The Main Ethiopian Rift (MER) is a classic narrow rift that developed in hot, weak lithosphere, not in the initially cold, thick, and strong lithosphere that would be predicted by common models of rift mode formation. Our new 1-D seismic velocity profiles from Rayleigh wave/receiver function joint inversion across the MER and the Ethiopian Plateau indicate that hot lower crust and upper mantle are present throughout the broad region affected by Oligocene flood basalt volcanism, including both the present rift and the adjacent Ethiopian Plateau hundreds of kilometers from the rift valley. The region of hot lithosphere closely corresponds to the region of flood basalt volcanism, and we interpret that the volcanism and thermal perturbation were jointly caused by impingement of the Afar plume head. Across the affected region, Vs is 3.6–3.8 km/s in the lowermost crust and ≤4.3 km/s in the uppermost mantle, both ∼0.3 km/s lower than in the eastern and western branches of the East African Rift System to the south. We interpret the low Vs in the lower crust and upper mantle as indicative of hot lithosphere with partial melt. Our results lead to a hybrid rift mode, in which the brittle upper crust has developed as a narrow rift along the Neoproterozoic suture between East and West Gondwana, while at depth lithospheric deformation is distributed over the broad region (∼400 km wide) thermally perturbed by the broad thermal upwelling associated with the Afar plume head. Development of both the East African Rift System to the south (in cold, strong lithosphere) and the MER to the north (in hot, weak lithosphere) as narrow rifts, despite their vastly different initial thermal states and depth-integrated lithospheric strength, indicates that common models of rift mode formation that focus only on temperature, thickness, and vertical strength profiles do not apply to these classic continental rifts. Instead, inherited structure and associated lithospheric weaknesses are the primary control on the mode of extension.
 There is an ongoing debate about the strength of crust and lithosphere in regions of extension. Knowledge of the rheological stratification of the lithosphere in these areas is a critical parameter in developing useful and accurate models of continental extension. Both simple analytical models, e.g., the commonly cited model of Buck , and numerous thermomechanical geodynamic models of large-scale lithospheric extension [e.g., Corti, 2004; van Wijk, 2005; Dyksterhuis et al., 2007; Sokoutis et al., 2007; Corti, 2008] depend upon the assignment of accurate rheologic parameters in the crust and upper mantle. The thermal structure of the lithosphere, rheological stratification, and crustal thickness are required for this modeling, yet detailed and three-dimensional information of this type is not commonly available. Adding further complication is the understanding that the rheological stratification may change at different stages in rifting, as has been suggested for the Iberia–Newfoundland and Alpine Tethys margins [Manatschal et al., 2007] and for the Main Ethiopian Rift [Keranen and Klemperer, 2008]. Broadly distributed data from the 2001–2003 EAGLE (Ethiopia-Afar Geoscientific Lithospheric Experiment [Maguire et al., 2003]) and the 2000–2002 EBSE (Ethiopian Broadband Seismic Experiment [Nyblade and Langston, 2002]) seismic experiments in Ethiopia provide an opportunity in the MER, a classic example of a “narrow continental rift,” to obtain a three-dimensional snapshot of crustal and upper mantle rheology immediately prior to breakup. We use shear wave seismic velocity profiles from our new joint inversion of surface wave dispersion curves and receiver functions using the methodology of Julia et al.  to infer lithospheric rheology over a region including the MER and the adjacent Ethiopian and Somali Plateaus. Dugda et al.  also used this joint inversion method in Ethiopia and Djibouti to study shear wave velocity structure using data from the EBSE array. Their study focused on the upper mantle and revealed thin lithosphere and low shear wave velocity in the uppermost mantle beneath the MER and the Ethiopian Plateau. Our study builds on Dugda et al.  by providing detailed crustal velocity profiles beneath both the EBSE and EAGLE Phase I broadband stations, and by studying spatial variations in the crustal and upper mantle velocity structure with respect to the domains previously recognized by Keranen and Klemperer . We use our joint inversion results to infer the spatially variant rheology of the MER and adjacent plateaus and on the basis of this rheologic structure we propose a hybrid mode of continental extension. We additionally provide modelers with constraints on rheology of the crust and upper mantle in a classic narrow rift.
2. Tectonic Background
 The MER is a seismically and volcanically active portion of the East African Rift System (EARS) trending NE across the Ethiopian plateau (Figure 1). The MER developed within the Mozambique Belt, a broad Neoproterozoic mobile belt that extends from Ethiopia south through Kenya, Tanzania and Mozambique. The Mozambique Belt is believed to represent a Himalayan-type continental collision zone [Burke and Sengor, 1986; Shackleton, 1986; Stern, 1994], and ophiolites mark numerous sutures within the basement complex of Ethiopia [Vail, 1985; Berhe, 1990; Stern et al., 1990] although the trend of the sutures is contentious [Church, 1991]. The suture is also evident in neodymium (Nd) isotopic data [Stern, 2002], which indicates a major crustal boundary in Ethiopia separating juvenile crust in the north and west (mean age 0.87 Ga) from reworked Paleoproterozoic and Archean crust (mean age 2.0 Ga) to the east [Stern, 2002].
 Because of the continuity of the Mozambique Belt (Figure 1), the MER is commonly thought to have developed in lithosphere with a rheological structure much like that adjacent to the EARS further south where, away from the Cenozoic rift, the lithosphere is cold, thick, and strong on the basis of high seismic velocities [Mechie et al., 1997], low heat flow [Nyblade et al., 1990], and deep crustal earthquakes [Foster et al., 1997]. The MER is unique, however, in that the heterogeneous Precambrian craton has been significantly modified by Tertiary plume magmatism. Flood basalts erupted primarily between 31 and 29 Ma cover much of the Proterozoic basement beneath the Ethiopian Plateau (Figure 1), extending over an area ∼1000 km in diameter and reaching up to 2 km in thickness [Baker et al., 1996; Hofmann et al., 1997; Ukstins et al., 2002]. In contrast to these Ethiopian flood basalts, which were erupted over a short time period ∼20 ma prior to the onset of extension and cover the broad region of topographic doming, magmatism in the Kenya Rift to the south has occurred in stages roughly coeval with extension and is located in or near the rift valley rather than over the Kenya dome [Macdonald, 2002]. A mantle slow velocity anomaly associated with the Afar plume underlies the MER and the NW plateau [Bastow et al., 2005; Benoit et al., 2006a; Bastow et al., 2008] down to at least 650 km depth beneath Afar and rising to at least 75 km depth beneath the Main Ethiopian Rift and NW Ethiopian Plateau.
 Since the eruption of the flood basalts in Ethiopia at 31–29 Ma, volcanism has continued both in the rift and on the adjacent plateaus. Two periods of volcanism at 22 Ma and from 10 to 7 Ma created a series of shield volcanoes on the northwest Ethiopian Plateau [Kieffer et al., 2004]. A period of volcanism from 10 to 12 Ma occurred within the MER, and bimodal Quaternary volcanism occurred both within the MER and on the northwest Ethiopian Plateau near Lake Tana, hundreds of kilometers from the rift valley [Abate et al., 1998; Conticelli et al., 1999].
 Lithospheric structures that developed within Ethiopia prior to the onset of Cenozoic rifting have controlled the evolution of the Main Ethiopian Rift, which is interpreted to have propagated in an irregular temporal and spatial manner [Bonini et al., 2005; Keranen and Klemperer, 2008]. The irregular propagation has created discontinuities along the strike of the rift, separating the northern (NMER), central (CMER) and southern (SMER) sectors (Figure 2). On the western shoulder two distinct domains are also interpreted, the NW and SW shoulders, on the basis of a sharp change in crustal thickness between the two domains from ∼40 km to ∼35 km respectively (Figure 2).
 We use the method of Julia et al.  for our joint inversion of two independent data sets: receiver function waveforms and Rayleigh wave group velocities. Jointly inverting these two data sets is now a standard technique in seismic investigations of the lithosphere. Although receiver function inversions provide detailed profiles of lithospheric discontinuity structure, they offer only weak constraints on the absolute velocity in each layer. Inversions of surface wave dispersion curves provide a smooth profile of the absolute velocity structure. Combining the two data sets in a joint inversion therefore provides a well-constrained profile of the lithospheric velocity and discontinuity structure. The inversion method of Julia et al.  has been used in numerous investigations [e.g., Julia et al., 2003; Horspool et al., 2006; Dugda et al., 2007; Gok et al., 2008], and other methods developed to jointly invert these two data types are also commonly employed [e.g., Lawrence and Wiens, 2004; Chang and Baag, 2005; Tkalcic et al., 2006]. Details on consistency and complementarity of the two data sets, as well as weighting and normalization of input data, are available in the original methodological paper of Julia et al. . While readers are urged to consult that paper for the complete methodology, we summarize the method and our slight modifications below.
 The method [Julia et al., 2000] that we employ makes use of a linearized inversion procedure that minimizes a weighted combination of least squares norms for each data set. These norms are a model roughness norm and a vector difference norm between inverted and preset model parameters. The velocity models obtained are, consequently, a compromise between fitting each of the observations, model simplicity and a priori constraints. To make the contribution of each data set to the joint least squares misfit comparable, a normalization of the data set is necessary, and this is done using the number of data points and variance for each of the data sets. An influence factor is then used to control the trade-off between fitting the receiver functions and the group velocity curves [Julia et al., 2000].
 Seismic data used in this study are from two networks. Data were collected from 2001 to 2003 by the EAGLE Phase I network [Stuart et al., 2006] and from 2000 to 2002 by the Ethiopia broadband seismic experiment (EBSE) [Nyblade and Langston, 2002]. The 30 stations of the EAGLE Phase I network were located primarily within the MER (Figure 2), with several stations on the nearby shoulders. Twenty-two of the 27 seismic stations in the EBSE were located either on the eastern or western side of the Ethiopian Plateau (Figure 2) with the rest of the stations in the MER or Afar Depression. In our joint inversion, we use our own computed receiver functions from 39 of the available stations and previously published fundamental mode Rayleigh wave group velocities [Pasyanos, 2005]. The remaining stations were either outside of our area of interest or had unreliable data quality (most as a result of equipment malfunction). Love wave group velocities were not included because there are few high-quality measurements available for Ethiopia [Benoit, 2005].
3.2. Rayleigh Wave Group Velocities
 Rayleigh wave group velocities of 15 to 75 s period from the model of Pasyanos  were used for the joint inversion. The model of Pasyanos includes group velocity measurements from Benoit et al. [2006b], who added new measurements made with data from the EBSE to the dispersion measurements of Pasyanos et al.  for a surface wave tomography study of eastern Africa. Typical uncertainties in group velocities range between 0.01 and 0.02 km/s but increase to 0.02 and 0.03 km/s for the shortest periods in the Pasyanos model. It is far better to use inverted group velocities, as we do in our joint inversion, than raw dispersion measurements because the inverted values represent laterally localized structure. Pasyanos  suggests the resolution of his model may be as low as 1° for periods less than 30 s, and 2° for longer periods. Velocity functions are extracted from his model at each station used in our inversion.
3.3. Receiver Functions
 Receiver functions were computed at 39 stations using 2462 seismograms from 148 teleseismic events between distances of 30° and 95° with magnitudes greater than 5.5 Mw. Most of the events are from the east (the Indonesian and Western Pacific subduction zones) or the northeast (Hindu Kush-Pamir region) (Figure 3). The maximum number of events stacked together to form a single receiver function was 46.
 We used the time domain iterative deconvolution method of Ligorria and Ammon  to compute the receiver functions, and we evaluated the quality of the receiver functions using a least squares misfit criterion. This misfit criterion provides a measure of the closeness of a receiver function to an ideal case, and is calculated by using the difference between the radial component seismogram and the convolution of the vertical component seismogram with the predicted radial receiver function. Receiver functions with a fit of 90% and above were used in the inversion. The receiver functions were filtered with a Gaussian low-pass filter using three distinct filter width parameters of 0.5, 1.0 and 2.5 (pulse widths of approximately 2.36, 1.67, and 1.0 s) to obtain low-, middle-, and high-frequency receiver functions, respectively. Radial and tangential receiver functions were visually examined for evidence of lateral heterogeneity and for dipping structure. Events with large-amplitude tangential receiver functions were not used.
 In the joint inversion, we used four groups of receiver functions each corresponding to a range of ray parameters from 0.04 to 0.049 s/km, from 0.05 to 0.059 s/km, from 0.060 to 0.069 s/km, and from 0.070 to 0.079 s/km. For each grouping of receiver functions, we computed and stacked the three sets of receiver functions with overlapping frequency bands described previously. The range of frequencies was chosen to be able to distinguish between gradational and sharp boundaries in lithospheric structure [Julia, 2007].
 For the receiver functions the lateral distance from the station sampled by the Ps phase at the Moho is 4 km and 7 km for crustal thicknesses of 25 km and 45 km for a ray parameter of 0.04 s/km, and 7 km and 13 km for a ray parameter of 0.079 s/km, so that the spatial resolution is significantly better than for the surface wave model.
3.4. Initial Models
 The initial models used for the inversion consisted of constant velocity layers that increase in thickness with depth. Layer thicknesses were 0.5 km for the top 1.5 km, 1 km to 5.5 km, 2.5 km between 5.5 and 53 km depth, 5 km between 53 and 260 km depth, 10 km between 260 and 330 km depth, and 20 km below 330 km depth. Dugda et al.  demonstrated that velocities at lithospheric depths trade off with velocities below ∼190 km depth. Following their approach we forward modeled a range of fixed shear wave velocity profiles below 190 km depth to find the model that best fit the long-period group velocities from the Harvard global model [Larson and Ekström, 2001]. In agreement with Dugda et al. , we found that a model for Ethiopia with shear wave velocities of 10% below PREM [Dziewonksi and Anderson, 1981] provided the best fit and we fixed the velocities below 190 km to these values in the inversion.
 For the top 190 km, we used 12 different starting velocity models (Figure 4) in our inversions to investigate the sensitivity of our model results to starting crustal velocity, crustal thickness, and Poisson's ratio. These included distinct combinations of models with two possible crustal velocity profiles (Figure 4), two different values of crustal thickness (25 and 45 km) that bracket the published range of crustal thicknesses across Ethiopia, and three Poisson's ratios of 0.27, 0.29, and 0.31 (corresponding to Vp/Vs ratios of 1.78, 1.84, and 1.91). Figure 5 shows results for station DMRK demonstrating our method and representative results. In an earlier phase of testing, we tested starting models both with and without sedimentary basins, but found that for many stations inversions using starting models without sedimentary basins were unstable whereas those with basins were stable at all stations. We also tested a broader range of Poisson's ratios before settling on the three values from 0.27 to 0.31, which were chosen to limit the number of initial models needed while still providing values representative of the range reported in Ethiopia by Dugda et al.  and Stuart et al. .
4.1. Sensitivity Testing
 To test the sensitivity of our models, we used a range of input parameters including the initial velocity and Poisson's ratio models as described above and a range of influence parameters (weighting the relative influence of receiver function data to dispersion data) at each station.
 For two stations we also used a bootstrapping method to determine the uncertainty in our inversions due to scatter in the data, one on the Ethiopian Plateau with a low-velocity lower crust and another on the northeastern shoulder with a higher-velocity lower crust. In the bootstrapping, for a given station we conducted n sets of inversions, where n is the total number of receiver functions for that station. In each inversion set for the selected station, we randomly selected n receiver functions from the total of n. By randomly selecting, we allow the same trace to be selected multiple times and other traces to not be used. We then divided these randomly selected n traces into the proper ray parameter and frequency groups and stacked each group. We inverted this new set of stacks using each of 12 input velocity and Poisson's ratio models and a range of influence parameters (0.3, 0.5, and 0.7), and repeated for each of n sets of new stacks. At the station DMRK, this resulted in nearly 3000 inversions for the single station. Because of the intense computing requirements, we only inverted two stations using this method. The tight clustering of the resulting profiles (Figure 6) provides confidence in our models at the other stations and in our interpretations. We estimate that our total uncertainty in velocity is ∼0.1 km/s in the lower crust and 0.2 km/s in the mantle, and the uncertainty in the depth to discontinuties is about 2.5 km in the crust and uppermost mantle (the thickness of our layers in this depth range).
4.2. Comparison of Inversion Results With Active Source Crustal Profiles
 The proximity of the EAGLE broadband stations to active source wide-angle profiles from the EAGLE experiment allowed us to independently confirm the reliability of our joint inversion results within the crust. Figure 7 shows active source models along and across the MER [Maguire et al., 2006], with 1-D profiles from our joint inversion located less than 15 km from the 2-D line (in most cases located directly along the 2-D line) overlain. Our results are very consistent with the active source modeling in discontinuity structure (depths to interfaces). A strong correlation is evident for the transition from low-velocity, near-surface layers to higher-velocity (∼6 km/s) layers, for the midcrustal discontinuity at 10–20 km depth, and at the Moho. The intracrustal reflector modeled in the active source profile at 20–30 km depth along the across-axis line (Figure 7a) is not visible in our profiles as a velocity increase and at several stations is more consistent with the top of a lower crustal low-velocity zone. Very little change in velocity is modeled across this interface in the active source profiles. The discrepancy between our joint inversion results and the active source discontinuity at this level may be a result of the different sensitivities of the methods: wide-angle refraction studies have little sensitivity to low-velocity zones, whereas our joint inversion has equal sensitivity to velocity changes of either polarity.
 A significant difference between the results of the active source profiling and the passive source seismic studies is the presence or absence, respectively, of a high-velocity lower crustal layer. Although on the active source profile in Figure 7 the lower crust beneath the NW shoulder includes a 10-km-thick high-velocity lower crustal layer, for a total crustal thickness of ∼50 km [Mackenzie et al., 2005], results from Dugda et al. [2005, 2007], Stuart et al. , and our joint inversion results show that this high-velocity body is a localized feature not characteristic of the NW shoulder, which more typically has a thickness ranging from 38 to 42 km. In the previous studies with Moho depths incompatible with those from the active source study [e.g., Dugda et al., 2005; Stuart et al., 2006] the source of the discrepancy between the results of the different methods was unclear. Our 1-D profiles, with both widespread coverage beneath the NW shoulder and information on the seismic velocity structure of the crust and upper mantle, can spatially map the lower crustal feature evident in the active source data. Velocity profiles from stations near the active source line beneath the NW shoulder show a distinctly different Moho gradient than the remaining stations (Figure 8). The anomalous stations in Figure 8b show a gradually increasing seismic velocity at the base of the crust over ∼10 km, approximately the thickness of the underplate modeled by Mackenzie et al. . Away from these three stations the Moho is a sharper contrast (Figure 8c) and our crustal thickness estimates are in agreement with crustal thickness estimates from Stuart et al.  and Dugda et al. . The anomalous thickness values of Mackenzie et al.  appear to reflect a local crustal thickening and possible underplating. The cause of this thickening is unknown but we note that the active source line was placed to cross the western boundary of the rift in this location to take advantage of the gentle relief and monocline associated with the intersection of the Yerer-Tullu Wellel Volcanotectonic Lineament and the NMER, both volcanically active, and perhaps the enhanced crustal underplating is related to this intersection of two rift trends.
 Results for our joint inversion are shown in Figures 9–12. Thirty-nine stations from the EAGLE and EBSE networks were used for our final inversions. In Figure 9, stations are grouped by domains on the basis of the discussion by Keranen and Klemperer  and as summarized in section 2. Our results for crustal thickness, lower crustal shear wave velocity, and upper mantle shear wave velocity are summarized in Table 1.
Table 1. Lower Crustal Shear Wave Velocity, Upper Mantle Shear Wave Velocity, and Crustal Thickness for Each Station in Our Joint Inversiona
 Beneath the NW shoulder, the Moho depth range is from 38 to 41 km, whereas beneath the SW shoulder Moho depth ranges from 33 to 36 km (Figure 10). Beneath the NMER, crustal thickness ranges from 28 to 36 km thick. To the south in the CMER, crustal thickness for AMME, the one station available to us within the CMER proper, is 38 km. Crustal thickness beneath the eastern shoulder is 38–41 km (Figure 10). Our crustal thickness estimates are in good agreement with the results reported by Dugda et al.  and Stuart et al.  using the Zhu and Kanamori  receiver function stacking technique.
 The uppermost mantle shear wave velocity in our results is similar to that found by Dugda et al.  and ranges between 4.0 and 4.3 km/s (Figure 11). These values are low with respect to 4.5 km/s in the uppermost mantle in the PREM model and in Tanzania and Kenya away from the Cenozoic rift [Julia et al., 2005]. Upper mantle velocity differs between domains within Ethiopia, with values of 4.0–4.2 km/s beneath the NW shoulder, 4.2–4.3 km/s beneath the SW shoulder, 4.0–4.1 km/s beneath the NMER, and 4.1 km/s beneath the eastern shoulder (Figure 11). The region with the lowest upper mantle seismic velocity (∼4.0 km/s) in the 5 km below the Moho is the NMER and the YTVL (Figure 11), which lies above the lowest upper mantle seismic velocity at 75 km depth [Bastow et al., 2005]. However, in our inversion a similar low-velocity anomaly is not seen beneath the CMER despite its observation at 75 km depth using travel time tomography [Bastow et al., 2005], either a result of a dipping low-velocity zone or reflecting the lower amplitude of the CMER low-velocity zone in the upper mantle tomography. The 20–40 km thick mantle lid visible on profiles from the Western Ethiopian Plateau in the work by Dugda et al. [2007, Figure 6, line A-A'] is located beneath the SW Ethiopian Plateau. The mantle lid is present in our results from the SW plateau (Figure 9) but is not evident beneath the NW plateau, where the upper mantle Vs is lower.
 Perhaps the most notable result of our inversions is the identification of a widespread, low-velocity lower crust throughout the region of flood basalt volcanism (Figure 12). In contrast to the marked contrast between rift and plateau structure of the EARS in Kenya [Dugda, 2007], the velocity structure of the crust and uppermost mantle in the MER is very similar to the velocity structure beneath the NW Ethiopian Plateau (Figure 9). The results from our inversion show that the lower crustal velocities beneath the MER (∼3.7 km/s), the NW shoulder (∼3.8 km/s), and the region of the eastern shoulder covered by flood basalt volcanism (∼3.8 km/s) are all reduced by 0.2–0.4 km/s (Figure 12) from velocities in the lower crust in the along-strike Mozambique Belt in Tanzania [Julia et al., 2005]. The largest difference between the crust and upper mantle profiles from the MER and NW Plateau is the ∼5 km difference in crustal thickness. This thinning is caused by upper crustal and midcrustal thinning within the rift valley itself, modeled in the across-axis wide-angle seismic refraction line from the EAGLE experiment [Mackenzie et al., 2005; Maguire et al., 2006], rather than by differences in lower crustal thickness.
6. Interpretation of Low Shear Wave Velocity in the Upper Mantle and Lower Crust
6.1. Upper Mantle Shear Wave Velocity of 4.0–4.2 km/s
 The primary potential causes of low upper mantle shear wave velocity are composition, increased temperature, the presence of fluids (such as partial melt), grain size reduction [Faul and Jackson, 2005], or some combination of these factors. These factors can be analyzed individually to assess the possible contribution of each to the low observed upper mantle shear wave velocity.
 Composition and grain size can be eliminated as predominant effects because of the large magnitude (5–10%) of the observed decrease. The composition of the mantle is less variable than the composition of the crust, and at most compositional variations cause an ∼1% change in shear wave velocity [Cammarano et al., 2003]. The effect of changing the grain size from 1 cm to 1 mm, the range commonly estimated for the upper mantle [Hirth and Kohlstedt, 2003, and references therein], results in an ∼100 m/s (∼2%) change in velocity at 1200°C (Figure 13), again not enough to account for the 300–500 m/s decrease observed in Ethiopia.
 We next examine the remaining factors: increased temperature and partial melt. These two factors are likely interdependent and combine to reduce the shear wave velocity. With the two effects superimposed, it is difficult to constrain the exact contribution from each. We instead seek to develop reasonable maxima and minima for each factor. Two recent publications [Faul and Jackson, 2005; Priestley and McKenzie, 2006] discuss the relationship between upper mantle temperature and shear wave velocity. The behavior of these two models differs (Figure 13), however both predict that an upper mantle temperature over 1300°C is required to produce a shear wave velocity of 4.1 km/s or lower, and that above 1300°C dVs/dT increases rapidly. On the basis of Faul and Jackson  and Priestley and McKenzie , if there is no contribution from partial melt, the estimated temperature of the upper mantle for seismic velocity ranging from 4.0 to 4.1 km/s (i.e., beneath the MER and NW plateau) is predicted to be 1300–1350°C. To consider the effect of only partial melt with no increase in temperature, we use the estimate of Faul et al.  of a 3.3% reduction in seismic velocity per percent melt. In this melt-only case, 3% melt is required in the upper mantle to reduce the seismic velocity by the observed 10 percent.
 Both high temperature and partial melt are capable of reducing the seismic velocity of the upper mantle. It is implausible however to assume that one of these effects would be present without the other. This is evident in the experimental results of Hirschmann et al. . The authors measured percent melt produced as a function of increasing temperature for upper mantle peridotites, including a peridotite composition similar to that present in Ethiopian xenoliths [Conticelli et al., 1999] (Figure 14). As Figure 14 demonstrates, as upper mantle temperature increases above ∼1270°C there is a rapid increase in the percent partial melt present from near zero to ∼5%. If a temperature of the uppermost mantle of 1300°C is required to reduce shear wave velocity to 4.1 km/s by a purely thermal effect as discussed above, we see that at this temperature ∼5% partial melt would be present (further reducing seismic velocity). The combined effect of high temperature and partial melt would then lower the shear wave velocity below that observed in Ethiopia.
 Thus the high temperature that would be required to account for the observed low seismic velocity cannot be present without a contribution from partial melt. In contrast, in principle the observed seismic velocity reduction could be the effect of transient partial melt without anomalous upper mantle temperature. To constrain the lower bound on the upper mantle temperature, we consider independent evidence regarding the thermal structure. Although we lack heat flow data in our study area, mantle xenoliths found in Oligocene and Quaternary basalts near Lake Tana on the Ethiopian Plateau [Conticelli et al., 1999] from a depth range of 40–50 km can be used as estimates of the temperature of the uppermost mantle at the time of eruption. These studies indicate minimum upper mantle temperatures between 920 and 1045°C for the Oligocene and Quaternary samples. This temperature is well above a steady state continental geotherm [Chapman, 1986]. We conclude that the uppermost mantle temperature is elevated by over several hundred degrees above stable continental regions, and that the observed seismic velocity reduction is the result of both high temperature and partial melt. We conclude that the most likely lower bound for the temperature is ∼1250°C, where melt begins to form [Hirschmann et al., 1998], and the most likely upper bound is ∼1300°C (at which the amount of melt produced would create a velocity effect greater than that permitted by observations). This temperature range is compatible with a melt percent of up to 3%. The cause of the high temperature in the uppermost mantle is interpreted to be the thermal perturbation associated with the Afar plume. The strong evidence from previous geophysical and geochemical studies for a thermal perturbation in the mantle down to depths of 600 km is summarized in section 1. Dugda et al.  model instantaneous thinning of the lithosphere due to plume impingement at the time of the flood basalt volcanism (∼30 Ma), with hot plume material remaining to the present. Their models show that the temperature increase (still present today) resulting from the transfer of heat from the plume material to the lithosphere is enough to account for the observed reduction in shear wave velocity [Dugda et al., 2007].
6.2. Lower Crustal Shear Wave Velocity of 3.7–3.8 km/s
 Unlike the upper mantle, where possible compositional effects on seismic velocity are small relative to the observed velocity reduction [Cammarano et al., 2003], in the crust one would commonly assume that the observed lower crustal velocity decrease (to 3.7–3.8 km/s) implies a felsic lower crust. In the following discussion we instead show that the composition is unlikely to be felsic and that the low velocity in the lower crust is most likely the result of a combination of high temperature and partial melt as in the upper mantle.
 Little information on the temperature of the lower crust is available for Ethiopia. Interpreting temperature from seismic velocity in the lower crust is difficult because the value of dVs/dT at high temperatures (>600°C) for crustal lithologies is not well constrained. For temperatures up to 600°C (still significantly below that plausible for the Ethiopian lower crust based on the previous arguments for a 1250–1300°C upper mantle) Kern et al.  show an average of ∼0.21 m/s/K decrease in shear wave velocity as temperature increases. Using a value of 0.21 m/s/K for dVs/dT requires a temperature increase of 1000–2000°C above that of the lower crust in Kenya (∼1500–2500°C) to create the observed decrease of 0.2–0.4 km/s. That temperature range is clearly unrealistic, likely because the dVs/dT value at lower temperatures is not applicable for higher temperatures. Kampfmann and Berckhemer  show a sudden decrease in the shear wave modulus at ∼800°C for mafic rocks, which would cause dVs/dT to rapidly increase above 800°C. Without better constraint on dVs/dT we are unable to quantitatively convert observed seismic velocity in the lower crust to temperature, but using either the Kern et al.  results or the Kampfmann and Berckhemer  results, the temperature in the lower crust is predicted to be 800°C or greater. A hot lower crust is also consistent with extrapolation of the predicted anomalously high upper mantle temperature.
 An independent constraint on the state of the lower crust comes from electrical conductivity data collected along a cross-rift line (Figure 2) by Whaler and Hautot . These data indicate a strong similarity in crustal structure between the NMER and the NW Ethiopian Plateau; the resistivity structure beneath the NMER and the Ethiopian Plateau is nearly the same (as are the seismic velocity profiles) with lower crustal conductivity of ∼4.10−2 S/m. In contrast the Somali Plateau shows a lower conductivity, below 1.10−2 S/m. The integrated lower crustal conductivity (conductance) beneath the rift and the NW Ethiopian Plateau is at least 400 S (siemens), the average conductance for Phanerozoic lower crust, while the conductance below the Somali Plateau is only ∼100 S [Whaler and Hautot, 2006]. High electrical conductivity in the lower crust implies saline fluids, graphite, or melt. Although we cannot a priori rule out any of these, the high temperatures implied by the previous discussion suggest anhydrous equilibrium conditions and the recent magmatism makes melt the natural explanation. If we make the assumption that melt is present in the lower crust and causes the high observed conductivity, we can constrain the amount present beneath the Ethiopian Plateau using the observed conductivity values [e.g., Schilling et al., 2006] independent of seismic velocity.
 Because of the large difference in the conductivity of basaltic melt (up to 10 S/m) and solid rock (<0.01 S/m) [Roberts and Tyburczy, 1999], electrical conductivity is very sensitive to melt percent. This electrical conductivity can be modeled using the Hashin-Shtrikman bounds [Hashin and Shtrikman, 1962], with the upper bound describing an ideal, connected melt network. This bound will describe the lower limit on the amount of melt present. The lower Hashin-Shtrikman bound describes isolated melt pockets and limits the maximum amount of melt present. If we model the upper Hashin-Shtrikman bounds using melt conductivity values of 1 S/m, 3 S/m, and 10 S/m, using a matrix conductivity of 0.001 S/m, we predict a melt percent between 0.5 percent (melt conductivity of 10 S/m) and 5 percent (melt conductivity of 1 S/m) as the lower bound on melt present (Figure 15). This assumes the ideal case of interconnected melt pathways. If some of the melt is present in isolated pockets the melt percent required to produce the observed bulk conductivity is higher. Archie's Law is a reasonable way to estimate a more complex melt system between the Hashin-Shtrikman end-members of completely isolated and completely connected networks [Hermance, 1979; Merzer and Klemperer, 1992]. If we use this method assuming a complex melt system [Hermance, 1979], we estimate between 6% and 20% melt in the lower crust (Figure 15) rather than 0.5 to 5% estimated for ideally interconnected pathways. The presence of melt in the crust beneath the NW plateau away from the MER is supported by elevated bulk crustal Vp/Vs ratios up to 1.9 [Dugda et al., 2005; Stuart et al., 2006]. A Vp/Vs ratio of 1.9 is commonly taken to represent a contribution from melt [e.g., Menke et al., 1998]. This lower crustal melt that we infer is not only present beneath the MER, it is present beneath the entire Ethiopian Plateau even hundreds of kilometers from the rift valley. The plateau has been the location of extensive off-axis Mio-Pliocene and Quaternary volcanism [Abate et al., 1998; Conticelli et al., 1999], possibly reflecting this continued presence of melt far from the rift axis.
 As a final possibility, a lower crust of felsic granulite across the entire region could cause the lower crust to have a shear wave velocity of ∼3.7 km/s [Christensen, 1996]. Such a composition, however, is unlikely. Mafic lower crustal xenoliths are described for both the Arabian-Nubian Shield to the north [McGuire and Stern, 1993; Al-Mishwat and Nasir, 2004] and along the trend of the Mozambique Belt to the south [Mansur et al., 2006]. These terranes meet in Ethiopia, therefore one might expect that the Ethiopian lower crust is composed of a similar mafic lithology. Vp/Vs ratios provide a stronger constraint on composition than either Vp or Vs alone. Vp/Vs ratios in Ethiopia have previously been estimated for the whole crust using the Zhu and Kanamori receiver function stacking method [Dugda et al., 2005; Stuart et al., 2006], but are estimated here for just the lower crust by using the ratio of the lower crustal shear wave velocity measured in our inversions to the lower crustal compressional wave velocity modeled using wide-angle data [Maguire et al., 2006]. The Vp/Vs ratio for the lower crust in Ethiopia ranges from 1.81 to 1.88. As shown in Figure 16, felsic granulite has a Vp/Vs ratio of 1.79, lower than that observed in Ethiopia, and mafic granulite has a Vp/Vs ratio of 1.82, within our observed range.
 The regional mafic lower crustal xenoliths and the observed Vp/Vs ratios cannot absolutely eliminate a felsic lithology in the lower crust as the cause of the low observed lower crustal shear wave velocity in Ethiopia. However, with strong evidence for (1) high temperature causing the upper mantle low-velocity anomaly (section 6.1), which would conduct into the lower crust to reach thermal equilibrium; (2) strong evidence for the presence of melt in the lower crust from conductivity measurements (section 6.2); and (3) suggestions of a mafic lower crust from Vp/Vs ratios and along-strike lower crustal mafic xenoliths (section 6.2), it is quite unlikely that a felsic lower crust is responsible for the low observed shear wave velocities.
7. A Hot, Ductile Lower Crust and Upper Mantle With Partial Melt
 Additional lines of evidence support our interpretation of a hot, ductile lower crust with partial melt above a hot mantle lithosphere, also with partial melt. Geodetic observations indicate that extension between the Nubian and Somalian plates is distributed over a broader region than the 80–100 km width of the MER [Bendick et al., 2006]. The authors conclude that the high temperature and mechanical stretching in the MER create distributed, continuous extension in a viscous lower crust over a wide region [Bendick et al., 2006]. If extension in the lower crust is broadly distributed but upper crustal extension is localized within the structural rift, lower crustal thickness should not change as much across our study region as upper crustal thickness. This can be tested by looking at the active source seismic profiles. As noted above, very little to no variation in lower crustal thickness is modeled in the profiles (if we exclude the modeled high-velocity lower crust beneath the NW shoulder for reasons discussed in section 4) either along or across the axis of the rift [Maguire et al., 2006], further supporting distributed extension at this depth.
 Results from shear wave splitting studies [Gashawbeza et al., 2004] (Figure 17) also support active deformation distributed far from the rift axis. In studies focused on the Ethiopian rift valley, observed splitting of similar magnitude and azimuth is interpreted to result from thin pockets of melt within the lithosphere aligned parallel to the rift [Ayele et al., 2004; Kendall et al., 2005]. The consistent delay times and azimuths of the fast polarization direction of splitting over the entire region of low lower crustal velocities extending hundreds of kilometers away from the rift itself (Figure 17) suggests that aligned pockets of melt may be present over this broad region, particularly when viewed in combination with the evidence for a hot upper mantle and lower crust over this same region. The previous interpretation of this data set uncritically assumed melt-filled cracks would only be present beneath the rift and so concluded that the uniformity of the splitting represented a Proterozoic lithospheric fabric [Gashawbeza et al., 2004]. We do not exclude preexisting fabrics as a contributing cause of the observed splitting pattern, but on the basis of the additional evidence presented here we believe that melt-filled cracks likely contribute to the observed splitting far from the rift itself and a combination of the two effects is our preferred interpretation of the splitting results over the entire region. In addition to the studies described previously, upper mantle seismic tomography [Benoit et al., 2006a; Bastow et al., 2008] shows a broad thermal upwelling beneath the Ethiopian Plateau down to depths of at least 400 km that is laterally extensive for about 400 km, providing further support for an elevated thermal regime.
 The combination of low shear wave velocity, high lower crustal conductivity, and high upper mantle temperature makes high lower crustal temperature with the presence of partial melt beneath the MER and the NW Ethiopian Plateau the most reasonable interpretation for our observations of low shear wave velocity in the lower crust. The melt is likely present in aligned pockets perpendicular to the extension direction [e.g., Ayele et al., 2004; Kendall et al., 2005], as suggested by shear wave splitting measurements, although even our minimum estimate of melt percent from lower crustal electrical conductivities and upper mantle seismic velocities (0.5%; Figures 14 and 15) is far higher than the 0.1% suggested by Ayele et al.  and Kendall et al. . Because melt fraction inferred from shear wave splitting relies on assumptions about crack aspect ratio, Ayele et al.  and Kendall et al.  simply chose the maximum plausible aspect ratio (100:1) to minimize the melt fraction required to explain their observations and the observations of Gashawbeza et al. . Simply reducing the crack aspect ratio to 20:1 fits the shear wave splitting data with the same (∼ 90 km) partial melt layer thickness and our increased minimum (0.5%) melt fraction. Within the crust itself the melt appears to be more pervasive beneath the NMER than beneath the NW plateau on the basis of higher delay times in crustal shear wave splitting [Keir et al., 2005] and higher values for Poisson's ratio within the rift itself [Dugda et al., 2005; Stuart et al., 2006].
8. Conclusions and Implications
 Our joint inversion of receiver function waveforms and surface wave dispersion curves in Ethiopia leads to the following conclusions:
 1. The shear wave velocity of the upper mantle beneath Ethiopia is low (4.0–4.2 km/s) over an area ∼400 km wide including both the Main Ethiopian Rift and the adjacent plateaus, ∼0.3 km/s lower than the shear wave velocity of the rift margins elsewhere in the EARS.
 2. The shear wave velocity of the lower crust across this same broad region is 3.7–3.8 km/s, ∼0.3 km/s lower than the shear wave velocity of the lower crust elsewhere in the EARS.
 3. The upper mantle immediately beneath the Moho is between 1250°C and 1300°C with up to 4% melt. The lower crust has between 0.5% and 20% melt (assuming a fully interconnected melt network and a more complex melt network, respectively) over a region much broader (∼400 km) than the ∼20 km wide zone of active intrusion within the MER [Keranen et al., 2004] or the ∼100 km wide rift valley.
 4. Strain in the lower crust is spread over the ∼400 km region with low lower crustal and upper mantle shear wave velocities, more broadly distributed than strain in the upper crust (localized within the ∼100 km wide rift).
 5. The MER is an example of a “narrow” rift forming in hot continental lithosphere.
 The MER is traditionally considered to have developed in initially thick, strong cratonic lithosphere. Previous studies [Buck, 1991; Brun, 1999] have demonstrated that the narrow rift mode preferentially evolves in cool, thick lithosphere with high depth-integrated strength. Here we show that this condition is unlikely for the MER, and that the rift instead appears to have developed in a region with a weak and ductile lower crust and hot upper mantle. The limits of the thermally modified lithosphere closely follow the limits of the Ethiopian flood basalt province (Figure 1). Plume impingement at ∼30 Ma led to the rapid thinning of the lithosphere and the subsequent transfer of heat from the plume material to the crust and upper mantle [Dugda et al., 2007].
 We have shown that the Ethiopian lithosphere was likely hot and weak prior to rifting, yet a narrow rift (<100 km wide at the surface) developed nonetheless. Dynamic models using finite element methods [e.g., Dyksterhuis et al., 2007] show that narrow rifts can develop in a region with a weak lower crust if a single weakness is present in the inherited structure. A Neoproterozoic, Pan-African suture runs through Ethiopia, and on the basis of the step in crustal thickness across the CMER from the eastern to the western plateaus, Keranen and Klemperer  interpreted the rift to follow the suture. Stern  uses geochemical evidence to make the same argument, albeit with less ability to spatially constrain the suture. This Neoproterozoic suture likely provided the initial weakness necessary for narrow rifting to occur within the upper crust in the MER even with a weak lower crust that is deforming over a region >400 km wide including the entire northwestern Ethiopian Plateau.
 In the EARS further south in Kenya and Tanzania, the eastern and western branches have also formed as narrow rifts. These two branches wrap around the boundaries of the Tanzania craton [Nyblade and Brazier, 2002], and Pan-African suture thickening of the crust against the eastern margin of the Tanzania craton is suggested as a control on the location of the eastern branch [Nyblade and Pollack, 1992; Tesha et al., 1997]. In Kenya and Tanzania the thermal structure of the region away from the Cenozoic rift is vastly different from the structure of the plateaus adjacent to the MER [Julia et al., 2005; Dugda, 2007]. In contrast to the low-velocity lower crust away from the Cenozoic rift in Ethiopia, the lower crust in Tanzania has shear wave velocities between 3.9 and 4.1 km/s away from the Cenozoic rift [Julia et al., 2005]. P wave tomography from the KRISP experiment [Mechie et al., 1997] and the Kenya Broadband Seismic Experiment [Park and Nyblade, 2006] show a narrow, elongate low-velocity region beneath the rift itself, with normal seismic velocity (Vp > 8.0 km/s) beneath both rift shoulders. The lower crust and upper mantle in Kenya and Tanzania appears mafic, cool, and strong, with only the rift itself hot and weak.
 The northern and southern regions of the EARS have very different thermal profiles and vertical strength profiles (Figure 18 and Table 2), with an order-of-magnitude greater integrated lithospheric strength in the region adjacent to the eastern and western branches than in the plateaus adjacent to the Main Ethiopian Rift. Nonetheless, both systems have developed as classic examples of narrow, continental rifts. This similar development in diverse settings indicates that inherited structures, which were present in both cases as the boundary of the craton to the south and as the suture to the north, were more important in controlling the mode of rifting than the rheological stratification. These inherited structures appear to be the primary control on the mode of continental rift development in East Africa.
Table 2. Parameters Used in the Calculation of the Geotherms and Strength Envelopes in Figure 18 for Rift Shoulders in Ethiopia and Kenyaa
Range of heat flow for Kenya from Nyblade et al. ; heat flow listed for Ethiopia is estimated on the basis of the heat flow required to match temperature constraints at depth using selected values for other parameters. Moho depth is based on the average depth beneath the eastern and northwestern shoulders, thickness of the upper crust is from active source modeling [Maguire et al., 2006], and strain rate is from both GPS [Bendick et al., 2006] and crustal thinning estimates for Ethiopia. Values for the remaining thermal parameters are from Artemieva and Mooney . Strength curves assume power law failure for granite, mafic granulite, and peridotite for the upper and lower crust and upper mantle, respectively [Ranalli, 1997].
Surface heat flow (mW/m2)
Upper crustal conductivity (W/m/K)
Lower crustal conductivity (W/m/K)
Upper crustal heat production (W/m3)
Lower crustal heat production (W/m3)
Thickness of upper crust (km)
Moho depth (km)
Strain rate (1/s)
 EAGLE data acquisition was funded by National Science Foundation (NSF) Continental Dynamics grant EAR0208475 and the Natural Environment Research Council. K. Keranen thanks the SEG Foundation and Anadarko for financial support during her Ph.D. We appreciate the continued willingness of Ethiopian scientists to host and collaborate with U.S. geophysicists. We thank Graham Stuart for sharing the EAGLE Phase I event data; EAGLE leadership for the coordination of the seismic project; the EBSE leaders for sharing their data; EAGLE and EBSE participants who helped with data collection; and Mulugeta Dugda, Tilahun Mammo, Ewenet Gashawbeza, Tyrone Rooney, Norm Sleep, Elizabeth Miller, and Chris Mattinson for valuable discussions. All five contributing authors of this manuscript significantly influenced and helped shape the final result. We appreciate very careful and constructive reviews from G. Corti and R. Bendick as well as helpful comments from the editor.