The lower boundary of the lithospheric plate is a very important parameter for understanding plate tectonics, although it is still one of the less known quantities. Our current knowledge of the lithospheric thickness beneath Turkey relies on studies that examined the data from several temporary and permanent seismic networks (e.g., Angus et al., 2006; Sodoudi et al., 2006, 2015; Gök et al., 2007, 2015; Vanacore et al., 2013; Vinnik et al., 2014). Interpretations from these studies are either confined to a limited region or to a limited depth extent, i.e., to crustal depths only. Thus, the variations of lithospheric thickness have not yet been homogeneously characterized in Turkey and surroundings. A robust estimate of the lithospheric thickness is an important constraint for our understanding of mantle deformation in response to tectonic forces (see Fischer et al., 2010, and Jones et al., 2010, for reviews). In the present study, we provide a complete image of the LAB topography below Anatolia and adjacent regions based on a combination of data from several different networks in the area. To achieve this, we have employed the S-receiver function technique, which is particularly suited to identify seismic discontinuities in the upper mantle, especially when low-velocity zones are involved.

The Cenozoic closure of the Tethys Ocean and the following continental collision of the northward-moving African and Arabian plates with the Eurasian plate are considered to be the main driving processes of the present tectonic setting in the eastern Mediterranean and Anatolia (Şengör and Kidd, 1979; Dewey and Şengör, 1979; Taymaz et al., 1990, 1991a,b; Jackson et al., 1992; Armijo et al., 1999; Taymaz et al., 2004; Faccenna et al., 2014; Schildgen et al.; 2014). Figure 1 summarizes the tectonic setting of the target area. GPS measurements have highlighted that the retreating Hellenic trench and associated slab roll-back of the subducting African lithosphere are currently the major driving forces for the western movement of the Anatolian plate (e.g., McKenzie, 1978; Reilinger et al., 2006). Current deformation is mainly accommodated through strain localization along the North Anatolian Fault (NAF) and the East Anatolian Fault (EAF). The deformation along the EAF is purely left lateral with no compressional component except for localized thrust ridges associated with strike-slip tectonics (Bulut et al., 2012). Body wave tomography images on global and regional scale indicate that the Hellenic subduction penetrates down to the lower mantle with an average dip of about 40∘ (Wortel and Spakman, 2000; Faccenna et al., 2006; Biryol et al., 2011; Salaün et al., 2012) while a recent waveform tomography study by Fichtner et al. (2013a, b) did not report any deep slab.

Crustal thickness ranges from 35 to 40 km along the North Anatolian Fault and shallows to 25–30 km in the Sea of Marmara region based on receiver functions and controlled source studies (Saunders et al., 1998; Sato et al., 2004; Zor et al., 2006; Laigle et al., 2008; Özacar et al., 2008; Bécel et al., 2009, 2010; Vanacore et al., 2013; Karabulut et al., 2013; Sodoudi et al., 2015). The lithosphere–asthenosphere boundary beneath the East Anatolian Accretionary Complex (EAAC) has been observed at 60–80 km depth based on S-receiver functions (Angus et al., 2006). They consider this lithospheric thickness to be anomalously thin and interpret this structure to be the remnant of the detachment of an oceanic slab. Further evidence for a thin lithosphere overlying hot and partially molten asthenospheric material beneath eastern Anatolia has been found in other P- and S-receiver function images (Zor et al., 2003; Özacar et al., 2008; Vanacore et al., 2013). Other indications of a shallow LAB are strong attenuation of Lg and Sn phases (Gök et al., 2003) and relatively low Pn- and uppermost mantle S-velocities (Al-Lazki et al., 2004, Maggi and Priestley, 2005). The slow velocity anomaly was attributed to the ascending asthenosphere that is emplaced beneath the plateau following the detachment of the northward-subducting Arabian oceanic lithosphere (Keskin, 2003; Faccenna, 2003; Şengör et al., 2003; Biryol et al., 2011; Fichtner et al., 2013a, b). Recently, tomography studies by Biryol et al. (2011) and Fichtner et al. (2013a) have confirmed that a hot and buoyant asthenospheric body supports the ∼ 2 km elevation of the Eastern Anatolian Plateau in the presence of an about 45 km thick crust (Şengör et al., 2003; Zor et al., 2003). At about 350 km depth, a fast anomaly has been found by tomographic studies and interpreted to represent a slab detachment (Lei and Zhao, 2007; Zor, 2008).

Recent extensions of seismic and geodetic networks in Turkey have considerably contributed to our understanding of seismicity patterns and crustal and lithospheric structures beneath Anatolia and the surrounding regions. These developments have been triggered in particular by the destructive 1999 Izmit (Mw7.4) and Düzce (Mw7.2) earthquakes. Our data set consists of teleseismic waveforms that were extracted from 1028 earthquakes recorded at 195 broadband stations (Fig. 2), operated by Kandilli Observatory and Earthquake Research Institute (KOERI; Kalafat et al., 2008), NOA-Net, Aristotle University Thessaloniki (HT-Net) and GEOFON (Hanka and Kind, 1994; GEOFON Data Centre, 1993).

We apply the S-receiver function method to this data set in order to image seismic discontinuities in the crust and upper mantle. A description of the S-receiver function analysis scheme and examples are found in Yuan et al. (2006) and Kind et al. (2012, 2014). The method employs teleseismic S phases, which are converted to P waves at a discontinuity below the station site.The main advantage of this approach over the P-receiver function technique is that there is practically no interference from multiples, which can otherwise significantly interfere with the imaging of deeper convertors, in particular for the depth range of 80–200 km in continental regions. S-receiver functions provide a broad sampling of the upper mantle comparable to the steep angle reflection images of the crust. This technique requires a sufficient number of high-quality observations of teleseismic S phases, typically at least about 60 records per station with signal-to-noise ratio above two. Permanent broadband stations or densely deployed temporary broadband networks can usually provide this amount of data. The limit of resolution is determined by the frequency content of teleseismic S phases, which usually only give convincing results for periods longer than a few seconds. Since the essential part of S-receiver functions consists of weakly converted P-precursors of the SV phase, the signal-to-noise ratio needs to be improved by stacking large number of records at the same station, or within a confined region of conversion points at a fixed depth, or by migrating the rays into the depth domain based on a 1-D reference model and the assumption that conversions occur at horizontal boundaries. The latter approach is known as a common-conversion-point stack (CCP). To reduce source side effects, the distribution of sources should have a well-balanced azimuthal and epicentral coverage. In general, events at epicentral distances between 60∘ and 85∘ are suitable for S-receiver function studies. Reviews of the application of the S-receiver function technique in investigations of the lithosphere–asthenosphere boundary are found in Rychert and Shearer (2009), Fischer et al. (2010) and Rychert et al. (2010). More recently, further studies of upper-mantle discontinuities have been published by Levander and Miller (2012), Kumar et al. (2012a, b), Hopper et al. (2013), Kind et al. (2013) and Sodoudi et al. (2013).

We have restricted the data set by visual inspection to reasonably simple and clear waveforms. Deconvolution results that fail to transform the SV signal (Q component) into a sharp spike have also been excluded based on a further visual inspection. This resulted in 11 660 S-receiver functions. For initial evaluation, all traces are combined within non-overlapping 0.5∘ epicentral distance bins irrespective of the station and source locations (Fig. 3). This figure shows all observable precursors of the S phases, including phases that might disturb the expected S-to-P conversion from upper-mantle discontinuities below the stations. A positive conversion (red) at -3 s, corresponding to the Moho, and a negative (blue) signal at -10 s, interpreted as the LAB, are clearly visible. We can also identify the SKS660p and ScS660p phases caused by S-to-P conversions of SKS and ScS at the 660 km discontinuity. These signals have significantly different slowness compared to the S-to-P conversions of the S phase at the LAB or Moho, meaning that they cut across all other phases before the S signal. It means that, if we apply proper delays before summation according to the slowness of the converted S phase, all other phases traveling with different slowness are suppressed. The strong signal following S is the SPn phase, which is caused by an S-to-P conversion traveling horizontally as a P wave through the uppermost mantle layer at the receiver side.

Since we expect to image in the south of the study region an inclined slab, we first use synthetic seismograms to quantify the magnitude of artifacts which may occur during the depth transformation of the time domain records and infer the maximum range of inclination at which the LAB of a slab can be imaged using S-receiver functions. Figure 4a–f show synthetic seismograms calculated with the RAYSUM software (Frederiksen and Bostock, 2000), which is capable of calculating receiver functions for inclined layers. In these figures, the converters used for the forward modeling are displayed by black lines. Positive and negative amplitudes of the synthetic S-RFs (S-receiver functions) are shown by red and blue, respectively. Based on available reference tomographic models (e.g., Biryol et al., 2011; Bakirci et al., 2012), an inclined 45 km thick high-velocity zone (slab) dipping to the north is used as the input model. The modeling has been performed for inclination angles between 0∘ and 50∘ with an increment of 10∘. The velocity contrasts are from 8.0 to 7.5 km s-1 at the lower boundary of the slab and from 6.5 to 8.0 km s-1 at its upper boundary, as depicted in Fig. 4a. S-RF traces are modeled for back azimuths from 0∘ to 180∘ with an increment of 10∘ and for a constant slowness of 9.8 s degree-1, which corresponds to an epicentral distance of 80∘. Events with back azimuths from 180∘ to 360∘ would yield identical information to that obtained for the range 0–180∘ due to the 180∘ periodicity of RFs for inclined structures. Figure 4g shows the incidence angle of the incoming S wave at the lower boundary of the dipping slab as a function of the back azimuth. For zero and small incidence angles of the high-velocity zone, the discontinuities are properly imaged at expected positions (see Fig. 4a, b). A signal is generated by events from nearly all azimuths. For a slab inclination greater than 20∘, the lower boundary of the slab is imaged only by events from southern directions since for back azimuths of 90∘ the emergent angle of the converted Sp wave reaches its critical value (see green line in Fig. 4g). Thus, for events from northern azimuths (0–90∘), no transmitted Sp conversion exists. For larger inclinations of the slab, the back azimuthal range for which Sp converted energy is transmitted becomes even smaller (see Fig. 4d–f and red, yellow and violet lines in Fig. 4g). Moreover, the amplitudes from the inclined structure show a reversal for slab inclinations larger than 30∘ and events from southern directions (160–180∘). This amplitude reversal occurs similarly for the Q-component of P-RFs from inclined structures (Schneider et al., 2013) and might cause destructive interference in CCP stacks.

We extracted several north–south and east–west profiles through Anatolia. The results from time domain stacks and depth transformations are shown in Figs. 5–10. The receiver functions are filtered with an 8 s low-pass filter in those figures. In the time domain, a moveout correction is applied using a reference slowness of 6.4 s degree-1 and time-corrected traces are stacked for S-to-P piercing points at 200 km depth within a certain window of latitude and longitude along the selected profiles. We chose the piercing point depth of 200 km in the time domain images in order to focus on potentially subducting structures. The summation windows have been placed along east–west and north–south profiles. The size of the windows is 0.3∘ in the north–south (or east–west) direction and the width of the section in the east–west (or north–south) direction given in each figure caption (e.g., Fig. 5). Neighboring boxes are not overlapping: no lateral smoothing is applied along the profiles. The relatively large profile widths (3–9∘) were required to collect a sufficient number of traces to ensure a good signal-to-noise ratio of the originally weakly converted phases in the summation trace. In order to obtain an image directly in the depth–distance domain, we also calculated CCP stacks, which, unlike the time domain summation, does not require the arbitrary choice of a piercing point depth. We assumed the IASP91 global reference model for the time-to-depth conversion. The receiver function depth transformation technique does not depend strongly on the given velocity model. The same profile width as in the time domain technique is used in the depth transformation technique, and all rays traveling within the profile width at any depth are summed. This means that, generally, more traces contribute to each profile in the depth domain because for the time-domain images only rays with piercing points in a particular depth are summed.

The depth of the LAB beneath the entire region of Anatolia varies between 80 and 100 km, and in particular does not change when crossing the North Anatolian and East Anatolian fault zones. This indicates that the thickness of the lithosphere is similar beneath Anatolia and the neighboring edges of the Eurasian and Arabian plates. An exception is the lithospheric thickness near 150 km in the northern Aegean (Sodoudi et al., 2006). In southern Anatolia, the subducting African lithosphere (LAB) between the Aegean and western Cyprus is imaged down to a depth of about 150 km, dipping to the north and east. Synthetic seismogram calculations have shown that a steeper-dipping deeper slab would not have been imaged by S-receiver functions, however. To the east of Cyprus a shallow LAB is observed (similar to the Arabian LAB) with apparently no indication of African subduction.