Introduction

Geological settings

The Afar triple junction (Fig. 1) is located in East Africa connecting three branches of a complex rift system. This area is an example of the last stage of continental rifting and the early stage of seafloor spreading, leading to the creation of oceanic crust (Mohr 1970; Tesfaye et al. 2003; Rychert et al. 2012). The Afar hotspot is located at the triple junction between the Red Sea rift, the Gulf of Aden, and the Ethiopian rift zone (Mohr 1970; Legendre 2013). The Red Sea rift is considered as the boundary between the Arabian and African (Nubian) plates, whereas the Gulf of Aden separates the Arabian and African (Somalian) plates (Bird 2003). The Ethiopian rift zone (also called East African Rift System), an active continental rift zone in East Africa, is a divergent tectonic plate boundary (Kusky et al. 2010) dividing the African Plate into two tectonic units: the Nubian and Somalian plates (Garfunkel and Beyth 2006). The Afar triple junction accommodates the divergent motions between the Arabian, Nubian, and Somalian plates along the Red Sea, Gulf of Aden, and East African rifts. The kinematics of the Afar triple junctions are generally studied using long-term deformation and geodetic observations (McClusky et al. 2010) as well as geodynamical modeling (Koptev et al. 2018).

The opening of the Red Sea and the Gulf of Aden started in the Late Eocene–Early Oligocene (Ghebreab 1998) and lead to the split between Africa and Arabia in the Early Miocene (Joffe and Garfunkel 1987). The early stage of the collision between the Mediterranean and Biltis segments (McQuarrie et al. 2003) occurred synchronously with the massive eruption of flood basalt associated with the Afar plume at around 30 Ma (Hofmann et al. 1997).

Fig. 1
figure 1

Tectonic setting of the Afar triple junction. The red triangles are active Holocene volcanoes (Venzke 2013). The magenta line represents the plate boundary (Bird 2003). The black circles represent the locations of the epicenters of Earthquakes (M\(_\text{w} > 4\)). Most of the Earthquakes in the region have shallow depths. The black text framed in magenta provides the names of the different plates (Bird 2003): AFn, African (Nubian) plate; AFs, African (Somalian) plate; Ar, Arabian plate. The green text framed in blue provides the names of the three rifts: ER, Eden rift; GA, Gulf of Aden; R,- Red Sea rift. The blue text framed in brown (Af) indicates the location of the Afar triple junction

Full size image

Previous geophysical studies

Several geophysical tools have been employed in this region to investigate its tectonics. In recent decades, the Global Positioning System (GPS) has been widely deployed in this region to investigate the relative motions of the different tectonic units (Walpersdorf et al. 1999; McClusky et al. 2010; Doubre et al. 2017).

The main results of geodetic measurements along the Afar triple junction have suggested counterclockwise rotation with respect to the Nubian Plate (with a rotation pole located in the central Red Sea). Doubre et al. (2017), based on a combination of GPS and Interferometric Synthetic Aperture Radar (InSAR) measurements, suggested that the northern part of the Somalia plate evolves independently from the entire African (Somalian) plate.

Previous seismological models from body-wave tomography (Benoit et al. 2006; Koulakov 2007) and surface-wave tomography (Sebai et al. 2006; Guidarelli et al. 2011; Legendre 2013) corroborated the presence of negative velocity anomalies beneath the triple rift systems. Also, they have suggested that the Afar hotspot and other volcanic provinces in the region (Wignall 2001; Mège and Korme 2004) are originating from the African superplume (Ni et al. 2002; Simmons et al. 2007), upwelling from the lower mantle.

The crustal thickness in the region is of great interest as it is crucial to unraveling the mechanisms of the tectonic evolution of the Afar triple junction. The Moho depth has been widely investigated using the receiver function technique (Dugda et al. 2005; Dugda and Nyblade 2006; Hammond et al. 2011; Reed et al. 2014; Thompson et al. 2015) and the crustal thickness ranges from ~ 15 km beneath the spreading centers up to ~ 40–45 km outside the continental break-up area (Hammond et al. 2011).

Previous anisotropy observations have mainly been achieved by the shear-wave splitting of teleseismic (Gashawbeza et al. 2004; Walker et al. 2005) or local Earthquakes (Keir et al. 2011). Several other shear-wave splitting measurements performed in the region and surroundings have been compiled into international shear-wave splitting databases (Wüstefeld et al. 2009; IRIS 2012). Deformation-related structures generally explain the anisotropic pattern in the Afar region. The general orientation of the anisotropies is associated with mantle flow; in contrast, subtle changes in the fast direction are related to the presence of fluids, isotropic media, or vertical mantle flow. Dyke-induced faulting and alignment of melt pockets near volcanic centers also have very distinct anisotropic patterns in the region (Gashawbeza et al. 2004).

Tectonics of the Afar triple junction

The tectonics of the Afar triple junction is dominated by the motion of three plates: Arabia, Africa (Nubia), and Africa (Somalia). The first rift to open is thought to have been the Gulf of Aden around 30 Ma (Ghebreab 1998). Simultaneously, the intense activity of the Afar hotspot (Schilling et al. 1992; Hofmann et al. 1997) led to a massive amount of flood basalt in the region. The opening of the Red Sea rift (Kusky et al. 2010) is thought to have started between 24 and 21 Ma. This event was followed by high magnetic anomalies (Wignall 2001; Mège and Korme 2004) and subsequently the opening of the Ethiopian rift between 18 and 15 Ma (Kusky et al. 2010). The local seismicity (Abdallah et al. 1979; Illsley-Kemp et al. 2018), surface displacement (Walpersdorf et al. 1999; Kusky et al. 2010) and geometry of the active faults (Doubre et al. 2017) were used to fine-tune the geometry and complexity of the tectonic units interacting in the region. Further information from body-wave tomography (Benoit et al. 2006) and surface-wave tomography (Guidarelli et al. 2011), as well as receiver function results (Hammond et al. 2011), have provided additional constraints on the nature of these tectonic units, such as their lateral and vertical extensions and the presence of rigid blocks in the region. Further geophysical modeling (Simmons et al. 2007; Reilinger and McClusky 2011) has enabled the kinematics and dynamics of the region to be refined. However, conflicting models have been suggested to explain the lateral variations of the anisotropy in the Afar triple junction region. Gao et al. (2010) suggested that the optimal source of anisotropy is located at approximately 300 km depth (in the asthenosphere), whereas Keir et al. (2011) found that this anisotropy originates mostly from deformation-related structures in the crust and lithosphere.

In this study, we performed a systematic investigation of the crustal structure and mantle anisotropy in the Afar region by means of receiver function and shear-wave splitting analyzes. Knowledge about the crustal thickness as well as crustal and lithospheric anisotropies (Legendre et al. 2017, 2020) can provide helpful insight into the crustal and mantle flow affecting the regional dynamics (Legendre et al. 2016; Fan et al. 2020). The main contribution of this study is a total of 431 new shear-wave splitting measurements in the region as well as 34 measurements of crustal thickness and ratio of seismic compressional and shear-wave velocities (Vp/Vs). In the first step, we will describe the data and methodologies used in this study (“Data and methods”). “Results” presents our new findings based on crustal thickness, velocity ratio between compressional waves and shear waves as well as shear-wave splitting of the SKS phase. “Discussion” discusses the implications of our computed crustal depth, \(\frac{V_\text{p}}{V_\text{s}}\) ratio, and anisotropy in terms of the structures, composition, and evolution of local and regional features.

Data and methods

Seismological data

We searched for all available broadband seismic stations deployed in the study region (Fig. 2). In total, 227 stations were found, but only 223 used for receiver functions and 224 used for shear-wave splitting constraints provided sufficient data for further investigation. To compute the receiver functions, we selected each available Earthquake with Mw > 5.5 and an epicentral distance between 30° and 90°, and to perform the shear-wave splitting measurements at each station, we selected Earthquakes with Mw > 6.5 with epicentral distances ranging from 90° to 120°.

We selected the accessible Earthquakes independently for each station. The number of Earthquakes mainly varies with the station operating time. We selected 427 events among all stations to compute the receiver functions and 514 for the shear-wave splitting measurements. Waveform data were downloaded from global open-access data centers (mostly from the GEOFOrschungsNetz program (GEOFON), GeoForschungsZentrum Potsdam, Observatories and Research Facilities for European Seismology Data Center and Incorporated Research Institutions for Seismology Data Management Center) (Megies et al. 2011; Trabant et al. 2012).

Fig. 2
figure 2

Locations of the 227 seismic stations deployed in Afar region. The color code indicates the networks to which the stations belong

Full size image

Receiver functions

Receiver functions are time series representing the internal boundaries of the Earth near the receiver, converting the incident P waves into S waves, or vice versa. The receiver function method is well established and widely used to investigate crustal and upper mantle velocity discontinuities (Langston 1979; Ammon et al. 1990; Zhu and Kanamori 2000; Kind et al. 2015; Maguire et al. 2018).

Herein, we illustrate the method implemented by Eulenfeld (2020) using a single station, ZF-MAYE. Figure 3 summarizes the potential events selected for this station and satisfying the magnitude and epicentral distance requirements.

Fig. 3
figure 3

Selected events used to compute the receiver functions for a single seismic station (ZF-MAYE). The seismic station is indicated by a black triangle, and the locations of the events are marked by circles, color-coded by epicentral depth

Full size image

The ZRT coordinates (vertical, radial, and transverse components) were conveniently adopted for the nearly vertical incidence of teleseismic events. Assuming the location of the station–hypocenter pair is known, it is possible to infer the wave coordinate system. The L component is associated with the P-wave polarization, whereas the Q component is tied to SV-wave polarization, and the T component is related to the SH-wave polarization.

In this study, the particle motions of the incident and converted waves were separated using the wave coordinates. To derive the P-receiver functions, the teleseismic waveform of the L (or Z) component (assumed to resemble that of a P wave striking the conversion boundary) is deconvolved from the QT (or RT) components to separate the source effects of the converted phases. The resulting receiver function (the Q component in particular) reveals the delay time (relative to the P-wave onset) and relative amplitudes of the P waves converted into S waves (Ps) by the significant discontinuity, as well as multiples (P\(_\text{P}\)P\(_\text{mP}\) P\(_\text{p}\)S\(_\text{mP}\), P\(_\text{P}\)P\(_\text{mS}\), and P\(_\text{p}\)S\(_\text{mS}\)) caused by reverberations within the layer. P indicates a P-wave in the mantle, p indicates a P-wave ascending to the surface, S indicates an S-wave in the mantle, s indicates an S-wave ascending to the surface, whereas m indicates a top-side reflection from the Moho.

Those phases are not easily identified on a seismogram, because they generally have low amplitude and are hidden in the coda of teleseismic P-wave coda. A conventional approach is to build the receiver function, and to highlight those compressional motion converted into shear-polarized motion. The underlying assumption is that the vertical component is primarily related to the compressional motion. Thus, deconvolving it from the horizontal components removes the signature of the Earthquake source (present in both horizontal and vertical components of the seismogram), as well as all compressional reverberations. The resulting product of the deconvolution is the time series of shear motion within the teleseismic P-wave coda. More details about the deconvolution process are available at Walpole et al. (2014) and Kumar and Legendre (2021).

The horizontal components of the seismogram are rotated into LQT components (L is aligned in the direction of P wave propagation; Q is aligned in the direction of the S\(_\text{v}\) phase movement whereas T is aligned in the direction of the S\(_\text{H}\) phase movement).

The L and Q receiver functions are obtained by deconvolving the vertical component seismogram of a single teleseismic event from the L and Q components seismogram, respectively, using the time domain source equalization method of Langston (1979), as shown in Eq. (1):

$$\begin{aligned} \text{RF}_\text{L}(\omega ) = c(\omega ) \cdot \frac{\text{L}(\omega )}{\text{Z}(\omega )}, \nonumber \\ \text{RF}_\text{Q}(\omega ) = c(\omega ) \cdot \frac{\text{Q}(\omega )}{\text{Z}(\omega )}. \end{aligned}$$
(1)

Figure 4 displays the receiver functions computed at the selected station (ZF-MAYE).

Fig. 4
figure 4

(Left) Q component of the receiver function for the seismic station indicated in Fig. 3. (Right) All 155 receiver functions sorted by distance (purple dots) and back-azimuth (blue dots). (Top) Stacked receiver function

Full size image

H–\(\kappa\) stacking

We stacked all receiver functions that were reasonable for each station to produce an average receiver function for that station (Fig. 4, top). Stacking receiver functions enhances the signal-to-noise ratio of the trace, thereby increasing the amplitude of the converted phases in the receiver function. Theoretically, all converted phases should be visible on the stacked receiver function. However, some converted phases have very low amplitudes and are not easy to discriminate.

To ensure the reliability of the H–\(\kappa\) stack, we selected only the receiver functions that were of high quality and discarded those with more uncertainty. In a first step, we discarded all receiver functions that had a signal-to-noise ratio lower than 3. Only the receiver functions with clear direct P, PS, and PPPS phases were retained. We used an automated tool (Eulenfeld 2020; Kumar and Legendre 2021) to identify the potential phases in the stacked seismogram (Fig. 5, bottom) where only the phases with strong amplitude contrasts were easily recognized. We set the first peak with the highest amplitude as the direct P-wave arrival. The first peak with an amplitude more significant than 10% of the P-wave arrival amplitude was considered to correspond to the P-to-S converted (PS) wave. The second peak with an amplitude greater than 5% of the P-wave arrival amplitude was considered the PPPS wave.

Fig. 5
figure 5

(Bottom) Stack of all receiver function for station ZF-MAYE (orange line) and automatically picked phases (blue cross and their name in black font). (Top) H–\(\kappa\) grid search following Zhu and Kanamori (2000). The best combination of Moho depth and \(\frac{V_\text{p}}{V_\text{s}}\) ratio is indicated by a black circle

Full size image

The conventional H–\(\kappa\) stacking method (Zhu and Kanamori 2000) requires another phase: the PpSs or PsPS that arrive 15–20 s after the direct P arrival. However, this phase has a negative amplitude and was not always picked by our automated picking method. Therefore, we only use those three arrival times to constrain the Moho depth (H) as well as the \(\frac{V_\text{p}}{V_\text{s}}\) ratio (\(\kappa\)), following:

$$\begin{aligned}&H = \frac{(t_1-t_0)}{\sqrt{(\frac{\kappa }{V_\text{p}})^2-p^2} - \sqrt{(\frac{1}{V_\text{p}})^2-p^2} }, \nonumber \\&H = \frac{(t_2-t_0)}{\sqrt{(\frac{\kappa }{V_\text{p}})^2-p^2} - \sqrt{(\frac{1}{V_\text{p}})^2-p^2} }, \end{aligned}$$
(2)

where \(t_0\) is the arrival time of the P wave, \(t_1\) is the arrival time of the PS wave, \(t_2\) is the arrival time of the PpPs wave, \(V_\text{p}\) is set to 6.3 km/s (based on the IASP91 reference model from Kennett et al. (1995), \(\kappa\) is the \(\frac{V_\text{p}}{V_\text{s}}\) ratio. H is the Moho depth (in km). A 65% weight was given to the PS arrival time in Eq. (2) and 35% was given to the PpPs arrival time, following Zhu and Kanamori (2000). Note that the calculated crustal thickness is strongly influenced by the initial velocity model, which is also affected by the thickness of the sedimentary layer.

For each station where the stacked receiver function displayed enough amplitude on the selected phases, both Moho depth (H) and \(\frac{V_\text{p}}{V_\text{s}}\) ratio (\(\kappa\)) could be retrieved, as displayed in Fig. 10.

Some receiver functions failed to provide enough energy at the predicted arrival times for the PPPS phase due to the instability of the deconvolution (Kind et al. 2015). Of the 223 seismic stations that provided data for receiver function inversion, 34 also supplied measurements for the H–\(\kappa\) stacking method (Zhu and Kanamori 2000). Most stations provided receiver functions results, but we imposed highly restrictive quality checks to ensure that the H–\(\kappa\) stacking method would yield consistent results. The quality checks consist of two steps: first, the receiver function requires a signal-to-noise ratio higher than 3. Then all the phases used for the H–\(\kappa\) stacking method need to be consistently determined for each station specific stacked receiver function.

Shear-wave splitting

Shear-wave polarization anisotropy from core phases is a well-established concept (Ando et al. 1983; Fukao 1984; Obrebski et al. 2010). To download and process the data, we followed the workflow of STADIUM-py (Walpole et al. 2014).

For each seismic event that matches the magnitude and epicentral distance requirements (Fig. 6), ZNE raw data were downloaded and rotated from ZNE to ZRT components. The predicted arrival time of the SKS phase was calculated using the IASP91 reference model (Kennett et al. 1995). The traces were cut around the SKS phase predicted arrival time (from 30 s before to 45 s after the predicted time), as displayed in Fig. 7.

Fig. 6
figure 6

Selected events for the computation of shear-wave splitting parameters for a single seismic station (ZF-MAYE)

Full size image
Fig. 7
figure 7

Raw waveforms of unfiltered ZNE components for an event recorded at station ZF-MAYE (8/9/2009, 10:55:56.28, epicenter (33.1474, 138.0594), epicentral depth of 302.2 km, and M\(_\text{w} =\)7.1

Full size image

We then used an automatic selection approach to identify the beginning and ending times of the SKS wave on the radial component (Fig. 8). Characteristic functions calculated using short-term-average/long-term-average (STA/LTA) algorithms are often employed for automated detection of P and S waves and arrival time estimation (Walpole et al. 2014). This study used an automatic selection method based on an STA/LTA algorithm to determine the SKS arrival time accurately. Two threshold values in the characteristic function were set to pick the beginning and end times of the SKS wave accurately (Walpole et al. 2014): the first threshold was for the beginning of the SKS phase, and the second was for the end of the SKS phase (red and blue lines in Fig. 8, respectively).

Fig. 8
figure 8

Automatic selection of the SKS phase using the recursive STA LTA method for the trace depicted in Fig. 7. The red and blue bars indicate the beginning and ending times of the SKS phase, based on values of the characteristic function (bottom)

Full size image

Once the SKS phase had been selected, we performed a grid search (Fig. 9) for the best-fitting splitting parameters following (Walpole et al. 2014). To pass the grid search successfully, we set some thresholds on the errors in the phase (\(\delta \phi\)) and the delay time (\(\delta t\)). Measurements with excessive uncertainties in the fast directions (\(\delta \phi > 7^\circ\)) and the delay time (\(\delta t > 1.5 \text{s}\)) were also been discarded. Those threshold values in the phase (\(\delta \phi\)) and the delay time (\(\delta t\)) are quite conservative and ensure that the final measurements do not include too many uncertainties.

Fig. 9
figure 9

(Top left) North and East components of the seismogram (same traces as in Figs. 7 and 8), with corresponding particle motion (center). (Bottom left) Radial and transverse components of the seismogram, with corresponding particle motion (center). (Right) Grid search results used to find the best-fitting parameters for the phase and delay time

Full size image

This grid search (Fig. 9) yielded the fast shear phase (\(\phi\)) and a delay time (\(\delta t\)). The procedure was repeated at each station and for all respective events. In the end, the average of all the measurements was computed at each station.

Results

We obtained successful measurements of H and \(\kappa\) for a total of 34 stations. Figures 10 and 11 display the receiver function inversions for H and \(\kappa\), respectively. Our results suggest that regional thick continental crust (with a thickness of up to 38 km) exists beneath the African and Arabian plates, which agrees with previous measurements (Dugda et al. 2005; Dugda and Nyblade 2006; Hammond et al. 2011; Reed et al. 2014; Thompson et al. 2015). In contrast, a very thin crust (with a thickness as low as 21 km) is found along the three branches of the continental rifting and beneath the Afar triple junction. This thin crust is consistent with the very thin crustal thickness found beneath the spreading centers and the area of continental break-up (Hammond et al. 2011). Hammond et al. (2011) found that the crustal thickness ranged from 16 km down the spreading centers in the northern Afar region to 20–25 km near the active rift segments and observed a northward thinning of the crust beneath the Afar triple junction.

Fig. 10
figure 10

Moho depths computed by the H–\(\kappa\) stacking method (Zhu and Kanamori 2000) near the Afar triple junction

Full size image
Fig. 11
figure 11

\(\frac{V_\text{p}}{V_\text{s}}\) ratio (\(\kappa\)) computed by the H–\(\kappa\) stacking method (Zhu and Kanamori 2000) near the Afar triple junction

Full size image

\(\frac{V_\text{p}}{V_\text{s}}\) ratio is expected to be in the range 1.7–1.9 in the continental crust (Zandt and Ammon 1995; Musacchio et al. 1997; Kandilarov et al. 2015); most of the measurements in the region (Fig. 11), the value of \(\frac{V_\text{p}}{V_\text{s}}\) is reasonable. However, the measurements observed near the continental rift display relatively high values (greater than 2.0). Note that the receiver function results are not displayed for all stations, as their main primary purpose was to compute H and \(\kappa\) using the H–\(\kappa\) stacking method (Zhu and Kanamori 2000).

Figure 12 summarizes the shear-wave splitting measurements, and Table 1 lists them in detail. In most of the regions on the Nubian and Somalian plates, the fast direction is oriented N030E. This fast direction orientation is highly consistent with previous shear-wave splitting measurements (Wüstefeld et al. 2009; IRIS 2012). Along with some additional measurements, we provide detailed information by classifying our results into three categories:

  • no data, unreliable measurements (due to insufficient operation time),

  • new measurements computed in this study,

  • and Null measurements (if number of Null measurements (Wüstefeld and Bokelmann 2007) was more significant than the number of regular measurements).

Null measurements can occur for several reasons, such as limited data coverage and azimuthal distribution of seismic events. They can also occur if the wave propagates through an isotropic medium, if the wave encounters fluid, if mantle upwelling occurs with a vertical flow of peridotite, or if the initial polarization direction is oriented along either the fast or slow axis. As the incoming shear wave is not split (Savage 1999), it is not possible to determine or accurately. However, in the central part of the rift, there are several stations with fast directions perpendicular to those of the others, at N120E. For several stations, we could only retrieve null measurements or no measurements (e.g., beneath the Arabian Plate), which contrasts with the high scattering present in previous measurements (Wüstefeld et al. 2009; IRIS 2012). In the null measurements, we do not report any direction of (\(\phi\)) or (\(\delta t\)) in the station summary (Table 1).

Fig. 12
figure 12

Shear-wave splitting results for the broadband stations around the Afar triple junction. Blue: previous measurements available from international databases (Wüstefeld et al. 2009; IRIS 2012). Yellow: due to the data quality, no convincing measurements could be obtained. Green: null measurements. Black: each line indicates the mean fast polarization direction given by the median of all observations at one station. The circle size is proportional to the delay time

Full size image
Table 1 Summary of all shear-wave splitting parameters measured in this study for the Afar triple junction region
Full size table

Discussion

Lateral variations of Moho depth

The results we obtained by inverting the Moho depth receiver functions (Fig. 10) agree well with previous measurements (Dugda et al. 2005; Dugda and Nyblade 2006; Hammond et al. 2011; Reed et al. 2014; Thompson et al. 2015). A thick African crust (40–45 km) is observed near the large igneous volcanic provinces (Wignall 2001; Mège and Korme 2004) beneath the stations located in the continental crust of the Nubian and Somalian plates. These results contrast strongly with the Moho depths we found along the rifts, with a crustal thickness of 21–25 km. This strong dichotomy between the unaltered continental crust and the continental crust thinned by the active continental rifting (Kusky et al. 2010) confirms the theory of the African Plate being split into two tectonic units: the Nubian and Somalian plates (Garfunkel and Beyth 2006).

\(\frac{V_\text{p}}{V_\text{s}}\) ratio reveals the presence of fluids

\(\frac{V_\text{p}}{V_\text{s}}\) ratio has been the topic of numerous studies in oceanic (Clague and Straley 1977; Hyndman 1979; Bloch et al. 2016), continental (Zandt and Ammon 1995; Musacchio et al. 1997; He et al. 2014), and passive margin domains (Kodaira et al. 1996; Kandilarov et al. 2015). Generally, \(\frac{V_\text{p}}{V_\text{s}}\) has been in the range of 1.7–1.9 for the continental crust.

These values agree well with the results obtained for the stations located far away from the rift system (Fig. 11). However, \(\frac{V_\text{p}}{V_\text{s}}\) increases drastically in the regions affected by active rifting (reaching 2.1, as high as the values found for oceanic domains). This strong contrast of between the unaltered continental crust and thinned crust regions is highly consistent with the mechanical extension of the continental thickness imaged by receiver functions (Fig. 10). The anomalously high \(\frac{V_\text{p}}{V_\text{s}}\), possibly caused by the reduction in shear-velocity (\(V_\text{S}\)), can be plausible indirect evidence for fluids, melts, or oceanic crust within the rift system branches.

Complex anisotropy beneath the rift system

Previous shear-wave splitting measurements (Wüstefeld et al. 2009; IRIS 2012) display a constant direction of the fast anisotropy axis throughout the African plate. However, some scattered measurements exist, mainly beneath the Arabian Plate. Our results (Fig. 12) display a similar trend for the stations located on both sides of the rift system beneath the African plates. However, beneath the Arabian Plate, the stations with reliable measurements show null measurements. This null identification suggests that the Arabian block could be isotropic, as it has not experienced substantial deformation since its formation. It is considered a substantially rigid block with limited internal deformation (Alothman et al. 2016; Tesauro et al. 2018) which is consistent with our results. In addition, the high level of inconsistency observed beneath the Arabian plate between the present results and previous shear-wave splitting measurements (Wüstefeld et al. 2009; IRIS 2012) displayed in Fig. 12, suggests that the previous measurements in this region were not classified as null, despite the strong angular variations of the fast direction of anisotropy. The highly scattered anisotropy directions measured in previous shear-wave splitting studies strongly suggest the presence of isotropic media beneath the Arabian block.

Close to the triple junction, most of the measurements are oriented N030E and agree with previous measurements. However, we also found a few measurements perpendicular to the general trend of N120E. The measurements of those outliers are mainly located close to the triple junction. Figure 12 also indicates that the stations located near the rifts tend to display null measurements. Most stations with null measurements are located near the active Holocene volcanoes (Fig. 1). Although null measurements are not present in the global database, the presence of null measurements can be an important indicator of the presence of fluid or melt beneath the rift branches.

Implications for local tectonics

In this study, 431 new shear-wave splitting measurements were obtained in the investigated region. The average fast direction measured is consistent with the results of Gao et al. (2010). However, the average delay time we found is smaller than that obtained by Gao et al. (2010), which is close to the global average for continents (1.0 s calculated from Kennett et al. (1995) and Silver (1996). The fact that the average delay time is slightly smaller than the global average for continental crust can be explained by the presence of thinned continental crust in our target region. The presence of several stations displaying null measurements also agrees with the results of Gao et al. (2010). The relatively homogeneous pattern of the splitting observed in the region suggests that the main contributor to the anisotropy has a deep origin, with local variations associated with crustal heterogeneities. The relatively homogeneous pattern of the splitting is also consistent with the model proposed by Barruol and Ismail (2001), involving a NE flow in the asthenosphere. In addition to the lateral flow of the asthenosphere, mantle upwelling is also responsible for the presence of fluids and melt, which is in agreement with the high amount of null measurements, as well as the very high values observed around the branches of the Afar rift system. Those observations are consistent with those of body wave (Benoit et al. 2006; Koulakov 2007) and surface wave (Sebai et al. 2006; Guidarelli et al. 2011; Legendre 2013) tomographic models. The presence of very slow velocities beneath the Afar region (for the global and regional models) and beneath the three branches of the rift system at all depths suggests the presence of fluids and melt, linked with potential mantle upwelling. Previous tomographic studies focusing on anisotropy on a large scale (Sebai et al. 2006) or local scale (Korostelev et al. 2015; Sicilia et al. 2008) have revealed large regions with surface wave slow anomalies that are associated with magmatism. Along the rift system, very slow velocities anomalies have been found, suggesting the presence of either partial melt or fluids released by cooling magmatic systems. The local seismicity (Abdallah et al. 1979; Illsley-Kemp et al. 2018) is also very shallow, suggesting the presence of shallow reservoirs in the crust along the rift system. In those regions, our results indicate both thinner crust, higher values of \(\frac{V_\text{p}}{V_\text{s}}\), and the presence of null measurements.

Conclusion

In this study, we obtained receiver function and shear-wave splitting measurements for all 227 available broadband seismic stations deployed in the Afar region. The receiver function results provide additional constraints on the crustal thickness and \(\frac{V_\text{p}}{V_\text{s}}\) ratio. The shear-wave splitting measurements provide some constraints on the average anisotropy over a column encompassing the crust and upper mantle beneath the seismic stations. Here, we found that the thick African crust (with a thickness of 40 km) has been strongly elongated and thinned (down to a thickness of 20 km) across the rift system. In addition, the exceptionally high \(\frac{V_\text{p}}{V_\text{s}}\) value within the crust along the rifts and numerous null measurements of shear-wave splitting for stations located near the rifts suggests the presence of melt and the early stage of oceanization. The results of this study provide essential constraints on the structure of the crust and average anisotropy in the Afar region. However, those measurements are limited by the availability of the seismic station deployed in the region. Besides, those measurements provide a 1D snapshot beneath each seismic station, and further modeling such as tomographic inversion could provide a 2D or 3D view of the region. The strong lateral variations of crustal thickness, \(\frac{V_\text{p}}{V_\text{s}}\) and seismic anisotropy suggest a more complex pattern with vertical variations that are not visible from our results. However, this study provides accurate measurements of several geophysical parameters that are directly interpreted in terms of the structure and dynamics of the region.