Seismic shear wave structure of the uppermost mantle beneath the Mohns Ridge



Crust produced at mid-ocean ridges with full spreading rates less than ∼20 mm/yr is observed to be only 0–4 km thick, well below the global average of 6–7 km for oceanic crust produced at faster spreading rates. The origin of this difference is unknown, but is speculated to result from a thicker thermal boundary layer at the axis of the slowest spreading ridges that either inhibits shallow melting or melt extraction. We present an analysis of regional broadband data, predominately Love and Rayleigh waves, collected along the very slow spreading Mohns Ridge (where crustal thickness is ∼4 km) in the Norwegian-Greenland Sea. The seismic data constrain lithospheric and asthenospheric velocities for lithospheric ages from 0 to 25 Ma. We find lithospheric thickness to closely match the prediction of a simple ridge half-space thermal model (via the temperature and pressure effects on the seismic properties of mantle materials). Asthenospheric shear wave velocities are consistent with this thermal model plus ≤2% melt at the youngest ages. Just at the top of the mantle, a thin zone with velocities intermediate between those of mantle and gabbroic rocks suggests that some melt is frozen into the mantle as might occur if a thin axial lithospheric lid inhibits melt extraction from the mantle and the melt is subsequently frozen into the mantle. If this process occurs to some degree at all slow to very slow spreading ridges, then there may be spatially large reservoirs of unaccounted for gabbroic melt frozen into the mantle throughout many ocean basins.

1. Introduction

While regional waveform studies from the Pacific to the Atlantic [e.g., Nishimura and Forsyth, 1989; Xu and Wiens, 1997; Gaherty, 2001; Dunn and Forsyth, 2003; Gu et al., 2005; Wiens et al., 2006; Delorey et al., 2007; Gaherty and Dunn, 2007] constrain the seismic structure of the upper mantle for mid-ocean ridges with full spreading rates greater than 20 mm/yr, few studies of this nature exist for mid-ocean ridges spreading at rates less than 20 mm/yr. Despite the lack of mantle seismic studies for these very slow spreading ridges, over the past decade much has been learned about this relatively new class of mid-ocean ridge. Ridges in this category display linked magmatic and amagmatic accretionary segments, a lack of transform faults, exposed mantle on the seafloor, and anomalously thin crust [Dick et al., 2003]. Measured crustal thicknesses range from 0 to 4 km, well below the 6–7 km global average for oceanic crustal thickness [Bown and White, 1994]. Abnormally thin crust has been discovered at the Gakkel Ridge [Jackson et al., 1982; Michael et al., 2003], the South-West Indian Ridge [Muller et al., 1997], the Knipovich Ridge [Ritzmann et al., 2002] and the Mohns Ridge [Klingelhöfer et al., 2000a]. For some sections of these ridges, the melt supply is suspected to at least temporarily drop to zero, resulting in the direct emplacement of mantle at the seafloor [e.g., Dick et al., 2003]. The process of crustal production at mid-ocean ridges and other spreading centers is fundamental to our understanding of mantle dynamics and melting and this observed spreading rate dependence on crustal production at very slow spreading rates indicates a fundamental change in sub-ridge processes.

At present, there are two end-member hypotheses to explain the reduced crustal production at very slow spreading ridges. While each emphasizes the importance of a thermal lid on controlling sub-ridge mantle processes, one suggests that the thermal lid suppresses melting while the other posits that it inhibits melt migration. In the melting-suppression model [Reid and Jackson, 1981] the sub-axial thermal structure is governed by conductive cooling and in very slow spreading environments a thick thermal boundary layer forms at the ridge axis. This shuts off melting at the top of the melting column at deeper depths than is predicted to occur for faster spreading ridges and, all else being equal, reduces the overall amount of melt produced in the mantle. Numerical calculations [e.g., Reid and Jackson, 1981; Bown and White, 1994; Niu and Hekinian, 1997] predict that melting shuts off at a depth of 30 km or more below the crust for the slowest spreading ridges. The global three-dimensional S wave velocity model of Zhang and Tanimoto [1992] has been interpreted to support the thick lithosphere theory [Niu and Hekinian, 1997]. Averaged S wave velocity variations plotted as a function of spreading rate indicate that slower spreading ridges are underlain by seismically faster material at depths ≤36 km. This increase in velocity can be interpreted as thicker lithosphere directly beneath the axis of slower spreading ridges as compared to faster spreading ridges, although the lateral resolution of the study at the ridge axis is poor.

In the inhibited-migration model [Cannat, 1996], melt production is normal but melt transport through the lithosphere is at least partially inhibited. This model does not require an exceptionally thick lithosphere along the ridge, but only a slightly thickened mantle lithosphere. The thickened lithosphere causes mantle melts migrating to the surface to become trapped and frozen beneath the crust; the final crustal thickness therefore does not represent the total amount of melt produced within the mantle. For example if the thickness of magma produced by the mantle is 6 km and the mantle lithosphere is 6 km thick at the ridge with 50% of the material composed of entrapped melt, then the final crustal thickness will be only 3 km. In support of this model, Cannat [1996] pointed to several lines of evidence including numerical estimates of lithospheric thickness, depth of seismicity, seismically determined crustal thickness, gravity, seafloor morphology and rock samples. Of importance is that regions of thin crust appear to have thickened lithospheres and contain outcrops of ultramafic rocks pervaded by frozen veins of gabbroic melts; some petrogenic studies also show evidence for deep onsets of crystal fractionation. More recently, a seismic study of Atlantic lithosphere found an abrupt drop in sub-Moho mantle velocities correlating with a change from slow to ultraslow paleospreading rate [Lizarralde et al., 2004]. The estimated volume of retained melt in the mantle balances an observed difference in crustal thickness for the two spreading rates.

These two hypotheses are observationally testable, and distinguishing between them can provide significant information on mantle dynamics beneath mid-ocean ridges. To help distinguish between them, we performed a detailed regional waveform study of the upper mantle seismic structure beneath the Mohns Ridge, a very slow spreading ridge with anomalously thin crust. By means of a joint inversion using both Love and Rayleigh waves, we developed lithospheric age-dependent velocity models of the Mohns Ridge. These models constrain the thickness and shear wave velocity of the lithosphere, asthenospheric velocities, and mantle anisotropy. The results are compared to a prediction derived from a thermal cooling model and to similarly determined seismic models from the north Atlantic and the Pacific.

2. Geologic Setting

The Mohns Ridge (Figure 1a) is centrally located in the Norwegian-Greenland Sea and is obliquely spreading at a full rate of 16 mm/yr [Vogt, 1986; Kreemer et al., 2003; Müller et al., 2008]. The location of the ridge axis has remained relatively stable since its formation ∼60 Ma [Talwani and Eldholm, 1977] thus resulting in nearly symmetric spreading. The strike of the ridge is 60° from north and its length is about 580 km.

Figure 1.

(a) Map depicting the age of the seafloor [Müller et al., 2008] surrounding the Mohns Ridge. The ages are roughly symmetric about the ridge with the oldest seafloor dating back to approximately 60 Ma. Areas of seafloor depicted in light gray are significantly older continental shelf; darker gray areas represent neighboring landmasses. The locations of seismic stations JMI and JMIC are denoted by a triangle. Earthquake epicenters are plotted as dots; the shaded lines represent great circle paths from events to stations. Black dots and lines represent events used to solve for mantle seismic structure within the lithospheric age range of 0–5 Ma, gray for 5–15 Ma, and white for 15–25 Ma. (b) A bathymetric map [Jakobsson et al., 2008] of the Mohns Ridge region showing the 27 earthquakes employed for the analysis of the seismic structure. Focal mechanism information for each event was extracted from the International Seismological Centre Online Bulletin (red) ( and Global CMT Project (blue) [e.g., Ekström et al., 2005] catalogs.

Water depth along the Mohns Ridge ranges from ∼2.8 km at the southern end to ∼3.2 km at the northern end [Géli et al., 1994]. The shallow southern end of the ridge is thought to result from a melting anomaly suggested by some to be the product of interaction of the ridge with a hot spot [Talwani and Eldholm, 1977; Vogt et al., 1981]. Possible evidence for such interaction comes from a regional tomographic study [Pilidou et al., 2004, 2005] revealing Sv wave velocities at 100–300 km depth that vary along the ridge with relatively lower velocities toward the southwestern end of the ridge and higher velocities toward the northeastern end. The anomalous structure may result from an interaction with the Iceland hot spot, but the existence of a unique plume beneath Jan Mayen has also been speculated [Neumann and Schilling, 1984; Schilling, 1985; Pilidou et al., 2004, 2005]. Geochemical analyses of Mohns Ridge lavas show variable incompatible element enrichments unique to the area [Haase et al., 1996; Schilling et al., 1999; Hanan et al., 2000] that are greatest at the Jan Mayen end of the ridge and fade with distance to the north. Also, selected isotopic ratios, such as 87Sr/86Sr and 206Pb/204Pb, trend from higher values at the southwestern end of the Mohns Ridge toward lower values at the northeastern end [Schilling et al., 1999; Hanan et al., 2000] and have been interpreted as an indicator of hot spot influence near Jan Mayen, although this interpretation is controversial [Haase et al., 1996].

An active-source seismic refraction and gravity study that sampled crust produced at the Mohns Ridge over the age range of 0–22 Ma, has discovered anomalously thin crust with a mean thickness of 4.0 ± 0.5 km [Klingelhöfer et al., 2000a]. The thin crust results from a very thin, underdeveloped, lower crustal layer (seismic layer 3) that is often interpreted to be composed of gabbros. The authors do not rule out that layer 3 may be at least partially composed (<30%) of serpentinized mantle rocks. The mantle immediately beneath the crust exhibits unusually low P wave velocities, ∼7.5 km/s on average, and as low as 7.2 km/s along a seismic profile located within the rift valley; the depth to which these low velocities penetrate is not well constrained. The authors suggest that faults and fractures, along which seawater penetrates, may reach greater depths than for faster spreading ridges, with the result that fluids chemically alter (serpentinize) the upper mantle and lower the bulk seismic velocity. They also suggest that the alignment of the fast-axes of olivine in the mantle flow direction, away from the ridge, may additionally lower the seismic wave speed along their ridge-parallel seismic profiles.

A rare earth element (REE) inversion assuming a primitive mantle yields a melt thickness, and equivalent crustal thickness, of 5.9 km for the Mohns ridge [Klingelhöfer et al., 2000b]. Because the evidence supporting a Jan Mayen mantle plume is controversial [Haase et al., 1996], the authors prefer the results of an additional REE inversion assuming a mantle with an average ɛNd (143Nd/144Nd ratio) of 7.0, as observed for Mohns Ridge basalts, that indicates melt production with an equivalent crustal thickness of ∼5 km. This thickness corresponds well with melt production inferred from Na concentrations corrected for fractionation [Klingelhöfer et al., 2000b]. The discrepancy between the geochemically derived equivalent thickness of crust and the seismically derived crustal thickness is ∼1 km; however, the associated errors are roughly 0.5–1.0 km. Conservatively, the authors suggest that at most 10% gabbro is trapped in the upper 15 km of the mantle. This, they claim, is too little to explain the observed low mantle seismic velocities. However, since the thickness over which gabbro might occur is as yet unknown, a similarly conservative estimate for a 5 km thick layer is 30% gabbro in the mantle.

3. Seismic Data

This study utilizes data from 27 earthquakes located along the Mohns and Knipovich ridges (Figure 1b). Source mechanisms predominantly indicate normal faulting with the direction of extension approximately parallel to the direction of spreading; only those events whose back azimuth do not lie along a node in the radiation pattern were used. Events range in magnitude from Mw 4 to 6 with source-to-receiver distances from 90 km to 950 km for the furthest Knipovich Ridge events. The data were recorded on broadband stations JMIC (operational from 10/31/2003 to present) and its predecessor JMI (operational from 8/31/1994 to 10/31/2003). Both stations employed Streckeisen STS-2 instruments owned and operated by Norwegian research groups, and waveform data are publicly available via the NORSAR and IRIS databases. These stations are ideally situated for studying Mohns Ridge mantle structure because of their location on Jan Mayen Island near the southern end of the ridge.

Each seismic trace was initially cut to 120 s before and 120 s after the expected S wave arrival time and the horizontal data were rotated to obtain radial and transverse waveforms; all data were corrected for instrument response. JMI data prior to October 2001 were corrected for a north/east channel swap problem while all JMI data were corrected for a 16° (azimuthal) instrument misalignment. Vertical and transverse broad band-pass filtered data sorted by epicentral distance are shown in Figure 2; strong surface wave arrivals for both Rayleigh and Love waves are clearly observed in the data. For waveform modeling, fundamental mode wavelets of Love and Rayleigh waves, centered about a given period, were extracted by narrow band-pass filtering the data as discussed below.

Figure 2.

Plots of broad band-pass filtered data with filter corners at (top) 18 and 65 s for the vertical channel and (bottom) 7 and 40 s for the transverse channel. Events are ordered according to distance, as indicated on the plots. Arrival times are reduced using a velocity of 4.0 km/s. Surface wave arrivals, as well as some body wave arrivals (for example, those observed on the vertical component records ranging from ∼150 to 350 km distance and indicated by the dashed line), are clearly observed. The relative phase shifts in waveforms for adjacent events can be attributed to the effects of variations in distance of the epicenters from the ridge axis and/or changing focal mechanisms. The data are dominated by ∼30 s (vertical channel) and ∼10 s period (transverse channel) waves.

4. Methods

The data were divided into three sets to model mantle structure for the following lithospheric age ranges: 0–5 Ma, 5–15 Ma, and 15–25 Ma. We solved for an average one-dimensional (depth varying) velocity profile for each age range. The 19 earthquakes distributed along the Mohns Ridge were used to invert for the near axis (0–5 Ma) seismic velocity profile while a total of 8 earthquakes (4 per age range) located along the Knipovich Ridge constrained the 5–15 Ma and the 15–25 Ma velocity profiles. Seismic wave paths for the 15–25 Ma group traverse lithosphere of variable age, up to ∼50 Ma, and as such may not accurately represent 15–25 Ma lithosphere. Nonetheless, the average age of the lithosphere traversed by the waves lays within the given age limits. Synthetic seismograms were calculated using a reflectivity code based on the method of Randall [1994] as modified by Xu and Wiens [1997] to include a water layer. The reflectivity method calculates the waveforms for both body and surface waves therefore eliminating the difficult task of separating the body wave arrivals from the surface wave arrivals at small source-to-receiver offsets. The reflectivity method requires the basic event parameters and moment tensor, along with 1-D models of P wave velocities, S wave velocities, and density. Before beginning the modeling, we verified the waveform code by comparing its output against known solutions for simple seismic structures.

For the 0–5 Ma age range, the initial S wave velocity model used to start the data modeling (Table 1) was determined by first computing a basic half-space cooling model for oceanic mantle (with a crust-mantle interface temperature of 650°C and mantle potential temperature of 1300°C) [Turcotte and Schubert, 2002] and then converting this to a seismic model using the methods of Faul and Jackson [2005]. A single profile is determined by laterally averaging the model over the 0–5 Ma age range. The Faul and Jackson method accounts for the temperature and pressure effects on mantle minerals, but neglects the possible effect of melt in the asthenosphere. Their method is also frequency dependent and we averaged the results over the appropriate frequency range for the data. The mantle P wave velocity was set to 1.8 times the S wave velocity. For the crust, values were taken from the results of a local active source seismic study [Klingelhöfer et al., 2000a]. The starting velocity model for the 5–15 Ma age group was the final solution for the 0–5 Ma age group, and the starting model for the 15–25 Ma age group was the solution for the 5–15 Ma age group.

Table 1. Initial Model
Layer Thickness (km)Vp (km/s)Vs (km/s)ρ (g/cm3)

Synthetic seismograms in the vertical and transverse directions were narrow band-pass filtered in the same manner as the data. A Gaussian shaped filter was used with thirteen center periods for Rayleigh waves (15, 18, 22, 26, 30, 34, 42, 50, and 65 s) and Love waves (9, 11, 14, 17, 20, 24, 29, 34, and 40 s). In the frequency domain, the filter is of the form F(w) = exp(− equation image) where w is frequency in radians, and the filter width, γ, ranged from roughly 0.02 to 0.03, depending on the center frequency, w0. The amplitudes of the wavelets were normalized by their standard deviations to minimize any effects of the uncertainty in the instrument response (particularly station JMI), the seismic moment, and the earthquake depth.

Direct comparison of filtered synthetic and observed waveforms via a computed misfit value (the mean absolute difference between the synthetic and observed seismograms) determined goodness of fit. The misfit function contains phase and group arrival information, but due to the amplitude normalization (previous paragraph), the sensitivity of the misfit function to waveform amplitude is suppressed. In order to down weight the importance of waveforms with low signal-to-noise ratios, the misfit for each wavelet was divided by a weighting factor, or relative waveform uncertainty. The weighting factor was calculated as the mean absolute difference between the observed wavelet and a theoretical (noise-free) wavelet with the same center period and phase. The radial component data yielded waveform uncertainties significantly larger than the other two components and were eliminated from the analysis. A few individual wavelets with very low signal-to-noise ratios were also removed. To account for crustal thickness and water depth variations along the great circle paths, surface wave phase corrections were calculated [Bozdağ and Trampert, 2008] and applied to the synthetic data resulting in a more accurate representation of the observed data (i.e., improved misfit); further discussion of these corrections is provided in section 5.

The inversion was performed using an iterative grid search method over the velocity layers of the lower crust and the upper mantle. The search began at the top of the model and moved downward over 13 velocity layers to ∼200 km depth. On the basis of a sensitivity analysis, the model below ∼200 km depth and within the uppermost crust has negligible influence on the data and was not perturbed during the inversion. A smoothness constraint (weak three-point averaging with weights of 0.2, 0.6, and 0.2) was applied with each iteration to avoid any spurious perturbations. Initially, only longer period data were included; shorter-period data were added as the model misfit improved. An obvious discrepancy between Rayleigh and Love data indicated the presence of anisotropy (VSH > VSV) which is often detected beneath mid-ocean ridges and is generally attributed to the horizontal alignment of the a-axis of olivine crystals due to shear deformation via mantle flow [e.g., Nishimura and Forsyth, 1989]. Radial anisotropy was included in the mantle portion of the model and the values were determined as part of the iterative scheme. Although crustal anisotropy is possible along mid-ocean ridges [e.g., Dunn and Toomey, 2001], allowing anisotropy in the crust did not yield a significant improvement of the data misfit and was not included in the final results.

5. Results

The final solution of the anisotropic 1-D velocity profile for each age range is shown in Figure 3 and waveform fits for several periods and events are shown in Figure 4. Average misfits of the wavelet phases, as a function of period, are shown in Figure 5a for the starting model (0–5 Ma age range) and Figures 5b5d for the final models. At each age range, a prominent high-velocity lithospheric lid is present at the top of the mantle (Figure 3a). The thickness of the lithospheric lid is least for the 0–5 Ma profile and thereafter increases substantially with lithospheric age. Beneath the lithospheric lid, shear wave velocities decrease with depth forming an upper mantle asthenospheric low-velocity region.

Figure 3.

Solutions for the near- and off-axis (a) VSV velocity profiles and (b) anisotropy profiles. In each case the vertical axis is referenced from the seafloor. Anisotropy is defined here as (VSH-VSV)/Vav, where VSV is the shear wave speed of vertically polarized waves, VSH is the shear wave speed of horizontally polarized waves, and Vav is the average of the two. Each VSV solution exhibits a ∼7 km thick region of anomalously low velocities just beneath the base of the crust. This region, shaded and extending from ∼4 to 11 km depth in the figure, has shear wave velocities intermediate between lower crustal and typical lithospheric values. The high velocity lithospheric lid is thinnest for the 0–5 Ma age range and thereafter increases with age. The 0–5 Ma profile, and to a lesser extent the 5–15 Ma profile, also exhibit a sub-lithospheric low velocity zone in the depth range of ∼20–70 km that is not apparent at the oldest ages. Anisotropy, such that VSH > VSV, is detected in the upper mantle; it is strongest just beneath the crust and decays with depth. The anisotropy is generally less at older ages, but this may simply be a consequence of the different azimuths of the seismic raypaths combined with azimuthal anisotropy.

Figure 4.

Observed data and corresponding synthetics for the 0–5 Ma age range for both (a) Love and (b) Rayleigh waves. The black curves are the observed data and the red curves are the synthetics (time increases downward). Select events are shown at several center periods (using a Gaussian-shaped band-pass filter), as indicated on the plots. Corresponding average misfit values of the wavelet phases are shown in Figure 5b.

Figure 5.

Plots of Rayleigh and Love wavelet misfit, calculated here as the phase difference in seconds between the observed and the synthetic wavelets, as a function of period. The value plotted for each period is the average misfit for all events at that period. These plots illustrate the modeling sensitivity to small changes in shear velocity. Positive values indicate the model is too fast (early arrivals) while negative values indicate the model is too slow (late arrivals). (a) The misfit relative to the starting isotropic seismic model for the 0–5 Ma age range. The starting model fits the data quite well, but with an obvious Love-Rayleigh (radial anisotropy) disparity. (b) The misfit relative to the final anisotropic velocity model of Figure 3, 0–5 Ma range. The trends and means of the misfit curves of (a) disappear. (c) Misfit for the 5–15 Ma range. (d) Misfit for the 15–25 Ma range. The older age ranges fit the data somewhat less well as compared to the 0–5 Ma range, a consequence of the more complicated lithospheric structure traversed by the waves. (e) The final anisotropic model, 0–5 Ma, with a thicker, ∼6 km, crust. (f) The final anisotropic model, 0–5 Ma, with an enforced 18 km thick lithosphere.

A notable feature of each of the three profiles is a region of Anomalous Low Velocities IN (ALVIN) the upper mantle just beneath the base of the crust. The ALVIN region is ∼7 km thick with shear wave velocities of ∼4.0–4.4 km/s. Shear wave values at the oldest ranges are slightly faster than those at the youngest age range, however this age-related difference is only weakly supported by the data. In general these shear wave values are less than expected for typical oceanic lithosphere (which can be 4.5 km/s or more), but in line with those estimated from the local refraction experiment of Klingelhöfer et al. [2000a].

Sensitivity tests were performed to determine to what extent the ALVIN region is required by the data. The ALVIN region was a persistent feature of all solutions to the problem irrespective of starting model, smoothness constraints, anisotropy, or assumption about the thickness or velocity of the crust. For example if the upper layers of the mantle were artificially held at 4.5 km/s, to coincide with the fastest lithospheric layer, the data misfit increased significantly, particularly for the shorter period data and a discrepancy between Rayleigh and Love phase fits formed. The tests do show some trade-off between the velocity of the layers within the ALVIN region and the ALVIN region's thickness as well as a trade-off with velocities just above or below it. While model uncertainties are difficult to obtain with this type of experiment, we estimate roughly a 2 km uncertainty in the ALVIN thickness and 0.2 km/s uncertainty in the shear wave speed.

The 0–5 Ma age range exhibits a mantle anisotropy of nearly 6% (VSH > VSV; anisotropy is defined here as equation image) in the top 20 km of the mantle that decreases at greater depths. As compared to a purely isotropic solution, the anisotropic solution removes a discrepancy between the data fits of the Rayleigh versus the Love data and the overall misfit drops by 13%. Sensitivity tests show that below ∼150 km depth, the data are not sensitive to anisotropy. For this test cosine tapers were applied to the bottom of the anisotropy model over a range of depths to determine to what depth the data were no longer sensitive to anisotropy. In addition, below ∼80 km depth there is some trade-off between isotropic and anisotropic values due to the differences in depth sensitivity of the Love and Rayleigh waves. As compared to the 0–5 Ma age range, the older age ranges exhibit lower degrees of anisotropy, with peak values near 4.5%. Since the great circle raypaths for the older age ranges are oblique to the axis of the Mohns Ridge, azimuthal anisotropy may reduce the Love-Rayleigh discrepancy that gives rise to our radial anisotropic result; this interpretation, however, is not unique.

The sensitivity of the results to crustal thickness is an important issue. While we assume the average crustal thickness along the raypaths is 4.2 km on the basis of the Klingelhöfer et al. [2000a] refraction study, the actual crustal thickness along the paths may be different. For example, seafloor bathymetry suggests thicker crust to the south, near the stations, and much thinner crust at the northern end of the Mohns Ridge. As noted above, the effect of crustal thickness changes was corrected for in the synthetic data. This was done by first assuming 4.2 km thickness for the central portion of the Mohns ridge and then using seafloor bathymetry and an isostasy argument to predict crustal thickness elsewhere along the raypaths. Given the predicted thickness difference relative to 4.2 km, we calculated the accumulated phase changes due to the crustal thickness variation [Bozdağ and Trampert, 2008] and added the result to the synthetic data. While this reduced the overall misfit, there is no assurance that the synthetics are corrected to the proper average crustal thickness. Therefore the sensitivity of the seismic profiles to changes in average crustal thickness was tested by increasing and decreasing the thickness of the lower crust. The results of the tests indicate that while there is some trade-off between the crustal thickness and velocities both above and below the base of the crust, unknown crustal thickness variations are not likely to explain away the ALVIN region itself. As an example, a 6 km thick crust produces significantly higher average misfits (Figure 5e) relative to the model with a 4.2 km thick crust, particularly affecting shorter periods where average misfits in the phase are greater than equation image.

The sensitivity of the data to the thickness of the high-velocity lithospheric lid was also examined. For the 0–5 Ma age range, the thickness was artificially increased by 6 and 12 km over that shown in Figure 3a. The first case resulted in slightly higher misfit values (2.3%) and early arrivals of 1–1.5 s on average for the 14–40 s Love data and the 15–30 s Rayleigh data. The second case produced similar results; misfit values increased by 10.8% and arrivals were 2–2.5 s on average too early for the same periods (Figure 5f). The misfit increases more rapidly if the lithosphere is thinned rather than thickened. Again, while it is difficult to place absolute bounds on model values, the data show enough sensitivity to lithospheric thickness that we suggest that our final model is close to having the appropriate average lithospheric values for the given age range.

6. Interpretation and Implications

Figure 6 shows the final VSV models for the three age ranges as compared to other velocity models of the same age: the Faul and Jackson [2005] prediction based on a half-space cooling model and the effects of temperature and pressure on mantle materials, the Nishimura and Forsyth [1989] model for the Pacific, and the Delorey et al. [2007] model for the northern Atlantic and Reykjanes Ridge. In Figure 6, all model depths are referenced to the base of the crust. At the oldest age range, the Mohns (apart from the ALVIN region) and Pacific models agree with the half-space cooling model, suggesting no significant effect from ridge-related mantle melting or other influences. At the 5–15 Ma age range the Mohns model has slightly lower sub-lithospheric velocities than predicted by the simple cooling model, perhaps indicating the addition of a small amount of melt originating from ridge-related upwelling and decompression melting. At the 0–5 Ma age range, the data-derived models all deviate from the simple cooling model, as is expected for the presence of melt in the upwelling region. For the Mohns model, ≤2% melt [e.g., Schmeling, 1985] can explain the observed discrepancy at 0–5 Ma and ≤1% melt can explain the discrepancy at the 5–15 Ma age range.

Figure 6.

Mohns Ridge VSV models as compared to other VSV models for the same age ranges. These models include a half-space cooling model converted to seismic shear wave speed via the method of Faul and Jackson [2005] (F&J), the Pacific mantle model of Nishimura and Forsyth [1989] (N&F; their results have been interpolated to our age ranges), and the north Atlantic, Reykjanes Ridge, model of Delorey et al. [2007] (DDG). All models are referenced to the base of the crust in this figure. The F&J model shown here is different from that used as our starting model. Here, the F&J model has a lower reference temperature at the top of the mantle (600°C versus 650°C) and a smaller grain size (2 mm versus 5 mm) that results in an overall better correlation with the Mohns Ridge models. The greatest variation between the four models occurs within the 0–5 Ma age range where each of the data-derived models has lower sub-lithospheric shear velocities than the F&J model, which is derived from temperature and pressure effects on mantle materials. Such a deviation may indicate the presence of a small amount of melt in the upper mantle. Only the Mohns model contains the sub-crustal low velocity (ALVIN) region in the topmost ∼7 km of the mantle.

For the 0–5 Ma age rage and below 60 km depth, the Mohns mantle has faster shear wave velocities than for either the Pacific [Nishimura and Forsyth, 1989] or Reykjanes [Delorey et al., 2007] models and is similar to the estimate from the half-space cooling model. Tests using different starting models indicated that these high velocities for the Mohns mantle are required by the data. There is some trade off between isotropic and anisotropic values below ∼80 km depth, but the anisotropy is required to be negative (VSV > VSH) at those depths to produce enough change in the isotropic model to match the other geographic regions. This also raises the data misfit and is therefore not the preferred solution. The deep low velocities in the Pacific model do not appear at the older age ranges, indicating that they arise from an upwelling phenomenon. Since melting in the presence of water is expected to occur below ∼60 km depth [Hirth and Kohlstedt, 1996], then perhaps some wet melting occurs in the Pacific, but little or no such deep melting occurs beneath the Mohns ridge on average. In addition, the velocities in the Mohns model below 60 km depth do not increase with distance from the ridge axis (or more accurately, the data do not require such an increase) and thus there is no obvious upwelling-related anomaly below this depth. While we cannot rule out melt at depths below 60 km in general, any broad upwelling-related melting that may take place occurs in undetectable quantities. The sensitivity of the long-period waves used to determine the velocities at these depths is broader than the 0–5 Ma age range, so the possibility of a narrow upwelling zone cannot be ruled out.

The Reykjanes model deviates the most from the thermal model predictions and is suggestive of a broad and deep region of melt and melting that may result from the influence of the Iceland hot spot [Delorey et al., 2007]. The Mohns ridge seismic structure does not suggest a similar influence from Iceland [Talwani and Eldholm, 1977; Vogt et al., 1981] or the proposed Jan Mayen hot spot [Neumann and Schilling, 1984; Schilling, 1985]. On the basis of the ridge geochemistry it is reasonable to expect some sort of anomalous mantle structure beneath the southern part of the Mohns ridge, however our seismic results provide the average mantle structure over the entire length of the ridge and therefore would not reflect any localized anomaly.

At very slow spreading rates it has been suggested that deeper, more pervasive fracturing of the crust enhances hydrothermal cooling [Francis, 1981] leading to a considerably thicker lithosphere than predicted by conductive cooling models. Our results do not substantiate this idea. The Mohns Ridge velocity structure for the 0–5 Ma age range (Figure 6a) is consistent with “normal” thickness lithosphere as predicted by a simple conductive cooling model. If we define the base of the lithosphere as the depth of maximum negative gradient of the velocity [e.g., Gu et al., 2005], then the Mohns Ridge velocity model suggests a lithospheric thickness of ∼22–34 km, neglecting the crust. The thickness of the mantle lithosphere predicted by the half-space cooling model is ∼25 km averaged over this age range. Lithospheric thicknesses estimated for the older age ranges also agree well with this theoretical calculation. Although the data do not sample in detail the axis of the ridge, the thickness of the mantle lithosphere at the ridge axis is estimated to be <10 km on the basis of the off-axis profiles and extrapolation of conductive cooling isotherms back to the ridge axis. In addition, gravity data indicate the presence of a low-density region in the lower crust beneath the ridge axis [Géli et al., 1994] and seismic wave speeds in the axial lower crust and mantle are lower than away from the ridge axis [Klingelhöfer et al., 2000a]. These observations indicate that crustal and mantle temperatures at the ridge axis are high and support the idea that the axial lithosphere is not unduly thick.

Irrespective of whether deep hydrothermal cooling occurs, several models have been developed to show that at the slowest spreading rates a decrease in melting occurs at shallow depths as a result of conductive cooling of the sub-ridge mantle [e.g., Reid and Jackson, 1981; Bown and White, 1994; Niu and Hekinian, 1997]. In these models a thicker mantle lithosphere results in a shorter melting column beneath the ridge and thus an overall thinner crust than for faster spreading ridges. Such a process might explain the difference between the thickness of typical mid-ocean ridge crust at faster spreading rates (6–7 km) and the geochemically estimated thickness of the column of melt produced beneath the Mohns Ridge (∼5 km) [Klingelhöfer et al., 2000b]. However, there are some problems with this hypothesis. For example, there is a ∼1 km disagreement between the geochemical estimate of the amount of melt produced and the seismic estimate of the crustal thickness. Furthermore, this hypothesis ignores melt transport and how mantle melts would pass through such a thick axial lithosphere. It also ignores whether melts freeze within the conductively cooled mantle lid and whether the heat released would inhibit deepening of the lithospheric-asthenospheric boundary (i.e., work against a deep cooling process).

We suggest that while some inhibition of melting may occur beneath the Mohns Ridge as predicted by the conductive cooling argument, a large portion of the ‘missing’ melt is frozen into the uppermost mantle (Figure 7) forming the observed ALVIN region with its shear wave velocities that are intermediate between mantle and crust. Assuming a gabbro velocity of 3.8 km/s and a mantle velocity of 4.5 km/s, ∼35% gabbroic material in the mantle would explain the velocities within the ALVIN region. This estimate is relatively high but is not precluded by the earlier seismic refraction [Klingelhöfer et al., 2000a] and geochemical [Klingelhöfer et al., 2000b] analyses, given their large uncertainties.

Figure 7.

Cartoon depicting our preferred interpretation of the Mohns Ridge mantle seismic structure. Lines with arrows indicate mantle flow lines. A slightly thickened axial lithosphere inhibits the migration of melt toward the surface. As a result, melt freezes into the top ∼7 km of the mantle resulting in shear wave speeds that are intermediate between crustal and mantle values. The cartoon depicts little to no melt below ∼60 km depth below the crust as is suggested by comparing the Mohns Ridge seismic results to the F&J temperature-only model in Figure 6. However, it is possible to make various shear wave predictions using the Faul and Jackson [2005] method and the possibility of trace amounts of melt below 60 km depth cannot be ruled out.

In addition to the Mohns Ridge, anomalously low sub-crustal seismic velocities have been documented for the Gakkel Ridge [Jokat et al., 2003] and a part of the western North Atlantic with very slow paleospreading rates [Lizarralde et al., 2004]; generally low lithospheric velocities have been documented for the Reykjanes Ridge as well [Gaherty and Dunn, 2007]. These studies indicate that ALVIN-like regions may be a common feature of slow to ultraslow spreading regimes. The explanation for these ALVIN-like regions is debated with some favoring serpentinization of the upper mantle [Klingelhöfer et al., 2000a; Jokat et al., 2003] and others favoring melt retention in the upper mantle [Lizarralde et al., 2004; Gaherty and Dunn, 2007]. Although given the available seismic data these two scenarios are seismically indistinguishable, we favor the melt retention interpretation for the following reasons: (1) geochemical models predict greater melt production than what is determined by the thickness of the crust [Klingelhöfer et al., 2000b]. Perhaps this excess melt is trapped in the mantle. (2) The deep penetration of hydrothermal cooling would result in an unusually thick lithosphere, which is not supported by the Mohns seismic model. (3) At very slow spreading rates the brittle-ductile transition is expected to be in the mantle, rather than in the crust, and steady state magma chambers are not likely to be present [Phipps Morgan and Chen, 1993]. This means that melts ascending from the mantle may become trapped beneath a sub-crustal permeability barrier and it seems likely that at least some of these melts would freeze into the uppermost mantle. (4) The pressure limitations on the depth of fluid penetration [Fisher, 1998] and volume expansion associated with hydrated olivine effectively seals fractures, restricting deep fluid flow into the mantle [Wilcock and Delaney, 1996; Schroeder et al., 2002]. Since faulting in this extending tectonic setting may re-open sealed fractures, a combination of some mantle alteration and ‘crustal emplacement’ to explain the presence of the ALVIN region cannot be ruled out.

A common interpretation of radial anisotropy is the alignment of mantle olivine due to sub-ridge flow. Six percent radial anisotropy is similar to, if not higher than, that found in other studies of the upper oceanic mantle near ridges spreading at slow to fast rates [e.g., Nishimura and Forsyth, 1989; Gaherty and Dunn, 2007]. On the other hand, a previous P wave study of azimuthal anisotropy in the Atlantic suggests that in ultraslow spreading environments the alignment of olivine is lower in the upper 10 km of the mantle [Gaherty et al., 2004] than is expected for faster spreading ridges. For the Mohns Ridge, the anisotropy could be due to strong olivine alignment, in disagreement with the Atlantic study, or could result from the layering of ultramafic and gabbroic material in the upper mantle. The layering could form by a number of mechanisms: gabbroic melt lenses or films that subsequently freeze into the lithosphere; the horizontally shearing of randomly oriented gabbroic bodies that freeze into the lithosphere; or by the development of melt-rich shear bands at low angles to the shear plane [Holtzman et al., 2003] that subsequently freeze into the lithosphere. A combination of weak olivine alignment and “layered” gabbroic bodies in the upper mantle would be consistent with both weaker azimuthal anisotropy, as predicted by the Atlantic study, and relatively stronger radial anisotropy, as observed here. The presence of gabbroic layering in the uppermost mantle is purely speculative, but does fit with our assertion that the upper several kilometers of the mantle contains a significant degree of solidified gabbroic material.

7. Conclusions

Mantle shear wave velocities near the Mohns ridge indicate the presence of a lithosphere whose thickness is similar to that predicted by a simple ridge cooling model. Low asthenospheric velocities above ∼60 km depth are consistent with small melt concentrations (≤2%), presumably produced by pressure-release melting in the ridge upwelling zone. At greater distances from the ridge (greater lithospheric ages), the asthenospheric anomaly fades and the lithosphere thickens, here the mantle seismic profile closely follows the prediction of a half-space cooling model.

A sub-crustal low-velocity lithospheric layer forms near the Mohns ridge and is present to at least ∼20 Ma. Shear wave velocities within this ∼7 km thick layer are intermediate between gabbro and peridotite and suggest either hydrothermal penetration and mantle alteration or a mantle impregnated by trapped, frozen basaltic melts. Since we find no evidence of an anomalously thick lithosphere that would be created by deep hydrothermal penetration and previous geochemical studies indicate that more melt is produced than is indicated by the thickness of the crust, we prefer the melt entrapment model.

Our results, taken together with results from previous studies of the Mohns area, suggest a scenario in which very slow spreading at the ridge results in a slightly thickened (<10 km thick) lithosphere at the ridge axis. This lithospheric lid may have some effect on decreasing the height of the melting column, but more importantly inhibits some melts from ascending out of the mantle. The trapped melts produce a zone <10 km thick of mixed gabbroic and ultramafic material. The crust produced at the ridge axis is thereby thinner than would be expected for the amount of melt generated in the mantle. If this process occurs to some degree at all slow to very slow spreading ridges, then there may be spatially large reservoirs of unaccounted for gabbroic melt frozen into the mantle throughout many oceanic basins, with profound implications for global-scale geochemical budgets.


All maps are produced using GMT software [Wessel and Smith, 1998]. All JMI and JMIC data and instrument responses come courtesy of NORSAR ( Initial data processing was performed using the software package SAC [Goldstein, 1996]. Thanks to Eric Hellebrand, Garrett Ito, Rolf Mjelde, and John Sinton for very helpful conversations and to Fred Duennebier, Cecily Wolfe, and John Sinton for their helpful comments regarding this paper. Don Forsyth and an anonymous referee provided constructive and helpful reviews. This research was supported by NSF grants OCE0648507 and OCE0426428.