Crustal structure of the Gulf of Aden southern margin: Evidence from receiver functions on Socotra Island (Yemen)

Breakup of continents in magma-poor setting occurs primarily by faulting and plate thinning. Spatial and temporal variations in these processes can be in ﬂ uenced by the pre-rift basement structure as well as by early syn-rift segmentation of the rift. In order to better understand crustal deformation and in ﬂ uence of pre-rift architecture on breakup we use receiver functions from teleseismic recordings from Socotra which is part of the subaerial Oligo-Miocene age southern margin of the Gulf of Aden. We determine variations in crustal thickness and elastic properties, from which we interpret the degree of extension related thinning and crustal composition. Our computed receiver functions show an average crustal thickness of ~28 km for central Socotra, which decreaseswestward alongthemarginto anaverageof~21km.Inaddition,thecrustthinswith proximity to the continent – ocean transition to ~16 km in the northwest. Assuming an initial pre-rift crustal thickness of 35 km (undeformed Arabian plate), we estimate a stretching factor in the range of ~2.1 – 2.4 beneath Socotra. Our results show considerable differences between the crustal structure of Socotra's eastern and western sides on either side of the Hadibo transfer zone; the east displays a clear intracrustal conversion phase and thick crust when compared with the western part. The majority of measurements across Socotra show V p / V s ratios of between 1.70 and 1.77 and are broadly consistent with the V p / V s values expected from the granitic and carbonate rock type exposed at the surface. Our results strongly suggest that intrusion of ma ﬁ c rock is absent or minimal, providing evidence that mechanical thinning accommodated the majority of crustal extension. From our observationswe interpret that the western part of Socotra corresponds to the necking zone of a classic magma-poor continental margin, while the eastern part corresponds to the proximal domain. to provide data coverage for the entire island of Socotra. We use teleseismic waveformdata from ournetwork toquantifyfor the ﬁ rst time the variation of the crustal thickness and elastic properties of Socotra. This study provides data and interpretation on the evolution of the whole crust of the southern Aden margin. This enables an interpretation of the breakup history in the context of the better-known northern conjugate margin in Oman, leading to an understanding of rift evolution. We use our results beneath Socotra to better understand plate scale deformationand compositional modi ﬁ cationthus improving our understanding of the breakup history of the youngest non-volcanic passive continental margin on Earth.


Introduction
The eastern Gulf of Aden continental margins are magma-poor oblique-rifted margins (Fig. 1a) (Bellahsen et al, 2013a;Leroy et al, 2010a). As such, long fracture zones generated by transform faults impose a first order control on the along-rift segmentation of the young ocean basin (Bellahsen et al, 2013b;Leroy et al., 2012). The Arabian plate is thought to have broken away from the Somalian plate at 20-18 Ma, with the young passive rifted conjugate margins still largely exposed subaerially (d' Acremont et al., 2010;Fournier et al., 2010). This makes it easy to interpret the extensional and sedimentary processes active during continental breakup. Post-rift sedimentary cover is generally relatively thin (0.5-1 km) (d' Acremont et al., 2005;Leroy et al., 2004) and seismic reflection profiles show an absence of seaward dipping reflectors (Autin et al., 2010a,b;d'Acremont et al., 2005;Leroy et al., 2010a,b) making syn/post-rift sediment and basement easily identifiable and mappeable. The structure of the margin is therefore interpretable on upper-crustal and crustal scale cross sections d'Acremont et al., 2005d'Acremont et al., , 2006Leroy et al., 2004Leroy et al., , 2010aLeroy et al., , 2012Tiberi et al., 2007).
Magma-poor rifted margins show a characteristic architecture within three distinct divisions from continent to ocean: the proximal, distal, and ocean-continent transition domains. The proximal domain is characterized by rift-basins bound by high-angle listric faults with Tectonophysics 637 (2014) [251][252][253][254][255][256][257][258][259][260][261][262][263][264][265][266][267] limited crustal thinning (β b 2) (e.g. Mohn et al., 2012). The crust of the distal domain is extremely thin with listric detachment faults commonly penetrating the whole crust. This differs from the proximal domain where fault does not cut through the whole crust (Franke, 2013;Péron-Pinvidic and Manatschal, 2008;Whitmarsh et al., 2001). Mohn et al. (2012) defined a narrow necking zone between proximal and distal domains. Necking of the continental crust occurs in a narrow transition zone from slightly stretched and thinned crust (~30 km) to extremely thinned crust (b10 km) with β N 2 (Mohn et al., 2012). The ocean-continent transition (OCT) is defined as the domain of exhumed sub-continental mantle (Franke, 2013;Mohn et al., 2012) if it has been sampled. If no samples are available then the OCT is defined as a domain of uncertain affinity between the thinned continental crust and the crust that displays clear oceanic geophysical characteristics (see Leroy et al., 2010a for details). Hence, crustal thickness variation along rifted margins from the proximal domain to the distal domain is a key parameter required to assess structural segmentation and the strain localization during the transition from continental break-up to active oceanic spreading.
During the last few decades, detailed geophysical and geological studies have been carried out within the Gulf of Aden to evaluate the initiation and development of the active spreading ridge (Sheba ridge) ( Fig. 1a and b). These studies have provided bathymetric (Leroy et al., 2004), seismic Bache et al, 2011;d'Acremont et al., 2005Laughton et al., 1970;Leroy et al., 2004Leroy et al., , 2010aStein and Cochran, 1985;Watremez et al., 2011), gravity, magnetic (Ali and Watts, 2013;d'Acremont et al., 2006;Fournier et al., 2010;Leroy et al., 2010a;Leroy et al, 2012), and heat flow (Lucazeau et al., 2008(Lucazeau et al., , 2009(Lucazeau et al., , 2010Rolandone et al, 2013) constraints. The interpretation of these data has resulted in a detailed model of the evolution of the eastern Gulf of Aden from rifting to oceanic spreading. In addition to these marine geophysical studies, detailed onshore stratigraphical, geological, structural, and geophysical studies have been carried out on the northern continental margin of the Gulf in eastern Yemen and Oman Al-Hashmi et al., 2011;Al-Lazki et al., 2012;Basuyau et al., 2010;Bellahsen et al., 2006Bellahsen et al., , 2013aChang and Van der Lee, 2011;Corbeau et al, 2014;Denèle et al, 2012;Huchon and Khanbari, 2003;Korostelev et Acremont et al., 2006;Leroy et al., 2010aLeroy et al., , 2012. c) Polar projection shows the distribution of seismic events, small circle interval is 30°. SSFZ: Shukra el Sheik fracture zone; AFFZ: Alula Fartak fracture zone; SHFZ: Socotra Hadbeen fracture zone; DSTF: Dead Sea transform fault, Owen TF: Owen transform fault. Platel and Roger, 1989;Platel et al., 1992;Robinet et al., 2013;Tiberi et al., 2007). These studies reveal extension directions of N20°E parallel to the plate motion and N160°E perpendicular to the main direction of the oblique rift Brune and Autin, 2013) and thinning of the crust from a thickness of 35 km at 30 km inland from the coast to a thickness of 26 km at the coast. Unfortunately, it has not been possible to carry out similar studies along the southern margin of the Gulf of Aden, where the surface geology investigation was previously limited to field mapping of the Socotra archipelago (e.g., Beydoun and Bichan, 1969;Samuel et al., 1997;Fournier et al., 2007;Leroy et al., 2012;Denèle et al., 2012;Bellahsen et al., 2013b;Pik et al., 2013), a shallow marine seismic reflection line on the southern Socotra platform Morrison et al., 1997) and field mapping in Somalia (e.g., Brown, 1931;Fantozzi, 1996;Fantozzi and Sgavetti, 1998;Granath, 2001). Socotra island constitutes an important part of the southern margin located at the easternmost part of the Gulf of Aden (Fig. 1b). It documents the full geological and tectonic history of rifting, including a record of the pre-rift continental architecture that existed when it was still attached to Arabia, as well as the complex geological and tectonic syn-rift stage and post-rift geological and tectonic processes . These previous studies do not include imaging of the deep crustal structure that is critical to evaluate crustal scale deformation history and evolution during rifting.
With this purpose in mind, the Socotra component of the YOCMAL (Young Conjugate Margins Laboratory) seismic project was conceived and implemented during 2010-2012. YOCMAL deployed 24 new broadband seismic stations to supplement the single permanent seismic station to provide data coverage for the entire island of Socotra. We use teleseismic waveform data from our network to quantify for the first time the variation of the crustal thickness and elastic properties of Socotra. This study provides data and interpretation on the evolution of the whole crust of the southern Aden margin. This enables an interpretation of the breakup history in the context of the better-known northern conjugate margin in Oman, leading to an understanding of rift evolution. We use our results beneath Socotra to better understand plate scale deformation and compositional modification thus improving our understanding of the breakup history of the youngest non-volcanic passive continental margin on Earth.

Geological and tectonic setting
Socotra is the largest of four islands (Socotra, Darsah, Samhah and Abd al Kuri) constituting the Socotra Archipelago located east of Somalia and representing the southeastern Gulf of Aden margin. Geographically Socotra is~380 km south of Ras Fartaq in mainland Yemen and~250 km east of the Horn of Africa (Fig. 1).
The greatest depth to the seafloor surrounding the islands is 2500 m, north of the archipelago. The~70 km wide Socotra Platform to the south of the islands is at water depths of less than~250 m. South of the Socotra Platform the seafloor drops to~5000 m in the North Somali Basin   (Fig. 1).
The continental rifting which eventually formed the Gulf of Aden conjugate margins began as early as the Oligocene (~34 Ma) with the formation of rift valley grabens (d' Acremont et al., 2005Acremont et al., , 2006Acremont et al., , 2010Leroy et al., 2010aLeroy et al., , 2012Robinet et al., 2013). Beginning in the early Miocene (~17.6 Ma), oblique seafloor spreading with a mean extension direction of N20°E began to separate the two N75°E trending conjugate margins (Leroy et al., 2004;Bellahsen et al. 2006;d'Acremont et al., 2006Leroy et al., 2010aLeroy et al., , 2012 (Fig. 1). Continental breakup resulted in distinct eastern and western structural domains of Socotra, separated by the NE-SW striking Hadibo transfer zone (HTZ) located through the middle of the island ( Fig. 2; Beydoun and Bichan, 1969;Leroy et al., 2012;Bellahsen et al., 2013b). The eastern domain is composed of a single large tilted block dominated by igneous rocks which forms the~1500 m high Haggier Mountain Range. The western domain is characterized by tilted blocks bound by~N110°E trending faults (Bellahsen et al., 2013b;Leroy et al., 2012).
The pre-rift geology of Socotra is a neoproterozoic basement and Cretaceous-Eocene carbonates. The Neoproterozoic basement, which is composed of metamorphic, plutonic and volcanic rocks , outcrops over more than 30% of the eastern domain in the Haggier Mountains (Fig. 2). In the western domain, basement consisting mainly of metamorphic and biotite granite rocks outcrops at Qalansiya and Sherubrub (Fig. 2). Pre-rift carbonatedominated Cretaceous strata were deposited uncomformably over the basement, followed by Paleocene/Middle Eocene carbonate shelf deposits (Beydoun and Bichan, 1969;Leroy et al., 2012).
The pre-rift strata were disturbed by the onset of rifting in the early Oligocene, with the regional stratigraphy providing age constraints on the rift history . Syn-rift carbonate sediments were deposited in three fault bound Oligocene basins preserved in the western domain of the island (Fig. 2). Of these the Allan-Kadama and Sherubrub-Balan basins are bound by~N110°E trending low-angle normal faults, and the Central Basin is aligned with the southern part of the HTZ (Bellahsen et al., 2013b) (Fig. 2). The HTZ is a preexisting structure associated with Permian-Triassic rifting which was reactivated during the opening of the Gulf of Aden (Bellahsen et al., 2013b;Denèle et al., 2012;Leroy et al., 2012). There are no syn-rift sediments on the eastern domain and the spatially limited rift faults are high angle (e.g. in Jebel Haggier) (Fig. 2).
Recent thermochronological dating indicates a slightly younger age of rift development in the eastern Socotra domain (38 ± 2 Ma) compared to the western Socotra domain (32 ± 2 Ma), which suggests that the HTZ was already reactivated at that time (Pik et al., 2013). A second phase of uplift and erosion is recorded only in the strata of the eastern part of the island at 20 ± 2 Ma. This post-dates the onset of rifting, but is synchronous with development of the OCT offshore Pik et al., 2013). The maximum denudation is close to the HTZ.
Post-rift sediments sealed the rifted-blocks faults activity with dominant vertical movement from the end of the Miocene to the Quaternary. This resulted in deposition of deltaic fans of continent-derived conglomerates in local basins during a major uplift phase . This was followed by subsidence and marine deposition on eroded, older pre-rift and syn-rift strata (Beydoun and Bichan, 1969;Leroy et al., 2012).

Data
A temporary seismic network of 24 three-component broadband stations in total was installed on Socotra. The network operated for a period of 12 months in two time periods between March 2010 and March 2011 and September 2011 to September 2012 as part of the YOCMAL seismic project. The seismic stations were distributed over the entire island of Socotra with station separation ranging from 5 to 40 km (Fig. 2). Firstly, the seismic network consisted of 6 Güralp CMG-3ESPD (60 s natural period) sensors, 13 Güralp CMG-6TD (30 s natural period) sensors and 2 Güralp CMG-40T (30 s natural period) sensors with RefTek-130 data acquisition systems. Additional data were collected from 3 temporary broadband stations (Fig. 2) equipped with Güralp CMG-40T sensors and RefTek-130 data acquisition systems, operated as an auxiliary part of the YOCMAL seismic project for 12 months between September 2011 and September 2012. In addition, data from one permanent GFZ broadband station SOCY, which was equipped with an STS-2 seismometer and a Quanterra-330 datalogger (shown as triangle in Fig. 2), were analyzed.
The seismic data were continuously sampled at 40 Hz by the temporary seismic stations and at both 100 and 20 Hz by the permanent network. All stations were equipped with GPS timing systems operated in continuous mode to synchronize the stations' internal clocks.
We used 100 teleseismic events with magnitudes N5.5 within an epicentral distance range of 25°-95°from each station (Fig. 1c). The raw waveform data were windowed to start 30 s before and ending 30 s after the computed IASP91 P-arrival time. We filtered the selected signal with a zero phase Butterworth bandpass filter with corner frequencies of 0.05-0.8 Hz. Based on the signal-to-noise ratio of P-wave phase and disturbances (e.g., spikes, glitches, sensor re-centering, calibration pulses, missing data…etc.) within 30 s after the P-wave arrival, we selected the best waveform data recorded at each station. The signal-to-noise ratio of P-wave phase is calculated using SN = A s /A n , where A s is the signal root mean square variance in the time window of 2 s before and 18 s after the P-arrival time, and A n is the noise root mean square variance in the time window of 22 s-2 s before the P-arrival time. Events with a signal to noise ratio (SN) of three or greater on the vertical component were selected and converted to radial and tangential RFs (Ligorria and Ammon, 1999).  Table 1 Crustal thickness (H) and V p /V s ratio results for seismic stations on Socotra. The average crustal velocity used is 6.4 km/s. The average crustal velocity is based on previous studies of the velocity structure for the Arabian platform (e.g., Al-Amri, 1998, 1999Mooney et al., 1985). The RFs then evaluated with careful visual inspection to accept those with a clear positive polarity P-wave arrival at, or shortly after zero time on the radial signal. In addition, accepted RFs have a tangential signal with smaller amplitudes compared to the radial signal and have no significant ringing on the radial or tangential signals. The number of events used for final analysis at each station varies from 18 to 56 (Table 1), depending on the background noise and the technical state of the station. Most of the selected events have an ENE backazimuth with some events from the south and some from the northwest (Fig. 1c).

Receiver function analysis
The receiver function method maps the seismic discontinuities within the Earth below three-component seismic stations by removing the effects of the source, propagation path and instrument response from the waveform of incident teleseismic P-waves that have been partially converted to S-waves. The amplitudes and polarity of the converted phases are sensitive to the impedance contrasts between the layers above and below the seismic discontinuity. The delayed arrival time of the P-waves converted to S-waves compared to the arrival time of the direct P-wave is a function of P-and S-waves velocities above the seismic discontinuity and the depth of the discontinuity (Fig. 3f). Hence, crustal thickness and elastic property information can be derived from receiver function analysis (0wens et al., 1984;Langston, 1977Langston, , 1979Vinnik, 1977;Zandt et al., 1995).
We employ the iterative time domain deconvolution technique developed by Ligorria and Ammon (1999) to compute the receiver functions. We calculate the receiver function by applying several processing steps. The N-S and E-W horizontal components were rotated to radial and tangential components. A 30 s time window (5 s before the theoretical P arrival time and 25 s after) was used to deconvolve the vertical component from the radial and transverse components. We applied a Gaussian filter of 2.5 to the deconvolved spike wave train, except for the noisiest stations where a Gaussian width of 2.0 is used to smooth the receiver function (e.g., Ligorria and Ammon, 1999). We adopt the H-k domain stacking technique (Zhu and Kanamori, 2000), to estimate the crustal thickness (H) and the average crustal V p /V s ratio (k). Inversion results can be seen in Fig. 5 and Fig. S2. The method sums the receiver function amplitudes of the Moho conversion of P-to S-waves and its multiple converted phases (PpPs and PpSs + PsPs) (Fig. 3f)    domain to H-k domain. This approach assumes an average V p for the whole crust and the best estimate for the crustal model is achieved with the maximum summation: where n is the number of receiver functions, Wj, (j = 1,2,3) is a weighting factor that represents the contribution of the corresponding seismic phase according to signal-to-noise ratio (with the constraint that W1 + W2 + W3 = 1) and r i (t) is the amplitude of the receiver function at time t of the associated seismic phase. The primary conversion phase that has the highest signal-tonoise ratio is associated with the highest weight. Low weights (W2 + W3 b W1) are assigned to the multiples (PpPs and PpSs + PsPs) in comparison to the primary phase Pms because the slopes of crustal multiples are very similar in the H-k plane (Zhu and Kanamori, 2000). In case of low signal-to-noise ratio the multiples are down weighted because they are often weak signals and masked by noise. Different values have been tested to evaluate the sensitivity of stacking results to the assigned weight. Then we assign weights of W1 = 0.6, W2 = 0.3 and W3 = 0.1 for the primary phase and multiples, respectively (Zhu and Kanamori, 2000). At a few stations these values have been adjusted based on the visibility of the multiples, using to set of weights W1 = 0.7, W2 = 0.2, W3 = 0.1 and W1 = 0.8, W2 = 0.2, W3 = 0.0. The theoretical travel times of Pms, PpPs and PpSs + PsPs can be calculated as follows: Due to the lack of a-priori velocity information, we tested the H-k technique with a range of average crustal velocities V p (5.8, 6.0, 6.2, 6.4, 6.6 and 6.8 km/s) to determine the most appropriate average velocity to be used in the inversion for H and k in the different regions of Socotra (an example of the results is shown in the supplementary materials, Fig. S4). We chose a value of 6.4 km/s for average crustal velocity (V p ), in agreement with the global average for continental crust (Christensen and Mooney, 1995) and previous receiver function results in the northern margin of the Gulf of Aden Tiberi et al., 2007).
A grid search in the 10-40 km interval Moho depth (H) versus 1.6 to 1.9 V p /V s ratio (k) is carried out to obtain the maximum stacked amplitude. In the minority of cases, we further limit the searched range of V p /V s ratios to avoid unrealistic crustal thickness and/or V p /V s results due to some crustal reverberations in the receiver function signal. We show the range of searched k values appear in the H-k diagrams.
We use the bootstrap resampling technique (Efron and Tibshirani, 1986) to estimate the errors of Moho depth and V p /V s ratio. The technique randomly samples subsets of receiver functions available for each station, recalculating the Moho depth and V p /V s ratio using the Zhu and Kanamori (2000) stacking method. This resampling is repeated 200 times and the standard deviation of these 200 results gives the error estimates for the stacking method and is displayed as error bars in the depth sections. We also use a bootstrap analysis to estimate the errors due to the sensitivity of the method to the assumed V p value. For each station, we recalculate H and V p /V s for 200 randomly chosen values of V p within the range 5.8 km −1 to 6.8 km −1 (Tiberi et al., 2007). Results of error estimates are shown in the supplementary material (Fig. S4).
To determine the velocity model of a layered crust that matches our RF results at selected sites of eastern and western domains of Socotra, we perform RF inversion with a stochastic method (Shibutani et al., 1996). The stacked radial receiver functions of each station are inverted for 1-D shear wave velocity structure using the Neighborhood Algorithm (NA) technique (Sambridge, 1999a,b). The initial model is based on the results of previous geophysical studies of the crustal structure beneath the Arabian plate on the northern conjugate margin (e.g., Al-Amri, 1998, 1999Mooney et al., 1985), and the V p /V s ratio is estimated from H-k stacking results. The model parameters are the thickness of the layer, the shear wave velocity at the top and bottom of each layer, the V p /V s ratio and the density. Our starting model is composed of 4 to 5 layers: sediment layer when needed, basement, upper crust, lower crust and uppermost mantle. Shear wave velocity construction at each layer and the computation of the synthetic waveforms follow the method implemented by Shibutani et al. (1996).

Results
Moho depth H and V p /V s results from receiver functions analysis are shown on the geological map of Socotra island Fig. (6a) and (b) respectively. The Pms Moho conversion phase appears clearly at all the stations from individual and stacked receiver functions (Figs. 3 and 4). Multiples (PpPs + PsSs) are clearly observed in the majority of stacked receiver functions; however, the Moho multiples may be disturbed by multiples from intracrustal and/or upper mantle interfaces at some stations (Fig. 4). Calculated crustal thickness and V p /V s ratio with uncertainties are compiled in Table 1. We organize the results into the eastern and western domains separated by the HTZ, on the basis of their differing geology and also in light of contrasting crustal properties found in our new study (Fig. 2). Most of the stations in the eastern domain display a clear intracrustal conversion phase and thick crust, contrasting the western domain which has thin crust and the absence of strong intracrustal discontinuities.

Eastern Socotra
With the exception of SAYH station, clear positive intracrustal phases are observed within 1.23-1.82 s of the direct P-wave arrival. At some stations (e.g., DESH, HEIL and GSFO) these phases are associated with crustal multiples (PpPs + PsSs) within 5-10 s after the direct P-wave arrival. In all the individual (Fig. 4) as well as stacked receiver functions from all the eastern stations (Fig. 3), a strong positive conversion phase arrives within a delay time of 3.1-3.8 s after the P-wave arrival. These conversions indicate a seismic velocity discontinuity that increases with depth and is interpreted as the Moho discontinuity.
The H-k stacking results ( Fig. 6a and b) indicate that the crustal thickness beneath eastern stations increases westward from 22.4 ± 0.6 km (SAYH) to its maximum value of 30.2 ± 1.1 km beneath BIRZ station on the plateau near the HTZ (Fig. 2). The associated V p /V s values for the eastern stations range from 1.70 (HEIL) located on the basement rock east of Hadibo to 1.82 (SAYH) located on sedimentary rock at the eastern part of the island. A very high value of 1.94 is estimated for TROB located near HTZ at the northern base of Haggier Mountain (Fig. 6b). The positive phase that arrives before the Moho conversion phase can be attributed to an intracrustal discontinuity located at a depth range of 10-16 km depending on the selected average velocity.

Western Socotra Island
The crustal thickness of the western domain of the island increases from the NNW to SSE, such that the western part can be divided into EW-trending zones with uniform crustal thickness. These zones broadly correlate with the three Oligocene basins (from north to south being the Allan-Kadama, Sherubrub-Balan, and Central basins). The northernmost zone, located on the northwest coast of the island, is narrow with a Moho Pms conversion phase of 2 to 2.5 s after the first P-wave arrival and multiples within a 7-10.5 s delay (Figs. 2 and 3). The corresponding crustal thickness in this zone is estimated at 16.4 km (RUSU) and 16.8 km (KADA) (Fig. 5 and S1-2). To the south in and around the Sherubrub-Balan basin five stations are characterized by a clear Moho conversion phase that arrives with a delay of 2.3-3.1 s after the arrival of the direct P phase. This zone, with an estimated crustal thickness of 19.6-21.6 km, makes up a large part of the western end of the island, mainly below Jebel Mala, Jebel Zala, and Jebel Allan, and on the coast at the SHRQ station (Figs. 2, 6a, and 8). Seven stations were located on the third E-W zone at the middle part of western Socotra between Sherubrub (SHER) on the coast and LISK station to the east (Figs. 2 and 8). It is characterized by Moho conversions that arrive within a time window of 2.67 to 3.17 s with associated depth ranges of 20.2 to 23.4 km. In our third, southernmost zone of the Central Basin, four stations constrain the crustal thickness to range between 26.4 and 28.2 km. Here the Moho conversion phase Pms delay time is of 3.5-3.7 s and multiples arrive 11.3-15.3 s after the first of P-wave arrival (Figs. 2, 6a, and 8).
The average value of the V p /V s ratio in the western domain of the island is~1.75. Lower V p /V s ratios (1.65-1.71) are observed beneath some stations, which are located on or near granitic bodies (RUSU, DIAA and SHER) (Fig. 6b). Higher V p /V s ratios of 1.77-1.84 are obtained for the stations located in the Sherubrub-Balan basin (RIGI, AMAD, SHUA, HAME, MDAP, LISK and SHRQ) and also beneath DANA station near the southern coast. A high value of 1.89 is observed for the station located near the northern coast of the western domain (KADA) (Fig. 6b).

Migrated cross sections
The migrated cross sections obtained from two~N-S profiles in the western domain of the island and two E-W profiles across the entire island are presented in Fig. 8. We utilized the IASP91 velocity-depth reference model (Kennett and Engdahl, 1991) to migrate the receiver functions from time domain to space domain, using the common conversion point (CCP) stacking technique of Dueker and Sheehan (1997) and of Zhu (2000). A bin size of 10 km width, 3 km length and 0.5 km depth is used.
The positive amplitudes, which indicate an increase of velocity with depth, are plotted in red, while negative amplitudes are shown in blue. Profile A-B, which is located at the westernmost part of the island, runs from Sherubrub on the southern coast to Kadama on the northern coast (Fig. 8b). The Moho discontinuity occurs at~23 km beneath the southern end of profile A-B and then ascends to~17 km beneath the northern end. The area located below KADA, RUSU and DIAA stations where the crust thins to~16 km shows some energy placed between the surface and Moho (Yellow-Orange colors), which may indicate a rather uniform high velocity gradient extending from the surface to the Moho underneath MALA, DIAA and RUSU stations (Fig. 8a). Another feature present near the northern end of this profile is a localized upper mantle discontinuity at 43 km depth (Fig. 8a). Profile C-D, which is located~17 km to the east of and approximately parallel to Profile A-B, extends from DANA station located on the southern coast to SHRQ station on the northern coast. The Moho depth decreases gently from~26 km in the south (DANA and WURE) to~20 km near the north coast (Fig. 8b).
The distribution of stations is sparser to the east along the E-W profiles. Profile E-F extends from SAYH station in the west of Ras Momi (eastern domain) to SHER station located at Sherubrub (western domain) (Fig. 8). The Moho depth is approximately equal at the two ends of the profile (~22 km) and increases to its maximum below the Haggier Mountains and plateau (Fig. 8c). Profile G-H runs parallel to the northern coast of the island from SAYH station in the west of Ras Momi (eastern domain) to MALA station located in the south of Qalansiya (western domain) (Fig. 8). Along this profile, the Moho depth increases from~22 km at the eastern end to a~28 km below Haggier mountains and then shallows to~16 km at its western end below the stations located near Qalansiya (Figs. 7 and 8d). The isolated upper mantle interface at~43 km depth below the western domain, which has been observed in the profile A-B, can also be seen on this profile. In addition we observe a constant velocity gradient that extends from the surface down to Moho below the stations located NE of Qalansiya (MALA, DIAA and RUSU) (Fig. 8d).
In the eastern part of the G-H and E-F profiles a clear positive phase is observed indicating velocity increase downward. It corresponds to a strong crustal convertor at all the eastern stations, with the exception of SAYH where it is weak and smeared. The positive phase is imaged at approximately 10 km depth and we interpret it to be mid-crustal interface between the upper and lower crust. A vertical step of about 2 km is observed near the location of HTZ and also imaged very clearly along the G-H section.
To further our understanding of the crustal structure beneath the stations in both eastern and western domains that our RFs and migrated cross sections show an intracrustal feature, we inverted the receiver function waveforms for SOCY (eastern Socotra) and DIAA (western Socotra) using the NA algorithm (Sambridge, 1999a,b). SOCY has a clear Moho conversion and multiples in addition to an intracrustal phase and thick crust (see CCP images Fig. 8c and d and RFs stacks Fig. 3a). In contrast DIAA, located in the Qalansiya Valley shows a thin crust, and the velocity increases with depth continuously rather than in discrete steps (see CCP images Fig. 8a and d). These two stations are also characterized by many events with representative RF's.
To further our understanding of the crustal structure underneath the stations in both eastern and western domains that shows an intracrustal feature in the RFs and in the migrated cross sections, we inverted the receiver function waveforms for SOCY (eastern Socotra) and DIAA (western Socotra) using the NA algorithm (Sambridge, 1999a,b). SOCY has a clear Moho conversion and multiples in addition to an intracrustal phase and thick crust (see CCP images Fig. 8c,d and RFs stacks Fig. 3a), while at DIAA, located in the Qalansiya Valley with a thin crust, the velocity increases with depth continuously rather than in discrete steps (see CCP images Fig. 8a and d). These two stations are also characterized by many events with representative RF's. Examples of inversion results are shown in Fig. 9. We obtain crustal thicknesses comparable to those obtained using the technique of Zhu and Kanamori (2000) (Figs. 4 and 5).

Eastern Socotra
Our crustal thickness estimates for eastern Socotra range from 27.6 km (GSFO) to 30.2 km (BIRZ) with an average of 28.6 km below the Haggier massif and 22.4-23.6 km with an average of 23.0 km to the east of the Haggier massif. Our results for the Haggier massif are very close to the crustal thickness (29.2 km) found for the southern Omani coast near Mirbat that represents the conjugate margin of the Socotra area (Tiberi et al., 2007). The crustal thinning from~28.6 km below Haggier massif to 23 km at the easternmost part of the island occurs over~15 km distance between HEIL/GSFO and DESH stations (Fig. 6a). Assuming a 28.6 km thick crust prior to rifting we estimate a stretching factor, β, of~1.24. This represents a lower bound estimate for total crustal stretching since 28.6 km thickness is likely to be representative of crust that has been thinned during early rifting. If we assume that the unstretched crustal thickness of the Arabian platform and/or Arabian shield on the northern conjugate margin is in the range of 35-40 km Al-Damegh et al., 2005;Al-Hashmi et al, 2011;Hansen et al., 2007;Tiberi et al., 2007) and that 23 km represents the final stretched crustal thickness, then β ranges from 1.52 to 1.74. These values are comparable to the stretching factor obtained for the northern conjugate margins studied in southern Oman and southwestern Yemen Tiberi et al., 2007). No clear N-S change in crustal thickness and inferred degree of extension is apparent at the eastern Socotra based on the crustal thickness distribution for the stations located on the Haggier massif and close to the HTZ (ca. 28.5 ± 0.9 km). The only clear indication of crustal thinning occurs at DESH and SAYH stations towards the east in the Momi transfer zone area (Figs. 2 and 8c and d).
For all the stations located on or near the Haggier massif a clear midcrustal interface occurs at depths between 10 and 12 km along the E-W profiles (Fig. 3). This interface appears on both profiles, but most clearly on profile G-H along the northern coast (Fig. 8d). Moreover, there is a clear vertical step in the depth to this interface from~10 km in the east to~12 km to the west in the vicinity of the HTZ. This interface is not imaged by the receiver functions west of the HTZ and the easternmost station SAYH. We interpret this interface to be the upper/lower crust boundary also known as the Conrad discontinuity. This discontinuity represents a sharp velocity step separating the upper crust (average V p of 6.3 km/s) from the lower crust (average V p of 7.0 km/s) (e.g., Ahmed et al., 2013;Mooney et al., 1985). Such an interface is a common feature in the Arabian plate (Egloff et al., 1991;Mooney et al., 1985;Stern and Johnson, 2010). Stern and Johnson (2010) suggest that it separates slower felsic upper crust from faster mafic lower crust, and may coincide with the transition from brittle to ductile deformation. A similar mid-crustal discontinuity has been reported in at least seven locations along the northern conjugate margin in southern Oman (Tiberi et al., 2007), below ASH station northern Oman (Al-Hashmi et al., 2011), and in western Yemen  towards the center of the rift axis. It disappears towards the southern edge of the syn-rift Ashawq graben on the southern coast of Oman (Tiberi et al., 2007).
Our results indicate a~2 km vertical offset of the intracrustal discontinuity on a westward dipping plane in the vicinity of the HTZ. This feature could be related to in depth displacement across a fault zone of 10 km wide. This fault zone could belong to the HTZ which is interpreted as a dextral N050°E-N070°E transform fault (Bellahsen et al, 2013b;Denèle et al, 2012;Fournier et al, 2007;Leroy et al., 2012;Pik et al, 2013). Contrary to Fournier et al. (2007) who concluded that the HTZ was a pre-existing shallow NE-SW fracture zone within the basement, our findings of the vertical shift of Conrad discontinuity across the structure indicates that the HTZ crosscuts the Haggier massif at its western side from the surface down to the lower crust. Moreover, both the Haggier massif and HTZ are pre-existing structures inherited from early history of the margin and are less affected by the Oligo-Miocene rifting episode (Bellahsen et al, 2013b;Denele et al, 2012;Leroy et al., 2012;Pik et al, 2013).

Western Socotra Island
Receiver functions from stations located on the western domain of the island (Fig. 10) do not image the Conrad discontinuity. This may be caused by the crust having been strongly affected by the Oligo-Miocene rifting. Interpretation of the ENCENS cruise seismic profiles at the northern Gulf of Aden margin (Ashwaq-Salalah, Taqah and Mirbat) indicates a gradual velocity increase within the continental domain from 6.0 km/s (upper crustal velocity) to 7.0 km/s (lower crust velocity) without any velocity step . This may indicate that the extensive thinning of the crust is accompanied by removal of the upper crust and/or modification of the crustal composition. Thinning of the crust in the western domain of the island occurred by a similar mechanism of dip slip along detachment faults dividing the western domain into four different areas of uniform E-W and variable N-S crustal thickness. Leroy et al. (2012), based on recent accurate field mapping of Socotra island (Razin et al, 2010), described two complex tilted blocks where previously Fournier et al. (2007) described three tilted blocks. The first block is located along the north-west coast with the low-angle fault bounding the southern side of the~15-km long, 5-km wide Allan-Kadama basin (Fig. 6). The two stations located in this area provide an estimated average crustal thickness of~16.5 km, which is the thinnest crust we observe along the southern margin of the Gulf of Aden. This result is consistent with 15-16 km thick continental crust obtained from gravity and seismic data at the southern end of a NE-SW profile along the Gulf of Aden to the west of the study area (d' Acremont et al., 2006) (Fig. 10).
The crustal thickness increases by 5 km away from the OCT over a horizontal distance of~5 km from the stations in the Allan-Kadama basin to stations in the north of the Sherubrub-Balan basin (Fig. 7). Then it increases slightly over a distance of~15 km from an average thickness of~21 km in this second zone to~23.0 km in a third zone within the Sherubrub-Balan basin. In the fourth zone in the Central basin, located at the southernmost western part of the island and furthest from the OCT, the average crustal thickness is~26.0 km (Fig. 7). The average crustal thickness in the fourth zone is comparable to the average crustal thickness reported for the southern Omani coast in the Dhofar region (Tiberi et al., 2007) (Fig. 10). Approximately the same stretching factor, β, of 1.1 occurs between the second, third, fourth zones and Mount Haggier, but it is higher (β = 1.27) between the first and second zones. If we assume the initial crustal thickness is the same as the average for the Haggier massif (~28.6 km), and a final crustal thickness is~16.5 km in the first block, then β = 1.7. Based upon our results for the final crustal thickness at the northwestern part of the Socotra island and assuming an initial crustal thickness of~35-40 km (Arabian shield and platform; Al-Damegh et al., 2005;Tiberi et al., 2007;Ahmed et al., 2013) the northernmost part of Socotra island has been stretched in the NE-NNE direction by a β factor of~2.12-2.42.
Although rifted margins show a large variability in their tectonic and sedimentary evolution related to their specific rift history and the presence of pre-existing structures, most rifted margins worldwide share a remarkably comparable large-scale architecture (Fig. 11) (Mohn et al., 2012). Based on the variations of crustal thickness and tectono-stratigraphic structure, eastern Socotra island appears to belong to the proximal margin and western Socotra to the necking zone as described by Mohn et al. (2012). The eastern domain consists of one large block tilted toward the south, with a large fault at the foot of Mount Haggier dipping toward the north on which there is at least 200 m of throw ( Fig. 7 and Fig. 17 of Leroy et al., 2012). The western domain is composed of two tilted blocks with low-angle faults dipping 18°to the north. This configuration could represent the necking zone similar to that observed in the West Iberia margin (Péron-Pinvidic and Manatschal, 2008), in the offshore Aden margin  and in an outcropping margin within an orogen such as in the Alps (Mohn et al., 2012). The uniqueness of the Gulf of Aden margin is that the necking zone, which is typically hundreds of meters below water, outcrops above sea level at Socotra.

V p /V s ratio and crustal composition
The spatial distribution of V p /V s ratios correlates well with that broadly expected from the mapped rock outcrops on the island (Fig. 6b). Metamorphic basement and granitic intrusions produce low values (1.65 ≤ V p /V s ≤ 1.71) and sedimentary and volcanic geological units produce velocities that correspond to felsic to intermediate compositions (1.73 ≤ V p /V s ≤ 1.84) (Fig. 6b) (e.g. Christensen and Mooney, 1995;Hammond et al., 2011;Thompson et al., 2010). The V p /V s values have no obvious relationships with either crustal thickness or age of the surface rocks ( Fig. 6a and b).
Five of our stations are located on or near metamorphic basement and granitic intrusions of Haggier massif, Qalansiya and Sherubrub (Figs. 2 and 6). The lowest value of 1.65 found at DIAA station, located in the Qalansiya Valley, may be related to the high silica content of paragneisses and quartzites that form the metamorphic basement (Denele et al., 2012). The other four stations (SHER, MALA, RUSU and DESH) are characterized by V p /V s values of 1.7-1.71, which are consistent with felsic composition of underlying rock units (Fig. 6b). On the northern conjugate margin in Dhofar-Oman very low V p /V s values were found at S09 station located on the metamorphic basement near Mirbat (Tiberi et al., 2007) and in western Yemen a V p /V s value of (~1.68) was reported for the stations located on granitic intrusions or on Precambrian quartz-rich rocks .
Results from thirteen stations, located within the sedimentary basins of the western domain of the island, are characterized by a low V p /V s value of~1.73 for the stations located on a thin sedimentary cover (mostly limestone) and a higher value of~1.83-1.84 for the stations located on a thick sedimentary cover near the middle of the basin. Similarly, in the eastern domain of Socotra, five stations located on the sedimentary cover and at or near the volcanic rocks are characterized by V p /V s ratios in the range 1.73-1.82. In general our finding of average V p /V s ratio of~1.7 for stations located at or near silica-rich  Fig. 9. a -Seismic velocity models for DIAA station (eastern Socotra) obtained from the NA method (Sambridge, 1999a) and waveform matches between the observed stacked receiver functions (red) and the predicted ones (green) based on the best models (red lines in the left diagrams). The gray area indicates the range of models employed by the algorithm. The best 1000 models are indicated in green, the red line corresponds to the best fit both for S-wave velocity and V p /V s ratio and the white line is the average velocity model. b -The same models for SOCY station (western Socotra). rock units and average V p /V s values of~1.77 for stations located on the sedimentary cover are consistent with the felsic and intermediate crustal composition (Christensen, 1996;Gallacher and Bastow, 2014;De Plaen et al., 2014). Broadly our V p /V s ratios contrast with those at the volcanic rifted western Gulf of Aden continental margin where V p /V s is generally 1.8 . Our results of low V p /V s show that the crust has experienced minimal intrusion of mafic rock during breakup, an interpretation consistent with the lack of mafic intrusions and basaltic flows observed at the surface.
The last two stations KADA and TROB are associated with very high V p /V s ratios of 1.89 and 1.94 respectively (Fig. 6b). TROB station is located within the complex structure of the HTZ (Bellahsen et al., 2013b;Leroy et al., 2012;Pik et al., 2013) showing the dipping interfaces and step of the mid-crustal interface (Fig. 8d). Hence, P-wave downdip traveling multiples generated at the dipping mid-crustal interface interfere with the Moho multiples and give a poorly constrained V p /V s ratio which may overestimate the real value of V p /V s ratio beneath the station (Wang et al., 2010). In contrast, the individual receiver functions from KADA are characterized by a clear Pms converted phase and good multiples. In addition the waves approach the station in the up-dip direction that has little effect on the estimate of the V p /V s ratio (Wang et al., 2010). The high V p /V s associated with the thin crust (~16 km) may be due to a more mafic crust with magmatic intrusion near the OCT as in the northern margin (Watremez et al., 2011(Watremez et al., , 2013.

Conclusion
Receiver functions analyzed in Socotra allow us to map Moho depth and V p /V s variations along the southern Gulf of Aden margin. Variability of crustal properties yields a good correlation with tectonic structures of Socotra. Twenty-five crustal thickness (H) and V p /V s ratio measurements obtained by using the H-k stacking technique of Zhu and Kanamori (2000) show that: -The crust of Socotra island is divided into distinct eastern and western domains. The eastern domain is characterized by an average of 28 km thick crust and pre-existing structures that date from the region's early pre-rift history. The western domain is  Tiberi et al., 2007), active seismic, gravity and magnetic studies offshore to the west of Socotra (d' Acremont et al., 2006), active seismic offshore Omannortheastern Gulf of Aden . The roman numbers at the western end of Socotra island indicate the areas with a uniform average crustal thickness as described in the result section. The blue line corresponds to the first magnetic anomaly identified (5d; 17.6 Ma; Leroy et al., 2010a). SHFZ: Socotra Hadbeen fracture zone; Alula Fartak fracture zone; MFZ: Momi fracture zone; HTZ: Hadibo fracture zone.
characterized by an average of 21 km thick crust that was modified during the Oligo-Miocene rifting that led to the opening of the Gulf of Aden. The two parts are separated by the HTZ, a westward dipping plane that may extend to the lower crust. -In the western domain, rifting resulted in progressive crustal thinning from 26 km in the south to 16.4 km in the north of the island. Four structurally distinct E-W trending areas of average crustal thicknesses of 16.5, 21, 23 and 26 km are defined. -Assuming the initial crustal thickness of 28.6 km, the maximum currently observed below Socotra, and thinnest crustal thickness of 16.5 km, we estimate a lower bound stretching factor of 1.7. However, a higher stretching factor in the range of~2.1-2.4 is obtained by assuming the initial crustal thickness more compatible with an undeformed continent such as that found for the Arabian plate (35 km). -An uppermost mantle reflector beneath the anomalously thin crust in the northernmost part of the island has been imaged by our computed receiver functions, but evidence for removal of lower crust and/or underplating could not be assessed with our dataset. -We computed V p /V s to vary between 1.7 and 1.77 at the majority of the stations. This is consistent with a felsic to intermediate crust beneath the stations located at the regions of granite intrusion or at sedimentary basin. The highest V p /V s ratios ranging between 1.78 and 1.84 are mainly associated with sediments. Broadly, the low V p /V s ratios are consistent with no or very low volume intrusion of mafic rock into the OCT during breakup, and therefore suggests that magma intrusion did not accommodate significant extension during the breakup of the eastern Gulf of Aden.
-From our observations we propose that the western part of Socotra corresponds to the necking zone of a classic magma-poor continental margin, while the eastern part corresponds to the proximal domain.