Seaﬂoor spreading event in western Gulf of Aden during the November 2010–March 2011 period captured by regional seismic networks: evidence for diking events and interactions with a nascent transform zone

of the crustal magma reservoir feeding the dikes. Over the whole period, the opening is estimated to be ∼ 1.76–8.8 m across the segment. A striking feature of this episode is that the seismicity remained conﬁned within one individual segment, whereas the adjacent en-echelon segments were totally quiescent, suggesting that the magma supply system of one segment is disconnected from those of the neighbouring segments. Second, we identify activity induced by the ﬁrst intrusion with epicentres aligned along an N035 ◦ E-trending, ∼ 30 km long at the northwestern end of the active opening segment. This group encompasses more than seven earthquakes with magnitude larger than 5.0, and with strike-slip focal mechanisms consistent with the faults identiﬁed in the bathymetry and the structural pattern of the area. We propose that a transform fault is currently in formation which indicates an early stage of the ridge segmentation, at the locus of the trend change of the spreading ridge, which also corresponds to the boundary between a clear oceanic lithosphere and the zone of transform between continental and oceanic crust.


Rifting episodes on the Aden/Red-Sea area
During the last decade, several important seismovolcanic events have occurred in Afar and its surrounding areas of the Red Sea and the Western Gulf of Aden (Fig. 1a), confirming the importance of volcanism in this extensional region above an abnormally warm mantle (Rooney et al. 2012;Ferguson et al. 2013;Hansen & Nyblade 2013). Since most of the events affected subaerial volcanotectonic segments, they offered an opportunity to study the tectonic deformation and the magmatotectonic interactions involved in the active rifting process using the combination of geodetic techniques and seismology. In the Afar Depression, the so-called «Ardukoba» event in the Asal-Ghoubbet rift segment (Djibouti) in 1978, one of the first seismovolcanic events to have been monitored with geophysical methods, shed light on the important role of dike intrusions on the observed surface deformation (Abdallah et al. 1979). More recently, 14 dike intrusions have been identified along the ∼60 km long Dabbahu-Manda-Hararo (D-MH) rift segment between 2005 and 2010 from InSAR data (Wright et al. 2006;Grandin et al. 2009Grandin et al. , 2010Hamling et al. 2009) and accompanied with hour to day long seismic swarms (Ebinger et al. 2008;Ayele et al. 2009;Keir et al. 2009;Grandin et al. 2010;Belachew et al. 2011;Grandin et al. 2011). Other signs of volcanic activity have been observed in southern Afar with the 2000 intrusion near the Ayelu-Amoissa volcanic chain . In northern Afar, Alu-Dalafilla erupted in November 2008 and was associated with sill deflation and dike intrusion , and in November 2010 Erta Ale lava lake rose to over-spill the inner crater (Field et al. 2012). The most recent documented magmatic event is the June 2011 eruption and subsidence of Nabro volcano (Ebinger et al. 2013;Hamlyn et al. 2014). Previous to these events, a dike intrusion and induced fault slip were observed near Dallol in October 2004 (Nobile et al. 2012). Further east, along the Sheba ridge and offshore Oman, volcanic activity is reported (Lucazeau et al. 2009;d'Acremont et al. 2010;Leroy et al. 2010a). Further north, in NW Saudi Arabia, May 2009 marks a new episode of seismic activity in the Harrat-Lunayyir volcanic field Al-Zahrani et al. 2012), where InSAR data helped to constrain this episode as a dike intrusion (Baer & Hamiel 2010;Pallister et al. 2010).
In agreement with previous studies of similar events in oceanic domain (e.g. Sigurdsson 1980), the displacement measurement associated with most of these seismovolcanic events, when they exist, are consistent with a horizontal opening across the rift, a narrow elongated subsidence zone along the rift inner floor, uplift of the margins and a local subsiding area below the main volcanic centre (Rubin & Pollard 1988). Together with the seismic data showing a lateral migration of the earthquakes (Brandsdöttir & Einarsson 1979), they confirm that dike intrusions accommodate significant deformation over short time-scales and that dikes are fed by sometimes complex crustal magma plumbing systems (Wright 2012). A less well-constrained magmatic event in the Red Sea is the relatively aseismic eruptions starting September 2007 at Jebel-Attair Island ending longer than a century of quiescence. In November 2011, NASA reported this activity caused birth of a volcanic island at Al-Zubair archipelago. Due to most activity occurring below the sea surface deformation could not be seen.
Along submarine mid-oceanic ridges, the monitoring of active rifting process is usually achieved through seismic data from local or regional oceanic ocean bottom seismometer and autonomous underwater hydrophone networks (Fox et al. 1995;Blackman et al. 2000;Dziak et al. 2009) or global land-based instruments (e.g. Tolstoy et al. 2001Tolstoy et al. , 2009). These seismic surveys reveal important oceanic spreading-related processes including the migration of the seismic activity (Fox et al. 1995), the role of fluids, and the slip activation on normal or transform faults in response to probable magmatic emplacements Dziak et al. 2009).
In this paper, we present the results from a regional seismic survey of the November 2010 episode along the oceanic ridge in the western Gulf of Aden. In doing so we improve significantly the precision in earthquake location, moment tensor inversion and magnitude cut-off of earthquakes achieved from global records (Shuler & Nettles 2012). We interpret our results in light of the recent seismic and geodetic observations made in the region along subaerial segments in order to better understand opening dynamics of the spreading segment and its relationship with fault activity. In addition, we use the high-precision earthquake locations along with highresolution bathymetry data to discriminate between dike-induced faulting near the dike and later induced slip on a nearby transform fault.

Geodynamical setting
The Gulf of Aden rifted margins transition from being volcanic in the west, to non-volcanic or including little evidence of magmatism in the east (Ahmed et al. 2014;Corbeau et al. 2014;Korostelev et al. 2014). The Gulf of Aden can be divided into three main parts based on sedimentation and structural styles Leroy et al. 2012;Bellahsen et al. 2013). With an extension direction of N020 • E, the eastern part is characterized by continuous calcareous sedimentation since the onset of the rifting (east of Alula-Fartak Fracture zone, AFFZ; Leroy et al. 2012;Robinet et al. 2013). The central part, between the AFFZ and Shukra-el-Sheik fracture zones (SSFZ), shows uplift and erosion with a subaerial to shallow sedimentation during the rifting (mudstones, evaporites and reefal limestones; Leroy et al. 2012). West of SSFZ, the western part of the Gulf, has been under the influence of Afar plume with a variable spreading ridge trend and extension direction (Tard et al. 1991;Hébert et al. 2001;Leroy et al. 2010a). The slow-spreading rate along the Aden and Sheba Ridges progressively increases from west to east from 1.6 cm yr −1 at the entrance of the Gulf of Tadjura to 2.4 cm yr −1 at the Aden-Owen-Carlsberg triple junction (Fig. 1a;Fournier et al. 2001Fournier et al. , 2010.
The westernmost part of the Aden Ridge presents a straight ridge with no first-order segmentation whereas the central part is more segmented with seven transform faults that offset the ridge in a leftstepping fashion by successive transforms ranging from 14 to 47 km ( Fig. 1). Detailed studies of the eastern part of the Gulf of Aden demonstrate that between the long-lived AFFZ and Socotra Hadbeen fracture zone (SHFZ; Fig. 1a) the oceanic ridge segmentation changes from three to two segments. Changes in volcanic feeding of the oceanic ridge and plate kinematics are proposed (d 'Acremont et al. 2006'Acremont et al. , 2010. Within the westernmost part of the Gulf of Aden, where the spreading ridge is linear, three distinct lithospheric domains are again identified based on gravity data (Hébert et al. 2001) and magnetic data (Leroy et al. 2012;Fig. 1b). The well-developed eastern domain, which lies between E044.30 • and E045 • (Fig. 1b), is characterized by an 8 km thick oceanic crust on either side of the ridge (Hébert et al. 2001) with magnetic anomalies defined as chron A5  (10 Ma) at the eastern part of the domain and A3a (6 Ma) at the western part of the domain (Tard et al. 1991;Audin 1999;Dauteuil et al. 2001;Hébert et al. 2001;Leroy et al. 2012). The central domain, located at E043.90 • to E044.30 • (Fig. 1b), marks the lithospheric transform between the well-developed oceanic lithosphere to a transitional lithosphere toward the African continent with the magnetic anomaly corresponding to chron A2a (2 Ma) on both sides of the spreading axis (Leroy et al. 2012). The crustal thickness of this domain is approximately 9-12 km below the ridge (Laughton et al. 1970;Hébert et al. 2001). Recent volcanic activity has been identified in this domain where most of the axial valley is covered with recent lava flows (Dauteuil et al. 2001). The western domain is located from E043.90 • to the entrance of the Gulf of Tadjura, and has a spreading ridge with a trend ranging from EW to ∼N075 • E with a 10-15 km wide axial valley. Four important basins, with two of them being the deepest basins west of Shukra El-Sheik fracture zone (SSTF), are 1650 m deep (Dauteuil et al. 2001). No oceanic crust has been clearly identified for this domain (Dauteuil et al. 2001), although the magnetic anomaly chron A2a (2 Ma) is observed on the southern shoulder of the spreading axis as well as the axial magnetic anomaly (Leroy et al. 2012;Fig. 1b). Courtillot et al. (1980) proposed the existence of an incipient oceanic crust within the Gulf of Tadjura and the subaerial domain in Djibouti from aeromagnetic data. In contrast, Hébert et al. (2001) suggested stretched and thinned continental crust of 15-20 km thick computed from gravity data. From oceanic magnetic anomalies identification and the bathymetric data, the Aden spreading ridge shows numerous clear volcanic edifices and cones providing evidence for magma involvement in the rifting process (Leroy et al. 2010a).

Location process and velocity
We combined the data recorded by both permanent and temporary seismic networks in Djibouti and Yemen presented in Fig. 2 and in Aux Mat-1 (Supporting Information). Both temporary networks were fully completed by 2010 December 15.
The determination of the velocity structure in the region is complex because of three main reasons: (1) the main active area is located along the ridge axis where oceanic spreading potentially occurs; (2) the stations in Yemen are located on the continental plateau; and finally (3) because the structure in Afar, and more specifically in Djibouti, is still debated between an oceanic crust (Ruegg 1975;Barberi et al. 1980), a continental crust (Makris & Grinzburg 1987) and a continental crust intruded by gabbroic and granitic plutons (Mohr 1987). Several distinct velocity models derived from active seismic profiles exist, and are presented in Aux Mat-2 (Supporting Information). All the models are characterized by a thin crust. Following Laughton & Tramontini (1969), this area is also characterized by an abnormal upper mantle velocity structure with a velocity (7.1 km s −1 ) lower than typical mantle velocity range (>7.8 km s −1 ). Similar results have been reported by Ruegg (1975) within the Gulf of Tadjura west of the activity area, with anomalous mantle velocities (6.7-6.9 km s −1 ). We then tested ten different crustal and upper mantle velocity models and the final model was selected using the criteria of minimum station residuals and minimum root mean square (rms) for the location of the 50 events encompassing at least 25 arrival times. We used a fixed Vp/Vs of 1.78 (Ahmed et al. 2013). The red and blue symbols correspond to the permanent and temporary stations, respectively. The locations of the 1122 events located in this study have been indicated by red circles. The insert at the top left corner is a zoom of the network in the centre of the Asal-Ghoubbet Rift. The insert at the bottom right corner is the seismicity recorded between 2010 November 10 and 2011 March 31. The blue and green dots are the events located by Shuler & Nettles (2012), by USGS, respectively. The locations determined in this study correspond to the red dots.
We analysed the seismic activity, from 2010 November 1 to 2011 March 31, during which several thousand of events were recorded. Using the selection criterion of picking at least a P-phase at seven stations and S-phase at two stations, 1400 events were selected for further analysis. A total of 21 649 P-wave arrival times were manually picked on the vertical component and 7202 S-wave arrival times were picked on horizontal components for three-channel stations and on vertical for single channel stations. The location process of some events has been difficult because most of the events have small amplitude emergent body-wave arrivals and because signals of distinct events are overlapping due to the intense seismic activity especially at the beginning of the period. Finally, we located a final set of 1122 regional earthquakes using the SEISAN software package (Ottemöller et al. 2011). In the final set, every earthquake has an average rms (root mean square of traveltime residuals) lower than 0.4 s. With the exception of 8 events, the horizontal and vertical errors are less than 3.75 and 5.0 km and below 2.0 and 2.5 km because of the network evolution on December 15, respectively (Aux Mat-3, Supporting Information). The complete network allows a wide cover of azimuths. Therefore, the values of the horizontal errors for the locations deduced from this regional network are significantly lower than the lateral location uncertainties given from centroid moment determinations which are between 1 and 13 km (Shuler & Nettles 2012).
The local magnitude of our set of localized seismic events ranges from 2.1 to 5.6.

Moment tensor inversion
We performed a moment tensor inversion to estimate the source parameters (seismic moment, moment magnitude and fault type) of strong and moderate earthquakes (M L ≥ 4.0) recorded by our seismic networks (Dziewonski et al. 1981;Kikuchi & Kanamori 1991;Dreger & Helmberger 1993;Kawakatsu 1995). Using Moment Tensor Inversion program from regional networks coded by Yagi (2004), we analysed the three-component seismograms recorded at broadband stations for events with M L above 4.0. Because of the network configuration, we chose to not take into account the isotropic component. The observed data were windowed 5 s before and 80 s after the P phase arrival time, then filtered from 11 to 25 s. Since we previously located events using a dense regional seismic network, we only explored the effect of centroid depths on the results of the moment tensor solution. Therefore to test the focal depth of the event determined by the location program and the stability of the used velocity model employed, we changed the depths of the hypocentre and selected the depth corresponding to the minimal variance. We estimated the depth accuracy of about ±1-2 km. In order to estimate the seismic energy released during the period of activity, we establish the linear fit between the local magnitude (M L ) and the logarithm of the seismic moment from the set of events for which we estimated the moment tensor (Aux Mat-4, Supporting Information).

Spatial distribution and main tectonic/volcanic features
During the analysed time period, the seismic activity in this area is high, with up to 73 located events d −1 on 2010 November 14.
From November 2010 to the end of March 2011, the activity is characterized by high-magnitude events, including 29 with M L ≥ 5.0, and a maximum recorded magnitude of 5.6. The activity is characterized by small isolated clusters located between (043.5 • E, 11.80 • N) and (044.6 • E, 12.25 • N), that is, within a 120 km EW-long and 40 km NS-wide rectangular area (Fig. 3a). More than 95 per cent of the located events are concentrated within the ridge valley. Below, we refer to segments (E-O) and canyons (1, 2) following the naming given by Audin (1999) from bathymetric and reflectivity data acquired during the Tadjouraden cruise in 1995 (Audin 1999;Dauteuil et al. 2001;Figs 3b and c).
The area affected by this intense seismic activity is located in the western part of the central domain (Section 1.2, Figs 1b and 3a) within the ∼20 km wide axial valley. Four en-echelon N115 • E-trending basins dissect the axial valley formed by N100 • E-to N140 • E-trending, ∼20 km long and ∼100 m high normal faults. The depth of the axial valley ranges from 1000 to 1500 m with the most superficial part corresponding to the central graben of segment J (Figs 1b and 4a). Near segment J, the vertical throw of the main normal faults bounding the ridge valley ranges from 300 to 500 m and decreases eastwards (Fig. 4b). On the northern ridge margin, two ∼300 m deep, 1-6 km wide, ∼NS-trending canyons run from Downloaded from https://academic.oup.com/gji/article-abstract/205/2/1244/689924 by IFREMER user on 17 July 2020 the southern Red Sea through the Bab El Mandab strait, to the axial valley and end by forming large fans along the northern ridge wall (fan 1 and fan 2; Audin 1999; Dauteuil et al. 2001). At the southern end of the western canyon, a NS-trending graben is bounded by four ∼N170 • E-striking normal faults. This graben is restricted in the north by chains of volcanic cones, and two EW-striking large faults in the south. Also, a narrow EW-trending graben intercepts this graben at right angles in its central part (Figs 3b and c). The large deposits of fan 2 have been shifted to the east and the canyon channel flows through a curved channel to the east of the graben, cutting the western side of the fan deposits and running southwestwards down to the inner-floor of segment L. The relative location of the eastern canyon and its associated fan 1 implies a right-lateral strikeslip movement along the E-W bounding normal fault. Assuming extension directed N037 • E, Dauteuil et al. (2001) ascribed a 5 km shift with a spreading rate of 2 cm yr −1 since 300 ky. Oblique normal faults with a right-lateral strike-slip component are observed in this domain of the Aden Ridge, consistent with the NW-SE-trending faults and the left-stepping en-echelon pattern with respect to the valley E-W-trending walls (Dauteuil et al. 2001).
The spatial distribution of the seismic activity clearly shows a main swarm centred at E044.000 • -N12.042 • (Fig. 3a). It includes 1032 earthquakes with magnitude ranging from 2.2 to 5.5, consisting of 90 per cent of the total number of events. The shape of the main N115 • E-trending and 40 km long swarm reveals that the episode is mainly related to the activity of one individual spreading segment, segment J, located east of the fan 2. Segment J extends from the southern ridge wall, characterized by short en-echelon NW-SE normal faults, to the northern ridge wall of the main axial valley. Near the segment centre and at the NW end of the central graben of the segment, at the base of northern ridge wall, chains of volcanic cones with a mean trend of N115 • E-N130 • E are aligned parallel to the bounding faults (Figs 3b and c). The majority of the epicentres (∼700 events) lie within the deep central graben formed by antithetic NW-trending normal faults. We note that the adjacent en-echelon segments K and I are completely inactive during this period.
Thirty kilometres west of the main swarm, a cluster of 39 moderate size seismic events was detected. Near segments M and N, strong events, eight of which have magnitude above 5.0, occurred along the northern and southern walls of the main valley as well as on the N130 • E-trending transverse ridge which is made of volcanic cones, recent basaltic flows and normal faults (Choukroune et al. 1988;Figs 3 and 4a). This area is characterized by steep E-W-trending normal faults forming most of the bathymetry along the northern and southern walls. More recent and smaller N130 • E-trending normal faults cut the ridge walls resulting to an en-echelon structure of the ridge wall. At this longitude, the cumulative height of the ridge walls exceeds 1600 m.
In addition, three small seismic clusters, encompassing ∼51 events, with magnitude below 4.7, occurred during the study at different locations along the ridge: longitude E043.36 • , E044.23 • and E044.52 • (Fig. 4). In agreement with the time-space analysis below, we do not consider these clusters to be directly related to the main activity of the episode. The westernmost cluster at E043.36 • occurred offshore within the Gulf of Tadjura, south of the city of Obock, near a volcanic cone located southwest of segment P (Figs 2 and 3). The events of the cluster at E044.23 • are aligned along the scarp resulting from the overlap of two deposit cones forming the fan 1 (Fig. 3). Finally, the cluster at E044.52 • is located between segments G and F, where the seafloor is structured by volcanic cones and ∼N135 • E-trending, SW-dipping normal faults (Fig. 3).
As shown in Fig. 2, the epicentres of the events recorded during this period and located by the NEIC agency are shifted to the south relative to our locations by 0.2 • . In contrast, the epicentre locations of Shuler & Nettles (2012) are shifted to the east by 0.2 • with respect to our results. Their locations of the main activity are centred at (E044.22 • , N12.07 • ), that is, at the base of fan 1 coming out of the eastern canyon, and they extend along the NW direction towards the northern ridge wall, north of segment I (Fig. 3). The seismicity of the westernmost cluster is more diffuse than the distribution we achieve since the epicentres are distributed from the northern wall to the southern wall of the ridge, west of segment J. Therefore, segment J appears aseismic according to the locations determined by Shuler & Nettles (2012), while it corresponds to the main active area in our analysis. The difference of epicentral locations between global determinations [Preliminary Determination of Epicentres (PDE) from United States Geological Survey (USGS)] and regional inversion are expected to be due to the difference of spatial cover of the monitoring network. In the case of the USGS locations (first arrival times), the heterogeneity of the stations explains the shift with our locations from the regional network. In the case of the centroid moment tensor (CMT) catalogue, the lateral (and vertical) shift results primarily from the Earth three-dimensional structure used for the inversion algorithm, which does not take into account lateral heterogeneity, as showed for instance by Hjörleifsdottir & Ekström (2010).

Depth distribution
Due to both the distribution of our seismic stations at the time of the swarm onset and the distance between the stations and the active area (∼80 km), any analysis of the hypocentre depth has to be done with caution. However, as we mentioned in Section 2.1, the network configuration was enhanced within the first month giving a better accuracy on the depth determination. The focal depths of the main swarm range from 1.6 to 12.8 km, which is consistent with other slow-spreading oceanic ridge segments (Toomey et al. 1985;Wolfe et al. 1995;Cannat 1996;Fig. 4). A large number of events are located at a shallow depth less than 2 km, especially at the central part of the segment J. The seismic activity is deeper, up to ∼10.0 km depth (Fig. 4b). We note however some alignments at the depth levels of 1.6, 2.0 and 2.1 km below the central graben probably resulting from the discontinuity of our crustal velocity model at 2.0 km. These alignments disappear at the SE and NW ends of the segment J where one notable feature is the ∼15 • -dipping ∼2.5 km thick gap of seismicity extending from 10.0 km depth below the northern margin to the area at 3.5 km depth below the central graben (Figs 4a and b). It is compelling to note that this gap of seismic activity and its shape appear early during the episode and does not evolve over the whole period of monitoring.

Focal mechanisms
Due to the strict selection criteria we adopted, the close occurrence of events in time and the relatively short hypocentre-station distance, several events with P onset in the coda of previous events have not been located. This reduces the total number of moment tensor solutions to 99 events compared to the 110 events CMT solutions determined by Shuler & Nettles (2012), who used teleseismic data recorded by remote seismic stations (epicentral distance >670 km) over a six-month period from November 2010 to April 2011.
In our analysis, 78 per cent of the solved 99 events occur within segment J and the remainder belong either to the high-magnitude western cluster or to the transfer zone between the central and western domains (Fig. 5). We distinguish three categories. First, 60 per cent of the earthquakes are characterized by normal faulting with a major trend of NW-SE. Second, 20 per cent of the earthquakes are characterized by normal faulting with a right-lateral strike-slip component. Third, 17 per cent of the earthquakes are characterized by strike-slip mechanism, including 11 per cent with pure strikeslip movement and 6 per cent with a slight component of normal faulting. We note only one event with left-lateral strike-slip faulting and two events with a reverse mechanism. The fault mechanism patterns are discussed in the following section.
Overall, our fault plane solution and the CMT solutions (Shuler & Nettles 2012) are very similar. The discrepancies are mainly due to the location differences of the hypocentres. Unlike Shuler & Nettles (2012), we do not constrain the hypocentral depth to 12 km, since it would be larger than the ∼10 km thick crust according to the seismic data in the western Gulf of Aden (Laughton & Tramontini 1969) and more generally according to the knowledge of the crustal structure along the spreading axis of mid-oceanic ridges (Cannat 1996) and its seismogenic behaviour (Weidner & Aki 1973;Toomey et al. 1985).
Our focal mechanism solutions are consistent with the structures inferred from the bathymetry and/or seismic profiles (Manighetti et al. 1997;Audin 1999;Dauteuil et al. 2001). First, the seismic activity concentrated within the inner floor of segment J is largely dominated by normal faulting mechanisms along N100 • Eto N140 • E-trending planes. This corresponds to the mean direction of the normal faults and the alignments of volcanic cones. We note that the mechanisms showing strike-slip faulting or normal faulting with a significant strike-slip component are located on the edges of segment J, at its two northwestern and southeastern ends and on its eastern border. This probably results from the complex structure of the ridge valley at this part of the Aden ridge, where the opening segments are perpendicular to the extension direction and the ridge walls are oblique to the N037 • E-trending extension direction, inducing lateral slip on E-W-trending normal faults. At the NW end of segment J, several types of focal mechanisms exist, consistent with the two sets of normal faults (NW-SE-trending and E-W-trending) and strike-slip faulting along the EW-trending direction (Fig. 5). Moreover, the complexity of the transform zone between the central and western domains is shown by three sets of strike-slip mechanisms with normal component for some of them and whose nodal planes are parallel to the following trends: N087 (Fig. 5). From bathymetry data and the seismicity, it is clear that there is a transform area between the central and western domains, since all the structural features from the east and west end at this level of the ridge and where both left and right strike-slip movement occur (e.g. Shuler & Nettles 2012).
Finally, at the westernmost area of the activity, the mechanisms are consistent with a dominant ∼N100 • -trending normal faulting for the events located along bounding faults of segments M and L. We note also that the events located on or near the southern wall are characterized by strike-slip faulting with a normal component, consistent with right-lateral slip on the N080 • -trending normal faults (Fig. 5).

Frequency content
Our seismic stations are more than ∼80 km from the source area, and as such the frequency content of the seismograms is likely influenced by path effects including attenuation. We therefore interpret our waveforms and spectra with caution. However, because all the events are located within a small area and therefore affected by the same path effect, we compare arrivals from different earthquakes at similar stations to identify distinct earthquake types. The first group of earthquakes is characterized by emergent low amplitude body waves with the whole spectrum below 2 Hz for all stations, implying that the LF content is neither due to the path nor to the site effects, but related to the source. An example of this kind of lowfrequency (LF) earthquake recorded at the station OBO2 in Obock in Djibouti is shown in Fig. 6(a). We observe a dominant frequency at ∼0.5 Hz (see signals and spectrograms at three other stations in Aux Mat-3, Supporting Information). Another characteristic of the LF earthquakes is the long coda, which lasts more than 100 s for a M L = 4.0 event. The second group includes earthquakes with a peak in their spectra above 3 Hz (Fig. 6b). It is important to note that for a similar range of magnitude, the events of this group have a shorter duration than those of the first group. Together with some energy picks at frequency below 4 Hz, most of seismic energy is released in the frequency band of 4-10 Hz with a dominant frequency at ∼6-7 Hz. The third group corresponds to events with a mix of high and LF contents (Fig. 6c).
Looking only at the earthquakes with magnitude above 5.0 (Fig. 6d), it is notable that the epicentral locations of the events belonging to each of these three groups are not randomly distributed within the active area. With the exception of one earthquake, the LF earthquakes of the first group are clearly located within segment J, where the volcanism and magmatism is concentrated. With a lower magnitude, the LF event shown in Fig. 6(a) that occurred on 31 December 2010 (18:09 UTC) is located near the volcanic cones bounding the middle of the central graben from the southwest (E043.995 • E, N12.013 • ), at a depth of ∼4.8 km.
The events of the second group with high-frequency content are mainly located along the SW-NE-trending zone crossing the ridge valley, where the focal mechanisms are consistent with clear strikeslip or normal slip. The high-frequency event of 2011 February 12 (14:54 UTC) shown in Fig. 6(b) is located on the fan 2 (E043.896 • , N12.056 • ) in the transfer zone between the central and western domains, at a depth of ∼5.1 km. Finally, the western domain where the events are mainly located within the ridge valley is characterized by earthquakes with a content of a combination of low and high frequencies.

Seismic moment release
The total cumulative seismic moment released during the monitoring period is ∼4.46 × 10 18 Nm (Fig. 7a). Two thirds of this energy was released the first month following the onset of the swarm and one third during the following four months, from December 2010 to March 2011 (Fig. 7a). The large magnitude events largely contribute to this high value of seismic energy till the end of 2010. From the beginning of 2011, the number of earthquakes remains relatively high but few large magnitude earthquakes are recorded leading to a subdued seismic moment release. The cumulative seismic moment estimated for the main active area, segment J, in the central domain, is ∼2.76 × 10 18 Nm, which accounts for ∼61.8 per cent of the total seismic moment release (Fig. 7a). The seismic cluster located 30 km southwest of the main swarm area (western domain) is characterized by strong events that account for 19 per cent of the 1254 A. Ahmed et al. seismic energy with a cumulative seismic moment of 0.85 × 10 18 Nm (Fig. 7). The seismic activity in the transform zone between the central and western domains is characterized by moderate seismic activity accounting for 17.7 per cent (0.79 × 10 18 Nm) of the seismically released energy (Fig. 7). Other small clusters located to the east and the west of the main swarm area release 1.5 per cent of the total seismic energy.
Looking at the time evolution of both the cumulative seismic moment and the number of events (Fig. 7a), we distinguish several main time periods of distinct character, that we describe in the following, in order to better understand the evolution of this seismovolcanic episode.

Pre-swarm
From November 10 to 13, 09:23 UTC, the first signs of activity in the area correspond to seven small and moderate earthquakes (3.2 ≤ M L ≤ 4.1) accounting for very little of the seismic energy (∼2.6 × 10 15 Nm). Five of them form an alignment along the 4 km long N105 • E-trending fault at the NW end of the segment J central graben. Two of them occurred below the edge of the northern wall of the axial valley at the NW end of the segment J (Fig. 8a).

Main swarm
The first and main swarm started with an acceleration of the event occurrence from November 13 (17:35 UTC). A total of 166 events were recorded till November 17 (02:02 UTC), accounting for more than the half of the total seismic energy of the whole episode. We decompose the main swarm into three distinct time spans based on the seismicity evolution and the particular patterns drawn by the hypocentres.
First, during the five initial hours of the main swarm, on 2010 November 13, from 17:35 UTC to the end of the day, the seismic activity corresponds to 29 earthquakes mainly located along a NS-trending axis at E044 • longitude (purple epicentres on Fig. 8a) and accounting for 0.02 × 10 18 Nm. The maximum magnitude is 4.4, corresponding to an event (18:26 UTC) located at the centre of the inner floor of segment J, with a focal mechanism consistent with a normal slip along a SE-trending plane parallel to the bounding structures of the central graben (Fig. 8a).
Second, the seismic activity on November 14 is characterized by a combination of small and moderate events (Fig. 8b). The events recorded during the six first hours are mostly low magnitude (3.2 ≤ M L ≤ 4.2, 0.01 × 10 18 Nm) and are concentrated below the centre of the segment J around 6 km depth. Except for one M L = 4.4 event (01:47 UTC) located near the northern border of the central graben and characterized by reverse faulting along a SE-trending plane, all the focal mechanisms reveal normal faulting and encompass a nodal plane with the direction ranging from N115 • E to N140 • E. This trend is consistent with the evolution of direction of the bounding faults of the central graben from the NE to the SW (Fig. 8a). It is important to note that the area active during this period at the centre of the segment J reveals a circular aseismic gap surrounding a volcano at the NW end of the central graben during the following periods (Fig. 8b). From 06:05 UTC to the end of the day, the earthquake magnitude increases significantly, since it is above 4.2 for the 59 located events, and above 5.0 for 13 of them and a corresponding seismic moment reaching 1.7 × 10 18 Nm. The largest event (M L = 5.4 at 06:32 UTC) is located at the mid-segment, at the foot of the slope of the northern ridge wall (Fig. 8b). The events that immediately follow between 6:00 and 09:30 UTC migrate SE-wards along the inner floor of the SE section of the segment J. It is interesting to note that all the focal mechanisms related with this time span are consistent with normal faulting including a high-dip plan (>80 • ). From 10:00 to 13:00 UTC, the events migrate northwestwards along the inner floor of the NW section of the segment J. The majority of the focal mechanisms show normal faulting, except for two events with strike-slip mechanisms along ∼EW-trending nodal plane that occurred near the intersection of the northern end of segment J with the northern wall E-W-orientated at this level of the ridge (Fig. 8a).
Following the significant activity within the segment J graben, the events recorded after 14:41 UTC are characterized by high magnitude events (13 events with M L > 5.0). They occurred first at the NW end of the segment J, below the northern margin of the rift valley and reached the transform zone between the central and the western domains. From 15:06 UTC on November 14, the activity shifted in the western domain and affected mainly the segments M and L (Fig. 8). Twenty events occurred in the area between segments L and O, with 18 of them with M L ≥ 4.0, including 8 with M L ≥ 5.0. This cluster started on November 14 in the western area includes the event with the largest magnitude in our catalogue (M L = 5.6), which occurred at 17:02 UTC. Both normal faulting and pure strike-slip types of focal mechanisms are determined. However, the majority of them encompasses an EW-striking plan, in agreement with the EWstriking faults identified from the bathymetry data and bounding the rift valley and forming the ridge wall. Events located along or near the southern ridge wall of the western domain exhibit a general trend of E-W-trending normal faulting with a considerable strike-slip component (Fig. 8a). However, It is important to note that this area remains inactive over the following months, till March 2011, when another cluster of events affected this area (Fourth period).
Finally, during the two following days, the occurrence rate of events significantly decreases with ∼30 events d −1 . The activity on November 15 is characterized by a swarm mainly located below the northwestern section of the segment J and the northern margin of the rift valley. The magnitudes remain high, with 3 events with magnitude above 5.0. The focal mechanisms are again consistent with a combination of strike-slip and normal faulting type. We note as well the occurrence of a moderate event (M L = 4.6) along the volcanic chain separating the segments N and M, with a strike-slip fault mechanisms. On November 16, no large magnitude events were recorded, but the activity remained high, mainly concentrated at the centre of the segment J. Very few focal mechanisms have been determined. We note however the reverse-type mechanism, consistent with a NE-striking slipping plane.

Second period
Daily event distribution of the period November 17 (20:45 UTC) to December 4 shows that the seismic activity occurs by clusters interrupted by quiet periods and accompanied only by four events with M L ≥ 5.0. The seismic energy released during this period is low (0.7 × 10 18 Nm) in comparison with the previous one. Fig. 9(a) shows that the activity is characterized by two swarms aligned along two orthogonal trends NW-SE and NE-SW. West of the longitude E043.9 • , the earthquakes tend to align between two NE-SW parallel lines with a mean trend of N026 • E. We note a 9 km long N157 • E line extending from the mouth of the western canyon channel to the NW end of the segment K (Figs 3b and 9a). Two large events with M L = 5.0 and 5.2 occurred at the NW end and near the SE end of this line, respectively. These two events are characterized by strike-slip mechanisms with a nodal plane trend of N158.5 • E and N153 • E, parallel to the trend of seismicity and the normal faults bounding the canyon channel (Fig. 9a). The seismic activity aligned along the NW-SE direction was confined to the NW end of segment J. Events with dominant normal faulting mechanisms occur at the centre of the segment. However, events with normal faulting mechanisms including an important strike-slip component and one E-W-trending nodal plane occur along the faults bounding the segment from the north and northeast. One event located at the base of the northern ridge wall exhibits a reverse faulting mechanism (Fig. 9a).

Third period
From December 5 to the end of 2010, we located 314 events with reduced location uncertainty due to the configuration of the network, which encompasses more stations on the Yemeni coast to the north of the active area (Aux Mat-1 in the Supporting Information, Fig. 2). This period of time is characterized by the highest activity with many small events at the centre of segment J partly due to the lower detection threshold related to the optimal configuration of our network from 2010 December 15. Also, at least 13 moderate to strong events occurred on the northern and southern edges of the segment, with eight of them of magnitude M L ≥ 5.0, four with M L = 4.9 and one with M L = 4.8. The total seismic energy released is 1.4 × 10 18 Nm. All the mechanisms of the events located on the northern edge of the segment are characterized by normal faulting with a nodal plane trend from N110 • E to N125 • E, except one event which is located near the volcanic highs bordering the central graben from the east and exhibits a strike-slip mechanism with an EWtrend plane (Fig. 9b). The events located on the southern edge of segment J are characterized by normal faulting with a small strikeslip component along planes with a direction between N130 • E and N150 • E (Fig. 9b). Some seismic activity aligned along the NE-SW direction at the western end of segment J is noticeable and 60 per cent of the events are clustered near the NW end of the central graben. Most of the shallow seismic activity (depth <4 km) was confined below the area situated between the volcanic cones at the segment centre and NW end of the central graben, with a seismic gap below the volcanic cones. Whereas there is no seismic activity in the western domain in December, two clusters occurred east of segment J. The first cluster corresponds to three events whose hypocentres lie at a depth between 3.5 and 7 km and are aligned along the fault that divides fan 1 into two parts (Figs 3 and 4a). The second cluster consists of 14 events located at a volcanic complex area between the segments G and F with a depth ranging from 2 to 11 km (Figs 3 and 4a).

Fourth period
The fourth period corresponds to the regular activity during the months of January to March 2011. A total of 411 events were recorded during this period accounting for a very low cumulative seismic moment (∼0.27 × 10 18 Nm). The highest magnitude recorded is 4.9 on January 30 (02:28 UTC). An average of 4 events d −1 were recorded during this period, but the seismic activity is better described by a series of burst of activity on January (5,16,18,30), February (3,12,19) and March (13,15), 2011. We locate 157 earthquakes in January 2011, and the area of the main activity remained in the same location as in December. Most of the events occurred in the middle of the graben and are aligned along the bounding faults, in particular the faults that run along the NE-N-NW side of the graben as shown in Fig. 9(c). We note some concentrations of events at the location of the volcanic cones near the middle and NW end of the central graben. In addition, some small activity along the NE-SW trend to the NW of the graben and two events were located near segment G at the eastern end of the central domain. Most of the moderate size events (M L = 4.0-4.9) are located in the central graben accompanied by normal faulting mechanisms with nodal planes parallel to the SE-NW trend (N100 • E -N140 • E) which is in agreement with the structures in the area (Fig. 9c).
The seismic activity during February and March 2011 has been clustered in seven distinct groups: the main seismic activity in the graben, a swarm centred on fan 2, and five small swarms to the east and west of the activity area (Figs 3a and 9c). Moreover, the slow migration of the seismic events is well represented by the seismic activity along the central graben in February. The event migration starts near the volcanic cone (location of the swarm onset) which is located near the NW end of the central graben and slowly moves southeastwards along the graben. This slow migration is accompanied by strong events with a normal faulting mechanism at the SE end of the graben at the vicinity of the base of the southern axial valley wall. This migration ends beyond this point by reaching an area located at shallow depth (<3 km) below the top edge of the wall at (044.253 • E, 11.923 • N). Eleven of the 21 moderate size events are located at the SE end of the segment, where eight of them are characterized by normal faulting along SEtrending planes (Fig. 9c). Two of the events are located near the volcanic highs bounding the central graben from the east and characterized by strike-slip mechanism with a small normal component with one EW-trending nodal plane (Fig. 9c). One of the events is located below the southern wall of the ridge and is characterized by a normal faulting mechanism with a strike-slip component.
The seismic activity in March is accompanied by small size events concentrated on fan 2 near the western channel and the SE end of the central graben. While the activity on fan 2 occurred at a small rate of 1-7 events d −1 , the seismic activity at the SE end is characterized by a strong burst of events on March 13 (Fig. 9c).
From 23:54 UTC March 15 to 00:39 UTC March 24, a swarm of 17 earthquakes at a depth ranging from 6.2 to 10.5 km is located at the base of the southern wall of the axial valley near segment N in the western domain. This swarm is characterized by small to moderate magnitude (2.5 ≤ M L ≤ 4.2) events distributed within a narrow, 6 km long, N115 • E-trending zone. Another swarm of 12 earthquakes occurred to the west from March 23 to March 26 near segment P. The small to moderate magnitude events of this swarm (2.1 ≤ M L ≤ 4.0) occurred at a depth ranging from 10.8 to 16.9 km. To the east, we observe some activity in the same locations that the seismicity recorded in December near fan 1 and segment G.

Bursts of activity
The activity recorded over the whole episode is characterized by several bursts of activity presented by Fig. 10. A striking feature shared by most of these bursts is that the associated epicentres are located within segment J, but also define an alignment parallel to the segment axis. These bursts last for a few hours, except for the first main burst on November 14, which continued on the following day. This latter one is particularly interesting, and especially the chronology of the hypocentre distribution as described above (Fig. 8).
All the others bursts are smaller in terms of number of events, duration and size of the affected area. Each time, the released energy is between ∼4.1 × 10 15 Nm and ∼4.6 × 10 16 Nm, corresponding to one event with a magnitude between 4.4 and 5.1, respectively. Two kinds of bursts can be identified. First the bursts of events concentrated at the centre of the segment J (2011 January 16 and 30) for which most of the hypocentres are deep (>4 km). Second, the bursts of events spread from the segment centre and its ends (3 and 5 December, 13 March) and where most of the hypocentres are shallow (<3 km, see Fig. 10). For the three bursts affecting both the centre and the ends of the segment J, significant lateral migration is unclear due to the uncertainties on the locations of the small events. However, at the scale of the segment, we do observe that the first events of each burst are located at the centre whereas the later events are located far from the centre. This would correspond to a migration rate ranging from 0.5 to 5 km h −1 .

D I S C U S S I O N
Together with high-resolution bathymetric data (Audin 1999), our detailed analysis of the seismicity during the November 2010-March 2011 episode in the Aden Gulf allows the clear identification of the activated structures belonging either to the ridge walls or the central valley. Both the space-time evolution and the characteristics of the seismicity confirm (1) the occurrence of volcanic activity at the centre of the ridge valley suggested by in situ measurements, and reveal (2) the occurrence of dike intrusions associated with a rifting episode, (3) the opening dynamics affecting one individual segment and (4) the interaction between the opening spreading segment J and the complex nascent transform zone.

Volcanic activity
Using a regional seismic network, it is difficult to interpret that underwater seismic activity is related to magmatic activity, that is, volcanic eruption and/or (a) magma intrusion(s), without any clear in situ field observations such as identification of fresh lava flows, bathymetric or morphological differences (e.g. Fox et al. 1992;Edwards et al. 2001). In our case, several observations suggest that this seismic episode corresponds to a significant magmatic activity, with probable lava flows over the sea floor within the ridge valley.
Our analysis of the space-time distribution of the earthquakes with respect to the main morphotectonic features within the active segment J reveals the role of the volcanic centres on the dynamics of this seismic event. In the case of D-MH segment activity in the Afar Depression, a ∼4 to 10 km deep magma chamber is inferred from transient surface displacements in a very local area during the inter-diking periods, which are interpreted as movements of ascending magma replenishing the central reservoir located slightly south of the main volcanic chain marking the topography (Grandin et al. 2010). The lack of earthquakes would result from specific thermal conditions due to a focused zone of melt supply and/or the inadequate seismic network for the detection of small-magnitude events. Along segment J of the western Aden Ridge, two types of gaps of seismicity are distinguishable: the small circular areas below the identifiable volcanic cones, and the large NW-dipping 1.5 km thick area (Fig. 4). Keeping in mind the larger uncertainties on the focal depth, in the same vertical columns of the small gaps, we see mostly moderate-magnitude deep events (located between 4 km and 12 km depth) and superficial small-magnitude events (located above 4 km) (see Fig. 4b for example). At the beginning of several peaks of activity, and particularly for the November 14, December 5 and January 30, several events occur at depth, below the seismic gap located underneath the central volcanic cone (Figs 8  and 10), providing probable evidence for the vertical ascent from deeper levels through a 'pipe', as observed in tomography images at the centre of oceanic segments (e.g. Canales et al. 2000;Madge et al. 2000;Dunn et al. 2005), either from a deep crustal magma chamber as inferred below the Krafla volcanic centre (Tryggvason 1986) or from a melting zone in the mantle. The large concentration of small magnitude and shallow earthquakes, particularly for the burst of January 16, could reveal faulting activity above the crustal magma reservoir, such as caldera faults or nearby faulting. This activity is associated with pressure changes into the reservoir, as it was shown during the inter-rifting period in the Asal-Ghoubbet rift in Djibouti, 20 yr after the 1978 Ardoukoba diking episode (Doubre et al. 2007a,b).
(1) The low-frequency events. The LF earthquakes, also called long-period or volcanic hybrid earthquakes have been the subject of many studies since they are commonly associated with magmatic activity and/or volcanic eruptions (Malone et al. 1983;Chouet et al. 1994;Lahr et al. 1994;Harrington & Brodsky 2007). The physical explanation of such events is still a key question and is usually attributed to the physical properties of the source material (liquid magma, hydrothermal fluids or gases) which generate LF signals during its movements (e.g. Aki & Koyanagi 1981;Chouet 1988;Leet 1991;Chouet 1996) or resonance of fluid-filled structure (Aki et al. 1977;Chouet 1985). In addition to the source, the LF content of the volcanic earthquakes is strongly dependent on the path effects (Harrington & Brodsky 2007;Cesca et al. 2008). LF signals from hybrid earthquakes have been observed and recorded over long durations across long distances during the 2009 earthquake swarm in Harrat-Lunayyir Saudi Arabia and are interpreted to be due to a dike intrusion (Pallister et al. 2010). In Afar, both types have been reported during the September 2005 D-MH magmatic events (Cote et al. 2010) and long period events have been observed at Kilauea Volcano in Hawaii (Cesca et al. 2008). LF earthquakes are observed in Icelandic volcanic regions, which are mostly due to magmatic intrusions or extrusions (Brandsdöttir & Einarsson 1992). In the western Aden ridge, the spatial distribution of the three groups of earthquakes based on their frequency content indicates that the LF events are located within segment J where volcanic cones are distinguishable in the bathymetry. The first large earthquake recorded on November 14 and located at depth below the main volcanic area is of very LF. Since we previously associated the occurrence of such earthquakes with the ascent of magma towards the shallow reservoir, they may result from the vibration of the magma conduits.
(2) The frequency-magnitude distribution. The frequencymagnitude distribution of the earthquakes recorded during the November 2010 episode shows high b-values, reaching 1.41 for the earthquakes located within segment J (Aux Mat-5, Supporting Information). Such high b-values are usually found in volcanic and hydrothermal areas, where the occurrence of low-magnitude earthquakes is high and large events do not occur. We find a b-value of 1.20 for the rest of data, confirming that the seismic activity along the segment J is specific (Aux Mat-5, Supporting Information). Taking into account that Hoffstetter & Beyth (2003) estimate a b-value of 1.05 for the period 1960-2000 in the Gulf of Aden, from Djibouti to E046 • , this confirms that volcanic activity is temporally restricted to our period of study. We note that the Saudi rifting event was associated with a b-value of 1.20 (Pallister et al. 2010), however except for the surface deformation pattern deduced from InSAR and consistent with a dike opening, the volcanic activity is not clearly confirmed. The high b-value is also indicative of the evolved nature of the spreading area, since an increase in b-value is generally observed from continental to oceanic rifts. For example, b-values have been reported to increase from 1.0 in Tanzania and Kenya, to 1.06 and 1.13 in Afar and the Main Ethiopian rifts respectively (Kebede & Kulhànek 1994;Hoffstetter & Beyth 2003;Keir et al. 2006).
(3) In situ observations. Finally, constraints provided by a global communications company running sea bottom cables in the Gulf of Aden report multiple cable breakages in the region of our seismic swarms one year after the period of our study, between January and March 2012. Taking into account the melting temperature of 1084 • C for the copper, used as a protection components of the cable, the very high temperatures in this area strongly suggest that volcanic or intense hydrothermal activity occurred. The wire did not show sign of lava residue. This in situ information is very important, since it indicates that the volcanic episode affecting segment J lasted at least 15 months, whereas the seismic activity recorded by both Djiboutian and Yemenite seismic observatories remained moderate and regular in time (the Arta Geophysical Observatory recorded more than 300 earthquakes in the area from April 2011 to December 2011 with 19 events with magnitude above 4.0, Observatory staff, private communication).

New diking event
By comparing our results with the seismic results acquired during previous seismovolcanic episodes in subaerial rift zones, for which both geodetic (e.g. Abdallah et al. 1979;Wright et al. 2006;Grandin et al. 2009Grandin et al. , 2012 and field constraints exist (e.g. Le Dain et al. 1979;Rowland et al. 2007), we infer that the seismic swarm in segment J of the western Gulf of Aden has nearly all the characteristics of seismic swarms related to diking episodes. This seismic activity and previous episodes share specific particularities described in the following, from which we attempt to determinate some quantitative characteristics.
(1) Peaks of activity. The evolution of the seismicity associated with rifting episodes is organized temporally by episodic peaks in activity, lasting from a few hours to a few days (Belachew et al. 2011;Grandin et al. 2011). During both the Ethiopian D-MH and the Icelandic Krafla sequences the peaks of activity have always been associated with individual diking events confirmed by geodetic measurements (e.g. Grandin et al. 2011). Between these peaks of activity, the duration of quiescent period ranges from one month to one year (see Wright 2012 and references therein). Ascending movement of magma to replenish the crustal magma chamber between dike injections occurred during the inter-diking periods (Grandin et al. 2010). During the five months of monitoring of the 2010 Aden episode, in addition to the main first swarm, we identify several small peaks associated with epicentres aligned along the rift axis (Fig. 10). Most of the hypocentres are concentrated within the three first kilometres of the sea floor, and could therefore correspond to dike intrusions. However, the time span between them ranges from one day to ∼one month. If these seismic peaks are all dike-induced, then the implication is that the necessary time to fill the magma chamber is very short, suggesting either a smallsize reservoir or a significant magma supply from the mantle. If these peaks are not dike-induced then the main faults are seismically slipping at shallow depth responding to the regional tectonic stresses.
(2) Magnitude. Most of the seismovolcanic episodes associated with dike intrusions are characterized by a seismic sequence starting with earthquake(s), whose magnitude ranges from 5.0 to 5.5. In western Aden, ten events of magnitude above 5.0 occurred along segment J. In the case of the D-MH episode, 15 large events (5.0 ≤ M ≤ 5.5) were recorded during the first mega-dike intrusion in September 2005 (Ayele et al. 2007) and at least 25 in the following month (Ayele et al. 2009). Among the starting series, one of them (M = 5.2) was recorded in the same half-hour as the explosive eruption near the Dabbahu edifice, which might indicate, the volcanic origin of the event . In the case of the 1978 Asal diking episode, the two main shocks (M ≥ 5.0) were located in the sea, in the Ghoubbet Gulf corresponding to the rift zone of the segment, across which the maxima in extensional deformation was measured (Ruegg et al. 1979). This range of magnitude is typical of the sea floor spreading seismicity (Sykes 1970;Solomon et al. 1988) and reveals the thin brittle crustal layer along the oceanic ridge. As mentioned above, the occurrence of the first large event (M = 5.4, 2010 November 14 06:32 UTC) located below the main volcanic centre, at a depth of 7.5 km, therefore seems to be associated with the dynamics, and probably the replenishment of the inferred underlying magma reservoir. The other events with magnitude above 5.0 are along the inner floor of segment J or in the vicinity of normal faults, suggesting that these events are associated with dike-induced faulting. Looking at the space-time decomposition of the activity, we first consider that the events recorded on 2010 November 14 and only located within segment J (see limits on Fig. 9a) are associated with the first intrusion. The total seismic moment release of this first series is ∼7.2 × 10 17 Nm, equivalent to an event of magnitude M w = 5.9 (Hanks & Kanamori 1979). Adding the events of the subsequent three days, the total seismic moment reaches 9.1 × 10 17 Nm. The seismic moment of the following peaks in activity do not exceed 4.6 × 10 16 Nm.
The increased number of monitored dike intrusions allowed Nobile et al. (2012) to examine the associated seismic moment/geodetic moment ratio (SM/GM) at a range of rift stages. With the exception of the Dallol intrusion in northern Afar (Nobile et al. 2012), this ratio does not exceed 4 per cent in incipient oceanic rift segments of Afar and reaches values of 35.0 in continental context (Calais et al. 2008;Pallister et al. 2010). Following this trend, we infer a SM/GM ratio between 1.0 and 5.0 per cent for the oceanic episode in segment J in western Aden. Since the first intrusion during November 14-17 is associated with a seismic moment release of 9.1 × 10 17 Nm, this results in a geodetic moment in the range of 1.82 to 9.1 × 10 19 Nm. We followed the approach of Shuler & Nettles (2012) to estimate 0.58-2.9 m of horizontal opening by normal faulting based on a relation of Solomon et al. (1988) which links the moment to the dimensions and geometry of the rupture. This is consistent with the lack of fringes in the interferograms built from ALOS-PALSAR data on land, on the southern Yemenite coasts. From simple direct elastic modelling of opening crack at a depth between 0 and 10 km, the width of the main intrusion below the segment J associated with the main swarm does not exceed 5.0 m.
(3) Along-axis alignment of seismicity. The epicentres located at the vicinity of the segment J draw a very clear alignment perpendicular to the extension direction and concentrate within an ∼8 km wide band parallel to the rift zone (Brandsdöttir & Einarsson 1979;Keir et al. 2009;Belachew et al. 2011;Grandin et al. 2011). During the dike sequence of D-MH, each intrusion following the September 2005 intrusion is associated with individual earthquakes swarms with a length of 10-13 km, and a width of 2-4 km, corresponding to the width of axial faults triggered by each intrusion (Grandin et al. 2011). From different studies in D-MH, the length of the seismically active band broadly corresponds to the length of the subsiding zone revealed by the deformation field deduced from InSAR studies and field observations (Wright et al. 2006;Rowland et al. 2007;Grandin et al. 2009). In addition, the seismic cluster length approximates the length of the intrusion at depth. In the Gulf of Aden episode, the total length of the N115 • -trending earthquake clusters along segment J is 40 km, and therefore suggests a similar length subsurface intrusion extending from the northern margin to the foot of the southern ridge wall. The first swarm corresponds to intrusions below both the northwestern and southeastern halves of segment J, while each of the following peaks occur in one direction. This pattern is similar to the intrusion sequence at D-MH, with the first mega dike intruded below both halves of the segment, with the following intrusions shorter and affecting one half of the rift. This pattern has been interpreted as the result of a high state of stress and/or a highly pressurized magma reservoir at the beginning of the sequence that induces the breaking of the crust along the whole segment length (Grandin et al. 2010). The following intrusions occur more or less in the same along-axis sections of the segment, suggesting that the deficit of opening is not fully filled after each of them. Also, we note that for most of the bursts the locations are next to the inferred crustal magma source below the main volcanic centre, which is consistent with a larger density of intrusions observed near the crustal magma reservoir along eroded volcanotectonic segments (Speight et al. 1982) and this confirms the role of the central crustal magma reservoir on the location of the highest magmatic dilatation.
From the comparisons of the displacement fields deduced from InSAR and field data in studies of the D-MH diking sequence, the width of the seismic band is indicative of the width of the subsiding zone and encompasses all the normal faults activated during the sequence and identified by the field observations (Rowland et al. 2007). Following the results of the InSAR inversions (Wright et al. 2006;Grandin et al. 2009Grandin et al. , 2010Hamling et al. 2009), the width of the seismically active band is wider than the opening width of the intrusion at depth, and underlines the dike-induced activity on normal faults located above the crack tip (Rubin & Pollard 1988). In our case, the 8 km wide alignment of the earthquakes along segment J suggests that the bounding normal faults already identified in the bathymetric data at segment J are activated in response to the intrusion, and that the construction of the topography of the segment probably results from a long-term repetition of dike intrusions along the segment axis. We note that the seismic band is clearly centred at the segment axis, suggesting that the 2010-2011 intrusions took place at the centre of the rift zone with no offset with respect to the topography. This contrasts the 2005 mega-dike in D-MH which was offset to the east of the rift axis (Grandin et al. 2009). Following Rubin & Pollard (1988) and assuming 65 • -dipping normal faults, the maximum in opening is approximately at 8 km depth. The activated area narrows after December 15 (Figs 8b and c), which could be the result of the evolution of the network configuration. Alternatively, this could also reveal the asymmetry of the along-axis crustal structure, with more superficial intrusions reflecting a thinner brittle crust in the southern half of the segment as can be inferred from the asymmetry of the bathymetry.
(4) Hypocentral distribution and relation with volcanic centres. We observe that the seismogenic layer extends from the seafloor to ∼12 km. The occurrence of small magnitude earthquakes at shallow depth suggests that the state of stress in the crust is high and that the dikes intrude at shallower levels because of the brittle crust is thin. Along segment J, 42 per cent of the events are less shallow than 3 km. In general, deep earthquakes occurred at the beginning of bursts. This is interpreted previously as a quick replenishment of the magma reservoir before the intrusion implying that the time residence into the shallow magma chamber is short since the magma transits in a few hours from the deep level to the magma chamber and to eventually intrude into the rift zone of the segment. This magma transfer of a few hours is much faster than the transfers of a few months in other volcanic areas (e.g. Peltier et al. 2009;Druitt et al. 2012). Recharging of the magma chamber just before the intrusion could explain a part of the discrepancy between the volume of magma intruded in the rift zone and the deflation volume of the magma chamber modelled as a Mogi source from geodetic data (e.g. Wright et al. 2006;Grandin et al. 2009;Nobile et al. 2012).
(5) Migration of seismicity. The primary evidence for dike propagation is the migration of seismicity away from the main volcanic centre towards the segment ends. Magma likely propagates into dikes following a process of hydraulic fracturing with the central crustal reservoir being the main pressure source (e.g . Gudmundsson 1998;Doubre & Geoffroy 2003). The deceleration of the seismic migration is consistent with a decrease in pressure within the reservoir coeval with the magma withdrawal. The migration rate of the seismicity in the western Gulf of Aden is 1-4 km h −1 , which is comparable to the migration rates reported for D-MH where typical velocities were 0.54-5.4 km h −1 range (Ayele et al. 2009;Keir et al. 2009;Belachew et al. 2011). Our rates are higher than the 0.36-1.8 km h −1 reported for Krafla-Iceland (Einarsson & Brandsdöttir 1980). We also note that after the intrusion reaches its final length, we observe the occurrence of small events along the dike (Fig. 10). This pattern is similar to what was observed in D-MH (Keir et al. 2009;Belachew et al. 2011;Grandin et al. 2011) and interpreted as earthquakes resulting from the inflation of the dike or by the thermal stresses near the dike walls.

Activity along one individual segment and magma intrusion as the main extension process
At this longitude, the structure of the Aden Ridge is characterized by left-stepping en-echelon segments which overlap by more than ∼20 km, corresponding to ∼80 per cent of their length. Nevertheless, the seismic activity during the five months following the sequence of November 2010 is confined within one individual segment. Such rifting episodes are usually seen as transient displacement accommodating the far field plate motion along the plate boundary. The total seismic moment for segment J during the five months under study is to ∼2.76 × 10 18 Nm (Fig. 7), from which we estimate a geodetic moment of ∼5.5 to 27.6 × 10 19 Nm, and magnitudes of horizontal opening of 1.76-8.8 m along the whole segment. This amount of opening is equivalent to ∼135-677 yr of spreading at this part of the ridge where the spreading rate is 1.3 cm yr −1 . The structure of the ridge axis with en-echelon segments overlapping over 80 per cent of their length, implies that if there is a deficit of opening along the divergent plate boundary, the concentration of extensional stresses along the ridge due to the far field plate motion is distributed over several overlapping en-echelon segments. The activity during the 2010-2011 episode occurring in one segment only and with adjacent segment quiescent provides evidence that the release of the stresses is controlled by the discrete supply of magma into the plumbing magma system of individual segments (Madge et al. 2000). Therefore, this observation shows that the dynamics of an oceanic spreading segment is independent from the adjacent segments. In addition, our work shows that the magma system feeding the crustal reservoir(s) of one segment is probably isolated from similar systems below other segments, even when adjacent segment overlap a significant distance. Further, this strongly supports a models by which the dynamics of a divergent plate boundary is controlled by the presence of a series of punctuated mantle sources connected independently to the shallow magma reservoirs of the segments.

Nascent transform zone
The seismic activity and focal mechanisms help constrain the interaction between rift opening and transform faulting. Near canyon 2, we note that several events with a strike-slip focal mechanism show both senses of horizontal slip along parallel planes. Oceanic transform faults are found along all spreading ridges except at along ultra-slow and ultra-fast settings. According to recent reviews (Choi et al. 2008;Gerya 2012), their origin remains unclear. Transform faults can nucleate spontaneously in accreting and cooling plate of wax as demonstrated by old analogue models (e.g. Freund & Merzer 1979). Analogue models of the oblique rifting specific to the Gulf of Aden evidence that such rift segmentation is structured by numerous transfer and transform zones. Block rotations cause steep shear in the rift centre. The steep faults are roughly parallel to the opening direction and well oriented to become potentially transform faults (Autin et al. 2010b).
The first group of strike-slip focal mechanisms include an E-W-trending focal plan, which is consistent with the dextral faults evidence in the bathymetry data and reported by previous studies done in the area (e.g. Audin 1999;Dauteuil et al. 2001). The focal mechanisms of earthquakes, mainly located along the northern end of segment J and on the southern shoulder of segment M, also show horizontal slip along the EW-trending planes. Such EW-trending faults are visible in the bathymetry, and their formation could be attributed to the E-W component of the regional stress due to oblique extension. The second group of focal mechanisms encompass one nodal plane mostly parallel to the ∼N40 • E-trending relative motion of the Arabia and Somalia plates. The earthquakes belonging to this group are located at the western end of the active segment J near the mouth of canyon 2 and draw an alignment perpendicular to the direction of the axis of the segments and parallel to the regional maximum tensile stress. These earthquakes result in right-lateral strike-slip motion along ∼N039 • ± 8 • E-striking faults planes that we interpret as an incipient transform fault between two ridge segments. This nascent transform is located where a change in ridge-axis trend occurred (Fig. 1b). Following previous studies, the transform area is the part located between 43.91 • E and the location where recent lava flows are observed east of 44.20 • E (Dauteuil et al. 2001;Hébert et al. 2001). However, the transform part is the area located between the two NNW-SSE trending normal faults at the locus of the two canyons cutting the northern shoulder of the axial valley. This area is the westernmost part of the E-W trending axial valley where well-developed en-echelon N115 • E-trending basins exist. West of this area, the trend of the axial valley is N075 • E with deep basins covered by a thick layer of interrupted deposits and the en-echelon structures built on the southern ridge wall (Choukroune et al. 1988), but not at the main valley.
The earthquakes belonging to third group of strike-slip mechanisms (N065 • E ± 8 • /N155 • E ± 8 • ) occur at the vicinity of the faults bounding the canyons, and especially along the western one (canyon 2, Fig. 3c). Near this latter canyon, tectonic structures are parallel to the ∼N150 • -155 • E direction, which corresponds to the direction of one of the nodal planes of the focal mechanisms, suggesting that left-lateral strike-slip faulting may occur along these planes (Figs 4c and 8c). It is not clear to which stress regime such movement could be attributed. Choukroune et al. (1988) proposed a model of N145 • E-striking rifts for the Gulf of Tadjura based on CYADEN dives results. Lépine & Hirn (1992) described left-lateral strike-slip motion on N150 • E-trending faults, resulting in clockwise rotations of the blocks between the faults (Fig. 1b). Manighetti et al. (1997) proposed two different kinematic models to explain the successive attempts of the Gulf of Aden ridge to propagate through Bab-El Mandeb strait resulting in the construction of the two canyons and bounding structures (Fig. 1). Their finite kinematic model extends from 5 to 0.7 Ma and explains a relative motion between Arabia-Somalia (1.73 cm yr −1 parallel to N039 • E direction), Arabia-Danakil (0.7 cm yr −1 along N65 • E) and Somalia-Danakil of 1.2 cm yr −1 along N025 • E direction. From these models, no clear mechanism could explain the eastward displacement of the fan 2 from the mouth of canyon 2. Our results which show both leftlateral and right-lateral motion at the same location and on distinct orientations may explain the displacement of fan-deposits 2 together with the curvature of the two canyon channels (Fig. 3a). Combining observations and results, we propose that this area is a transfer zone between the E-W and N75E-oriented ridges and corresponds to the western boundary of the transform zone between true oceanic lithosphere in the east and a transitional lithosphere in the west.

C O N C L U S I O N S
By combining data from permanent and temporary networks in Djibouti and Yemen, we obtained accurate locations of the earthquakes associated with the large seismic episode started on November 2010 in the western Gulf of Aden. From precise location process, the analysis of the space-time distribution of the magnitude 2.1-5.6 events, their source mechanisms and their frequency content, we are able to distinguish several seismic swarms related to distinct tectonic features. A large part of seismicity is concentrated along an active spreading segment, and the characteristics of this seismicity share numerous similitudes with recent dike intrusions monitored along subaerial segments. The dynamics of this opening segment confirms the mechanisms and processes involved during an active rifting episode, in particular the crucial role of the magmatic system. The seismicity struck only one segment of a highly oblique, en-echelon and overlapping system, and suggests the independence of the magma plumbing systems below each discrete ridge segment.
Finally, the opening dynamics of the segment induces movement along oblique planes related to a transfer zone, that we interpret as a nascent transform zone. This transform zone corresponds to the western limit of the 'true' oceanic lithosphere in the Gulf before the transitional or continental lithosphere of Afar in the vicinity of the Gulf of Tadjura. Therefore, this episode demonstrates the close interaction between the active rifting and/or spreading processes and the development of the transform faults in the oceanic lithosphere. Our results may suggest that transform fault onset and evolution may be controlled by the opening of the segment and more precisely follow the focus magma supply on one segment. At the scale of the structure of a spreading ridge, the magmatic plumbing system could trigger the initiation of a transform zone.

A C K N O W L E D G E M E N T S
The instruments temporary deployed in Djibouti and Yemen belong to the French national pool of portable seismic instruments Sismob-RESIF, the IUEM-UBO and EOST-IPGS temporary seismic networks. This work was supported by the following projects funded by the French National Agency (ANR, France): ANR-YOCMAL, ANR-DoRA and Rift2ridge, CNRS-INSU-PICS Yemen, GSMRB Yemen and Actions Marges. We thank the French Embassy in Sana'a (J.G. Sarkis, J. Dechezlepretre and C. Bousquet) and the French army in Djibouti for logistical support during the installation of the networks. This work could not have been done without the precious collaboration with OGA and CERD in Djibouti (OGA), SVOC and GSMRB in Yemen. We are grateful to Encarnation, and Regis Coquelin de Lisle, the local governors and the people of the Yemen governorates for their help during the field work. Our deep thoughts go to our colleague Kassim Mohamed, now deceased, and his family. We gratefully acknowledge the information provided to us regarding the damage to submarine cables in the Gulf of Aden during 2011. Due to reasons of confidentiality we are unable to release details of the company or individuals that pro-vided us with the information. DK is supported by NERC grant NE/L013932/1.