

To understand the physical state of a major fault segmenttoward the end of its seismic cycle, we installed a local seismicnetwork within the eastern part of the Marmara Sea seismicgap in order to monitor the present-day seismic activity at alow detection threshold. In a first stage, two 5-seismographseismic arrays were deployed on the two outermost PrincesIslands, Sivriada and Yassıada, offshore İstanbul at only∼3 km distance to the submarine outcrop of the major NAFZfault branch below the Çınarcık basin, the Princes Islandssegment of the NAFZ [Prince Islands Realtime EarthquakeMonitoring System (PIRES) campaign, Bohnhoff et al.,2007; Bulut et al., 2009] (Fig. 1b). Stations of the two arrayswere deployed in a cross-shaped layout with an average spa-cing of ∼150 m (Fig. 1c). Additionally, we deployed a singleseismograph onBurgazada to improve the azimuthal coverageof target hypocenters from close proximity. Data from thePIRES arrays were combined with recordings from selectedpermanent stations of the Turkish Kandilli Observatory andEarthquake Research Institute network (KOERI) and theArmutlu peninsula network (ARNET; Ş. Baris, H. Woith,personal commun., 2010; Fig. 1b,c). Coherent waveformsrecorded across the PIRES arrays at >3 km distance to thetarget area allowed us to better identify the microearthquakescomparedwith individual stations on themainland. ThePIRES

1759

Bulletin of the Seismological Society of America, Vol. 101, No. 4, pp. 1759–1768, August 2011, doi: 10.1785/0120100215

arrays allowed us to improve the magnitude of completenessalong the Princes Islands segment of theNAFZ down toM 2:0

(Bulut et al., 2009).The microseismicity observed along the study area

defines a set of distinctly aligned clusters delineating two∼110°N northwest–southeast trending subparallel fault seg-ments beneath the Çınarcık basin (Bulut et al., 2009) that are

also seen in high-resolution seafloor imaging (Armijo et al.,2005). The depth distribution of hypocenters suggests thatsubclusters and fault segments merge with increasing depthstoward a single master fault below the seismogenic layer thatis supported by results from multichannel seismic profilesacross the Çınarcık basin that show merging fault brancheswithin shallow depths (Carton et al., 2007, Laigle et al.,2008). Fault plane solutions for the eastern Sea of Marmarashow a predominant dextral strike-slip regime along thenorthern strand of the NAFZ bounding the Çınarcık basinin agreement with the analysis of İzmit aftershocks (Örgülüand Aktar, 2001; Pınar et al., 2003; Bohnhoff et al., 2006).The deformation pattern changes toward the west at thebending point of the NAFZ south of İstanbul where a substan-tial thrust component is present in the mechanisms (Bulutet al., 2009). This suggests an added component of faultnormal compression along the east–west trending NAFZ seg-ment west of İstanbul as opposed to pure strike-slip motionalong the northern slope of the Çınarcık basin. Faultingcharacteristics suggest that currently normal faulting playsa minor role offshore İstanbul as opposed to the southernÇınarcık basin below the Armutlu peninsula where normalfaulting dominates (Örgülü et al., 2011).

In this study, we analyze spatiotemporal earthquakesequences occurring along secondary faults close to thePrinces Islands segment of the NAFZ at the transition be-tween the 1999 İzmit rupture and the Sea of Marmara seismicgap. We make use of the close spatial proximity of events toimprove the resolution of hypocenters using relative reloca-tion techniques. The relative earthquake locations are deter-mined using the double-difference earthquake locationtechnique based on cross-correlation derived relative traveltimes (Waldhauser and Ellsworth, 2000). Refined spatiotem-poral distribution of hypocenters provides insight into theinteraction of colocated earthquakes at a scale comparableto or better than the earthquake source dimensions.

Data Analysis

Database

The Princes Islands are located ∼15 km southeast fromthe megacity of İstanbul; they provide good noise conditionsfor collecting the seismic data in an otherwise heavily popu-lated and noisy region and within a few kilometers of theNAFZ. Permanent subarrays on two islands (PIRES) are con-figured with a cross-shaped geometry with one of the diago-nals parallel to the main branch of NAFZ (Fig. 1c). Based onPIRES recordings an average seismicity rate of 20 events permonth for M < 2:5 was observed along the Princes Islandssegments (Bulut et al., 2009). The aperture of the subarraysis limited to ∼300 m due to the size of the islands. Seismicstations are equipped with a L4-C short period sensor provid-ing almost flat velocity response above 1.0 Hz (samplinginterval is 0.005 s). In addition to PIRES recordings, weuse four stations of a newly installed permanent seismic

Figure 1. (a) Recent sequence of large earthquakes along theNAFZ in northwest Turkey. The rectangle marks the area enlargedin (b). (b) Red dots indicate the location of the individual seismicsequences analyzed in this study. Dashed blue line indicates thetransition zone between 1999 İzmit rupture and Marmara SeaSeismic Gap that did not generate a major earthquake since1766. Stars (PIRES arrays) and squares (single permanent stations)marks stations of the combined seismic network to monitor themicroseismicity along the transition zone (Princes Islandssegment). PIRES arrays on the islands of Sivriada (‘S.’ in b) andYassıada (Y. in b). (c) Stations are deployed in cross-shape layoutto optimize monitoring conditions for seismic events occurringalong the Princes Islands segment of the NAFZ at the transitionfrom the 1999 İzmit rupture to the seismic gap.

1760 F. Bulut, W. L. Ellsworth, M. Bohnhoff, M. Aktar, and G. Dresen

network in the greater ARNET region (Ş. Baris, H.Woith, per-sonal commun., 2010). We also included data from selectedstations of the permanent seismic network that is operated byKOERI. The combined network consists of 26 seismic stationsproviding a maximum azimuthal coverage of 282° (Fig. 1b).Depth control of hypocenters is also satisfactory because theearthquakes forming the clusters analyzed here have an epi-central distance less than 15 km from the PIRES stations.

Hypocenter Determination

We use the double-difference earthquake locationmethod to obtain the highest precision of spatial offsetbetween the earthquake hypocenters (Waldhauser and Ells-worth, 2000). Use of relative arrival time data for closely

spaced events suppresses the effect of unmodeled velocitystructure on hypocentral offsets because the ray paths ofpaired events are almost identical. The method also allowsthe use of differential travel times, which can be measuredmuch more precisely than arrival time onsets, resulting inmore accurate differential locations.

We search for similar waveforms using the cross-correlation technique to locate the clustered seismicity occur-ring within small volumes (Fig. 2). The technique is usedalso to measure differential travel times. The time windowrange is �0:1 to 0.9 s framing the manually picked P- andS-wave arrival times. The cross-correlation coefficientthreshold for data selection is investigated comparing thecross-correlation coefficient and the hypocentral-precision/data-misfit obtained from the corresponding data set

Figure 2. (a) Daily seismicity rate in the Çınarcık basin from PIRES recordings: A high number of events compared to the backgroundseismicity is observed for particular time periods indicating temporal clustering of earthquakes along the Princes Islands segments of theNAFZ. The squares mark the sequences detected here, while sequence S2 contains the largest number of events (77). (b) Cumulativemagnitudes of individual sequences (squares) do not exceed the commonly observed maximum magnitudes despite very high numberof earthquakes for each sequence. (c) Single station recordings for sequence S1: Highly coherent waveforms as well as similar S–P timesconfirm close spatial proximity of the events within the temporal clusters.

Spatiotemporal Earthquake Clusters along the North Anatolian Fault Zone Offshore İstanbul 1761

(Fig. 3a). The best average precision of hypocenters isobtained for the cross-correlation coefficient range of 0.6–0.75. However, selecting the highest possible threshold(0.75) significantly reduces the number of data used from∼8000 to only ∼3000 (Fig. 3b). Therefore, the data vectoris restricted to cross-correlation coefficients >0:6. We adopta 1D reference velocity model that was optimized earlier forthe eastern Sea of Marmara region (Bulut et al., 2009) usingthe VELEST simultaneous inversion code (Kissling et al.,1994). The uncertainties of initial absolute locations weretypically 1850 m and 2470 m in the horizontal plane anddepth axis, respectively. Using the double-difference methodlowered the average misfit from 0.17 s to 0.04 s resulting inimproved precision of the relative hypocentral locations to∼40 m and ∼60 m for the horizontal and depth axis, respec-tively (Fig. 3c,d). The temporal distribution shows an aver-age seismicity rate of 20 events per month while the strongestof the temporal activity clusters includes 77 events withinless than 24 hrs. We identify a total of seven clusters (Fig. 2a).Individual clusters do show a high similarity of waveformsindicating that the temporal clusters are concentrated withindistinct spatial activity sports (Fig. 2c).

Source Dimensions

Initial pulse widths of seismograms are analyzed tomeasure the earthquake source size. The time between the

initial onset and the first zero crossing is used to representthe duration of the rupture. Following O’Neill (1984), wefirst measure initial pulse widths of P waves to determinetheir minimum value. Assuming that the initial pulse mustbe characterized by at least 5 samples (4 intervals), the mini-mum initial pulse that we can measure is 0.02 s based onsampling rate of our data (200 Hz). In reality, we observeminimum pulse widths between 0.016 and 0.021 s, corre-sponding to source radii of smaller than ∼20 m (Fig. 4a).The minimum pulse widths are used to estimate the attenua-tion and the required corrections for the quality-factor (Q).Following Gladwin and Stacey (1974), initial P-wave pulsewidth (τ ) is represented as

τ � τ 0 � ctQ�1: (1)

Here, c is a constant (assumed to be 1.0), t is travel time, andτ0 is the initial pulse width at the earthquake source (t � 0).A linear system of equation (1) is constructed combiningminimum initial pulses observed at different distances inorder to average the Q-factors (Fig. 4a). Least square analy-sis of equation (1) allows determination of the Q-factorto 441� 21 and 368� 38 for the islands of Sivriada andYassıada, respectively. The trend of pulse widths versusmagnitudes shows that the pulse widths are reliably mea-sured for the magnitudes above 0.9, which corresponds to65% of the events analyzed here. The recordings are

Figure 3. (a) Average location error versus cross-correlation coefficient plotted to determine the threshold for selecting the data vector.The best performance is obtained for the range between 0.6 and 0.75. (b) Number of data used versus cross-correlation coefficient threshold.The threshold range of 0.6–0.75 corresponds to number of data ranging from ∼8000 to ∼3000. (c, d) Error distribution of relocatedhypocenters estimated using variance-weighted covariance matrix and bootstrap resampling technique.

1762 F. Bulut, W. L. Ellsworth, M. Bohnhoff, M. Aktar, and G. Dresen

corrected for instrument response. In a second step, initialpulse widths are corrected for corresponding Q-factors usingequation (1). Rupture durations are then multiplied by a con-stant rupture velocity that is assumed to be 2:6 km=s(Fig. 4b). We also assume that the duration of the outward-expanding circular rupture equals half of the Q-correctedinitial pulse width (τ0). The standard deviations of the mea-surements from different stations show that the sourcedimensions are determined typically better than 15.0 m(Fig. 4c).

Results and Discussion

Source Overlaps

The seismicity clusters analyzed here consist of a totalnumber of 125 events in 7 separate episodes with M < 2:5

earthquakes occurring sequentially one after another with aduration of each episode of less than 24 hrs (Fig. 2a,b). Wecalculate cumulative magnitudes for each sequence summingup the seismic moments of individual events. Conversionfrom seismic moment to magnitude and vice versa is per-formed using the relation MD � 0:9 log10�M0� � 16:4,which is calibrated for the Çınarcık basin area using spectra-derived seismic moments for the magnitude range of0.5–2.5 (MD: duration magnitude; andM0: Seismic momentin dyn · cm). We first calculate individual event seismicmoments from magnitudes, then sum the seismic momentsof closely spaced events and finally convert cumulative seis-mic moments to cumulative magnitudes. Despite a high num-ber of events, the cumulative magnitudes of the sequences donot exceed the maximum magnitude observed in the presentbackground seismicity along the region (Fig. 2b). Visualinspection of the recordings shows coherent waveforms with-in a sequence, indicating spatial proximity of earthquakesources within the temporal sequences (Fig. 2c). Here, theprimary focus is to resolve whether the event pairs of similarwaveforms represent failure of adjacent fault patches or repet-itive failure of the same particular area. To address this objec-tive, we analytically calculate overlaps of the source areasusing the source radii and the double-difference derived in-terevent distances. A circular rupture is assumed. It is centeredon the hypocenter determined using cross-correlation mea-surements because this location is effectively the hypocen-troid of the event and not the point of initial rupture. Thesource overlap is calculated using the ratio between overlap-ping and nonoverlapping area of the circles representing thearea of earthquake sources resulting in values between 0.0 (nooverlap) and 1.0 (100% overlap, Fig. 5).

The source overlaps are mostly small or even zerodocumenting that the majority of the events represent failureof adjacent source patches on a fault plane. For a smallnumber of events we observe significant source overlaps thatmight indicate repetitive failure of the same fault patch(Fig. 5a, left part). However, the confidence range of the cal-culations strongly depends on the precision of input data.

Figure 4. (a) Initial pulse width measurements versus traveltimes for small-size earthquakes which are used to approximateQ-factors. (b) Initial pulse width measurements and correspondingsource dimensions. Error bars represent source dimension errorsestimated by standard deviation of pulse width measurements.(c) Error distribution of source dimensions obtained from standarddeviation of initial pulse measurements at different stations.

Spatiotemporal Earthquake Clusters along the North Anatolian Fault Zone Offshore İstanbul 1763

Therefore, we incorporate the precision of double-differencelocations and standard deviation of source radius measure-ments to estimate the confidence range of source overlaps.The error bars of highly overlapping events remain above0.5 except for two event pairs (Fig. 5b). We note that thesource dimension also depends on an assumed rupturevelocity; thus, varying rupture velocities could affect ourresults on source overlap.

Rerupturing an individual source patch in a short timeinterval might be explained by an incomplete initial stressdrop. However, stress drops associated with overlappingsources are significantly higher than the median stress dropaccording to results from the analysis of spectral ratios (Bulutet al., 2010). Alternatively, amuch larger-scale slow slip eventor ongoing aseismic creep might be driving the processproviding rapid reloading. Lohman and McGuire (2007)concluded that the swarm activity observed in the SaltonTrough in 2005 was driven by long-term aseismic creep ona surrounding surface of the fault based on geodetic measure-ments and the spatiotemporal occurrence of local seismicactivity. However, around the eastern Sea of Marmara, theGPS-derived velocity field is not precise enough to verify thisissue for the time period of the sequences observed, becausethe number of onshore locations is limited to the PrincesIslands north of the fault while no onshore location closeto the fault exists toward the south (e.g., Ergintav et al., 2009).

Temporal Characteristics

We further investigate the spatiotemporal relation ofcolocated events using relative hypocenters with locationaccuracy comparable to or better than the source dimensions.

The most prominent sequence observed along the PrincesIslands segment consists of 77 relocated microearthquakesthat all occurred within less than 20 hrs on 7 July 2007 andwithin an area of about 2:5 km2 (S2 in Figs. 1b, 2a). Thissequence is located at the northern slope of the Çınarcıkbasin along the Princes Islands segment (see S2 in Fig. 1bfor location and Fig. 6a for relocated hypocenters). The epi-central alignment of the earthquakes suggests a northwest–southeast striking orientation in map view. This is rotated 15°clockwise with respect to the local trend of the NAFZ. Thestrike of the aligned events and the local trend of NAFZ

represent converging lines and suggest a junction betweenthe main fault branch and a splay fault at the southernmostpart of the cluster (Fig. 6a). Multichannel seismic profilesshow that the main branch of the NAFZ at the northernescarpment of the Çınarcık basin is steeply inclined to thesouthwest and surrounded by a network of subsidiary faults(Carton et al., 2007, Laigle et al., 2008). Interestingly, thehypocentral depths of sequence S2 range from 10.0 to11.5 km and also suggest an 80° steep dip to the southwest(Fig. 6b).

The temporal evolution of the events most likely repre-sents progressive failure of adjacent patches of the splay faultneighboring the main branch of NAFZ. To further analyzethis, we subdivided the sequence into 3-hr snapshots andplotted the hypocenters on a depth section trending north-west–southeast, that is, along the average trend of thesequence as identified in Figure 6a. During the first 6 hr, only7 microearthquakes are observed that all occur at the deepestpart of the section close to the junction of the splay fault andthe NAFZ (Fig. 7a,b). These events have relatively smallmagnitudes (M < 1:4). During the next 3-hr period (hours6 to 9 ), the activity expands to shallower depths but remainslimited to the southeast part of the plane. Thereafter theearthquakes start to systematically spread along the planeof the entire splay fault with a gradual increase of number

Figure 5. Source overlaps of colocated events plotted with theirratio of the source size. Source overlap is given between 0 (no over-lap) and 1.0 (100% overlap). Our analysis shows that source over-laps are mostly small or even zero indicating failure of adjacent faultpatches rather than repetitive failure of an individual fault patch.The error bars show source overlap errors calculated combiningrelative uncertainty of hypocenters as well as source dimensionerrors.

Figure 6. (a) Epicentral distribution of sequence S2 consistingof 77 relocated microearthquakes (see Fig. 1b, S2 for location). Thedouble-difference locations characterize a northwest–southeasttrend in first order approximation. Most events of this sequenceoccurred on a secondary fault (splay fault) close to the major NAFZfault branch (The Princes Islands segment). (b) Depth section of theseismic sequence S2 shown in a) indicating a southwest inclinedhypocentral distribution with a dip angle of ∼80°.

1764 F. Bulut, W. L. Ellsworth, M. Bohnhoff, M. Aktar, and G. Dresen

of events during the most active part of the sequence (hours 9to 15). The largest event occurs after 12 hr with a magnitudeof 2.5. The centroid of the activity then migrates to the mostnorthwestern part of the splay fault, and the seismicity ratedecreases. Strikingly, earlier hypocenters clearly define thestarting point of the sequence followed by a systematicmigration throughout the splay fault. The initial events occurin close proximity to one another compared with the subse-quent spreading of activity. The initial events allow locatingthe nucleation point of the progressive failure along the fault(Fig. 7). The complete spatiotemporal pattern clearly showsan upward as well as lateral migration of the activity front,which initiates on the junction and then propagates into thesplay fault away from the major NAFZ fault branch.

Structural Implications

The temporal distribution of magnitudes as well as thenumber of events observed within the sequences increasedgradually and then decayed without the occurrence of a dis-

tinct mainshock. Thus, S2 as well as the other sequencesobserved here represent swarmlike behavior rather thantypical mainshock/aftershock type sequences (Sykes, 1970).The earthquake swarms are considered to be hosted withinthe structures of heterogeneous materials under nonuniformstress fields (Mogi, 1963).

The small splay fault hosting cluster S2 resembles asubsidiary Riedel-type shear fault forming part of the evolv-ing fault network in the Çınarcık basin (Tchalenko, 1970).During our period of observation, spatiotemporal sequencesare located in the eastern Çınarcık basin, which is surroundedby a transtensional regime despite local transpressive featuresin the west. Composite fault plane solutions for the sequencesshow a predominant strike-slip mechanism with no evidencefor normal faulting (see also Bulut et al., 2009). Interestingly,scatter in nodal plane orientation between these clusteredevents is small, irrespective of the distance between individualsequences reaching up to 20 km (Fig. 8a). This indicatesthat the currently active structures associated with the spatio-temporal sequences display similar kinematics throughout

Figure 7. Six 3-hr snapshots of sequence S2 showing (a) spatiotemporal evolution along strike in a southeast–northwest direction and(b) in depth section. The activity is first localized on the junction of splay fault and the NAFZ and then starts migrating through the splay faultand toward shallower depths (see text for details).

Spatiotemporal Earthquake Clusters along the North Anatolian Fault Zone Offshore İstanbul 1765

the eastern Çınarcık basin. Furthermore, the epicentraldistribution of double-difference locations allows us to deter-mine the northwest–southeast trending subvertical planes ofthe fault plane solutions as the nodal planes (Fig. 8c).

Seismological observations in California showed the ex-istence of strike-slip related earthquake swarms under similartectonic settings (Hill, 1977). Hill’s fault mesh model pro-posed for earthquake swarms consists of a network of faultsconnecting fluid-filled extensional cracks oriented parallel tothe direction of the maximum principal stress (Fig. 8b). In theÇınarcık basin, the direction of the maximum principal stressis oriented in a northwest–southeast direction and thus nearlyparallel to the local trend of the NAFZ along the PrincesIslands segment (Pınar et al., 2003). The map view of thedouble-difference locations shows a northwest–southeastalignment of epicenters. This might possibly represent faultpatches connecting extensional cracks oriented parallel to thedirection of maximum principle stress (Fig. 8c). Note thatgas-emissions have been reported to occur along the faultsegments in the eastern Çınarcık basin (Geli et al., 2008).

For the time period covered by our analysis, we do notobserve any swarm activity west of 29.0° E, that is, towardthe western margin of the Çınarcık pull-apart basin. If theobserved seismic clusters are related to fluid transport orfluid pressure variations along a fault network, it is concei-vable that swarm activity is confined to local areas ofactive extension such as the Çınarcık pull-apart structure.The evolving network of active faults may provide effectivefluid conduits releasing the observed gas emissions.

Interestingly, the target area is surrounded by on-landhot springs (gray triangles in Fig. 8). The hot springs havebeen interpreted to document an increase in vertical perme-ability of the uppermost crust due to local extension of thearea (Pfister et al., 1998). Also, sea-floor acoustic measure-ments indicate fluid emissions within the surface sedimentsalong the Çınarcık basin (Geli et al., 2008). However, gasbubbles sampled along the same area do not show character-istics of mantle or lithosphere origins and rather consist ofshallower origins such as biogenic methane and thermogenicethane (Bourry et al., 2009).

Present-day seismicity in the eastern Sea of Marmarareveals lateral variations of the seismicity rate along theÇınarcık basin (Bulut et al., 2009). The seismic activity islow in the west and increases toward the east beyond 29.0° E.In particular, the sequences presented here are located onlyeast of 29.0° E. This region accommodates the western ter-mination of the 1999 İzmit rupture zone. Double-differencelocations suggest that the sequences are slightly off the mainsegments of the NAFZ. Because the main fault is probablylocked and at a very late stage in its seismic cycle, slipmay be transferred to secondary structures surrounding themain fault trace (Waldhauser and Ellsworth, 2002).

Conclusion

Spatiotemporal clustering of microearthquakes is de-tected along the eastern part of the Çınarcık basin in theSea of Marmara hosting the western end of the 1999 İzmitrupture zone. The sequences consist of up to 77 colocated

Figure 8. (a) Composite fault plane solutions for the event sequences throughout the Çınarcık basin that indicate a predominantnorthwest–southeast oriented right-lateral strike-slip regime in good correspondence with the regional tectonic setting of NAFZ. Trianglesshows the location of thermal springs. (b) Conceptual fracture-mesh model for earthquake swarms proposed by Hill 1977. Circles indicateminor faults joining en echelon, fluid-filled extension cracks, which might be observed as swarmlike earthquake sequences. (c) Relocatedhypocenters of sequences S1, S2, and S3 that emphasize northwest–southeast oriented semiparallel structures in accordance with the localsegments of the NAFZ (lines represent faults after: Armijo et al., 2005) as well as the direction of the maximum principal stress (after Pınaret al., 2003).

1766 F. Bulut, W. L. Ellsworth, M. Bohnhoff, M. Aktar, and G. Dresen

events occurring within less than a 24-hr period, while theaverage seismicity rate is 20 events per month. The temporaldistribution of magnitudes and number of events versus timesuggests that the sequences may represent swarmlike ratherthan mainshock/aftershock sequences.

Precise hypocenters document the orientation of thestructures associated with the earthquake swarms and allowdiscriminating of the fault plane from the auxiliary plane oftheir focal mechanisms. We find that the sequences representnorthwest–southeast oriented structures developed subparal-lel to the local trend of the NAFZ as well as the direction ofthe maximum principal stress. The temporal distribution of asingle sequence represents upward and lateral migration ofseismicity on an evolving subsidiary fault.

The analysis of interevent distances and the source radiiindicate that the events associated with the earthquakesequences mostly represent progressive failure of adjacentfault patches along planes of activity. However, we alsodetected several event pairs with substantial source overlapthat are interpreted to represent repeated failure on the samesource patch.

Data and Resources

The seismic data used in this study are mainly obtainedfrom PIRES network that we have been operating sinceAutumn 2006. In addition to PIRES recordings, we includefour stations of a newly installed ARNET permanent seismicnetwork in the greater Armutlu Peninsula region (Ş. Baris,H. Woith, personal commun., 2010) and selected stationsof the permanent regional network that are operated by Kan-dilli Observatory and Earthquake Research Institute (http://www.koeri.boun.edu.tr/sismo/, last accessed March 2011).We use Generic Mapping Tools to plot some of the figures.

Acknowledgments

We are grateful to Tom Parsons, Dave Hill, and two anonymous re-viewers for constructive comments improving the paper, to Sebastian Hainzland Erik Rybacki for useful discussions about potential models. We thankDogan Kalafat, Claus Milkereit, Şerif Barış, and Heiko Woith for sharingseismic data. The Geophysical Instrument Pool Potsdam provided seismicstations for the initial period of the PIRES campaign. PIRES has been co-funded by GFZ-Potsdam and Boğaziçi University Research Foundation(grant number: BAP-09M103). We also thank the Helmholtz Foundationfor funding the project within the Young Investigators Group “From micro-seismicity to large earthquakes”. Vanessa Fremd helped us with data proces-sing. We thank Birsen Can for useful discussions on data evaluation andcontinuous support for the maintenance of PIRES network.

References

Armijo, R., N. Pondard, B. Meyer, G. Uçarkuş, B. Mercier de Lépinay,J. Malavieille, S. Dominguez, M. Gustcher, S. Schmidt, C. Beck,N. Çağatay, Z. Çakır, C. İmren, K. Eriş, B. Natalin, S. Özalaybey,L. Tolun, I. Lefèvre, L. Seeber, L. Gasperini, C. Rangin, Ö. Emre,and K. Sarıkavak (2005). Submarine fault scarps in the Sea ofMarmara pull-apart (North Anatolian fault): Implications for seismichazard in Istanbul, Geochem. Geophys. Geosyst. 6, Q06009, doi10.1029/2004GC000896.

Barka, A., H. S. Akyüz, E. Altunel, G. Sunal, Z. Çakir, A. Dikbas, B. Yerli,R. Armijo, B. Meyer, J. B. de Chabalier, T. K. Rockwell, J. R. Dolan,R. D. Hartleb, T. E. Dawson, S. A. Christofferson, A. Tucker,T. E. Fumal, R. M. Langridge, H. D. Stenner, W. R. Lettis,J. Bachhuber, and W. D. Page (2002). The surface rupture and slipdistribution of the 17 August 1999 İzmit earthquake (M 7.4), NorthAnatolian fault (in The İzmit, Turkey, earthquake of 17 August1999, Toksoz), Bull. Seismol. Soc. Am. 92, no. 1, 43–60.

Bohnhoff, M., M. Aktar, F. Bulut, D. Childs, E. Görgün, H. Karabulut, andG. Dresen (2007). The North Anatolian fault in the broader Istanbul/Marmara region: Seismic monitoring of a seismic gap, EGU-HumboldtConference, The Role of Geophysics in Natural Disaster Prevention,Lima, Peru, March 2007.

Bohnhoff, M., H. Grosser, and G. Dresen (2006). Strain partitioning andstress rotation at the North Anatolian fault after the 1999 İzmit Mw �

7:4 earthquake, Geophys. J. Int. 66, no. 1, 373–385.Bourry, C., B. Chazallon, J. L. Charlou, J. P. Donval, L. Ruffine, P. Henry,

L. Geli, M. N. Çağatay, S. İnan, and M. Moreau (2009). Free gas andgas hydrates from the Sea of Marmara, Turkey, Chem. Geol. 264,no. 1–4, 197–206, doi 10.1016/j.chemgeo.2009.03.007.

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. doi 10.1029/2008JB006244.

Bulut, F., M. Bohnhoff, W. L. Ellsworth, and G. Dresen (2010). Character-ization of the closely spaced earthquakes along the North Anatolianfault zone, NW Turkey, Seismol. Res. Lett. 81, no. 2, 253–283, doi10.1785/gssrl.81.2.253.

Carton, H., C. Singh, A. Hirn, S. Bazin, B. deVoogd, A. Vigner, A. Ricolleau,S. Cetin, N. Oçakoğlu, F. Karakoç, and V. Sevilgen (2007). Seismicimaging of the three-dimensional architecture of the Çınarcık basinalong the North Anatolian fault, J. Geophys. Res. 112, B06101, doi10.1029/2006JB004548.

Ergintav, S., S. McClusky, E. Hearn, R. Reilinger, R. Cakmak, T. Herring,H. Ozener, O. Lenk, and E. Tari (2009). Seven years of postseismicdeformation following the 1999,M � 7:4 andM � 7:2, İzmit-Düzce,Turkey earthquake sequence, J. Geophys. Res. 114, B07403, doi10.1029/2008JB006021.

Géli, L., P. Henry, T. Zitter, S. Dupré,M. Tryon,M. N. Çağatay, B.Mercier deLépinay, X. Le Pichon, A. M. C. Şengör, N. Görür, B. Natalin,G. Uçarkuş, S. Özernd, D. Volker, L. Gasperini, P. Burnard,S. Bourlange, and The Marnaut Scientific Party (2008). Gas emissionsandactive tectonicswithin the submerged sectionof theNorthAnatolianfault zone in the Sea of Marmara, Earth Planet. Sci. Lett. 274, 34–39.

Gladwin, M. T., and F. D. Stacey (1974). Anelastic degradation of acousticpulses in rock, Phys. Earth Planet. In. 8, 332–336.

Hill, D. P. (1977). A model for earthquake swarms, J. Geophys. Res. 82,347–352.

Kissling, E., W. L. Ellsworth, D. Eberhart-Philipps, and U. Kradolfer (1994).Initial reference models in local earthquake tomography, J. Geophys.Res. 99, 19,635–19,646.

Laigle, M., A. Becel, B. Voogd, A. Hirn, T. Taymaz, S. Ozalaybey, andMembers of SEISMARMARA Leg1 Team (2008). A first deep seis-mic survey in the Sea of Marmara: Deep basins and whole crustarchitecture and evolution, Earth Planet. Sci. Lett. 270, 168–179.

Le Pichon, X., N. Chamot-Rooke, C. Rangin, and A. M. C. Şengör (2003).The North Anatolian fault in the Sea of Marmara, J. Geophys. Res.108, no. B4, 2179, doi 10.1029/2002JB001862.

Lohman, R. B., and J. J. McGuire (2007). Earthquake swarms driven byaseismic creep in the Salton Trough, California, J. Geophys. Res.112, B04405, doi 10.1029/2006JB004596.

McClusky, S., S. Balassanian, A. Barka, C. Demir, S. Ergintav, I. Georgiev,O. Gurkan, M. Hamburger, K. Hurst, H. Kahle, K. Kastens, G. Keke-lidze, R. King, V. Kotzev, O. Lenk, S. Mahmoud, A. Mishin, M. Nadar-iya, A. Ouzounis, D. Paradissis, Y. Peter, M. Prilepin, R. Reilinger,I. Sanli, H. Seeger, A. Tealeb,M. N. Toksöz, and G. Veis (2000). GlobalPositioning System constraints on plate kinematics and dynamics in theeasternMediterranean andCaucasus, J. Geophys. Res. 105, 5695–5719.

Spatiotemporal Earthquake Clusters along the North Anatolian Fault Zone Offshore İstanbul 1767

Mogi, K. (1963). Some discussions on aftershocks, foreshocks, andearthquake swarms—The fracture of a semi-infinite body caused byan inner stress origin and its relation to the earthquake phenomenon,3, Bull. Earthq. Res. Inst. Tokyo Univ. 41, 615–658.

O’Neill, M. E. Source dimensions and stress drops of small earthquakes nearParkfield, California, 1984, Bull. Seismol. Soc. Am. 74, no. 1, 27–40.

Örgülü, G. (2011). Seismicity and source parameters for small-scale earth-quakes along the splays of the North Anatolian fault (NAF) in theMarmara Sea, Geophys. J. Int. 184, 385–404. doi 10.1111/j.1365-246X.2010.04844.x.

Örgülü, G., and M. Aktar (2001). Regional moment tensor inversion forstrong aftershocks of the August 17, İzmit earthquake (Mw � 7:4),Geophys. Res. Lett. 28, 371–374.

Parsons, T. (2004). Recalculated probability of M ≥ 7 earthquakes beneaththe Sea of Marmara, Turkey, J. Geophys. Res. 109, doi 10.1029/2003JB002667.

Pfister, M., L. Rybach, and S. Simsek (1998). Geothermal reconnaissanceof the Marmara Sea region (NW Turkey): Surface heat flow density inan area of active continental extension, Tectonophysics 291, 77–89.

Pınar, 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.

Şengör, A. M. C., O. Tüysüz, C. İmren, M. Sakınç, E. Eyidoğan, N. Görür,X. Le Pichon, and C. Rangin (2005). The North Anatolian fault: A newlook, Annu. Rev. Earth Planet. Sci. 33, 1–75.

Tchalenko, J. S. (1970). Similarities between shear zones of differentmagnitudes, Geol. Soc. Am. Bull. 81, 1625–40.

Sykes, L. R. (1970). Earthquake swarms and sea-floor spreading,J. Geophys. Res. 75, 6598–6611.

Waldhauser, F., and W. L. Ellsworth (2000). A double-difference earthquakelocation algorithm: Method and application to the northern Haywardfault, Bull. Seismol. Soc. Am. 90, 1353–1368.

Waldhauser, F., and W. L. Ellsworth (2002). Fault structure and mechanicsof the Hayward fault, California, from double-difference earthquakelocations, J. Geophys. Res. 107, no. B3, 2054, doi 10.1029/2000JB000084.

Helmholtz Centre PotsdamGFZ German Research Centre for GeosciencesTelegrafenberg, Haus DPotsdam D-14473, Germany

(F.B., M.B., G.D.)

U.S. Geological Survey345 Middlefield RoadMenlo Park, California 94025–3591

(W.L.E.)

Boğaziçi UniversityKandilli Observatory and Earthquake Research Institute34342 BebekÇengelköy, Istanbul 34684 Turkey

(M.A.)

Manuscript received 6 August 2010

1768 F. Bulut, W. L. Ellsworth, M. Bohnhoff, M. Aktar, and G. Dresen