You are on page 1of 17

Geophys. J. Int. (2023) 232, 1276–1292 https://doi.org/10.

1093/gji/ggac390
Advance Access publication 2022 October 08
GJI Seismology

Crustal and uppermost mantle structure beneath Tristan da Cunha


using surface wave phase velocity from horizontal components OBS
ambient seismic noise

Hao Zhang ,1,2 Wolfram H. Geissler,2 Mechita C. Schmidt-Aursch2 and


Raffaele Bonadio3

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


1 KeyLab of Submarine Geosciences and Prospecting Techniques, Ministry of Education, and College of Marine Geosciences, Ocean University of China,
Qingdao 266100, China. E-mail: zhanghao@stu.ouc.edu.cn
2 Geophysics Section, Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research, Bremerhaven 27570, Germany
3 Geophysics Section, Dublin Institute of Advanced Studies School of Cosmic Physics, Dublin D02Y006, Ireland

Accepted 2022 September 30. Received 2022 August 19; in original form 2022 May 17

SUMMARY
We applied ambient noise tomography on data from an ocean bottom seismometer (OBS)
experiment around Tristan da Cunha in the South Atlantic, in order to determine the crust
and uppermost mantle shear velocity structure beneath this region. The determination of
the orientation of the horizontal seismometer components allowed to perform ambient noise
cross-correlation with all three components of 19 broad-band OBSs and two land stations. We
extracted the phase velocity dispersion curves from the first higher mode Rayleigh waves and
fundamental mode Love waves at periods of 3–8 s, which were only observed in the radial
and transverse components, respectively. Following the two-steps inversion of surface wave
tomography, we finally obtained a 3-D shear velocity model around Tristan da Cunha. Our
results concur with previous studies in this region. The 3-D shear velocity model documents that
there is a shallow conduit of low shear velocity underneath the volcanic archipelago, extending
to at least ∼25 km depth. This conduit in the lithosphere may represent the magmatic plumbing
system. We also observe indications for low shear velocities in the uppermost mantle beneath
the seamounts in the southwest of the island, in an area where the deeper Tristan mantle plume
is assumed.
Key words: Atlantic Ocean; Seismic noise; Seismic tomography; Surface waves and free
oscillations; Hotspots.

the first higher mode and the fundamental mode Rayleigh wave, but
1 I N T RO D U C T I O N
the first higher mode is stronger than the fundamental mode on the
Since the surface wave impulse response can be extracted from am- radial components. In addition to those experiments in the Pacific
bient noise cross-correlation, ambient noise tomography is widely Ocean, the ambient-noise Rayleigh waves from several recent OBS
used for imaging the shear velocity structure in the crust and upper- experiments in the Atlantic Ocean (e.g. Harmon et al. 2020; Saikia
most mantle (e.g. Shapiro & Campillo 2004; Snieder 2004; Zhang et al. 2021) and Mediterranean Sea (Wolf et al. 2021) were also
et al. 2020; Feng 2021; Zhang et al. 2021). However, its application used to study the local ocean crust or lithosphere.
for deep-sea ocean bottom seismometer (OBS) data appear still to In this work, we used an amphibious data set from OBSs and land
be limited. Harmon et al. (2007) and Yao et al. (2011) found the stations deployed around and on Tristan da Cunha (TdC). We applied
fundamental and first higher mode Rayleigh wave using vertical seismic ambient noise cross-correlation on all three seismometer
components of OBSs and used them to invert for 1-D shear velocity components. By using the phase velocity at dominating periods
structure. Due to the unknown orientation of the horizontal compo- retrieved from 1st higher mode Rayleigh and fundamental mode
nents, they could only utilize the vertical components. Other studies Love waves on horizontal components, we inverted the 3-D crustal
(Takeo et al. 2013; Zha et al. 2014) observed only fundamental mode and uppermost mantle shear velocity structure in this region.
Rayleigh wave on vertical components. Furthermore, Takeo et al. TdC is a small volcanic island in the South Atlantic, located
(2013) found the first higher mode Rayleigh wave on the horizontal ∼400 km to the east of the Mid-Atlantic Ridge (MAR, Fig. 1). It
components, while Zha et al. (2014) observed only the fundamental is the main island of the TdC archipelago, consisting of the islands
mode Rayleigh and Love waves. Tomar et al. (2018) observed both of TdC, Nightingale and Inaccessible. The latest eruption on TdC

1276 
C The Author(s) 2022. Published by Oxford University Press on behalf of The Royal Astronomical Society.
OBS ambient seismic noise 1277

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 1. Bathymetrical map of study region using the ETOPO2 bathymetry data. The red triangles mark the location of both OBSs and land stations used
in this paper. The black lines indicate the age of oceanic crust (Müller et al. 2008). The inset in the right-bottom shows the position of TdC within the South
Atlantic. MAR, Mid-Atlantic Ridge; TdC, Tristan da Cunha island and RGR, Rio Grande Rise.

occurred in 1961, and there were reports on a submarine eruption off-axis toward the Tristan side, which may suggest existence of a
south of the island in 2004 (e.g. O’Mongain et al. 2007). deep mantle plume.
The classic plume theory assumes that there is a deep mantle However, some scientists have different explanations (Foulger
plume (Tristan hotspot) beneath the region, affecting already the & Natland 2003; Anderson 2005; Fairhead et al. 2005) for the
Gondwana plate ∼132 Ma ago and contributed to the separation volcanism in this region; they argued that it is not the product of
of the South American and the African continents and finally the a deep mantle plume or the Tristan hotspot, but possibly related
breakup of the South Atlantic (Morgan 1971; Morgan 1997; Cour- to shallow geodynamic processes as a consequence of the South
tillot et al. 2003; Hoernle et al. 2015). The aseismic Walvis Ridge Atlantic opening.
is suggested to represent a classical hotspot track, with TdC as the Recently, local geophysical studies were applied in order to obtain
active end of it. The Walvis Ridge–Tristan hotspot track represents more detailed structures in this region (Baba et al. 2017; Geissler
the classic evolution of a hotspot (Richards et al. 1989). It has been et al. 2017; Ryberg et al. 2017; Schlömer et al. 2017; Bonadio
termed one of the seven hotspots most likely to be sourced from et al. 2018; Geissler et al. 2020), yet the results of those stud-
the lowermost mantle from a ‘primary’ plume (Courtillot et al. ies show different indications. The magnetotelluric results of Baba
2003) with its base currently located at the margin of the African et al. (2017) show no evidence of a mantle plume. Schlömer et al.
large low shear velocity provinces (LLSVP, Torsvik et al. 2006; (2017) obtained local P-wave velocity structure with potential deep
Burke et al. 2008). French & Romanowicz (2015) also define the upper-mantle plume conduit in the southwest. Ryberg et al. (2017)
Tristan hotspot as ‘clearly resolved’ plume based on global wave- found a shallow low shear velocity anomaly directly under the is-
form tomography. Geochemical and geochronological researches lands. Receiver functions from Geissler et al. (2017) show weak
(Humphris et al. 1985; O’Connor & Duncan 1990; Gibson et al. hotspot indications in the west/southwest at mantle transition zone.
2006; Hicks et al. 2012; O’Connor et al. 2012; Rohde et al. 2013; Petrological modelling and probabilistic Rayleigh-wave inversion
Weit et al. 2017) show that volcanism within the Walvis Ridge–Rio from Bonadio et al. (2018) indicates a moderately hot mantle in
Grande Rise volcanic system is fundamentally age-progressive. The this region. The study of multibeam echosounders and sub-bottom
global P- and S-wave velocity structures from Zhang & Tanimoto profiler data (Geissler et al. 2020) indicate reduced magma supply
(1993), Masalu (2015) and Celli et al. (2020) show that the MAR in this region and weak plume–ridge interaction at present time,
low-velocity anomaly in this region extent deeper and is shifted but on the other hand documents sparse volcanic activity at the
1278 H. Zhang et al.

seafloor to the west/southwest of the islands. Since most of the pre- from other signals with relatively good signal-to-noise ratio (Fig.
vious studies focus on the deep structures, the analysis of shallower S3, Supporting Information). We thus applied the method of Stach-
structures is lacking. In this paper, we obtain crustal and uppermost nik et al. (2012) to determine the horizontal orientations of OBS
mantle shear velocity structure using ambient noise tomography, stations, which is based on the Rayleigh-wave polarization charac-
which would provide a new insight into the formation of volcanic teristic of a 90◦ phase shift between vertical and radial directions.
archipelagos in this region. Following Stachnik et al. (2012), we keep rotating the two horizon-
tal component and calculating the correlation coefficient between
the each ‘radial’ components and the Hilbert transform (includes
2 D ATA A N D M E T H O D S the 90◦ phase shift) of vertical component. Taking the maximum
correlation coefficient, we could determine the true radial direction
The data analysed in this paper stem from a 11-month deployment
between the earthquake and OBS. This way, the orientation also can
of 24 broadband OBS from the German DEPAS pool (Deutscher
be calculated. Fig. 2 illustrates the general progress of this method
Geräte-Pool für Amphibische Seismologie, Schmidt-Aursch &

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


using one earthquake event as example. For each OBS, the orien-
Haberland 2017). The stations were deployed and recovered during
tation is determined by the following three steps: (1) we download
two cruises with R/V Maria S. Merian in early 2012 (cruise MSM
the events catalog of earthquakes (Mw ≥ 6.0) from USGS, and cut
20/2, Jegen et al. 2015) and end of 2012/early 2013 (cruise MSM
the Rayleigh waves waveforms of each event; (2) for each event,
24, Geissler 2014). All OBS are equipped with a three-components
the peak correlation coefficient and the peak angle are calculated as
broad-band seismometer (Güralp CMG-40T) and an ultralow fre-
mentioned above; (3) we selected the events with the correlation co-
quency hydrophone (HTI-01-PCA/ULF and HTI-04-PCA/ULF).
efficient larger than 0.6, which indicate the waveforms are probably
Usable data were recovered from 20 OBSs, which recorded con-
well recorded, and the final orientation is estimated by the median
tinuously for ∼350 d from 2012 February to 2013 January. The
angles of the selected events (Fig. S4, Supporting Information). We
network included a temporary land station (NIG01) on Nightingale
finally obtained reliable orientations for 19 OBS stations (Table S1,
Island. We also retrieved few data of a permanent station TRIS
Supporting Information). One of OBSs (TDC18) has a technical
(Global Seismograph Network) (Albuquerque Seismological Labo-
problem with its horizontal components.
ratory 1988) on TdC island. The locations of all stations are shown
in the Fig. 1. The probabilistic power spectral density (PPSD, McNa-
mara & Buland 2004) of each channel of all stations are calculated
first for quality control and spectral analysis. The spectra as well 2.2 Ambient noise cross-correlation
as the power spectral density (PSD) curves suggest that noise level
Impulse response or the empirical Green’s function retrieved from
of oceanic environments is relatively high comparing with the stan-
cross-correlation of ambient noise can be used to analyse the sur-
dard model, with dominating signal periods around 1–10 s period
face wave dispersion between two stations, which is widely used
(Fig. S1, Supporting Information).
to investigate the shear velocity structure of the subsurface (e.g.
OBS cannot rely on the precise GPS time unlike land seismo-
Shapiro & Campillo 2004; Snieder 2004).
logical stations. The clock synchronization can therefore only be
We used the three seismometer components of all usable stations.
achieved before the deployment and after the recovery. The time
The data were split into one-day long segments and decimated into
drift between the OBS log and GPS is referred to as skew. There-
two samples per second for further analysis and process. After
fore, time correction is crucial for the analysis and interpretation
time correction, the two horizontal components are first rotated into
of OBS data (e.g. in our case, it directly affects the stacking of
radial-transverse coordinate system according to the estimated ori-
ambient noise cross-correlations). We applied the noise-correlation
entation (see above). The daily seismograms are split into 2000
method (Hannemann et al. 2014) to calculated the clock drift for
s windows which overlap 50 per cent. We performed the cross-
the OBSs of which clock synchronization failed, and also check the
correlation for each time window and the daily cross-correlation
trend of the time drift. An example before and after time correction
functions (CCFs) are averaged of all the windows in one day. Fi-
are shown in Fig. S2 (Supporting Information).
nally, we stacked all daily CCFs to obtain yearly CCFs. Often, the
normalization methods such like one-bit or running mean average
(Bensen et al. 2007; Ritzwoller & Feng 2019) are suggested to
2.1 Orientation determination
suppress natural earthquakes in the data. We found they does not
The orientation of the two horizontal seismometer components is affect the signal at the dominating periods (1–10 s), but only make
unknown when the OBS settles down onto the seafloor. In order the impulses at longer period (10–30 s) weaker (see Fig. S5 and
to utilize all three components, the determination of orientation is Table S2, Supporting Information). Thus, we decided to not use
crucial but also difficult for the passive source OBS data, compar- any normalization in our case for the further analysis. Fig. 3 shows
ing with the active source experiments using the artificial source all the CCFs of vertical, radial and transverse components. We can
such like air guns to help determining the direction of horizontal observe the empirical Green’s function of Rayleigh waves on the
components. Several methods were proposed using different sig- vertical and radial components, and Love waves on the transverse
nals such like: body wave polarization, surface wave polarization, components, according to the particle motion of surface waves.
ambient noise, or ship noise (e.g. Stachnik et al. 2012; Zha et al. As shown in the Fig. 3, the final CCFs are bandpass filtered in
2013; Scholz et al. 2016). For our OBS data set, the orientation two different period bands (1–10 and 10–30 s), and the dominating
was previously estimated by P and Ps wave polarization, neces- periods are 1–10 s which are consistent with ambient noise spectrum
sary to analyse receiver functions (Geissler et al. 2017). In this from PPSD.
study, we used the polarization of Rayleigh surface waves of nat- In the period of 10–30 s, there are weak impulses/signals on both
ural earthquakes. Compared with P waves and other signals, the vertical and radial components. We tried to apply methods such as
natural Rayleigh waves have the strongest energy and highest am- spectral whitening (Bensen et al. 2007; Ritzwoller & Feng 2019) or
plitude, and their lower frequency content make it distinguished three-station interferometry (Zhang et al. 2020; Feng 2021; Zhang
OBS ambient seismic noise 1279

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 2. An example of orientation calculation for OBS TDC01 using one earthquake (Mw = 7.1, 2012 March 25). (a) Shows the waveforms of Rayleigh
waves on three components (bandpass filter: 0.02–0.5 Hz). (b) and (c) are envelopes and correlation coefficients at each orientation angles calculated following
Stachnik et al. (2012). The blue line is the true orientation angle with the highest coefficient. (d) and (e) show the particle motion in transverse and radial
directions after orientation correction. An ellipse in the radial direction is shown, which demonstrates the polarization characteristics of Rayleigh waves.

et al. 2021) to enhance the longer period range (10–30 s) but failed. In contrast to the Rayleigh waves recorded on land, the Rayleigh
The real source of those impulse responses is questionable, since waves observed at the seafloor are actually called Scholte–Rayleigh
they only occur in the daily CCFs of days with strong teleseismic waves (Scholte 1947; Yao et al. 2011). For simplicity, we call
earthquakes. Thus, those impulse responses might be not caused by them Rayleigh waves in this paper. Those waves propagate near
ambient noise but natural earthquakes, and we decided to not to use the liquid–solid interface, and their fundamental mode is very sen-
this period band for the further analysis. No impulses/signals are sitive to the water layer, especially in period range of 1–10 s. On the
observed in the transverse components, which indicate that those other hand, the presence of the water layer makes the slow funda-
signals are related to Rayleigh waves from earthquakes. This way, mental mode and the fast first higher mode Rayleigh waves clearly
it proved again that the orientation corrections calculated before are separated. Therefore, it makes it easier to observe and analyse first
reliable. The impulse responses at dominating periods 1–10 s can be higher mode Rayleigh waves.
observed in all daily CCFs independent of the occurrence of large In the dominating periods of 1–10 s, only fundamental mode
teleseismic earthquakes. Rayleigh waves are visible on the vertical components. The group
1280 H. Zhang et al.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 3. Record section of the CCFs for three different components of vertical (ZZ), radial (RR) and transverse (TT). The CCFs are filtered in two period
bands: 1–10 and 10–30 s. All traces are normalized, and the amplitude of 10—30 s is much weaker than that of 1–10 s. The red and blue lines represent
velocities of 1.5 and 4.5 km s−1 , respectively.

velocity of those is relatively slow, less than 1.5 km s−1 (Fig. 3). Rayleigh surface waves are extremely weak and invisible on the
However, on the radial component we observe both fundamental vertical components while strong enough on the radial components,
mode and first higher mode Rayleigh waves. On the transverse which concur with observations of Takeo et al. (2013).
component, the fundamental mode of the Love waves is visible. Due
to the strong sensitivity of the velocity of the fundamental Scholte–
2.3 Phase velocity measurement
Rayleigh wave to the water layer (instead of deeper structure) in
this period range, we choose to analyse only the fundamental mode We measured the phase velocity dispersion in frequency domain
of Love waves and first higher mode of the Rayleigh waves on the based on the theory of Aki (1957) following the method proposed by
two horizontal components. Unlike the observations of Yao et al. Ekström et al. (2009) which used the zero-crossings of the spectrum
(2011) and Harmon et al. (2007) with two clear modes of Rayleigh in the frequency domain. The advantage of this method compared
waves on the vertical components, in our case, the first higher mode to the time domain method (Bensen et al. 2007) is, that we can
OBS ambient seismic noise 1281

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 4. Examples of measuring the phase velocity dispersion from fundamental mode Love and first higher mode Rayleigh waves between OBS TDC01 and
OBS TDC02 (distance = 87.03 km). (a) Stacked waveforms on the radial components, blue line is the entire CCF including fundamental mode Love wave and
black line is the waveform applied a time window for subsequent measurement. (c) Colorful dots are the sets of possible phase velocity values calculated from
the zero-crossings of the real part of spectra shown in (e). The blue line is a reference curve calculated from modified AK135 model (Table S2, Supporting
Information) to guide the selection, and the red line is the selected dispersion curve. (e) Real part of spectra curve of the black waveform in (a). (b), (d) and (e)
are similar but on the radial components.

measure the phase velocity when the interstation distance is in the With
same order as the wavelength, which means it works well in the case
of short interstation distances and high frequency. Q = α 2 FT F + β 2 HT H (2)
In the implementation of this method, we applied at first a win- where vector d and vector m are the perturbations of traveltimes
dow on the CCFs in time domain to remove the irrelevant signals and slowness comparing with a reference model, respectively. The
and to smooth the spectrum. Then, the CCFs are transformed into matrix G is the forward operator containing the length of each path
frequency domain, and the zero-crossings of the real part of the in every cell. The first term represents the misfit of data, where C is
frequency spectrum are used to calculated the possible phase ve- the covariance matrix which should be diagonal. The second term
locities at the corresponding frequency, following Ekström et al. include two parts.
(2009). Then we picked the dispersion curves manually. An exam- The matrix F controls the spatial smoothing:
ple is shown in the Fig. 4. Due to the circumstance that the frequency
Si j
of phase velocities between different paths are not the same, the dis- Fi j = δi j +  (3)
persion curves had to be interpolated in a next step with an interval j Si j
of 0.1 s. We finally obtained reliable measurements in the period
δi j is the Kronecker symbol and Si j ∝ exp(−di2j /2σ 2 ), where di j
range 3.0–8.0 s with the peak count of 162 and 155 measurements
is the distance between the ith and jth nodes.
on the radial and transverse components, respectively. For the sub-
The matrix H is used to penalize the weighted norm of the model:
sequent tomography, we only use the dispersion curves at 3.5–7.5 s
with more than 100 individual measurements at each period. Fig. 5 Hi j = exp (−λρi ) δi j (4)
shows all the dispersion curves we extracted from the radial and
transverse components and their averages. where the ρi is the paths density of ith cell. The parameters α, β, λ
and σ are used to ensure the model smoothness and reduce extreme
value, after many tests, they are chosen as 23, 35, 100 and 0.9 in
our case, and their resolution test will be analysed later.
2.4 Phase velocity map tomography With these definitions, the optimal slowness vector m was calcu-
lated by:
To invert 2-D phase velocity maps at each period, we applied the
 −1 T
method of Barmin et al. (2001) based on the generalized least- m = GT C−1 G + Q G d (5)
squares theory. We used a uniform grid of 0.2◦ ×0.2◦ for tomog-
raphy. During the inversion, we used the average velocity from all
the dispersion curves as the input velocity at each period. The ray 2.5 Shear velocity inversion
paths between station pairs are computed using the fast marching
method (Rawlinson & Sambridge 2004). In the inversion problem, We jointly inverted the phase velocity of fundamental mode Love
the penalty function was defined following Barmin et al. (2001) as: waves and first higher mode Rayleigh waves at each grid node
using code surf96 (Herrmann 2013), which is based on the iterative
(Gm−d)T C−1 (Gm−d)+mT Qm (1) linearized inversion. The velocity model contains a water layer of
1282 H. Zhang et al.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 5. All the phase velocity from first higher mode Rayleigh and fundamental mode Love waves in the study region. Bottom: the grey curves are the
individual dispersion curves of the first higher mode Rayleigh and fundamental mode Love waves. The blue and green lines are the corresponding averages,
respectively. The top panels show the number of individual measurements at each period.

variable thickness on the top, 10 layers of 1 km thickness, 5 layers of the phase velocities of both surface wave types are significantly
2 km thickness below and 2 layers of 5 km thickness at the bottom. reduced in the vicinity of the islands, which already indicates that
At first, we invert for an average 1-D model using the average the shear velocity around the volcanic island is lower in both the
Rayleigh and Love wave dispersion curves. The start model in- crust and the uppermost mantle based on the sensitivity kernels.
cludes a 3.4 km thick water layer which is the average depth of The velocity anomaly of the Love waves is much weaker than that
seafloor in the region, and a constant velocity of 5 km s−1 below. of the first higher mode Rayleigh waves, meanwhile, as the period
We tested different start models (also with velocities >5 km s−1 ), increases, the amplitude of velocity anomalies decreases for both
and the inversion results always converge to same model (see Fig. wave types. At the periods of 4.0 and 5.0 s, the maximum pertur-
S6, Supporting Information). The final 1-D velocity model and the bation of the low-velocity anomaly is around 20 per cent for Love
corresponding dispersion curves are shown in Fig. 6. and 15 per cent for Rayleigh waves. At the longest studied period
The average 1-D velocity model was used as initial model to invert of 7.0 s, the perturbation drops to 10 per cent and 5 per cent, re-
the shear velocity structure at each grid node to construct a 3-D spectively. Additionally, the positions of this anomaly for those two
velocity structure. Since the water depth is crucial for the inversion surface wave types are slightly different, as anomaly of first higher
(Ryberg et al. 2017), we use the average water depth around each mode Rayleigh waves is offset to southwest compared with that of
single grid node with a diameter of the average interstations distance Love waves at each period. Compared to the low-velocity anomaly
using to ETOPO2 bathymetric data. near the islands, the amplitude of the velocity variation in other
areas seem not to be significant. The phase velocity of Love waves
northeast of the islands is relatively higher at all periods, suggesting
3 R E S U LT S a higher shear velocity in the crust.

3.1 Phase velocity maps 3.2 Resolution and sensitivity kernels analysis
The 2-D phase velocity maps of first higher mode Rayleigh and As proposed by Barmin et al. (2001), the resolution matrix R was
fundamental mode Love waves are shown in Fig. 7. At each period, calculated by (GT C−1 G + Q)−1 GT C−1 G, the rows of which are the
OBS ambient seismic noise 1283

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 6. The average 1-D shear velocity model. (a) Red line is the inverted average model using the average dispersion curves of both first higher mode
Rayleigh and fundamental mode Love waves; black dashed line is shear velocity model of equatorial MAR from Saikia et al. (2021). Green marks are the
shear velocity of 20–50 Ma Pacific lithosphere from Nishimura & Forsyth (1989). (b) Black dots are the measured average phase velocity of first higher mode
Rayleigh waves. The black line is the predicted dispersion curve for the average 1-D velocity model. (c) Same but for fundamental mode Love waves.

resolution maps at each grid node. The spatial resolution was defined velocities in the southwestern area are slightly reduced compared
as the radius of the fitted cone in each resolution map (Goutorbe to the northeastern area. but are still lager than for the central TdC
et al. 2015). We calculated the spatial resolution as can be seen area. Due to missing depth sensitivity, we cannot say if the shear
from Fig. 8, the relatively higher resolution near the TdC island is velocity profiles from the southwestern and the central areas may
∼70 km with better ray coverage, and the average spatial resolution converge at deeper depths.
is ∼100 km. To test the resolution of our tomographic grid in The horizontal slices of relative shear velocity are shown in
more qualitative way, we also applied checkerboard tests. The tests Fig. 11. Limited by the station distribution and data quality, we
confirm that potential velocity anomalies can be recovered well by only analyse the velocity anomalies with horizontal width larger
both first higher mode Rayleigh and fundamental mode Love waves, than 70 km based on the resolution analysis. Our shear velocity
especially near the islands. model clearly shows that the shear wave velocity beneath the TdC
We use the average 1-D shear velocity model in the Fig. 6 to islands is about 10 per cent lower than the average 1-D model at
calculate the sensitivity kernels of the phase velocity inversion, all depths. In addition, the low-velocity anomaly below the vol-
as shown in Fig. 9. We find that the Love waves at period of 4– canic island shifted slightly to the southwest in the upper mantle
7 s are mainly sensitive to the velocity structures above 10 km, (>10 km). The velocity in the northeast is higher at a depth of
while the first higher mode Rayleigh waves are more sensitive 6 km. A continuous low-velocity anomaly appears in the south-
to the deeper structures up to ∼25 km. Therefore, we assume west at ∼14 km depth, and its lateral extent expands as the depth
that the velocity structure of the joint inversion is reliable above increases.
25 km. Fig. 12 shows one latitudinal, one longitudinal vertical and one
southwest–northeast slice of the 3-D shear velocity model, all cross-
ing the island. The slices show obvious low shear velocity at the
3.3 Shear velocity structure shallow depth above ∼6 km (L1), and a low-velocity conduit at
As shown in Fig. 6, our average shear velocity model is consistent depth from ∼10 to 25 km (L2) underneath the TdC islands. Com-
with the study on the equatorial MAR (Saikia et al. 2021), and the pared with the average shear velocity model, the diameter of the low-
shear velocities for the deeper parts are also similar to the Pacific velocity conduit in the latitudinal cross-section is around ∼100 km,
lithosphere (20–50 Ma, Nishimura & Forsyth 1989). Fig. 10 shows while that in longitudinal cross-section is narrower with ∼70 km
the local average 1-D shear velocity profiles in the northeast, south- diameter at depths from 12 to 20 km. The cross-section (c) shows
west and central TdC island region calculated from local grid nodes. a low velocity (L3) appears in the southwest at upper mantle and
The TdC islands have the lowest shear velocities from seafloor to at seems to be connected to the low-velocity conduit below the island.
least 25 km depth. In the northeastern and southwestern areas, the Our result can’t image the fracture zone close to TdC due to the
shear velocities are similar above 15 km. Below 15 km depth, shear limitation of resolution.
1284 H. Zhang et al.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022

Figure 7. (a) Fundamental mode Love and (b) first higher mode Rayleigh wave phase velocity anomalies (per cent) at period of 4, 5, 6 and 7 s.

4 DISCUSSION Philippine Sea (Takeo et al. 2013) but different to observations


in the southern central Pacific (Harmon et al. 2007; Yao et al.
4.1 Surface waves responses extracted from ambient noise 2011).We think that this may be due to the sediments and shallow
of OBS structures close to seafloor affecting the ellipticity characteristics of
the Rayleigh surface waves (Boore & Nafi Toksöz 1969; Tanimoto
In our case, the horizontal components are very important, from
& Rivera 2008) and also the difference of the ellipticity between
which we retrieved not only the fundamental mode Love but also
the fundamental and first higher mode Rayleigh. The reason for this
the first higher mode Rayleigh waves to jointly invert shear velocity.
should be further discussed and investigated, but it is beyond the
The first higher mode Rayleigh wave is too weak to utilize on
scope of this study.
the vertical components, which is similar to observations in the
OBS ambient seismic noise 1285

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 8. Resolution analysis of tomographic phase velocity maps. (a) and (c) Resolution lengths calculated from the resolution matrix at 5 s in radial and
transverse components, respectively. (b) and (d) Checkerboard test results. Input perturbation anomalies have size of 1◦ by 1◦ and amplitude of 20 per cent.
The station locations are shown as red dots.

Figure 9. (a) The phase velocity dispersion curves and (b) 1-D depth sensitivity kernel of fundamental mode Love, (c) fundamental mode Rayleigh and (d)
first higher mode Rayleigh waves at different periods. All simulation results are calculated using the average 1-D velocity model in Fig. 6.
1286 H. Zhang et al.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 10. 1-D shear velocity profiles for different local areas. (a) Central/volcanic islands (red); southwest area (blue) and northeast area (green). Each profile
is the average of local area velocity structure. (b) Bathymetric map shows the positions of the subregions (circles). The small triangles mark the grid nodes
used to calculate the average regional velocity model in (a).

The phase velocity of fundamental mode Rayleigh waves from depths shallower than ∼10 km. From the cross-sections through the
ambient noise cross-correlations cannot be utilized in this study. 3-D model (Fig. 12), it also can be seen that at the depth greater
For the longer periods of 10–30 s, we mentioned the reasons before than ∼9 km, the velocity change is no longer significant. Therefore,
in Section 2.2. For the fundamental mode Rayleigh wave(s) with we assume that the crustal thickness in this area is about 7±2 km
dominating periods (1–10 s) it is not possible to pick the phase after subtracting the thickness of oceanic water layer, which is in
velocity (see Fig. S7, Supporting Information) when the interstatio accordance with the results of receiver-function analysis (Geissler
-distance is large (>50 km) and the velocity is low (<1.5 km s−1 ). et al. 2017).
Moreover, the dispersion curve for our average 1-D shear velocity The crustal age of this region is ∼10–30 Ma (Müller et al.
model shows that phase velocity of the fundamental mode Rayleigh 2008) and the normal average thickness of the Atlantic Ocean is
wave decreases dramatically with decreasing periods below ∼10 s 6.97 ± 0.57 km (White et al. 1992). If there is strong mantle plume
(Fig. 9a). Also, for the periods less than 10 s, the phase velocity directly under the whole region influencing also the magmatic pro-
of the fundamental mode Rayleigh wave becomes sensitive and duction at the mid-ocean spreading center, the crust may be thick-
closer to the compressional (P wave) velocity in the ocean (Ew- ened due to the plume–ridge interaction. However, already Geissler
ing et al. 1957) and is therefore less sensitive for the structure of et al. (2017) measured the crustal thickness to ∼5.5–7.5 km, which
the crust and uppermost mantle. The first higher mode Rayleigh should be the normal crustal thickness. Our observations also show
wave in the oceanic environment is in a sense analogous to the that there is no obvious crustal thickening in the entire region, and
fundamental-mode wave on the continent (Harmon et al. 2007; Yao there is only a low-velocity crustal anomaly underneath and close
et al. 2011; Takeo et al. 2013). For these reasons, we finally utilize to the volcanic islands. This proofs that the oceanic crustal struc-
only the dominating periods of the fundamental mode Love and the ture around the volcanic area is affected only locally by the mantle
first higher mode Rayleigh waves in our ambient noise tomography. plume and not regionally. In the cross-sections (Fig. 12), we inter-
These two types of surface waves are observed only in the hori- pret the shallow low-velocity anomaly L1 (above ∼6 km) as related
zontal components of our OBS data. Due to the short periods band to the volcanic edifices constructed by volcanic activity onto the
of our dispersion data and the large difference in depth range of pre-existing oceanic crust (Fig. 13).
sensitivity between those two types of surface waves (Fig. 9), we As shown in Fig. 10, our regional 1-D shear velocities away
chose to use joint inversion without considering the ‘Rayleigh–Love from the islands show a good agreement with the average shear
discrepancy’ caused by radial anisotropy (e.g. Feng & Ritzwoller velocity model of equatorial MAR (Harmon et al. 2020; Saikia
2019). In addition, compared with using only the vertical compo- et al. 2021), which represent a fast lithosphere. Their model also
nent as studied by Ryberg et al. (2017), the joint inversion of the shows the boundary between crust and upper mantle at the depth of
fundamental mode Love and first higher mode Rayleigh waves can ∼10 km. For our case, we thus draw a velocity contour of 4.4 km s−1
obtain a more reliable/constrained and deeper-reaching (now up to to indicate roughly the Moho as well as the low-velocity zone in the
25 km depth) shear (wave) velocity structure beneath the study area. mantle in Figs 12 and 13 for interpretation.

4.2 Regional crust and uppermost mantle structure 4.3 Magmatic plumbing system in the uppermost mantle
beneath TdC archipelago
It is difficult to obtain the sharp discontinuity like the crust–mantle
boundary (Moho) by surface wave inversion, therefore we can From the regional 1-D profiles (Fig. 10) and cross-sections (Fig. 12),
only get rough depth estimates of the Moho in this work. In our we image a shallow conduit of low shear velocity underneath the
1-D velocity profiles (Fig. 10), the velocity changes dramatically at volcanic islands (anomaly L2). We observe that the low-velocity
OBS ambient seismic noise 1287

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 11. Horizontal slices of velocity perturbation relative to the average Vs model at different depths. The average velocities and depths are shown in the
top left corner.

conduit L2 extends into the upper mantle down to at least 25 km Alaska–Bai et al. 2021). However, still it is very challenging, since
depth, with a diameter of ∼100 km in the east–west direction and a not always all methods provide concurring results, as it is also the
narrower width of ∼70 km in the north–south direction. We interpret case for the deeper upper-mantle structure beneath the Tristan da
the reduced shear velocities as related to the presence of melts. Cunha region that was studied by magnetotellurics and teleseis-
Reduced seismic velocities were also inferred by 1-D modelling mic earthquake tomography (e.g. Baba et al. 2017; Schlömer et al.
of receiver functions (Geissler et al. 2017; Weit et al. 2017), see 2017). Since those studies focus on the deeper structure (>50 km)
Fig. 13. Petrologic and geochemical studies of magmatic rocks and have limited lateral resolution (>200 km) at shallower part of
suggested magma storage and crystallization at depths of about 24– this region, more detailed shallow information might be missed in
36 km, and the melting depth with its base at 80–100 km and a top both studies.
at 60 km (Gibson et al. 2005; Weit et al. 2017). Combining our results (Fig. 12c) with the ones of the study
Deep magmatic plumbing system were previously imaged, both of magmatic system from Weit et al. (2017), we believe that the
in continental and oceanic setting by a variety of geophysical meth- volcanic islands of TdC archipelago and nearby seamounts share
ods, for example, seismicity (El Hierro, Canary Islands–González the same primary magmatic plumbing system as show in Fig. 13,
et al. 2013); local earthquake tomography (Kamchatka, Russia– causing the ongoing volcanic activity around the TdC. The TdC
Koulakov et al. 2017, Mayotte, Comoros Islands–Foix et al. 2021), main island formed ∼1.3 Ma ago (O’Connor and le Roex 1992),
magnetotellurics (Tongariro, New Zealand–Hill et al. 2015) and and the South Atlantic mid-ocean ridge is spreading at a rate
ambient noise (Mount St. Helens, USA–Wang et al. 2017, Katmai, of 3.2 cm yr−1 (Argus et al. 2011). Thus, this recent plumbing
1288 H. Zhang et al.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 12. Vertical cross-sections of absolute Vs. The line on the top show the seafloor topography of the cross-section. Due to the variable seafloor topography,
the depth of velocity section starts from 4 km. (a) Slice at latitude = 37.40S. (b) Slice at longitude = 12.40 W. (c) SW–NE slice cross the island (d) the position
of three slices. The red dash circle marks the location of deeper Tristan conduit at depths greater than 50 km (Schlömer et al. 2017). The red solid circle marks
the location of the L2 conduit in the uppermost mantle at depth of 14 km. The black solid lines in each section represent the velocity contour of 4.4 km s−1 . N,
Nightingale island; R, Rockhopper Seamount and T, Tristan da Cunha island.

system exist since at least ∼1.3 Ma ago, and the position of vol- anomaly shifts to southwest as depth increases from the crust to the
canic activities around TdC should also be affected by eastward uppermost mantle. Since the ongoing volcanic activities around the
drift of the oceanic crust and lithosphere of about 50 km. In fact, TdC archipelago are fed by this same existing magmatic plumbing
a prominent seamount (Rockhopper Seamount) is indeed found at system located at the southwest side, we believe that the future lo-
∼50 km southwest of the main TdC island. The topographic section cal volcanic activities will concentrate at the southwest/west of the
of Fig. 12(c) clearly shows three islands/seamounts from northeast archipelago (see also Schlömer et al. 2017; Geissler et al. 2020).
to southwest, the sizes of which also seem to indicate different time
of formation. However, due to the lack of age data or samples for
all the islands/seamounts, the order of formation is still unknown. 4.4 Southwest upper-mantle anomaly
The vertical slices (Fig. 12) show that the L2 low-velocity conduit The upper-mantle low-velocity anomaly L3 is located partly beneath
in the uppermost mantle is located at the southwest/west of the TdC the area of ‘western volcanic cones’ of Geissler et al. (2020). Also,
main island with the largest size instead of directly below. In our the 1-D regional shear velocity model (Fig. 10) hint at reduced
view, the reason for this is the spreading of the mid-ocean ridge and velocities in the uppermost mantle compared to the area northeast
plate motion, causing the islands to drift towards northeast, together of the islands. Unfortunately, the anomaly L3 and its extent could not
with their volcanic edifices ontop of the crust (shallow low-velocity be imaged due to the limited extent of the seismological network.
anomaly L1). As shown in Fig. 12(c), the conduit L2 in the upper- Also, due to the missing of depth sensitivity, we cannot get the
most mantle, interpreted as magmatic plumbing system, is closer velocity structure of the whole lithosphere/asthenosphere system.
to the seamount/island (Rockhopper Seamount and Nightingale Is- Therefore, any interpretation has to be considered with cautions.
land) at the southwest. Therefore, the low-velocity anomaly shifts By teleseismic P-wave tomography, Schlömer et al. (2017) found a
slightly to southwest from the crust to the uppermost mantle. The deeper (>50 km) and larger (>200 km in diameter) Tristan plume
horizontal slices (Fig. 11) also show that the centre of low-velocity conduit in this region (Fig. 12d). Geissler et al. (2017) also observed
OBS ambient seismic noise 1289

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Figure 13. Sketch to illustrate our interpretation. It is based on the cross-section in Fig. 12(c). The figure on the right is shear velocity mode under Nightingale
Island from receiver function which indicates two melt reservoirs in the uppermost mantle. Arrows indicate potential magma pathways. N, Nightingale island;
R, Rockhopper Seamount and T, Tristan da Cunha island.

indications of a deep-reaching plume at mid-mantle in the area. in the uppermost mantle that might be above the present head of
Bonadio et al. (2018) mentioned the main thermal anomaly may Tristan mantle plume located in the southwest of TdC.
be located closer to the MAR instead of directly beneath the TdC.
Celli et al. (2020) also found strongly decreased velocities in the
upper mantle between MAR and TdC, which could, in our view, AC K N OW L E D G M E N T S
possibly represent the shallowest part (head) of the Tristan mantle
We thank the DFG (Deutsche Forschungsgemeinschaft) and ‘Sen-
plume. We therefore speculate that the low-velocity anomaly L3 in
atskomission für Ozeanographie’ for funding SAMPLE (South At-
the uppermost mantle could be spatially related to the mantle plume
lantic Margin Processes and Links with onshore Evolution) Priority
underneath. L3 seems unlikely to be a consequence of shallow
Program 1375 (JE296/9–1, GE 1783/4–2). Also, we thank the mas-
causes less than 20 km depth since we find no abnormal crustal
ter Ralf Schmidt and the crew of Maria S. Merian. We thank the
structure (crustal thickness) in the southwest area. The actual deeper
DEPAS (Deutsche Geräte-Pool für amphibische Seismologie, ht
conduit seems to be offset to the west/southwest by ∼200 km from
tps://jlsrf.org/index.php/lsf/article/view/165) instrument pool. This
TdC as documented by (see also Schlömer et al. 2017). Actually,
study was also sponsored by China Scholarship Council (no.
the L2 is at the position where the top of the plume was expected
202006330084). We thank Wessel & Smith for the free use of GMT
by plate kinematic calculations, whereas L3 is directly above the
software. Most figures were plotted by GMT Open-source software
deeper lying imaged plume conduit. The anomaly L3 may represent
(https://www.generic-mapping-tools.org) and Obspy python Tool-
a new growing plumbing system since already several seamounts
box (http://obspy.org) was used for data processing. Furthermore,
could be mapped on the seafloor above it (Geissler et al. 2020).
we thank anonymous reviewers for their valuable suggestions that
greatly improved the manuscript. The OBS data used for this study
are archived at Alfred Wegener Institute (www.awi.de) and can be
5 C O N C LU S I O N
requested from ’wolfram.geissler@awi.de’.
Analysing OBS data around TdC, we extracted the Green’s func-
tions of surface waves from three seismometer components and
separated different modes of Rayleigh surface waves on the radial D ATA AVA I L A B I L I T Y
components, which are different from the observations on the land,
The data used in this study are available upon request by contact with
as well as Love waves on the transverse components. We applied
’wolfram.geissler@awi.de’. The data of station TRIS are available
ambient noise tomography to obtain a 3-D shear velocity structure
at IRIS (https:/ds.iris.edu/mda/IU/TRIS/).
in this region, using the phase velocities measured from the first
higher mode Rayleigh and fundamental mode Love waves. The re-
sults show that the crustal structure and thickness is normal in most
REFERENCE
parts of the study area and that they are only anomalous close and
Aki, K., 1957. Space and time spectra of stationary stochastic waves, with
underneath the volcanic islands and seamounts. The shear velocity special reference to microtremors, Bull. Earthq. Res. Inst., 35, 415–456.
model also shows a significant low-velocity conduit (diameter of Albuquerque Seismological Laboratory/USGS, 1988. Global Seismograph
∼100 km) within the lithosphere below the TdC archipelago reach- Network (GSN - IRIS/USGS), doi:10.7914/SN/IU.
ing to a depth of at least 25 km that we interpret as the magmatic Anderson, D.L., 2005. Scoring hotspots: the plume and plate paradigms,
plumbing system. Our study also hints at a low-velocity anomaly Spec. Papers—Geol. Soc. Am., 388, 31, doi:10.1130/0-8137-2388-4.31.
1290 H. Zhang et al.

Argus, D.F., Gordon, R.G. & DeMets, C., 2011. Geologically current motion geochemistry of alkaline igneous rocks from the Paraná–Etendeka large
of 56 plates relative to the no-net-rotation reference frame, Geochem. igneous province, Earth planet. Sci. Lett., 251, 1–17.
Geophys. Geosys., 12(11), 1–13. Gibson, S., Thompson, R., Day, J., Humphris, S. & Dickin, A., 2005. Melt-
Baba, K., Chen, J., Sommer, M., Utada, H., Geissler, W.H., Jokat, W. & Jegen, generation processes associated with the Tristan mantle plume: constraints
M., 2017. Marine magnetotellurics imaged no distinct plume beneath the on the origin of EM-1, Earth planet. Sci. Lett., 237, 744–767.
Tristan da Cunha hotspot in the southern Atlantic Ocean, Tectonophysics, González, P.J. et al., 2013. Magma storage and migration associated with the
716, 52–63. 2011–2012 El Hierro eruption: implications for crustal magmatic systems
Bai, T., Nayak, A., Thurber, C., Zeng, X. & Haney, M., 2021. Ambient noise at oceanic island volcanoes, J. geophys. Res., 118, 4361–4377.
tomography of the Katmai volcanic area, Alaska, J. Volc. Geotherm. Res., Goutorbe, B., de Oliveira Coelho, D.L. & Drouet, S., 2015. Rayleigh wave
419, 107373, doi:10.1016/j.jvolgeores.2021.107373 . group velocities at periods of 6–23 s across Brazil from ambient noise
Barmin, M.P., Ritzwoller, M.H. & Levshin, A.L., 2001. A fast and reliable tomography, Geophys. J. Int., 203, 869–882.
method for surface wave tomography, Pure appl. Geophys., 158, 1351– Hannemann, K., Krüger, F. & Dahm, T., 2014. Measuring of clock drift
1375. rates and static time offsets of ocean bottom stations by means of ambient

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


Bensen, G., Ritzwoller, M., Barmin, M., Levshin, A.L., Lin, F., Moschetti, noise, Geophys. J. Int., 196, 1034–1042.
M., Shapiro, N. & Yang, Y., 2007. Processing seismic ambient noise Harmon, N., Forsyth, D. & Webb, S., 2007. Using ambient seismic noise to
data to obtain reliable broad-band surface wave dispersion measurements, determine short-period phase velocities and shallow shear velocities in
Geophys. J. Int., 169, 1239–1260. young oceanic lithosphere, Bull. seism. Soc. Am., 97, 2009–2023.
Bonadio, R. et al., 2018. Hot upper mantle beneath the Tristan da Cunha Harmon, N., Rychert, C.A., Kendall, J.M., Agius, M., Bogiatzis, P. & Thari-
hotspot from probabilistic rayleigh-wave inversion and petrological mod- mena, S., 2020. Evolution of the oceanic lithosphere in the equatorial
eling, Geochem. Geophys. Geosyst., 19, 1412–1428. atlantic from Rayleigh wave tomography, evidence for small-scale con-
Boore, D.M. & Nafi Toksöz, M., 1969. Rayleigh wave particle motion and vection from the PI-LAB experiment, Geochem. Geophys. Geosyst., 21,
crustal structure, Bull. seism. Soc. Am., 59, 331–346. e2020GC009174, doi:10.1029/2020GC009174.
Burke, K., Steinberger, B., Torsvik, T.H. & Smethurst, M.A., 2008. Plume Herrmann, R.B., 2013. Computer programs in seismology: an evolving tool
generation zones at the margins of large low shear velocity provinces on for instruction and research, Seismol. Res. Lett., 84, 1081–1088.
the core–mantle boundary, Earth planet. Sci. Lett., 265, 49–60. Hicks, A., Barclay, J., Mark, D.F. & Loughlin, S., 2012. Tristan da Cunha:
Celli, N.L., Lebedev, S., Schaeffer, A.J., Ravenna, M. & Gaina, C., 2020. constraining eruptive behavior using the 40Ar/39Ar dating technique,
The upper mantle beneath the South Atlantic Ocean, South America and Geology, 40, 723–726.
Africa from waveform tomography with massive data sets, Geophys. J. Hill, G.J. et al., 2015. Structure of the Tongariro volcanic system: in-
Int., 221, 178–204. sights from magnetotelluric imaging, Earth planet. Sci. Lett., 432,
Courtillot, V., Davaille, A., Besse, J. & Stock, J., 2003. Three distinct 115–125.
types of hotspots in the Earth’s mantle, Earth planet. Sci. Lett., 205, Hoernle, K., Rohde, J., Hauff, F., Garbe-Schönberg, D., Homrighausen, S.,
295–308. Werner, R. & Morgan, J.P., 2015. How and when plume zonation appeared
Ekström, G., Abers, G.A. & Webb, S.C., 2009. Determination of surface- during the 132 Myr evolution of the Tristan Hotspot, Nat. Commun., 6,
wave phase velocities across USArray from noise and Aki’s spectral 1–10.
formulation, Geophys. Res. Lett., 36. doi:10.1029/2009GL039131. Humphris, S.E., Thompson, G., Schilling, J.-G. & Kingsley, R.H., 1985.
Ewing, W.M., Jardetzky, W.S. & Press, F., 1957. Elastic Waves in Layered Petrological and geochemical variations along the Mid-Atlantic Ridge
Media, McGraw-Hill, New York. between 46 S and 32 S: influence of the Tristan da Cunha mantle plume,
Fairhead, J.D., Wilson, M., Foulger, G.R., Natland, J.H., Presnall, D.C. & An- Geochim. cosmochim. Acta., 49, 1445–1464.
derson, D.L., 2005. Plate tectonic processes in the South Atlantic Ocean: Jegen, M., Geissler, W., Maia, M., Baba, K. & Kirk, H., 2015. TRISTAN:
do we need deep mantle plumes?, in Plates, Plumes and Paradigms, Electromagnetic, Gravimetric and Seismic Measurements to Investigate
Geological Society of America, 388, 537–553. the Tristan da Cunha Hot Spot and its Role in the Opening of the South-
Feng, L., 2021. Amphibious shear wave structure beneath the Alaska- Atlantic-Cruise No. MSM20/2, January 17-February 15, 2012, Walvis
Aleutian subduction zone from ambient noise tomography, Geochem. Bay (Namibia)–Recife (Brazil).
Geophys. Geosyst., 22, e2020GC009438, doi:10.1029/2020GC009438. Koulakov, I. et al., 2017. Three different types of plumbing system be-
Feng, L. & Ritzwoller, M.H., 2019. A 3-D shear velocity model of the neath the neighboring active volcanoes of Tolbachik, Bezymianny, and
crust and uppermost mantle beneath Alaska including apparent radial Klyuchevskoy in Kamchatka, J. geophys. Res., 122, 3852–3874.
anisotropy, J. geophys. Res., 124, 10468–10497. Masalu, D.C.P., 2015. Global mid-ocean ridges mantle tomography profiles,
Foix, O., Aiken, C., Saurel, J.-M., Feuillet, N., Jorry, S.J., Rinnert, E. & Earth, 4, 80–88.
Thinon, I., 2021. Offshore Mayotte volcanic plumbing revealed by local McNamara, D.E. & Buland, R.P., 2004. Ambient noise levels in the conti-
passive tomography, J. Volc. Geotherm. Res., 420, 107395, doi:10.1016/ nental United States, Bull. seism. Soc. Am., 94, 1517–1527.
j.jvolgeores.2021.107395. Morgan, J.P., 1997. The generation of a compositional lithosphere by mid-
Foulger, G. & Natland, J., 2003. Is“ hotspot” volcanism a consequence of ocean ridge melting and its effect on subsequent off-axis hotspot up-
plate tectonics?, Science, 300, 921–922. welling and melting, Earth planet. Sci. Lett., 146, 213–232.
French, S.W. & Romanowicz, B., 2015. Broad plumes rooted at the base of Morgan, W.J., 1971. Convection plumes in the lower mantle, Nature, 230,
the Earth’s mantle beneath major hotspots, Nature, 525, 95–99. 42–43.
Geissler, W., 2014. Electromagnetic, gravimetric and seismic measurements Müller, R.D., Sdrolias, M., Gaina, C. & Roest, W.R., 2008. Age, spreading
to investigate the Tristan da Cunha hot spot and its role in the opening of rates, and spreading asymmetry of the world’s ocean crust, Geochem.
the South Atlantic Ocean (MARKE)-Cruise No. MSM24-December 27, Geophys. Geosyst., 9, doi:10.1029/2007GC001743.
2012-January 21, 2013-Walvis Bay (Namibia)-Cape Town (South Africa), Nishimura, C.E. & Forsyth, D.W., 1989. The anisotropic structure of the
MARIA S. MERIAN-Berichte, MSM24, 1–56, doi:10.2312/cr msm24. upper mantle in the Pacific, Geophys. J. Int., 96, 203–229.
Geissler, W.H. et al., 2020. Seafloor evidence for pre-shield volcanism above O’Connor, J.M. & Duncan, R.A., 1990. Evolution of the Walvis Ridge-
the Tristan da Cunha mantle plume, Nat. Commun., 11, 1–15. Rio Grande Rise hot spot system: implications for African and
Geissler, W.H., Jokat, W., Jegen, M. & Baba, K., 2017. Thickness of the South American plate motions over plumes, J. geophys. Res., 95,
oceanic crust, the lithosphere, and the mantle transition zone in the vicinity 17475–17502.
of the Tristan da Cunha hot spot estimated from ocean-bottom and ocean- O’Connor, J.M., Jokat, W., Le Roex, A.P., Class, C., Wijbrans, J.R., Keßling,
island seismometer receiver functions, Tectonophysics, 716, 33–51. S., Kuiper, K.F. & Nebel, O., 2012. Hotspot trails in the South Atlantic
Gibson, S., Thompson, R. & Day, J., 2006. Timescales and mecha- controlled by plume and plate tectonic processes, Nat. Geosci., 5, 735–
nisms of plume–lithosphere interactions: 40Ar/39Ar geochronology and 738.
OBS ambient seismic noise 1291

O’Connor, J.M. & le Roex, A.P., 1992. South Atlantic hot spot-plume sys- White, R.S., McKenzie, D. & O’Nions, R.K., 1992. Oceanic crustal thick-
tems: 1. Distribution of volcanism in time and space, Earth planet. Sci. ness from seismic measurements and rare earth element inversions, J.
Lett., 113(3), 343–364. geophys. Res., 97, 19683–19715.
O’Mongain, A., Ottemoller, L., Baptie, B., Galloway, D. & Booth, D., 2007. Wolf, F.N., Lange, D., Dannowski, A., Thorwart, M., Crawford, W., Wiesen-
Seismic activity associated with a probable submarine eruption near Tris- berg, L., Grevemeyer, I. & Kopp, H. the AlpArray Working, G., 2021. 3D
tan da Cunha, July 2004–July 2006, Seismol. Res. Lett., 78, 375–382. crustal structure of the Ligurian Basin revealed by surface wave tomogra-
Rawlinson, N. & Sambridge, M., 2004. Wave front evolution in strongly phy using ocean bottom seismometer data, Solid Earth, 12, 2597–2613.
heterogeneous layered media using the fast marching method, Geophys. Yao, H., Gouedard, P., Collins, J.A., McGuire, J.J. & van der Hilst, R.D.,
J. Int., 156, 631–647. 2011. Structure of young East Pacific Rise lithosphere from ambient noise
Richards, M.A., Duncan, R.A. & Courtillot, V.E., 1989. Flood basalts and correlation analysis of fundamental-and higher-mode Scholte-Rayleigh
hot-spot tracks: plume heads and tails, Science, 246, 103–107. waves, C.R. Geosci., 343, 571–583.
Ritzwoller, M.H. & Feng, L., 2019. Overview of pre- and post-processing of Zha, Y., Webb, S.C. & Menke, W., 2013. Determining the orientations of
ambient-noise correlations, in Seismic Ambient Noise, Cambridge Uni- ocean bottom seismometers using ambient noise correlation, Geophys.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022


versity Press, 144–187, doi:10.1017/9781108264808.007. Res. Lett., 40, 3585–3590.
Rohde, J.K., van den Bogaard, P., Hoernle, K., Hauff, F. & Werner, R., 2013. Zha, Y., Webb, S.C., Wei, S.S., Wiens, D.A., Blackman, D.K., Menke, W.,
Evidence for an age progression along the Tristan-Gough volcanic track Dunn, R.A. & Conder, J.A., 2014. Seismological imaging of ridge–arc
from new 40Ar/39Ar ages on phenocryst phases, Tectonophysics, 604, interaction beneath the Eastern Lau Spreading Center from OBS ambient
60–71. noise tomography, Earth planet. Sci. Lett., 408, 194–206.
Ryberg, T., Geissler, W., Jokat, W. & Pandey, S., 2017. Uppermost mantle Zhang, S., Feng, L. & Ritzwoller, M.H., 2020. Three-station interferometry
and crustal structure at Tristan da Cunha derived from ambient seismic and tomography: coda versus direct waves, Geophys. J. Int., 221, 521–541.
noise, Earth planet. Sci. Lett., 471, 117–124. Zhang, S., Wang, H., Wu, M. & Ritzwoller, M.H., 2021. Isotropic and
Saikia, U., Rychert, C., Harmon, N. & Kendall, J.M., 2021. Upper mantle azimuthally anisotropic Rayleigh wave dispersion across the Juan de Fuca
anisotropic shear velocity structure at the equatorial mid-Atlantic ridge and Gorda plates and U.S. Cascadia from earthquake data and ambient
constrained by Rayleigh Wave group velocity analysis from the PI-LAB noise two- and three-station interferometry, Geophys. J. Int., 226, 862–
experiment, Geochem. Geophys. Geosyst., 22, e2020GC009495, doi:10 883.
.1029/2020GC009495. Zhang, Y.S. & Tanimoto, T., 1993. High-resolution global upper mantle
Schlömer, A., Geissler, W.H., Jokat, W. & Jegen, M., 2017. Hunting for the structure and plate tectonics, J. geophys. Res., 98, 9793–9823.
Tristan mantle plume–An upper mantle tomography around the volcanic
island of Tristan da Cunha, Earth planet. Sci. Lett., 462, 122–131.
Schmidt-Aursch, M. & Haberland, C., 2017. DEPAS (Deutscher Geräte-Pool S U P P O RT I N G I N F O R M AT I O N
für amphibische Seismologie): german instrument pool for amphibian
Supplementary data are available at GJ I online.
seismology, JLSRF, 3, doi:10.17815/jlsrf -3-165.
Scholte, J.G., 1947. The range of existence of Rayleigh and Stoneley waves, Figure S1. PPSD for the entire deployment for the three seismome-
Geophys. Suppl. Mon. Not. R. Astron. Soc., 5, 120–126. ters and one hydrophone components of OBS station TDC01. The
Scholz, J.-R., Barruol, G., Fontaine, F.R., Sigloch, K., Crawford, W. & black lines are the mean PSD curves. The standard low and high
Deen, M., 2016. Orienting ocean-bottom seismometers from P-wave and
noise model (McNamara & Buland 2004) are also plotted.
Rayleigh wave polarisations, Geophys. J. Int., 208, 1277–1289.
Shapiro, N.M. & Campillo, M., 2004. Emergence of broadband Rayleigh
Figure S2. An example of time correction (Hannemann et al.
waves from correlations of the ambient seismic noise, Geophys. Res. Lett., 2014). (a) The cross-correlation and daily time drifts before time
31, L07614, doi:10.1029/2004GL019491. correction between land station NIG01 and OBS TDC14 (dis-
Snieder, R., 2004. Extracting the Green’s function from the correlation of tance = 56.88 km), noted that NIG01 has no clock drift. The time
coda waves: a derivation based on stationary phase, Phys. Rev. E, 69, drift is linear. The jump at the middle (2013 July 1) is caused by
046610, doi:10.1103/PhysRevE.69.046610. a leap second of GPS time system. (b) Same after time correction.
Stachnik, J., Sheehan, A.F., Zietlow, D., Yang, Z., Collins, J. & Ferris, A., The clock drift is eliminated, and leap second was removed.
2012. Determination of New Zealand ocean bottom seismometer orien- Figure S3. An earthquake (Mw = 7.1, 2012 March 25, latitude
tation via Rayleigh-wave polarization, Seismol. Res. Lett., 83, 704–713. = −35.2◦ and longitude = −72.217◦ ) recorded by all stations on
Takeo, A., Nishida, K., Isse, T., Kawakatsu, H., Shiobara, H., Sugioka, H. &
vertical components. We can observe strong and clear Rayleigh
Kanazawa, T., 2013. Radially anisotropic structure beneath the Shikoku
Basin from broadband surface wave analysis of ocean bottom seismometer
waves.
records, J. geophys. Res., 118, 2878–2892. Figure S4. Orientation results for all earthquakes following method
Tanimoto, T. & Rivera, L., 2008. The ZH ratio method for long-period of Stachnik et al. (2012) in Fig. 2. (a) All the max peak correla-
seismic data: sensitivity kernels and observational techniques, Geophys. tion coefficient and the peak angel of station TRIS. Each red dot
J. Int., 172, 187–198. represents the results of one event. Blue dashed line indicates the
Tomar, G., Stutzmann, E., Mordret, A., Montagner, J.-P., Singh, S.C. & final orientation of TRIS. (b) Rose Chart show the directions of the
Shapiro, N.M., 2018. Joint inversion of the first overtone and fundamen- selected events (coefficient > 0.6). (b) and (d) Same but for TDC02.
tal mode for deep imaging at the Valhall oil field using ambient noise, (e) and (f) Same but for NIG01. (g) and (h) Same but for TDC22.
Geophys. J. Int., 214, 122–132. Figure S5. A test of different normalizations for the CCF from the
Torsvik, T.H., Smethurst, M.A., Burke, K. & Steinberger, B., 2006. Large
station pair TDC01–TDC02 following Ritzwoller & Feng (2019).
igneous provinces generated from the margins of the large low-velocity
provinces in the deep mantle, Geophys. J. Int., 167, 1447–1460.
The left-hand table shows different combination of normalization.
Wang, Y., Lin, F.C., Schmandt, B. & Farrell, J., 2017. Ambient noise tomog- The diagrams in the middle and right show the corresponding wave-
raphy across Mount St. Helens using a dense seismic array, J. geophys. forms in two different period bands. The shade parts indicate the
Res., 122, 4492–4508. trailing noise used to calculate SNRs in Table S2.
Weit, A., Trumbull, R.B., Keiding, J.K., Geissler, W.H., Gibson, S.A. & Figure S6. Average models inverted using different start models
Veksler, I.V., 2017. The magmatic system beneath the Tristan da Cunha with constant upper-mantle shear velocity. Black dashed line is
Island: insights from thermobarometry, melting models and geophysics, shear velocity model of equatorial MAR from Saikia et al. (2021).
Tectonophysics, 716, 64–76.
1292 H. Zhang et al.

Figure S7. A failure example of measuring the phase veloc- Table S3. Modified AK135 model for the reference dispersion
ity dispersion (Ekström et al. 2009) from fundamental mode curves.
Rayleigh waves between OBS TDC01 and OBS TDC02 (dis- Table S4. Average velocity model in our study region.
tance = 87.03 km), similar as Fig. 4 shows. However, the potential
Please note: Oxford University Press is not responsible for the con-
phase velocities are too close to pick the right curve.
tent or functionality of any supporting materials supplied by the
Table S1. Location, orientation and skew of stations.
authors. Any queries (other than missing material) should be di-
Table S2. SNR suing trailing noise for Fig. S4 for different normal-
rected to the corresponding author for the paper.
ization test in two period bands.

Downloaded from https://academic.oup.com/gji/article/232/2/1276/6754323 by Cardiff University user on 10 November 2022

You might also like