of 11
Geophys. J. Int.
(2022)
228,
1713–1723
https://doi.org/10.1093/gji/ggab429
Advance Access publication 2021 October 22
GJI Seismology
Crust and upper-mantle seismic anisotropy variations from the coast
to inland in central and Southern Mexico (2): correlations with
tectonic tremor
Allen Husker
,
1
Jorge C. Castellanos,
1
Xyoli P
́
erez-Campos,
2
Ra
́
ul W. Valenzuela
2
and
William B. Frank
3
1
Seismological Laboratory, California Institute of Technology, Pasadena, CA
91125
, USA. E-mail:
ahusker@caltech.edu
2
Instituto de Geof
́
ısica, Universidad Nacional Aut
́
onoma de M
́
exico, Mexico City, CDMX
04510
, Mexico
3
Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA
02139
,USA
Accepted 2021 October 20. Received 2021 August 15; in original form 2021 February 17
SUMMARY
Seismic anisotropy in the flat slab region of Mexico is compared with tectonic tremor (TT)
activity. The anisotropy is observed in three separate horizontal layers using a novel technique
with receiver functions. Those layers are identified as the continental crust and the subducted
flat oceanic slab and a thin (
10 km thick) remnant mantle wedge between those two layers.
The TT is located in two zones: (1) the Sweet Spot where most of the tremor is observed
(
160–180 km from the coast) and (2) the Transient Zone (
80–110 km from the coast).
Anisotropy within each layer is observed to be different within each of the tremor zones than
just outside them. The changes are explained as due to hydration within those zones. Water
releasing phase changes have previously been modelled to occur within those two zones in the
subducted slab (Manea & Manea). Water rising through each of the layers should generate the
observed differences in anisotropy in those zones as the fast polarization direction and split
times can differ between dry and hydrated material. This observation also correlates with the
many observations of high pore fluid pressure associated with TT.
Key words:
North America; Seismic anisotropy; Subduction zone processes.
1 INTRODUCTION
A novel approach to analyse the anisotropy with receiver functions
(RF) was developed by Castellanos
et al
.(
2017
). Starting from the
Earth’s surface and going through greater depths, the technique iso-
lates the anisotropy of individual layers by removing the anisotropy
effects of the preceding layers within the RFs, and can be subtracted
from the overall anisotropy observed by
SKS
splitting. This process
allows for the anisotropy of each layer to be analysed separately. This
is an advance from previous work that relied solely on
SKS
splitting
in the forearc regions of subduction zones where the fast directions
were generally observed to be trench parallel (e.g. Long & Silver
2008
). SKS splitting only provides the cumulative total anisotropy
along the ray path from the core–mantle boundary (CMB) to the
surface, and so its analysis may not necessarily be the best tool
to investigate the small-scale anisotropic structure of subduction
zones. During the RF analysis of Castellanos
et al
.(
2017
), the au-
thors from that paper and this one observed that local variations
in anisotropy from the RFs correlated with slow slip and tremor
in Guerrero, Mexico. However, Castellanos
et al
.(
2017
) focused
on the large-scale crustal and mantle anisotropy variations under
central Mexico, and did not explore further the anisotropy varia-
tions and their relation to the slow slip and tremor along the section
where the slab is flat. In this second paper, we use the same data and
method to investigate the correlation between seismic anisotropy
and slow slip phenomena.
Slow slip and tremor in the Mexican subduction zone exhibit a
very complicated behaviour that extends trench-parallel throughout
the whole zone. The large-scale behaviour is highly dependent on
the geometry of the subducted slab throughout the subduction zone.
Slow slip phenomena are often observed in subduction zones where
the slab interface is near 30–50 km depth (e.g. Audet & Kim
2016
).
The Mexican subduction zone has steep subduction at the NW and
SE edges with the flattest part of the slab in the middle (Fig.
1
).
The RF study of Castellanos
et al
.(
2017
) used the MesoAmerican
Subduction Experiment (MASE) data set, which was a temporary
linear array of seismometers deployed from 2005 to 2007 (Fig.
1
)
to determine the geometry of the slab (P
́
erez-Campos
et al
.
2008
).
We limit our discussion of the slow slip phenomena to the region
beneath the MASE array.
The Mexican flat slab region, where the MASE array was located,
hosts one of the only double slow slip event (SSE) zones in the world.
C

The Author(s) 2021. Published by Oxford University Press on behalf of The Royal Astronomical Society.
1713
Downloaded from https://academic.oup.com/gji/article/228/3/1713/6408464 by California Institute of Technology user on 16 December 2021
1714
A. Husker
et al
.
Figure 1.
The study area is shown in map view. MASE is shown as inverted blue triangles. The Trans-Mexican Volcanic Belt (TMVB) (dark grey) and the
slab iso-depth lines are from Ferrari
et al
.(
2012
). The red outline indicates the large SSE region (Graham
et al
.
2016
). The exact shape of the downdip SSE
region is not well defined, but results suggest it is coincident with tremor in the Sweet Spot and occupies a similar space (Villafuerte & Cruz-Atienza
2017
).
The Transient Zone (yellow) and the Sweet Spot (red) are shown to indicate the tremor regions (Husker
et al
.
2012
;Frank
et al
.
2014
).
The double SSE zone is defined by the occurrence of two SSE zones
directly updip/downdip from each other. The updip SSEs have a 4-yr
recurrence interval and a 6-month to 1-yr duration and are referred
to as long term (e.g. Radiguet
et al
.
2012
). They are also the largest
SSEs observed in the world with
M
w
7.0–7.5. The downdip SSEs
are considered short term since they occur every 60–90 d between
long-term SSEs, last for about a week and are
M
w
6.4 (Frank
et al
.
2015
; Rousset
et al
.
2017
;Husker
et al
.
2019
). The long-term SSEs
have been observed to migrate on the order of 100 km parallel to the
trench, but it is unclear if the short-term SSEs migrate (Radiguet
et
al
.
2010
).
Along with the SSEs there are tectonic tremor (TT) and low-
frequency earthquakes (LFEs). TTs are suggested to be made up of a
series LFEs (Shelly
et al
.
2007
). Independent TT and LFE detection
and location studies in Mexico suggest they come from the same
source (Frank
et al
.
2014
; Cruz-Atienza
et al
.
2015
;Husker
et al
.
2019
). For the purpose of explaining their behaviour in Mexico,
we assume it is the same phenomenon, with LFEs giving a more
detailed point source identification of TT over time. TT, as expected,
follows the double SSE zone and is also split into two corresponding
regions (Fig.
1
). Previous studies have named the subregions based
on the quantity of TT found (Fig.
1
). The largest quantity of TT is
found in the Sweet Spot, with an order of magnitude less found in the
Transient Zone (Fig.
1
). The Sweet Spot is so-called since it is the
location with the largest quantity and highest amplitude TT despite
being located in the small SSE region and not the large SSE region
(Husker
et al
.
2012
;Frank
et al
.
2014
). High pore fluid pressure has
been suggested as a possible reason for the TT difference in the two
regions and has been recognized as important in the generation of
TT worldwide (Manea & Manea
2011
;Husker
et al
.
2012
; Audet
&Kim
2016
). Between the Sweet Spot and Transient Zone there
is a region with almost no TT called the Buffer Zone. Sweet Spot
TT are directly associated with displacements from the short-term
SSEs (Frank
et al
.
2015
; Villafuerte & Cruz-Atienza
et al
.
2017
).
Husker
et al
.(
2019
) and a series of papers by William Frank
(Frank
2016
;Frank
et al
.
2018
; Frank & Brodsky
2019
)have
shown that the total geodetic response of slow slip can be captured
by LFEs/tremor suggesting that they are the direct result of the
geodetic movement. Therefore, they must be collocated, if not two
measurements of the same phenomena. Cruz-Atienza
et al
.(
2015
)
were able to locate tremor near the interface between the overriding
plate and the subducted plate, but did not have the resolution to fix
it into a specific region. SSEs are observed in the GPS record as the
reversal of the slip accumulation between the overriding crust and
the subducted crust (e.g. Radiguet
et al
.;
2012
;Frank
et al
.
2014
;
Cruz-Atienza
et al
.
2015
;Graham
2016
). Slip inversions of SSE
always place it at that interface (e.g. Radiguet
et al
.
2012
;Cavali
́
e
et al
.
2013
;Graham
et al
.
2016
). The anisotropy measurements we
explore here help limit the slip region within the layers.
2 METHODS AND RESULTS
RF are the transfer function between different components of seis-
mic motion that highlight seismic discontinuities such as the inter-
face between the crust and the mantle. When the incident angle of
a
P
wave is close to perpendicular at an interface, a weak
S
wave
is generated after passing through the interface. The deconvolution
of the
P
wave from the horizontal seismogram is the
P
-wave RF
with the
S
waves seen as pulses indicating sharp velocity contrasts.
If there is anisotropy embedded within the layers, the converted
S
waves will also split into a fast and slow pulse and may, therefore,
also be used to characterize anisotropy within the subsurface (Casidi
1992
; Levin & Park
1998
;Savage
1998
; Farra & Vinnik
2000
;Liu
&Niu
2012
). Castellanos
et. al
.(
2017
) used this feature of the RF’s
to investigate the layered anisotropic structure beneath the MASE
experiment. As the particulars of the RF analysis are explained in
their paper, here we only summarize the steps of the technique in
some detail.
RF pulses are generated at interfaces and then rise through the
layers all the way to the surface, which allows for an interpretation
of the layered structure beneath the station. The RF pulses clearly
indicate that there is a 10
±
3 km low-velocity layer (LVL). The
Downloaded from https://academic.oup.com/gji/article/228/3/1713/6408464 by California Institute of Technology user on 16 December 2021
Seismic anisotropy variations
1715
LVL could possibly be altered oceanic crust (OC) or remnant man-
tle wedge (RMW) (P
́
erez-Campos
et al
.
2008
). Previous authors
working with MASE data interpreted the LVL as altered oceanic
crust which meant the RFs along this transect corresponded to the
following layers: (1) continental crust, and the (2) upper (UOC)
and (3) lower oceanic crust (LOC, e.g. P
́
erez-Campos
et al
.
2008
;
Kim
et al
.
2012
; Fig.
2
a). Kim
et al
.(
2012
) developed this inter-
pretation by looking at impedance contrasts at the upper and lower
edges of the OC and the outer layers (bottom blue and red pulses in
Fig.
2
a). In the first step of their analysis, Castellanos
et al
.(
2017
)
manually identified these three pulses at all available RF across the
array. Once the arrival time of all of the impulsive
Ps
converted
phases in the time-series were retrieved, they characterized their
anisotropic behaviour at each individual layer. However, because
the RF arrivals carry the structure’s response from their conversion
points to the seismic station, it was necessary to take into consid-
eration that each pulse was affected by the mechanical properties
of all the layers they propagated through. This particularity implied
that, in order to accurately quantify the anisotropy of a given in-
terface, they had to first account for the anisotropy effects of all of
the layers that lie directly above it. For this reason, Castellanos
et
al
.(
2017
) applied a systematic processing scheme where, starting
from the Earth’s surface and going through greater depths, they
characterized the anisotropy of individual layers and compensated
for their effects to all of the proceeding layers until the anisotropic
properties of all interfaces were isolated. This process allowed for
the anisotropy of each layer to be analysed separately and is dif-
ferent from conventional approaches that are only able to provide
average measures of the entire anisotropy structure beneath the
station.
To measure the anisotropy at each layer, Castellanos
et al
.(
2017
)
averaged the results that they obtained through a particle motion
analysis and a waveform cross-correlation procedure (Bowman &
Ando
1987
; McNamara & Owens
1993
). For the particle motion
analysis, each radial and tangential RF pair was windowed around
the arrival of the
Ps
phase of interest, and the orientation of the
particle motion diagrams was taken as the fast polarization direc-
tion,
φ
,
of the pulse. This angle was then used to rotate the RFs
into a new coordinate system, in which the time difference between
the fast and slow pulses,
δ
t,
was measured directly from the time-
series. For the waveform cross-correlation procedure, each of the
windowed RF pair was rotated in 1
increments from a
90
to
90
range to search for the angle that resulted in the maximum of
the cross-correlation function. This angle was taken to be the fast
polarization direction,
φ
,
of the pulse and the time split between the
two pulses,
δ
t
, was, again, measured directly from the waveforms.
The anisotropy parameters for a given RF pair and structural in-
terface were then taken as the average of the results obtained from
both methods, with the exception of the cases that both approaches
yielded substantially different results. For those cases, Castellanos
et al
.(
2017
) attributed the discrepancy of the methods to the diffi-
culties of properly estimating the particle motion orientation (due
to high noise levels) and, instead, only considered the output of
the cross-correlation grid search. As an example, supplementary
Fig.
3
shows the anisotropy characterization process for an RF pair
at an individual station. The final anisotropy measurements for a
single station and layer were then obtained by averaging the results
from all available RFs in which the interface pulses were robustly
identified. The standard errors of their measurements were then
computed by bootstrapping, with a total of 200 repetitions, each
parameter and uncertainties for the grid search algorithm were ob-
tained by the 95 per cent bootstrap confidence interval and by the
bootstrap standard error for the particle motion analysis and cross-
correlation procedure results. The shear wave splitting values, along
with their uncertainties, can be found in the Supporting Information
by Castellanos
et al
.(
2017
).
Once the anisotropy at a station and given layer was successfully
characterized, Castellanos
et al
.(
2017
) removed the anisotropy ef-
fect of said layer on the RFs by rotating the time-series to the
fast/slow polarization directions, correcting for the time-shift and
rotating it back to the radial and tangential coordinate systems. Af-
ter correction, the procedure described above was systematically
repeated for the proceeding RF pulses until the anisotropy of all the
layers of interest was characterized. It is important to note that the
uncertainties and errors from one layer to the other are propagated
throughout this process and, as a result, the inferences that are drawn
from deeper layers should be interpreted with more caution than the
ones made from the shallower ones.
Now, because signals in RFs are generated at the interfaces be-
tween layers, the anisotropy values,
δ
t
and
φ
, represent an average
over an entire layer. Although the technique of Castellanos
et al
.
(
2017
) isolates the anisotropy of a layer, it is still impossible to know
its distribution throughout the layer thickness. If the anisotropy is
distributed, it implies that the
δ
t
of a layer is dependent on the width
of the layer. Therefore, to eliminate this effect, we show anisotropy
percentage instead of just
δ
t
, which removes the dependence on the
width of the layer. It is defined as
Anisotropy %
=
v
fast
v
slow
v
mean
×
100
,
where
v
fast
=
d
layer
t
ave
(
δ
t
/
2
)
,
and
v
slow
=
d
layer
t
ave
+
(
δ
t
/
2
)
.
d
layer
is the thickness of the layer and
t
ave
is the
Ps
traveltime through
the layer, as given by P
́
erez-Campos
et al
.(
2008
) and Iglesias et al.
(
2010
).
Although the depth-dependent anisotropy interpretation of
Castellanos
et al
.
2017
was made in consistence with the RF re-
sult of Kim
et al
.(
2012
), there is the possibility that the identified
pulses may not necessarily correspond to the interfaces that were
mentioned above. In particular, Castellanos
et al
.(
2017
)usedthe
first prominent positive pulse in the RFs (Fig.
2
) to determine the
anisotropy of the continental crust. However, recent investigations
have shown that, in the presence of serpentinization, the Moho sig-
nal in the RFs may be absent or inverted (e.g. Bostock
et al
.
2002
;
Abe
et al
.
2013
; Kawakatsu and Watada
2007
). If this scenario hap-
pened to be the case in south-central Mexico, then this would imply
that the anisotropy that was measured for this particular pulse does
not correspond to the entirety of the continental crust but rather to
some mid-crustal layer. Nonetheless, the continuity of the red pulse
along the RFs suggests that the
P
-to-
S
conversion occurs at a con-
tinuous interface rather than at discrete regions of highly hydrated
slab at the continental Moho and that, therefore, the upper oceanic
slab and the continental crust are not in direct contact. Moreover,
there is ample evidence to suggest that the top of the slab and the
continental crust are not in contact, as coupling between the two is
very low and do not generate earthquakes (e.g. Manea
et al
.
2004
;
P
́
erez-Campos
et al
.
2008
). The lack of local earthquakes in the flat
slab region also mean that they cannot be used to resolve the thick-
ness of these layers as precisely as observed with RFs. We prefer an
interpretation of the LVL as remnant mantle wedge (RMW) which
leads to the following three layers (Fig.
2
): (1) the continental crust,
(2) the ancient RMW that remained after the slab became flat and
Downloaded from https://academic.oup.com/gji/article/228/3/1713/6408464 by California Institute of Technology user on 16 December 2021
1716
A. Husker
et al
.
Continental
Crust
LVL
Oceanic
Crust
Mantle
V
S
V
P
Receiver
function
V
S
V
P
Continental
Crust
RMW
LVL-Oceanic
Crust
Mantle
Receiver
function
(a)
(b)
(c)
Receiver
function
station PLAT
0
2
4
6
8
10
Time [s]
Figure 2.
Synthetic isotropic RF for (a) an LVL and OC, and (b) RMW and an OC that is an LVL, (c) stacked receiver function at station PLAT, located at the
flat-slab region. The amplitude ratios of the positive/negative and negative/positive pulses are closer to the second model.
(3) the OC. This corresponds to RF pulses generated at the follow-
ing interfaces: (1) the bottom of the continental crust (Moho), (2)
the top of the slab and (3) the bottom of the OC. Fig.
2
shows that
the OC could still be separated into UOC and LOC. However, it is
beyond the scope of this paper as the anisotropy cannot distinguish
separate layers between the pulses corresponding to the top of the
slab and the bottom of the OC and so we only refer to the OC. It is
important to note that the wavelength of the seismic waves allows
for some sampling between layers. The region we refer to as the
RMW is about 10 km thick between the RF generating interfaces.
The RFs’ wavelength is on the order of a few km. Therefore, the
RF will sample across a very wide distance compared to the thick-
nesses of these layers, and outside layers may be partially sampled
as well due to the long RF wavelength. However, the largest effect
should still be from the specific layer mentioned. Fig.
4
depicts the
anisotropy percentage in each of the different layers as compared
with LFE zones. It is worth taking into consideration that an addi-
tional source of bias that is likely to be present in the anisotropy
measurements, and may affect the overall shape of the waveforms,
is the one associated to the instabilities of the deconvolution of the
RF’s (i.e. different filter widths and water levels can affect the ap-
parent duration of the pulses). However, because of the consistency
in the processing of the teleseismic signals across the entire array,
we argue that their joint analysis still allows a robust interpretation
of the relative change in the anisotropic structure beneath the sta-
tions. We are interested in the relative change of the magnitude of
anisotropy rather than its absolute magnitude.
The high anisotropy percentage regions in the subducted slab
are coincident with the Sweet Spot and Transient Zone tremor/LFE
regions and two water releasing phase changes computed for the
phase equilibria and P–T water content for sediments, metabasalt
and the serpentinized mantle (Fig.
4
, Manea & Manea
2011
;Frank
et al
.
2014
). The ancient mantle wedge only has high anisotropy
in the Sweet Spot (Fig.
4
). The continental crust is reversed with
low anisotropy above the high anisotropy zones seen in the ancient
mantle wedge and the subducted slab (Fig.
4
). Huesca-P
́
erez
et al
.
(
2016
) found similar differences in the anisotropy percentage in the
continental crust between the Sweet Spot and Buffer Zone but with
a more limited data set. The anisotropy directions in the continental
crust and the upper mantle are discussed in detail by Castellanos
et al
.(
2017
) and are found to be aligned with the major geodetic
strains in the region. They also presented the anisotropy within the
subducted slab and the RMW (which they referred to as the lower
and upper parts of the subducted slab), and noted that the polariza-
tion directions in the RMW were more varied than the subducted
OC. To explain this observation, they proposed that the anisotropy
in the subducted OC was due to foliated serpentine minerals gen-
erated during the formation and spreading of the oceanic plate.
This interpretation is slightly different from Faccenda
et al
.(
2008
),
who interpreted trench parallel anisotropy as serpentinization that
occurred as hydrous minerals entered cracks that formed with the
bending of the plate at the subduction zone. In either case the fast
direction within the OC is trench parallel. Here, we show correla-
tions between the anisotropy in the subducted OC and RMW and
the local slow slip phenomena using the data from Castellanos
et
al
.(
2017
).
The anisotropy fast polarization directions in the OC are mainly
trench-parallel as expected, except in two zones where they are
closer to trench perpendicular (Fig.
5
). The anisotropy fast axes in
the RMW vary much more along the array (Fig.
5
). Castellanos
et al
.(
2017
) suggested that the greater variation in the anisotropy
within the RMW was mostly due to the fragility of the low strength
hydrous material of which it may be composed (Kim
et al
.
2010
),
but did not have an interpretation for the origin of such variation.
The anisotropy fast direction in the RMW varies greatly around the
part of the SSE that crosses the array in the Transient Zone and
appears to be directly affected by the large SSEs (Fig.
5
). The OC
shows some more limited variations in anisotropy fast direction near
the SSE (red outline).
Fig.
5
shows that anisotropy in the RMW and the subducted OC
vary across the Sweet Spot. However, to observe correlations with
LFEs, Fig.
6
shows the anisotropy variations in both regions com-
pared to a single LFE burst (Frank
et al
.
2014
). Only one LFE burst
is shown for clarity, but the variations in LFE quantity hold for the
entire duration of MASE array (blue bars in Fig.
5
). The burst corre-
sponds to activity during one short-term SSE. There is a significant
difference within the Sweet Spot at 140 km from the coast. There
are more than four times the quantity of LFEs/length further from
the trench (light pink region), that closer (dark pink region). This
can be seen within Fig.
6
as the streaks that cross from the light
pink region in the Sweet Spot to the dark pink region, diminish sig-
nificantly and can be seen as much sparser. The opposite is true for
streaks running in the other direction (Fig.
6
). The fast polarization
direction of anisotropy is trench parallel within the Sweet Spot and
the Transient Zone, and returns to trench perpendicular north of the
Sweet Spot further from the trench (Fig.
6
).
Downloaded from https://academic.oup.com/gji/article/228/3/1713/6408464 by California Institute of Technology user on 16 December 2021
Seismic anisotropy variations
1717
0
1
2
3
4
5
6
7
8
9
10
5.4
5.6
5.8
6.0
6.2
6.4
−0.4
−0.2
0
0.2
Particle motion
analysis
−50
0
50
Time [s]
S
/
F
T
/
R
F/S
l
a
i
t
n
e
g
n
a
T
l
a
i
t
n
e
g
n
a
T
Tangential
Time [s]
Polarization
diagram
Waveform
cross-correlation
−0.4
−0.2
0
0.2
−0.4
−0.2
0
0.2
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
Amplitude
Radial
Radial
Phi [°]
Lag [s]
Particle motion analysis
dt: 0.16 s
Phi: -26°
Waveform cross-correlation
dt: 0.15 s
Phi: -18°
RF backazimuth: 147º
Final parameters (0.16 s, 125°)
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
Radial
−0.4
−0.2
0
0.1
0.2
0.3
0.4
0.5
-0.3
-0.1
(a)
0
0.5
-0.5
(b)
(c)
(d)
(f)
(e)
Figure 3.
Example of the anisotropy characterization process for the pulse corresponding to the OC at station VEVI. (a) Shows the RF in the radial and
tangential coordinate system (
R
/
T
), and the fast–slow coordinate system (F/S) as determined by the particle motion analysis (middle panel) and waveform
cross-correlation procedure (right-hand panel). The grey waveforms in the first panel correspond the radial RF, while the blue waveforms correspond
to the
tangential RF. The green line marks the peak of the OC
Ps
phase, while the red dashed lines mark the window applied to isolate the pulse. (b) Shows the
windowed OC
Ps
phase. (c) Shows the particle motion of the radial and tangential RFs (waveforms in b). (d) and (e) Shows the particular motion of the
‘corrected’ oceanic crust
Ps
phase, as determined by the particle motion analysis (d) and waveform cross-correlation (e) procedure. Note the increase in
linearity of the polarization diagrams after the anisotropy correction is applied. (f) Shows the normalized correlation coefficients of the oceanic
crust
Ps
phase
for different values of
φ
. The white lines mark the angles determined by both the particle motion analysis and waveform cross-correlation procedure. The
parameters that were retrieved from both methods are presented in the panel next to (f). For this particular phase and RF pair, the results of both analy
ses were
averaged and yield a fast direction of anisotropy of 125
and a time split of 0.16 s.
3 DISCUSSION
Anisotropy comes from different sources, and the superposition of
those sources may be constructive or destructive, thereby increasing
or decreasing the total observed shear wave splitting time delay in
a layer. We follow Faccenda
et al
.’s (
2008
) work by considering
the sources of anisotropy as due to crystal preferred orientation
(CPO) and shape preferred orientation (SPO) and adjusted CPO for
hydration in the different layers. A-type CPO olivine fabric is fre-
quently used to interpret the shear wave splitting measurements of
waves with nearly vertical incidence angles. A-type olivine is a dry
olivine as experimental results have shown that it develops under
conditions of relatively low stress, high temperature and low water
content (Karato
et al
.
2008
). One can assume hexagonal anisotropy
with a horizontal axis of symmetry where the fast polarization direc-
tion becomes oriented subparallel to the shear direction, or nearly
parallel to the direction of horizontal mantle flow (Kneller
et al
.
2005
; Karato
et al
.
2008
; Long
2013
). Other olivine fabric types
develop under different conditions of hydration, temperature and
stress (Kneller
et al
.
2005
; Jung
et al
.
2006
;Karato
et al
.
2008
;
Long
2009
). In particular, the relationship between the direction of
mantle flow and the orientation of the fast axis of anisotropy is the
same for the C-, D- and E-types as for the A-type, that is, subparal-
lel (Kneller
et al
.
2005
; Jung
et al
.
2006
;Karato
et al
.
2008
; Long
2009
). For B-type fabric, however, the fast polarization direction
is perpendicular to the direction of mantle flow. B-type fabric is a
wet olivine that develops in response to relatively low temperatures,
high stresses and high water content (Kneller
et al
.
2005
; Jung
et
Downloaded from https://academic.oup.com/gji/article/228/3/1713/6408464 by California Institute of Technology user on 16 December 2021
1718
A. Husker
et al
.
Figure 4.
Magnetotelluric and anisotropy vertical profiles beneath the MASE array. The slab geometry beneath the MASE array was determined by P
́
erez-
Campos
et al
.(
2008
) as shown in the vertical cross-sections. The anisotropy percentage is separated into the different layers and scaled to each layer to facilitate
the interpretation. The colour indicates anisotropy per cent. The Transient Zone (yellow) and the Sweet Spot (red) are shown in all views to indicate t
he tremor
regions (Husker
et al
.
2012
;Frank
et al
.
2014
). The resistivity taken from J
̈
odicke
et al
.’s (
2006
) magnetotelluric sounding is shown at the top for comparison.
Low resistivity (i.e. high conductivity) is red, while high resistivity (i.e. low conductivity) is blue.
al
.
2006
; Karato
et al
.
2008
; Long
2009
). SPO, on the other hand,
includes faults, cracks and layered media (Faccenda
et al
.
2008
).
Different combinations of the SPO, CPO and hydrated CPO change
the measured anisotropy direction and the
δ
t
. We explain how these
differences could cause the variations observed in the different
layers.
The model that best explains our results for the OC is that hy-
drated material in parallel trending faults in the subducted OC
formed during subduction dominate the total anisotropy and cre-
ate trench parallel SPO anisotropy (Faccenda
et al
.
2008
). Our
explanation is derived from previous models that suffered from not
being able to separate the anisotropy into different layers. They
were
SKS
splitting studies where the anisotropy is summed as the
wave passes through the different layers.
SKS
splitting observations
are frequently interpreted based on the physical conditions such as
state of stress, temperature and hydration of olivine that prevail in
the mantle. These conditions are determined based on experimental
work and geodynamic modelling. The fast polarization directions
are usually aligned in the direction of mantle flow except for hy-
drated olivine (B-type) in which case the fast axes become oriented
perpendicular to the direction of mantle flow (Kneller
et al
.
2005
;
Jung
et al
.
2006
; Faccenda
et al
.
2008
; Karato
et al
.
2008
).
SKS
split-
ting observations in Mexico agreed with this analysis, exhibiting a
roughly trench normal
SKS
fast axis above the flat slab (80–220 km
from the coast) and a more mixed fast direction above the hydrated
mantle wedge (
>
220 km from the coast, Bernal-L
́
opez
et al
.
2016
).
Castellanos
et al
.(
2017
) also observed the same pattern within the
mantle. However, the subducted slab exhibits anisotropy that is
Downloaded from https://academic.oup.com/gji/article/228/3/1713/6408464 by California Institute of Technology user on 16 December 2021