Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 A 500-km wide region separating the East African and Ethiopian plateaus is one of the few places within the African Superswell with elevations below 1 km. The region encompasses much of southeastern Sudan and northern Kenya and experienced both Mesozoic and Cenozoic rifting. Crustal and uppermost mantle structure is investigated in the region by modeling Rayleigh wave dispersion measurements. Modeling results give Sn velocities of 4.1–4.3 km/s and average crustal thickness of 25 ± 5 km, some 10–15 km thinner than the crust beneath the East African and Ethiopian Plateaus. The isostatic response from 10 to 15 km of crustal thinning is sufficient to account for the low elevations between the plateaus.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The African Superswell, as described by Nyblade and Robinson , is comprised of the Ethiopian, East African and southern African Plateaus, as well as a bathymetric swell in the southeastern Atlantic Ocean basin. One of the few regions within the Superswell where elevations are less than 1 km is between the East African and Ethiopian plateaus in southeastern Sudan and northern Kenya (Figure 1). This region is ∼500 km wide and has experienced both Mesozoic and Cenozoic rifting. The Mesozoic rifts belong to the Central African Rift System that formed during the breakup of Gondwana [Browne et al., 1985; Binks and Fairhead, 1992; Bosworth, 1992] (Figure 1). During the Cenozoic, some of the Mesozoic rifts were reactivated, and new rifts formed as part of the East African rift system (Figure 1) [Hendrie et al., 1994; Ebinger et al., 2000].
 A possible explanation is that multiple episodes of rifting thinned the crust across the region, and that isostatic adjustments from the thinning resulted in lower elevations. Previous seismic studies of crustal structure in northern Kenya show that the crust has been thinned, at least locally [Simiyu and Keller, 1997, Prodehl et al., 1997 and the references therein], however whether or not the crust has been thinned across the entire region remains unknown. Hence, in this study we investigate crustal structure between the plateaus to determine if the lower elevations everywhere could have resulted, isostatically, from crustal thinning.
2. Surface Wave Inversion
 To investigate crustal structure, we first generated surface wave group velocity maps by inverting dispersion measurements of fundamental mode Rayleigh waves. The dispersion measurements were made using data from permanent and temporary broadband seismic stations in eastern Africa, the Arabian Peninsula, the Middle East, and the Seychelles for periods of 10 to 100 s. Permanent station locations are given by Pasyanos . Temporary stations that were used come from the Saudi Arabian PASSCAL Network [Vernon et al., 1996], the Tanzania Broadband Seismic Experiment [Nyblade et al., 1996], the Ethiopian Broadband Seismic Experiment [Nyblade and Langston, 2002], and the Kenya Broadband Seismic Experiment [Nyblade and Langston, 2002]. We utilize surface waves recorded on stations surrounding the study area for this study because the study area is not easily accessible for field work.
 Group velocity measurements were made using the PGplot Surface Wave Multiple Filter Analysis code (PGSWMFA) (Ammon, personal communication, 2003) by applying a narrow-band Gaussian filter to vertical component seismograms, and then picking the maximum amplitude at each period. To maximize ray path coverage, dispersion measurements made in this study were combined with previous dispersion measurements from Eurasia and North Africa from Pasyanos et al.  and Pasyanos and Walter  yielding a total of ∼50,000 Rayleigh wave measurements. Ray paths for the new dispersion measurements along with the previous measurements for the periods of interest (10–60 s) are shown in the supplemental material.
 Group velocity dispersion measurements were inverted using the method outlined in Pasyanos et al.  to produce maps showing spatial variability in group velocity for periods of 10 to 100 s. The inversion was performed using a 1.5° × 1.5° grid and a conjugate gradient method, which works well on sparse linear systems. In the inversion, the data were weighted using quality estimates of the dispersion measurements. A smoothing constraint was also applied to the data to control the tradeoff between fitting the data and smoothing the models. Group velocity maps obtained from the inversion for a number of periods are provided as supplemental material.
 Results of the inversion show substantial lateral velocity heterogeneity, and this can be easily seen using average group velocity dispersion curves for the different tectonic regions (Figure 2). Shorter period group velocities (10–30 s) generally reflect shallower structure, such as sedimentary basins and areas of thinner crust, while longer period group velocities (40–60 s) begin to sample deeper (mantle) structure. In southeastern Sudan, slower than average velocities can be seen at 10 s period, faster than average velocities from 20–40 s period, and slower than average velocities at 50 and 60 s period (Figure 2). At 10 and 20 s periods, group velocities in northern Kenya are faster than average, but slower than average from 30 to 60 s period (Figure 2). The group velocity maps (supplemental material) also show velocity variations outside the study region, and a discussion of these variations can be found in Pasyanos .
 To evaluate the resolution of the group velocity models, we generated synthetic travel times using the same ray paths as our data for different ‘checkerboard’ models, and then inverted the synthetic dispersion data using the same parameters we used to invert our data. The results of three checkerboard tests at 20, 40 and 60 s using 2° × 2° squares are provided in the supplementary material. The checkerboards are well resolved for each test in continental regions of the model.
3. Constrained Grid Search
 Next, the surface wave group velocities were used with a grid search technique to create a layered earth model. A grid search method was chosen to do this because a priori information can be easily included in the model search, and because it is straightforward to see how various tradeoffs in model parameters affect the results.
 The model domain for the grid search was parameterized using 1° × 1° blocks and three layers in depth: a near surface sediment layer, a crustal layer, and an upper mantle layer. The upper mantle was parameterized as a 30 km thick layer over the ak135 model [Kennett et al., 1995] to a depth of 150 km. In the grid search, crustal thickness (H) was allowed to vary between 15–45 km, sediment thickness (S) from 0–6 km, upper mantle S velocities (Sn) from 4.1–4.6 km/s, and crustal Poisson's ratio (σc) from 0.24–0.34. We assumed that sediment S velocities were 1.9 km/s from the surface to 2 km depth, and then 2.8 km/s for sediments deeper than 2 km. The crustal S velocity and the mantle Poisson's ratio were held constant for each inversion, though several values of each were tested. We fixed the crust and upper mantle structure in parts of Ethiopia, Tanzania, Uganda and Kenya where constraints on crustal thickness, sediment thickness, Poisson's ratio, and Pn velocity exist from seismic refraction, Pn tomography, and receiver function studies [Brazier et al., 2000; Last et al., 1997; Dugda et al., 2005; Mackenzie et al., 2005; Prodehl et al., 1997 and references therein] (Figure 3).
 Best fitting models were selected based on a misfit estimation that minimizes the group velocity residual divided by the uncertainty measurements of the group velocity dispersion curve generated from the surface wave velocity maps. The uncertainty measurements from the surface wave maps are estimated by using a bootstrapping method. The misfit equation is:
where Um is the group velocity from the model, Ud is the group velocity from the surface wave velocity maps (i.e., the data), sd is the data uncertainty, and i is the loop over the periods from 1 to n [Pasyanos and Walter, 2002]. A perfect fit to the data would have a misfit of 0, while a model fitting all points at one standard deviation would produce a misfit of 1, providing a scale to assess how well the data is being fit by the models. Using this scale, all of the models with misfit functions <1 are considered reasonable models with respect to the data uncertainty.
Figures 3a–3d show maps of the model parameters (H, S, Sn, and σc) obtained from the grid search. Throughout the region of low elevations between the Ethiopian and East African Plateaus, crustal thickness ranges from 30 to 20 km (Figure 3a), and average sediment thicknesses of 1 km are obtained, with thickness increasing to 3 km along the Kenya coast and in the northwestern portion of the rifted region in Sudan (Figure 3b). Sediment thickness locally within the rift basins probably exceeds these estimates, which represent average structure across the basins as a result of the grid spacing used in the modeling. The Sn velocities range between 4.1–4.3 km/s throughout the region of low elevation (Figure 3c), and crustal Poisson's ratio ranges from 0.26–0.30 with an average of 0.28 (Figure 3d).
 To assess the effect of using a single average S wave velocity (set as an a priori constraint) for the crust, we ran separate grid searches using S velocities that ranged from 3.52–3.71 km/s. The effects on the model parameters are shown in Table 1. While changing the average crustal velocity had little effect on the average Moho depth, average Sn velocities, or average crustal Poisson's ratio, the thickness of the sediment layer increased by ∼0.25 km for every ∼0.3 km/s increase in average crustal velocity. The mantle Poisson's ratio was also varied in separate grid searches from 0.27 to 0.29, but this had little effect on the grid search results.
Table 1. The Effects of the Model Parameters When the Average Crustal Velocity and Average Mantle Poisson's Ratio are Held Constant for Each Grid
Average Crustal Vs, km/s
Average Mantle σ
Average Sediment Thickness, km
Average Crustal Thickness, km
Average Sn Velocity, km/s
Average crustal σ
 By examining the rms misfit values associated with the parameters used in the grid search, we can evaluate the uncertainties in each parameter. 1-D slices through the parameter space and the corresponding rms misfit for crustal thickness, Sn velocity, crustal Poisson's ratio, and sediment thickness for southern Sudan and northern Kenya are shown in the supplementary material. In these regions, the uncertainties in crustal thickness are ±5 km, ±0.15 km/s for Sn velocity, ±0.02 for Poisson's ratio, and ±1 km for sediment thickness.
 The results of the grid search indicate that the crust is thinned significantly in the region of low elevation between the Ethiopian and East African Plateaus. The average crustal thickness in the plateaus is ∼38 km, while the average crustal thickness between the plateaus is ∼25 ± 5 km. Low Sn velocities (4.1–4.3 km/s) also characterize the region of low elevations. These Sn velocities are comparable to Sn velocities found under the Cenozoic Kenya Rift and Main Ethiopian Rift [Simiyu and Keller, 1997; Prodehl et al., 1997 and references therein; Mackenzie et al., 2005], and indicate that upper most mantle temperatures in this region are elevated.
 Using the crustal thickness estimates shown in Figure 3a, we can now investigate whether the low elevations between the Ethiopian and East African Plateaus could result from crustal thinning. Assuming Airy isostasy, we calculate the expected difference in average elevation across the region of thinned crust between the Ethiopian and East African Plateaus. For Airy isostasy, uplift U = r(ρm − ρc)/ρc, where r is the difference in crustal thickness, ρm is the mantle density, and ρc is the crustal density. For r ∼ 13 km (i.e., 25 km versus 38 km thick crust), ρc of 2.8 g/cm3, and ρm of 3.2 g/cm3, 1.8 km of differential elevation can be accounted for by isostasy. This result is consistent with the observed difference in elevation between the Ethiopian and East African Plateaus and the region of low elevation in between them (Figure 1).
 Estimates of crustal thickness shown in Figure 3 are also consistent with gravity models of the region. Bouguer gravity anomalies within the region of low elevation are ∼−60 to 100 mGals [Simiyu and Keller, 1997], at least within the vicinity of the Cenozoic Turkana rift. An anomaly of this size can be accounted for with a thin crust (∼20 km) over a hot (i.e. low density) mantle [Simiyu and Keller, 1997].
 The results of this study indicate that lower elevations found between the Ethiopian and East African Plateaus may reflect an isostatic response to crustal thinning. The crust is probably thinner than normal across this region because of the superposition of the multiple phases of rifting in the Mesozoic and Cenozoic. If the crust in this region had not been thinned by ∼10–15 km, then the high elevation of the Ethiopian and East African Plateaus would in all likelihood be continuous and the plateaus would not be seen as geographically distinct regions of uplift. Because the variations in elevation can be readily attributed to crustal thinning, there is little reason to suspect fundamental changes in mantle structure between the Ethiopian and East African Plateaus.
 We thank Laike-Mariam Asfaw, Atalay Ayele, and the technical staff of the Geophysical Observatory of Addis Ababa University for their help with the Ethiopian Broadband Seismic Experiment, and Silas Simiyu for his help with the Kenya Broadband Seismic Experiment. We also thank two anonymous reviewers for their helpful comments. This research has been funded by the National Science Foundation (grants EAR 993093 and 0003424).