The Mid-Atlantic Ridge at 8–9°S is characterized by a transition from axial valley to axial high and recent episodes of ridge jumping and ridge propagation. We present constraints on the structure of 0–4 Ma crust in this region on the basis of the analysis of wide-angle seismic data from a grid of profiles across and parallel to the current and abandoned spreading centers. A 350–800 m thick oceanic layer 2A, interpreted as high-porosity extrusive basalts, is underlain by a ∼2.0–2.5 km layer 2B with velocities which increase with age and decrease in the vicinity of the pseudofaults. Layer 3 velocities are uniform across the area except for a possible localized anomaly at the ridge axis. The crustal thickness varies from 6–7 km near the pseudofaults formed by ridge propagation to 9–10 km at the segment center of the recently (∼0.3 Ma) abandoned spreading center. Seismically determined crustal thickness and density variations and age-related lithospheric cooling can plausibly account for all observed variations in gravity across the area, and there is no requirement for the thicker crust at the segment center to be underlain by hot mantle. The transition from axial valley to axial high occurs at a crustal thickness of ∼8 km.
 Seismic experiments on oceanic crust formed at slow-spreading ridges such as the Mid-Atlantic Ridge have revealed a layered structure consisting of an upper layer with high velocity gradient, oceanic layer 2, underlain by a lower layer with much lower gradient called oceanic layer 3 [e.g., Spudich and Orcutt, 1980; White et al., 1992]. On-axis experiments at slow-spreading ridges have also revealed lateral variations in both the thickness and velocity of these layers, that can be related to the complex magmatic, tectonic, and hydrothermal processes taking place there. Such experiments have revealed that the crust is thicker at segment centers than at segment ends [e.g., Tolstoy et al., 1993; Hooft et al., 2000], and that the mean crustal thickness can vary significantly between segments [e.g., Muller et al., 1999; Hooft et al., 2000]. Upper crustal velocities commonly decrease toward segment ends, probably due to enhanced tectonism [e.g., Tolstoy et al., 1993; Hosford et al., 2001]. Oceanic layer 2A, a thin, surficial low velocity layer commonly associated with extrusive basalts, typically shows a rapid increase in seismic velocity in the first 10 myr of crustal evolution [Grevemeyer and Weigel, 1996; Carlson, 1998], though its thickness varies little with time [Grevemeyer et al., 1998]. Age-dependent variations in the remainder of layer 2 and in layer 3 are less pronounced, though rapid lateral variations in lower crustal velocity occur where a recent magmatic episode has emplaced hot or molten rocks at the ridge axis [e.g., Navin et al., 1998; Canales et al., 2000a]. Uppermost mantle velocities may also increase with age, from values as low as ∼7.5 km/s beneath zero-age crust [Hooft et al., 2000] to ∼8.0 km/s in mature oceanic crust.
 The axial morphology of slow-spreading ridges is highly variable. Axial rift valleys are caused by necking of the lithosphere at the ridge axis and disappear when the axial lithosphere can no longer support significant stresses [Tapponnier and Francheteau, 1978; Chen and Morgan, 1990a]. Hence rift valleys are not seen at spreading rates greater than ∼80 mm/year [Small and Sandwell, 1989; Malinverno, 1993], because the axial lithosphere is too thin to support them. Rift valleys are also absent at slow-spreading ridges where the crust is anomalously thick. The crustal thickness is a key control on axial relief because oceanic crustal rocks are inherently weaker than mantle rocks [Chen and Morgan, 1990b] and/or because the increased magmatism associated with thickened crust results in a hotter, weaker lithosphere. Indeed, the spreading rate at which the rift-valley/no rift-valley transition occurs has been shown to vary systematically with axial depth, which may be taken as a crude proxy for crustal thickness [Malinverno, 1993]. Similarly, the crustal thickness at which the morphological transition takes place decreases as the spreading rate increases. Hence at 35°N on the Mid-Atlantic Ridge, where the spreading rate is 25 mm/year, a clear rift valley morphology is present where the crustal thickness is 8 km [Hooft et al. 2000], while at 33°S, where the spreading rate is 35 mm/year, an axial high is present where the crustal thickness is 8 km [Tolstoy et al., 1993]. At the Galapagos spreading center where the spreading rate is 50 mm/year, the axial valley disappears when the crustal thickness reaches ∼7 km and there is a broad region of transitional axial morphology when the crustal thickness is between 7 and 8 km [Canales et al., 1997].
 A broad region of axial highs lies between the Ascension and Bode Verde fracture zones at 9–11°S (Figure 1), where the spreading rate is 32–33 mm/year [Bruguier et al., 2003]. These axial highs coincide with deep mantle Bouguer anomaly gravity lows centered at 9°S and 9°40′S (Figure 2) [Bruguier et al., 2003]. Such lows are normally attributed to thickened crust at segment centers [e.g., Tolstoy et al., 1993]. The whole axial region is also anomalously shallow. The depth anomaly has been attributed to asthenospheric flow toward the ridge axis from a robust mantle plume centered 400–450 km east of the ridge axis [e.g., Kincaid et al., 1996]. However, gravity data indicate that the depth anomalies are supported isostatically by thickened crust [Minshull et al., 1998; Bruguier et al., 2003]. The relationship between these depth and crustal thickness anomalies, associated geochemical anomalies, and the so-called Ascension or Circe hot spot, is discussed in detail elsewhere [Bruguier et al., 2003]. Here we report wide-angle seismic constraints on the crustal structure on- and off-axis in the vicinity of the morphological transition from axial valley to axial high.
2. Seismic Data Acquisition and Processing
 Our seismic survey was focused on a single ridge segment between 8°30′S and 9°S, where a complex history of ridge propagation and ridge jumps has been inferred from bathymetric and magnetic data [Brozena and White, 1990; Bruguier et al., 2003]. The spreading history has been interpreted in several ways; in our preferred model based on forward modeling of magnetic anomaly data [Bruguier et al., 2003], the spreading center jumped at ∼0.3 Ma from around 13°30′W to around 13°10′W. Both before and after the ridge jump, the ridge has been propagating northward, leaving distinct pseudofault traces in the bathymetry (Figure 1).
 Seismic data were acquired from RRS James Clark Ross in 1993 using a tuned array of ten air guns with a total volume of 4462 in3 (73 L), towed at a nominal depth of 15 m and fired at 20 s interval to give a nominal shot spacing of 50 m. Shots were recorded on a 48-channel, 2.4-km hydrophone streamer and by deployment of 55 disposable sonobuoys with a typical spacing of ∼15 km. Three east-west lines (CAM83, CAM84 and CAM88) were designed to determine variations in crustal structure perpendicular to the ridge axis. Six closely spaced lines (CAM90–94 and CAM96) were acquired near the active spreading centre and parallel to it, including one extending northward across the propagating rift tip at 8°S [Minshull et al., 1998]. An additional line (CAM86) was acquired over ∼1.1 Ma crust on the western flank of the Ridge (all ages are according to the interpretation of Bruguier et al. ). Crustal arrivals were recorded from 40 of the sonobuoys, and for many of these the close shot spacing allowed relatively weak arrivals to be picked unambiguously to maximum ranges of ∼40 km (Figure 3). The close (∼4 km) spacing of profiles near the ridge axis meant that in some cases shots on one line were recorded on sonobuoys deployed on another, resulting in proper reversal of the wide-angle profiles (Figure 4).
 Shot locations were determined using GPS navigation and sonobuoy positions were determined from direct water wave arrivals and seabed reflections using the method described by Bruguier and Minshull . Sonobuoy data were bandpass filtered to remove wave and waterborne noise, with a low cut at 3 Hz and a high cut between 20 Hz and 10 Hz depending on sonobuoy and range.
3. Wide-Angle Data Analysis
3.1. Travel Time Picking
 Four distinct crustal arrivals were identified (Figure 3): a low velocity second-arrival phase from oceanic layer 2A that emerges tangentially from the seafloor reflection (P1); a high-amplitude refracted phase from layer 2B that is visible to ranges of ∼15 km (P2); a lower-amplitude refracted phase from layer 3 that continues to ranges of 40 km or more (P3); and a high-amplitude wide-angle reflection from the Moho observed on some sonobuoys as a distinct second arrival at 20–40 km range (PmP). A total of 20,755 travel times were picked from the 40 sonobuoy record sections, mainly with a semi-automatic picking algorithm that uses cross-correlation between adjacent traces. Pick uncertainties were estimated using an empirical relationship between uncertainty and signal-to-noise ratio, defined as the ratio of energy in a 250 ms window after the pick to that of a 250 ms window before the pick [Zelt and Forsyth, 1994]. Uncertainties assigned were 25, 35, 50, 75, and 100 ms for signal-to-noise ratios of >10, 4–10, 2–4, 1–2, and <1, respectively.
 Uncertainties in source-receiver ranges come from uncertainties in the water column velocity, estimated at ∼0.3%, uncertainties in the water depth at seafloor reflection points (100–200 m), and uncertainties in the water-wave travel time picks, which were typically 50 ms at long ranges. The combined effect is a range uncertainty that increases systematically with range to 150–200 m at the longest ranges, resulting in an additional travel time uncertainty of ∼33 ms for an arrival at 6 km/s. This source of error was represented in our models as a travel time uncertainty of 1 ms per kilometer of range. The picking and range-related uncertainties were added in quadrature to give a total uncertainty for each pick, which had a mean of ∼90 ms for the P1 arrival, ∼60 ms for P2 and ∼100 ms for P3 and PmP.
3.2. Travel Time Inversion
 Velocity models for each profile were determined using a modified, regularized version of the travel time inversion technique of Zelt and Smith . The method is described in detail by Bruguier  and is summarized only briefly here. Zelt and Smith  use a damped least squares approach in which the objective function ∥Aδm − δt∥2 + λ2 ∥δm∥2 is minimized, where A is the matrix of partial derivatives of travel times with respect to model parameters, δm is the model parameter adjustment vector, δt is the vector of travel time residuals and λ is a constant which controls the trade-off between the model fit and the size of the parameter adjustments. To avoid the strong lateral variations that can result from such an approach, we minimize a modified objective function ∥Aδm − δt∥2 + λ2∥D(m0 + δm)∥2, where m0 is the initial model, D is a discrete difference operator that generates the horizontal derivatives of parameter values, and λ now controls the trade-off between the fit to the data and the model smoothness. The objective function is minimized by a model adjustment δm given by
which can be weighted by data and model covariance matrices Ct and Cm, to give
The smoothing factor λ is initially set at a high value and gradually reduced for successive iterations until the data are fit within their estimated uncertainties, i.e., a normalized χ2 statistic of 1.0 in the terminology of Zelt and Smith .
3.3. Synthetic Seismogram Modeling
 Relative amplitudes of the different arrivals provide constraints on the velocity gradients within layers and the velocity contrasts across boundaries. Synthetic seismograms were calculated using the asymptotic ray theory algorithm of Zelt and Ellis . Poisson's ratios of 0.30 and 0.25 were used for layer 2A and the other crustal layers respectively [Spudich and Orcutt, 1980]. To compensate for an elastic attenuation, the amplitudes were calculated assuming a Q of 25 and 200 for layers 2A and 2B, respectively [Jacobson and Lewis, 1990] and 450 for layer 3 [Spudich and Orcutt, 1980]. Densities were calculated from velocities using the relationship of Christensen and Shaw . In the absence of refracted arrivals from the mantle, the synthetic seismograms provided confirmation that the high-amplitude wide-angle reflection observed at large source-receiver offsets was indeed from the Moho.
3.4. Modeling Strategy
 Differences in travel time through the water column due to off-line sonobuoy drift were corrected using the approach described by Bruguier and Minshull , and in-line drift was dealt with by binning data from each sonobuoy into a series of common receiver gathers spread along the shooting track. In the travel time models, the seafloor was parameterized at a spacing of 1 km by median filtering the higher resolution swath bathymetry, to approximate the diameter at the seafloor of the first Fresnel zone for crustal arrivals. The modeling approach consisted of three steps. First the uppermost crustal structure was determined from the P1 and P2 arrivals. Then the deeper crustal structure was determined from the P2 and P3 arrivals, and finally the Moho depth was determined from the PmP arrivals. At each step, all ten seismic profiles were modeled simultaneously following the joint inversion technique of Zelt , modified to include the regularization as described above. Models were parameterized with velocity nodes at each sonobuoy location and additional nodes at line intersections, to give a horizontal node spacing of 5–20 km. In the joint inversion scheme, parameters considered as common to more than one node (i.e., velocity or depth nodes at intersections) are treated as a single parameter, so that models are forced to be consistent at their intersection points. An initial one-dimensional inversion of the entire data set generated a starting model for the regularized inversions. To further stabilize the inversion, the velocity was required throughout to be a continuous function of depth at the layer 2/3 boundary.
 After each stage of the travel time inversion, amplitudes of synthetic seismograms computed from the resulting model were compared visually with observed amplitudes and where systematic differences were seen on more than one sonobuoy, velocity gradients and/or contrasts across boundaries were adjusted to improve the fit. Reflections from the layer 2/3 boundary were not observed in our data, so the main constraint on the depth of this boundary came from synthetic seismogram modeling of the associated decrease in amplitude. PmP arrivals were detected on nine sonobuoys; for a further seven instruments the Moho depth was constrained by modeling the amplitude increase associated with the Moho triplication. The range and range extent of the PmP amplitude peak depends on the velocity at the based of the crust, on the thickness of the Moho transition zone, and on the velocity of the uppermost mantle. These parameters were not constrained by travel time data, so for simplicity a first-order velocity discontinuity at the Moho and a mantle velocity of 8.0 km/s were assumed. The velocity at the base of the crust was adjusted to match synthetic seismograms to observed data, from a starting model with a vertical velocity gradient of 0.1/s in layer 3. The velocity at the base of the crust might be slightly overestimated, and therefore crustal thickness overestimated by 100–200 m, if mantle velocities are as low as those observed by Hooft et al.  and Canales et al. [2000b] at 35°N on the Mid-Atlantic Ridge.
 Model parameter uncertainties were estimated from a perturbation analysis of profile CAM83, chosen as a representative example, using the approach described by Zelt and Smith . All the velocity and depth nodes corresponding to a particular interface were perturbed until the fit to the data was significantly worse than the best fitting model. This procedure combined with an F test suggested uncertainties of ±0.3 km/s for the layer 2A velocity, ±0.1 km/s for the average velocities at the top of layer 2B and the top of layer 3, and ±0.5 km for the Moho depth. The analysis shows also that the velocity at the base of the crust is very poorly constrained by the data; the main constraint on this value came from synthetic seismograms. The analysis does not take into account trade-offs between parameters, so uncertainties may be underestimated, but nor does it take into account the additional constraints provided by synthetic seismograms.
 In the final model, χ2 values substantially less than 1.0 were reached for some phases on some profiles (Table 1). These low values of χ2 result from our use of a single value of λ for each phase across the entire survey, and indicate that for those phases and profiles a smoother model would have sufficed.
Table 1. Summary of Root-Mean Square and χ2 Errors Between the Observed and Modeled Travel Times for the P2, P3, and PmP Picks
Number of Picks
Number of Picks
3.5. Gravity Modeling
 A detailed analysis of gravity and bathymetry data from our survey area suggested that, in general, mantle density variations contribute little to along-axis gravity anomalies [Bruguier et al., 2003]. However, the same cannot be said of across-axis anomalies, and the complex spreading history of the survey area makes an accurate prediction of mantle density anomalies from thermal models more difficult. Therefore gravity data were used only as a check on seismic models, and no attempt was made to adjust the models to fit gravity data. The gravity effects of off-line bathymetry were removed using the method of Muller et al. . Seismic velocities were converted to densities using the velocity-density relationship of Christensen and Shaw  for the crust and a density of 3.28 Mg/m3 for the mantle, and gravity anomalies were computed for each seismic profile using the method of Talwani et al. .
4.1. Layer 2A
 Seismic reflection data showed that sediment thicknesses are negligible in the survey area, so the low-velocity P1 phase (Figure 4), which was identified on 21 of the sonobuoy records, is interpreted as energy returned from the uppermost igneous crust. Refracted arrivals from the uppermost crust at slow- to intermediate-spreading ridges in general have very low amplitudes [e.g., Cudrak and Clowes, 1993; Smallwood and White, 1998]. The relatively high amplitude of the P1 phase suggests that it is a wide-angle reflection, and synthetic seismograms confirm this interpretation (see below). Hence this arrival places constraints on the thickness and velocity of the near-surface, low-velocity layer 2A, which must be present even where the P1 phase is not seen because the P2 phase is not tangential to the seabed (Figure 4). The velocity and thickness were determined for each of the sonobuoys where the arrival was visible by applying the regularized travel time inversion described above to the P1 phase and to P2 arrivals from the P1/P2 crossover range to 1 km beyond this range. The inversion used an initial layer thickness of 0.5 km and was stabilized by using a fixed velocity gradient of 0.5/s based on the maximum range of the P1 phase, and a fixed velocity of 5.0 km/s for the top of layer 2B. The inclusion of only short-range P2 arrivals means that the inversion is relatively insensitive to the layer 2B velocity. This procedure constrains the properties of layer 2A typically 2–3 km either side of each sonobuoy.
 The result was a mean layer 2A velocity of 2.74 ± 0.24 km/s (1σ) and a mean layer 2A thickness of 0.51 ± 0.12 km. Systematic variations of layer 2A velocity with age, such as those that have been observed elsewhere at slow-spreading ridge axes [Grevemeyer and Weigel, 1996; Hussenoeder et al., 2002a, 2002b] were not detected. However, the age range in our survey area may be too small for such variations to be pronounced enough to be observed. The layer thickness varies between 0.30 and 0.85 km (Figure 5). Taking into account the 0.3 km/s uncertainty in velocity, the uncertainty in these thickness estimates is only ∼100 m, so the observed variations are likely to be real. The layer is thinnest close to the interpreted present-day spreading center at 13°10′W, and thickens on the east flank of the ridge, but in the tectonically complex central part of our survey area, no systematic variation with crustal age could be resolved.
 On the basis of the sonobuoy data, a velocity contrast of ∼2 km/s exists at the base of layer 2A, which is larger than the velocity contrast at the seafloor. Such a large contrast could be expected to generate a strong normal-incidence reflection that would be visible in the multichannel reflection data. However, no such reflection event could be identified in the stacked seismic sections. Common depth point supergathers from the region with the shallowest seabed, however, show a shallow crustal reflection with an apparent velocity significantly higher than the seabed velocity (Figure 6). This arrival is similar to shallow crustal arrivals seen in densely sampled wide-angle data from both the Mid-Atlantic Ridge [Hussenoeder et al., 2002a] and the East Pacific Rise [Vera and Diebold, 1994; Hussenoeder et al., 2002b]. It is interpreted as energy turning within a steep velocity gradient at the base of layer 2A. This reflection appears intermittently on profile CAM96, where the effects of seafloor topography may move the arrival in and out of streamer range [Harding et al., 1993], but could not be identified on other profiles, where the seafloor is generally deeper. Attempts to image the reflection by stacking were unsuccessful due to its close proximity to the seafloor bubble pulse (Figure 6), but travel time and amplitude modeling provided additional constraints on layer 2A structure. One-dimensional reflectivity modeling [Sandmeier and Wenzel, 1986] of the reflection where it is seen most clearly, which probably corresponds to the thinnest layer 2A as well as to the shallowest water depth, suggests a ∼100-m-thick layer 2A underlain by a 200 m transition to layer 2B velocities (Figure 6). Even allowing for the fact that the step in velocity modeled in the sonobuoy data will correspond approximately to the centre of such a transition, layer 2A is thinner at the location modeled than beneath any of the sonobuoys. Hence the multichannel data reveal lateral variability unresolved by modeling the sonobuoy data.
4.2. Cross-Axis Profiles
 In this section we present seismic velocity models for the cross-axis profiles, beginning with line CAM83, which extends from 2.4 Ma crust in the west to 4.2 Ma crust in the east and crosses the ridge axis at 8°56′S (Figure 1). This profile also crosses a rifted high interpreted, on the basis of the coincident magnetic anomaly profile, as the segment center of a spreading axis abandoned at 0.3 Ma [Bruguier et al., 2003]. The crustal structure along line CAM83 is constrained by nine sonobuoys and by crossings with five axis-parallel profiles (Figure 7). Upper crustal structure is well constrained by arrivals on all nine sonobuoys (Figure 8), but only sonobuoys 7 and 8 recorded significant lower crustal arrivals. A good travel time fit was achieved with very little lateral variation in velocity (Figure 8; Table 1).
 The velocity at the top of layer 2B is 5.1 km/s close to the centre of the profile and 5.3–5.4 km/s near the ends. Layer 2 is ∼3 km thick, and the velocity is 6.8 km/s at the top of layer 3, increasing to 7.2 km/s at the base of the crust. Moho reflections are very clear on sonobuoys 7 and 8 (Figures 3 and 9) and suggest an abrupt change in crustal thickness from 9–10 km near the crossing point with CAM90 to 7–8 km beneath sonobuoy 9. Synthetic seismograms from the model matched reasonably well the PmP critical distance of ∼27 km for sonobuoy 7 and ∼24 km for sonobuoy 8 (Figures 3 and 9). The velocity at the top of layer 3 is well constrained by P3 arrivals in this region, so the change in PmP travel times cannot be explained entirely by a reduced layer 3 velocity on axis, though such a reduction may make a small contribution.
 Line CAM84 also crosses both the abandoned axial high and the active spreading center, and spans crust of a similar age range to line CAM83 (Figure 7). The velocity structure is constrained by eight sonobuoys, five of which recorded data beyond 30 km range (Figure 8). The structure of layer 2B is constrained along the whole profile, and the data require an increase in velocity at the top of layer 2B from 4.9–5.0 km/s in the central part of the profile to 5.2–5.4 km/s near the ends (Figure 7). Layer 2 is again ∼3 km thick. The upper part of layer 3 is constrained by a large number of P3 arrivals (Figure 8; Table 1). PmP was only identifiable as a separate arrival on sonobuoy 17 (Figure 10), though amplitude increases at 20–30 km range, which probably correspond to the Moho triplication, were seen on several other profiles (e.g., sonobuoy 12, Figure 4). Travel times and synthetic seismograms for sonobuoy 17 (Figure 10) indicate a crustal thickness of 8 km to the west of this sonobuoy.
 Line CAM88 crosses the entire extinct rift tip from the western to the eastern pseudofault and from 2 Ma crust in the west to 3 Ma crust in the east (Figures 1 and 7). Velocities are constrained by six sonobuoys, although only two show data beyond 20 km range (Figure 8). In contrast to the other two cross-axis profiles, layer 2B velocities on line CAM88 decrease from 5.0 km/s at the abandoned spreading center to 4.6–4.7 km/s near the pseudofaults (Figure 7). Near the ends of the line, the observed travel times for the P2 arrivals are larger than the predicted times and the apparent velocity of this phase is slower than that predicted by the model (Figure 8). This observation suggests that the regularization strength used, which was appropriate for the complete P2 data set, was too high on this profile, and that true lateral velocity variations are larger on this profile than those shown in the final model. The thickness of layer 2, which is constrained by changes in amplitude as well as by limited P3 picks, is ∼2.5 km at the center of the profile, 0.5 km less than on lines CAM83 and 84, and decreases to ∼2.0 km at the ends. High amplitude arrivals at ∼30 km range on sonobuoy 27 (Figure 4) are interpreted as Moho reflections and indicate a crustal thickness of ∼9 km. In contrast, PmP appears earlier and at ranges of 20–25 km on sonobuoy 28 (Figure 8), indicating a crustal thickness of only ∼6 km. This step in the Moho, which may be even steeper than that shown in the smooth final model since the travel time fit of the PmP arrivals on sonobuoy 28 is relatively poor (Figure 8), is approximately colinear with a seabed scarp (Figure 7). This alignment may indicate a tectonic origin for both due to low-angle normal faulting associated with initiation of the new spreading center.
 Gravity anomalies computed from the seismic models (Figure 11) show a good fit to short-wavelength signals in the observed gravity, but there are long-wavelength misfits of 10–20 mGal. These misfits are most likely due to lateral variations in mantle density. For profile CAM83, the residual anomaly matches well the predicted thermal anomaly from a lithospheric cooling model based on a simplified plate geometry centered on the eastern spreading axis [Bruguier et al., 2003], in particular for the western part of the profile. The match is poorer, however, for the other two profiles. The magnitude of the misfit can be explained by density variations due to lithospheric thermal structure, but the nature of those variations is poorly explained by the simplified model used. The lack of a systematic increase in residual gravity west of the current spreading center may indicate that at these latitudes, there is still a residual thermal effect from the abandoned spreading center, and that the spreading center was abandoned later in the north than in the south.
4.3. Axis-Parallel Profiles
 Seven axis-parallel profiles were acquired (Figure 1). Apart from line CAM96, which has been presented elsewhere [Minshull et al., 1998; Bruguier et al., 2003], these lines were all much shorter than the cross-axis profiles and constrained by fewer sonobuoys.
 Line CAM86 was located 17 km west of the abandoned spreading center on 1.1 Ma crust. Velocities are constrained by four sonobuoys (Figures 12 and 13). The velocity at the top of layer 2B decreases from 5.2 km/s in the south, near the segment center, to 4.8 km/s in the north, and layer 2 thins by ∼1 km between sonobuoys 23 and 24, in agreement with results from the cross-axis profiles. PmP arrivals were picked on sonobuoys 22 and 24 and indicate a decrease in crustal thickness from ∼8 km in the south to ∼6 km in the north.
 Lines CAM90–94 form a series of closely spaced profiles on the west flank of the current spreading center, with line CAM90 lying just within the shallow rift valley (Figure 1). If the ridge jump occurred at 0.3 Ma, as interpreted from magnetic data [Bruguier et al., 2003], CAM90 is the only one of the five which samples crust created at the new spreading center. Velocities on line CAM90 are constrained by three sonobuoys, two of which recorded clear arrivals to 25–30 km range (Figure 13). Velocities at the top of layer 2B decrease from 5.2 km/s near the crossing with CAM83 to 4.7 km/s near the pseudofault, and layer 2 thins to the north by ∼1 km. P3 arrivals from sonobuoy 34 were poorly fit by the smooth model from the joint regularized inversion, with a root mean square misfit (RMS) of 129 ms for this phase. The RMS misfit for these arrivals is reduced to 33 ms if layer 3 velocities are not required to match other models at line intersections, and there is only a minor deterioration of the fit to other arrivals (Figure 13). In this preferred model, the velocity at the layer 2/3 boundary varies significantly along the line from 6.7 km/s near the crossing with line CAM83 to 6.3 km/s in the north. Similar lateral velocity variations at the base of layer 3 were introduced to keep the vertical gradient constant and are generally consistent with observed amplitude variations; these variations are not constrained by travel time data. PmP arrivals on sonobuoys 34 and 35 indicate a decrease in crustal thickness from ∼10 km at line the intersection with line CAM83 to ∼8 km at the intersection with line CAM84.
 Line CAM91 samples ∼2.2 Ma crust formed at the abandoned spreading center and is constrained by data from four sonobuoys. Velocities at the top of layer 2B decrease from 5.1 km/s at the segment center to 4.7 km/s in the north and 4.3 km/s in the south. P3 arrivals are consistent with a uniform velocity at the layer 2/3 boundary, and PmP reflections on sonobuoys 36 and 38 sample the same region of the Moho and indicate a crustal thickness of ∼8 km. Lines CAM92–94 have poorer constraints, with only the upper crust well sampled and no identified PmP arrivals (Table 1). In all three lines, a northward decrease of the velocity at the top of layer 3 is observed, from 5.1 km/s at the segment center to 4.7 km/s near the intersections with line CAM88. A similar variation was observed on profile CAM96 [Bruguier et al., 2003], but there the velocities increase again to the north of the eastern pseudofault.
 Gravity anomalies computed from the seismic models fit the observed gravity anomaly well (Figure 14). Residual anomalies attributable to mantle density variations are generally less than 10 mGal, and long-wavelength trends are similar to those predicted by the thermal model.
5.1. Upper Crust
 Oceanic layer 2A has been studied extensively at the East Pacific Rise, where it has been interpreted as a layer of extrusive lava flows [e.g., Vera and Diebold, 1994; Harding et al., 1993; Christeson et al., 1996; Hussenoeder et al., 2002b] and to a lesser degree at the Juan de Fuca Ridge [e.g., Cudrak and Clowes, 1993; McDonald et al., 1994]. However, since layer 2A rarely generates first arrivals in seismic data from sea surface shots, at slow-spreading ridges, where scattering from the seafloor is stronger, arrivals from this layer are usually obscured. In the absence of such arrivals, regularized tomographic inversions generate a smooth velocity-depth profile in which the velocity rises steeply in the upper 1 km of the crust [e.g., Canales et al., 2000a]. The thickness and seismic velocity of layer 2A determined by our study match well those observed at and near the ridge axis at slow- to intermediate-spreading ridges, and a few kilometers off-axis at the East Pacific Rise (Figure 15a). There is a remarkably good agreement of velocities in the upper 1 km with those obtained by Hussenoeder et al. [2002a] from 0–2 Ma crust on the Mid-Atlantic Ridge at 35°N. At 8–9°S, the mean layer 2A thickness is greater and there is more variation in this thickness, but both differences may be simply a consequence of the greater age range and complex spreading history of our study area. The thickness of the transition zone at the base of layer 2A matches better the on-axis model of Hussenoeder et al. [2002a] than their off-axis model; it is unlikely that the small differences between these models could be resolved by our noisier data. For both these Mid-Atlantic Ridge studies, the velocity at the top of layer 2B is considerably higher than those observed at Reykjanes Ridge and at Mohns Ridge (Figure 15a).
 The velocity variations in the upper 1 km of the crust may be interpreted in terms of variations in porosity and pore aspect ratio distribution. For the case of porous basalt, where seismic velocities of the matrix and pore space are very different, Hashin-Shtrikman bounds are too broad to be useful. Narrower and more useful bounds were developed by Berge et al.  using an assumed pore aspect ratio distribution. Application of their approach suggests a porosity of 23–30% at the seabed, decreasing rapidly to ∼12–14% at the top of layer 2B (Figure 15b). This porosity decrease may represent the base of the extrusive lavas, or alternatively a porosity change within the extrusive sequence [e.g., Harding et al., 1993; Cudrak and Clowes, 1993]. The layer 2A thickness matches direct observations of a 280–420 m extrusive layer at the south wall of the Vema transform [Auzende et al., 1989], and the appearance of dykes at ∼500 m depth in old Atlantic crust at Deep Sea Drilling Project holes 417D and 418A [Bryan et al., 1979]. The large variations (±300 m) in the thickness of layer 2A (Figure 5), which are similar to those observed at the Juan de Fuca Ridge [Cudrak and Clowes, 1993; McDonald et al., 1994], may then be explained by a combination of the episodicity of magmatism, the preferential accumulation of extrusive lavas in topographic depressions, and the dismemberment of the crust by extensional tectonics.
 Velocity-depth profiles fall generally within White et al.'s  envelope for 0–7 Ma Atlantic oceanic crust away from fracture zones and mantle plumes, though velocities at the top of layer 2B are slightly higher (Figure 16). The difference may be largely an artifact of modeling approaches which assume a smooth velocity gradient in the uppermost crust. The thickness of layer 2 is generally 2–3 km across the survey area, and is also typical of Atlantic oceanic crust. The velocity structure also resembles that determined for 6–7 Ma crust in the vicinity of Ascension Island [Klingelhoefer et al., 2001].
 The velocity at the top of layer 2B varies systematically across our survey area (Figure 17). First there is a decrease in velocity from south to north, which likely reflects the presence of less altered, higher porosity material to the north, consistent with inference from gravity anomalies that the spreading center may have been abandoned later in the north [Bruguier et al., 2003]. Velocities are lowest near the pseudofaults, where extensive large-scale porosity due to tectonic strain may be expected [e.g., Kleinrock and Hey, 1989]. A similar velocity reduction was observed at the pseudofault marking the southern limit of ridge segment OH-1 at 35°N by Hosford et al. . Secondly, on lines CAM83 and 84, there is a small but significant increase in this velocity with crustal age, at a rate of ∼0.1 km/s per Ma (Figure 18). On line CAM88 this variation with age, if present, is obscured by the velocity decrease toward the pseudofaults. Such variations in layer 2B velocity near ridge axes often can be obscured by wide-angle data analyses which require smooth variations of velocity with depth since large changes in the velocity and thickness of layer 2A are also present [e.g., Carlson, 1998]. For example, at 35°N the mean velocity of layer 2 increases by 0.7–0.8 km/s between zero-age and 2 Ma crust, but this increase may be largely due to changes in layer 2A velocity [Hosford et al., 2001]. A similar, but more rapid increase with age has been observed at Reykjanes Ridge [Smallwood and White, 1998], while no such systematic variation is apparent at the East Pacific Rise [Grevemeyer et al., 1998].
5.2. Lower Crust
 Layer 3 velocities are remarkably uniform across the survey area, with good fits achieved to both travel times and amplitudes using a constant velocity on all lines except for the northern end of line CAM90. A similar uniformity of layer 3 velocities has been observed in 0–2 Ma crust at 35°N on the Mid-Atlantic Ridge [Hosford et al., 2001]. As elsewhere on the Mid-Atlantic Ridge and Reykjanes Ridge [e.g., Hooft et al., 2000; Smallwood and White, 1998], wide-angle Moho reflections are observed from beneath the current spreading axis (Figures 8 and 13).
 Reduced velocities at the top of layer 3 in the northern part of line CAM90 (Figure 12) are not detected on either the adjacent profiles or the crossing profiles, so must be limited to a region only a few kilometers wide around the active spreading center. Velocities are reduced by up to ∼0.4 km/s, a change that is too large to be accounted for by plausible changes in porosity. The velocity reduction may result from a thermal anomaly due to a recent intrusive event, or from the presence of a small amount of partial melt. If the P wave velocity of layer 3 gabbros varies with temperature at a rate ∂V/∂T = −0.57 × 10−3 km s−1 K−1 [Christensen, 1979], the maximum velocity anomaly corresponds to a temperature anomaly of around 700 K. If the off-axis temperature at the top of layer 3 is ∼200–300°C. as suggested by thermal models incorporating hydrothermal cooling [e.g., Henstock et al., 1993], no partial melt is required to explain the anomaly, though the presence of partial melt could contribute to it.
 The crustal thickness increases systematically from north to south (Figure 19), toward the center of the abandoned spreading segment, the southern half of which remains active [Bruguier et al., 2003]. The total variation is ∼4 km, from 6 km to 10 km. The magnitude of the variation is similar to that seen along the magmatically robust OH-1 segment at 35°N [Hooft et al., 2000], but the mean crustal thickness is about 2 km greater due to the presence of a significant mantle melting anomaly at 9–10°S [Bruguier et al., 2003]. The layer 2 thickness also increases southward, from 2.0–2.5 km on line CAM88 to ∼3 km on lines CAM83 and CAM84, but as elsewhere [Mutter and Mutter, 1993], changes in crustal thickness appear to be taken up dominantly by changes in layer 3 thickness. The southward increase in crustal thickness coincides with increasing enrichment in incompatible trace elements of basalts dredged at the ridge axis [Schilling et al., 1985]. We interpret the thickness variations as due to the presence of a more fertile mantle source to the south of our survey area [Bruguier et al., 2003], as has been suggested for the increased crustal thickness of the OH-1 segment [Niu et al., 2001].
 If a thick, hot mantle plume head were present to the south of our survey area, we would expect to see a greater southward increase in crustal thickness derived from mantle Bouguer gravity anomalies that inferred from seismic data. No such trend is inferred (Figure 19b). However, there is a large and well constrained discrepancy in crustal thickness around the abandoned spreading center on lines CAM84 and CAM88, where the gravity inversion predicts a crust 2–3 km thicker than suggested by the seismic models. A difference of 1–2 km is also present around the segment center of the active spreading center. These differences are unlikely to be due to changes in mean crustal density since only small variations in mean crustal density are expected for these crustal thicknesses [Minshull, 1996], and a large change in density would be expected to be accompanied by a corresponding change in seismic velocity. Therefore we attribute them to reduced mantle densities and higher temperatures than those predicted by the thermal model, which neglects the effects of ridge jumps and ridge propagation. Although the ridge jump likely occurred at depth before its surface expression [Bruguier et al., 2003], hot, low-density asthenospheric mantle may remain at relatively shallow depths beneath the recently abandoned spreading center, particularly to the north. Hot asthenosphere may be present also beneath the active segment center.
 The thinnest crust is observed beneath the pseudofault trough at the northern end of line CAM86 (Figure 12), beneath the new pseudofault at the northern end of line CAM91, and beneath the propagating rift tip at the northern end of line CAM96 [Bruguier et al., 2003]. A reduced crustal thickness in these locations might be expected due to the juxtaposition of the spreading center against older lithosphere, but the age discontinuity involved is in all cases relatively small (generally <1 myr). The reduced thickness is therefore analogous to that which occurs at small-offset fracture zones on slow-spreading ridges [e.g., Minshull et al., 1991; Tolstoy et al., 1993], and may be due to focused magmatic accretion at the segment center. Our observations contrast with those of Hasselgran et al. , who infer from seismic reflection data the presence of thickened crust beneath pseudofaults with similar age differences on the flanks of the Juan de Fuca ridge.
 The inferred termination points of the pseudofaults (Figure 1) suggest that the abandoned spreading center began propagating at ∼2.5 Ma. This age coincides with the age of the crust where the Moho steps down to the west on line CAM83. Perhaps, then, the onset of excess melt production caused ridge propagation to initiate as a result of excess gravitational-spreading stresses [Phipps Morgan and Parmentier, 1985]. However, it is clear from the northward decrease in crustal thickness that such stresses were not required to maintain ridge propagation once it had been initiated.
 Our crustal thickness measurements also allow an examination of the relationship between crustal thickness and axial morphology at constant spreading rate. In the south of our survey area, where seismic data indicate that the axial crust is 10 km thick and gravity data indicate a similar thickness beneath the abandoned spreading center, both the active and the abandoned spreading centers have rifted axial highs (Figure 20). Where the crustal thickness is reduced to 7 km, there is a deep axial valley. However, at intermediate thicknesses of 8–8.5 km, the morphology varies widely. Where line CAM84 crosses the abandoned axis, there is a clear axial high, though it is broad and subdued in amplitude (Figures 1 and 20). An axial high is also present at 33°S on the Mid-Atlantic Ridge, where the spreading rate and crustal thickness are similar [Tolstoy et al., 1993]. Conversely, where this line crosses the new axis, and the crust is as thick or perhaps slightly thicker (Figures 7 and 19), there is a clear axial valley (Figures 1 and 20). The presence of an abandoned axial high on line CAM84 favors a flexural origin for the high such as that suggested by Buck . The presence of an axial valley at the new spreading centre may also be consistent with such an origin since the release of flexural stresses generated by the abandoned axial high will favor the formation of a valley until sufficient flexural stress has been accumulated from new accretion.
 Travel time and synthetic seismogram modeling of wide-angle seismic data from sononbuoys and a multichannel streamer deployed on 0–4 Ma crust on the Mid-Atlantic Ridge at 8–9°S lead to the following conclusions:
 1. Oceanic layer 2A is 300–850 m thick and is underlain by a ∼100 m transition to layer 2B. The layer thickness matches estimates of the extrusive layer thickness in Atlantic oceanic crust from drilling and submersible observations.
 2. Velocities at the top of layer 2B increase with age at ∼0.1 km/s per Ma.
 3. Anomalously low layer 3 velocities detected by one sonobuoy at the current spreading center may indicate the presence of hot or partially molten rock. Elsewhere, no lateral variation in layer 3 velocities is resolved by our data.
 4. A spreading center abandoned at ∼0.3 Ma is characterized by an axial high and reduced mantle densities; both the mantle density and the layer 2B velocity are lower to the north, where we infer that the spreading center was abandoned later. Otherwise, little variation in mantle density is required beyond that due to lithospheric cooling with age. Specifically, increased crustal thicknesses to the south are not associated with a mantle density decrease that has a detectable gravity signature, and no mantle plume is required.
 5. Pseudofaults due to two episodes of ridge propagation are marked by lower layer 2B velocities, indicating greater tectonic strain, and thinner crust, indicating a reduced melt supply. Pseudofault propagation may have been initiated by a pulse of increased magmatism, but increased magmatism was not required to sustain it.
 6. At a spreading rate of 32–33 mm/year the transition from axial valley to axial high occurs at a crustal thickness of ∼8 km, but the equilibrium morphology may take >0.3 myr to establish itself after a ridge jump.
 Data acquisition was funded by the Office of Naval Research. We thank the officers, crew and scientific party of RRS James Clark Ross cruise 5 for their assistance, and A. Hosford and J. P. Canales for thorough reviews. NJB was supported by a NERC research studentship, TAM by a Royal Society University Research Fellowship, and JMB under Office of Naval Research 6.1 funding (Program Element 61153N). Figures were generated using the GMT package of Wessel and Smith .