We describe a geophysical study of oceanic core complexes (OCC) and surrounding seafloor on the Mid-Atlantic Ridge at 13°N–14°N and off-axis to ∼1.9 Myr. Data include a detailed, deep-towed side scan sonar, magnetic field and bathymetry survey, supplemented by concurrent sea-surface bathymetry, magnetic field and gravity measurements. Using side scan and bathymetry, we infer areas and relative ages of seafloor volcanism, revealing a complex pattern of melt accretion across the median valley including close to its walls. We estimate tectonic and magmatic extension throughout the area, and find that average tectonic extension since chron 2 on plates containing OCCs is up to three times that on their conjugates. Deep-towed magnetic data reveal asymmetric spreading (faster on OCC-containing plates) and crustal magnetization that is highly heterogeneous on a scale of ∼5 km, suggesting that exhumed domes of OCCs have highly variable lithologies, perhaps comprising both serpentinized peridotite and gabbro. Improved fits to magnetic data are provided by models incorporating ∼45°of OCC footwall rotation. An axial zone of normal magnetization, of presumed Brunhes epoch, has highly variable width and amplitude, with parts of the ridge axis displaying very low or apparently reversed magnetization. Gravity requires that OCCs have dense cores capped by lower density zones several kilometers thick. Gravity data indicate longer term patterns of crustal thickness and melt distribution that are broadly consistent with numerical models of OCC formation and show that waxing magmatism may terminate OCCs.
 Mid-ocean ridges display significantly varying morphology, and generate oceanic lithosphere in a variety of distinct processes, as a function of varying spreading rate and mantle temperature [Cannat et al., 2006; Macdonald, 1982]. Much of the slow-spreading Mid-Atlantic Ridge (MAR) displays a well-defined axial valley and well-lineated abyssal hill terrain, with crust some 6–8 km thick and well-defined Vine-Matthews-Morley magnetic lineations formed by magmatic accretion at the ridge axis. We refer to this as ‘lineated’ terrain. Elsewhere the axial valley and abyssal hills are less distinct, crust is thinner, and abyssal hills are shorter, blockier and more widely spaced. We refer to this as ‘blocky’ terrain. Lineated and blocky terrain are often interpreted as reflecting magma-rich and magma-poor accretion, respectively.
 Oceanic Core Complexes (OCC) are often associated with blocky terrain and are thought to be characteristic of limited magmatism [Cann et al., 1997; Escartín and Canales, 2011; Karson and Dick, 1983; Tucholke et al., 1998]. They are topographic massifs several kilometers across whose domed surfaces display morphological corrugations aligned along the spreading direction. These surfaces are interpreted as the outcrops of detachment faults which, while active, may take up a significant proportion of relative plate motion. Detachment faults are associated with asymmetric plate accretion [Okino et al., 2004] and have been suggested to occur along some 50% of the MAR [Escartín et al., 2008]. Numerical modeling suggests OCCs may form when approximately 50% of plate separation is taken up by magmatic accretion, compared to 80–90% for lineated terrain [Buck et al., 2005; Tucholke et al., 2008].
Smith et al. [2006, 2008] described the ridge south of 14N, which they called the ‘13°N segment’ (hereafter ‘13N’) (Figure 1). It is mostly characterized by blocky terrain with extensive seismicity, and contains numerous OCCs, mostly on the western flank. Two of these appear to intersect the ridge axis and are thought to be actively extending (OCC1320 and OCC1330, Figure 1). (Following MacLeod et al. , we name OCCs according to their latitude in degrees and minutes; the13°20′ OCC of Smith et al. [2006, 2008] was called OCC1319 by MacLeod et al. but is hereafter called OCC1320). The area lies close to the ill-defined North America (NA)-South America (SA)-Africa (AF) triple junction [Roest and Collette, 1986]. In 2007, as part of a broader study, we carried out a high-resolution geophysical study of the 13N segment, and report that survey here.
2. Data Acquisition
 Our survey extends approximately 70 km southwards from 13°50′N (near the southern limit of the 14N segment; Figure 1) and incorporates several OCCs. Our primary tool was the TOBI deep-tow, carrying a 30 kHz side scan sonar, three-component fluxgate magnetometer, echosounder and depth recorder [Flewellen et al., 1993]. TOBI was towed along E-W track lines, mostly 6 km apart (Figure 1). Details of processing the TOBI data are given in auxiliary material. The side scan data are shown in Figure 2, TOBI-derived high-resolution seafloor gradients inFigure S1, and TOBI total magnetic field in Figure S2.
 While towing TOBI we measured multibeam bathymetry (Figure 1) using a hull-mounted Simrad EM120. We measured sea-surface gravity and reduced it to the free-air anomaly (FAA,Figure 3a) with an estimated total error of 2 mGal. Sea-surface total magnetic field was measured with a proton precession magnetometer. Details are given inauxiliary material.
3. Seafloor Geology
 We analyzed seafloor morphology using ship-based multibeam depth data displayed as gray scale and color-coded shaded relief maps, with and without superimposed contours, together with seafloor gradient maps (e.g.,Figure S3), and profiles (Figure S1). We interpreted the side scan data in map view and draped over bathymetry, using visual appearance, texture, and quantitative backscatter levels to distinguish volcanic and tectonic features similar to those we have used elsewhere [Searle et al., 2010]. These are listed in Table S1 and illustrated in Figures S5 and S6. We combined all these data sets to make our final interpretations. For example, fault scarps were identified from a combination of side scan texture, bathymetry, slope maps, and high-resolution gradient.Figure 4 presents a geological map interpreted from these acoustic terrains.
3.2. Regional Morphology
 Our study area lies within an ∼100 km wide region of blocky terrain between 13°50′N and 12°50′N (Figure 1). The ridge axis generally trends N-S, offset by a 20-km-long second-order non-transform offset (NTO) between 13°35′N and 13°44′N. The V-shaped boundary between the 14N lineated and the 13N blocky segment lies at the northern end of the NTO. From the azimuth of this boundary we calculate that the 14N magmatic segment has expanded southwards at ∼15 km Ma−1 for the last 1.8 Ma.
 Two small, isolated exposures of lineated terrain associated with numerous small, closely spaced faults lie within the blocky region (A and B, Figures 1 and 4). Region A parallels the southern boundary of the 14N segment, and is bounded to north and south by bathymetric lows trending NW-SE (blue lines inFigures 1 and 4).
 We recognize a neovolcanic zone (NVZ) containing the most recent seafloor volcanism and characterized by the brightest sonar backscatter levels (terrain V1, Table S1 and Figure S5). A similar, slightly older, terrain is V1a. These terrains are composed of thousands of small volcanic hummocks [Yeo et al., 2012], often aligned in rows up to 8 km long (V5, V6). Progressively older versions of this terrain, with lower backscatter and inferred greater sediment cover, are mapped as V2 and V2a. Hummocky seafloor constitutes approximately 90% of the visible volcanic seafloor, similar to the MAR between 27°N and 30°N [Briais et al., 2000].
 Several large, volcanically constructed ridges occupy the axial valley (e.g., C, Figures 1 and 4). These ridges comprise numerous, coalesced volcanic cones as observed elsewhere on the MAR [Karson et al., 1987; Lawson et al., 1996; Searle et al., 2010; Smith and Cann, 1990] and identified as axial volcanic ridges (AVRs) [Parson et al., 1993]. However, at 13N they are generally sediment covered and thus relatively old compared with the brighter NVZ, which contains mostly minor volcanic lineations. The large ridges may be mature AVRs, now spread off-axis [MacLeod et al., 2009]. These ridges lie to the east of the current NVZ, implying a relatively recent shift of volcanic activity ∼6 km westward in the region of the two active OCCs, which may be related to their development as we discuss later. Between OCC1320 and OCC1330, the highest acoustic backscatter occurs not on the major volcanic ridge (Figure 1, C), nor even on the minor volcanic lineaments at 44° 52.5′W (Figure 1, D), but at the foot of the E-facing normal fault scarp at 44° 53.5′W (Figure 1, E).
 The robust NVZ near inactive OCC1348 (Figures 1 and 4, F) contains no major ridge, but comprises many minor volcanic lineaments spanning the axial valley floor. This suggests either that volcanism here produces low-relief flows, that melt emplacement has rapidly switched back and forth across the axial valley so there is no preferential zone of volcanic construction, or that melt emplacement at the MAR axis has recently resumed after a prolonged hiatus with insufficient time to form a full AVR.
 There are relatively minor occurrences in the NVZ and its flanks of terrains exhibiting a more uniform backscatter with superimposed hackly texture, labeled V3, V3a and V4. Such terrains are usually associated with relatively smooth seafloor, though not necessarily built from sheet flows [Cann and Smith, 2005; Searle et al., 2010]. As elsewhere on the MAR, there are significant numbers of flat-topped seamounts (V7).
 The width of both the NVZ and other volcanic terrains varies systematically along axis. The detailed distribution of these components along an axial section is given in Figure S7. Because we have used backscatter level as an additional discriminant in this study, our measures of NVZ width are more robust and precise than those of MacLeod et al. , who used a qualitative visual comparison. Although details differ, we confirm that the NVZ is widest between active OCCs and narrows or disappears adjacent to them. The average NVZ width is 2.5 km, with a maximum of 6 km at 13°22′N (immediately north of OCC1320), and vanishing entirely adjacent to OCC1320 and OCC 1330.
 There are gaps of 2 km and 4 km respectively in V1 and the slightly older V1a adjacent to OCC1320, and gaps of 10 km and 7 km at OCC1330. These gaps contain the oldest volcanic terrain V2a and additionally, at OCC1330, smooth terrain V3 and V3a. The narrowing exposures of V1towards the OCCs imply propagation along the MAR axis at rates of ∼13 km Ma−1 (Figure S8). Minor volcanic lineaments (V6) swing from roughly N or S to NW and SW, toward the OCCs, as they approach them from south and north respectively. In contrast, there is a continuous, 6 km wide NVZ opposite the inactive OCC1348 comprising V1and V3.
 We estimate the age of the neo-volcanic hiatuses by noting that the adjacent axial seafloor has texture V2a, with average acoustic backscatter level 370 (Table S1). Assuming that the sonar signal penetrates a maximum of 5 m [Lawson et al., 1996], that the mean backscatter level over unsedimented basalt and thick sediment is 1270 and 100 units, respectively (Table S1), that intensity falls off linearly with increasing sediment cover, and that the sedimentation rate is 5 m Ma−1 [Mitchell et al., 1998], we obtain a maximum age of 0.77 Ma for V2a. If the sedimentation rate were 10 m Ma−1 and the sonar penetration only 2.5 m, the age of V2a would be 0.19 Ma. Thus, the hiatus has lasted at least several hundred thousand years. If OCC1320 and OCC1330 have taken up the full plate separation by slip on their detachments, their widths imply minimum ages of 0.38 Ma and 0.41 Ma, respectively; if they take up only half the plate separation by tectonic slip, they have maximum ages of 0.77 Ma and 0.82 Ma. Since these estimated ages of OCCs and volcanic hiatuses overlap, we are unable to say whether waning axial volcanism preceded (perhaps causing) OCC initiation or followed it.
3.4.1. Oceanic Core Complexes
Smith et al.  identify 24 OCCs in the 13N segment. Seven of these are within our study area, including the active OCC1320 and OCC1330 and inactive OCC1348. We have identified a further four inactive OCCs around (13°25′N, 45°08′W) using our side scan sonar imagery and higher resolution bathymetry data (orange shading, Figure 4). They have faintly corrugated surfaces and are mostly smaller, with surface areas <30 km2, than those previously documented (average 65 km2). OCC1326 and OCC1330a exhibit distinct, hooked breakaway ridges. All four are part of a group of OCCs with a distinct NW-SE-trending northern boundary, sub-parallel to the oblique bathymetric valley discussed above (G,Figures 1 and 4). This strengthens our view that this is a migrating boundary between magma-rich and magma-poor regions.
 In addition to the brief descriptions given by MacLeod et al. of the three near-axis OCCs, we make the following observations:
 Bathymetric corrugations on the smooth dome have mean wavelength of 470 m, amplitude 25 m, and strike of 271° ± 1°. The striations visible on side scan sonar have mean wavelength <50 m and strike 270° ± 5°. Both strikes fit the 273° ± 2° relative plate motion predicted by the NUVEL-1a plate velocity model for SA-AF [DeMets et al., 1994]. Striations in the high, predominantly mafic, blocky massif west of 44° 56′W [see MacLeod et al., 2009] are rotated 11° clockwise in the south and 8° counterclockwise in the north.
 The breakaway zone comprises two major, steeply back-tilted fault blocks (H, J,Figures 1 and 4) with a joint E-W extent of 2.3 km at 13°20′N. Side scan sonar shows they have a hummocky texture suggesting volcanic composition. Ridge J is bisected by two E-W offsets in line with the northern and southern edges of the OCC dome, forming three segments with a cumulative length of 15.9 km, average trend 342° and dip of 35°–42°W. This is joined at 13°21.4′N by ridge H, which is continuous along axis for 20.4 km, trends 356°and dips 35°–40°W. If H marks the breakaway, J may be the crest of a rider-block derived from the hanging wall [Smith et al., 2008]; if J may mark the breakaway, H would simply be a normal fault block that pre-dated OCC formation and has been back-tilted during OCC formation.
 OCC1330 has a rounded plan, whereas OCC1320 is triangular. There is no distinct topographic boundary between the upper massif and the breakaway region. The ‘upper massif’ lies in a topographic depression several hundred meters shallower than the summit of the smooth dome, and has a pronounced, highly irregular and mottled backscatter pattern suggestive of prolonged tectonic deformation. Although the striated surfaces of OCC1320 and OCC1330 have similar mean backscatter intensities (480 and 455), there is less variation at OCC1330, indicative of thicker sediment cover. OCC1330's dome is cut by several minor, N-S trending faults that post-date the striations. There is no clear breakaway ridge and, if the continuous detachment model ofSmith et al.  applies, this may mean there are no rider blocks at OCC1330. Along the central axis of OCC1330, the dome changes gradient at 4.7° km−1 and dips 14°E at the footwall/hanging wall boundary, similar to OCC1320 (Figure S1). Striations on the domal section are less well-defined than at OCC1320; bathymetric corrugations have a mean wavelength of 615 m and amplitude of 32 m, slightly larger than at OCC1320. The mean azimuths of corrugations and striations are 272° ± 2° and 271° ± 6°, respectively, consistent with OCC1320 and the NUVEL-1a prediction for SA-AF.
 OCC1348 is the oldest and largest of the near-axis core complexes, and is almost entirely obscured by sediments on the side scan imagery. The domal section has a less regular slope than the others, and is bisected by several ridge-parallel faults. Neither a prominent, 300 m offset, N-S-trending fault at 44°56′W (K,Figure 4), nor the smaller fault 1′ farther west, can be the termination, since corrugated and striated seafloor occurs west of them. The upper massif has an irregular, elevated topography more like OCC1320 than OCC1330. The eastern boundary is marked by a single, clearly defined ridge at 44°51.8′W that is continuous for 20 km N-S and whose eastern side dips 38°E. The outward-facing part of this ridge is volcanic [MacLeod et al., 2009]: either a ‘breakaway ridge’ or part of a rider-block. Corrugations across the domal surface trend 281° ± 1°, 10° clockwise of the mean for the younger OCCs to the south, and consistent with the 282° ± 1° predicted for NA-AF by NUVEL-1a [DeMets et al., 1994].
3.4.2. Tectonic Strain
MacLeod et al. estimated tectonic strain by summing fault heaves measured along twelve of our deep-tow profiles. Here we provide an improved estimate by measuring all faults in the study area as imaged by side scan sonar and mapped inFigure 4. Total fault area divided by area of survey gives the tectonic extension, assuming extension is entirely ridge-normal. This is a minimum estimate as it ignores extension on unidentified faults, though we suspect that to be minimal. To assess asymmetry, we divide the area into four quadrants (Figure 1). Results are given in Table 1. Because OCCs may develop from ordinary normal faults [MacLeod et al., 2009], the distinction between large normal faults and small OCCs is not always clear. However, the vast majority of faults have areas ≤7 km2 (Figure 5), and for the purpose of Table 1 we arbitrarily define OCCs as having areas >15 km2.
Quadrants are defined in Figure 1, and are bounded by the MAR axis, magnetic Chron2, the NTO and the north and south limits of the TOBI survey.
Along-axis length of quadrant, km
Area of quadrant, km2
Faulted surface area, km2
E-W tectonic strain since chron 2, %
E-W strain asymmetry (100(W-E)/W+E),%
Number of inward-facing faults <15 km2
Number of OCC > 15 km2
Number of outward-facing faults
Fault density including OCC, km2/fault
Average inward-facing fault area, km2
SD of inward-facing fault area, km2
Average OCC area, km2
Standard deviation OCC area, km2
Average fault area including OCC, km2
Standard deviation all fault areas, km2
 Most faults are inward facing, as elsewhere on the MAR [Escartín et al., 1999]. Of the few outward-facing faults that we identified, most are small and grouped on the outer slopes of large normal faults (e.g., L,Figures 1 and 4); they are rare near OCCs. Small faults with uncertain dip were arbitrarily classified as inward-facing.
 Faulting is distinctly asymmetric. Ten of the eleven OCCs in our study area are on the western ridge flank, though the presence of the single, large OCC1348 on the east flank results in only an approximately 3:1 asymmetry in OCC fault area. In the SE quadrant tectonic strain is partitioned entirely onto 51 small inward-facing and 6 small outward-facing faults, whereas in the SW quadrant tectonic extension is taken up primarily on 6 large detachments and only 16 small faults. Average fault density shows a similar pattern, with approximately one fault per 50 km2 in the SW compared with one per 13 km2in the SE. The NW-quadrant is similar to the SE-quadrant, neither containing OCCs. The NE-quadrant contains just one OCC.
 Average tectonic strain (including exposed detachments) since magnetic chron 2 (section 5.2.1) is 31%, 10%, 15%, and 24% in the SW, SE, NW and NE quadrants, respectively. Slight differences from the estimates of MacLeod et al.  are due to our improved methodology, measuring all faults in the area, compared with MacLeod et al.'s restriction to faults under the survey tracks. Even in quadrants lacking OCCs, the strain is somewhat higher (15% and 10%) than the average strain of 10% found at MAR 29°N, although the ∼2:1 average strain asymmetry between OCC-containing quadrants and their conjugates is similar to that between inside and outside corners there [Escartín et al., 1999].
 The proportion of plate separation taken up by magmatic accretion, M, is the complement of tectonic strain, T (i.e., M = 1 − T) [Buck et al., 2005; MacLeod et al., 2009]. Using our estimates of T, we find average M since chron 2 is about 85% and 75% in the NW and NE, compared to 70% and 90% in the SW and SE. Thus total M for both plates is the same (80%) throughout our study area, and is lowest (75% and 70%) in quadrants containing OCCs.
4. Gravity and Crustal Structure
 The mantle Bouguer anomaly (MBA) and residual mantle Bouguer anomaly (RMBA, corrected for thermal effects) were calculated using standard methods [Kuo and Forsyth, 1988] to compute the gravitational effect of several interfaces parallel to the seafloor. Details are given in auxiliary material.
4.1. Regional RMBA and Crustal Thickness
 The RMBA (Figure 3b) is generally lower in the north than the south and includes an axial low reflecting low density material associated with and propagating southwards from the magmatically robust 14N segment. There is a strong, linear, southwards increase of about 2.2 mGal km−1in average RMBA across our 13 survey profiles. The northern boundary of the southwestern RMBA high runs NW-SE, matching the oblique trend of OCCs and bathymetric depression G discussed above, and provides further support for a propagating magmatic region. The RMBA minimum in the SE quadrant coincides with the localized zone of abyssal hills B, also supporting the idea of a spatially and temporally localized and asymmetric episode of robust magmatism there.
 The most striking feature of the RMBA is the positive anomaly across the SW part of the survey area, which at its maximum is over 19 mGal greater than at the conjugate position in the SE. We carefully checked the validity of this anomaly. Though hard to see in the FAA and bathymetry, there is a subtle NE-SW bathymetric asymmetry that is revealed by low-pass filtering and appears in the seafloor, intracrustal and Moho interface corrections. The asymmetry also appears in the MBA and RMBA ofSmith et al. , giving us confidence in its reality. This regional RMBA asymmetry also matches the asymmetric pattern of faulting and tectonic strain. Cross-axis RMBA asymmetry is less pronounced north of the NTO, matching the lower degree of asymmetry in tectonic strain there.
 We computed regional variations in crustal thickness by assuming the RMBA arises entirely from variations in depth to a Moho interface of density contrast 600 kg m−3, equivalent to taking crustal density as 2700 kg m−3 [Fujiwara et al., 2003] (Figure 3c). Details are given in auxiliary material. Crustal thickness thus defined comprises all material of density 2700 kg m−3, which may include serpentinized peridotite as well as basalt and gabbro; this analysis may thus overestimate the thickness of magmatic crust. In particular, values of relative crustal thickness >−6 km over OCCs in Figure 3c do not necessarily imply that some (or any) magmatic crust is present in their footwalls. Seafloor sampling here has recovered mainly serpentinized peridotite [Beltenev et al., 2007; MacLeod et al., 2009].
 Along the MAR axis, where we expect the crust is mostly basalt and gabbro so that gravity-derived thickness fairly accurately reflects the magmatic crustal thickness, there is a thickening toward segment 14N of ∼0.3 km. In the SW survey quadrant there is apparent crustal thinning of 1.2 km between the MAR axis at 13°25′N and the thickness minimum in the SW, consistent with thinning of ‘at least 1 km’ for the same area found bySmith et al. . This broad thickness anomaly could be effectively averaging a number of even shorter-wavelength, higher amplitude anomalies from the numerous individual OCCs in the area.
4.2. Inferred Component of Magmatic Extension
 Crustal thickness has previously been used to infer the relative variation in melt supply to mid-ocean ridges, with thicker crust occurring at magmatically robust segment centers and thinner crust in magma-poor regions such as segment ends [Lin et al., 1990; Tucholke et al., 1997]. Thus the gradual decrease in crustal thickness from north to south over the 13°N region can be interpreted as a decrease in melt supply to the ridge axis. At 14N the crustal thickness has been estimated to be ∼7 km [Fujiwara et al., 2003] to ∼8.5 km [Smith et al., 2008]. Assuming this thickness represents the ‘normal’ amount of magmatic accretion for a magmatically robust environment with M = 95% [Buck et al., 2005] or M = 90% [Escartín et al., 1999], we can scale our gravity estimates of crustal thickness to the corresponding component of magmatic accretion:
where MT is the inferred percentage magmatic accretion at any point, RCTis the gravity-derived residual crustal thickness at that point,ACTRMBA is the thickness of crust assumed for the RMBA calculation (i.e., 6 km), ACTSC is the assumed thickness of crust and MSCis the assumed melt supply at the magma-rich segment center. TakingACTSC = 8 km and MSC = 95%, we obtain a variation in MT across our survey area from approximately 56% in the SW quadrant to 75% near the northern MAR axis (Figure 3c). The inferred component of magmatic extension at the MAR axis, where the crust is unlikely to comprise much serpentinised mantle material, ranges from ∼75% north of 13°45′N to ∼70% within the localized, magmatically robust zone between OCC1320 and OCC1330 (E, Figure 4). These are comparable to the estimates made from fault exposures: ∼80% averaged across the entire survey area (section 3.4.2), or ∼60% for single profiles through the less magmatically robust parts of the survey area [MacLeod et al., 2009].
4.3. OCC Structure
4.3.1. The 2.5D Models
 ‘2.5D’ models consist of polygons of which the shape, density contrast and length perpendicular to the plane of the profile are prescribed [Pedley, 1991]. We gave all polygons a half-length (perpendicular to profile) of ±10 km, and extended the polygons at either end of the models to ±1000 km along the profile to minimize edge effects.
Figure 6ashows RMBA profiles with long-wavelength regional gradient removed across OCC1320 and OCC1348. OCC1330 is similar, and is illustrated inFigure S9. Each OCC is associated with a positive RMBA, confirming that they are underlain by dense material. The RMBAs are a few kilometers wider than the topographic expressions of each OCC and are generally centered over the OCC summits. Figure 6bshows 2.5D models in which a 2-layer crust overlying a uniform mantle has been upwarped along a detachment fault, exposing mantle in the smooth OCC domes. These give a poor fit (dotted and dot-dashed lines inFigure 6a). Figure 6cshows models in which a 3–5 km-thick low-density zone (LDZ) of 2900 kg m−3 is included in the detachment footwall. These provide excellent fits for OCC1320 and OCC1330, but OCC1348 requires an additional 2 km thinning of the crust beneath the breakaway ridge and a further reduction in density of the LDZ just below the Moho outcrop (Figure 6c). The density and thickness of the LDZ can be changed by ±100 kg m−3and ∓1 km without exceeding the ±2 mGal error in the gravity data, as there is a trade-off between LDZ thickness and density. Given that the LDZ can be modeled with a density similar to that of the lower crust (2900 kg m−3), the Moho position is insensitive to gravity.
4.3.2. Linked Detachment Surfaces
Smith et al.  and Schouten et al. propose a model for the 13°N region in which multiple OCCs may be linked along- and across-axis. This model suggests that the group of ridges around point H inFigures 1 and 4, which we have interpreted as the breakaway to OCC1320, are actually rider-blocks formed by high-angle faults cutting through the hanging wall and rooting into the detachment (Figure 7b, −9 km to −15 km). Similar rider-blocks have been proposed at Atlantis Massif [Blackman et al., 1998]. Figure 7 compares such a model with one comprising multiple detachments. Figure 7b models the subsurface density structure for a single, continuous detachment with the breakaway to OCC1320 coinciding with the ridge at 45°04.1′W (−22 km). The variable thickness LDZ has a similar effect to introducing a horizontal density gradient in which the older part of the footwall is less dense than the younger domal part; representing increased weathering, more pervasive serpentinization and/or temporal variations in melt capture. Such a model is indistinguishable using gravity from the multiple detachment model (Figure 7c).
4.3.3. 3D Gravity Models
 Although 2.5D modeling is suitable to address some problems, the detailed density structures of OCCs are fully three-dimensional. We used a novel method to represent the expected 3D mass distribution beneath OCCs and thus to further constrain the density of the footwall and examine the subsurface shape of the footwall/hanging wall boundary. The method is a slight modification of that ofBlackman et al. , in which the bathymetry grid is manipulated to produce density interfaces that parallel the seafloor in the far field but are upwarped under OCC domes. The gravity anomaly associated with these interfaces is calculated using the method of Parker . Details are given in the auxiliary material.
 The RMBA assumes a Moho interface 6 km below seafloor, which is inappropriate for OCCs. We therefore model the FAA directly. We correct the FAA for the thermal anomaly as described above, and remove a low-order polynomial regional trend (seeauxiliary material). We then calculate the combined attraction of three interfaces-the seafloor, an intracrustal (upper/lower crust) layer, and the Moho-and compare this with the corrected FAA. The intracrustal interface is between 100 m and 300 m below the seafloor under the dome; in the far-field, it is 1.5 km below seafloor; the Moho is everywhere 4.5 km below the intracrustal interface (Figure 8). In all models, the sum of the density contrasts at the two interfaces is 600 kg m−3. Full details are given in Table 2.
Models in bold are illustrated in Figures 10 and 11. All models have intracrustal interface at 1.5 km below seafloor outside defined dome.
 All the models presented fit the gravity within ±2 mGal error, and illustrate the variety and limitations of acceptable models. Model G3D_1 is illustrated in Figure 9. Here, the shallow, domed part of the intracrustal interface coincides with the outline of the smooth dome and the footwall/hanging wall boundary as identified by side scan sonar (Figure 9, red dashed lines), and is 100 m below the seafloor. The place where this interface reaches1.5 km below seafloor (blue dashed line) is chosen to give the detachment surface on the younger side of the OCC a dip similar to the 20° inferred for the shallow part of the TAG OCC [deMartin et al., 2007]. These models have peak misfits of 3 to 5 mGal and RMS misfits of 0.2 to 0.5 mGal for all three OCCs (Table 2).
 Slightly improved fits can be achieved for some or all OCCs by varying model parameters. These include: simulating a greater thickness of low density, highly fractured and serpentinised material on the upper part of the domal section by increasing the depth of the intracrustal interface to 300 m (OCC1348, model G3D_3b), and changing the density contrast across both the intracrustal and Moho interfaces to 300 kg/m3 (model G3D_2). Uncertainties in the plan view shape of each OCC core may also account for some of the residual anomaly (perhaps eliminating the apparent need for older OCCs to be less dense at shallow depths). For example, by extending the shallow, high density zone at OCC1320 by a few kilometers west and south (Figure 9, dashed white line), the modeled anomaly increases, reducing the residual to under 0.2 mGal (model G3D_4b, Table 2).
5. Magnetic Anomalies, Crustal Structure and Spreading History
 We performed inversions of the TOBI deep-towed magnetic data both on the individual measured profiles (2D) and on gridded data (3D), using the methods ofParker and Klitgord , Parker and Huestis , and Guspí  (Figure 10). Details are given in the auxiliary material. 2D inversions retain the high spatial frequencies achieved by close sampling along track and generally less severe upward continuation, but are only appropriate if the magnetic structures are truly 2D. 3D inversion deals better with 3D structures, but at the cost of lower spatial resolution. In general, however, the shape of the 2D and 3D solutions match well, and the main difference is in the amplitude of magnetization. For example, the 2D inversion shows a well-developed, ridge-centered 20 A m−1 Brunhes anomaly adjacent to OCC1320, compared to a 10 A m−1 peak in the 3D solution. Slight differences between the 2D and 3D solutions may reflect different levels of upward continuation, breakdown of the 2D assumption, or large profile spacing. Our 3D solution is comparable in shape and amplitude (but of higher resolution) to that obtained by Smith et al. using sea-surface data.
5.1. Anomaly Identification and Spreading History
5.1.1. Anomaly Identification
 Crustal magnetization across the 13°N region is highly disorganized and anomaly identification is difficult. We used the anomaly picks of Fujiwara et al.  and Smith et al.  based on sea surface data, together with the expected pattern for a symmetric spreading rate of 26 km Ma−1, as a guide.
 The onset of anomaly 2a (2.58–3.58 Ma [Cande and Kent, 1995]) is intermittently resolved at the western ends of profiles north of 13°30′N (Figure 10). Anomaly 2 (1.77–1.95 Ma) is apparent through most of the survey area, but not between 13°36′N and 13°45′N in the east. Over much of the area, the Brunhes anomaly is unclear or has a width different than the expected 20 km. We approximate a central zone of positive magnetization (central positive anomaly, CPA) by the blue line in Figure 10. The CPA forms a component, but may not be the entirety, of the Brunhes chron. It is virtually absent at 13°27′N immediately south of OCC1330, and reaches a maximum width of 30 km at 13°30′N (center of OCC1330), at 13°35′N (in the region of the NTO) and at 13°42′N. The CPA is characterized by a number of isolated peaks of typically 8–14 A m−1, sometimes with negatively magnetized (presumably Matuyama-age) material in between (e.g., on survey lines 1, 6 and 10); these are not fully resolved in the 3D solution. The CPA is relatively strong and continuous north of the NTO (though it is anomalously narrow on line 12), and there is a strong 10 A m−1 peak in the axial valley between OCC1320 and OCC1330, though narrower than expected for the Brunhes. Elsewhere, particularly over and around OCCs, the CPA is narrow or discontinuous.
5.1.2. Spreading History
 We cannot use the highly variable width of the CPA to reliably infer spreading rate; the only usable isochron is anomaly 2. We measured to the oldest (1.95 Ma), peak (1.86 Ma) and youngest (1.77 Ma) sides, yielding three estimates of full-spreading rate that average 25.5 ± 1.5 km Ma−1. This is consistent with the 25.7 ± 1.0 km Ma−1derived from NUVEL-1a [DeMets et al., 1994] and 25 km Ma−1 measured by Fujiwara et al.  around Fifteen Twenty Fracture Zone.
 We calculated separate half spreading rates for the two ridge flanks and spreading rate asymmetries on each 2D profile, using four independent estimates of the spreading axis position based on side scan, bathymetry, magnetization and gravity (Figure 11). These four estimates range from 1–10 km apart; almost coinciding just north of the NTO, at 13°40′N, and being most widely separated within it. We measured the distances from each of these axes to the peak of anomaly 2 (since this position is least sensitive to addition of the annihilator) on each profile.
 In the south, the western flank on average has spread faster than the east. The asymmetry averaged across lines 1–7 is −10%, though there is considerable variation, depending on the ridge axis adopted. North of the NTO, the opposite sense of asymmetry prevails and averages +20%. Thus, the plate containing OCCs tends to accrete faster. However, measured relative to either the central magnetic peak (green in Figure 11) or the hummocky volcanic lineaments where magnetic anomalies are currently forming (red), the asymmetry is small or non-existent, as also reported byMacLeod et al. . Nevertheless, this position is unlikely to be stable for long periods, so long-term spreading asymmetry seems likely. Measuring from the RMBA low, which probably represents a more stable long-term axis, suggests that asymmetry may be up to ±40%. This is similar to the 30% and 25% values of tectonic strain measured for the SW- and NE-quadrants, respectively, and within the 30–70% range of spreading rate asymmetries proposed byFujiwara et al.  for 1 Ma of spreading near Fifteen Twenty Fracture Zone (FTFZ). However, it is less than the 70–100% asymmetry estimated at FUJI Dome [Searle et al., 2003], the Australian-Antarctic-Discordance [Okino et al., 2004] and Atlantis Massif [Grimes et al., 2008]. Thus, averaging over 1.86 Ma probably masks shorter-period, higher amplitude variations in asymmetry associated with the formation of a single OCC. Two-dimensional magnetic models with asymmetric spreading produce reasonable fits to the observed position of anomaly 2 with −20% asymmetry since anomaly 2, −75% asymmetry for just the past 0.39 Ma, or any combination between (Figure 12).
5.1.3. OCC History
 Because our data suggest there is no hanging wall volcanism opposite active OCCs, we consider it likely that these OCCs take up most or all of the plate separation on detachments slipping at or near the full spreading rate of 25.7 km Ma−1. We use this assumption to determine the history of the near-axis OCCs.
 OCC1320 is between 8.8 km and 11.1 km wide, depending on the position of its breakaway. If it is forming at the full spreading rate, the detachment must have been active for 0.34–0.43 Ma. Similarly, OCC1330 and OCC1348 would have had full-spreading rate durations of 0.34–0.46 Ma and 0.32–0.45 Ma, respectively. Our side scan data suggest that OCC1320 and OCC1330 either are or have very recently been active, so they formed at 0.34–0.43 Ma and 0.34–0.46 Ma, respectively. (We note that they must have initiated near 44°54′W, approximately the same longitude as the axial-valley-wall fault immediately to its north (near E,Figure 1), where Smith et al. suggest an OCC may currently be forming). OCC1348 is interpreted as inactive. We assume that its breakaway ridge was initially at the along-axis projection of the present-day axial valley wall-fault, and that after the OCC terminated, it migrated to its current location at the half-spreading rate. Thus its ages of initiation and termination are 0.82–0.89 Ma and 0.37–0.57 Ma, respectively.
Figure 13 shows the inferred spreading history for the last 0.8 Ma. OCC1348 begins forming within the northern inside corner of the NTO at 0.8 Ma (Figure 13a) and is active until 0.4 Ma. During this time there is no magmatic accretion within the axial valley, which becomes sedimented and volcanically inactive, and magnetic anomaly formation by cooling of lavas is disrupted. Instead, the OCC footwall acquires a thermo-remanent magnetization in intruded gabbro or a chemical remanent magnetization via serpentinization. Because the initiating ridge jump was into Matuyama-age crust, the eastern part of OCC1348 has reversed magnetization (Figure 13b).
 At approximately the same time that OCC1348 was being terminated (0.4 Ma), OCC1320 and OCC1330 began forming within the southern inside corner of the NTO (Figure 13b). On the basis of backscatter intensity, OCC1330 probably initiated slightly earlier than OCC1320, though by an amount irresolvable by magnetic anomalies. These OCCs are thus expected to have formed almost entirely within the Brunhes chron. The present-day axial valley floor should contain positively magnetized, Brunhes-age material that formed between 0.4 Ma and 0.8 Ma by conventional magmatic processes prior to OCC formation (Figure 13c).
5.2. Magnetization Distribution
 Inferred magnetization varies significantly between adjacent profiles. For example, profile 1 shows negative magnetization over the MAR axis, compared to the (expected) positive magnetization on the adjacent profile 2, six kilometers to the north. We could make all of profile 1 positive by adding ∼15 annihilators, but would then not see the expected negative magnetization for the Matuyama chron. Similarly in the center of OCC1330, a negative anomaly on profile 6 is flanked by positive magnetization on profiles 5 and 7, only three kilometers to the S and N. This difference is only partly resolved in the 3D solution. We were concerned that these rapid variations might arise from an error, but upward continuation of the TOBI magnetic field to the sea-surface and comparison with the field measured independently there confirmed the correctness of the deep-towed results. We conclude that spatially rapid (3–6 km) 3D variations in crustal magnetization in this region are real, and need closer-spaced measurements to fully resolve them. They could potentially arise from (1) jumping of the axis of magmatic accretion (e.g., so that some Matuyama age crust remains under the axial valley on profile 1), (2) rapid variations in lithology (e.g., between highly magnetized basalt and poorly magnetized gabbro), or (3) rotation of magnetization in OCCs (which we model below).
 OCC1320 is generally associated with low amplitude, negatively magnetized crust (Figure 10). The area of weakly or negatively magnetized crust within the axial valley on survey line 1 coincides with robust, widespread recent volcanism seen in the side scan data, suggesting the NVZ here may form only a thin veneer of recent volcanism overlying a thicker accumulation of negatively magnetized material, although we cannot suggest a plausible tectonic explanation for this. At the latitude of OCC1320, the CPA is well-formed but only 8–13 km wide, representing 0.3–0.5 Ma of Brunhes magmatic accretion, consistent with the low acoustic backscatter, inferred volcanic hiatus and tectonic history (section 5.1.3). A similar ‘Brunhes deficit’ is observed at TAG, where 0.35 Ma of extension on a fault that extends 3.9 km in the spreading direction has apparently created a −5 A m−1 magnetic trough within the Brunhes by exhumation of nonmagnetic dikes and gabbro [Tivey et al., 2003].
 We model this in Figure 14a, which shows two scenarios. If the domal section (pink in the figure) has zero magnetization (perhaps reflecting low magnetization of gabbros in the footwall), it produces the magnetic anomaly indicated by the dotted line, which gives a reasonable but not good fit. This is analogous to the TAG model [Tivey et al., 2003]. A better fit is obtained if the smooth dome is modeled with a magnetization of 3 A m−1, typical of serpentinised peridotite and gabbro [Bina and Henry, 1990; Oufi et al., 2002; Rao and Krishna, 2002] and the upper massif with −4 A m−1, with both magnetizations rotated down 45° to the west, yielding the modeled field shown by the solid black line. This would roughly reflect the rotation of magnetization vectors as the footwall is exhumed, similar to the 46° ± 6° rotation inferred palaeomagnetically at Atlantis Massif [Morris et al., 2009] or the 50°–80° inferred near FTFZ [Garcés and Gee, 2007]. Negative magnetization in the upper massif implies either that the OCC initiated off-axis within Matuyama-age crust or that Matuyama-age rider-blocks have been clipped off the hanging wall and now sit atop the upper massif [Smith et al., 2008].
 OCC1330 was surveyed with three lines only 3 km apart, yet shows very heterogeneous magnetization in both 2D and 3D solutions (Figure 10). Its northern and southern flanks are associated with magnetization highs, while the center has a magnetization low (negative in the 2D but weakly positive in the 3D solution). Here we expect the 3D solution may be more accurate because of the complex topography and closer-spaced survey lines. This heterogeneity may reflect heterogeneous lithology, as at Kane [Dick et al., 2008] and Atlantis [Blackman et al., 2002; Boschi et al., 2006; Canales et al., 2008; Karson et al., 2006] Massifs. The area of low magnetization is bounded on its older and younger sides by positive magnetization peaks; thus the Brunhes anomaly appears to have been split by formation of the OCC. This can be modeled in a similar way to OCC1320 with rotated and variable amplitude magnetization, including some zero magnetization near the breakaway (Figure 14b). Although a weak, positive, rotated magnetization over the younger OCC produces an excellent fit, it does mean that the total width of normally magnetized material here somewhat exceeds the expected width of the Brunhes. However, further modeling is not justified until denser magnetic observations and sampling are available.
 Acoustic backscatter suggests the OCC1330 footwall is slightly older than OCC1320, implying that OCC1330 may have formed in older, pre-Brunhes, crust. This would not be expected to produce the magnetization peaks that flank the OCC to east and west, but they can be explained if the longitudinal location of magmatic accretion is highly variable. For example, between OCC1320 and OCC1330, the bulk of the Brunhes anomaly is not centered on the bathymetric ridge axis but over the western axial valley wall-fault. Side scan shows this area to be highly volcanically active, implying that crust is being actively constructed close to the fault—possibly due to footwall capture of ascending melt [MacLeod et al., 2009; Standish and Sims, 2010]—rather than at the center of the axial valley several kilometers to the east. Downward bending of the hanging wall (indicated by the large bathymetric depression running parallel to this fault–E, Figure 1) may have resulted in melt being concentrated here. If this fault subsequently develops into an OCC-forming detachment [Smith et al., 2008], the Brunhes would be split, as seen at OCC1330. This implies that very large normal faults—precursory structures to detachment faults [MacLeod et al., 2009]—may act to focus melt emplacement prior to OCC formation. However, the absence of the Brunhes on the older side of the apparently younger OCC1320 shows that such melt focusing is not ubiquitous.
 OCC1348 is associated with negatively magnetized material at the breakaway ridge and (in the 2D solution) weakly positive magnetization over the dome. This pattern is consistent with our 0.37–0.89 Ma predicted age of formation, the older parts being more negative. The upper massif has similar morphology to OCC1320, so may also comprise Matuyama-age rider-blocks that contribute to the overall negative magnetization. At the MAR axis, the CPA is narrow at the latitude of OCC1348 and wider immediately to the north and south, similar to the pattern predicted inFigure 13c, suggesting that magmatic accretion was more widespread at locations along-axis from where the OCC formed. Survey line 13 shows a depression within the CPA at 45°02′W that coincides with widespread recent volcanism in the axial valley. This pattern is similar to that observed along lines 1 and 5, immediately south of OCC1320 and OCC1330, respectively. The 8 km width of this magnetic depression implies that magmatism has been relatively weak during the last ∼0.32 Ma. Immediately west of OCC1348 (around 45°00′W), widespread recent volcanism is also associated with low- or negatively magnetized crust.
6.1. Tectonic Strain
 Overall, tectonic strain increases from north to south through our study area, implying a consequential decrease in magmatic extension southwards from the magmatically robust 14N segment. This is broadly reflected in the crustal thickness which also reduces southwards. Superimposed on this are strong E-W asymmetries. We confirm the findings ofMacLeod et al. that tectonic strain is asymmetrically distributed. Tectonic extension averaged over 2 Ma is greater, by a factor up to3, on plates containing OCCs than on their conjugates. Because we have limited temporal control, the asymmetry could be considerably greater while OCCs are actively extending. Plates containing OCCs, especially in the SW of our study area, also have markedly thinner crust, in particular in the SW quadrant of our survey. This may reflect a deepening of isotherms and depression of the brittle-plastic boundary where melt supply is low, allowing faults to penetrate deeper and grow larger [Shaw, 1992], predisposing to OCC formation. Similarly, spreading rates are asymmetric, with faster spreading on plates containing OCCs. Similar patterns of asymmetric spreading and crustal thickness variations are found in other areas of ultra-slow spreading and OCC formation on the MAR, Southwest Indian Ridge (SWIR) and Australian-Antarctic Discordance [Cannat et al., 2003, 2009; Searle and Bralee, 2007; Searle et al., 2003; Allerton et al., 2000; Okino et al., 2004].
 Many volcanic ridges and faults in our study area hook toward actively slipping OCC detachments. Those that do not lie within areas of low backscatter, suggesting they are older. These hooked features may be caused by deflection of the minimum compressive stress toward the stress-free boundary between OCC footwall and hanging wall. These may become precursors of hooked breakaway ridges. For example, a major hooked fault between OCC1320 and OCC1330 (M,Figure 4) is the current axial valley-wall fault and has been proposed as a site for future OCC formation [Smith et al., 2006]. It could form a hooked breakaway of the future OCC.
6.2. Melt Accretion
 Side scan sonar, coupled with gravity and to some extent crustal magnetization, reveals a complex pattern of shifting melt accretion. Some areas show robust NVZs building large axial volcanic ridges near the center of the axial valley. Elsewhere, the most recent volcanism occurs away from the topographic axis, building small volcanic ridges and, in at least one instance, being focused around an axial valley wall-fault. Off-axis volcanism has been found in other areas of poorly magmatic accretion including the SWIR [Standish and Sims, 2010] and Mid-Cayman Rise [Connelly, 2010; Murton et al., 2010].
 There is a clear boundary between lineated, magmatically robust crust in the 14N segment and the blockier, magma-poor crust of 13N. However, small areas of apparently magmatically robust crust lie within the generally blocky regions, often bounded by oblique topographic depressions similar to the traces of second-order ridge discontinuities [Gente et al., 1995; Tucholke et al., 1997]. This suggests that small areas of migrating enhanced magmatism may occur in narrow zones within magma-poor regions. Some OCCs cluster around the edges of these migrating zones of focused melt, suggesting these boundaries may mark places where there is a melt fraction critical for OCC formation [Buck et al., 2005; Karson and Winters, 1992; Tucholke et al., 2008]. As the zone migrates along the ridge axis, renewed volcanism within the axial valley may terminate slip on adjacent OCC detachment surfaces, while new OCCs may then initiate further along-axis, toward the edge of the zone of melting.
Behn and Ito , Buck et al. , and Tucholke et al.  have proposed, based on numerical modeling, that OCC initiation requires the magmatic component of extension to be about 30–50%, averaged over both ridge flanks. We calculate that this component, MT, is at most 63% in the SW quadrant (point A, Figure 3c) and 71% at the conjugate point B, giving a mean maximum of 67%. (Our estimates are maxima because RCT may reflect the presence of serpentinite as well as magmatic crust (section 4.1).) By comparison, Schouten et al.  estimated T and hence M at OCC1320 by fitting a flexural faulting model, which yields an average M of 57% to 71% for a plate with elastic thickness of 1.0 km to 0.75 km. The relatively low MTconfirms that magma-poor accretion has dominated our SW quadrant for at least the last 2 Ma and, because short-wavelength variations may be aliased, could allowM to be as low as 50% during initiation of individual OCCs. We note that Olive et al.  have shown that OCCs can form at relatively higher MT if significant melt is intruded into the lower, ductile crust; however, there is still a strong asymmetry in crustal thickness arising from the absence of melt injected into the upper, brittle crust of the hanging wall. To improve the time resolution of our analysis (for example, to estimate Mduring times of active slip on OCC detachments, or just prior to OCC initiation or following termination) would require dating these events precisely and thus a higher-resolution magnetic survey.
 [Tucholke et al., 2008] suggested that waning magmatism, evidenced by an increase in RMBA across the footwall/hanging wall boundary, may have terminated an OCC at MAR 15°45′N by initiation of a new master fault. In contrast, waxing magmatism has been proposed to promote termination by renewed melt emplacement at the MAR axis [Karson and Winters, 1992; MacLeod et al., 2009; Tucholke et al., 1998; Tucholke et al., 2001] or by rheological changes following melt accretion to the footwall [Cannat et al., 2009]. Our data are consistent with OCC termination by waxing magmatism. NVZs propagate from melt-rich areas toward the tips of active OCCs (OCC1320 and OCC1330), and eventually emplace a continuous zone of volcanism through their footwalls (OCC1348). As the footwall/hanging wall boundary migrates back toward the regional position of the MAR axis, any combination of: (1) renewed magmatism at the ridge axis, perhaps due to natural episodicity, (2) along-axis propagation of volcanism, or (3) significant melt being captured by the OCC footwall, would thermally alter and weaken the lithosphere near the MAR axis and OCC toe, making it easier to create a new fault rather than to maintain slip on the existing detachment.
 Such gabbroic material would have to extend 2–5 km beneath the detachments if it had a density of 2900 ± 100 kg m−3. The LDZ must be thicker or of lower density at OCC1330 and OCC1348 than at OCC1320. This may reflect greater serpentinization and increased faulting and degradation of the footwall, and perhaps also more mechanical weathering [Blackman et al., 2008] in older OCCs. Alternatively, the thicker northern LDZ may reflect a greater component of gabbro intruded into the footwall as a result of the regionally greater magmatism there. At the OCC1348 breakaway, thinner crust may mark reducing magmatic accretion prior to OCC commencement, and lower density LDZ may indicate additional serpentinization in the oldest exposed mantle.
 Our estimates of LDZ thickness are consistent with the 1.4 km of predominantly gabbroic material drilled from the dome of Atlantis Massif [Ildefonse et al., 2006], and with Blackman and Collins's  finding that seismic velocities in the top 6.8 km there do not exceed 7.8 km s−1.
6.4. OCC Corrugations and Striations
 If bathymetric corrugations are produced by casting of ductile footwall against the brittle hanging wall [Spencer, 1999; Tucholke et al., 2008], then variations in their amplitude and wavelength between OCCs may reflect subtle changes in the topography of the brittle-ductile transition, consequent on irregular magma intrusion in a magma-poor environment. Striations have been observed in the upper massif at OCC1320, which appears to contain mafic, crustal rocks. If they, like corrugations, are produced by casting of relatively ductile footwall against brittle hanging wall, then their occurrence on the upper massif indicates a transition to relatively ductile crust at depths of ∼1.5–3.5 km (the distance from the breakaway to the oldest, most-westerly striations). This could indicate the development of abnormally warm lithosphere during OCC formation.
 We observed a significantly different trend between corrugations on OCC1348 compared with OCC1320 and OCC1330. The former match the predicted NA-AF plate motions [DeMets et al., 1994], while the latter match predicted SA-AF motion. This suggests that the NA-SA-AF triple junction lies between 13°30′N and 13°48′N, somewhat south of the 14.3°N position suggested byEscartín et al. .
6.5. Magnetic and Lithological Heterogeneity
 The inferred crustal magnetization pattern in our study area is highly heterogeneous. It can be partly resolved by taking account of tectonic rotations associated with footwall exhumation, but this cannot explain all the anomalies. Three-kilometer-spaced survey lines over OCC1330 show completely different magnetizations over what appear to be tectonically coherent terrains. This probably reflects lithological heterogeneity between varyingly serpentinized peridotite and gabbro intrusions [Dick et al., 2008; Blackman et al., 2002; Boschi et al., 2006; Canales et al., 2008; Karson et al., 2006]. Resolution of these variations must await higher resolution magnetic observations and detailed sampling.
 The central positive magnetization anomaly, possibly representing Brunhes-age crust, is of very variable width. In some places there appears to be no positive magnetization at the ridge axis. Part of this ‘Brunhes deficit’ may arise from low magnetization of serpentinised peridotite or gabbro intrusions in OCC footwalls, which will not necessarily produce a distinct, positive magnetic anomaly similar to normal magmatic spreading.
 In magmatically formed Brunhes-age crust, multiple magnetization peaks at the same latitude suggest that the locus of melt emplacement jumps across-axis episodically at 0.25–0.40 Ma intervals by up to 12 km (e.g.,Figure 10, survey line 10). This is supported by our side scan sonar data, which show a range of inactive and active volcanic lineaments across the axial valley-floor displaying varying acoustic backscatter.
 Magmatic accretion and crustal thickness broadly decrease from north to south through the study area, but with strong superimposed E-W asymmetries. Observed patterns of recent volcanism are consistent with OCCs being terminated by waxing magmatism, and variations in inferred crustal thickness hint that waning magmatism may be involved in initiating them. There is a complex pattern of shifting melt accretion within the median valley, including some areas where magmatism appears to be concentrated very near median valley wall faults.
 Average tectonic strain since chron 2 is three times greater in plates containing OCCs than in their conjugates. The azimuths of faults and volcanic ridges are deflected away from MAR-parallel toward the emerging boundaries of OCCs, in a process of stress refraction that may also explain curved breakaway ridges. The azimuths of corrugations on OCC footwalls are different north and south of about 13°40′N, suggesting they may reflect the presence of the NA/SA plate boundary near there.
 Gravity data show OCCs to be underlain by dense cores, but require that the tops of their domes are underlain by zones several kilometers thick of material of lower density than normal mantle; this could be gabbro and/or serpentinised peridotite, which are indistinguishable gravimetrically.
 Inferred crustal magnetization is highly variable, including anomalously wide and narrow parts of the presumed Brunhes anomaly. Spatially rapid variations in magnetization over OCCs suggest highly heterogeneous lithology within them.
 The Natural Environment Research Council funded this work through grant NE/B500058/1 and a tied studentship to CM. During the writing of the paper, RCS was partially supported by Emeritus Fellowship EM/4/EM/2010/0086 from the Leverhulme Foundation. We are indebted to the Officers, crew and shipboard scientists of RRS James Cook cruise 007 for their assistance in data acquisition. Special thanks to Debbie Smith and colleagues for providing us with bathymetry and other data prior to publication. We thank our many colleagues who contributed to the ideas presented here, particularly Chris MacLeod, Bramley Murton, Kay Achenbach, Tim Le Bas, Debbie Smith, and Jean-Arthur Olive. Debbie Smith, Pablo Canales and two anonymous reviewers made many suggestions that have significantly improved the paper.