You are on page 1of 13

The two 2021 close temporal and spatial earthquakes in Sandwich

Island

Abstract

1. Introduction,
On August 12, 2021, a seismic event occurred near the South Sandwich Islands. Initially, the USGS reported
an Mw 7.5 at a depth of 47 𝑘𝑚, however, this earthquake generated a globally propagating tsunami that only
could be expected for larger and shallower events. As time passed, the aftershocks of the earthquake
illuminated a length of ~400 𝑘𝑚 in the subduction zone, this contradicts the previously calculated magnitude.
The next day, the USGS updated its results and reported that ~180 𝑠 after the beginning of the first event, a
second earthquake, Mw 8.1, would have triggered and would probably be responsible for the instrumental
tsunami recorded by tide gauges in the South Atlantic and the Indian Ocean. Finally, it was concluded that the
first event produced the second event, the main one. In this scenario, it can be mentioned that the short lapse
of time and space between the two events caused an overlap in the arrival of both event waves in the different
stations, and that was the reason why the seismic agencies weren’t able to identify the two events
automatically.

All the seismological agencies have algorithms that allow them to provide a quick description of the main
parameters of the seismic source. The calculation of these parameters is based on the implementation of
inverse models that consider the physics of the generation and propagation of waves from the source. Then,
to achieve satisfactory solutions, it is required that the arrival of the different phases of the waves would not
be contaminated with overlapping waves from different sources, otherwise, the automatic algorithms could
not be able to understand the signals.

The seismic parameters calculated by different agencies for the main event of August 12, 2021, have huge
differences between their results: For example, the early report (PDE) of the National Earthquake Information
Center (NEIC) of the US Geological Survey observed two earthquakes that occurred in a lapse of 3 minutes.
(1) The first earthquake occurred at 18:32:52 (𝑈𝑇𝐶; 25.03°𝑊, 57.68°𝑆, depth 47.2 𝑘𝑚, 𝑀𝑤 = 7.5) and (2)
The second occurred at 18:35:17 (25.26 °𝑊, 58.38°𝑆, depth 22.8 𝑘𝑚, 𝑀𝑤 = 8.1). Following the
nomenclature used by Jia et al., (2022), the first event will be called NEIC1, while the second one will be called
NEIC2. The Global Center Moment Tensor Project (GCMT) (Ekström et al., 2012) also reported two events: (1)
GCMT1 (202108121835a, with centroid time 18:35:25, 𝑀𝑤 = 8.3; 24.34°𝑊, 59.48°𝑆 and depth 20.0 km).
And (2) GCMT2 (202108121835A, with centroid time 18:36:13, 𝑀𝑤 = 7.9; 25.15°𝑊, 60.47°𝑆 and depth 15.1
km). USGS NEIC also reported a solution of the moment tensor, giving a centroid time of 18:36, 𝑀𝑤𝑐 =7.98
and depth 35.5 𝑘𝑚.

Jia et al., (2022) propose a rupture sequence from a spectral analysis of the magnitudes of body and surface
waves and an inversion of multiple sub-events using seismograms of the regional and teleseismic field. After
studying the signals using low frequencies, they concluded that the event had a duration of 260 𝑠 and was
made up of a sequence of 5 sub-events, two would have occurred early in the north of the rupture zone, then
would have been a slow event that migrated to the south, which had the characteristics of a tsunami-
earthquake and would have released 70% of the total seismic moment. Finally, another two earthquakes
would occur in the south area of the main rupture. The different sub-sources would have had hypocenters
with very different depths that trench between 7 to 39 𝑘𝑚, so we would be facing a rupture with
characteristics of a deep and a superficial event at the same time.

There are several reasons why seismic signals from different sources overlap with each other. Examples of
this are the existence of an early aftershock/foreshock, whose waves arrive at the stations before the waves
associated with the main rupture fade, there are some examples of this in Fan & Shearer (2016); Yue et al
(2017); Fan & Shearer (2018); Another explanation could be that the signals may be contaminated by P-wave
reverberations in the trench water column, which are usually seen as a monochromatic excitation in the P-
wave coda, Ihmle & Madariaga (1996); Yue et al (2017); Qiang et al (2019); Lay & Rhode (2019); Lay et al
(2019) and Wu et al (2019). These studies have tried to separate the different signals that come from more
than one source, by using methodologies that are summarized in the observation of seismograms and their
spectrograms, inversion of the source using far-field seismograms, and tide gauges, and P-wave
backpropagation over the source.

Some examples of these methodologies are Ihmle & Madariaga (1996), who studied the 1994 Kurile Islands
earthquake and the 1995 Chile earthquake, identifying monochromatic signals in the coda of the P wave,
concluding these signals didn’t belong to the pulse of the seismic rupture but were the result of reverberation
of P waves in the water column of the trench. Another example is Fan & Shearer (2018), who studied the
signals of the Sumatran earthquake 𝑀𝑤 7.2, 2012, and concluded the seismic signal shows P-wave
reverberations in the water that were generated by early aftershocks and not by the main event. Qian et al
(2019) recognize that there are pwP reverberations in the 2015 Illapel Earthquake, concluding t

hat the presence of a water layer over the trench is necessary to explain the waveform in the far-field
seismograms, so there would also be 𝑝𝑤𝑃 reverbs.

The objective of this study is to describe the rupture process of the event that occurred on August 12, 2012 in
the Sandwich Islands, based on a methodology that allows separating the different sources that overlap the
same seismic signal. This methodology consists of analyzing the traces in time and generating spectrograms
associated with the stack of those signals and producing filtered backpropagation images that highlight the
space-temporal location of the different energy pulses of these events.

2. Sisototonic context of the South Sandwich Islands


The South Sandwich Islands events occurred in the subduction zone of these 700 𝑘𝑚 long arc-shaped islands.
There, the South American plate is being subducted westward at a rate of ~7 𝑐𝑚 per year (Pelayo & Wiens,
1989), under the Sandwich plate, which is a small piece of young oceanic crust formed just 300 𝑘𝑚 to the
west of East Scortia Ridge. At the southern edge, the Sandwich Plate borders the Antarctic Plate through a
short section of a transform fault. The plate is marked by a series of regular islands, formed along the volcanic
arc of the subduction zone, those islands are the Sandwich Islands, which give this region its name. The limited
observation of these events is due to these islands being small and inhospitable, doesn’t have a native
population, and counts with a large portion is covered by ice, which makes it difficult to install instruments in
this area.
Figure 1 Map of the seismotectonic context of the two events in the South Sandwich Islands. In the lower-left section, a projection of the
terrestrial map is shown with the distribution of stations used in the analysis of this earthquake, the blue triangles correspond to the
South American network, the green ones to the Antarctic-Australian network, and the yellow ones to the South American network.
African. Red stars are the epicenters of earthquakes. In the main image, the epicenters of both earthquakes and fault geometries are
represented by the focal mechanisms, the solid red curves are the plate boundaries in this area, and the colored circles are the
accumulation of aftershocks from 90 days after the event. both earthquakes, where the colors represent the hypocentral depth of each
event.

3. Methodology
The presence of seismic wave signals from different sources near in time and space generates a superposition
of the signals, which makes them quite difficult to analyze using conventional methods. Below, is a brief
description of the different methodologies that will be implemented for the analysis of signals in the far-field
from velocity seismograms and their displacement derivative.

3.1 Pre-processing

We used velocity seismograms from teleseismic stations located at angular distances between 30° and 90°,
with large azimuthal coverage. The analysis shown here was made with South American, African, Antarctic,
and Oceanian stations (SA, AF, and A-O). They were selected from a visual inspection, looking for a balance in
the spatial distribution of the network. 61 stations were used, and the cross-correlation was made with the
SJG station. We worked with the vertical component of the velocity seismograms of these sets of networks,
which were downloaded from the Python library, ObspyDMT. The spatial location of the different networks
that we used is shown in the lower-left part of Figure 1. The analysis and conclusions of this work consider the
global network, in addition to the information of each continent separately. The graphs associated with the
local networks are available in the supplementary material.

A selection of traces was made from the following steps: (1) visual inspection to eliminate defective traces,
(2) Elimination of stations with low cross-correlation with the reference station. (For the global network, it
was required a cross-correlation greater than 0.4, while for the local networks, it was required 0.6). (3)
Elimination by visual inspection to prevent the analyzed networks had small locations with a high density of
stations in comparison to other regions that don’t have it. Finally, (4) Normalization of the amplitudes and
elimination of the linear trend of the signal.

3.2 Stack making from aligning traces concerning the P wave of both events

It was shown that the USGS and other agencies such a have made estimations of the rupture time and the
hypocenter of the main event with great dispersion. This happened because the arrival of the waves of the
main event came while the foreshock earthquake was still disturbing the surface from teleseismic distances,
which caused both signals to be superimposed, and it was more difficult to separate the two events in the
seismogram. This inaccuracy in the time of the main event motivated us to recalculate the start time of the
earthquake. For the arrival time of the foreshock, 𝑃1 , we used the results from the USGS since initially there
was no superimposition of waves from different sources. In the case of the main event, we observed the
arrival times of its different phases were contaminated by the first earthquake, which makes it difficult to
determine the exact time of the beginning of the main event.

To achieve a first-order approximation to the arrival time of the main event, we ordered the teleseismic traces
concerning their distance to the source, (Figure 2). A Butterworth bandpass filter is generated between 0.01 −
0.04 𝐻𝑧 which allowed us to highlight the main characteristics of the phases 𝑃, 𝑃𝑃 and 𝑆 of each earthquake,
which facilitated the temporary identification of the arrival of each wave. The different phases are
represented in Figure 1 by curves of the tau-p model. The arrival of 𝑃2 was recognized from the coherence of
the stack in the vicinity of the time of the main event shown by USGS, (146 s after 𝑃1 ), we took SJG as a
reference station and calculated the cross-correlation with the rest of the stations, moving our value of time
of 𝑃2 to the right, until generating a coherent stack, with a pulse associated to the location of 𝑃2 .

The stack allows us to visualize waves coming from the source, by amplifying the coherent phases through a
constructive superposition and cancelation of the incoherent phases that comes from the ambient noise and
the local propagation effects, (Ishi et al., 2007). We are interested in generating a stack that shows the
consistent information that exists in the time window that contains the 𝑃 and 𝑃𝑐𝑜𝑑𝑎 . To do that, is necessary
to align the traces concerning arrivals of 𝑃1 and 𝑃2 , calculating the best cross-correlation between the
different traces, concerning model trace. Then these traces are superimposed on each other to generate the
stack. The first group of stacks is generated in the time window of 0 − 200 𝑠, associated with the foreshock
event, and a second group of stacks in the time window of 180 − 400 𝑠, located to the main event.

For the back projections, the local arrays achieve a better resolution/coherence ratio at high frequencies,
while the global network achieves a balance by taking the low frequencies of the signal (Koper et al., 2012).
of the global network and each continental network, as is shown in the supplementary material. These
observations will allow us to visualize with more precision the spatial and temporal evolution of ruptures and
depending on the filtering frequency used, we can associate them to different depths, as is shown by Yao et
al., (2013) and also will allow us to detect the seismic signals coming from other types of sources, such as
reverberations in water, Ihmle and Madariaga (1996); Koper et al, (2012); An et al (2017); Fan and Shearer
(2018); Yue et al (2017), among others. In Figure 2, we appreciate that the filter that visually distinguishes the
seismic phases of both events the most is a 0.01 − 0.04 𝐻𝑧 bandpass filter, although in the following sections
of this work the observations were made by taking another frequency bands. (range 0.001 − 0.1 𝐻𝑧.) For the
rest of the work, we determined two-time windows associated with the 𝑃 and 𝑃𝑐𝑜𝑑𝑎 of the foreshock and the
main event, whose size is proportional to the duration of the source and the associated reverberation interval.

3.5 Spectrogram generation

We suggest a new approach in the frequency and time-frequency domain to analyze the spectral content of
the P and 𝑃𝑐𝑜𝑑𝑎 wave window stack. Calculations of time-frequency spectrograms of the stack allow us to
identify different coherent energy packets, which are associated with resonance frequencies of specific wave
segments, (Ihmle and Madariaga, 1996 and et al, 2017). Teleseismic P-wave packets related to the foreshock
and the main event will be analyzed using Fourier transforms and wavelets method. It is expected that if the
signal has been contaminated by more than one source of seismic or water reverberation rupture, this should
be observed in the time-frequency spectrum, separating each energy radiator into different regions, similar
to mountains with local highs of the function. Those mountains would be associated with different
incremental windows: ∆𝑓 and ∆𝑡. The time-frequency spectrograms were used to identify different types of
seismic sources, such as P-wave reverberations in many works such as Ihmle and Madariaga, (1996); Yue et
al, (2017); An et al, (2017); Lay et al, (2019).

We will calculate the time-frequency spectra of the two earthquakes separately, each earthquake will be
represented by the stacks coming from each station array. In both cases we will analyze the stack spectrum
of the velocity and displacement seismograms, aligning the traces in the foreshock and later in the main event.
The wavelet spectra are shown in Figure 3, where we get the stack shape in the function of to time (lower
box); the Fourier spectrum, with a frequency window between 1 × 10−3 − 1 𝐻𝑧, (left box) and the 2D time-
frequency spectrum, in a time window of 0 − 500 𝑠, (large upper right box).

3.6 P-wave back-projection imaging

The back-projection analysis of earthquakes is a type of stacking procedure of seismograms from a matrix of
stations that allows reconstructing the radiation source of waves, from the coherent records of the stations
of the entire seismic network. This method was developed following the Mw 9.2 Sumatra-Andamar
earthquake in 2004, shown in works such as Ishii et al, (2005); Ishii et al, (2007); Kiser et al, (2011) and Kiser
& Ishii, (2012). An advantage of this method is that allows us to describe complex breaking processes from a
few assumptions of the source. BP images are used to understand the first-order characteristics of the rupture,
it is not expected that this location shows the precise localization of the rupture, nor provide us an estimated
calculation of the source's Green functions, due to the alterations coming from the geometry of the station
array that is being used, then, the different sets of BP images don’t have to match exactly the illuminated
surfaces (Koper et al., 2012). The back-projection method allows us to solve rupture behaviors with variable
speed, complicated geometries, multiple events, and early aftershocks. Some examples are treated in (Ihmle
and Madariaga, (1996); Fan and Shearer, (2017); Yue et al., (2017); Fan and Shearer, (2018); Fan et al., (2017)
and An et al, (2017)).

To generate the images for this study, we used a standard procedure: First, we created the stacks, as we
described in point 3.4. In order do to that, we aligned the seismograms concerning the reference phases 𝑃1
and 𝑃2 , by using cross-correlation algorithms. This generated a set of static time corrections that explain the
blurring of the image created by 3D variations in the internal structures of the earth. We generated a grid over
the hypocenter of 14° × 12° latitude-longitude, in order to back-propagate the waves from the stations
towards the source, with ∆𝑙𝑎𝑡 = ∆𝑙𝑜𝑛 = 0.1°.r each point of the grid, we formed a beam by stacking the
displaced seismograms, so he coherent information of the traces is superimposed and the incoherent
information is destroyed. In addition to that, we consider that there are no variations in the vertical, so the
model assumes 𝑧 = 𝑐𝑡𝑒.

This method is more efficient when there is a higher coverage of azimuthal stations (Cite), to use a global
network, or a uniform distribution of these stations in a limited area in azimuth, as in the case of local
networks. In both cases, the best images generated with this method have a balance between signal resolution
and signal coherence (Koper et al., 2012). In the case of the global network, that balance is achieved by using
low-frequency filters (~0.01 − 0.1 𝐻𝑧), while in local networks, the balance is achieved by using
intermediate-frequency filters (~0.05 − 1.0 𝐻𝑧). Another consideration is the type of stack used in the results
of back-projection images. In our study we generated 4th order stacks, since these stacks contain coherent
signals, improving the image resolution at the cost of losing information of absolute amplitude. (Fan and
Shearer, 2017). Finally, the traces we used to generate the back-projection images will be filtered according
to the windows 𝑡 − 𝑓 that excited in the wavelet spectra, because we are interested in knowing the spatial
location of the coherent wave radiating source.

4. Results
4.1 Stack and waveform analysis
Figure 2 shows the seismic traces in a window of ∆𝑡 = 450 𝑠, associated with stations in the far-field ordered
by their angular distance from the source. These traces were aligned with respect to the foreshock event, 𝑃1 .
Initially, we considered the values that were calculated by the USGS to locate the arrival of 𝑃2 , which would
have been located at 146 𝑠 after the foreshock event, (𝑡0 = 0), but by aligning the traces of each station
concerning the instant 𝑡 = 146 𝑠 of any of the reference stations already mentioned in section 3.1, we can’t
observe a great coherence in the stack. Therefore, following the method of recognition of the arrival of 𝑃2
from the degree of stack coherence, explained in Section 3.3, we changed the alignment time, until the
maximum constructive amplification of the stack was achieved. This happened 180 s after the arrival of 𝑃1 .

In Figure 2, we also marked the arrivals of the main phases 𝑃1,2 , 𝑃𝑃1,2 and 𝑆1,2 of both events. Using visual
inspection, we could see that the best consistency of signal/arrival time of phases 𝑃2 , 𝑃𝑃2 and 𝑆2 , was achieved
approximately 180 𝑠 after the arrival of 𝑃1 , which matches with the most coherent search method of stack
coherence described in Section 3.3. The traces are filtered at a very low frequency with a passband of 0.01 −
0.04 𝐻𝑧, because this filter allows better visualization of the main seismic phases. Finally, on the 𝑌 axis, at a
distance of 94° concerning the source, the stack of the global network has been plotted, as a result of the
superposition of all the normalized signals, where the pulses associated with waves 𝑃1,2 can be distinguished;
and It is also observed that, if we define two-time windows of 150 𝑠 that contains the arrivals of each direct
phase P, ∆𝑡𝑝1,𝑝2 we can see that both windows are contaminated by different arrivals. Figure 2 includes the
temporal correspondence of the generation of the sources detected by USGS: NEIC1,2 (red) and GCMT:
GCTM1,2 (green). We also stand out a rough temporal correspondence between NEIC1 vs. the foreshock event
and GCTM2 vs. the main event.

At the time when we only see the foreshock signal 𝑡 = [0 − 180] s, by aligning the traces and generating the
stack with respect to 𝑃1 , we observed two almost monochromatic wave packets of period 𝑇 ≈ 17 and 15 s in
the time interval 𝑡 = 30 − 90 𝑠 and then, 𝑡 = 110 − 180 𝑠, represented by the two green rectangles shown
in Figure 2 at distances between 80° − 90°. Finally, we observed that the time window after the arrival of 𝑃2 ,
between 180 − 360 𝑠, has many phases of different amplitude and period. In this window there is a
superposition of the signals of both earthquakes, and the phases of 𝑃𝑃1 , 𝑃𝑃2 y 𝑆1 and 𝑆1 . are coupled. Then,
if there are monochromatic signals, they are contaminated with the phases of both earthquakes.

(¿a qué se deben estos dos pulsos monocromaticos?)


Figure 2 Traces of global network velocity seismograms. They are ordered on the Y axis by angular distance between the station and the
epicenter of the foreshock event at 𝑡 = 0. The P (blue), PP (black), and S (red) phases of the foreshock and main event are labeled. We
consider that the main phases are better seen if we filter all the seismograms at very low frequency (we use a filter of 0.01 − 0.04 𝐻𝑧).
At 90° away from the source, the waveform of the stack is generated with this order of traces. The time windows associated with the
𝑃𝑐𝑜𝑑𝑎 of each earthquake are contaminated with phases that could be P-wave reverberations in the trench. Two green rectangles are
indicating almost monochromatic phases, probably after the main event there are also similar packages, but they may be contaminated.
In addition, at the bottom of the figure, it is indicated the time of the different sources identified by USGS (red) and GCMT (green).

4.2 Spectral Analysis


We developed a time-frequency analysis of the stack associated with the different networks that were
studied: (G; SA; A; A-O). Figure 3 shows four time-frequency spectrograms of the velocity stack (a) and (b) and
displacement (c) and (d), generated with the global network, aligning the traces concerning 𝑃1 , (a) and (c) and
with respect to 𝑃2 , (b) and (d); in a window of time of 450 s. We can see the 1D frequency spectrum on the
left of each sub-Figure 3 and the stack time series on the lower right section of each sub-Figure 3.

For the velocity stack aligned concerning 𝑃1 , in Figure 3, (a) the time-frequency spectrum shows 5 high-
amplitude regions, energized at different frequency ranges. Two of them are correlated with the arrivals of
the P waves of the main event and the foreshock. The first, correlated with the foreshock event, is observable
in the time window between 0 − 80 𝑠, associated with very low frequencies between ~0.03 − 0.07 𝐻𝑧. While
the second, correlated with the main event, is in the time window of 180 − 290 𝑠 and energized by a range
of frequencies even lower than the previous one, 𝑓 ~ 0.02 − 0.05 𝐻𝑧. The other three pulses observed in
Figure 3 (a) can be seen in a range of frequencies higher than those earthquakes, 𝑓 ~ 0.07 − 0.1 𝐻𝑧. Two of
them are observed with a delay of the arrival of 𝑃1 and 𝑃2 (the first is at 𝑡 ~ 180 − 200 𝑠 and the second at
𝑡 ~ 250 − 400 𝑠, although this changes a little when aligning the stack concerning 𝑃2 ) and depending on
which of the two 𝑃𝑖 we use to align the stacks, these signals will have greater or lesser coherence, following
the behavior of said phase 𝑃.

Figure 3 (b) shows the stack of global network stations in speed, aligning the traces concerning the main event.
Here the foreshock event is very dimmed since the traces are now not aligned with it. The stack time series
shows a large amplitude pulse after 200 𝑠, which is more clearly observed in the wavelet spectrum, due to
the time and frequency window in which it is found, 2 × 10−2 − 4 × 10−2 𝐻𝑧, we think it represents the main
event. The amplitude spectrum to the left of (b) shows two important energy pulses in the stack, one of which
corresponds to the main event and the other, of higher frequency, is in the range where the reverberations
in (a) are found. In the wavelet spectrum, this is seen as a long-lived energized fringe above the main event.
Aligning the traces with respect to 𝑃2 , the nearly monochromatic signals appear to merge into a single
continuous signal, which has two slopes, one positive, starting with frequency 5 × 10−2 𝐻𝑧 at 𝑡~90 − 180 𝑠,
and a zero slope that starts approximately at the time of the main event, at 𝑡~180 − 300 𝑠, some point pulses
of greater amplitude are also present.

Deriving the signal from the stack speed to have offset generates a low-frequency filter effect, as can be seen
in Figures 3 (c) and (d). In the case of (c), where the stack is aligned concerning the foreshock event, we notice
that the energetic pulses that we associate with the almost monochromatic signals, the reverberations, do
not look good, unlike in (a) and (b). However, the foreshock and main events are in the same frequency range
described in (a) and (b). On the other hand, there is a very long signal in time, which illuminates the spectrum
in more than ~400 𝑠, which is in the very low frequency range 2 × 10−3 − 8 × 10−3 𝐻𝑧. This long signal of
very low frequency can be associated with event 𝐸3 described by Jia et al., (2022), who define it as a slow
event with tsunami-earthquake characteristics, which would have released 70% of seismic energy from the
South Sandwich Islands event.

Finally, in the case of Figure 3(d), the scroll stack centered on the main event is shown. Like (b), in this
displacement signal, it is observed that most of the energy is concentrated in the main event, around 𝑡~200 𝑠,
this is observed both in the amplitude of the time series, and in the amplitude spectrum and wavelet. As in
(c), the almost monochromatic signals associated with reverberations are not observed, but if there is a source
of very low frequency and long duration, which could be related to a slow sub-event, only in this case the
signal has less duration and amplitude than in (c), this may be the result of aligning the stack traces with
respect to 𝑃2 , we are blurring the signal associated with the slow event and since it has a greater amplitude
when aligning the signal Regarding 𝑃1 , (Figure 3c), we can assume the source of this sub-event is close in time
with the foreshock.

Although the stacks of Figure 3 show 4 more important or less visible phases in the traces: 𝑃1 , monochromatic
phase (𝑃𝑐𝑜𝑑𝑎1 ), 𝑃2 and contaminated monochromatic (𝑃𝑐𝑜𝑑𝑎2 ). The time-frequency spectrogram manages to
separate the frequency content of the different energy packets associated with each of these traces and
shows us the fifth source of very low frequency and long duration that we can associate with a slow sub-event.

For the case of the foreshock event, in all the networks we observe between 𝑡 ~ 150 − 200 𝑠 the spectral
signal shows that the characteristic frequencies have a peak energy of 𝑓~7 × 10−2 𝐻𝑧 (𝑇~14.2 𝑠). On the
other hand, if we consider the approximate thickness of the water layer in the depth contour zone near this
radiator is ~6 𝑘𝑚 (trench depth) and if we assume that the compressive wave speed in the water is of 𝑉𝑝 =
1.6 𝑘𝑚/ℎ, then we can conclude that the theoretical resonance period of the seismic waves associated with
the water layer is 𝑇0 ~16 𝑠, (Ihmle and Madariaga, 1996). This period is within the characteristic range of the
energy peak associated with this window of between 150 − 200 𝑠 (𝑇 ~ 14.2 𝑠). Therefore, the results of the
time-frequency spectrum and the characteristic frequency/period model are consistent with this signal being
a P-wave reverberation in the trench (𝑃𝑤𝑝). Similar periods are observed in the 𝑡 − 𝑓 spectra of the main
event in the 240 − 300 𝑠 window of the global network, with small variations of ±40 𝑠 for the other networks
(see supplementary material). Therefore, we can also conclude that the second energized signal associated
with the main event, (Figure 3(b)), is due to P-wave reverberations in the water layer above the trench,
contaminated with other faces of the foreshock.

Figure 3 Spectrogram of wavelets associated with the speed stack, (a)-(b) and displacement, (c)-(d), of the global network; centered on
the foreshock event (a) and (c) and the main event, (b) and (d), for a window of 450 s, containing the two earthquakes and their
reverberations. The lower part of each subfigure shows the waveform of the stack over time, highlighting the arrival of the P waves from
each event. To the left of each subfigure, we have the Fourier spectrum associated with the entire signal stack, and to the right, the
wavelet spectrum. The lower right part of each subfigure corresponds to the time series of the velocity or displacement stack, depending
on the case. The most energized zones, represented by warm colors in the wavelet spectrum, can be recognized as the foreshock event,
the main event, and the two associated reverberation packages. The color palette shows amplitude variations in the energy of the stack,
but since this is built from the superposition of normalized seismograms, we are more interested in the degree of coherence of the signal,
instead of the numerical scale of the amplitudes.

4.4 Back-projection analysis


Figures 4 (a) and (c) show the foreshock and main event seismic energy radiators, represented by BP images
considering amplitudes greater than 40% of the maximum value of the original BP signal, for the global
network. The complementary material shows the results of these images for local networks, where amplitudes
greater than 70% of the maximum are considered. This difference is since local networks generate BP images
with energy peaks associated with hills with slopes. very low, this makes it difficult to limit the location to a
small area, which does not happen in the BP images of the global network, where the energy explosions are
mountains with steeper slopes.

In (a) and (c) of Figure 4 we show the radiators associated with P-wave rupture and reverberation. The colors
of the contour lines representing these radiators in the BP images coincide with those used in the rectangles.
of the 𝑡 − 𝑓 spectrogram of the same figure, in both cases ,the signals related to the main event and the
reverberations are being shown. The back-projections shown here are filtered with band-pass Butterworth
filters, whose windows are a function of the space delimited by the colored rectangles that enclose the
energized zones in the time-frequency spectrum. For the back-projections of the foreshock, a filter between
0.028 − 0.063 𝐻𝑧 was considered, in the initial time window between 0 − 80 𝑠; For the section of the stack
that represents the monochrome phases after the foreshock, between 140 − 185 𝑠, a 0.05 − 0.1 𝐻𝑧 filter
was used; In the case of the main event, the time window of 165 − 260 𝑠 was taken, with a filter of 0.02 −
0.045 𝑠; Finally, for the almost monochromatic signal after the main event, we take a window of the stack
between 𝑡 = 240 − 320 𝑠, filtering it between 0.05 − 0.1 𝐻𝑧.

Both in the back-projections image of the global network, and the local networks shown in the supplementary
material, a separation is observed, in space; time, and frequency, between the signals associated with
earthquakes and the almost monochromatic signals that follow them. For the time window between 0 and
80 𝑠, filtered at a very low frequency, it is observed that there is consistency between the arrival of the
foreshock P wave and the spatial location of the energy radiator it generates in the back-projection images
(level curves Green color). This can also be seen for all local networks in the supplementary material, but the
figure where it is seen most clearly is for the global network. The 140 − 185 𝑠 window, shown in Figure 4 (a),
low frequency filtered, contains a second radiator near the trench, represented by the yellow curves, which
is to the south, centered at latitude 59.7 ° 𝑆. In the local networks SA, AA and AF, shown, in the supplementary
material, it is observed that the spatial distribution of P-wave radiators is very similar to that seen in the global
network, and the differences in the geometry of the radiator are attributable to the geometry of the station
arrangements that exist in each local network.

For the subsequent time window between 165 − 320 𝑠, shown in Figure 4 (c), two energy radiators are also
observed, the largest we associate with the main event (𝑡 = 165 − 260 𝑠), since it coincides temporarily with
that signal, represented by the light green contour lines. This signal is displaced to the east, concerning the
epicenter of the main event, on the trench, we attribute it to the distortion produced by the superposition of
earthquakes on the stack. We also have that the red color signal, associated with the almost monochromatic
pulse at 𝑡 = 240 − 320 𝑠, remains located towards the south of the main event. So, the two monochromatic
signals that follow each event are spatially radiated from the same area.

This differentiation between seismic sources and monochromatic signals can be seen from the point of view
of the time evolution of the source. In the supplementary material, snapshot figures of the back-projection
images show the time evolution of the energy radiation. These figures were built by centering the stack on
the epicenter of the foreshock (for the first 180 𝑠 of the signal) and centered on the epicenter of the main
event (for the time after 𝑡 = 180 𝑠) in two frequency bands, very low frequency 0.01 − 0.05 𝐻𝑧, which we
can associate with breakage and low frequency 0.05 − 0.1 𝐻𝑧, which is associated with reverberations.

Figure 4 Spectral and back-projection analysis of the stack generated by the traces of the global network. (a) shows the P-wave back-
projection of the stack centered at the epicenter of the foreshock. The green level curves show the energy radiation associated in time
and frequency with the foreshock, the yellow ones are the radiation associated with the time and frequency window of the
monochromatic signal, the light green star corresponds to the epicenter of the foreshock, and the celestial star, is the epicenter of the
main event. The blue dotted lines are the fault limits of the area surrounding Sandwich Plaza. (b) shows the wavelet spectrum associated
with the centered foreshock stack in a window of 200 s from the arrival of the 𝑃1. The signal at the bottom of the figure is the global
network stack centered at the epicenter of the foreshock, and the signal on the left is the Fourier spectrum of that signal. The dark green
and yellow rectangles are the most exciting time and frequency windows of that spectrum and are also the windows used to generate
the figures in (a). (c) and (d) have the same meaning as (a) and (b), but for the stack centered on the arrival time of the waves of the main
event. In this case, the light green level curves in (c) represent the main event, while the red level curves are the almost monochromatic
signal after that event.

5. Discussion and Conclusion


On August 12, 2021, two seismic events Mw 7.5 and 8.1 occurred near the South Sandwich Islands. These
events were recorded by the global network of seismological stations. The seismic signals analyzed in this
work show an unusual complexity, they are contaminated with waves from different sources. Jia et al., (2022)
generated source inversions from regional and teleseismic stations, using low-frequency filters (< 0.02 𝐻𝑧).
They conclude that this event was the result of the combination of 5 sub-events, 𝐸1 ; 𝐸2 ; 𝐸3 ; 𝐸4 ; 𝐸5 in a period
of 260 s, where surface and deep sources would have combined with a range of between 7 − 25 𝑘𝑚, in
addition to a slow event, 𝐸3 , with tsunami-earthquake characteristics would have occurred.
In our study we observe two seismic sources that we associate with a foreshock and main event, two higher
frequency signals, with monochromatic characteristics, associated with P wave reverberations and a very low
frequency signal between 2 × 10−3 − 8 × 10−3 Hz and in a long-time window lasting between 100 − 400 𝑠,
shown in the time-frequency spectrum of Figure 3 (c) and (d). Although our spectrum is not very precise in
temporally delimiting this source, since the stacked spectrum will be focused or defocused depending on the
time, we use to align the traces, we agree with Jia et al., (2022) that part of the earthquake has a slow
component, represented by a very low-frequency source, 𝐸3 . We can observe this in the frequency spectra
and wavelets, although we were unable to generate satisfactory BP images to observe it, due to the very low
resolution, a product of the extremely low frequency ranges is observed, 𝑂(10−3 )𝐻𝑧.

Our methodology doesn’t allow us to observe the sub-events 𝐸1 − 𝐸2 and 𝐸4 , −𝐸5 , we see two earthquakes
that temporarily coincide with the sub-events of Jia et al., (2022): foreshock and main event, observed both
in the images spectral as in those of BP. We also have higher frequency signals, bounding a narrow band,
almost monochromatic, which could be P-wave reverberations in the trench, this is concluded by the
frequencies associated with said sources and the phase shift they have with the foreshock and main event.
These sources are not mentioned in the work of Jia et al., (2022), probably because they are observable in a
higher frequency range than the one used in their analysis, (0.05 − 0.1 𝐻𝑧).

We attribute the reverberations that we observe in the spectrograms and images of BP to the foreshock and
main event and not to 𝐸3 , since the windows where we observe these reverberations do not temporarily
coincide with the duration of 𝐸3 , according to Jia et al., (2022). Therefore, we believe that the instrumental
tsunamis may be associated with the main event, the largest of those observed by us, and not with 𝐸3 . A back-
projection of the very low-frequency signal identified in Figure 3 (c) and (d) is not shown, since at that
frequency range the resolution of the back-projection images is very low.

Both (a) waveforms, (b) Fourier spectra, (c) wavelets, and (d) time-integrated P-wave back-projections support
the idea seismic signals from the Sandwich Island events are contaminated with 5 different sources,
superimposed throughout 350 𝑠: (I) a foreshock event lasting 100 𝑠, (II) P-wave reverberations in the trench
south of the foreshock breakup entered at latitude 59.5° 𝑆, between 140 − 185 𝑠 after the start of the
foreshock breakup; (III) Main Event between 180 − 260 𝑠 after the foreshock earthquake, which has greater
amplitudes than this one in the stack and seismic traces, (IV) P-wave reverberations in the trench, generated
by the main event and located to the south of this, over the trench, near the location of the first
reverberations, although with a lower resolution of the data and while all these sources occur, the Fourier and
wavelet spectra also detect (V) a slow event of very low frequency.

The present study showed a relatively simple procedure allows to identify the origin of the different sources
superimposed on the seismic signal, these can be very useful to analyze very complex signals that are
contaminated by the superimposition of more than one source. We believe that this methodology can be
useful as a first-order approximation to obtain information on the temporal and spatial origin of seismic
sources, and we believe that the use of fast techniques based on the use of spectra and BP images can have a
practical application. in the development of tsunami early warning methods.
Bibliography
An, C., Yue, H., Sun, J., Meng, L., & Báez, J. C. (2017). The 2015 M w 8.3 Illapel, Chile, earthquake: direction‐reversed along‐
dip rupture with localized water reverberation. Bulletin of the Seismological Society of America, 107(5), 2416-2426.

Ekström, G., Nettles, M., & Dziewoński, A. M. (2012). The global CMT project 2004–2010: Centroid-moment tensors for 13,017
earthquakes. Physics of the Earth and Planetary Interiors, 200, 1-9.

Fan, W., & Shearer, P. M. (2016). Fault interactions and triggering during the 10 January 2012 Mw 7.2 Sumatra
earthquake. Geophysical Research Letters, 43(5), 1934-1942.

Fan, W., & Shearer, P. M. (2018). Coherent seismic arrivals in the P wave coda of the 2012 Mw 7.2 Sumatra earthquake:
Water reverberations or an early aftershock?. Journal of Geophysical Research: Solid Earth, 123(4), 3147-3159.

Ihmlé, P. F., & Madariaga, R. (1996). Monochromatic body waves are excited by great subduction zone
earthquakes. Geophysical research letters, 23(21), 2999-3002.

Ishii, M., Shearer, P. M., Houston, H., & Vidale, J. E. (2005). Extent, duration and speed of the 2004 Sumatra–Andaman
earthquake imaged by the Hi-Net array. Nature, 435(7044), 933-936.

Ishii, M., Shearer, P. M., Houston, H., & Vidale, J. E. (2007). Teleseismic P wave imaging of the 26 December 2004 Sumatra‐
Andaman and 28 March 2005 Sumatra earthquake ruptures using the Hi‐net array. Journal of Geophysical Research: Solid
Earth, 112(B11).

Jia, Z., Zhan, Z., & Kanamori, H. (2022). The 2021 South Sandwich Island Mw 8.2 earthquake: a slow event sandwiched
between regular ruptures. Geophysical Research Letters, 49(3), e2021GL097104.

Kiser, E., & Ishii, M. (2017). Back-projection imaging of earthquakes. Annual Review of Earth and Planetary Sciences, 45, 271-
299.

Kiser, E., Ishii, M., Langmuir, C. H., Shearer, P. M., & Hirose, H. (2011). Insights into the mechanism of intermediate‐depth
earthquakes from source properties as imaged by back projections IRIS or CMT of multiple seismic phases. Journal of
Geophysical Research: Solid Earth, 116(B6).

Koper, K. D., Hutko, A. R., Lay, T., & Sufri, O. (2012). Imaging short‐period seismic radiation from the 27 February 2010 Chile
(Mw 8.8) earthquake by back‐projection of P, PP, and PKIKP waves. Journal of Geophysical Research: Solid Earth, 117(B2).

Kiser, E., & Ishii, M. (2012). Combining seismic arrays to image the high-frequency characteristics of large
earthquakes. Geophysical Journal International, 188(3), 1117-1128.

Pelayo, A. M., & Wiens, D. A. (1989). Seismotectonics and relative plate motions in the Scotia Sea region. Journal of
Geophysical Research: Solid Earth, 94(B6), 7293-7320.

Qian, Y., Wei, S., Wu, W., Zeng, H., Coudurier‐Curveur, A., & Ni, S. (2019). Teleseismic waveform complexities caused by
near trench structures and their impacts on earthquake source study: Application to the 2015 Illapel aftershocks (Central
Chile). Journal of Geophysical Research: Solid Earth, 124(1), 870-889.

Lay, T., & Rhode, A. (2019). Evaluating the updip extent of large megathrust ruptures using Pcoda levels. Geophysical
Research Letters, 46(10), 5198-5206.

Lay, T., Liu, C., & Kanamori, H. (2019). Enhancing tsunami warning using P wave coda. Journal of Geophysical Research:
Solid Earth, 124(10), 10583-10609.

Wu, Z., Lay, T., & Ye, L. (2020). Shallow megathrust slip during large earthquakes that have high P coda levels. Journal of
Geophysical Research: Solid Earth, 125(1), e2019JB018709.

Yue, H., Castellanos, J. C., Yu, C., Meng, L., & Zhan, Z. (2017). Localized water reverberation phases and its impact on
backprojection images. Geophysical Research Letters, 44(19), 9573-9580.

You might also like