Volume 18, Issue 8 p. 2906-2929
Research Article
Free Access

Mantle dynamics beneath the discrete and diffuse plate boundaries of the Juan de Fuca plate: Results from Cascadia Initiative body wave tomography

Joseph S. Byrnes

Corresponding Author

Joseph S. Byrnes

Department of Earth Sciences, University of Oregon, Eugene, Oregon, USA

Correspondence to: J. S. Byrnes, [email protected]Search for more papers by this author
Douglas R. Toomey

Douglas R. Toomey

Department of Earth Sciences, University of Oregon, Eugene, Oregon, USA

Search for more papers by this author
Emilie E. E. Hooft

Emilie E. E. Hooft

Department of Earth Sciences, University of Oregon, Eugene, Oregon, USA

Search for more papers by this author
John Nábělek

John Nábělek

College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, Oregon, USA

Search for more papers by this author
Jochen Braunmiller

Jochen Braunmiller

School of Geosciences, University of South Florida, Tampa, Florida, USA

Search for more papers by this author
First published: 01 July 2017
Citations: 21
This article was corrected on 31 MAY 2019. See the end of the full text for details.


We use the delay times of teleseismic S phases recorded by ocean bottom seismometers during the plate-scale Cascadia Initiative community experiment to constrain the heterogeneity of seismic velocity structure beneath young oceanic lithosphere. Our study area covers the entire Juan de Fuca (JdF) and Gorda plates, from their creation at the JdF and Gorda Ridges to their subduction beneath the North American continent, and the entire length of the Blanco transform fault. The range of the observed Vs anomalies requires variations in the melt fraction of the asthenosphere. The data require that low Vs anomalies extend to depths of at least 200 km, which is within the carbonatite melting regime. In the upper 200 km of the mantle, Vs increases rapidly to the east of the JdF Ridge, while there is no clear relationship with the age of the lithosphere in the Gorda region. The distribution of melt is asymmetric about both the JdF and Gorda Ridges. Dynamic upwelling – due to the buoyancy of the mantle – and accompanying downwelling can explain the rapid decrease in melt fraction to the east of the JdF Ridge, the asymmetry about the JdF Ridge, and the sinuous pattern of upwelling near the Blanco transform fault. Finally, mantle flow beneath the diffuse Gorda and Explorer plate boundaries is distinct from that beneath the discrete plate boundary of the JdF Ridge. In particular, shear between the Pacific and JdF plates appears to dominate mantle deformation over seafloor spreading beneath the Gorda Ridge.

Key Points

  • Tomography reveals larger variations in seismic velocity beneath the Juan de Fuca and Gorda plates than predicted for lithospheric cooling
  • Dynamic upwelling with off-axis downwelling best explains the mantle structure beneath the Juan de Fuca Ridge
  • Distinct patterns of mantle flow are inferred beneath discrete and diffuse plate boundaries

1 Introduction

The Cascadia Initiative community experiment [Toomey et al., 2014] provides a novel opportunity to investigate mantle structure beneath an entire oceanic plate, from its formation at the Juan de Fuca (JdF) and Gorda Ridges to its subduction beneath North America. While the Cascadia Initiative study area is relatively compact, within its aperture there are all types of plate boundaries, including spreading centers, transform faults, a convergent margin capable of generating large earthquakes, and diffuse plate boundaries to the north and south (Figure 1a). In addition, decades of interdisciplinary studies indicate that the JdF Ridge interacts with hotspots of the northeastern Pacific. A primary motivation for the Cascadia Initiative was to better understand the regional scale structures that can affect processes occurring near the Cascadia subduction zone, which poses a significant hazard to the Pacific Northwest.

Details are in the caption following the image

(a) Major tectonic features in the study area. Black lines are transform faults, double lines are spreading centers, and the line with triangles is the Cascadia megathrust. Bold arrows indicate the direction of absolute plate motion, and thin arrows indicate the direction of ridge migration. Numbers next to the bold arrows are estimates of the velocity of absolute plate motion in mm/yr [DeMets et al., 2010]. Streaked areas indicate diffuse plate boundaries. Tectonic features of interest are marked by capital letters; starting near 48°N and moving south along the ridge-transform-ridge system, A marks the Endeavour Segment, B the Cobb Offset, C Axial Seamount, D the Cascadia Depression, and E the Escanaba Segment. The scale bar in the bottom-left approximately shows 200 km for the Mercator projection at these latitudes. (b) Age of the seafloor for the JdF, Gorda, and Pacific plates [Wilson, 1993].

Mantle processes beneath young oceanic plates, particularly those that are bounded by such diverse, discrete and diffuse tectonic boundaries, are not well constrained by geophysical observations. Factors that can influence mantle convection, melt generation, and internal deformation of young oceanic lithosphere include the motion of the overlying lithosphere, the viscosity of the asthenosphere, and the heterogeneity of the upwelling mantle. A key unknown is whether mantle upwelling occurs as a passive response to the spreading of the plates [e.g., Phipps Morgan and Forsyth, 1988], or if dynamic upwelling occurs due to the buoyancy of the upwelling mantle [Scott and Stevenson, 1989; Buck and Su, 1989; Katz; 2010]. A range of observations from different parts of the globe [e.g., Lin et al., 1990; Spiegelman and Reynolds, 1999; Hung et al., 2000; Wang et al., 2009; Key et al., 2013] has yet to resolve the pattern of mantle convection beneath spreading centers.

Here, we invert the delay times of teleseismic S waves for velocity anomalies in the upper mantle. We use data recorded by ocean bottom seismometers (OBSs) that were deployed across the JdF and Gorda plates during the Cascadia Initiative [Toomey et al., 2014] and Blanco transform experiments [Ghorbani et al., 2015]. The observed range of Vs anomalies requires significant variations in asthenospheric melt content. We infer that dry melting occurs within 100 km of the JdF Ridge, but is more broadly distributed around the Gorda Ridge. Deeper, volatile-induced melting extends to at least 200 km depth and around the Blanco transform fault. Models of dynamic upwelling better explain both the asymmetric distribution and large lateral gradients of the retained melt fraction near the JdF Ridge than models of purely passive upwelling. Dynamic upwelling also provides an explanation for the sinuous pattern of upwelling we observe near the Blanco transform fault. In contrast, we do not observe evidence for dynamic upwelling beneath the diffuse plate boundaries of the Gorda and Explorer regions. Results from both seismic tomography and SKS splitting from the Gorda deformation zone show a strong association with the deformation of the Gorda plate and a weak association with the Gorda Ridge. These results suggest distinct patterns of mantle flow beneath discrete and diffuse plate boundaries.

2 Geological Setting

Seafloor spreading in the northeast Pacific occurs along three spreading centers, known as the Explorer, JdF, and Gorda Ridges, which are offset by the Sovanco and Blanco transform faults (Figure 1a). The full spreading rate is 56 mm/yr [Wilson, 1993] except along the southern half of the Gorda Ridge, where the spreading rate decreases to ∼10 mm/yr [Riddihough, 1984]. The ridge system migrates to the northwest at 25 mm/yr due to the higher velocity of the Pacific plate in a hotspot reference frame [Small and Danyushevsky, 2003]. Both the Explorer and Gorda plates are undergoing significant internal deformation and reorganization [Wilson, 1986; Wilson, 1989; Braunmiller and Nábělek, 2002; Chaytor et al., 2004; Dziak, 2006]. The recently developed Nootka transform fault separates the Explorer and JdF plates [Hyndman et al., 1979; Riddihough, 1984], whereas the boundary between the JdF and Gorda plates is diffuse. With this in mind, we refer throughout this paper to the plates subducting beneath North America north and south of the Blanco fracture zone as the JdF and Gorda plates and the regions where crust was created at the JdF and Gorda Ridges as the JdF and Gorda regions, respectively. We do not have seismic data from the Explorer plate; however, we do have coverage of a diffuse deformational region that lies on the Pacific plate south of the Sovanco transform and north of the Cobb Offset along the northern JdF Ridge.

Several observations suggest significant differences in the mantle structure beneath the JdF and Gorda Ridges, even along segments with similar spreading rates. The JdF Ridge is characterized by an axial high, while the Gorda Ridge is characterized by an axial valley [Hooft and Detrick, 1995]. This suggests that either the temperature or the supply of melt is greater beneath the JdF than Gorda Ridge [Chen and Morgan, 1990]. Results from SKS splitting indicate that mantle flow beneath the JdF plate is to first approximation driven by absolute plate motion, whereas mantle deformation beneath the Gorda region is driven by the relative motion of the JdF and Pacific plates with no detectable contribution from the overriding Gorda plate [Bodmer et al., 2015; Martin-Short et al., 2015]. Results from surface wave tomography show low Vs anomalies beneath the JdF Ridge that are likely due to the retention of melt [Tian et al., 2013] and that are asymmetric about the ridge axis [Bell et al., 2016]; in contrast, weaker variations in seismic velocity are observed beneath the Gorda Ridge [Bell et al., 2016]. Each of these results suggests different mantle structures and processes beneath the intact JdF and the internally-deforming Gorda plates.

The Blanco transform fault formed ∼6 Ma [Embley and Wilson, 1992; Wilson, 1993] and has lengthened to 350 km. Seismicity occurs along the entire length of the fault [Braunmiller and Nábělek, 2008], and is segmented by several pull-apart basins and one active intra-transform spreading center (Figure 1a). Active seafloor spreading likely occurs at the largest of these, the Cascadia Depression [DeCharon, 1989; Braunmiller and Nábělek, 2008]. Along the northwestern section of the Blanco transform fault, from the Cascadia Depression to the JdF Ridge, large, negative residual mantle Bouguer anomalies are consistent with thickened crust [Gregg et al., 2007]. The major element composition of basalts erupted along this section of the fault suggests that some melting occurs beneath the Pacific plate south of the JdF Ridge [Gaetani et al., 1995].

Relatively little is known about how the aging of the lithosphere affects the JdF and Gorda plates. Before being subducted beneath the North American continent, the JdF and Gorda plates reach ages of 8–10 Myr and 4 to 7 Myr, respectively [Wilson, 1993]. The age of neither plate corresponds directly to distance from the ridge because of offsets in age at propagator wakes and the internal deformation of the Gorda plate (Figure 1b). The age of crust that may have been created at intra-transform spreading centers is also not known. Surface wave tomography results show that seismic velocity in the upper mantle appears to increase rapidly on the eastern flank of the JdF Ridge [Tian et al., 2013; Bell et al., 2016], while an age progressive pattern is less clear for the Gorda plate [Bell et al., 2016]. Onshore seismic studies have identified a seismic discontinuity beneath the recently subducted JdF and Gorda plates at depths near 50 km [Kumar and Kawakatsu, 2011] and 25 km [Liu et al., 2012] below the oceanic crust, respectively, likely consistent with a seismic structure that is controlled in part by the aging of the oceanic lithosphere [Kumar and Kawakatsu, 2011]. Seafloor sediments are generally thicker where oceanic crust is older and near the subduction zone, and thicknesses range from near zero to approximately 2 km [Divins, 2003; Ruan et al., 2014; Bell et al., 2015; Han et al., 2016].

3 Data and Methods

3.1 Seismic Experiments

We analyzed teleseismic S phases recorded by instruments deployed during the Cascadia Initiative [Toomey et al., 2014] and the Blanco transform experiment [Ghorbani et al., 2015] (Figure 2). We used data from both the onshore and offshore components of the Cascadia Initiative when measuring the delay times, since signal-to-noise ratios are generally higher in the onshore data, and used data from only the offshore component during the inversions. Stations deployed on the forearc were not used due to high ambient noise levels. Delay times were measured with 29, 28, and 31 OBSs from the first, second, and third years of the Cascadia Initiative, which were deployed from 26 July 2011 to 22 July 2012, 14 July 2012 to 10 August 2013, and 2 August 2013 to 7 July 2014, respectively. Stations deployed during the first and third years were predominately in the northern half of the study area (blue triangles in Figure 2) and all but three stations from the third year reoccupied sites from the first year. Stations deployed during the second year (red triangles in Figure 2) were predominately in the southern half of the study area. Triangles which are half red and half blue in Figure 2 mark sites that were reoccupied during at least one northern and southern leg of the Cascadia Initiative and are necessary to connect relative travel time measurements from different deployments. The Blanco transform experiment occurred from 18 September 2012 to 5 October 2013, which largely overlaps with the second year of the Cascadia Initiative. We used data from 30 of these broadband OBSs (yellow triangles in Figure 2).

Details are in the caption following the image

(a) Map of the study area and the locations of the OBSs used. Blue and red triangles show stations deployed during the northern and southern legs of the Cascadia Initiative, respectively. Reoccupied sites that tie the northern and southern deployments together are shown by half red and half blue triangles. Yellow triangles show broadband OBSs deployed during the Blanco transform experiment. Background colors show bathymetry in kilometers below sea level. Thick and thin contours show 1 and 2.5 km below sea level, respectively.

3.2 Measurement of Delay Times

We measured the relative delay times of teleseismic S and sS phases on the transverse channel. Before measuring delay times, all seismograms were corrected for instrument response, rotated into the radial-transverse coordinate system, and filtered. Transfer functions describing the instrument response for all stations were provided by the IRIS DMC and deconvolved from each seismogram, using the method of Haney et al. [2012]. The orientations of the horizontal components for OBSs from the Cascadia Initiative were determined by Sumy et al. [2015]. Data from all stations were resampled to a common sampling rate of 40 samples per second. Finally, we applied a third-order bandpass filter with corner periods of 12 and 33 s to each seismogram. Signal-to-noise ratios for all arrivals observed on the horizontal components were sensitive to the short-period limit of the bandpass filter, which we attribute to noise from microseisms [Webb, 1998]. Examples of aligned seismograms processed in this manner are shown in Figure 3.

Details are in the caption following the image

Example seismograms for teleseismic S phases recorded on the transverse component of OBSs. The vertical, dashed lines show the part of the seismogram used for cross-correlation. (a) Seismograms of a magnitude 7.0 event in Japan, recorded during the first year of the Cascadia Initiative. (b) Seismograms of a magnitude 6.6 event offshore of New Guinea recorded during the second year of the Cascadia Initiative and the Blanco transform experiment.

We measured the delay times with the method of VanDecar and Crosson [1990]. Arrival times were first predicted with the IASP91 velocity model [Kennett and Engdahl, 1991] and after cross-correlation the delay times were demeaned. This method preserves the relative arrival time of a phase and therefore is only sensitive to relative variations in seismic velocity. We measured the sS phase for one event with estimated depth of 150 km. The minimum standard error of the delay times is assumed to be 0.25 s, which is discussed in supporting information section S1. By convention, a positive (negative) delay time represents a delayed (advanced) arrival with respect to the mean for an event.

Variations in the thickness of seafloor sediments across the study area contribute to the delay times of teleseismic arrivals. Before imaging mantle structure, we corrected the delay times for sediment thickness variations using two methods. The first method, described by Ruan et al. [2014], measures the thickness and mean Vs of a sediment layer from the ratio of vertical and pressure displacements of Rayleigh waves, from which one-way travel times may be calculated. We used corrections found with this method for 43 unique sites by Ruan et al. [2014] and Bell et al. [2015]. Sediment corrections for reoccupied sites agreed with in 0.05 s between deployments [Bell et al., 2015] and we used the corrections measured for the first station deployed at a given site. Stations deployed on exposed basement were assigned a sediment thickness of zero. For OBSs where the first method could not be used, we interpolated sediment thickness from a global compilation provided by the National Geophysical Data Center [Divins, 2003], which have been smoothed laterally over a minimum distance of 10 km. We then calculated the travel time through the sedimentary layer using the mean Vs profile of the regional sediments calculated by Ruan et al. [2014]. For the 43 sites where both methods could be used, the mean difference in the correction is 0.13 s, with a standard deviation of 0.46 s, which is less than the range of measured delay times and the sediment corrections (see section 4 and Table 2).

3.3 Tomographic Method

The tomographic method used to image upper mantle structure is described in Hammond and Toomey [2003] and Bezada et al. [2013]. Within the model domain, seismic ray paths and travel times are calculated with Dijkstra's algorithm [Dijkstra, 1959; Moser, 1991]; outside the model domain travel times are calculated through a 1-D model using the tau-p method [Crotwell et al., 1999]. The inverse problem is solved by simultaneously minimizing the Euclidean norm of the prediction error, the model perturbational vector relative to the starting model, and the roughness of the perturbational model (for details, see Toomey et al. [1994]). The method also solves for event and station statics, both with units of seconds. The event statics account for the differing distribution of stations that record each event and for unknown structure outside of our image volume. The station statics account for uncertainties in the shallow structure beneath a station or station timing errors. The station statics are damped during the inversion to avoid removing signal from mantle structure (see supporting information section 3). The partial derivatives of a travel time with respect to the model parameters are calculated by fitting approximations of “banana-doughnut” kernels [Schmandt and Humphreys, 2010a] around the ray paths calculated within the model domain [Bezada et al., 2013]. This approach approximates the sensitivity of finite-frequency delay times measured by cross-correlation to seismic structure located near a ray path [Dahlen et al., 2000]. Iterations of the forward and inverse problem are repeated until improvements in data misfit are negligible.

For the results presented below, we adopted the following inversion parameters, which are summarized in Table 1. For seismic ray tracing, the velocity model was gridded at 10 km intervals. The differences between the travel times calculated by our ray tracing approach and by the tau-p method [Crotwell et al., 1999] have a standard deviation of 0.12 s, considerably less than the relative travel time errors. We solved for perturbations to the starting model every 25 and 50 km in the horizontal and vertical directions, respectively. The bounds of the model are 500 km to the east and west of 128°W and 800 km north and south of 45°N. Nodes that are masked in the plots of the velocity models have values of the derivative-weight-sum (DWS) less than 10, indicating no ray coverage [Toomey and Foulger, 1989; Toomey et al., 1994]. Weights for the penalty, vertical smoothness, and horizontal smoothness were set to 1, 75, and 150, respectively (for details, see Toomey et al. [1994]). We chose these parameters by comparing the recovery of anomalies and the misfit to the data in many different inversions of synthetic and actual data. All results shown were obtained by repeating the forward and inverse problems over 6 iterations.

Table 1. Summary of the Inversion Parameters
Parameter Value
Ray tracing grid spacing 10 km
Standard deviation of ray tracing errors 0.12 s
Perturbational grid vertical spacing 50 km
Perturbational grid horizontal spacing 25 km
Iterations of the forward and inverse problem 6
Bounds of the model domain 37.8°N-52.2°N and 121.6°W-134.4°W
Penalty, vertical smoothing, and horizontal smoothing weights 1, 75, 150
RMS of the station terms for the preferred model 0.22 s
RMS of the misfit to the delay times 0.43 s
Table 2. Data Statistics
Parameter Value
Number of arrivals 1,343
Number of events 108
Picking error 0.25 s
RMS of the measured arrivals 0.95 s (1.82 s with event terms applied)
Bandpass filter limits 12 to 33 s
RMS of the sediment corrections 0.65 s
RMS of the sediment-corrected arrivals 1.42 s (2.1 s with event terms applied)

4 Results

4.1 Teleseismic Delay Times

The measured delay times are summarized in Figure 5 and Table 2. We measured 1,343 delays times of teleseismic S and sS waves on the transverse channel from 108 events at epicentral distances between 30 and 100° (Figure 4a). We measured an average of 12 delay times per event on OBSs. The azimuthal distribution of the delays is dominated by events from Japan, South America, and Tonga (Figure 4b). After measuring the delay times, eight stations from the Cascadia Initiative were removed from our analysis because their timing base is suspect. To identify these sites, we used results from this study and an ongoing analysis of teleseismic P wave data (M. Bodmer, personal communication). The latter provides a larger data sample and thus better statistical estimates. Suspect stations were identified by either an average delay time after shallow corrections at each station that fell outside the 2.5 sigma range for both onshore and offshore sites in the Pacific Northwest, or a difference in the mean S wave delay time at reoccupied sites that exceeded 0.8 s.

Details are in the caption following the image

(a) Locations of the 108 events used in this study. (b) Rose diagram of the azimuth of station-event pairs used in this study. Bins are 18° wide and begin at 0°.

Figure 5a shows the mean delay by station after applying corrections for sediment thickness and event statics; these delays are averaged at reoccupied sites; see supporting information section 2 for mean delays by deployment. Figure 5b shows the mean station delay without a correction for sediment thickness. Both data sets have been demeaned. We did not make corrections for variations in crustal thickness; typical variations are likely less than 0.5 km [Carbotte et al., 2008; Han et al., 2016] and would contribute a delay time of at most 0.01 s, which is insignificant in comparison with the uncertainty of a measurement (0.25 s). For the entire set of delays, the RMS and range is 0.95 s and −2.3 s to 3.6 s, respectively, without sediment corrections, and these values increase to 1.42 s and −3.2 s to 4.3 s when sediment corrections are applied. For comparison, the RMS of S wave delay times measured across the Western United States is 1.18 s [Schmandt and Humphreys, 2010b]. The sediment corrections are shown in Figure 5c and vary between 0 and 2 s. This range is approximately one third of the range of the measured delay times. The corrections for sediment thickness generally increase with lithospheric age and are greatest near the continental margin and within the JdF plate. Corrections for sediment thickness within the Gorda region are systematically less than for the JdF region and do not systematically increase with the age of the crust. These patterns are expected because of the sediment loads of the rivers in the Northwestern United States. Figure 5d shows the standard deviation of the delays in seconds measured at each station. This is not the uncertainty of the observations, instead it provides an indication of the heterogeneity of seismic structure beneath a station. The standard deviation at a given station is much smaller than variations of the mean delays between stations, which indicates that upper mantle structure primarily varies at depths shallow enough for rays from different azimuths to sample similar structure. We note that if we were to solve for undamped station statics during the inversion, a large portion of the signal related to upper mantle heterogeneity would be removed (supporting information Figure S9); this underlies the importance of making sediment corrections.

Details are in the caption following the image

Summary of the delay times. (a) Mean delay times in seconds recorded at each station after applying event statics and sediment corrections. This is the set of delays used to construct the preferred model. (b) Mean delay times in seconds recorded at each station after applying only the event statics. The data sets in Figure 5a and 5b have been separately demeaned. (c) One-way travel times through seafloor sediments at each station in seconds, which were subtracted from the measured delay times for Figure 5a. Circles are stations where the method of Ruan et al. [2014] was used, squares are stations where interpolation was used. (d) The standard deviation of the delay times in seconds recorded at each station (not the uncertainty of the observations).

Geographic patterns in the mean station delays are apparent in Figures 5a and 5b. In the JdF region, the mean delays are more positive (delayed arrival) at stations near the spreading center and become more negative (advanced arrival) where the plate is older; this pattern indicates an increase in upper mantle velocity with plate age. We note that this pattern is more pronounced in the set of delay times that include the sediment corrections. The mean delay times are also asymmetric about the JdF Ridge whether or not the sediment corrections are applied, with more delayed arrivals for sites on the Pacific plate, except toward the southern JdF Ridge where it intersects the Blanco transform fault. In the Gorda region, no clear relation between the mean delay times and the age of the crust is observed in either Figures 5a or 5b; this suggests a weaker relationship between plate age and upper mantle Vs. In the region of the Blanco and Mendocino transform faults, the mean station delays vary abruptly near the ridge-transform intersections (RTI), with more positive delays where the crust is younger. We do not observe a region of more positive delays near the Cascadia Depression. Curiously, the most negative delays are on the oldest sections of the JdF plate, despite the presence of older lithosphere south of both the JdF-Blanco RTI and the Mendocino Transform fault (Figure 1b); this is true for delays with and without sediment thickness corrections.

4.2 Tomographic Model

Our preferred tomographic results are shown in Figures 6 and 7. To construct our preferred model, we inverted sediment corrected delay times for mantle structure and damped station terms. We prefer this approach because the sediment corrections account for known structure that cannot be resolved by teleseismic delays, and the damped station statics account for uncertainties in sediment corrections, instrumental clock errors, or the seismic velocity of the oceanic crust. After conducting trial inversions, we chose to damp the stations statics to an RMS of 0.22 s (supporting information Figure S5); when larger station terms are allowed, the statics develop long-wavelength trends that are due to mantle structure (see supporting information). The supporting information also shows tomographic results for delay times without sediment corrections, the structure introduced by sediment corrections, differing inversion parameters, and the effects of including OBSs with suspect timing.

Details are in the caption following the image

Horizontal cross sections through the tomographic model at (a) 50 km depth, (b) 100 km depth, (c) 200 km depth, (d) the average over 0 to 200 km depth, (e) 300 km depth, and (f) 400 km depth. The images are masked where there is no ray path coverage.

Details are in the caption following the image

Vertical cross sections through the tomographic model. The images are masked where there is no ray path coverage. The locations of select features are indicated at the top of the plots. (a) Cross section across the northern JdF plate. (b) Cross section across the southern JdF plate. (c) Cross section along the Blanco transform. (d) Cross section across the Gorda Ridge. (e) Cross section along the JdF Ridge. The map in the upper right shows the locations of the cross sections.

Figure 6 shows that in the upper 200 km significant variations in shear wave velocities are observed throughout the study area. Overall, anomalies vary from approximately −5% to 5% (Figures 6a–6d), comparable to the total variation in the upper 200 km beneath the entire United States [Schmandt and Lin, 2014]. Without sediment corrections, the range of velocity anomalies is less, −3% to 3% (supporting information Figure S7), but still considerable. At 200 km depth, the overall magnitude of the velocity anomalies decreases to −2.5% to 2.5%. The largest negative and positive velocity perturbations in the region are located near the two ridges and the oldest portions of the JdF plate, respectively (Figures 6a–6d). However, significant variations in velocity are observed beneath areas with similarly aged lithosphere, with generally lower velocities beneath the southern half of the JdF plate, relative to the north (Figures 6a–6d, 7a, and 7b). By 200 km depth, velocities are higher beneath the central region of the JdF Ridge (north of Axial Seamount) than elsewhere along the ridge, with the possible exception of a high velocity anomaly that is at the edge of the study area to the north (Figure 6c). To the east of the Cascadia megathrust, Vs begins to decrease; however, this occurs at the edge of the study area and is not well constrained by our results.

The velocity anomalies beneath the JdF Ridge are not uniform or symmetric. Where we have good resolution, anomalously low velocities are observed farther to the west of the JdF Ridge than to the east, indicating an asymmetric anomaly beneath the spreading center. This is particularly evident north of Axial Seamount (∼46°N); for example, Figure 8 shows that Vs at ∼4 Myr old crust is up to 3% lower on the western flanks of the JdF Ridge than to the east, which is large in comparison to an asymmetry of 1–2% in the upper 200 km of the mantle near the southern East Pacific Rise [Hammond and Toomey, 2003]. South of Axial Seamount anomalously low velocities extend farther to the east of the ridge axis than to the north (Figures 6a, 6b, 7a, and 7b), particularly near the JdF-Blanco RTI. On the Pacific side of the JdF Ridge south of Axial seamount, Vs does not appear to be as asymmetric as to the north; however, due to a lack of instruments on the Pacific plate, we cannot fully assess the degree of asymmetry in this region. Using surface wave data from Cascadia Initiative stations, Bell et al. [2016] also reported an asymmetric low-velocity anomaly in this region. For the results of Bell et al. [2016], the degree of asymmetry about the southern JdF Ridge is reduced if the lateral smoothing constraint is relaxed and the velocity anomalies then appear more similar to the results presented here (S. Bell and D. Forsyth, personal communication, 2017). Along the northern JdF Ridge — including the region of the Endeavour segment and south of the Sovanco transform — the amplitude of the low-velocity anomaly decreases and again extends farther to the east of the ridge axis than beneath the central section of the JdF Ridge. This region is associated with a diffuse plate boundary [Dziak, 2006] near the southern limits of the Explorer plate (Figure 1a).

Details are in the caption following the image

(a) Shear-wave velocity anomalies as a function of the age of the lithosphere to the east and west of the JdF Ridge (blue), the Gorda Ridge (orange), and the East Pacific Rise (yellow) averaged over the upper 200 km. Each profile was set to 0% at an age of 0 Myr. The vertical bars show the standard deviation, not the uncertainty, of the velocity anomaly at a given age. (b) Map showing the regions where velocities were taken from for the JdF (blue) and Gorda (orange) Ridges.

Near the Gorda Ridge, Vs is generally low, and does not increase with distance from the spreading center as clearly as near the JdF Ridge (Figures 6a–6d). Velocities are asymmetric beneath the Gorda Ridge (Figure 8), with lower velocities imaged beneath the Pacific plate, but in contrast to the JdF Ridge the degree of asymmetry is at most 1%, and is only apparent in the sediment-corrected results (supporting information Figure S7). We observe a north to south trend in velocities that correlates with the southward decrease in spreading rate along the Gorda Ridge. Bell et al. [2016] also observed less asymmetry beneath the Gorda than JdF Ridge and a north to south trend in Vs, though at depths shallower than 55 km they observed higher Vs beneath the Gorda than JdF Ridge. Velocities are low in our model within the whole region bounded by the Blanco and Mendocino transform faults east of 129°W and increase rapidly across the transform faults. We do not observe a change in Vs in the northeastern corner of the Gorda region, where Bodmer et al. [2015] inferred that mantle flow changes from being driven by absolute plate motion to the relative motion of the JdF and Pacific plates.

We observe distinct patterns of velocity anomalies along the strike of the Blanco transform fault. In the upper 200 km, Vs along the transform fault is average or slightly low for the region. Velocities are lowest near the JdF Ridge, higher near the Gorda Ridge, and highest beneath the center of the transform fault and the Cascadia Depression. Lateral variations in velocity are pronounced near each RTI in the upper 200 km of the model, with velocities increasing in the ridge parallel direction toward the older lithosphere. At each RTI, the velocities beneath the older lithosphere are lower than those observed beneath similarly aged lithosphere of the JdF plate. Relative to the direction of ridge migration, two RTIs (JdF-Blanco and Gorda-Mendocino) are leading and one RTI (Gorda-Blanco) is trailing [Carbotte et al., 2004]. The velocity anomalies imaged near the RTIs do not clearly correlate with the direction of ridge migration.

4.3 Model Resolution

The rows of the resolution matrix [Backus and Gilbert, 1968; Jackson, 1972] and inversions of synthetic data demonstrates the resolving capabilities of our study. Because the number of model parameters exceeds the number of independent data, individual parameters cannot be uniquely resolved. Instead, weighted averages of model parameters are resolved, with the weights given by the resolution matrix. Since our inversion is regularized with off-diagonal elements of a covariance matrix (for details, see Toomey et al. [1994]), we use the equations of Vasco and Johnson [2003] to calculate the resolution matrix after singular value decomposition with the partial-derivative matrix from the final iteration. This approach allows us to identify how well our inversion can resolve an individual model parameter.

Figure 9 shows examples of averaging kernels for our study. Horizontal resolution (Figures 9a–9d) was evaluated by normalizing the off-diagonal elements of the resolution matrix by the corresponding diagonal elements and contouring the normalized values at 0.5 and 0.25. Vertical resolution (Figures 9e–9h) was estimated by averaging the values of the resolution matrix at different depths within the region where the normalized, horizontal resolution was greater than 0.25, and renormalizing the resulting profile. At 100 km depth, the lateral dimension of the averaging kernel varies from 70 to 150 km (Figure 9a), and increases to 100 to 200 km at 300 km depth (Figure 9b). Model parameters are not independent within the upper 200 km (Figures 9e–9h), and model parameters at 300 km depth depend on those from approximately 200–400 km depth (Figure 9f). Horizontal resolution is to first approximation limited by station spacing.

Details are in the caption following the image

Examples of averaging kernels. (top row: a, b, c, and d). Averaging kernels for chosen model parameters in map view. The inner and outer contours show the 0.5 and 0.25 levels, respectively, for kernels normalized to 1. Black circles show the location of the model parameter. (a) Three parameters at 100 km depth throughout the study area. (b) Three model parameters at the same location as in Figure 9a but at 300 km depth. (c) Three parameters near ridge-transform intersections at 100 km depth. (d) Parameters on each side of the JdF and Gorda Ridges at 100 km depth. (bottom row: e, f, g, and h) Line plots of averaging kernels against depth for the model parameter with the matching color in the map directly above. See text for discussion.

Figure 10 shows an inversion of synthetic data predicted for a model with two layers of regularly spaced anomalies at 50–100 and 250–300 km depth; inversion parameters are the same as the preferred model. The sign of the anomalies changes every 100 and 150 km in the upper and lower layer, respectively. In both depth intervals, the inversion recovers lateral variations well (Figures 10a–10c) and overestimates the depth extent of the anomalies (Figure 10g). When the recovered anomalies are averaged over the upper 200 km of the model space, the amplitudes better approximate the input values (Figures 10c and 10d). The supporting information shows results for the same pattern of anomalies with even station and event coverage to demonstrate that resolution is primarily limited by the available ray paths.

Details are in the caption following the image

Inversion of a synthetic data set generated for a model with a regular pattern of anomalies. The recovered model is shown at (a) 100 km depth, (b) 300 km depth, (c) averaged over the upper 200 km, and along longitude 128°W (g). Anomalies from the starting model are shown for the same regions in Figures 10d, 10e, 10f, and 10h.

We use the resolution matrix and results of synthetic inversions to ascertain what features of our model may be reliably interpreted. Figures 9c and 9d show lateral averaging kernels near the 3 RTIs and to either side of the JdF and Gorda Ridges, respectively. For the RTIs, model parameters beneath the older plate are independent of all model parameters beneath the ridges, confirming that the low velocities observed in these regions are resolved independently from ridge structure. Where the spreading centers are encompassed by OBSs, Figure 9d shows that structure to either side of the ridge is independently resolved; this demonstrates that the asymmetric structure imaged near the JdF and Gorda Ridges is robust. Resolution tests also indicate that sharp, vertical contrasts in Vs will be averaged laterally, producing gradients with a full width of approximately 75 km. We thus estimate that the region of low Vs beneath the JdF Ridge (Figures 6 and 7) could be as narrow as 100 km wide if the actual transition to higher Vs is a sharp, vertical boundary (supporting information section 6).

The data require that structure extends to depths of at least 200 km. For a model domain that is 500 km deep (Figures 6 and 7), the final RMS misfit of the data is 0.43 s. When anomalies are restricted to depths above 100, 200, or 300 km, the RMS misfit increases to 0.51 s, 0.46 s, and 0.44 s, respectively. After converting the misfits to a χ2 distribution, we used an F-test to find if the degraded fits to the data by the squeezed models are statistically significant. We find p-values for the extension of the model domain below 100, 200, and 300 km depth of 0.01, 0.15, and 0.25, respectively, and conclude that structure must extend to depths of at least 200 km. At 300 and 400 km depth, the largest velocity anomalies are generally near the edges of the model domain and are of low amplitude. While our inversion can theoretically recover velocity anomalies at these depths (Figure 10), the details of the velocity structure below 200 km are poorly constrained by our results.

5 Discussion

Our tomographic results provide new insights into variations in oceanic upper mantle structure with plate age and near tectonic boundaries, including spreading centers, ridge-transform intersections, transform faults, and the Cascadia subduction zone. We use our results to infer physical properties of the upper mantle and then discuss the implications for geodynamic processes.

5.1 Physical Properties

Figure 11a shows velocity anomalies by plate age for three spreading centers — the JdF Ridge, Gorda Ridge, and East Pacific Rise (EPR) — as well as predictions for half-space cooling models. The EPR result is from Hammond and Toomey [2003, Figure 12b]. For each spreading center the velocity anomalies are averaged about the ridge axis to remove asymmetries. The velocity anomalies are averaged vertically over the upper 200 km and at 1 Myr intervals, using bin widths of 1 Myr; ages for the JdF and Gorda plates were estimated by linearly interpolating the isochron data of Wilson [1993] (Figure 1b). For each curve the velocity anomaly at 0 Myr is set to zero so that we can assess age-dependent variations. The temperatures of the half-space cooling model were calculated assuming a mantle potential temperature of 1623 K and a seafloor temperature of 273 K [Turcotte and Schubert, 2002; Stixrude and Bertelloni, 2005; Harmon et al., 2009].

Details are in the caption following the image

Trends of Vs anomalies averaged over the upper 200 km against the age of the lithosphere. (a) Plots of velocity against age for the JdF region (blue), Gorda region (orange), the East Pacific Rise (yellow), and two half-space cooling models (black and dashed). Each profile was set to 0% at an age of 0 Myr. Vertical bars show the standard deviation, not the uncertainty, of the velocity anomalies at a given age. (b) Map showing the regions where the velocities were taken from in Figure 11a for the JdF (blue) and Gorda (orange) region.

Details are in the caption following the image

Contours of depth-averaged variations in melt fractions in the upper 200 km (in %). The regions masked are the same as in Figure 6d; in addition, the regions south of the Mendocino transform fault and east of the Cascadia megathrust are masked. The calculation and associated uncertainties are described in the text.

The results for both the JdF Ridge and the EPR show systematic increases in upper mantle velocities with plate age that are too large to be explained by variations in temperature due to lithospheric cooling (Figure 11a). Moreover, the velocities beneath the JdF plate continue to rise rapidly to 4 to 6 Myr, as opposed to leveling off between 2 to 4 Myr near the EPR. Beneath the JdF plate, the increase in seismic velocity with age is greater than that observed in larger scale studies, which typically have poorer lateral resolution [Nishimura and Forsyth, 1989; Nettles and Dziewonski, 2008; James et al., 2014]. For the Gorda region the age dependence of velocities is less apparent; the Gorda anomalies are discussed separately below.

Since the cooling of the lithosphere alone cannot explain our results, we considered the effects of conductive cooling on the depth to a seismic G discontinuity. The G discontinuity is a generic term for a decrease in seismic velocity with increasing depth beneath oceanic lithosphere, and variations in the depth to such a discontinuity will impart a teleseismic delay time. While an age-dependent discontinuity may not exist beneath young oceanic lithosphere, one could be caused by the presence of melt [e.g., Kawakatsu et al., 2009; Holtzman, 2016] or by a thermally controlled anelastic process [Karato, 2012], and such a discontinuity has been previously proposed to exist beneath the JdF plate [Kumar and Kawakatsu, 2011]. We assume an ad hoc 8% decrease in velocity at depths greater than the 1100 K isotherm [Rychert and Shearer, 2011] of the half-space cooling model. This hypothesis mimics a G discontinuity at the base of the thermal lithosphere, as suggested by onshore studies of the recently subducted JdF [Kumar and Kawakatsu, 2011] and Gorda [Liu et al., 2012] plates. Including such a discontinuity increases the cumulative age-dependent increase in velocity to 3% for the upper 200 km, which is still significantly less than the observed increase in Vs for the JdF plate. Moreover, Vs increases most rapidly beneath 3 to 5 Myr old lithosphere, while both half-space cooling models predict that Vs should increase at a decreasing rate as the lithosphere ages. On the basis of these comparisons we conclude that thermal evolution of oceanic lithosphere or its effects on the G discontinuity cannot explain mantle structure in the JdF region.

Our tomographic results require age-dependent variations in the physical properties of the asthenosphere. We discuss the likely effects of temperature, anelasticity, and melt fraction and conclude that the most likely explanation is age-dependent variations in the amount of partial melt. We first examine if only variations in temperature can explain the observed velocity anomalies. Assuming that variations in temperature occur between the 1100 K isotherm and 200 km depth, a 300 K contrast in temperature is required to explain the variations in Vs beneath the JdF plate, assuming a Qs of 50 [Karato, 1993]. If the anomalies are confined to shallower depths, then the required variations in temperature will be higher. As the required temperature anomaly is greater than the range of temperature in all MORB and OIB sources [e.g., Putirka, 2005], we conclude that other mechanisms are required to explain our results.

A solid-state anelastic process is also unlikely to explain our results. We estimate the required variations in Qs to explain our results with the equations of Anderson and Given [1982]. We assume that Qs reaches a maximum beneath the JdF plate where we observe the highest Vs, and then calculate the required reduction in Qs to explain the observed velocity anomalies. For maximum values of Qs of 50, 100, and 200, Qs beneath the JdF Ridge would need to reach minimum values of 13, 14, and 15, respectively. These values are lower than the inferred value of 25 beneath the JdF Ridge, and are significantly lower than the range of Qs values that can be explained by variations in temperature, grain size, or water content [Eilon and Abers, 2017].

We infer that variations in the melt fraction of the asthenosphere are required to explain our observations, a result consistent with those of previous seismic studies [Tian et al, 2013; Bell et al., 2016, Eilon and Abers, 2017]. To estimate lateral variations in melt fraction we first remove the predicted effects of the half-space cooling model and then assume the following: melt is uniformly distributed between the 1100 K isotherm and 200 km depth; 1% of retained melt reduces Vs by 8% [Hammond and Humphreys, 2000a]; and melt has a negligible effect on seismic attenuation [Hammond and Humphreys, 2000b]. Figure 12 shows a map of lateral variations in the melt fraction (as a percentage) of the asthenosphere in the upper 200 km that are consistent with these assumptions. If melt is retained in high aspect-ratio bands [Holtzman et al., 2003; Kawakatsu et al., 2009] or if melt influences Qs [Faul et al., 2004; McCarthy and Takei, 2011; Abers et al., 2014; Holtzman, 2016, Eilon and Abers, 2017], then smaller variations in melt fraction are required by our results; however, if melt is retained in triple-junction tubules at low melt fractions [Hammond and Humphreys 2000a] then larger variations are required. We also note that because we used an average of velocities in the upper 200 km of the mantle, greater melt fractions in narrower depth intervals are possible and our estimates are a minimum bound on the retained melt fraction. For example, in regions of mantle upwelling we expect relatively higher melt fractions above the anhydrous solidus due to increased productivity.

Figure 12 shows that the inferred, depth-averaged asthenospheric melt fraction varies throughout the study area. We attribute the higher average melt fractions near the JdF and Gorda Ridges to regions of increased melt production and retention. Beneath the Blanco transform fault and its bounding RTIs, the inferred asthenospheric melt fractions are generally less than that beneath the spreading centers, implying lower rates of melt production. Beneath the older JdF plate, melt fractions are less than near the ridges but not uniform, with the smallest melt fractions inferred between 45°N and 47°N. Given the uncertainties in our observations and imaging, regions where melt fraction is less than 0.1% may be melt free. Our results do not constrain structure east of the Cascadia deformation front (Figures 10a–10c), thus we cannot determine if anomalously high melt fractions are present beneath the subducting JdF slab [Hawley et al., 2016]. However, beneath the older Juan de Fuca plate, we question the existence a sub-slab melt anomaly east of deformation front in view of the low asthenospheric melt fractions.

Our tomographic results are consistent with relatively low, but nonzero, melt fractions at depths near 200 km. A 3% contrast in Vs (Figure 6c), for example, requires unrealistically large temperature anomalies of 130 and 260 K for Qs of 50 and 200, respectively [Karato, 1993]. The Vs anomalies may indicate the presence of carbonatitic melts [Dasgupta and Hirschmann, 2006; Dasgupta et al., 2013], with retained melt fractions on the order of ∼0.1% or less [Hirschmann, 2010]. We consider two possible distributions of melt near 200 km depth. First, a 3% change in Vs could correspond to a ∼0.3% contrast in melt fraction if melt is distributed in elongate inclusions at grain boundaries [Faul et al., 1994; Hammond and Humphreys, 2000a]. However, smaller melt fractions are thought to be present within the carbonatitic melting regime [Hirschmann, 2010]. A second possibility is that the deep melt fractions are as small as 0.01% and are distributed in a connected network of thin films along grain boundaries, which may reduce Vs relative to melt-free rock by ∼5% [Holtzman, 2016]. Thus, the deep Vs anomalies either reflect the presence of variable fractions of melt, or regions of melt-free and melt-bearing rock. Seismic and conductivity anomalies beneath the EPR and Galápagos Archipelago have been attributed to deep melting [Hammond and Toomey, 2003; Key et al., 2013; Villagómez et al., 2014], suggesting that carbonatitic melts are commonplace in regions of mantle upwelling near 200 km depth.

5.2 Geodynamic Processes

Here we discuss how variations in upper mantle melt fraction can occur in our study area and their relations to geodynamic processes. To emphasize relations between our tomographic results, estimates of melt distribution, and tectonic features, Figure 13 shows the absolute value of the gradient of the Vs anomalies in the upper 200 km of the mantle; a value larger than 0.03%/km (bold contour in Figure 13) outlines regions of large lateral velocity gradients. We expect the largest and most rapid variations to occur within the dry melting regime because melt productivity is an order of magnitude lower below than above the dry solidus [Hirth and Kohlstedt, 1996] and because of the likelihood that mantle upwelling is more broadly distributed at deeper depths. Processes that can cause lateral variations in melt content include melt production during mantle upwelling, melt freezing during mantle downwelling, and variable rates of melt transport. In the following sections, we discuss how such processes can give rise to the inferred variations in melt fractions within the different tectonic settings of our study area. We do not consider melt freezing at the base of the thermal lithosphere, since the deepening of a G discontinuity cannot explain either the range or lateral gradients of the velocity anomalies.

Details are in the caption following the image

Magnitude of the gradient of the Vs anomalies shown in Figure 6d. The bold lines show the 0.03%/km contour. The same regions are masked as in Figure 6b. The calculation is described in the text.

5.2.1 Juan de Fuca Ridge and Plate

Figure 13 shows that near the JdF Ridge, the largest lateral gradients are located east of the spreading center. We attribute the largest gradients in velocity to melt freezing due to mantle downwelling. Our reasoning is as follows: if only melt production and transport rates varied — without freezing — then melt fraction would decrease gradually with distance from a spreading center since melt fractions below ∼1% do not readily segregate from the mantle [Spiegelman, 1996; Faul, 2001]. In this scenario, abrupt gradients in velocity structure would not develop. We note that both the location and magnitude of the gradient is not uniform near the JdF Ridge. The largest gradients are closer to the spreading center to the north than to the south of Axial Seamount (∼46°N), consistent with the westward offset of Vs anomalies north of Axial Seamount. Farther to the north, and near the Endeavour segment (∼48°N), the gradients east of the ridge axis are both lower in magnitude and farther from the ridge axis, which suggests that downwelling is not present; synthetic tests confirm that we would detect a large gradient in this region if it were present (supporting information Figure S14).

In Figure 14, we illustrate a first-approximation, two-dimensional scenario for mantle flow and melting beneath the central JdF Ridge that is consistent with our tomographic results, geodynamic predictions, and previous work in the region. Characteristics of this simplified scenario include asymmetric mantle flow due to northwest migration of the JdF Ridge, asymmetric melting that extends farther west of the spreading center due to an entrained thermal anomaly, and mantle downwelling with melt freezing beneath the eastern flank. Geodynamic studies predict that ridge migration alone does not give rise to significant asymmetric melting [Toomey et al., 2002; Conder et al., 2002; Katz, 2010]. However, a ridge migrating over a thermal anomaly does result in asymmetric melting [Toomey et al., 2002; Conder et al., 2002], particularly if there is a component of dynamic upwelling, which enhances mantle downwelling and melt freezing beneath the trailing flank. Assuming that the temperature of the melt is near the pressure dependent solidus, the melt fraction will reach zero with less than 10 km of downwelling [Katz, 2010]. Dynamic upwelling and off-axis downwelling are triggered when the viscosity of the mantle is low enough for buoyancy forces—due to the composition, temperature, or retained melt fraction of the mantle—to contribute to the focusing of mantle flow beneath the ridge axis [Scott and Stevenson, 1989; Jha et al., 1994; Katz, 2010].

Details are in the caption following the image

Scenario for upwelling and melting beneath the JdF Ridge. Melt fractions are qualitatively shown in shades of orange, the lithosphere in blue, streamlines by black lines with arrows, the edge of station coverage by the vertical dashed line, and approximate depths are given on the left. The base of the melt-bearing region deepens to the west due to an entrained thermal anomaly. See text for discussion.

We attribute the primary characteristics of mantle structure near the JdF Ridge to a modest thermal anomaly beneath the Pacific plate, in conjunction with ridge migration and dynamic upwelling. A likely source of this thermal anomaly are the hotspots in the northeastern Pacific; for example, the Cobb-Eickelburg hotspot is estimated to have an excess temperature of at least 30 to 40 K [Rhodes et al., 1990; Hooft and Detrick, 1995]. We note that if buoyancy forces do not contribute to mantle flow, then either a thermal anomaly on the order of 100 K or pressure driven flow in the asthenosphere is required to explain the mantle velocity structure [e.g., Toomey et al., 2002; Conder et al., 2002]; we consider such scenarios for this region unlikely.

Our preferred interpretation for the mantle structure of the central and southern JdF Ridge is that the thermal anomaly responsible for the asymmetry is preferentially located north of Axial Seamount and west of the ridge axis. To support this view, in Figure 15 we show the difference between Vs over the upper 200 km along transects 75 km east of and directly beneath the JdF Ridge. The eastward increase in Vs (Figure 13) begins closest to the ridge axis near 47°N and farthest from the ridge axis south of 45°N. This is consistent with the largest westward offset in the mantle low-velocity anomaly occurring near 47°N and suggests that thermally-induced dynamic upwelling is stronger in this region. While we image a more symmetric pattern of anomalies about the southern than central ridge axis, only one OBS was deployed west of the southern JdF Ridge. However, the asymmetry of mantle Bouguer anomalies (MBA) — which constrains mantle density anomalies since it accounts for variations in seismically measured crustal thickness — from three ridge-perpendicular transects are consistent with our interpretation (dashed lines and text in Figure 15 [Marjanović et al., 2011]). For example, mass deficits in the mantle are most asymmetric at 47°N, with lower densities to the west of the ridge (positive ΔMBA in Figure 15), and are more ridge-centered at 45°N with some eastward offset (negative ΔMBA). Intriguingly, 3He/4He ratios vary with the asymmetry of mantle structure (blue circles in Figure 15 [Lupton et al., 1993]). This trend, which negatively correlates with 87Sr/86Sr, is consistent with the presence of a modest thermal anomaly where 3He/4He ratios are higher [Graham et al., 2001, 2014]. One issue with our interpretation is that the crust is moderately thicker along the southern JdF Ridge [Carbotte et al., 2008], and it is commonly assumed that crustal thickness increases with increasing mantle temperature. In some geodynamic models, however, crustal thickness decreases as the strength of dynamic upwelling increases and the melt-producing region narrows [Katz, 2010]. Our results may be consistent with this prediction, since we observe both a wider and stronger low velocity zone beneath the southern JdF Ridge.

Details are in the caption following the image

Comparison between Vs anomalies (orange), the asymmetry of mantle Bouguer anomalies (ΔMBA, black, dashed lines and text, from Marjanović et al. [2011]), and helium ratios (blue, from Lupton et al. [1993]) along the JdF Ridge. The orange line shows the difference between Vs anomalies along two mantle transects, one running 75 km east of the JdF Ridge, the other directly on the ridge; both vertically average Vs over the top 200 km of the model and positive values indicate increasing Vs to the east of the ridge axis. ΔMBA is the difference between gravity anomalies 50 km east and west of the JdF Ridge from three ridge-perpendicular transects; positive values indicate a greater mass deficit to the west of the ridge.

Farther north along the JdF Ridge, and beneath the Explorer deformation zone, we infer that mantle upwelling is relatively broad and occurs at a reduced rate. This region is characterized by a decreased magnitude of the low-velocity anomaly (Figure 6), which we attribute to lower melt fractions (Figure 12), and Vs gradients that are both lower in magnitude and farther off axis than to the south (Figure 13). This suggests slower rates of mantle upwelling and that mantle downwelling is weaker or not occurring to the east of the spreading center. Moreover, the helium ratios and asymmetry in gravity anomalies about the ridge axis also decrease in this region (Figure 15). These trends to reduced anomalies north of the Cobb Offset all occur near a diffuse plate boundary (Figure 1) associated with the Explorer plate and the region south of the Sovanco transform fault [Dziak, 2006]. This region of the JdF Ridge has undergone significant deformation and reorganization in the last 3.5 Myr [Riddihough, 1984; Wilson et al., 1984; Karsten et al., 1986; Braunmiller and Nábělek, 2002]. As deformation has progressed, basalts of more variable and enriched compositions are being erupted along the Endeavour segment that may originate from a small degree of melting of an enriched component that reaches the surface without mixing with more depleted melts during ascent [Karsten et al., 1990; Goldstein et al., 1991; Cousens et al., 1995; Sours-Page et al., 1999]. Based on these observations, we infer that beneath the diffuse plate boundary the rate of mantle upwelling and the in situ melt fractions have decreased, and the region of mantle upwelling has broadened as mantle downwelling waned.

Our results are consistent with a region of nearly melt-free asthenosphere beneath the older portions of the JdF plate. As noted above, to explain the abrupt gradients in velocity we infer that melt freezing has occurred within the primary region of melt production due to mantle downwelling. Hence, the asthenosphere nearest to the Cascadia subduction zone at latitudes between approximately 45°N and 48°N may be largely melt free. While our results only constrain relative variations in velocity, Bell et al. [2016] likewise inferred a nearly melt-free asthenosphere in this region from surface wave tomography. We do not resolve structure east of the Cascadia megathrust (Figure 10), thus we cannot constrain if melt fractions are higher beneath the subducting plate [e.g., Hawley et al., 2016].

5.2.2 Ridge-Transform-Ridge Plate Boundaries

We infer that mantle upwelling and melt production occurs along the full length of Blanco transform fault in a broad and sinuous pattern. The relatively low velocities imaged throughout the transform domain (Figures 6a-6d) and the inferred melt fractions (Figure 12) support this view. Our interpretation is also consistent with the inference from petrology that some melt is produced beneath the Pacific plate near Blanco transform fault [Gaetani et al., 1995]. We further note that the lowest Vs anomalies near the JdF-Blanco RTI do not follow the ridge-transform-ridge geometry observed on the surface. Instead, low-velocity anomalies elongate along the strike of the Blanco transform fault. For example, anomalies near the southern JdF Ridge extend toward the east at shallow depths (Figures 6a and 6b) and the discord between geometry of the plate boundaries and the low Vs anomalies becomes more pronounced near 200 km depth (Figure 6c).

The sinuous character of the velocity structure beneath the Blanco transform is consistent with predictions of dynamic models of mantle flow. Passive flow models predict that upwelling and melt production will be centered beneath the spreading centers with a gradual reduction in upwelling toward an RTI and continued upwelling beyond the ridge end. The distribution of upwelling at mantle depths in passive flow models largely maintains the geometry of the ridge-transform-ridge system [Phipps Morgan and Forsyth, 1988, Figure 6]. In contrast, for dynamic flow models the mantle flow displays a smoother geometry than the surface offsets, resulting in a more sinuous pattern that is asymmetric about the ridge axis and weaker within approximately 100 km of the RTI [Sparks et al., 1993, Figure 4; Magde et al., 1997]. In concert with our interpretation for the asymmetry and Vs gradients of the JdF Ridge, we attribute the sinuous pattern of anomalies beneath the Blanco transform fault to the influence of dynamic upwelling beneath the ridge-transform-ridge system.

5.2.3 Gorda Ridge and Plate

The seismic structure of mantle beneath the Gorda Ridge and plate shows a strong association with the Gorda deformation zone that rivals that of seafloor spreading processes. South of the Blanco transform fault, variations in melt fractions are generally less than 0.2% (Figure 12) and appear unrelated to the Gorda Ridge or lithospheric age (Figure 11), which is in contrast to the structure beneath the JdF Ridge and plate. Instead, the clearest trend is an abrupt decrease in Vs from the central Gorda Ridge toward the south that terminates at the Escanaba Trough (Figures 6a–6d). Results from surface wave tomography for the same region show a decrease in Vs toward the south at 55 km depth [Bell et al., 2016]. However, this trend is not apparent in the surface wave results at shallower depths, suggesting that melt may be primarily retained at deeper depths than beneath the JdF Ridge. Results from SKS splitting are consistent with the view that seafloor spreading processes are being overwhelmed by diffuse deformation. Fast polarization directions become more closely aligned with the relative Pacific-JdF plate motion from north to south and, like the Vs anomalies, do not vary with distance from the ridge axis [Bodmer et al., 2015]. Therefore, seismic velocity and anisotropy appear to be better related to the deformation of the southern Gorda plate than with spreading at the Gorda Ridge. We also do not observe evidence for mantle downwelling (Figure 13) and instead observe a continuous distribution of melt at longitudes nearest to the Cascadia subduction zone. Tomographic studies using onshore data also observe lower Vs anomalies underneath the slab south of 43°N than farther north [e.g., Schmandt and Humphreys, 2010a; Xue and Allen, 2010; James et al., 2011], which is consistent with the retention of melt farther east than our results can constrain (Figure 10).

We conclude that the dynamics of mantle flow beneath the diffuse plate boundaries of the Gorda and Explorer deformation zones are fundamentally different in comparison with that beneath the JdF Ridge and plate. The age-independence of seismic velocity and anisotropy across the Gorda Ridge and their association with the deformation of the Gorda plate suggests that the shear zone between the JdF and Pacific plates dominates mantle deformation over seafloor spreading. Similarly, the Endeavour segment of the JdF Ridge, at the southern edge of the Explorer deformation zone, also features gradual variations in Vs anomalies that are more similar to the tomographic results for the Gorda Ridge than to the large gradients along the JdF Ridge south of the Cobb Offset. Plate boundary reorganization around both the Gorda and Explorer Ridges results in a broad region of intraplate deformation, volcanism, and seismicity [Wilson, 1986, 1989; Chaytor et al. 2004; Karsten et al, 1986; Braunmiller and Nábělek, 2002; Dziak, 2006], and is associated with the eruption of increasingly enriched basalts derived from smaller degrees of melting [Karsten et al., 1990; Davis et al., 2008]. These contrasting observations between the JdF Ridge and plate, and the adjacent deformation zones, suggests distinct patterns of mantle flow beneath discrete and diffuse plate boundaries. Dynamic upwelling and mantle downwelling occur in response to spreading at the JdF Ridge south of the Cobb Offset (Figure 14), while forces associated with the deformation zones appear to disrupt or even dominate over the influence of seafloor spreading and mantle buoyancy beneath the diffuse plate boundaries.

6 Conclusions

The Cascadia Initiative and Blanco transform experiments provide a novel opportunity to study the structure of the upper mantle beneath an entire tectonic plate. In this study, we have tomographically imaged shear-wave velocity anomalies in the upper mantle beneath the Juan de Fuca, Gorda, and Pacific plates using the relative arrival times of teleseismic S waves recorded on the transverse components of OBSs. The range of observed Vs anomalies requires variations in the retained melt fraction of the asthenosphere. We attribute large Vs anomalies beneath the JdF and Gorda Ridges to melt production shallower than the dry solidus and smaller Vs anomalies across the transform faults and at least as deep as 200 km to volatile-induced melting. Three lines of evidence suggest that dynamic upwelling—driven by the buoyancy of the mantle — occurs beneath the JdF Ridge. First, large gradients in Vs to the east of the JdF Ridge require melt freezing due to downwelling, which is predicted to occur in tandem with dynamic upwelling. Second, the asymmetry of the low Vs region beneath the JdF Ridge is most likely due to the interaction of a modest thermal anomaly in the asthenosphere and dynamic upwelling. Third, the low Vs anomalies beneath the JdF-Blanco transform intersection display a smoother pattern than the surface offsets, as predicted by some models of dynamic upwelling. Distinct patterns of mantle flow are inferred beneath the diffuse plate boundaries of the Gorda and Explorer deformation zones from that beneath the discrete plate boundary of the JdF Ridge. We do not observe evidence for dynamic upwelling north of the Cobb Offset and beneath the Gorda Ridge, where the shear zone between the Pacific and JdF plates dominates mantle deformation over seafloor spreading.


We thank Z. Eilon and an anonymous reviewer for their helpful comments, which greatly improved both the content and presentation of this manuscript. Data used in this research were provided by instruments from the Ocean Bottom Seismograph Instrument Pool (http://www.obsip.org) which is funded by the National Science Foundation. OBSIP data are archived at the IRIS Data Management Center (http://www.iris.edu). The seismic data used in this paper are available from the Data Management Center under the network codes 7D and X9 for the Cascadia Initiative and the Blanco transform experiment, respectively. This research was supported by the National Science Foundation under grants OCE-1139701, OCE-1333196, and EAR-1520694 to the University of Oregon, and OCE-1031858 and OCE-1131767 to Oregon State University.


    In the originally published version of this article, Figure 1 was incorrect. The figure has since been corrected, and this version may be considered the authoritative version of record.