Seismic velocity measurements provide a useful tool for investigating spatial variations in uppermantle properties, such as temperature and anisotropy, with implications for melt supply and mantle heterogeneity (e.g. Lin and Phipps Morgan, 1992;Dunn et al., 2005). These measurements are relatively straightforward to obtain on the continents (e.g. Chulick and Mooney, 2002;Chulick et al., 2013). However, it remains challenging and expensive to measure upper-mantle seismic velocity in the deep ocean, due to its remote location and difficulties in deploying long-term instruments on the seafloor. Pn phases are rays that are critically refracted at the Moho and propagate along the top of the uppermost mantle (e.g. Linehan, 1940;Brandsdottir and Menke, 1997). At the Mid-Atlantic Ridge (MAR) from 10°N to 35°N, Pn arrivals from 48 individual ray paths were recorded with hydrophones, and used to investigate upper-mantle velocities, giving a mean velocity of 8.0 ± 0.1 km s -1 (Dziak et al., 2004). This velocity estimate was higher than that from nearby active source seismic experiments along the ridge axis (7.5-7.9 km s -1 ; Canales et al., 2000), probably due to the effects of younger and thinner oceanic lithosphere being sampled by the refraction profiles, and the effects of averaging velocities across all rays. Despite such advances, upper-mantle velocities in the deep oceans remain poorly constrained, and the potential for hydrophone-recorded Pn phases to resolve spatial variations in upper-mantle velocity has not yet been sufficiently tested.
Confidential manuscript submitted to Bulletin of the Seismological Society of America 3 Here, we use Pn arrivals from regional earthquakes to constrain upper-mantle velocity in the equatorial Atlantic Ocean. Arrivals were recorded by a combination of five moored hydrophones and a single seismograph station installed on the St. Peter and St. Paul islets, giving 152 ray paths that sample mantle conditions both on-and off-axis, and across the St. Paul transform system. Our study is coincident with several mantle velocity estimates from a wide-angle seismic experiment (Le Pichon et al., 1965), and hence has the opportunity to validate spatial variations in velocity revealed by groups of similar ray paths.
In the equatorial Atlantic Ocean (10ºN-5ºS and 34°W-21°W), the MAR is offset by some of the longest transform faults on Earth, including the Strakhov, St. Paul, and Romanche transforms (Figure 1). The St. Paul transform system consists of four transform faults and three intra-transform ridge segments that accommodate an offset of 630 km. The northwest transform fault is currently undergoing transpression, giving rise to the ~200 km-long and ~30 km-wide Atobá ridge (Maia et al., 2016), and also uplift of 1.5mm yr -1 at the St. Peter and St. Paul islets (Campos et al., 2010;Maia et al., 2016). Other transforms in the system do not host topographic highs or an island related to transpression, and hence presumably are not experiencing uplift. In the three intervening spreading segments, seafloor spreading is slow, at ~16 mm yr -1 average half rate (DeMets et al., 2010). Faulting plays an important role in crustal accretion, and seismicity rates are relatively high, providing a useful tool to investigate the properties of the crust and upper mantle, as well as deformation at long-offset strike-slip systems (e.g. Francis et al., 1978;Abercrombie and Ekstrom, 2001;de Melo and do Nascimento, 2018).
We analyzed Pn arrivals in waveform data recorded by a combination of five moored autonomous hydrophones and one land-based seismograph (Figure 2). The five autonomous hydrophone instruments were deployed during two separate experiments: stations EA2 and EA8 were part of the Equatorial Atlantic (EA) array (Smith et al., 2012). Data were recorded at 16-bit resolution and a sampling rate of 250 Hz; for further details on these hydrophone instruments see Fox et al. (2001). Hydrophones H2, H4, H5 were deployed during the COLd Mantle Exhumation and Intra-transform Accretion experiment (COLMEIA; Maia et al., 2014Maia et al., , 2016)), and recorded data at 24 bit-resolution with a sampling rate of 240 Hz; for further instrument details see D 'Eu et al.(2012). We also used waveform data recorded by a three-component broadband
Prior to manually picking Pn arrivals, we applied a 6-20 Hz Butterworth bandpass filter to the hydrophone data in order to suppress unwanted noise. A bandpass filter with range 4-12 Hz was applied prior to picking arrivals from the ASPSP seismograph, to suppress additional microseism noise due to its island location. Based upon origin time, events were manually associated with earthquakes in the International Seismological Center Bulletin (ISC), yielding hypocenter locations, origin times, and magnitudes ranging from 3.5 to 5.4 MW. Earthquakes mostly occur due to strike-slip faulting along the Strakhov, St. Paul, and Romanche transform faults, with additional events due to extension along the intervening spreading ridge segments (Figure 2a). Example arrivals from three events are shown in Figures 3 and4, highlighting the typical response to strike-slip and normal faulting earthquakes ranging in magnitude from 4.6 to 5.3 MW.
Typical Pn-arrivals are emergent, and have low signal-to-noise ratio (SNR; noted in Figures 3 and4), making pick identification challenging. Given the mixed nature of our network and often noisy arrivals, picks were made based on the onset of emergent energy combined with changes in SNR, waveform character and amplitude. The observation of linear move-out, consistent with upper mantle velocity, added confidence to our picks, since this moveout is evident across the hydrophone array stations due to wave propagation along the crust-mantle interface (see common-receiver plots in Supplementary Figures S1-S6). P-arrivals are easily distinguished from T-phase arrivals, which arrive much later than P-arrivals, are emergent in character, and are higher in amplitude than P-arrivals (see hydrophone H5 in Figure 4). The catalog of detected events is given in Table S1.
In order to further test whether the detected arrivals were Pn phases, we compared the observed travel times to those predicted by the global iasp91 velocity model (Kennett and Engdahl, 1991). For each source-receiver ray path, we calculated the predicted Pn arrival time using iasp91, with the addition of a station-dependent delay to account for the propagation time from seafloor to hydrophone. This delay (1.2-2.5 s, see Table 1) was estimated using the hydrophone mooring cable length at each station, and the local water sound velocity estimated from the Global Ocean Sound Confidential manuscript submitted to Bulletin of the Seismological Society of America 6 Speed Profile Library (Barlow, 2019). The predicted Pn arrival times differ from the observed Pn arrivals by 5-10 s (Figures 3 and4), a difference which arises since the iasp91 model contains a crustal layer that is much thicker (30 km) than that expected in the oceans (~6 km). Hence, the differences in observed and predicted Pn arrival time are probably dominated by this additional crustal layer thickness in the velocity model, plus earthquake location and origin time uncertainties. Although these differences are evident, the waveform character and linear move-out velocity give us confidence in our identification of these emergent phases as Pn arrivals.
ISC origin times were subtracted from the Pn arrival times to obtain travel times for each ray path (i.e. each event-station pair). We account for travel time in the oceanic crust by subtracting ray path distances and travel times for the portion of the path that travels through the crust, assuming that all events occurred at 10 km depth (ISC catalog), and that crustal thickness is uniformly 6.0 km with a crustal velocity of 6.5 km s -1 (Christeson et al., 2019). For each station, we then calculate the distance and travel time for the portion of the ray path that extends from an earthquake in the crust to the Moho, and back from the Moho to the receiver. Pn velocity is obtained by dividing the distance travelled in the mantle by the travel time in the mantle. Details of these corrections for each station are given in Table 1.
Our approach yieldeded 152 Pn velocity estimates from the catalog of 50 regional earthquakes (Figure 5). Although epicentral distances range from 32 km to ~1095 km, all 50 events were detected at nearly all available stations, implying that the detection threshold of the combined hydrophones and ASPSP station is at least MW 3.5. Since most stations were located either near to, or to the north of, the St. Paul fracture zone, our ray path coverage is more comprehensive in the northern part of the study area. Ray paths sampling upper-mantle velocities to the south of the St. Paul fracture zone are restricted to events detected by hydrophone EA8, and those originating from four earthquakes located at the eastern end of the Romanche transform fault (Figure 5).
The two most significant potential sources of error in our analysis are hypocenter locations of events in the ISC Catalog, and Pn arrival time picks. We estimated hypocenter location (and hence epicentral distance) error to be ± 10 km, based upon ISC catalog location and typical error in global earthquake location (Lohman and Simons, 2005;Weston et al., 2012). This hypocenter location error implicitly includes other uncertainties associated with ISC catalog locations, such as those caused by un-modeled three-dimensional velocity structure and picking errors, which result in trade-offs between origin time and location (Bondár and Storchak, 2011). Arrival time pick (and hence also travel time) errors were investigated by estimating SNR for each arrival via two methods, one using the amplitude ratio between peak signal and root mean square noise, and another via the ratio between the short time average amplitude and long time average amplitude (STA/LTA; Figure S7). We find that both SNR estimates are only weakly dependent on epicentral distance and magnitude, however we do observe station-dependent variations in the scatter in SNR.
We quantify this scatter in terms of the standard deviation of SNR of arrivals for a particular station (Figure S7e), which likely is due to persistent local noise sources. Hence we estimated arrival time pick error based on the emergent character of arrivals and the standard deviation of SNR, with station-dependent errors defined as ± 0.5 s for EA2 and EA8; ± 1.0 s for H2, H4 and H5; and ± 0.3 s for ASPSP.
The total uncertainty in our velocity estimate, 𝛿𝑣 , was estimated by assuming that epicentral distance, d, and travel time, t, have errors that are uncorrelated and random. This Confidential manuscript submitted to Bulletin of the Seismological Society of America 8 assumption is valid since we attribute the main source of travel time error to uncertainty in picking of Pn arrivals (which in turn depends on waveform character and noise level), and the distance error is most significantly affected by error in earthquake location from the ISC catalog, which is assumed to be constant and hence is independent from hydrophone Pn pick error. We formally propagate the errors in d and t, as follows
where δd is epicentral distance error, and δt is travel time error (e.g. Taylor, 1997).
Although receiver location uncertainty is negligible for the land station ASPSP (located with meter-scale accuracy via the Global Positioning System), there is potential location uncertainty for the moored hydrophones in our network. Moored hydrophone locations were obtained by acoustic triangulation between the mooring acoustic release and the deployment vessel soon after the moorings settled on the seafloor, within error of several meters. In order to account for the possibility of abnormally strong current motion, each instrument was fitted with a pressure and temperature logger below the floatation package, so that any significant hydrophone depth changes would be recorded (e.g. Fox et al., 2001). Significant depth changes were not detected during depolyments, and thus we assume that the hydrophone location was constant during data collection, and hence hydrophone location uncertainty is less than 10 m.
The resulting 152 Pn ray paths (Figure 5b) and travel times (Figure 6) indicates uppermantle velocities that vary considerably across the study area, with estimates ranging between 7.2
Confidential manuscript submitted to Bulletin of the Seismological Society of America 9 and 11.1 km s -1 , and uncertainties ranging from 0.1 to 1.9 km s -1 (Table S2). Variability in reduced travel time increases with epicentral distance (Figure 6), although SNR does not show a similar trend (Figure S7). Hence the epicentral distance-dependent scatter in reduced travel time is likely due to variations in the depth of ray penetration (which increases with epicentral distance), and not due to increasing pick uncertainty. At the center of the study area there appears to be a longitudinal variation in Pn velocity, with events originating near the St. Paul transform system, and sampling adjacent lithosphere, having higher velocities than those from the adjacent spreading centers (Figure 5a). The best constrained estimate for sub-axis, ridge-parallel mantle velocity comes from ray paths that sample the portion of the spreading axis between the Strakhov fracture zone and stations near the St. Paul fracture zone (H2, H5 and ASPSP). Here, Pn travel times consistently imply relatively low velocities, with a mean of 7.7 km s -1 . Slightly higher velocities ranging between 7.8 and 8.2 km s -1 are indicated by ray paths between hydrophone EA2 and the Strakhov fracture zone, oriented roughly parallel to a plate spreading flowline. Ray paths oriented southwest-northeast (azimuth ~060°), i.e. oblique to the spreading direction, between events on the St. Paul fracture zone and detected at hydrophone EA2, have some of the highest mantle velocities (between 7.6 and 8.5 km s -1 ) compared to other rays sampling areas unaffected by fracture zones. Velocity estimates in the vicinity of the St. Paul fracture zone itself (from transform faulting events detected by hydrophones H2, H4 and H5, and ASPSP) show considerable variation, ranging from 8.0 to 9.1 km s -1 and a mean of 8.4 km s -1 , and little apparent spatial consistency.
Among these events, we encountered one of the highest Pn velocities (9.0 km ± 0.
In general, rays originating from the St. Paul transform system have higher velocities than those originating from active spreading centers to the east and west (Figure 5a), probably due to cooler conditions at the Moho along the transform. Our estimates of upper-mantle Pn velocities broadly agree (within error) with Pn velocities from radially stratified velocity models such as PREM (Dziewonski and Anderson, 1981) and iasp91 (Figure 6; Kennett and Engdahl, 1991). Our
Pn velocity estimates are also consistent with mantle velocity estimates from a series of reversed wide-angle refraction seismic profiles (i.e. with multiple shot points giving overlapping coverage) collected in the equatorial Atlantic during R/V Atlantis cruise A180 (Figure 5b; Le Pichon et al., 1965). The modal difference in velocity between refraction profiles from Le Pichon et al. (1965) and all intersecting ray paths is 0.2 km s -1 (see histogram in Figure 5c), although our Pn velocity estimates are consistently lower than those reported by Le Pichon et al. (1965), with a maximum disagreement of 1.2 km s -1 . A mantle velocity of 8.30 km s -1 was reported along profile A180-48, which is 283 km-long, and crosses the eastern side of the St. Paul transform fault (near ~26.3°W), trending northeast-southwest (Figure 5b). This velocity is consistent with that inferred from Pn (Growe et al., 2019). The general agreement between upper-mantle velocities from the refraction profiles and our Pn arrivals validates our results, and implies that spatial trends observed in the study area are likely to be real.
Elsewhere along the MAR, between 10º to 40ºN, a mean upper-mantle velocity of 8.0 km ± 0.1 km s -1 was estimated using a similar method to this study with Pn arrivals detected by an array of autonomous hydrophones (Dziak et al., 2004). Ray paths used by Dziak et al., (2004) often crossed the ridge axis, spanned a series of fracture zones, and extended onto older crust, which may explain the close agreement in results. This result suggests that off-axis and on-axis upper mantle characteristics are similar in the northern and equatorial Atlantic Ocean.
Near the Oceanographer transform fault on the MAR (~35°N), a two-dimensional tomographic inversion of wide-angle seismic refraction data suggests velocities of 7.4-7.8 km s -1 (Canales et al., 2000;Hooft et al., 2000). These results agree within error with our estimates of Pn velocity from rays sampling on-axis upper-mantle to the north of the St. Paul transform fault (Figure 5b), which are typically 7.2-8.0 km s -1 .
Seismic velocities in the upper-mantle near to the ridge axis, i.e. in young lithosphere, are expected to be lower than in off-axis areas, due to upwelling of hot material (e.g. Turcotte and Schubert, 2002). Following the removal of minor gridding artifacts associated with fracture zone traces, we used a global crustal age model (Müller et al., 2008) to assign a mean crustal age along each ray path, for comparison with Pn velocity (Figure 7a).
Ray paths sampling lithosphere younger than 10 Myr show a wide range of velocities, with a mean of 7.9 km s -1 and standard deviation of 0.5 km s -1 . Twenty ray paths yield velocities less than 7.5 km s -1 . Pn velocities for ray paths sampling lithosphere older than 20 Myr are slightly higher, with a mean of 8.1 km s -1 and standard deviation of 0.3 km s -1 , while only two ray paths give velocities lower than 7.5 km s -1 (Figure 7a). Most rays cover a wide range of crustal ages, so this geometry, and our averaging approach, may smear the possible effects of lithospheric aging.
The lack of rays travelling exclusively via older lithosphere may also obscure any progressive trend between upper-mantle velocity and crustal age. However, the tendency toward the inclusion Confidential manuscript submitted to Bulletin of the Seismological Society of America 13 of lower velocities in younger crust (Figure 7a) reflects the expected variation with respect to the zone of axial upwelling.
Laboratory experiments have shown that the mantle can experience significant shear strain during corner flow at the ridge axis, leaving an anisotropic fabric in the lithospheric mantle as minerals (e.g. olivine) are aligned into a lattice preferred orientation (LPO; e.g. Zhang and Karato, 1995;Nicolas and Christensen, 2011). Anisotropy consistent with a LPO formed by twodimensional mantle flow has been measured at some locations in the oceanic upper mantle, in particular at the fast-spreading East Pacific Rise (e.g. Raitt et al., 1969;Lin et al., 2016), however the strength of anisotropy varies widely, and debate remains about its origins (e.g. Mark et al., 2019). Since isochrons in this region are fairly uniform (Figure 5), VPn anisotropy could be expected parallel to paleo-relative plate motion, although this assumption has been shown to not apply everywhere (VanderBeek and Toomey, 2017).
We investigated the dependence of mantle velocity with azimuth, and use epicentral distance as a proxy for depth of mantle penetration to group rays (Figure 7b). No discernable pattern is evident in rays grouped by epicentral distance, including those expected to sample deepest in the mantle with epicentral distances > 700 km (blue lines in Figure 7c). Removing rays with VPn error > 0.4 km s -1 also does not resolve any azimuthal dependence (Figure 7d), nor does separating rays by mean crustal age (Figures 7e and7f).
The apparent lack of such azimuthal dependence could be due to several reasons. First, azimuthal dependence may be too subtle to be resolved by our VPn estimates, given the uncertainties in hypocenter location and crustal thickness discussed above. Second, the slow Confidential manuscript submitted to Bulletin of the Seismological Society of America 14 spreading rate of the MAR (~32 mm yr -1 total rate; (DeMets et al., 2010)) may result in a thickened lithosphere that is dominantly cooled by conduction, thus inhibiting corner flow (e.g. Sleep, 1975).
As a result, deformation could be accommodated by faulting at depths of 5-10 km beneath the Moho, reducing the viscous strain in the mantle at these depths, and suppressing the anisotropy recorded in the mantle (e.g. Ribe, 1989). Observations of weaker or anomalous anisotropy elsewhere in the Atlantic Ocean are consistent with our findings (e.g. Gaherty et al., 2004;Dunn et al., 2005). Third, complex, three-dimensional upwelling patterns near the ridge axis could result in anisotropy on relatively short wavelengths (Lin and Phipps Morgan, 1992), which would be smeared along our relatively long ray paths, and hence not be resolved.
To explore the relationship between VPn and the thermal structure of the asthenospheric upper-mantle, we compared our velocity estimates with a global, vertically polarized shear speed model SL2013sv (Schaeffer and Lebedev, 2013). Our objective is to evaluate our observations of uppermost mantle properties in the context of deeper mantle properties. We do not aim to directly validate our VPn estimates via this comparison. This model was chosen because it is particularly sensitive to anomalies within the upper-mantle, and hence provides a window into the upper mantle structure directly beneath our Pn ray paths (Schaeffer and Lebedev, 2013). We extracted values of vertically polarized tomographic shear velocity anomaly (%dVs) at 100 km intervals along each ray path, from slices through the SL2013sv model at depths of 25, 50, 75 and 150 km. We then calculated the mean %dVs along each ray path, at each depth interval (Figure 8). At 25 and 50 km depths, the effects of the ridge axis are evident, with higher velocities associated with ray paths travelling off-axis (detected by EA2 and EA8), and hence not sampling the relatively low-velocity
Confidential manuscript submitted to Bulletin of the Seismological Society of America 15 axial region (Figures 8a and8b). This effect is less pronounced at 75 km depth (Figure 8c), and is not apparent at 150 km depth, which presumably reflects sub-plate velocities. The lack of correlation between SL2013sv and Pn velocities at 150 km suggests that our VPn estimates, sensitive to the velocity structure directly beneath the Moho, do not record deeper, larger-scale sub-plate (i.e. asthenospheric) processes and anomalies. Hence our observed VPn variability may instead arise due to local variations in melt supply, lithospheric thickness, or faulting.
We used a network of five autonomous hydrophones and a broadband seismograph to detect
Pn arrivals from regional earthquakes in the equatorial Atlantic Ocean over a period of ~12 months between 2012 and 2013. Our estimates of upper-mantle velocity from the travel times of 152 Pn arrivals broadly agree (mostly within 0.2 km s -1 ) with those from nearby seismic refraction experiments.
We find that the upper-mantle near the St. Paul transform system has consistently high velocities (>8 km s -1 ), compared to relatively low velocities (~7.5 km s -1 ) in the adjacent MAR spreading segments northwest of the transform. This spatial pattern is consistent with the notion that Pn ray paths sample lower velocity mantle near the ridge axis, and higher velocity material near transforms, which are generally cooler, despite the presence of intra-transform spreading segments. We do not resolve any dependence between VPn and azimuth, which could either be due to observational uncertainty, or due to the combined effects of thickened lithosphere and more complex mantle upwelling patterns under slow-spreading conditions. We also do not find any correlation between VPn and vertically polarized shear speed from the global SL2013sv model, indicating that our method is not sensitive to properties of the asthenosphere. et al., 1996;Gasperini et al., 1997;Maia et al., 2016).