Hydrous upwelling across the mantle transition zone beneath the Afar Triple Junction

The mechanisms that drive the upwelling of chemical heterogeneity from the lower to upper mantle (e.g., thermal versus compositional buoyancy) are key to our understanding of whole mantle convective processes. We address these issues through a receiver function study on new seismic data from recent deployments located on the Afar Triple Junction, a location associated with deep mantle upwelling. The detailed images of upper mantle and mantle transition zone structure illuminate features that give insights into the nature of upwelling from the deep Earth. A seismic low‐velocity layer directly above the mantle transition zone, interpreted as a stable melt layer, along with a prominent 520 km discontinuity suggest the presence of a hydrous upwelling. A relatively uniform transition zone thickness across the region suggests a weak thermal anomaly (<100 K) may be present and that upwelling must be at least partly driven by compositional buoyancy. The results suggest that the lower mantle is a source of volatile rich, chemically distinct upwellings that influence the structure of the upper mantle, and potentially the chemistry of surface lavas.


Introduction
The upwelling of material from the lower mantle to the base of the lithosphere is hypothesized as being a primary planetary geodynamic process [Morgan, 1971], and it is widely believed that the driving force behind these upwellings is thermal convection initiated due to heating at the core-mantle boundary [Beier et al., 2008]. However, it is unlikely that convection in the Earth is purely isochemical [Stixrude and Lithgow-Bertelloni, 2012], and heterogeneous mantle composition may also lead to variations in density and hence buoyancy. Despite this expectation, the extent to which upwellings and their associated surface volcanism are driven by temperature or compositional variations in the mantle is currently poorly constrained by observations [Ito and van Keken, 2007]. The mineralogical phase changes that produce the 410, 520, and 660 km seismic discontinuities (olivine ! wadsleyite, wadsleyite ! ringwoodite, and ringwoodite ! perovskite 1 magnesiow€ ustite, respectively, and herein referred to as the 410, 520, and 660) are sensitive to both temperature and composition [Helffrich, 2000], equipping the mantle transition zone (MTZ) with mechanisms to test upwelling hypotheses. Due to the opposite signs of the Clapeyron slopes associated with the 410 and the 660, regions of warmer than average mantle, a likely situation if upwelling is driven by thermal buoyancy, should produce a thinner than expected MTZ. This has been observed in certain localities (e.g., 20 km beneath Iceland suggesting an excess temperature of 150 K) [Shen et al., 1998], but other global observations have suggested that there is little correlation of a thinner MTZ thickness with the location of mantle upwellings [Tauzin et al., 2008].
In this study, we analyze new seismic data from recent deployments located on the Afar Triple Junction (Figure 1), the youngest flood basalt province in the world and where the geochemistry of recent volcanism strongly suggests a lower-mantle source [Pik et al., 2006]. The region is characterized by widespread 30 Ma volcanism, initiation of continental rifting 29 Ma and continued yet less voluminous volcanism to the present day ( Figure 1). Although some authors argue for a single plume located beneath Kenya and associated lateral flow of material in the upper mantle toward Ethiopia [Ebinger and Sleep, 1998;Owens et al., 2000;Huerta et al., 2009;Hansen et al., 2012;Hansen and Nyblade, 2013], other seismic tomographic images hint at low velocities beneath Afar extending into the lower mantle indicating the potential presence of an Afar plume [Montelli et al., 2006;Chang and Van der Lee, 2011]. However, to date, the spatial extent, morphology, and thermochemical nature of the lower-mantle African Superplume and its connectivity with the upper-mantle beneath East Africa remains unclear.
Incorporating new dense, high-quality data located across the Afar Triple Junction allows us to significantly improve our spatial coverage and hence our understanding of the geodynamical system beneath East Africa. Tomography and teleseismic receiver functions (RFs) provide evidence for ongoing upwelling across the MTZ that is likely driven, at least in part, by compositional buoyancy.

Data and Method
Earlier studies of MTZ structure beneath Ethiopia have been limited to the Main Ethiopian Rift (MER) and southernmost Afar due to the geographical distribution of stations associated with previous experiments [e.g., Nyblade et al., 2000b;Benoit et al., 2006b;Cornwell et al., 2011]. Incorporation of data from new seismic initiatives located throughout the Afar depression and Western Plateau have led to a significant improvement in station coverage and opened the upper mantle in this region to detailed seismic investigation ( Figure 1) [e.g., Hammond et al., 2013;Rychert et al., 2012].
The S wave seismic tomography model has been presented previously, where interpretations focused on the upper 400 km of the mantle [Hammond et al., 2013]. We present here for the first time the deeper parts of the S wave model (<700 km). Details of the inversion can be found in Hammond et al. [2013], but in summary, we include 13,161 S/SKS-wave travel time picks and invert the data using a standard regularized, linear least squares inversion [VanDecar et al., 1995], jointly inverting for slowness, near-surface corrections, and earthquake corrections. Hammond et al. [2013] show that resolution is excellent in the upper 400 km of the model beneath the study region. We have extended these checkerboard tests using the approach of Hammond et al. [2013], including normally distributed noise to the synthetic travel time data. These tests show that good resolution, sufficient to draw robust inferences, is also present to depths of 700 km beneath both the Main Ethiopian Rift and Afar Depression.
The new receiver function data set contains 5158 high-quality waveforms from 138 broadband seismic stations (Figure 2), between 2 and 4 times the data volume incorporated into previous RF studies of the MTZ beneath Ethiopia [Benoit et al., 2006b;Cornwell et al., 2011]. The RFs were calculated using the iterative time domain approach with a Gaussian pulse width of 0.625 [Ligorria and Ammon, 1999] and migrated to depth using a 3-D common conversion point method where the size of the stacking bins were determined by the Fresnel zone [Wilson and Aster, 2003;Angus et al., 2009;Hammond et al., 2011;Thompson et al., 2011]. A common slowness assumption (i.e., that the P-to-S conversions have the same slowness as the teleseismic P wave) was used in the 3-D migration, potentially leading to slight overestimates in MTZ thickness (maximum of 5 km, although as the mean epicentral distance of earthquakes used in this study is 73.8 it is unlikely to be more than 3 km) [Lawrence and Shearer, 2006]. A minimum RF amplitude of 0.005 was required for the arrivals from the MTZ discontinuities for a given stacking bin in order to use the observation to calculate MTZ thickness. Corrections for both crustal structure and upper-mantle velocity structure from the regional P and S wave tomographic models [Hammond et al., 2011[Hammond et al., , 2013 applied to the ak135 velocity model [Kennett et al., 1995] were also performed. Since the tomographic anomalies are relative to an undetermined background mean, absolute depths of the discontinuities are only indicative. However, there is good resolution of lateral variations in mantle velocity structure and MTZ thickness. We also migrate the data through the 1-D ak135 velocity model and, as did Cornwell et al. [2011], the global absolute P and S wave models of Montelli et al. [2006] for comparison.
In summary, the recent seismic deployments in the Afar Depression in combination with previous seismic experiments in the Main Ethiopian Rift (MER) [Nyblade and Langston, 2002;Bastow et al., 2011] provide unparalleled coverage and resolution at MTZ depths beneath this large igneous province both in terms of seismic tomography and RFs from the MTZ discontinuities (Figures 1 and 2). Furthermore, with the unique position of this purported plume locality being on land, data coverage and quality are superior to other extensively studied locales such as Hawaii or Iceland [Wolfe et al., 1997[Wolfe et al., , 2009].

Tomography
The S wave tomographic models show a persistent low-velocity zone (1-2% slow) throughout the transition zone beneath northern Afar ( Figure 3). A second weaker anomaly is also present beneath the MER ( Figure  3). It is unclear if this continues into the lower mantle as the models lose resolution at these depths. At the top of the transition zone, a broad anomaly is observed (Figure 3), similar to that shown for the uppermost mantle in Hammond et al. [2013]. These features are well resolved at depths extending to at least 600 km ( Figure 4). More recent travel time inversions that extend the Hammond et al. [2013] models by including data from Tanzania and Saudi Arabia better constrain the velocity structure in the MTZ beneath the Horn of Africa (Civiero et al., Multiple mantle upwellings beneath the Northern East-African Rift System from relative P-wave traveltime tomography, submitted to Earth and Planetary Science Letters, 2015). They show extremely similar features within the MTZ, but with lower amplitudes.

Receiver Functions
The arrival times of P-to-S conversions from the MTZ discontinuities (P410s and P660s) relative to the direct P wave at a reference slowness of 6.5 s deg 21 are 49.2 6 0.3 and 73.7 6 0.9 s beneath northern Afar and Geochemistry, Geophysics, Geosystems  Figure 2 for distribution of stations). These are consistently more than 5-6 s slow across the study region compared to that predicted by ak135 (43.9 and 67.9 s, respectively), making them some of the slowest P410s and P660s arrivals in the world along with Hawaii and Iceland [Tauzin et al., 2008]. Despite these large delay times, mean differential arrival times (P660s -P410s 5 24.5 6 0.9 and 23.7 6 0.3 s beneath northern and central Afar, respectively) closely correspond to that predicted by ak135 (24.03 s) and similar to several nonhot spot locations [Tauzin et al., 2008]. This implies that, due to the similar paths taken in the upper mantle by both phases, most of the travel time anomaly is accrued above the 410.
In the vicinity of the tomographically determined low-velocity anomaly seen beneath Afar, the MTZ thickness changes little (mean thickness of 251 km with standard deviation of 13 km determined from the 3-D migration, Figure 7) and lies close to, even slightly above, the global average (ranging from 242 to 247 km) [Flanagan and Shearer, 1998;Gu and Dziewonski, 2002;Lawrence and Shearer, 2006;Tauzin et al., 2008]. This would argue against a significant temperature perturbation at MTZ depths across this region (certainly less than 100 K) [Helffrich, 2000], consistent with previous single station observations [Nyblade et al., 2000a;Benoit et al., 2006a]. There are large perturbations to the apparent topography and MTZ thickness in regions characterized by rapid lateral changes in upper-mantle velocity structure and anisotropy [Hammond et al., 2011[Hammond et al., , 2013[Hammond et al., , 2014Kendall et al., 2005;Gao et al., 2010;Rychert et al., 2012;Bastow et al., 2008], i.e., the border fault regions in Figure 7 where the MER opens into the Afar triangle (9 , 40 E). Rapid changes in structure were also seen by Cornwell et al. [2011]; however, this is in a region where we observe incoherent stacking of the signal (note the reduction in amplitude of the 410 and the apparently diffuse 660 at the southern edge of Figure 8 and in the vicinity of the southern border faults in Figures S1-S3) and choose to focus our interpretation on Afar where the crust, lithosphere, and upper-mantle structure is more coherent.
Despite the lack of evidence for MTZ thinning associated with the tomographic anomaly beneath Afar, the RF waveforms show evidence for a negative conversion above the 410 irrespective of the velocity model used for migration (Figures 5-8 and S1-S4). This implies the presence of a distinct, depth-localized lowvelocity layer (LVL) above the 410, which is most pronounced where the lowest seismic velocity anomalies are observed in the tomography (Figures 7 and 8). A similar LVL has been previously observed in diverse tectonic settings globally including beneath station ATD in Djibouti [Tauzin et al., 2010], but our study maps its detailed spatial extent beneath Afar for the first time. A number of features in the RF data show that this is a robust observation. The arrival from the top of the LVL exhibits positive moveout with slowness ( Figures  5 and 6), indicative of direct P-to-S conversions as opposed to reverberations from shallower discontinuities. It is not an artifact of the RF estimation technique [Ligorria and Ammon, 1999] because there is no symmetric side lobe of comparable amplitude on the underside of the 410 (Figures 5 and 6). Moreover, the ray sampling is densest here, making results from northern and central Afar extremely robust (Figure 2). Although anomalously slow, the crust and upper-mantle structure across much of Afar is relatively uniform [Hammond et al., 2011[Hammond et al., , 2013 with little anisotropy compared to the MER [Hammond et al., 2014], resulting in the migrated RF signal stacking coherently (Figures 8 and S1-S4). The approximate spatial extent of the LVZ, estimated from its robust amplitude detection from the crust and upper-mantle corrected 3-D CCP migration ( Figure S3), is shown in Figure 7.

Discussion
Seismic tomography, especially in the case where surface-waves are not used, can result in significant vertical smearing. Therefore, the presence of a LVL above the 410 could contribute to the low seismic velocities within the MTZ. Figure S5 shows a synthetic test where a 4% slow, 60 km thick low-velocity layer sits directly above the 410 [e.g., Tauzin et al., 2010]. It is evident that such a thin layer is not resolvable by tomography alone and that smearing occurs, but it is not of significant amplitude to explain the observed anomaly (Figure S5). This suggests that there must still be reduced seismic velocities within the MTZ beneath Afar to explain the observations. The simplest way to explain this is through the presence of a warm thermal anomaly. There has been much debate over the strength of any thermal anomaly in the upper mantle beneath the Afar Depression. Cornwell et al. [2011], based on MTZ structure from the MER and southern Afar, suggested a thermal anomaly of 1250 K beneath the region. Alternatively, S wave receiver functions suggest that little anomaly is required [Rychert et al., 2012], but recent geochemical analyses and petrological constraints suggest an excess temperature of 1100 to 150 K [Rooney et al., 2012;Ferguson et al., 2013]. Based on our estimates of TZ thickness from across Afar, 1150 K appears too high as we would expect to see significant variations in MTZ thickness associated with this, but a thermal anomaly of <100 K is consistent with our results. This observation is supported by recent estimates from P wave tomography and joint seismic, geochemical, and numerical modeling that suggest a mild thermal anomaly of 100 K exists in the MTZ and upper mantle beneath the Afar Depression(Armitage et al., Upper mantle temperature and the onset of extension and break-up in Afar, Africa, submitted to Earth and Planetary Science Letters, 2015;Civiero et al., submitted manuscript, 2015). To test this, we build a synthetic model that mimics a broad, 100 K anomaly in the mantle, together with the melt layer. The 100 K anomaly has 1.5% anomaly above the transition zone and 0.5% anomaly within the transition zone [Styles et al., 2011;Civiero et al., submitted manuscript, 2015]. This simple model is consistent with the tomography model (Figures 7, 8, and S6). Amplitudes are still lower than that seen in the data inversions; however, Figure S7 shows a case where we increase the velocity contrast associated with the melt layer. It is clear that, while the vertical extent of the melt layer is not constrained by the tomography alone, the melt layer can play a controlling influence on the amplitudes recovered in the model. However, given that shallow low-velocity anomalies are also likely smeared deeper, care must be taken when interpreting the amplitudes of the anomalies within the MTZ (Civiero et al., submitted manuscript, 2015). Despite this, it is important that the lowest seismic velocities imaged from tomography beneath Afar coincide with the presence of a stable melt layer imaged using the RFs.
A low-velocity layer above the 410 is commonly ascribed to a stable, depth-localized region of partial melt [Revenaugh and Sipkin, 1994]. It is difficult to reconcile the depth-localized nature of this feature using elevated temperature alone [Tauzin et al., 2010], consistent with our MTZ thickness estimates. Other mechanisms are often invoked, most notably the presence of water. Recent observations have confirmed that the MTZ can be locally hydrated, containing up to 1.5 wt % water [Pearson et al., 2014]. Within the upper part of the MTZ, water storage capacity of wadsleyite is expected to be 5 times what it is above the 410 [Inoue et al., 2010], meaning upwelling material transforming into olivine has the potential to exceed the uppermantle water storage capacity and induce partial melting [Hirschmann, 2006]. This so-called transition zone water filter model [Bercovici and Karato, 2003] predicts a global LVL above the 410 due to the passive upwelling of ambient mantle, a conclusion that some seismic results support [Tauzin et al., 2010]. The original model also predicts that this process would be suppressed in plume-affected regions due to increased water solubility with increasing temperature and short residence times in the MTZ [Bercovici and Karato, 2003]. These predictions have been questioned recently, instead suggesting that ambient mantle (the expected source of mid-ocean ridge basalt) does not have sufficient water content (<200 ppm) to exceed the upper-mantle storage capacity directly above the 410, even with the reduced values predicted by the most recent studies (440-850 ppm) [Tenner et al., 2012;F erot and Bolfan-Casanova, 2012]. Thus, a global LVL may not be present. In contrast, plume source mantle has been shown to have sufficiently high water content (up to 1000 ppm) to significantly exceed the upper-mantle storage capacity [Nichols et al., 2002]. This would permit the formation of a localized melt layer [Tenner et al., 2012], as is seen here, without recourse to a global model that is controversial as to the amount of water assumed to be in normal mantle. An associated broadening of the 410 may also be expected [Frost and Dolejs, 2007], but the presence of a pronounced LVL above the discontinuity complicates observation of this broadening through its amplitude. The presence of melt may also act to truncate the region of coexistence between olivine and wadsleyite [Helffrich, 2000], creating a sharper than expected seismic discontinuity.
Our results provide some of the most compelling and well-constrained evidence to date for the formation of this melt layer in the presence of currently active upwelling in a volcanic environment. Such low-velocity layers have been found beneath some other Mesozoic flood basalts [Vinnik and Farra, 2007], yet this spatial correlation requires the controversial assumption of long-term coupling of the continental lithosphere and Figure 7. (a-c) Slices through the S wave tomography model at 410, 520, and 660, respectively (see Figure 3). (d-f) Apparent topography of the 410, MTZ thickness and apparent topography of the 660. Regions of anomalous MTZ thickness can be seen coincident with the MER and other border faults where rapid changes in upper-mantle velocity structure and anisotropy are present. The line of cross section is 42 E. The dashed region is the limit of CCP bins with 150 hits per bin, and the region with shaded droplets is the approximate extent of the LVL determined by its detection in the 3-D CCP migration.
Geochemistry, Geophysics, Geosystems underlying mantle to depths 400 km. A LVL has also been recently suggested beneath Hawaii in conjunction with a significant depression of the 410 (from 400 to 450 km, interpreted as a temperature anomaly of >300K [Huckfeldt et al., 2013]). The results from Afar are novel for the fact that no significant variation in discontinuity depth or MTZ thickness is correlated with it. Consistent with the latest tomographic studies (Civiero et al., submitted manuscript, 2015), this suggests that only a weak thermal anomaly (<100 K) can be invoked to drive upwelling beneath northern and central Afar. Hydration can also contribute to positive buoyancy, with the addition of 2000 ppm water having the same effect on the density of olivine as a temperature increase of 100 K [Smyth et al., 2006]. With upwelling of hydrous mantle being a fundamental requirement to initiate dehydration melting on top of the 410, it must be driven in part by compositional buoyancy beneath the Afar Triple Junction. Volcanics from the study region have suggested a volatile-rich and hydrous source for metasomatism of the lithospheric mantle [Baker et al., 1998], and the origins of the upwelling may lie with recycled, metasomatized subducted oceanic lithospheric mantle, explaining both the hydrous and compositionally buoyant nature of the inferred upwelling [Niu and O'Hara, 2003;Niu et al., 2012].
Compositional buoyancy has been suggested beneath East Africa previously, but in terms of negative buoyancy at depths of 1000 km [Simmons et al., 2007]. This is below where we have resolution, so cannot confirm or refute the presence of this feature. However, complex geodynamics are expected in the lower mantle in the presence of recycled, compositionally layered oceanic lithosphere and/or a dense basal layer at the bottom of the mantle [Tackley, 2011;Olson and Kincaid, 1991], hence our interpretations are not necessarily contradictory.
In association with the LVL, there is also a strong positive polarity arrival within the MTZ beneath northern Afar that we interpret as the 520 seismic discontinuity (Figures 5, 6, and 8), an arrival not commonly observed in higher-frequency RF studies. As with the conversion from the top of the LVL, the moveout characteristics of the 520 indicate a direct P-to-S conversion (Figures 5 and 6). The width of the two-phase loop associated with this discontinuity is expected to shrink under hydrous conditions, producing a stronger converted seismic signal [Inoue et al., 2010]. The 520 appears more visible where the LVL is at its thickest and the low-velocity anomaly in the tomography is at its greatest (Figure 8), providing further evidence for localized hydration of the mantle. There appears to be a doubling of the 520 in the migrated RFs beneath northern Afar (Figure 8), but this is likely due to the stacking of a single discontinuity with laterally variable depths. With water expected to alter the depth of the 520, it is plausible that the variability in depth is also related to water content [Inoue et al., 2010]. A recent study of the 520 beneath the African continent, incorporating data from station ATD, also attests to the variability of this discontinuity beneath our study region [Juli a and Nyblade, 2013].
While we try to avoid interpreting the more subtle features of our models, there appears to be a minor thickening of the MTZ due to a depression of the 660 moving from central into northern Afar (the location of the tomographically determined low-velocity anomaly, Figure 8). Assuming the dominance of the olivine system under anhydrous conditions, this is opposite to what would be expected for a hot mantle [Helffrich, 2000]. While the depth of the 410 is expected to be relatively unaffected by the presence of water (<5 km variation over plausible mantle water contents) [Frost and Dolejs, 2007], recent work has suggested that the depth of the 660 could be more sensitive to it (potentially up to 40 km deeper than under anhydrous conditions) [Ghosh et al., 2013]. Therefore, this feature may also be consistent with a hydrous MTZ. Another explanation for this feature is that high temperatures can lead to the 660 region being controlled by the garnet to perovskite phase transition [Weidner and Wang, 1998], with an opposite Clapeyron slope to that of the ringwoodite to perovskite transition. This would imply an increase in temperature could in fact cause a deepening of the 660 and a relatively normal, or even increased, MTZ thickness [Deuss, 2007]. For example, Cornwell et al. [2011] interpreted variations in bulk composition (i.e., variations in the amount of garnet) within a ubiquitously hot mantle (1250 K) to explain the strong variations in MTZ thickness observed beneath the MER. As stated previously, we choose to confine our interpretations to the Afar region where the stacking coherence is high. Due to the lack of any correlated topography on the 410 associated with the deepening of the 660 and the latest tomographic studies consistent with only a mild temperature elevation (100 K) (Civiero et al., submitted manuscript, 2015), we discard phase changes associated with the garnet system at high temperature as a viable interpretation for this feature.
We showed that compositional buoyancy, whose source lies in the lower mantle, must play a role in the development of the seismic structure below Ethiopia. Various observations of isotopic and major element heterogeneity in East African lavas [Pik et al., 2006;Meshesha and Shinjo, 2008] suggest compositional variability of their sources, some of which are inferred to lie in the lower mantle. Our results support both the inference of a deep source (certainly beneath the 410), and also the extreme heterogeneity. Duncan and Richards [1991] showed that compositional buoyancy diminishes mixing by entrainment, leading to juxtaposed compositions arising from different sources. We speculate that the surface chemistry of Afar rocks may map lower-mantle source chemistry.

Conclusions
The new high-resolution images of the MTZ constrain upwelling beneath the Afar Triple Junction. The RF data and seismic tomography show that only a weak thermal gradient is required in the MTZ across East Africa, but small, volatile-rich, and compositionally buoyant plumelets which rise through the MTZ from a heterogeneous deep mantle source likely exist. It is these plumelets that give rise to the complex geochemical signatures observed across Ethiopia and East Africa, and when sufficiently hydrous lead to the formation of a stable melt layer above the 410. The results show that mantle plumes maybe more complex than previously thought, and that in some regions are likely driven by differences in composition rather than temperature alone. This is an important consideration not just for East Africa, but in understanding how mantle plumes interact with the MTZ and the transport of chemical heterogeneity from the deep Earth.