New estimates of seismic anisotropy from shear wave splitting measurements in eastern Africa reveal a pattern of seismic anisotropy dominated by a NE alignment of fast polarization directions with local changes around the thick Archean lithosphere of the Tanzania craton. The overall pattern is consistent with mantle flow from the African superplume but not with absolute plate motion, a plume head, or fossil anisotropy in the lithosphere. In combination with tomographic images of the African superplume, this finding suggests that plateau uplift, volcanism, and continental breakup along the Afro-Arabian rift system is strongly influenced by flow from the lower mantle and indicates a connection between lower mantle processes and the tectonic deformation of the Earth's surface.
 The high plateau, rift valleys, and volcanic fields in eastern Africa and western Arabia are first-order features of the African and Arabian plates whose origins have long been attributed to a variety of mantle processes, including plumes [Pik et al., 2006], edge flow convection [King and Ritsema, 2000], and the upwelling of hot buoyant rock in the lower mantle [Forte et al., 2010]. Understanding the flow field in the upper mantle beneath these features is not only key to determining how they formed but also for unraveling the geodynamic processes responsible for continental breakup along the Afro-Arabian rift system, characterizing how the mantle convects beneath the African and Arabian plates, and understanding the nature of one of the largest seismic anomalies in the mantle, the so-called African superplume [Hansen et al., 2012; Simmons et al., 2007; Simmons et al., 2011]. Patterns of seismic anisotropy in the upper mantle can reveal the direction of flow and shed light on mantle processes, provided that alternative causes of anisotropy can be ruled out. Here we present new estimates of seismic anisotropy in eastern Africa from shear wave splitting measurements, combine them with published measurements, and use the data ensemble to map out the pattern of seismic anisotropy along the entire Afro-Arabian rift system.
2 Data, Methods, and Results
 Broadband seismic data used in this study come from temporary and permanent seismic stations in Uganda, Tanzania, Zambia, and Malawi that were installed and operated as part of AfricaArray [Nyblade et al., 2011]. The temporary stations were deployed in three phases starting in August 2007. Twenty stations were initially deployed in Uganda and northwestern Tanzania for 18 months, and then most of the stations were redeployed in January 2009 across southern Tanzania (Figure 1). In August 2010, the stations were redeployed for a second time in Zambia and operated through July 2011 (Figure 1). The permanent stations were installed at various times beginning in 2006 (Figure 1).
 Shear wave splitting measurements were made using recordings of earthquakes with magnitudes (Mw) ≥5.5 and epicentral distances from each station of 90° to 140°. To calculate individual splitting parameters (i.e., delay time and fast polarization direction), SKS and SKKS phases were first identified and windowed using Splitlab [Wüstefeld et al., 2008]. These phases were then analyzed using the method of Silver and Chan  and stacked using the method of Wolfe and Silver  modified by Restivo and Helffrich , resulting in a single splitting measurement for each station (Figure 2). Stacking assumes that anisotropy in the region is due to a single non-dipping layer that has very little azimuthal dependence. Walker et al.  tested dipping-axis and two-layer models for this region and concluded that these models were inconsistent with their observations. We have also investigated variations in splitting parameters as a function of initial polarization angle and found no visible trends (Figures S2 and S3 in the auxiliary material) although for many stations there are insufficient data to make firm conclusions. Obreski et al.  illustrated that multilayer anisotropy, if present, might be best developed within the regions of Cenozoic rifting.
 To minimize the influence of noise, the data were bandpass filtered between 0.02 and 0.20 Hz. This choice of corner frequencies prevents noisy data from producing results with erroneously high signal-to-noise ratios [Walker et al., 2004]. Individual results for each station were visually inspected, and only the highest-quality results were included in the final stack. To determine which results were of high quality, we examined the energy on the transverse component (Figure 2a), the linearity of particle motion (Figure 2b), and the misfit grid (Figure 2c). The complete dataset includes 771 individual splitting measurements (687 SKS phases and 84 SKKS phases) from 306 unique events (no null measurements were observed). Stacked splitting results can be found in Table S1, details regarding the number of events and phases included in each stack can be found in Table S2, individual splitting times can be found in Table S3, and the misfit grids of each stack are shown in Figure S1. In addition, Table S4 summarizes information from events used in this study for which both SKS and SKKS phases were be analyzed. There are too few results (Table S4) to draw any conclusions S4 about the presence or absence of lower mantle anisotropy [e.g., Wang and Wen, 2007].
 Results are shown in Figure 1, along with shear wave splitting measurements from several other studies in eastern Africa and Arabia [Walker et al., 2004; Vinnik et al., 1992; Wolfe et al., 1999; Gao et al., 1997; Barruol and Ben Ismail, 2001; Ayele et al., 2004; Schmid et al., 2004; Kendall et al., 2005; Hansen et al., 2006; Kaviani et al., 2011] and major geologic features of the region. The Precambrian tectonic framework of eastern Africa includes the Archean Tanzania craton, situated in the middle of the East African Plateau, surrounded by several Proterozoic orogenic belts. The Cenozoic rift system, which in eastern Africa extends from central Mozambique northward to the Afar Depression in Ethiopia, developed within the Proterozoic orogenic belts and splits into two branches (eastern and western) around the Tanzania craton (Figure 1). The Precambrian basement of northern Zambia consists of a number of Proterozoic terrains, as does the Arabian Peninsula.
 The pattern of fast polarization directions in northern Zambia and Malawi generally trend to the NE, but in southernmost Tanzania, the fast polarization directions are generally parallel to the margin of the Tanzania craton, bifurcating around its southern end. To the east and west of the Tanzania craton, the fast polarization directions are also nearly parallel to the margins of the craton, but along the northern margin, the fast polarization directions at stations ROTI, MALE, TALE, and KAKA are oriented NNE, almost perpendicular to the craton' s margin. Average splitting times in the Proterozoic orogenic belts around the craton do not show any clear pattern between stations with fast polarization directions parallel versus orthogonal to the craton margin (Figure S4).
 Within the craton, where the lithosphere is thickest [Adams et al., 2012], the splitting times tend to be smaller, but the overall pattern of fast polarization directions are similar to the pattern in the surrounding mobile belts (Figure S4). For example, three stations along the eastern margin of the Tanzania craton have fast polarization directions oriented to the north and NE, consistent with fast polarization directions in the orogenic belt to the east of the craton, while stations in the center and western side of the craton mostly have fast polarization directions oriented to the north and NW, consistent with the fast polarization directions in the orogenic belts to the west of the craton. In Ethiopia, the fast polarization directions strike to the NE, with a small rotation between the stations inside and outside the Main Ethiopia Rift [Kendall et al., 2005], and in the Arabian Peninsula, the fast polarization directions are oriented to the north [Hansen et al., 2006]. Overall average splitting times from East Africa tend to be somewhat lower than in Ethiopia or Arabia (Figure 1).
 Interpretations of seismic anisotropy are commonly based on xenolith studies and laboratory experiments which indicate that upper mantle anisotropy is most likely caused by the lattice-preferred orientation (LPO) of olivine resulting from shear strain [Mainprice and Silver, 1993; Zhang and Karato, 1995; Tommasi et al., 2000]. This LPO may indicate shear strain beneath the lithosphere caused by mantle flow or within the lithospheric mantle caused by past tectonic events (e.g., fossil anisotropy) [Silver and Chan, 1988]. Alternatively, seismic anisotropy can develop from shape-preferred orientation of structures in the lithosphere [Holtzman and Kendall, 2010].
 Previous studies in eastern Africa have found that the fast polarization direction in volcanically active rift valleys in Ethiopia and Kenya is to the first-order rift-parallel or NNE [Walker et al., 2004; Gao et al., 1997; Kendall et al., 2005; Gao et al., 2010] (Figure 1). This pattern is not consistent with the expected flow direction in the sublithospheric mantle orthogonal to the strike of the rift, such as observed in mid-ocean ridges [e.g., Nowacki et al., 2012], or to absolute plate motion, leading the authors of those studies to attribute the pattern to magma-filled cracks in the rifted lithosphere or to a combination of magma-filled cracks and mantle flow in the asthenosphere [Gao et al., 2010]. Away from the rift valleys, previous studies have attributed the fast polarization directions in eastern Africa to fossil anisotropy because they are consistent with the strike of Precambrian orogenic systems and not consistent with the direction of anisotropy predicted by absolute plate motion or a radial pattern created by flow outward from the center of a plume head [Walker et al., 2004; Gashawbeza et al., 2004]. In Arabia, the north striking fast polarization direction has been attributed to a combination of plate- and density-driven flow in the sublithospheric mantle [Hansen et al., 2006], and on a larger scale, the pattern of fast polarization directions across the African and Arabian plates has been attributed to plate-scale flow in the upper mantle associated with the African superplume [Forte et al., 2010; Behn et al., 2004].
 The broader pattern of anisotropy shown in Figure 1, compared to the earlier studies, remains incompatible with explanations invoking mantle flow in the direction of absolute plate motion or mantle flow within a plume head. Nowhere in eastern Africa or Arabia does the pattern of fast polarization directions exhibit a radial pattern, as would be expected from mantle flow away from the center of a plume head.
 For many of the stations in the Proterozoic orogenic belts around the Tanzania craton, the observed margin-parallel fast polarization directions are consistent with the strike of the orogenic belts and therefore with a fossil anisotropy interpretation [e.g., Savage, 1999]. Similarly, the fast polarization directions in the center and western side of the craton can be associated with WNW structural trends [Walker et al., 2004; Begg et al., 2009; Holmes, 1951]. However, the north and NE directions within the eastern third of the craton are not easily explained by fossil anisotropy, nor are the NNE-oriented directions at stations along the northern edge of the Tanzania craton (Figure 1). Stations ROTI and MALE, for example, are located in the Mesoproterozoic Rwenzori belt, which accreted to the craton margin in an N-S direction [Leggo, 1974; Begg et al., 2009]. If fossil anisotropy were at play, then the fast polarization directions would be oriented E-W, not NNE as observed. Stations KAKA and TALE are in the Neoproterozoic Mozambique belt just to the east of the Aswa shear zone, which marks the suture between the Rwenzori and Mozambique belts [Leggo, 1974; Mosley, 1993]. The NNE fast polarization orientation found at these stations is also inconsistent with a fossil anisotropy interpretation, which would predict an NW-SE orientation aligned with the strike of the Aswa shear zone. Thus, while it may be possible to invoke fossil anisotropy to explain some of the shear wave splitting observations, the ensemble of observations cannot be attributed to fossil anisotropy.
 In contrast to the fossil anisotropy interpretation, all of the fast polarization directions, from northern Zambia to the northern edge of the Arabian Peninsula, are consistent with flow in the upper mantle associated with the African superplume [Forte et al., 2010]. Figure 3 illustrates the low-velocity structure of the superplume and its location with respect to the main geologic structures and the shear wave splitting measurements. Within the superplume structure, warm, low-density lower mantle rock beneath southern Africa flows upward to the NE, reaching the upper mantle beneath northern Zambia, where it then flows to the northeast beneath eastern Africa and Arabia (Figure 3). As the flow moves to the northeast, the bifurcation of the fast polarization directions around the Tanzania craton is consistent with the lateral flow in the upper mantle beneath the thinner Proterozoic lithosphere and around the thicker Archean lithosphere of the Tanzania craton [Adams et al., 2012; Obrebski et al., 2010; Walker et al., 2004; Fouch et al., 2000; Ritsema et al., 1998]. The margin-parallel fast polarization directions along the eastern and western sides of the craton are also consistent with the flow continuing in a generally northerly direction. In addition, the NNE fast polarization directions along the northern margin of the craton (stations ROTI, WALE, KAKA, and TALE), which cannot be explained by fossil anisotropy, are fully consistent with mantle flow extending northward beyond the East African Plateau toward Ethiopia and then on to the Arabian Peninsula.
 Thus, we argue that the first-order pattern of fast polarization directions is best explained by mantle flow associated with the African superplume. However, we also acknowledge that there are some second-order variations to the overall pattern of fast polarization directions, especially over short-length scales, and that these variations might be better explained by complex anisotropy (i.e., multilayer) arising from fossil anisotropy in a heterogeneous lithosphere above the superplume-driven flow field in the sublithospheric mantle or, alternatively, in the case of stations within or near to volcanic regions, magma-filled cracks in the rifted lithosphere. In addition, we suggest that the difference in average splitting times between East Africa and Ethiopia and Arabia could result from the combined anisotropic effects of mantle flow and magma-filled cracks.
 We find that only one interpretation, the superplume interpretation, can explain all of the fast polarization directions away from the volcanic rift valleys, from northern Zambia to northern Arabia. The ensemble of observations is not consistent with interpretations invoking absolute plate motion, a plume head, or fossil anisotropy in the lithosphere. This conclusion, in combination with tomographic images of the African superplume, suggests that plateau uplift, volcanism, and continental breakup along the Afro-Arabian rift system is strongly influenced by mantle flow that originates in the lower mantle.
 We thank Samantha Hansen for providing the tomographic image shown in Figure 3 and Jim Gaherty and two anonymous reviewers for their constructive reviews. Field support is gratefully acknowledged from IRIS-PASSCAL, the Ugandan, Tanzanian, and Zambian Geological Surveys, the University of Dar es Salaam and from many individuals at those institutions and at Penn State University. This study was funded by the National Science Foundation (grants OISE-0530062, EAR-0440032, EAR-0824781).