Mapping Mantle Flows and Slab Anisotropy in the Cascadia Subduction Zone

The Cascadia margin is an unusual subduction zone characterized by the downdip movement of young and thin oceanic plates, where mantle flow and intraslab deformation are still unclear. Here we present new anisotropic tomography of the Cascadia subduction zone, in which the hexagonal symmetry axis of anisotropy is tilting rather than horizontal or vertical as assumed in previous studies of seismic anisotropy. Subduction‐induced entrained and toroidal flows under the Cascadia margin are discriminated well by the spatial relationship between tilting‐axis anisotropy and slab geometry. The obliquely entrained flow is trapped in a narrow zone (<100 km wide) above and below the subducting slab and reaches ∼200 km depth, which is surrounded by large‐scale sub‐horizontal toroidal flow. The intraslab anisotropy is trench‐normal above 80 km depth but changes to trench‐parallel at 100–400 km depths, which may reflect fossil anisotropy overprinted by deep deformation beneath the arc, or joint effect of serpentinization and hydrous faulting.


Introduction
Seismic anisotropy is one of the most powerful tools to detect mantle flow and lithospheric deformation because deformation yields lattice-preferred orientation of anisotropic minerals or shape-preferred orientation of isotropic minerals (Ismaıl & Mainprice, 1998;Karato et al., 2008).In the past decades, shear-wave splitting (SWS) measurements have provided abundant information on seismic anisotropy in subduction zones (e.g., Bodmer et al., 2015;Currie et al., 2004;Eakin et al., 2010;Long, 2016;Martin-Short et al., 2015;Zhao, Wang, et al., 2023).When penetrating an anisotropic medium, a shear wave splits into fast and slow polarized waves in orthogonal directions, and the fast wave through the subslab mantle exhibits a predominantly trench-parallel direction globally (Figure 1a) (Long, 2013).This phenomenon has been explained by several mechanisms, including trench-parallel 3-D mantle flow (Long & Silver, 2008), olivine fabric transition (Karato et al., 2008), and hydrated faults in the subducted slab (Faccenda et al., 2008).However, the subslab mantle in Cascadia is an exception because such a trench-parallel SKS splitting pattern has not been reported in the whole Cascadia subduction zone yet (Figure 1c) (Long, 2016), which makes some aspects of Cascadia subduction dynamics elusive.
Near the Cascadia margin, the anisotropic pattern beneath the Pacific Northwest has been mapped by using several geophysical techniques including SWS measurements (e.g., Bodmer et al., 2015;Currie et al., 2004;Eakin et al., 2010;Long, 2016;Martin-Short et al., 2015), body wave tomography (Buehler & Shearer, 2010;Huang & Zhao, 2013;Zhao & Hua, 2021), surface wave tomography (Gosselin et al., 2020;Wagner et al., 2013), Eikonal tomography derived from ambient noise (Lin et al., 2011), full-wave inversion (Krischer et al., 2018;Rodgers et al., 2022;Zhou et al., 2022;Zhu et al., 2020), and joint inversion of SKS and surface waveforms (Yuan et al., 2014;Yuan & Romanowicz, 2010), which have greatly improved our understanding of subduction Abstract The Cascadia margin is an unusual subduction zone characterized by the downdip movement of young and thin oceanic plates, where mantle flow and intraslab deformation are still unclear.Here we present new anisotropic tomography of the Cascadia subduction zone, in which the hexagonal symmetry axis of anisotropy is tilting rather than horizontal or vertical as assumed in previous studies of seismic anisotropy.Subduction-induced entrained and toroidal flows under the Cascadia margin are discriminated well by the spatial relationship between tilting-axis anisotropy and slab geometry.The obliquely entrained flow is trapped in a narrow zone (<100 km wide) above and below the subducting slab and reaches ∼200 km depth, which is surrounded by large-scale sub-horizontal toroidal flow.The intraslab anisotropy is trench-normal above 80 km depth but changes to trench-parallel at 100-400 km depths, which may reflect fossil anisotropy overprinted by deep deformation beneath the arc, or joint effect of serpentinization and hydrous faulting.

Plain Language Summary
The slab deformation and mantle flow under the Cascadia margin are controversial.We obtain new anisotropic tomography of the Cascadia subduction zone, which can reveal tilting 3-D fast velocity directions (FVDs) and planes.Because the observed trench-normal FVD in the Cascadia margin can be explained by both 3-D toroidal and 2-D entrained flows, we discriminate the two types of mantle flow with 3-D FVDs that contain dip angle information.The 2-D entrained flow above and below the subducting slab is trapped in a narrow zone of ∼100 km wide and is surrounded by large-scale sub-horizontal toroidal flow.The slab anisotropy above 80 km depth is consistent with that in the Juan de Fuca plate before subduction, but it is reshaped beneath the arc due to the slab deformation or phase transition.
LIANG ET AL. dynamics.However, the mantle convection pattern and deformation processes in this region are still puzzling.
Specifically, E-W or ENE-WSW trending fast velocity directions (FVDs) appear in many SKS splitting measurements near the Cascadia margin and the Juan de Fuca-Gorda plate (e.g., Bodmer et al., 2015;Currie et al., 2004;Eakin et al., 2010;Liu et al., 2014;Martin-Short et al., 2015), whereas these similar FVDs onshore and offshore reflect two distinct mantle flow patterns (toroidal flow and entrained flow) simultaneously (Eakin et al., 2010;  (Wüstefeld et al., 2009), respectively, with our fast HSA model, whose length denotes the delay time as shown at the lower-left corner of (c).Offshore colors denote the age of oceanic lithosphere, whose scale is shown at the bottom-right. 10.1029/2023GL105527 3 of 12 Long, 2016;Zandt & Humphreys, 2008).Although anisotropic tomography has better depth resolution, the obtained images are less consistent with each other.For example, some models of azimuthal anisotropy tomography show trench-parallel FVDs (Gosselin et al., 2020;Huang & Zhao, 2013;Wagner et al., 2013) in the Cascadia margin, whereas other anisotropic models show FVDs sub-normal to the trench (Yuan & Romanowicz, 2010;Zhu et al., 2020).In addition, seismic reflection profiles (Han et al., 2016) show that faults are produced in the upper part of the Juan de Fuca plate due to the upward plate bending in the outer-rise area, but the deep portion of the subducted slab is not imaged.The seismicity and tomographic studies have revealed the general morphology of the subducting Juan de Fuca-Gorda plate (Chen et al., 2015;Gao, 2018;Hawley et al., 2016;Hayes et al., 2018;McCrory et al., 2012) with a kink beneath Washington, whereas along-strike variations in seismicity and seismic velocity suggest fragmentation of the subducting slab (Obrebski et al., 2011;Sigloch, 2011).Because the young Juan de Fuca-Gorda plate is thin (Chen et al., 2015;Zhao & Hua, 2021), its deformation and hydration/dehydration processes before and after subduction are hard to constrain.
In this work, we assume that the hexagonal symmetry axis (HSA) of the anisotropic medium in the study region is orientated freely in 3-D space.By inverting a great number of P-wave travel-time data of local and teleseismic events, we obtain two high-resolution tomographic models of isotropic P-wave velocity (Vp) and anisotropy with fast and slow HSAs.Our results shed new light on mantle flow and intraslab deformation in the Cascadia subduction zone.

Data and Methods
We The approach of tilting-axis anisotropy (also known as tilted transverse isotropy) is more reasonable than those of azimuthal and radial anisotropies because its HSA can be freely orientated in 3-D space, which can provide more detailed information on subduction dynamics (Wang & Zhao, 2021;Zhao, Liu, et al., 2023).Under a fast HSA assumption, anisotropy is characterized by one fast axis (i.e., 3-D FVD) along the lineation and two slow axes perpendicular to it (Nicolas & Christensen, 1987).Under a slow HSA assumption, the corresponding model contains a fast velocity plane (FVP) perpendicular to the slow HSA (Table 1 and Figure S6 in Supporting Information S1).In the following, we adopt a T-shaped symbol to express the FVP or FVD projected on the paper plane (Figure 2 and Figure S6 in Supporting Information S1).
In the tomographic inversion for tilting-axis anisotropy, we need more seismic data and updated inversion strategy, because there are one isotropic Vp parameter and three anisotropic parameters at each grid node.The following new inversion strategies are adopted for tilting-axis anisotropy: (a) Partial derivatives of travel-time with respect to the anisotropic parameters are expanded to the second-order Taylor series (Wang & Zhao, 2021).(b) The L-BFGS-B algorithm (Morales & Nocedal, 2011) is used to solve the non-linear inversion problem due to involving the high-order terms into the system of observation equations (Figure S7 in Supporting Information S1).(c) A pre-defined high-Vp zone with ∼35 km thickness and +2% Vp perturbation (Figure S8 in Supporting Information S1) is included in the starting velocity model to express the thin Juan de Fuca-Gorda slab, because the depth resolution of our tomography is lower than the slab thickness (Zhao & Hua, 2021).For details of the tomographic inversion, see Text S2 and Figures S9-S12 in Supporting Information S1.

Results
Figure 2 shows the obtained 3-D fast HSA model.The isotropic Vp images show a clear high-Vp anomaly with an average thickness of 35 km, which reflects the subducted Juan de Fuca-Gorda plate.The FVDs within the slab are sub-normal to the slab interface beneath both the forearc and backarc areas.Above the slab, low-Vp anomalies appear in the lower crust and mantle wedge beneath the arc volcanoes, reflecting dehydration of the slab and serpentinization in the mantle wedge (Bostock et al., 2002;Zhao & Hua, 2021).In the mantle wedge, the FVDs are parallel to the trench beneath Oregon but oblique to the trench beneath Washington and northern California.
In the subslab mantle, the FVDs are relatively complex, and they are NE-SW, N-S, and NW-SE under Washington, Oregon, and northern California, respectively.The FVDs exhibit small dip angles in the mantle wedge and the subslab mantle.To evaluate the slow and fast HSA models, we used the bootstrap method, checkerboard resolution tests, tradeoff tests, and F tests to calculate their uncertainty, resolution, tradeoff between the isotropic Vp and anisotropy, and the effect of the pre-defined slab, respectively (for details, see Text S3, Text S4 and Figures S14-S35 in Supporting Information S1).Our isotropic Vp models have <0.5% uncertainty and <50 km resolution, whereas the tilting-axis anisotropy models have <1.0%uncertainty and <100 km resolution.The test results also show that our isotropic Vp and anisotropy have weak tradeoff, and the intraslab anisotropy is not an artifact due to the pre-defined high-Vp slab in the starting model.
Our predicted SKS splitting with the slow and fast HSA models shows 28.5° and 25.2° direction differences, respectively, with the observed SKS splitting on average.In the forearc and arc areas, most stations show a misfit <20°.Conversely, the NE and southern parts of the study region present a misfit >45° due to significant Fast HSA model A tilting-axis anisotropy model under the assumption that Vp is the fastest along the HSA (Figure S6a in Supporting Information S1).This assumption is suitable for an anisotropic medium with one fast axis along the lineation, for example, deformed peridotite or blueschists Slow HSA model A tilting-axis anisotropy model under the assumption that Vp is the slowest along the HSA (Figure S6b in Supporting Information S1).This assumption is suitable for an anisotropic medium with one slow axis perpendicular to the foliation or fault plane, especially foliated serpentinite in the mantle wedge and subducted slab Table 1 Acronyms uncertainties in the anisotropic parameters or substantial anisotropy in the deep upper mantle below 150 km depth.Previous studies have detected deep upper mantle flow (Mondal & Long, 2020) and toroidal flow (Eakin et al., 2010;Zandt & Humphreys, 2008) beneath the NE and southern parts of the study region, respectively.Hence, the deep portion of the upper mantle may play an important role in the SKS splitting.
Azimuthal and radial anisotropies are widely investigated by previous studies with an assumption of horizontal and vertical HSAs, respectively.Tilting-axis anisotropy has a more general assumption in which the HSA can be freely orientated in 3-D space.Here we compare our slow and fast HSA models with previous azimuthal and radial anisotropy models (See  & Romanowicz, 2010; Zhou et al., 2022;Zhu et al., 2020).These models are generally similar in the forearc and arc areas, suggesting that the tilting-axis anisotropy can reconcile the contradictory assumptions of azimuthal and radial anisotropies.
Nevertheless, the NE part of our fast and slow HSA models is different from the previous azimuthal and radial anisotropy models and the SKS splitting measurements, probably due to the lack of seismic stations and few local earthquakes there.Therefore, the crustal structure there is not well constrained due to the relatively sparse rays (Figure S3 in Supporting Information S1), which may bring systematic errors to anisotropic parameters in the NE part of the study region.

Entrained Flow and Toroidal Flow
There are two endmembers in subduction-induced mantle flow: 2-D corner flow entrained by downdip movement of the subducted slab (Hall et al., 2000) and 3-D toroidal flow induced by slab rollback (Long & Silver, 2008;Russo & Silver, 1994).They can be differentiated well in general because, in most cases, toroidal and entrained flows show trench-parallel and trench-normal FVDs, respectively.In the Cascadia margin, however, the trench-normal FVDs are part of a circular SKS splitting pattern in the western US, which may correspond to toroidal mantle flow induced by a slab window (Zandt & Humphreys, 2008).Therefore, the interpretation of the trench-normal FVDs is challenging as they could reflect both entrained and toroidal flows and are hard to discriminate (Long, 2016).Recent tomographic studies for azimuthal anisotropy, with a better depth resolution, have revealed potential mantle flow patterns (Zhao & Hua, 2021;Zhu et al., 2020), but they still cannot exclude the case that the trench-normal FVDs may reflect toroidal flow (Zandt & Humphreys, 2008).Hence, this paradox (Long, 2016) has not been addressed adequately.Because the tilting-axis anisotropy contains information on azimuthal and dip angles of the fast HSA, it is able to discriminate toroidal and entrained flows and provide a unique perspective to reveal more complex mantle flows.
To discriminate the entrained and toroidal flows, we calculate the included angle between FVDs and oblique entrained flow vectors in both lateral and depth directions (Figures 3b and 3c).At 80 km depth, the FVDs coincide with the oblique corner flow (angle <30°) within ∼100 km width of the subducting slab, which reflects the entrained flow above and below the slab.With the increasing distance from the slab, the FVDs show a larger included angle with the oblique corner flow (∼80°), which may indicate a transition from the entrained flow to toroidal flow.The entrained flow above and below the slab becomes thinner at ∼200 km depth, disappears at ∼300 km depth, and is gradually replaced by the toroidal flow.Hence, we deem that the entrained flow is trapped in a narrow zone (<100 km width; <200 km depth) above and below the slab and surrounded by large-scale sub-horizontal toroidal flow in the deep portion of the subduction zone (>200 km depth).

Along-Strike Variation of Anisotropy
Figure 4 and Movies S1 and S2 show the relationship between the FVDs and slab geometry, in which the azimuthal and dip angle variations of the subducting slab are taken into account.The azimuths and dip angles of subslab FVDs exhibit along-strike variations above 100 km depth.The subslab FVD azimuth changes from NE-trending in the northern part to SE-trending in the southern part (Figure 4d), which align with the inferred direction of mantle flow beneath the Juan de Fuca and Gorda plates based on the offshore SKS splitting results (Bodmer et al., 2015;Martin-Short et al., 2015).Therefore, the subslab anisotropy may be influenced by the flow pattern in the oceanic asthenosphere.
The isotropic Vp and the FVD dip angle in the subslab mantle also exhibit along-strike variations that are correlated with the density of episodic tremor and slow-slip events (e.g., Dragert et al., 2001;Schwartz & Rokosky, 2007).
The subslab mantle beneath Washington and northern California is characterized by subhorizontal FVDs and low Vp, whereas the Oregon subslab mantle shows subvertical FVDs and relatively high Vp (Figures 4c and 4e).These variations may indicate the existence of northern and southern toroidal flows and central entrained flow in the subslab mantle (e.g., Zhao & Hua, 2021).Notably, the tremor density near the Cascadia margin is high in Washington and northern California but low in Oregon (Figures 1b and 4e), being correlated with variations of the isotropic Vp and the FVD dip angle in the subslab mantle.Previous studies have shown that a low-Vp subslab anomaly reflects high buoyancy (Bodmer et al., 2018;Hawley et al., 2016) due to thermal anomalies and partial melting, which may affect the along-strike variations in tremor density (Bodmer et al., 2018;Porritt et al., 2011) and megathrust earthquakes (Fan & Zhao, 2021).The variations in temperature and melt content may be associated with the different patterns of subslab mantle flow.

Intraslab Anisotropy
The intraslab anisotropy is thought to be more complex, including contributions of fossil anisotropy of the oceanic lithosphere (Audet, 2013;Song & Kim, 2012), strain-related anisotropy due to slab bending (Faccenda et al., 2008;Wang et al., 2022), and fabric transition of anisotropic minerals (Karato et al., 2008).In our fast HSA model, the intraslab FVDs are parallel to the slab interface, but the FVPs vary from E-W trending beneath the forearc to N-S trending beneath the back-arc (Figure 5, Movies S3 and S4).The following two mechanisms may explain this result.
1. Fossil anisotropy under the forearc and slab deformation under the back-arc.SWS studies using data recorded at dense oceanic bottom seismometers have revealed NE-SW polarized direction in the Juan de Fuca plate and E-W polarized direction in the Gorda plate (Bodmer et al., 2015;Martin-Short et al., 2015) (Figure 1c).This feature is consistent with the azimuth of intraslab FVDs and FVPs beneath the forearc in our models (Figure 5).In the forearc, the slab is shallower than 50 km depth and its vertical bending is relatively small, so fossil anisotropy may dominate the intraslab anisotropy.But beneath the arc and back-arc areas, the slab bends abruptly with its dip angle increasing from ∼20° to ∼60° (Figure 3a).The intraslab FVP changes from NE-SW trending and upright beneath the forearc to N-S trending and eastward-dipping beneath the arc and back-arc, suggesting that the fossil anisotropy under the forearc is overprinted by newly-formed anisotropy.Tilting-axis anisotropic tomography beneath Japan shows that hydrous faulting contributes to the intraslab FVPs under the forearc (Wang et al., 2022), where fossil anisotropy in the subducting Pacific slab is overprinted in the strongly bending part (with a dip angle of ∼45°) (Wang & Zhao, 2021).Another example is in South America where the subducted Nazca slab fabric was modified from the fossil spreading direction to a direction sub-parallel with the slab strike (Eakin et al., 2016).Hence, fossil anisotropy of the Juan de Fuca plate overprinted by intraslab deformation could explain the FVP change beneath the arc area.2. Joint effects of serpentinization and hydrous faulting.In a global view, serpentine has strong anisotropy and is widespread in the forearc area, especially in warm subduction zones such as Cascadia (Abers et al., 2017;   et al., 2002;Reynard, 2013;Wagner et al., 2013) and SW Japan (Zhao et al., 2021).Hydrous faulting is commonly developed near the trench due to upward bending of the subducted slab, which is thought to be an important source of anisotropy in the subduction zone (Faccenda et al., 2008;Wang et al., 2022).
Experimental evidence shows that serpentine would break down at 60-120 km depths in the thermal condition of a warm subduction zone (Reynard, 2013), which coincides with the depth of FVP change in our slow HSA model (∼80 km depth; Figures 5a and 5b).Therefore, the intraslab anisotropy may be affected by serpentinization predominantly in the forearc.When serpentine breaks down with the subducted slab going deeper, hydrous faulting could be responsible for the intraslab anisotropy under the back-arc.We used the seismic data downloaded from the data centers of the U.S. EarthScope Transportable Array and the International Seismological Center.Drs.Fan-Chi Lin and Hejun Zhu kindly provided their azimuthal anisotropy models for the model comparison.We appreciate helpful discussions with Dr. Zewei Wang on the tomographic inversion for tilting-axis anisotropy.We are very grateful to Prof. Daoyuan Sun (the Editor) and four anonymous referees for their thoughtful review comments, which have greatly improved this paper.We acknowledge the Beijing Super Cloud Computing Center (BSCC) for providing HPC resources.This work was partially supported by the fundamental and applied fundamental research major program of Guangdong Province (2019B030302013), grants from the National Natural Science Foundation of China (No. 42288201, 42106066), Japan Society for the Promotion of Science (No. 19H01996), PI Project of Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou) (GML2022006), and Tuguangchi Award for Excellent Young Scholar (SKLaBIG-TJ-22-01).Most of the figures are plotted using the GMT and ParaView software packages.This is contribution NO.IS-3410 from GIGCAS.

•
Tilting-axis anisotropy can reconcile contradictory assumptions of azimuthal and radial anisotropies • Entrained flow occurs within ∼100 km depth above and below the subducting slab and is surrounded by toroidal flow • Fossil anisotropy in the slab is overprinted beneath the arc due to intraslab deformation or phase transition Supporting Information: Supporting Information may be found in the online version of this article.

Figure 1 .
Figure 1.(a) Summary map of subslab mantle anisotropy worldwide derived from shear-wave splitting (SWS) measurements (from Long, 2013), in which the fast polarized wave is trench-parallel in most subduction zones except for the Cascadia margin (black rectangle).The length and orientation of red bars denote the SWS shortest delay time and polarized direction in the subslab mantle, respectively.The size of pink circles denotes the longest SWS delay time in the subslab mantle.The delay time scale is shown at the lower-left corner.Background colors denote surface topography whose elevation scale is shown at the bottom-left.(b) Red triangles denote active volcanoes.Red patches onshore denote density contours of tremors.Blue lines offshore and onshore outline boundaries of the Juan de Fuca and Gorda plates and contours of the subducting slab (from the Slab2 model; Hayes et al., 2018), respectively.Blue arrows offshore denote the motion direction of the Juan de Fuca plate relative to the North American plate.(c) Red and blue bars denote observed and predicted SKS splitting results(Wüstefeld et al., 2009), respectively, with our fast HSA model, whose length denotes the delay time as shown at the lower-left corner of (c).Offshore colors denote the age of oceanic lithosphere, whose scale is shown at the bottom-right.
collected P-wave arrival-time data from the International Seismological Center (ISC)-EHB bulletins during 2000-2018 and the Array Network Facility component of the EarthScope USArray program during 2004-2021.We selected local and teleseismic events according to the following three criteria: (a) Each event was recorded at 10 or more stations in the study region.(b) Epicentral distance from a teleseismic event to a station is limited to 30° to 90°.(c) All local earthquakes are relocated and their uncertainties are <7 km in the epicenter, <8 km in focal depth, and <1.0 s in origin time.After merging the two data sets (ISC-EHB and USArray) and removing the duplicates, our final data set includes 91,066 first P-wave arrival times of 6805 local earthquakes and 584,187 P-wave relative travel-time residuals of 12,602 teleseismic events (Figures S1-S5 and Text S1 in Supporting Information S1).These data were recorded at 687 seismic stations including 197 USArray stations (FigureS1ain Supporting Information S1).

Figure
Figure S13 in Supporting Information S1 shows the obtained 3-D slow HSA model.The isotropic Vp images are almost the same as those of the fast HSA model, though some high-Vp anomalies exhibit a lower amplitude than the fast HSA model.The slab subducts at a low angle beneath the forearc and exhibits upright and ENE-WSW trending FVPs.Beneath the back-arc area, however, the slab becomes steep and its FVPs change to NNE-SSW trending and sub-normal to the slab interface.The FVPs in the mantle wedge are sub-parallel to the slab interface beneath Washington and northern California, but they change to NE-SW trending and perpendicular to the slab interface beneath Oregon.The subslab mantle is characterized by low-Vp anomalies and sub-horizontal FVPs.

Figure 2 .
Figure 2. (a-c) Map views at depths of 40, 80, and 200 km and (d-f) vertical cross-sections along 41°N, 44°N, and 47°N latitudes of the fast HSA model.The background blue and red colors denote high and low isotropic Vp perturbations, respectively, whose scale is shown below (d).Each T-shaped symbol denotes a 3-D fast velocity direction (FVD) projected on the paper plane as shown below (e) and (f), whose size denotes the anisotropic amplitude with a scale as shown below (d).The red open triangles and white tiny dots denote active volcanoes and earthquakes, respectively.In (a-c), the blue line denotes location of the slab upper boundary at each depth, and white lines denote locations of the vertical cross-sections in (d-f).In (d-f), blue solid lines denote the Moho discontinuity and the slab upper boundary, whereas the blue dashed line denotes the slab lower boundary.The gray patches atop (d-f) denote the surface topography.

Figure 3 .
Figure 3. (a) Distribution of azimuthal (blue dots) and dip (green dots) angles of fast velocity planes (FVPs) within the slab along an east-west profile at 43°N latitude.The vertical red bar denotes the arc location.(b) Schematic diagram showing domains of toroidal (yellow) and entrained (pink) flows, as well as the reorientation of intraslab anisotropy beneath the arc.The red triangle denotes the volcanic front.Two blue solid lines denote the Moho discontinuity and the slab upper boundary, whereas the blue dashed line denotes the slab lower boundary.(c) Distribution of the included angle between the fast velocity direction (FVD) and the oblique entrained flow vector along the 43°N latitude profile.The blue, pink and yellow areas denote domains of the subducting slab, entrained flow and toroidal flow, respectively.

Figure 4 .
Figure 4. (a, b) Map views showing the fast velocity directions (FVDs) in (a) the mantle wedge (50 km above the slab upper boundary) and (b) the subslab mantle (50 km below the slab lower boundary) of our fast HSA model.The colors denote the spatial relationship between the FVDs and the slab interface as shown at the bottom-right.Each T-shaped symbol denotes a 3-D FVD projected on the paper plane as shown on the right of (g).The blue arrows denote the subduction direction.Blue lines denote depth contours of the mantle wedge or the subslab plane.(c-e) North-south variations of (c) the dip angle and (d) azimuth of subslab FVDs, and (e) subslab isotropic Vp at 80 km depth (red dots) and tremor percentage (pink patches), along 124.5°W longitude.Two horizontal lines in (c, e) denote state boundaries of Washington, Oregon and California.The gray patches in (d) denote inferred directions of mantle flow beneath the Juan de Fuca and Gorda plates(Bodmer et al., 2015;Martin-Short et al., 2015).(f, g) Stereographic projections and 3-D views of the FVDs (red bars) in the mantle wedge (f) and the subslab mantle (g).In the stereographic projection, the red, blue and yellow colors denote areas close to the dip, azimuthal and normal vectors of the slab interface, respectively.The dashed black line denotes the general slab interface.The crosses denote stereographic projections of the FVDs in the mantle wedge (f) or the subslab mantle (g).

Figure 5 .
Figure 5. (a) Map view showing the fast velocity planes (FVPs) within the subducting slab of our slow HSA model.The colors denote the spatial relationship between the HSA and the slab interface as shown below (c).Each T-shaped symbol denotes a FVP projected on the paper plane as shown below (d).Blue arrows denote the subduction direction.Blue lines denote depth contours of the slab interface.(b) Stereographic projections of the slow HSAs (black crosses) and 3-D view of FVPs (green ellipses) within the subducting slab.In the stereographic projection, the red, blue and yellow colors denote areas close to the dip, azimuthal and normal vectors of the slab interface, respectively, whereas the dashed black line denotes the general slab interface.(c, d) The same as (a, b) but for intraslab fast velocity directions (FVDs, blue bars) of the fast HSA model.Each T-shaped symbol in (c) denotes a FVD projected on the paper plane as shown below (d).Black crosses in the stereographic projection (d) denote fast HSAs within the subducted slab.