Electrical characterization of the North Anatolian Fault Zone underneath the Marmara Sea, Turkey by ocean bottom magnetotellurics

The ﬁrst magnetotelluric study in the Marmara Sea, Turkey, was undertaken to resolve the structure of the crust and upper mantle in the region, and to determine the location of the westward extension of the North Anatolian Fault (NAF) in the C¸ınarcık area. Long-period ocean bottom magnetotelluric data were acquired at six sites along two proﬁles crossing the C¸ınarcık Basin, where a signiﬁcant increase in microseismic activity was observed following the devastating 1999 ˙Izmit and D¨uzce earthquakes. 2-D resistivity models indicate the existence of a conductor at a depth of ∼ 10 km in the middle of both proﬁles along with a deeper extension into the upper mantle, implying the presence of ﬂuid in the crust and partial melting in the upper mantle. The northern and southern boundaries of this conductor are interpreted to represent the northern and southern branches of the NAF in the Marmara Sea, respectively. These conductors have been previously identiﬁed farther to the east along the NAF, suggesting that the electrical characteristics of this fault are continuous from onland areas into the Marmara Sea. Microseismic activity in the C¸ınarcık area is located above the conductor documented here, and indicates a possible seismogenic role of crustal ﬂuids present in the conductive zone. In comparison, resistive zones along the NAF may act as asperities that could eventually result in a large earthquake.


I N T RO D U C T I O N
The 1600-km-long North Anatolian Fault (NAF) is an intercontinental dextral strike-slip fault that is located between the northern Eurasian Plate and the southern Anatolian block (Fig. 1a). After closure of the Neo-Tethyan Ocean during the Late Mesozoic and Cenozoic, collision of the Arabian and Eurasian plates resulted in relative westward movement of the Anatolian Plate (McKenzie 1972;Toksöz et al. 1979). This westward movement is considered to be the main cause of major tectonic events along the NAF. During the past century, the epicentres of destructive earthquakes along the NAF have migrated westward starting with the 1939 Erzincan earthquake (M s 7.9) and extending to the 1999İzmit (M w 7.4) and Düzce (M w 7.2) earthquakes (Pınar et al. 2010). The fault rupture of the 1999İzmit (M w 7.4) earthquake extended into the Marmara Sea however, the fault segment between the 1912 Ganos and 1999 Izmit fracture zones (Fig. 1b) has not ruptured since 1766 (Toksöz et al. 1979); this segment in the Marmara Sea is considered to be a 'seismic gap' that may be capable of generating a M ≥ 7 earthquake (Hubert-Ferrari et al. 2000).
The onland outcrop of the NAF is well known as a result of previous geological and geophysical studies (Ketin 1948;Barka  fragment of continental lithosphere that ruptured from Gondwanaland during the Triassic (Yılmaz et al. 1997). The Armutlu-Almacık Zone contains the remnants of the Intra-Pontide suture between thė Istanbul-Zonguldak Zone and the Sakarya Continent, and consists of a tectonic melange of two zones (Fig. 1). All of these tectonic zones form constituent parts of the Western Pontides, a series of east-west trending Tethyan orogenic belts; this means that all three zones keep almost the entire evolutionary record of the Tethysides (Yılmaz et al. 1997). The occurrence of theİzmit and Düzce earthquakes on the northern side of these tectonic zones (Tank et al. 2003(Tank et al. , 2005Kaya et al. 2009;Tank 2012) indicates the significance of the extension of these zones into the Marmara Sea, in terms of possible locations for the next devastating earthquake along the NAF.
However, the juxtaposition of these zones beneath the Marmara Sea and the relationship between these zones and the NAF are currently poorly understood. A number of marine studies undertaken after theİzmit earthquake enabled the formulation of various tectonic models for the Marmara Sea, namely pull-apart , single dextral strike-slip fault (Le Pichon et al. 2001), and extensional, crustal thinning models (Becel et al. 2009). The pullapart model suggests that the Marmara Sea is a large pull-apart basin that includes a number of smaller pull-apart basins formed in a transtensional tectonic regime . This model requires segmented faulting of the NAF within the Marmara Sea. According to Le Pichon et al. (2001), the NAF crosses the Marmara Sea as a single dextral strike-slip fault that follows the northern escarpment of the Ç ınarcık Basin (Ç B) in the east and cuts the Central Basin to the west. In comparison, crustal thinning model suggests the presence of an extensional regime in the Marmara Sea that is dependent on both normal and strike-slip faulting regime (Becel et al. 2009). Crustal thinning has been documented in the southern part of the Central Basin and is also observed beneath theİmralı and Ç Bs (Laigle et al. 2008;Becel et al. 2009). The precise hypocentral distribution of microseismic activity (Bulut et al. 2009) is consistent with the down-dipping structures imaged by Carton et al. (2007) in the Marmara Sea and with present-day deformation controlled by a right-lateral strike-slip regime (Örgülü 2010). Local earthquake tomography (Karabulut et al. 2003;Barış et al. 2005) has imaged low P-wave velocity (Vp) and low Vp/Vs (where Vs = S-wave velocity) ratio zones down to 15 km depth beneath the Marmara Sea, in addition to high Vp and Vp/Vs ratio zones towards the northern and southern edges of the Ç B. Although previous studies have provided valuable information related to the form and tectonism of the NAF, the westward extension of this fault zone and the deeper structure beneath the Marmara Sea remain controversial.
Magnetotellurics (MT) is an electromagnetic method that utilizes naturally occurring electric and magnetic fields to map the subsurface electrical resistivity at depths ranging from the near surface to the upper mantle (Vozoff 1991). Since fluids significantly lower the electrical resistivity of rocks, this technique is highly useful in fault zone investigations (Ritter et al. 2005;Becken et al. 2011). Previous MT research undertaken around active fault zones indicates a strong correlation between the presence of fluids and seismic activity (Unsworth et al. 2000;Ogawa et al. 2001;Ogawa & Honkura 2004;Wannamaker et al. 2009;Becken et al. 2011), with the majority of fault zones being associated with resistor-conductor boundaries, and devastating earthquakes occurring in or around asperity zones identified by local resistive areas spatially associated with zones of low resistivity (Honkura et al. 2000;Oshiman et al. 2002;Tank et al. 2003Tank et al. , 2005Kaya et al. 2009).
Previous onshore MT studies of the NAF have identified conductors at and below a depth of 10 km along the NAF, with northern and southern edges of the conductor coinciding with the surface traces of the NAF. The occurrence of main shocks and major aftershocks within brittle resistive zones (Tank et al. 2003(Tank et al. , 2005Kaya et al. 2009), and earthquake swarm activity around more ductile conductive regions (Tank et al. 2003) emphasizes the importance of fluids during seismic events. Our objective here is to image the western extension of the NAF under the Marmara Sea in an area known to be a seismic gap. Fig. 1b shows the locations of magnetotelluric stations used during this study. We used 12 land sites (yellow triangles) and six ocean bottom sites (yellow squares) in the eastern part of the Marmara Sea, and defined two profiles (P1 and P2) for later 2-D modelling, as identified by dashed blue rectangles in the figure. The NE-SW oriented profile (P1) includes five ocean bottom sites, four of which are located in the Imralı and Ç Bs, together with a single land site in the southern Marmara region. The second N-S oriented profile (P2) is composed of two ocean bottom sites and 11 land sites. The northernmost ocean bottom site is common in both profiles.

DATA
Ocean bottom magnetotelluric sites deployed during this study used ocean bottom electromagnetic instruments (OBEM) developed by Kasaya & Goto (2009). We measured two horizontal electric fields using 4 m dipoles with Ag-AgCl electrodes, and threecomponent magnetic fields using flux-gate sensors. The data were acquired with an 8 Hz sampling rate for about 3 weeks and all OBEM were successfully recovered after data acquisition. Time-series data acquired during deployment were analysed using a robust processing code (Chave et al. 1987) and usable MT transfer functions were obtained for a period range of 10-11 000 s, as shown in Figs 2 and 3.
Land site deployments during this study used broad-band Phoenix MTU5 MT instruments that cover a period range between 0.003 and 2000 s. However, in order to match the ocean bottom MT period range, data from the land MT sites were used at periods >10 s as shown in Figs 2 and 3. One land site at the southwestern end of profile P1 was a new deployment, with the other 11 land sites on profile P2 from Tank et al. (2003). These data were jointly inverted with the new ocean bottom data acquired during this study.
Figs 2 and 3 show the data obtained during this study as sounding curves, incorporating both diagonal and off-diagonal components. Fig. 2 shows observed data sounding curves across profile P1, comprising, from northeast to southwest, ocean bottom sites 101, 102, 103, 104 and 105 together with land site 106. The short period range (< ∼250 s) at sites 101, 102 and 103 contains clear phase rolling out of quadrant (PROQ; Chouteau & Tournerie 2000), with no such phase responses recorded at ocean bottom site 105. These differences can be explained by changes in bathymetry, with site 105 located on a shallow (50-m water depth), flat-lying section of seafloor, whereas the other ocean bottom sites are located near sharp changes in bathymetry. Electric currents in the ocean, perpendicular to bathymetric gradients, can generate secondary magnetic fields that are comparable to or even stronger than the primary magnetic field (Constable et al. 2009;Worzewski et al. 2010). Given this, the modelling discussed here did not incorporate short-period ocean bottom data that were affected by bathymetry-derived PROQ.
The dimensionality of the data set needs to be defined prior to 2-D or 3-D modelling. Here, we use the tensor-decomposition code of McNeice & Jones (2001), an expanded version of the Groom-Bailey decomposition (Groom & Bailey 1989) that incorporates multiple sites and periods, with strike directions inferred from the site-and period-dependent decomposition parameters shown in Fig. 4. The optimum strike directions in the 100-11 000 s period bands were N90 • E and N62 • E for profiles P1 and P2, respectively. These estimated 2-D strike directions were used to decompose the profile data, enabling the definition of TE and TM modes, which represent flows of electric current along and perpendicular to these strike directions.

M O D E L L I N G
We inverted the data by 2-D modelling using the appropriate strike directions defined within the preceding section. Although the strike directions along profiles P1 and P2 differ by some 28 • , we assumed a quasi-2-D structure for the study area.   It is known that TM mode 2-D modelling over a 3-D anomaly can robustly recover the resistivity section under the profile (Wannamaker et al. 1984). We tested such situation for an OBEM data set beneath the Marmara Sea. Fig. 5 shows a simplified 2-D and 3-D resistivity model that mimics the Marmara Sea; this model has a uniform earth of 100 m and a box-shaped ocean of 0.3 m with dimensions 180 km (E-W) × 60 km (N-S) × 1.2 km (depth). Fig. 6 compares the 2-D (Ogawa & Uchida 1996) and corresponding 3-D (Mackie et al. 1994) responses along the profile. The two northern stations (402 and 404) are on land, with the following two ocean bottom stations (406 and 408) close to the coast in the Marmara Sea, respectively. Solid lines denote 2-D TE (red) and TM (blue) responses, with asterisks denoting the corresponding 3-D responses. It is interesting to note that both TE and the corresponding 3-D responses have shorter period PROQ, primarily caused by the strong electrical current in the sea oriented parallel to the coastline, which can generate a large secondary horizontal magnetic field component with the opposite polarity to the primary magnetic field (Constable et al. 2009;Worzewski et al. 2010). The difference in TE and the corresponding 3-D response is evident for both ocean bottom and land sites; in comparison, the TM and the corresponding 3-D responses are in good agreement. The TM responses from the modelled sea are dominated by a galvanic charge build-up at the vertical ocean-land interface, providing a good approximation, even for 3-D structures. Given this, we decided to use TM mode responses that, as shown here, are robust even in 3-D situations.
We used a modified version of the code of Ogawa & Uchida (1996) for 2-D inversion, with modifications detailed below. The Ogawa and Uchida code used Rodi's (1976) algorithm (MOM's method) to calculate spatial derivatives on the ground. This method is used for multiple levels of seafloors in this study. In addition, we modified the definition of roughness norm to include an a priori model in order to stabilize the inversion. The original roughness norm was |Cm| 2 , where m is a model vector representing the log resistivity of the model, and C is the roughening matrix of the Laplacian operator. The modified version has a |C(m − m 0 )| 2 norm, where m o is a vector comprising the log resistivity of the a priori model. The static shift was used as a constraint in inversion as in the original version of Ogawa & Uchida (1996) code.
We started with the initial model construction as described here. Initially, sea water was assigned a fixed value of 0.3 m by referencing the bathymetric data. It is also important to include a sedimentary layer as a priori information, as we do not have the short-period data required to constrain the resistivity of shallow areas immediately under individual ocean bottom sites. During this study, we used the distribution of sediments identified using seismic reflection data (Carton et al. 2007), with Fig. 7 showing the shallow resistivity distribution that was incorporated into the model as a constraint. During modelling, a thick sedimentary layer was assigned a fixed resistivity of 10 m, extending to a depth of 4-5 km below the Ç B (Okay et al. 2000;Carton et al. 2007), with the initial model having a uniform below-sediment resistivity. Error floor values of 10 per cent and 3 • were used for apparent resistivity and phase values, respectively, with the initial model also used as an a priori model.
We formulated three initial models that used uniform belowsediment resistivities of 10, 100 and 1000 m, with the final models dependent on the initial (a priori) models. Of the three initial models, the 100 m model showed the best fit with the data, with root mean square misfit values of 2.1 and 1.9 for profiles P1 and P2, respectively.
The identified 2-D electrical resistivity structures in the eastern Marmara Sea along profiles P1 and P2 are shown in Figs 8(a) and (b). The electrical resistivity models have similar   resistivity distributions beneath both profiles, with a shallow conductor (C1) starting a few kilometres beneath the sedimentary layer and merging with a deeper conductor (C2) in the central part of the profiles. Another shallow conductor (C3) in P1, located between two resistive layers, is also present beneath the Armutlu Peninsula in P2. Both models contain deep resistors that horizontally bound the deep C2 conductor in the north (R1) and in the south (R2). The shallow resistive layer underlain by a conductive anomaly (C3) beneath the Armutlu Peninsula was also observed in the previous MT study by Tank et al. (2003) in which the structures down to almost 20 km were investigated.
A comparison of observed and calculated responses is shown in Figs 9 and 10, for profiles P1 and P2, respectively. These curves  indicate a generally good recovery of the observed data. The apparent resistivity and phase responses are compared in pseudosections in Figs 11 and 12, confirming the recovery of the observed data.
We also tested the main features identified within the final models by forward modelling using resistivity changes. Figs 13 and 14 show sensitivity tests for the major anomalies [R1 (a), C1 (b), C2 (c) and R2 (d)] beneath profiles P1 and P2, respectively. These tests confirm that the modelling accurately represents the data. According to these tests, both conductive anomalies C1 and C2 in both profiles have resistivity values that range between 1 and 10 m.

D I S C U S S I O N
The results of this study provide the first electrical images of structures between the seafloor and the upper mantle beneath the Marmara Sea. The tectonic and geological implications of the major anomalies identified above are discussed in this next section.

Implications for tectonic configurations
The final models given in Figs 8(a) and (b) show three domains; one is consisting of two central subvertical conductors (C1, C2), and others including the surrounding resistors (R1, R2). This distribution is consistent with the known tectonic provinces in the study area. The northern resistor (R1) and the overlying 10-km-thick conductor which belong to theİstanbul-Zonguldak Zone represent Precambrian bedrock and Ordovician to Carboniferous sediments (Yılmaz et al. 1997). The southern resistor (R2) corresponds to the Sakarya and Armutlu zones, and represents Paleozoic metamorphic rocks of the Sakarya Continent. The zone with the subvertical conductive anomaly corresponds to the collision zone between theİstanbul-Zonguldak Zone and the Sakarya Continent.

Upper crustal conductor (C1)
An upper crustal shallow conductive anomaly, here named C1, is present in the middle of both profiles with a resistivity of 1-10 m. Experimental data indicate that electrical resistivity of aqueous crustal fluids is in the range of 0.01-0.1 m (Nesbitt 1993); meaning that, using the Hashin-Shtrikman upper bound (Hashin & Shtrikman 1962), a bulk resistivity of 1-10 m can be explained by porosity of 0.15-15.0 per cent. In addition, a seismic tomography study detected low Vp and Vp/Vs ratio zones at depths of 5-15 km beneath the Ç B (Barış et al. 2005). These zones were also interpreted as areas of high-fluid content. The distribution of microseismic epicentres around the Ç B is shown in Fig. 1(b). In Fig. 8, projected hypocentres of earthquakes that were at a distance of ±10 km from the profiles P1 and P2 are given. This projection indicates a good correlation between resistivity and seismicity, with the majority of the microseismic activity clustering outside the rims of the C1 conductor beneath the profile P1, and close to the C1 and C3 conductors beneath the profile P2 (Fig. 8).
This configuration can be explained by the existence of an interconnected fluid network and the associated triggering of earthquakes by the migration of fluids into the surrounding crust. Migration of fluid into less-permeable crust can reduce the effective normal stress and trigger earthquakes (Sibson et al. 1988;Cox 1999;Sibson 2000). Such seismicity-resistivity relationships are known in a number of seismically active regions Ogawa & Honkura 2004;Wannamaker et al. 2004Wannamaker et al. , 2009Jiracek et al. 2007;Mitsuhata et al. 2001).

Lower crust to upper-mantle conductor C2
The research discussed here identified a vertical conductor, here named C2, runs from the lower crust to the upper mantle. This conductor may be associated with the presence of high-salinity fluids and/or the partial melting of mantle material due to the asthenospheric upwelling. The bulk resistivity (1-10 m) of the C2 zone indicates either 0.15-15.0 per cent fluid or 1.5-39 vol per cent melt fraction by the Hashin-Shtrikman upper bound, using pure melt and aqueous fluid resistivities of 0.1-0.3 and 0.01-0.1 m, respectively (Presnall et al. 1972;Waff 1974;Tyburczy & Waff 1983;Nesbitt 1993;Yoshino et al. 2010;Pommier & LeTrong 2011;Evans 2012).
The presence of partial melt is supported by the upwelling of asthenospheric material by other magnetotelluric studies documented for the onland extent of the NAF (Gürer 1996;Tank et al. 2005;Türkoglu et al. 2008). In addition, seismological studies indicate that the crustal thickness around the Marmara Sea varies between 29 and 32 km (Gürbüz et al. 2003;Zor et al. 2006). Within the Marmara Sea, it is almost 31 km (grey dashed line in Fig. 8), except areas to the south of the Central Basin and beneath the Ç ınarcık andİmralı basins. Here, the crustal thickness decreases to 26 km due to the thinning of the upper crust associated with lithospheric-scale extension (Laigle et al. 2008;Becel et al. 2009). In addition to that, Straub & Kahle (1994) documented a NE-SW oriented extensional regime along the northern branch of the NAF, in the present study area. Furthermore, the presence of re-gions with high heat flow (100-140 mW m −2 ) in the Marmara Sea (İlkışık 1995;Tezcan 1995) and mantle-derived He (>50 per cent of total He concentration) along the west to central segment of the NAF (Güleç et al. 2002) may indicate the presence of zones with significant extension and high temperature that may be related to the upwelling of asthenospheric material. Dilek & Altunkaynak (2007 suggested partial melting as a source of the Eocene volcanism in and around the Marmara Sea; as well, Altunkaynak (2007) showed that the geochemistry of the Marmara granitoids was indicative of significant crustal contamination during magma ascent.
These data suggest that fluids are migrating from a deep ductile region to the upper brittle zone beneath the Marmara Sea. If this conductor does represent partial melting, one possibility is that the high temperatures, partial melting and fluid-releasing dehydration reactions documented in this area may be related to asthenospheric upwelling caused by subduction of Neo-Tethyan oceanic lithosphere beneath the Sakarya Continent, which was followed by slab breakoff.
The form of the deep C2 conductor changes beneath both profiles, with the conductor appearing narrower, but with a higher resistivity, beneath profile P2 than beneath profile P1. One possible reason for this is that profile P2 crosses the easternmost part of the Ç B and the Armutlu Peninsula, and does not intersect structures beneath theİmralı Basin which is associated with highly conductive aqueous fluids and/or molten crustal and mantle material. The basement rocks of the Armutlu Peninsula consist of a succession of highand low-grade metamorphics overlain by sedimentary rocks-a similar lithological sequence to that found within theİstanbul-Zonguldak units and Sakarya Continent (Yılmaz and Tüysüz 1991;Yılmaz et al. 1995). This indicates that profile P1 includes structures beneath both theİmralı and Ç Bs, leading to a wider conductive anomaly than observed in the profile P2.
Deep subvertical conductors beneath large strike-slip faults have also been recently imaged along the NAF (Tank et al. 2005), the Alpine Fault in New Zealand (Wannamaker et al. 2002) and the San Andreas Fault (Becken et al. 2011), suggesting that these conductors may be a common feature in areas containing major strike-slip faults. The existence of this type of conductor extending to 50km depth below the NAF nearİzmit was documented by Tank et al. (2005), who interpreted the deeper part of the conductor to be a region of partial melt, and the shallow part (just under the brittle-ductile transition) to be a region containing saline fluids. A similar interpretation, combining regions of fluid and underlying partial melt, was also provided in Tibet by Li et al. (2003). Wannamaker et al. (2002) and Becken et al. (2011), on the other hand, showed that deep vertical conductor is fluid source supplying crustal fluid.

Continuous tectonic features along the NAF
We have documented significant similarities between the features resolved in MT models obtained for the Marmara Sea area, and in models for the region to the east of the study area along the NAF. The Marmara Sea profiles, P1 and P2, are dominated by a conductor that is horizontally bounded by resistive zones to the north and south (Fig. 15). This structure was also observed in previous onshore MT studies in the eastern part of the present study area along the NAF (Gürer 1996;Honkura et al. 2000;Oshiman et al. 2002;Tank et al. 2005;Kaya et al. 2009;Kaya 2010). The distribution of geothermal fields from east to west along the NAF (Aydın et al. 2005) is also consistent with the location of the deep conductors, suggesting that similar structures may also be present along the western part of the NAF in the Marmara Sea area. In terms of extending geoelectrical structures from the eastern Marmara region to the Ç B, we suggest that, from north to south, the Istanbul-Zonguldak and Armutlu-Almacik zones and the Sakarya Continent tectonic zones are continuous from the previously documented onland areas to the area beneath the Marmara Sea.

Branches of the NAF
Previous MT studies identified a correlation between resistorconductor boundaries and onland branches of the NAF (Tank et al. 2003(Tank et al. , 2005Kaya et al. 2009). In this respect, subvertical resistorconductor boundaries beneath the Marmara Sea may also indicate the location of NAF branches, suggesting that the northern resistorconductor boundary in the study area represents a northern branch of the NAF (NAF1) that extends west from theİzmit earthquake rupture zone to the Marmara Sea. Fig. 15 shows both extension of the NAF branches and resistivity distribution along the NAF from Düzce region to the Marmara Sea.
The second boundary, located at the southern escarpment of the Ç B beneath profile P2, is also observed beneath profile P1, suggesting that this minor branch of the NAF extends farther to the west. The boundary at the southern part of the C2 zone, beneath profile P1, implies the existence of another branch of the NAF, as shown by a dashed line in Figs 1 and 15. This has often been referred to as the middle branch though it is one strand of the southern branch of the NAF (Yilmaz et al. 2010). Here we refer to this branch as NAF2 which means the extension of the southern branch of the NAF in the Marmara Sea (Fig. 15). Extension of these subvertical resistorconductor boundaries to at least 50 km depth suggests deeply rooted NAF that is also supplied by a teleseismic tomography study indicating a P-wave velocity contrast, represented by relatively high P-wave velocity perturbations (δVp) to the north of the NAF and low P-wave velocity perturbations to the south, down to a depth of 150 km in the Marmara Sea (Biryol et al. 2011). In order to improve our knowledge of the extension of these structures farther to the west, more OBEM instrument data from offshore areas to the west are required.
Resistivity structures defined around the rupture zones of the 1999İzmit and Düzce earthquakes showed that the main shocks occurred within high-resistivity zones that were bounded to the south by lower-resistivity zones (Tank et al. 2003;Kaya et al. 2009). This suggests that along the NAF, comparisons between resistive structures beneath the Marmara Sea and the epicentres of the 1999 earthquakes may be crucial in relating high-resistivity zones to possible asperity zones that may initiate a large devastating event within the Marmara Sea, which is an area of increased seismicity after theİzmit earthquake. This indicates that more MT research needs to be undertaken on the resistive zone at sites farther to the north in the Ç B.

Tectonic models under the Marmara Sea
The single dextral strike-slip fault model in the Marmara Sea suggests westward extension of the northern branch of the NAF along the northern escarpment of the Ç B (Le Pichon et al. 2001) corresponding to the northern resistor-conductor boundary in our final models. In order to be able to support or decline continuation of a single fault in this model, more OBEM instrument data from offshore areas to the west are needed to figure out if there exists either many or a single resistivity boundary corresponding to the fault branch. The pull-apart model, on the other hand, suggests segmented faulting of the NAF due to both normal and strike-slip faulting regimes in the Marmara Sea . The location of the branches of the NAF in the eastern Marmara Sea and extensional regime suggested by this model are supported by our resistivity models although it seems from our models that extension is not strictly limited to crustal scale as suggested by this model. However, the crustal thinning model suggests both strike-slip and normal faulting where branches are consistent with resistivity boundaries in our models, and lithospheric scale extension in the Marmara Sea. The continuation of deep conductive material to shallow depths, as indicated by our electrical resistivity modelling, supports ongoing crustal thinning beneath the eastern Marmara Sea.

C O N C L U S I O N S
We performed 2-D modelling of MT data collected at both offshore and onshore sites along the western part of the NAF. Electrical resistivity models determined by 2-D inversions of TM mode data contain some characteristics related to the NAF. The presence of a conductor (C2) bounded by resistive blocks (R1 and R2) is similar to that previously documented in profiles crossing theİzmit earthquake rupture zone, and characterizes the western extension of the NAF from the east to the Marmara Sea. The resistor-conductor boundaries observed to the north and south of C2 correspond to the northern and southern branches of the NAF in the Marmara Sea, respectively. The shallow conductor C1, the presence of which implies the existence of a zone of interconnected pore fluid, seems to be related to microseismic activity through movement of fluids towards a more resistive seismic zone, whereas the deep conductor C2, which is probably related to the presence of fluid originating from dehydration and partial melt derived from upwelling of hot asthenospheric material, may be the source of a shallow interconnected fluid zone (i.e. C1). Resistive zones near resistive-conductive boundaries are potential sites for large earthquakes along branches of the NAF; in particular, the northern branch of the NAF may be likely to generate future large earthquakes beneath the Marmara Sea and as such requires further study.

A C K N O W L E D G E M E N T S
This study was supported by KAKENHI (19253002). TK was supported by Monbukagakusho (Japanese Government) Scholarship. SBT was partly supported by the JSPS-PD fellowship program. We thank the Office of Navigation, Hydrography and Oceanography of the Turkish Naval Forces, and Turkish Airlines for their assistance during this study. We are grateful to the Marine Research Coordinating Office of the General Directorate of Mineral Research and Exploration (MTA) for providing bathymetry data for the Marmara Sea. We thank Dr. Fatih Bulut (GFZ) for sharing his high-resolution microseismic locations in and around the Ç B, and are grateful to Tadanori Goto and Elif Tolak for his help during instrument preparation. We also thank two anonymous reviewers and Dr. Gary Egbert for their valuable comments and contributions to this paper. Some Figure 15. Upper panel: map showing resistivity variation at ∼15 km depth across and along the NAF from the 1999İzmit and Düzce earthquake rupture zones in the east to the Marmara Sea in the west. Red stars denote epicentres of theİzmit and Düzce earthquakes, blue lines indicate MT profiles and hot-coloured rectangles show the location of the conductors. Lower panel: from left-to right-hand side, 2-D resistivity models generated during this study and for two profiles crossing theİzmit and Düzce ruptures; resistivities are plotted in the same colour range. The presence of the middle conductor bounded by resistive zones from east to west in all profiles is indicative of the continuation of the NAF. NAF1, NAF Northern branch; NAF2, NAF Southern branch. of the figures presented here were generated using the Generic Mapping Tools (GMT) code developed by Wessel & Smith (1995).