RESEARCH ARTICLE TX2019slab: A New P and S Tomography Model

10.1029/2019JB017448
Incorporating Subducting Slabs
Key Points:
• We developed new P and S
Chang Lu1 , Stephen P. Grand1, Hongyu Lai2 , and Edward J. Garnero2
tomography models incorporating 1
3‐D subducting slabs
Department of Geological Sciences, Jackson School of Geosciences, University of Texas at Austin, Austin, TX, USA,
2
• Mislocation effects caused by School of Earth and Space Exploration, Arizona State University, Tempe, AZ, USA
subducting slabs were reduced by
inverting for velocity and source
location simultaneously Abstract Large numbers of earthquakes occur in subduction zones that are marked by dipping, narrow
• The spin transition and the
high seismic velocity slabs. The existence of these fast velocity slabs can cause serious earthquake
post‐perovskite phase transition
may explain lower mantle mislocation problems that can bias estimates of seismic travel time residuals. This can affect the recovery of
heterogeneities subducting slabs in tomography as well as introduce significant artifacts into lower mantle structure in
tomography models. In order to better account for known subducting slabs, we performed a new P and S
Supporting Information:
wave joint tomography inversion incorporating a three‐dimensional thermal model of subducting slabs in
• Supporting Information S1
the starting model. In addition, velocity and source locations were inverted for simultaneously. Our new P
and S models feature higher‐amplitude subducting slabs compared with previous global tomography results.
Correspondence to: The S to P heterogeneity ratio based on the new tomography model indicates that thermal elastic effects
C. Lu, alone cannot explain all the heterogeneities in the lower mantle. Much of the observed abnormal S to P
[email protected]
heterogeneity ratio can be explained by anelastic effects, the spin transition, and phase transitions of
bridgmanite to post‐perovskite in the lower mantle.
Citation:
Lu, C., Grand, S. P., Lai, H., & Garnero, Plain Language Summary Seismic tomography uses seismic travel time data to image deep
E. J. (2019). TX2019slab: A new P and S earth velocity structure. However, it has been shown that the existence of subducting slabs can
tomography model incorporating
significantly bias the imaging result. In order to reduce this effect, we produced a new P and S wave
subducting slabs. Journal of Geophysical
Research: Solid Earth, 124, tomography model that included theoretical three‐dimensional subducting slab structures in the starting
11,549–11,567. https://doi.org/10.1029/ model. The new model has higher‐amplitude subducting slabs compared with other models. Based on the
2019JB017448
new model, we conclude that it is difficult to explain the P and S velocity anomalies found in the deep mantle
by temperature variations alone without invoking complex mineral phase transitions, large anelastic effects,
Received 26 JAN 2019
Accepted 27 SEP 2019 or chemical variations.
Accepted article online 22 OCT 2019
Published online 11 NOV 2019

1. Introduction
Subduction of oceanic lithosphere is believed to play a critical role in large‐scale mantle convection (e.g.,
Billen, 2008; Kellogg et al., 1999). The investigation of deep subducting slabs, therefore, has been an active
field in geophysics. It is widely accepted that subducting slabs are relatively cold and thus have high seismic
velocity compared to surrounding mantle (e.g., Fukao & Obayashi, 2013; Zhao et al., 2017). Global seismic
tomography, a method to image deep earth seismic velocity structure, has been used to constrain the loca-
tion of subducting slabs. Generally, shear wave global tomography has limited ability to image short‐
wavelength structures such as subducting slabs (e.g., French & Romanowicz, 2014; Grand, 2002;
Kustowski et al., 2008; Moulik & Ekström, 2014; Panning & Romanowicz, 2006; Ritsema et al., 2011).
Compressional wave global tomography has provided higher‐resolution images of upper mantle slabs
(e.g., Amaru, 2007; Li et al., 2008; Obayashi et al., 2013; Simmons et al., 2012). However, detailed waveform
modeling studies suggest that the amplitude of subducting slabs is underestimated in global P wave tomogra-
phy models (Zhan et al., 2014). Tao et al. (2018) performed full‐waveform inversion using upper mantle tri-
plicated waves to image the subducting slabs beneath Eastern Asia. Both of their P and S models show much
higher‐amplitude velocity anomalies inside the slabs than in global tomography models. Other seismic mod-
elling results, as well as theoretical thermal models of subducting slabs, also imply far stronger velocity
anomalies within slabs relative to those seen in global tomography (Chen et al., 2007; Kawakatsu &
Yoshioka, 2011; Syracuse et al., 2010; Wang et al., 2014). The discrepancies between detailed seismic studies
©2019. American Geophysical Union. of slabs with global tomography models imply a potential problem with using tomography models to infer
All Rights Reserved. density anomalies for use in mantle convection studies.

LU ET AL. 11,549
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

The existence of sharp, dipping fast velocity anomalies can also bias earth-
quake locations and impact seismic travel time data used in tomography
(Creager & Jordan, 1984; Ding & Grand, 1994; Sleep, 1973). Therefore,
the incorrect imaging of subducting slabs could degrade tomography
models even far from subduction zones. Through synthetic testing, Lu
and Grand (2016) found that the incorrect imaging of subducting slabs
could introduce up to 0.5% amplitude false velocity anomalies in the lower
mantle in global shear wave tomography. This is comparable to the ampli-
tude of velocity anomalies found in the lower mantle. These artifacts may
have significant implications for interpretation of lower mantle heteroge-
neities. For example, some studies use the S to P heterogeneity ratio to
identify potential chemically distinct heterogeneities (e.g., Koelemeijer
et al., 2016; Masters et al., 2000; Saltzer et al., 2001; Tesoniero et al.,
2016). Artifacts produced by incorrectly accounting for subducting slabs
can impact the reliability of results using this approach.
Lu and Grand (2016) compare several strategies to best account for sub-
Figure 1. Distribution of travel time residuals measured at station MAW for
ducting slabs in global tomography. They suggest including a priori sub-
a representative 30 × 30‐km earthquake group used in the EHB data selec-
ducting slabs in the starting model and performing structure and source
tion process. The group is centered at 7.3°S, 155.9°E, and 45‐km depth. The
station MAW is located in Antarctica 67.6°S, 62.9°E. The residuals outsidelocation jointly in tomographic inversions. In this study, we present a
±2σ were treated as outliers and eliminated from the inversion. new P and S global tomography model using a three‐dimensional (3‐D)
subducting slab structure in the initial model. This enables us to produce
a global tomography model with more realistic subducting slabs. In the lower mantle, we evaluate the effect
of subducting slabs on S to P heterogeneity ratio estimation.

2. Data and Method

2.1. P Wave Data
The P wave data used in this study are from the database of travel time residuals maintained by E. R.
Engdahl and coworkers (EHB data; Engdahl et al., 1998). They use arrival times reported to the
International Seismological Center (ISC) and the U.S. Geological Survey's National Earthquake
Information Center (NEIC), relocate earthquakes and re‐identify phases, and then produce a final data set
after quality control. In this study, we use P, pP, and Pn phases in the EHB data, which contains approxi-
mately 14 million rays from ~553,000 earthquakes from the years 1966 to 2008.
Even though the quality of EHB data is significantly higher than the raw ISC catalogue data, there are still
errors in the data set. For example, we have found cases where two nearby earthquakes have more than a 10‐
s difference in travel time residuals to a common station. In order to further improve the data quality, we
designed a new data selection process for the EHB data. We first group nearby earthquakes together, then
evaluate the travel time residuals of each group at each station. For earthquakes close to subduction zones,
a group covers about a 30 × 30‐km area in the horizontal direction, while larger 300 × 300‐km areas form
groups away from subduction zones. In the radial direction, we separate earthquakes using a 30‐km interval.
Figure 1 shows the distribution of travel time residuals for a representative 30 × 30‐km group (centered at
7.3°S, 155.9°E; 45‐km depth) to station MAW in Antarctica (67.6°S, 62.9°E). The residuals outside ±2σ (cor-
responding to ~95% confidence interval) were labeled as outliers and eliminated from the inversion. For
groups with very few arrivals to a station (less than 10 arrivals), the standard deviation of the whole EHB
data (~1.8 s) is used to define outliers. This process rejected about 5% of the EHB data.
2.2. S Wave Data
We include two groups of shear wave data in this study. The first group (TX data) was used in Lu and Grand
(2016). The data set consists of ~70,000 global S, ScS, SKS, and SKKS phase travel times from 540 earth-
quakes. Their upgoing and surface bounce equivalents are also included. Upper mantle triplicated waves
are also included to provide better coverage in the upper mantle and transition zone. All the travel times
were manually measured using a seismic waveform approach. The data measurement method, with exam-
ples, is discussed in Grand (1994). The other group of shear wave data is presented in Lai et al. (2019). They
use a semiautomated process to determine the onset time of horizontally polarized shear waves bandpassed

LU ET AL. 11,550
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

from 0.01 to 0.1 Hz. The measurements involve the derivation of an

empirical wavelet, which is matched to seismic waveforms from a given
earthquake, through an iterative cross‐correlation process. The onset tra-
vel time is measured based on a Gaussian function which best fits the
empirical wavelet. This data set includes ~226,000 S, SS, SSS, ScS, and
ScSScS phases from 360 earthquakes all of which turn within the lower
mantle; that is, upper mantle triplicated phases are excluded. These new
data provide better ray coverage than the first data set in the lower mantle,
especially in the southern hemisphere (Figure S1).

2.3. Input Slab Model

We use a starting 3‐D slab model based on a theoretical global 3‐D thermal
Figure 2. Comparison of the subducting slab beneath East Asia found in a model of subducting slabs given by Stadler et al. (2010). The slab locations
regional tomography model FWEA18 (Tao et al., 2018) and our input slab are defined by seismicity and represented on a 0.1° × 0.1° grid. The model
model. (a) Horizontal slice of shear wave model FWEA18 at 300 km. The red does not include seismically inactive subduction zones so it is a relatively
line shows the location of the cross sections in (b) and (c). (b) Cross section
conservative representation of slabs. The geometry of the slabs has been
of the same tomography model in (a). (c) Cross section of the input shear
wave slab model. Solid black lines represent the 410‐ and 660‐km disconti- checked against model Slab 1.0 (Hayes et al., 2012). The high degree of
nuities, respectively. similarity between the two models, in terms of slab geometry, confirm
the reliability of the input slab model. The thermal structure of the slab
model is determined by the age of the subducting plate at the time of sub-
duction and the convergence rate. When compared with slab thermal models produced by detailed dynamic
modeling (e.g., Kawakatsu & Yoshioka, 2011; Syracuse et al., 2010; van Keken et al., 2011), the Stadler et al.
model generally has larger thermal anomalies. Therefore, we changed the thermal anomalies in the Stadler
et al. model using the slab thermal model of van Keken et al. (2011) while keeping the geometry the same.
Because the van Keken et al. model only contains 2‐D thermal structures across different subducting slabs,
we interpolated their model for each slab in Stadler et al.'s model to make the average thermal anomaly
across slabs to be consistent with van Keken et al.'s model. The scaling we used to adjust the amplitude of
the Stadler et al. model varied from ~0.7 to ~0.9.
We followed Lu and Grand (2016) to convert the thermal slab model to a 3‐D perturbation model of P and S
velocity. The slab model contains uncertainties due to both the thermal structure as well as the mineral phy-
sics modeling procedure. However, our derived slab model generally agrees with some detailed regional stu-
dies (e.g., Chen et al., 2007; Gaherty et al., 1991; Tao et al., 2018; Vidale, 1987) in terms of geometry as well as
seismic amplitudes. In Figure 2, we compare our input slab model for a cross section in East Asia with a
recent regional full‐waveform tomography model FWEA18 (Tao et al., 2018). Both models have about a
100‐km‐thick undeformed subducting slab dipping to 660‐km depth. Therefore, we feel the model is a good
first‐order approximation to global slab structure.

2.4. Forward Modeling

In our previous work, we used the open‐source package “SPECFEM3D GLOBAL” Spectral‐Element Method
(SEM; Komatitsch & Tromp, 2002a, 2002b) to model the effect of the 3‐D subducting slabs on shear wave
travel times for the handpicked data set discussed above (Lu & Grand, 2016). For each earthquake, we com-
puted SEM synthetics for the 1‐D starting model then another simulation for a second model that included
slabs embedded in the 1‐D model. The calculations are accurate to periods of about 12 s. Comparing the two
simulations, for each measured phase in the data, allowed us to adjust the travel time residuals for the effect
of the slabs. In other words, we produce a set of residuals relative to an Earth model with slabs.
For the presumably higher frequency and far larger EHB data, it is not feasible to use the SEM method to
model the effect of slabs on the data. Therefore, instead of using SEM, we used 3‐D ray tracing to evaluate
the effect of slabs on our P wave data. The shear wave data set from Lai et al. (2019) are measurements of
onset time so are presumably the equivalent of high‐frequency travel times. Thus, we also used 3‐D raytra-
cing to correct the Lai et al. data set for slab structure. The open‐source LLNL‐Earth3D 3‐D ray tracing pack-
age (Simmons et al., 2012) was used for this purpose. The 3‐D Earth can be represented by different levels of
spherical tessellation grids. A higher tessellation recursion level means a finer grid and more computation
cost. In order to check the accuracy of the 3‐D ray tracing code as well as determine the optimal level of

LU ET AL. 11,551
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 3. Distribution of travel time residuals caused by the input slab model. Plotted are the differences between a 1‐D
prediction of travel times and predictions for the same 1‐D model with slabs embedded for all the data used in our
inversion. P wave residuals were derived from 3‐D ray tracing (Simmons et al., 2012), while the S wave residuals were
obtained from 3‐D ray tracing and SEM simulations (Komatitsch & Tromp, 2002a, 2002b).

the spherical tessellation grids, we compared 3‐D ray tracing results with SEM simulation results for a few
test earthquakes. We find that the when the tessellation recursion level is above 9 (corresponding to ~0.25
arc degree average node spacing), the difference between 3‐D ray tracing and SEM simulations agree to
within 0.2 s, which we consider acceptable for our purposes. The effect of subducting slabs on the seismic
travel time residuals are summarized in Figure 3. The largest travel time residual caused by our starting
model is ~8 s for P wave data while it is ~12 s for S wave data.
The “TX” data set contains upper mantle turning waves. As discussed in Grand (1994), due to the large het-
erogeneity in the shallow mantle, at a given distance the ray paths of S or SS waves can be quite different
depending on the specific region being sampled. Relative to standard 1‐D models the difference is especially
significant in cratonic regions. For this reason, different seismic models were used to determine ray paths for
upper mantle waves in Grand (1994, 2002). This process required detailed waveform analysis, and thus is not
suitable for large‐volume catalogue data. The EHB P data set contains upper mantle turning waves. In order
to correct the ray paths that sample cratonic regions for EHB data, we first derived a 3‐D S wave craton
model and then converted it to a P wave model (Figure S2). The fast velocity anomalies in our previous S
wave tomography at the shallowest depths (Lu & Grand, 2016) are used to determine the location of cratons
if they have Archean or Proterozoic crust (Laske et al., 2013). The fast velocity anomalies down to 210 km
which are close to these regions are used to define the craton roots. These cold cratons are believed to have
distinct chemical compositions relative to surrounding mantle and contain high percentages of Mg. Lee
(2003) showed that the effect of Mg# (100 × Mg/(Mg + Fe)) on Vp/Vs ratio was significantly larger than
the effect of temperature in peridotite. Therefore, we estimated the Mg# in cratons using a linear interpola-
tion based on our S wave craton model, assuming that normal mantle has a low (87) Mg# while the fastest
craton has the highest (94) Mg# (Lee, 2003). Using this approach, the average variation of Mg# in cratons is
~3.5, which agrees with the value reported by Deschamps et al. (2002) and is slightly higher than predictions
by Forte and Perry (2000) and Perry et al. (2003). Then we adopted the linear relationship between Mg# and
Vp/Vs ratio reported by Lee (2003) to derive a P wave craton model. Using the P wave craton model, 3‐D ray
tracing (Simmons et al., 2012) was used to determine the ray paths for rays turning above 800‐km depth
(Figure S3). For rays turning below 800‐km depth, the effect of cratons on seismic ray paths is negligible.
2.5. Joint P and S Inversion Starting From 3‐D Slab
In our inversion, the mantle was divided into 99,146 blocks. The blocks are about 275 × 275 km in lateral
dimension and vary from 75 to 240 km in thickness. Both P and S wave sensitivity kernels were calculated
using the ray theory approximation based on a 1‐D velocity model. For S waves, we use a combined
TNA/SNA model (Grand & Helmberger, 1984), which is the same 1‐D model as we used in previous TX mod-
els (Grand, 2002). The AK‐135 model (Kennett et al., 1995) is used as the P wave starting model. The P and S
wave data are known to have very different ray coverage, especially in the shallow mantle. At shallow
depths, the P wave data have limited sampling of the oceans, while S wave data have much better resolution
there. Therefore, we correlated the P wave and S wave models by introducing another term XP/S into our

LU ET AL. 11,552
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

inversion, which is the ratio of the P and S wave slowness perturbations in each layer. This parameter can be
obtained using thermoelastic parameters from mineral physics measurements as a function of depth assum-
ing a thermal origin for mantle seismic heterogeneity (see Lu & Grand, 2016 for details). The XP/S in cratons
vary in 3‐D and are consistent with the values used in our forward modeling. Also, Lu and Grand (2016)
showed that correcting for earthquake mislocation can be better achieved if one inverts for velocity structure
and earthquake location simultaneously. Therefore, we include earthquake location in our inversion as well.
The linearized seismic tomography problem can be written as
2 3 2 3
GP AP rP
2 3
6 λAS 7 6 λr S 7
6 λGS 7 mP 6 7
6 76 7 6 7
6 DP 76 mS 7 6 0 7
6 76 7 6 7
6 76 LP 7 ¼ 6 7 (1)
6 74 5 6 7
6 DS 7 6 0 7
4 5 LS 4 5
λX −λX X P=S 0

where GP and GS are the P and S wave sensitivity kernel matrices and rP and rS are the corresponding travel
time residuals caused by velocity anomalies and earthquake mislocation. The travel time residuals r are the
leftover residuals after we removed the effect of subducting slabs, which is the most significant difference
from previous global tomography studies. We use λ to represent the relative weight of S wave data to P wave
data. mP and mS are the P wave and S wave slowness perturbations and L represents the relocation para-
meters to be inverted that include changes in event latitude, longitude, and depth, as well as origin time.
A is the relocation matrix formed of the partial derivatives for these parameters. D is a smoothing operator,
which is a Laplacian filter with 76% of the weight applied to horizontal nearby blocks and 24% of the weight
applied to the vertical nearby blocks. A weighting term λX is used to control how strong the connection
between the P and S models is enforced. Through a trial and error process, λX was chosen to be 500 which
was the maximum value before the connection term began to decrease the variance reduction of the S wave
data (less than 0.5%; Figure S4).

3. Results
3.1. TX2019slab Model
We show our new model (TX2019slab) in horizontal slices at selected depths in Figure 4. The blocks used in
the inversion and the input slab model have been resampled onto a 1° × 1° grid when making the plot. Our P
and S models generally agree with previous tomography results at large scale. At 150‐km depth, our model
shows fast velocity anomalies in cratonic regions as expected. Slow velocity anomalies are found beneath
mid‐ocean ridges as well as the East African Rift. Due to poor data coverage in the oceans, most previous
global P wave tomography models do not have slow mid‐ocean ridges (e.g., Amaru, 2007; Li et al., 2008;
Obayashi et al., 2013). Our shallow oceanic P structure is mainly constrained by the S wave data and the rela-
tionship between P and S anomalies we introduced into the inversion. In the upper mantle and transition
zone, short‐wavelength subducting slabs appear to be the most heterogeneous structures in our model.
Interestingly, the slab signature near 600‐km depth is much broader than at 300‐km depth. In the mid‐lower
mantle, as in previous tomography studies, two elongated fast velocity anomalies in P and S are seen beneath
North/South America and South Asia. The locations of these fast anomalies generally agree with previous
studies (e.g., Amaru, 2007; French & Romanowicz, 2014; Grand, 2002; Li et al., 2008; Moulik & Ekström,
2014; Obayashi et al., 2013; Ritsema et al., 2011). Beginning in the mid‐lower mantle, slow anomalies are
seen in both P and S beneath the south central Pacific and Africa. They increase in strength and size with
depth and at the bottom of the mantle are quite broad. These two structures have been called Large Low
Shear Velocity Provinces (LLSVPs; see Garnero et al., 2016 for a review). Note that the P wave amplitude
is muted at the base of the mantle inside the Pacific LLSVP.
Our model yields a 33.8% overall variance reduction for the P wave data and 91.4% for S wave data. In com-
parison, we derived another P and S model using the same data, inversion method, and regularization but
starting from a 1‐D velocity model. We refer to this model as TX2019. The variance reduction, using model
TX2019, is slightly less than 33.8% for the P wave data and 91.3% for the S wave data. The inversion starting

LU ET AL. 11,553
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 4. Lateral velocity variations in model TX2019slab at representative depths. Each row corresponds to the model at a
particular depth, and the two columns show (left) P wave and (right) S wave models, respectively. The color scales change
for each plot according to the amplitude (X) labeled at the bottom right corner. Purple lines in the second row show the
locations of cross sections in Figures 7 and 8.

from a model with 3‐D slabs results in a slightly better data fit than the inversion starting from the 1‐D model
although the difference is minimal. This is a clear indication of the nonuniqueness of models in global
seismic tomography. The addition of slabs in our starting model, however, results in a model more
consistent with regional studies and geodynamic models without sacrificing fit to global data.

3.2. Checkerboard Test

In order to test the ability of the data to image mantle structure, we performed a resolution test. The input
checkerboard consists of 5° × 5° blocks with alternating P wave velocity perturbations of ±1.5% for each
layer. S wave velocity perturbations are scaled from the P wave, using the thermally introduced S to P rela-
tionship discussed above. Synthetic travel times were generated using the sensitivity matrix from forward
modeling, and inverted using the same inversion scheme as for the real data (Figure 5). The correlation

LU ET AL. 11,554
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

coefficient (CC) between the input (min) and output models (mout) at each depth has been calculated and
labeled in Figure 5. The CC of two scalar fields can be calculated as

N  
∑i¼1 X 1;i X 2;i
CC ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffi (2)
N  N 
∑i¼1 X 1;i ∑i¼1 X 2;i

where X1,i and X2,i are the ith elements in X1 and X2 after having mean values removed, respectively. We also
calculated amplitude recovery (AR) to illustrate the resolution of the inversion. AR is defined as the ratio of
the root‐mean‐square (RMS) amplitudes of min and mout as
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
N N
AR ¼ ∑i¼1 mout 2 = ∑i¼1 min 2 ×100% (3)

The CCs vary from 0.68 to 0.98, while amplitude recoveries are generally below 83% because of the regular-
ization. Because of the S to PP wave velocity scaling used in the inversion, the shallow oceanic regions in the
P wave model show some structures simply scaled from the S wave model. However, these regions are still
the least resolved regions in the mantle. The derived P wave model in these regions highly depends on the S
wave model as well as the scaling relationship used in the inversion.

3.3. Relocation Results

Earthquake hypocenters were relocated in our inversion. The average epicenter shift for the whole data set is
about 9 km with a 6‐km standard deviation. We also find an average 3 ± 10‐km depth shift for all the events
used in our inversion. For the origin time, our inversion results found a 0.1 ± 1.2‐s average shift. These
results generally agree with earthquake relocation results for EHB data reported by Amaru (2007).
Figure 6 shows the shifts in event locations in two subduction regions. For better visualization, only earth-
quakes with more than 250 travel time residuals and a focal depth greater than 100 km are included in the
plots. In regions that are azimuthally well sampled such as the Ryukyu arc and Izu‐Bonin arc, the epicenter
shifts are relatively small. In regions, such as South America, where the station coverage around the subduc-
tion zone is less complete, the epicenters shift more (Figure S5). Note that in the Kuriles, there is a systematic
shift oceanward in epicenters due to the northwest dipping slab. There is also an interesting change in pat-
tern along Central America that may be related to a change in slab dip angle along the subduction zone.
Further analysis of the earthquake relocation results is left for future work.

4. Discussion
Compared with previous global tomography studies, the most significant difference in our model is that we
performed our inversion starting with a 3‐D subducting slab model and invert for velocity and earthquake
location simultaneously. We compare our new model to previously published models in vertical cross sec-
tions across four major convergent plate boundaries. Figure 7 compares P models across the sections
(Amaru, 2007; Li et al., 2008; Obayashi et al., 2013), and Figure 8 compares S models (French &
Romanowicz, 2014; Moulik & Ekström, 2014; Ritsema et al., 2011). The cross‐section locations are shown in
Figure 4.
For the P wave models, similar fast anomalies are seen in all four tomography models (Figure 7). Across the
northern Honshu arc, our model shows stagnant slab above the 660 discontinuity, and agrees with all the
other models. The cross section across the western Java arc, beneath which the Indo‐Australian plate is sub-
ducting, shows large fast anomalies in the uppermost lower mantle, with an extension of fast velocity into
the deeper mantle to the north. Similar features can be seen in the other models, although our new model
has a larger, stronger anomaly in the deepest mantle. The Tonga and South American cross sections show
more differences among models. This is likely due to worse azimuthal station coverage around these regions.
The cross sections across the Tonga arc show very complex structures. In our model, a stagnant slab above
the 660 discontinuity is seen but there is also an anomaly in the lower mantle with a gap between the two.
This is most similar to the GAPP4 model except in that model there is no gap between the deeper structure
and the anomalies in the transition zone. Some studies (e.g., Bonnardot et al., 2009; Brudzinski & Chen,
2005) suggest that the stagnant slab above the 660 is the southwestward flattening of the downgoing slab

LU ET AL. 11,555
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 5. Output from a checkerboard test for (left column) P wave and (right column) S wave data at selected depths.
Correlation coefficients (CCs) and amplitude recovery (AR) between the input checkerboard and the inversion results
are labeled on each plot.

beneath the Lau basin, while the slab below the 660 is the westward extension of the downgoing slab from
the southern Tonga trench (see Fukao & Obayashi, 2013, Figure 11). Richards et al. (2011), however, treat
the stagnant slab above the 660 as a slab remnant detached from the Vanuatu trench. In South America,
both our model and UU07P show a dipping slab in the upper mantle but the slab in our model penetrates
deeper into the lower mantle. Possibly due to poor ray coverage, models GAPP4 and MIT08P have less clear
subducting slabs in the upper mantle and little continuation of slab into the lower mantle.
Compared with the other models, the most significant difference with our model is the higher amplitude of
velocity anomalies within subducting slabs. In our model, the amplitude of P wave velocity anomalies inside
the slab in the upper mantle and transition zone are mostly higher than 2% while in the other models the
anomalies are less than 1.5%. The difference is more significant in the regions where ray coverage is limited.
The difference, of course, is because we include slabs in our starting model (Figure S6). The differences this

LU ET AL. 11,556
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 6. Representative epicenter shifts derived from velocity and source location joint inversion. The circles show the
original earthquake epicenters. The red lines show the direction and amplitude of epicenter shifts. Only earthquakes
with more than 250 travel time residuals and a focal depth greater than 100 km are shown for better visualization.

causes elsewhere in the model are smaller in amplitude but can still be significant. Given that the slabs in our
model are consistent with regional studies and theoretical modeling, and that the data variance reduction is
the same or slightly improved over a model derived without the slabs in the starting model, we feel that our
new model has some advantages over previous models.
The comparison of our model with other S wave models also shows similar features (Figure 8). Generally, S
wave models are longer wavelength and narrow slabs, such as in our model, are not well resolved. The slabs
we start with are consistent with geologic inferences and also result in a slightly higher variance reduction of

Figure 7. Comparison of cross sections across four major subduction zones in representative P wave tomography models.
Each column shows specific tomography models, including TX2019_slab (this study), GAPP4 (Obayashi et al., 2013),
MIT08P (Li et al., 2008), and UU07P (Amaru, 2007). Each row corresponds to cross‐section locations indicated in Figure 4.
Solid black lines show 410‐, 660‐, and 1,000‐km depth, respectively.

LU ET AL. 11,557
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 8. Comparison of cross sections across major subduction zones in representative S wave tomography models. Each
column shows specific tomography models, including TX2019_slab (this study), S40rts (Ritsema et al., 2011), UCBSEM
(French & Romanowicz, 2014), and S362+M (Moulik & Ekström, 2014). Each row corresponds to cross‐section locations
indicated in Figure 4. Solid black lines show 410‐, 660‐, and 1,000‐km depth, respectively.

the S wave data. Our S wave model shows similar features as seen in our P wave model, including dipping,
stagnant, and penetrated slabs (Figure S6). The other three S wave models, however, only show very long‐
wavelength fast velocity anomalies, which makes it challenging to identify the location and shape of sub-
ducting slabs. Also, the amplitude of the subducting slabs in our S wave model are mostly higher than 3%,
which are again much higher than in the other models.
The S to P heterogeneity ratio, defined as

∂lnV S
RV S =V P ¼ (4)
∂lnV P

has been widely used as an important diagnostic parameter to determine the compositional state of the
Earth's mantle, especially in the deeper mantle (e.g., Della Mora et al., 2011; Houser et al., 2008; Karato &
Karki, 2001; Koelemeijer et al., 2016; Masters et al., 2000; Robertson & Woodhouse, 1996; Saltzer et al.,
2001; Tesoniero et al., 2016). Several approaches have been used to calculate the S to P mantle heterogeneity
ratio (see Tesoniero et al., 2016 for a summary). It has been shown that different approaches can lead to dif-
ferent heterogeneity ratios, even using the same tomography model (Koelemeijer et al., 2016). We used two
methods to determine the 1‐D S to P heterogeneity as a function a depth using our TX2019slab model
(Figures 9 and S7). In the first method, a point to point division is computed except for regions where Vp
or Vs variations are less than 0.1%. Regions that have opposite sign Vp and Vs anomalies, which contribute
to about 5% of the volume of the mantle, were also excluded. The median values at each depth are chosen
as the 1‐D heterogeneity ratio. In the second method, we divided the RMS of the velocity variations for P
and S at each depth after excluding small or opposite variations as in the first method (Figure 9).
Throughout most of the mantle, the S to P ratio derived by RMS division is higher than by point to point divi-
sion. A similar observation is seen in the comparison done by Koelemeijer et al. (2016) using tomography
model KRDH16. This is due to the fact that the RMS value is more sensitive to outliers than the median
value. The LLSVPs have large shear anomalies with smaller P anomalies and are the primary cause of the
difference between the two methods in the deepest mantle. Even though the two methods described above
lead to different 1‐D S to P ratios, both S to P ratio models show similar trends.

LU ET AL. 11,558
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 9. (a) S to P heterogeneity ratio derived using the TX2019slab model as a function of depth. Two S to P ratio profiles are derived by point to point (P2P)
division and RMS value division, respectively. (b) Distribution of P2P division results at selected depths.

Lu and Grand (2016) showed that earthquake mislocation caused by subducting slabs could bias the tomo-
graphy results in the lower mantle. This bias has the potential to further affect our estimation of the S to P
ratio. To evaluate this effect, we also derived the S to P heterogeneity ratio for model TX2019 model using the
two approaches described above (Figure 10). We find very little difference in the S to P heterogeneity ratio
between the slab model and the model without starting slabs. We also calculated the point to point S to P
ratio differences between the TX2019slab and TX2019 models and plot them in map view in Figure 11.
The S to P ratio can differ by more than 0.5 between the models in specific regions. This is a significant dif-
ference since the lateral average S to P ratio in the lower mantle is between 1.5 and 3.5 in our model. Our
analysis shows that the 1‐D average of P to S heterogeneity ratio is unlikely to be significantly affected by
the bias introduced by unmodeled subducting slabs. However, the bias may be quite large for the S to P het-
erogeneity ratio on a regional scale.
In the deeper mantle, our 1‐D S to P heterogeneity ratio profiles show an increase with depth, which is con-
sistent with most previous tomography studies (Antolik et al., 2003; Della Mora et al., 2011; Ishii & Tromp,
1999; Koelemeijer et al., 2016; Masters et al., 2000; Mosca et al., 2012; Resovsky & Trampert, 2003;
Romanowicz, 2001; Su & Dziewonski, 1997; Tesoniero et al., 2016) (Figure 12). We followed Koelemeijer
et al. (2016) and used the RMS division method for all the models in Figure 12. The S to P heterogeneity ratio
profiles derived using body wave seismic data alone are plotted separately (Figure 12a) from results that also
used normal model data (Figure 12b). Among these profiles, the RV S =V P range from about 1.5 to 4 throughout
the lower mantle. The RV S =V P profiles vary more dramatically in the deep lower mantle than at shallower
depths among the different models. In particular, most of the RV S =V P profiles agree with each other above
1,500‐km depth. These models were produced using different data sets and different inversion strategies;
therefore, the similarity indicates a convergence of RV S =V P estimations in this depth range. In the deep lower
mantle, RV S =V P in KRDH16 (Koelemeijer et al., 2016) increases rapidly with depth and reaches ~4 near 2,450
km, then decreases again to the CMB. This feature is not seen in the other models. Most of the other models,
including TX2019slab, show a gradual increasing RV S =V P with depth in this depth range. In contrast, there are
several models have a small decrease above the CMB, including DBTNG11 (Della Mora et al., 2011), R01
(Romanowicz, 2001), TCB16 (Tesoniero et al., 2016), and SFBG10 (Simmons et al., 2010).
Figure 12c compares our RV S =V P in the lower mantle with predictions assuming that seismic heterogeneity
is due to thermal variations alone. All the predicted RV S =V P profiles in Figure 12c assume a simple

LU ET AL. 11,559
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 10. Comparison of S to P heterogeneity ratios using TX2019slab (solid lines) and TX2019 (dash lines). S to P ratio
profiles are derived by point to point (P2P) division (green lines) and RMS value division (red lines), respectively.

“pyrolite” mineral assemblage model (e.g., Stixrude & Lithgow‐Bertelloni, 2012) without invoking more
complex proposed scenarios in the lower mantle such as spin transition effects (Lin et al., 2013; Wu &
Wentzcovitch, 2014) or the phase transition from bridgmanite to post‐perovskite (pPv; Murakami et al.,
2004; Oganov & Ono, 2004; Tsuchiya et al., 2004). The estimate of RV S =V P by Karato and Karki (2001) is
lower than values we obtained using the RMS division method from about 1,500‐km depth and below

Figure 11. The difference in point to point S to P heterogeneity ratio between the TX2019slab and TX2019 models in map
view at representative depths. The S to P heterogeneity ratio was calculated in each individual block. Although the 1‐D
average S to P heterogeneity ratio is not significantly different between these two models, the incorrect imaging of sub-
ducting slabs could bias the tomography results in the lower mantle in specific regions, as shown by high‐amplitude
regions in this figure.

LU ET AL. 11,560
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 12. Comparison of S to P heterogeneity ratios in the lower mantle from (a) pure body wave seismic study, (b) normal mode data involved seismic study, and
(c) mineral physics predictions assuming a thermal cause for heterogeneity in the lower mantle. Results in this study are shown using red solid (RMS division
method) and green solid lines (point to point method). S to P heterogeneity ratio derived from mineral physics modeling from this study and used in the P and S joint
inversion is also shown as a red dot dash line. (a) Results from previous pure body wave studies include AGED03 (Antolik et al., 2003), HMSL08 (Houser et al.,
2008), SD97 (Su & Dziewonski, 1997), SHK01 (Saltzer et al., 2001), DBTNG11 (Della Mora et al., 2011), SFBG10 (Simmons et al., 2010), and TCB16 (Tesoniero et al.,
2016). All the S to P heterogeneity ratios were obtained using the RMS value division method. (b) Results from models that used normal mode data include MLDB00
(Masters et al., 2000), IT99 (Ishii & Tromp, 1999), R01 (Romanowicz, 2001), KRDH16 (Koelemeijer et al., 2016), and MCDRT12 (Mosca et al., 2012). All the S to P
heterogeneity ratios again were obtained using the RMS value division method. (c) Thermally induced S to P heterogeneity ratios derived from mineral physics
modeling including the effect of anelasticity (Trampert et al., 2001; Karato, 1993; Karato & Karki, 2001; Yang et al., 2016) are shown in colored dot dash lines.

but is similar to values we derived using point to point division. However, the Karato and Karki (2001)
estimates are significantly higher than other mineral physics predictions (Karato, 1993; Trampert et al.,
2001; Yang et al., 2016) as well as ours. Two studies have argued that Karato and Karki (2001)
overestimated the effect of anelasticity in the lower mantle, which would overpredict RV S =V P since the
temperature derivative of shear wave velocity is more sensitive to the anelasticity effect (Brodholt et al.,
2007; Matas & Bukowinski, 2007). If this is the case, our derived RV S =V P implies factors other than
simple thermal variations, starting from near 1,500‐km depth, contribute to seismic heterogeneity. We
also plotted the predicted RV S =V P from mineral physics modeling in this study in Figures 12a and 12b.
All of the seismically derived RV S =V P profiles have higher values than mineral physics predictions in the
deep lower mantle, which implies the existence of more complex thermal‐chemical heterogeneities,
although the amount and depth of these heterogeneities still vary among models and methods for
estimating 1‐D RV S =V P .
Anelasticity in the Earth for P and S waves is quantified by quality factors QP and QS, respectively. The qual-
ity factor Q due to viscoelastic relaxation can be written as (Karato, 1993)
 
α αH *
Qðω; T Þ ¼ Q0 ω exp (5)
RT

where ω is the seismic frequency, α is the exponent describing the frequency dependence of the attenuation,
T is the temperature, R is the gas constant, H* is the activation enthalpy, and Q0 is a normalization constant
that can be constrained using seismically observed attenuation. Assuming that bulk attenuation is

LU ET AL. 11,561
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

negligible, the relationship between elastic (“el”) and anelastic (“anel”) values of RV S =V P can be given by
(Matas & Bukowinski, 2007)
 
AP 3V P 2
RV S =V P anel
¼ RV S =V P el
þ −RV S =V P el
(6)
V P;T þ AP 4V S 2

where

αH * cot ðαπ=2Þ
AP ¼ − (7)
2QP RT 2

∂lnV P el
V P;T ¼ (8)
∂T

The measurements of several critical anelasticity‐related parameters, including H*and α, still vary a lot
among different studies (see Brodholt et al. (2007) and Matas and Bukowinski (2007) for summaries).
Therefore, the predicted RV S =V P from mineral physics after taking anelasticity into account may contain sig-
nificant uncertainties. In Figure 13, we evaluated the potential effect of anelasticity on predicted RV S =V P to
check whether the uncertainty in anelasticity could explain the apparent discrepancies between seismic
observed and mineral physics predicted RV S =V P in the deep lower mantle (Figure 12c). To make the compar-
ison more diagnostic, we plotted the variation of RV S =V P as a function of S velocity perturbation at selected
depths instead of using a 1‐D profile. We varied the anelasticity‐related parameters within their possible
ranges to fully explore the possible anelasticity effect. Figures 13b (and 13c only show estimates at 2,000‐
km depth but we found similar results at all other depths. According to our test, when the anelasticity effect
is strong, mineral physics predicted RV S =V P becomes larger, but the amount of variation is limited. For exam-
ple, the maximum RV S =V P is ~2.5 at 2,000‐km depth, which is comparable to our observed 1‐D RV S =V P in
TX2019slab (Figure 12). But this maximum value requires the combination of extreme values for multiple
parameters (i.e., H* = 800 kJ/mol, α = 0.4), that Brodholt et al. (2007) claim unlikely. For better visualization,
we binned and averaged seismically derived RV S =V P based on a 0.2% S velocity perturbation interval and
applied a moving average filter in Figure 13a for comparison. At all selected depths, the seismically derived
RV S =V P profiles show “V” shapes: RV S =V P values increase with increasing S wave perturbation, both in the
positive and negative directions. The RV S =V P for large S wave perturbation regions (larger than 1% for exam-
ple) can easily be over 2.8 and even reach 5. Tesoniero et al. (2016) has shown that the “V” shape is a common
feature in all the tomography models they tested and argued that the V shape could be a result of errors in
seismic tomography. The effect of small errors on RV S =V P values can be amplified when the true velocity per-
turbation is small because of the division between S and P wave anomalies (see Tesoniero et al. (2016) for
more discussion). If Tesoniero et al. (2016) are correct, it means that the RV S =V P in the large S wave perturba-
tion regions are more reliable than in the small‐velocity perturbation regions. In other words, the actual
RV S =V P could be higher than our 1‐D seismic observation in Figure 12, which means that anelasticity effects
alone cannot explain the observed RV S =V P assuming that heterogeneity is due to thermal effects in a chemi-
cally homogeneous mantle. More complex thermal‐chemical structures are needed to explain our seismic
observed RV S =V P in the deep lower mantle.
An interesting feature in our S to P heterogeneity profile is that the ratio increases starting from 1,400‐km
depth and reaches a local maximum around 1,800‐km depth, stays relatively constant to 2,100‐km depth,
then decreases before a large jump near the CMB. Both our derived S to P heterogeneity ratios have this pat-
tern, although the S to P ratio derived by RMS division shows this feature more clearly. Model SD97 (Su &
Dziewonski, 1997) and KRDH16 (Koelemeijer et al., 2016) show a similar trend above 2,100‐km depth but
RV S =V P continues to increase with depth below 2,100 km. Several other models have similar peaks but at dee-
per depth, including SFBG10 (Simmons et al., 2010), SHK01 (Saltzer et al., 2001), IT99 (Ishii & Tromp, 1999),
(Romanowicz, 2001), and MLDB00 (Masters et al., 2000). Wu and Wentzcovitch (2014) reported that the P
wave velocity of pyrolitic lower mantle becomes insensitive to temperature variations due to the spin transi-
tion between ~1,400‐ and ~2,100‐km depth. This would increase the S to P heterogeneity there even if varia-
tions were solely due to thermal effects. Thus, the abnormal RV S =V P we observe between 1,400 and 2,100 km

LU ET AL. 11,562
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 13. Comparison of seismically observed S to P heterogeneity ratios versus S velocity perturbations and mineral physics predictions assuming different ane-
lasticity parameters. (a) Results calculated for TX2019slab model at selected depths. The average scaling factors are calculated for each 0.2% velocity perturbation
*
interval. A three‐point moving‐average filter was applied before plotting for better visualization. (b) Mineral physics predictions at 2,000‐km depth with varying H .
α is fixed at 0.2 and Q0 is derived following Matas and Bukowinski (2007) method assuming average lateral temperature variation dT/T = 0.4 and seismic observed
quality factor Qs = 350. (c) Similar as (b) but α is fixed at 0.4.

may partially be due to the spin transition in a mantle with heterogeneity dominated by temperature
variations. In Figure 14 we compare the predicted RV S =V P from Wu and Wentzcovitch (2014) to our
results. Our result matches the mineral physics prediction for an aggregate consisting of 5 wt %
ferropericlase (Mg0.92Fe0.08)SiO3 and the rest bridgmanite the best, which generally agrees with a
bridgmanite‐enriched mantle model proposed by Murakami et al. (2012). However, the modeling result of
Wu and Wentzcovitch (2014) may oversimplify the effect of the spin transition in the mantle since it does
not consider several critical effects such as iron‐partitioning variations crossing the spin transition (e.g.,
Kobayashi et al., 2005; Lin et al., 2013), and the spin transition in Fe‐
bearing bridgmanite (e.g., Chantel et al., 2012; Fu et al., 2018). We leave
to further work a more detailed analysis of the effect of the spin
transition on the S to P heterogeneity ratio.
Karato and Karki (2001) claim that RV S =V P cannot exceed 2.7 for an iso-
chemical deep lower mantle. Our model, as well as several other mod-
els shown in Figure 12, shows the bottom 200 km of the mantle to
have heterogeneity ratios well above 2.7. One major limitation of the
mineral physics predictions we use is that they did not consider poten-
tial phase transitions occurring in the lower mantle. Bridgmanite could
transform to pPv in the deepest ~400‐km mantle (Murakami et al.,
2004; Oganov & Ono, 2004; Tsuchiya et al., 2004). Compared with
bridgmanite, pPv has similar P velocity and higher S velocity
(Tsuchiya et al., 2004; Wookey et al., 2005). Wentzcovitch et al.
(2006) reported a very large RV S =V P (>6) caused by the pPv transition
along their predicted phase boundary (Figure 14). Lateral variation in
phase abundances can explain the high RV S =V P we found in our model
in the bottom few hundred kilometers of the mantle. The amplitude
differences between the results of Wentzcovitch et al. (2006) and seis-
Figure 14. Comparison of the S to P heterogeneity ratios in the TX2019slab
model and mineral physics predictions including the effects of the spin mic observations could be caused by uncertainties in the mineral phy-
transition and the pPv phase transition. The mineral physics predictions sics modeling, such as the pressure and temperature sensitivity of the
accounting for the spin transition (dash lines) are adopted from Wu and bridgmanite to pPv phase transition (Cobden et al., 2015), and the pos-
Wentzcovitch (2014). Thermally induced S to P heterogeneity ratios in sible presence of subducted MORB or iron at the lowest mantle
aggregates along adiabatic (Adi) and superadiabatic (SAdi) geotherms are
(Grocholski et al., 2012; Tateno et al., 2007).
shown. Aggregates consist of 5 or 10 wt % ferropericlase (Mg0.92Fe0.08)SiO3
and bridgmanite. The predicted S to P heterogeneity ratio due to the bridg- Figure 15 shows radial depth profiles of the correlation coefficient (CC)
manite to pPv transition along the phase boundary (solid black line) is between P and S wave perturbations, the RMS velocity anomalies for P
shown with uncertainty (shaded purple area) from Wentzcovitch et al.
and S, and the CC between bulk sound and shear speed perturbations.
(2006). The light yellow background color indicates the proposed depth
range for the spin transition effect and light background below shows the P and S velocity heterogeneities are highly correlated throughout the
depths where the pPv phase transition could occur. mantle in our model (Figure 15a). The RMS amplitude of S wave

LU ET AL. 11,563
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Figure 15. Depth profiles calculated for the TX2019slab model. (a) Correlation between P and S wave heterogeneities. (b)
RMS amplitude of P and S wave anomalies. (c) Correlation between bulk and shear wave anomalies. Bulk and shear
speeds become decorrelated in the shallow lower mantle (~700‐ to ~1,500‐km depth) and anticorrelated below 1,500‐km
depth.

perturbations is higher than P waves at all depths, which agrees with previous studies (e.g., Koelemeijer
et al., 2016; Simmons et al., 2010; Figure 15b). The depth profile of the correlation between bulk sound
and shear speeds shows a more complex variation (Figure 15c). In the upper mantle, the bulk sound
speed correlates well with shear speed. In the shallow and mid‐lower mantle, the bulk sound
anomalies are almost decorrelated with S wave anomalies. Sound and shear speeds become
anticorrelated below ~1,500‐km depth. Similar anticorrelation of sound and shear speeds in the deep
lower mantle has been reported in several other tomography studies (e.g., Koelemeijer et al., 2016;
Masters et al., 2000; Simmons et al., 2010; Su & Dziewonski, 1997). The anticorrelation between
shear wave and bulk sound speed anomalies has been used to argue for chemical variations in the
deep mantle. However, previous investigations have shown that both the spin transition effect (Wu &
Wentzcovitch, 2014) and the phase transition from bridgmanite to pPv (Wookey et al., 2005) can
cause anticorrelation between sound and shear speeds in the deep mantle. Thus, the anticorrelation
we observe in our model could still be explained by thermal effects alone for an isochemical mantle,
although it likely requires a strong effect of the spin transition.
It is still challenging to draw concrete conclusions about the thermal‐chemical structure in the lower mantle
using P‐S tomography results alone. Although the observed RV S =V P is higher in the deep mantle than pre-
dicted by simple calculations assuming that heterogeneity is due to thermal effects alone, the effects of the
spin transition, phase changes, and anelasticity can increase the scaling beyond what has been considered
in the past for an isochemical mantle. Detailed evaluation of several critical topics are still lacking to defini-
tively rule out a thermal explanation for mantle heterogeneity. These include the effect of iron‐partitioning
variations crossing the spin transition (e.g., Kobayashi et al., 2005; Lin et al., 2013), the spin transition in Fe‐
bearing bridgmanite (e.g., Chantel et al., 2012; Fu et al., 2018), the pressure and temperature sensitivity of
the bridgmanite to pPv phase transition (Cobden et al., 2015), and the vertical smearing effect in tomography
models.

LU ET AL. 11,564
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

5. Conclusions
In this study, we produce a new P and S global tomography model. The most significant improvement in our
inversion, relative to our past work, is that we include a priori 3‐D subducting slabs in our starting model and
inverted for velocity and source parameters simultaneously. Both of our P and S wave models feature higher‐
amplitude subducting slabs compared with previous global tomography results, which better matches obser-
vations using other approaches. We also calculated the S to P heterogeneity ratio using our model. Although
the 1‐D S to P heterogeneity ratio is not significantly different using models with starting slabs relative to
models without, we found that the incorrect imaging of subducting slabs could bias the tomography results
in the lower mantle in specific regions. Our derived S to P ratio features a broad peak at around 1,800‐ to
2,100‐km depth, which is consistent with mineral physics predictions of the spin transition effect. The high
S to P ratio right above CMB can be due to bridgmanite to pPv phase change. However, our observations
alone cannot rule out the possibility of the existence of chemical heterogeneities in the deep lower mantle.

Acknowledgments References
The TX2019slab model is available at
IRIS EMC website (http://ds.iris.edu/ Amaru, M. L. (2007). Global travel time tomography with 3‐D reference model (Doctoral dissertation). Utrecht, Netherlands: Utrecht
ds/products/emc‐earthmodels/). We University.
thank ISC (International Seismological Antolik, M., Gu, Y. J., Ekström, G., & Dziewonski, A. M. (2003). J362D28: a new joint model of compressional and shear velocity in the
Centre, http://www.isc.ac.uk/isc‐ehb/), Earth's mantle. Geophysical Journal International, 153(2), 443–466.
IRIS DMC (Incorporated Research Billen, M. I. (2008). Modeling the dynamics of subducting slabs. Annual Review Earth Planetary Sciences, 36, 325–356.
Institutions for Seismology, Data Bonnardot, M. A., Régnier, M., Christova, C., Ruellan, E., & Tric, E. (2009). Seismological evidence for a slab detachment in the Tonga
Management Center, https://ds.iris. subduction zone. Tectonophysics, 464(1), 84–99. https://doi.org/10.1016/j.tecto.2008.10.011
edu/ds/nodes/dmc/), ORFEUS Brodholt, J. P., Helffrich, G., & Trampert, J. (2007). Chemical versus thermal heterogeneity in the lower mantle: The most likely role of
(Observations & Research Facilities for anelasticity. Earth and Planetary Science Letters, 262(3), 429–437. https://doi.org/10.1016/j.epsl.2007.07.054
European Seismology, https://www. Brudzinski, M. R., & Chen, W.‐P. (2005). Earthquakes and strain in subhorizontal slabs. Journal of Geophysical Research, 110(B8), B08303.
orfeus‐eu.org/), NECDC (Northern https://doi.org/10.1029/2004JB003470
California Earthquake Data Center, Chantel, J., Frost, D. J., McCammon, C. A., Jing, Z., & Wang, Y. (2012). Acoustic velocities of pure and iron‐bearing magnesium silicate
http://ncedc.org/), F‐net (F‐net perovskite measured to 25 GPa and 1200 K. Geophysical Research Letters, 39, L19307. https://doi.org/10.1029/2012GL053075
Broadband Seismograph Network, Chen, M., Tromp, J., Helmberger, D., & Kanamori, H. (2007). Waveform modeling of the slab beneath Japan. Journal of Geophysical
http://www.fnet.bosai.go.jp/), and Research, 112, B02305. https://doi.org/10.1029/2006JB004394
CNSN (Canadian National Seismic Cobden, L., Thomas, C., & Trampert, J. (2015). Seismic Detection of Post‐perovskite Inside the Earth. In A. Khan, & F. Deschamps (Eds.),
Network, http://www. The Earth's Heterogeneous Mantle (pp. 391–440). Switzerland: Springer.
earthquakescanada.nrcan.gc.ca/ Creager, K. C., & Jordan, T. H. (1984). Slab penetration into the lower mantle. Journal of Geophysical Research, 89, 3031–3049. https://doi.
stndon/CNDC/) for providing seismic org/10.1029/JB089iB05p03031
data used in this study. We thank Georg Della Mora, S., Boschi, L., Tackley, P., Nakagawa, T., & Giardini, D. (2011). Low seismic resolution cannot explain S/P decorrelation in the
Stadler, Michael Gurnis, and Ellen lower mantle. Geophysical Research Letters, 38, L12303. https://doi.org/10.1029/2011GL047559
Syracuse for providing thermal slab Deschamps, F., Trampert, J., & Snieder, R. (2002). Anomalies of temperature and iron in the uppermost mantle inferred from gravity
model. We also thank Nathan data and tomographic models. Physics of the Earth and Planetary Interiors, 129(3), 245–264. https://doi.org/10.1016/S0031‐9201(01)
Simmons, Alessandro Forte, Jung‐Fu 00294‐1
Lin, Thorsten Becker, and Suyu Fu for Ding, X. Y., & Grand, S. P. (1994). Seismic structure of the deep Kurile subduction zone. Journal of Geophysical Research, 99(B12),
the valuable discussions. This research 23,767–23,786. https://doi.org/10.1029/94JB02130
was supported by National Science Engdahl, E. R., van der Hilst, R. D., & Buland, R. (1998). Global teleseismic earthquake relocation with improved travel times and proce-
Foundation (NSF) grants EAR‐1648770 dures for depth determination. Bulletin of the Seismological Society of America, 88(3), 722–743.
and EAR‐1648817 and by the Jackson Forte, A. M., & Perry, H. C. (2000). Geodynamic evidence for a chemically depleted continental tectosphere. Science, 290(5498), 1940–1944.
School of Geosciences at the University https://doi.org/10.1126/science.290.5498.1940
of Texas at Austin. French, S. W., & Romanowicz, B. A. (2014). Whole‐mantle radially anisotropic shear velocity structure from spectral‐element waveform
tomography. Geophysical Journal International, 199(3), 1303–1327. https://doi.org/10.1093/gji/ggu334
Fu, S., Yang, J., Zhang, Y., Okuchi, T., McCammon, C., Kim, H.‐I., et al. (2018). Abnormal elasticity of Fe‐bearing bridgmanite in the
Earth's lower mantle. Geophysical Research Letters, 45(10), 4725–4732. https://doi.org/10.1029/2018GL077764
Fukao, Y., & Obayashi, M. (2013). Subducted slabs stagnant above, penetrating through, and trapped below the 660 km discontinuity.
Journal of Geophysical Research: Solid Earth, 118, 5920–5938. https://doi.org/10.1002/2013JB010466
Gaherty, J. B., Lay, T., & Vidale, J. E. (1991). Investigation of deep slab structure using long‐period S waves. Journal of Geophysical Research,
96(B10), 16349–16367. https://doi.org/10.1029/91JB01483
Garnero, E. J., McNamara, A. K., & Shim, S.‐H. (2016). Continent‐sized anomalous zones with low seismic velocity at the base of Earth's
mantle. Nature Geoscience, 9, 481–489. https://doi.org/10.1038/ngeo2733
Grand, S. P. (1994). Mantle shear structure beneath the Americas and surrounding oceans. Journal of Geophysical Research, 99,
11,591–11,621. https://doi.org/10.1029/94JB00042
Grand, S. P. (2002). Mantle shear–wave tomography and the fate of subducted slabs. Philosophical Transactions of the Royal Society of
London A: Mathematical, Physical and Engineering Sciences, 360(1800), 2475–2491.
Grand, S. P., & Helmberger, D. V. (1984). Upper mantle shear structure of North America. Geophysical Journal International, 76(2), 399–438
. https://doi.org/10.1111/j.1365‐246X.1984.tb05053.x
Grocholski, B., Catalli, K., Shim, S.‐H., & Prakapenka, V. (2012). Mineralogical effects on the detectability of the postperovskite boundary.
Proceedings of the National Academy of Science, 109(7), 2275–2279.
Hayes, G. P., Wald, D. J., & Johnson, R. L. (2012). Slab1. 0: A three‐dimensional model of global subduction zone geometries. Journal of
Geophysical Research, 117(B1), B01302. https://doi.org/10.1029/2011JB008524
Houser, C., Masters, G., Shearer, P., & Laske, G. (2008). Shear and compressional velocity models of the mantle from cluster analysis of
long‐period waveforms. Geophysical Journal International, 174(1), 195–212. https://doi.org/10.1111/j.1365‐246X.2008.03763.x

LU ET AL. 11,565
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Ishii, M., & Tromp, J. (1999). Normal‐mode and free‐air gravity constraints on lateral variations in velocity and density of Earth's mantle.
Science, 285(5431), 1231–1236. https://doi.org/10.1126/science.285.5431.1231
Karato, S.‐i. (1993). Importance of anelasticity in the interpretation of seismic tomography. Geophysical Research Letters, 20(15), 1623–1626.
https://doi.org/10.1029/93GL01767
Karato, S.‐i., & Karki, B. B. (2001). Origin of lateral variation of seismic wave velocities and density in the deep mantle. Journal of
Geophysical Research, 106(B10), 21,771–21,783. https://doi.org/10.1029/2001JB000214
Kawakatsu, H., & Yoshioka, S. (2011). Metastable olivine wedge and deep dry cold slab beneath southwest Japan. Earth and Planetary
Science Letters, 303(1), 1–10. https://doi.org/10.1016/j.epsl.2011.01.008
Kellogg, L. H., Hager, B. H., & van der Hilst, R. D. (1999). Compositional Stratification in the Deep Mantle. Science, 283(5409), 1881–1884.
https://doi.org/10.1126/science.283.5409.1881
Kennett, B. L. N., Engdahl, E. R., & Buland, R. (1995). Constraints on seismic velocities in the Earth from traveltimes. Geophysical Journal
International, 122(1), 108–124. https://doi.org/10.1111/j.1365‐246X.1995.tb03540.x
Kobayashi, Y., Kondo, T., Ohtani, E., Hirao, N., Miyajima, N., Yagi, T., et al. (2005). Fe‐Mg partitioning between (Mg, Fe)SiO3 post‐
perovskite, perovskite, and magnesiowüstite in the Earth's lower mantle. Geophysical Research Letters, 32, L19301. https://doi.org/
10.1029/2005GL023257
Koelemeijer, P., Ritsema, J., Deuss, A., & van Heijst, H. J. (2016). SP12RTS: a degree‐12 model of shear‐ and compressional‐wave velocity
for Earth's mantle. Geophysical Journal International, 204(2), 1024–1039. https://doi.org/10.1093/gji/ggv481
Komatitsch, D., & Tromp, J. (2002a). Spectral‐element simulations of global seismic wave propagation—I. Validation. Geophysical Journal
International, 149(2), 390–412.
Komatitsch, D., & Tromp, J. (2002b). Spectral‐element simulations of global seismic wave propagation—II. Three‐dimensional models,
oceans, rotation and self‐gravitation. Geophysical Journal International, 150(1), 303–318.
Kustowski, B., Ekström, G., & Dziewoński, A. (2008). Anisotropic shear‐wave velocity structure of the Earth's mantle: A global model.
Journal of Geophysical Research, 113, B06306. https://doi.org/10.1029/2007JB005169
Lai, H., Garnero, E. J., Grand, S. P., Porritt, R. W., & Becker, T. W. (2019). Global travel time dataset from adaptive empirical wavelet
construction. Geochemistry, Geophysics, Geosystems, 20. https://doi.org/10.1029/2018GC007905
Laske, G., Masters, G., Ma, Z., & Pasyanos, M. (2013). Update on CRUST1. 0‐A 1‐degree global model of Earth's crust. Paper presented at
EGU General Assembly, Vienna, Austria.
Lee, C. T. (2003). Compositional variation of density and seismic velocities in natural peridotites at STP conditions: Implications for seismic
imaging of compositional heterogeneities in the upper mantle. Journal of Geophysical Research, 108(B9), 2441, ECV 6‐1‐6‐20. https://doi.
org/10.1029/2003JB002413
Li, C., van der Hilst, R. D., Engdahl, E. R., & Burdick, S. (2008). A new global model for P wave speed variations in Earth's mantle.
Geochemistry, Geophysics, Geosystems, 9, Q05018. https://doi.org/10.1029/2007GC001806
Lin, J.‐F., Speziale, S., Mao, Z., & Marquardt, H. (2013). Effects of the electronic spin transitions of iron in lower mantle minearls: impli-
cations for deep mantle geophysics and geochemistry. Reviews of Geophysics, 51, 244–275. https://doi.org/10.1002/rog.20010
Lu, C., & Grand, S. P. (2016). The effect of subducting slabs in global shear wave tomography. Geophysical Journal International, 205(2),
1074–1085. https://doi.org/10.1093/gji/ggw072
Masters, G., Laske, G., Bolton, H., & Dziewonski, A. (2000). The Relative Behavior of Shear Velocity, Bulk Sound Speed, and Compressional
Velocity in the Mantle: Implications for Chemical and Thermal Structure. In S. i. Karato, A. Forte, R. Liebermann, G. Masters, & L.
Stixrude (Eds.), Earth's Deep Interior: Mineral Physics and Tomography From the Atomic to the Global Scale, Geophysical Monograph
Series (Vol. 117, pp. 63–87). Washington, DC: American Geophysical Union.
Matas, J., & Bukowinski, M. S. T. (2007). On the anelastic contribution to the temperature dependence of lower mantle seismic velocities.
Earth and Planetary Science Letters, 259(1), 51–65. https://doi.org/10.1016/j.epsl.2007.04.028
Mosca, I., Cobden, L., Deuss, A., Ritsema, J., & Trampert, J. (2012). Seismic and mineralogical structures of the lower mantle from prob-
abilistic tomography. Journal of Geophysical Research, 117, B06304. https://doi.org/10.1029/2011JB008851
Moulik, P., & Ekström, G. (2014). An anisotropic shear velocity model of the Earth's mantle using normal modes, body waves, surface
waves and long‐period waveforms. Geophysical Journal International, 199(3), 1713–1738.
Murakami, M., Hirose, K., Kawamura, K., Sata, N., & Ohishi, Y. (2004). Post‐perovskite phase transition in MgSiO3. Science, 304(5672),
855–858. https://doi.org/10.1126/science.1095932
Murakami, M., Ohishi, Y., Hirao, N., & Hirose, K. (2012). A perovskitic lower mantle inferred from high‐pressure, high‐temperature sound
velocity data. Nature, 485, 90–94.
Obayashi, M., Yoshimitsu, J., Nolet, G., Fukao, Y., Shiobara, H., Sugioka, H., et al. (2013). Finite frequency whole mantle P wave
tomography: Improvement of subducted slab images. Geophysical Research Letters, 40, 5652–5657. https://doi.org/10.1002/
2013GL057401
Oganov, A. R., & Ono, S. (2004). Theoretical and experimental evidence for a post‐perovskite phase of MgSiO3 in Earth's D″ layer. Nature,
430(6998), 445–448.
Panning, M., & Romanowicz, B. (2006). A three‐dimensional radially anisotropic model of shear velocity in the whole mantle. Geophysical
Journal International, 167(1), 361–379. https://doi.org/10.1111/j.1365‐246X.2006.03100.x
Perry, H. K. C., Forte, A. M., & Eaton, D. W. S. (2003). Upper‐mantle thermochemical structure below North America from seismic–
geodynamic flow models. Geophysical Journal International, 154(2), 279–299. https://doi.org/10.1046/j.1365‐246X.2003.01961.x
Resovsky, J., & Trampert, J. (2003). Using probabilistic seismic tomography to test mantle velocity‐density relationships. Earth and
Planetary Science Letters, 215(1), 121–134. https://doi.org/10.1016/S0012‐821X(03)00436‐9
Richards, S., Holm, R., & Barber, G. (2011). When slabs collide: A tectonic assessment of deep earthquakes in the Tonga‐Vanuatu region.
Geology, 39(8), 787–790. https://doi.org/10.1130/G31937.1
Ritsema, J., Deuss, A., van Heijst, H. J., & Woodhouse, J. H. (2011). S40RTS: a degree‐40 shear‐velocity model for the mantle from new
Rayleigh wave dispersion, teleseismic traveltime and normal‐mode splitting function measurements. Geophysical Journal International,
184, 1223–1236. https://doi.org/10.1111/j.1365‐246X.2010.04884.x
Robertson, G., & Woodhouse, J. (1996). Ratio of relative S to P velocity heterogeneity in the lower mantle. Journal of Geophysical Research,
101(B9), 20041–20052. https://doi.org/10.1029/96JB01905
Romanowicz, B. (2001). Can we resolve 3D density heterogeneity in the lower mantle? Geophysical Research Letters, 28(6), 1107–1110.
https://doi.org/10.1029/2000GL012278
Saltzer, R. L., van der Hilst, R. D., & Karason, H. (2001). Comparing P and S wave heterogeneity in the mantle. Geophysical Research Letters,
28(7), 1335–1338. https://doi.org/10.1029/2000GL012339

LU ET AL. 11,566
 21699356, 2019, 11, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2019JB017448 by Cochrane Saudi Arabia, Wiley Online Library on [16/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2019JB017448

Simmons, N. A., Forte, A. M., Boschi, L., & Grand, S. P. (2010). GyPSuM: A joint tomographic model of mantle density and seismic wave
speeds. Journal of Geophysical Research, 115, B12310. https://doi.org/10.1029/2010JB007631
Simmons, N. A., Myers, S. C., Johannesson, G., & Matzel, E. (2012). LLNL‐G3Dv3: Global P wave tomography model for improved regional
and teleseismic travel time prediction. Journal of Geophysical Research, 117, B10302. https://doi.org/10.1029/2012JB009525
Sleep, N. H. (1973). Teleseismic P‐wave transmission through slabs. Bulletin of the Seismological Society of America, 63(4), 1349–1373.
Stadler, G., Gurnis, M., Burstedde, C., Wilcox, L. C., Alisic, L., & Ghattas, O. (2010). The Dynamics of Plate Tectonics and Mantle Flow:
From Local to Global Scales. Science, 329(5995), 1033–1038. https://doi.org/10.1126/science.1191223
Stixrude, L., & Lithgow‐Bertelloni, C. (2012). Geophysics of Chemical Heterogeneity in the Mantle. Annual Review of Earth and Planetary
Sciences, 40(1), 569–595. https://doi.org/10.1146/annurev.earth.36.031207.124244
Su, W.‐j., & Dziewonski, A. M. (1997). Simultaneous inversion for 3‐D variations in shear and bulk velocity in the mantle. Physics of the
Earth and Planetary Interiors, 100(1), 135–156. https://doi.org/10.1016/S0031‐9201(96)03236‐0
Syracuse, E. M., van Keken, P. E., & Abers, G. A. (2010). The global range of subduction zone thermal models. Physics of the Earth and
Planetary Interiors, 183(1), 73–90. https://doi.org/10.1016/j.pepi.2010.02.004
Tao, K., Grand, S. P., & Niu, F. (2018). Seismic structure of the upper mantle beneath eastern Asia from full waveform seismic tomography.
Geochemistry, Geophysics, Geosystems, 19, 2732–2763. https://doi.org/10.1029/2018GC007460
Tateno, S., Hirose, K., Sata, N., & Ohishi, Y. (2007). Solubility of FeO in (Mg, Fe)SiO3 perovskite and the post‐perovskite phase transition.
Physics of the Earth and Planetary Interiors, 160(3), 319–325.
Tesoniero, A., Cammarano, F., & Boschi, L. (2016). S‐to‐P heterogeneity ratio in the lower mantle and thermo‐chemical implications.
Geochemistry, Geophysics, Geosystems, 17, 2522–2538. https://doi.org/10.1002/2016GC006293
Trampert, J., Vacher, P., & Vlaar, N. (2001). Sensitivities of seismic velocities to temperature, pressure and composition in the lower mantle.
Physics of the Earth and Planetary Interiors, 124(3), 255–267. https://doi.org/10.1016/S0031‐9201(01)00201‐1
Tsuchiya, T., Tsuchiya, J., Umemoto, K., & Wentzcovitch, R. M. (2004). Phase transition in MgSiO3 perovskite in the earth's lower mantle.
Earth and Planetary Science Letters, 224(3‐4), 241–248.
van Keken, P. E., Hacker, B. R., Syracuse, E. M., & Abers, G. A. (2011). Subduction factory: 4. Depth‐dependent flux of H2O from sub-
ducting slabs worldwide. Journal of Geophysical Research, 116, B01401. https://doi.org/10.1029/2010JB007922
Vidale, J. E. (1987). Waveform effects of a high‐velocity, subducted slab. Geophysical Research Letters, 14(5), 542–545. https://doi.org/
10.1029/GL014i005p00542
Wang, T., Revenaugh, J., & Song, X. (2014). Two‐dimensional/three‐dimensional waveform modeling of subducting slab and transition
zone beneath Northeast Asia. Journal of Geophysical Research: Solid Earth, 119, 4766–4786. https://doi.org/10.1002/2014jb011058
Wentzcovitch, R. M., Tsuchiya, T., & Tsuchiya, J. (2006). MgSiO3 postperovskite at D conditions. Proceedings of the National Academy of
Sciences of the United States of America, 103(3), 543–546. https://doi.org/10.1073/pnas.0506879103
Wookey, J., Stackhouse, S., Kendall, J.‐M., Brodholt, J., & Price, G. D. (2005). Efficacy of the post‐perovskite phase as an explanation for
lowermost‐mantle seismic properties. Nature, 438(7070), 1004.
Wu, Z., & Wentzcovitch, R. M. (2014). Spin crossover in ferropericlase and velocity heterogeneities in the lower mantle. Proceedings of the
National Academy of Sciences, 111(29), 10,468–10,472. https://doi.org/10.1073/pnas.1322427111
Yang, J., Lin, J.‐F., Jacobsen, S. D., Seymour, N. M., Tkachev, S. N., & Prakapenka, V. B. (2016). Elasticity of ferropericlase and seismic
heterogeneity in the Earth's lower mantle. Journal of Geophysical Research: Solid Earth, 121, 8488–8500. https://doi.org/10.1002/
2016JB013352
Zhan, Z., Helmberger, D. V., & Li, D. (2014). Imaging subducted slab structure beneath the Sea of Okhotsk with teleseismic waveforms.
Physics of the Earth and Planetary Interiors, 232, 30–35. https://doi.org/10.1016/j.pepi.2014.03.008
Zhao, D., Fujisawa, M., & Toyokuni, G. (2017). Tomography of the subducting Pacific slab and the 2015 Bonin deepest earthquake (Mw
7.9). Scientific Reports, 7, 44487. https://doi.org/10.1038/srep44487

LU ET AL. 11,567