WEBVTT
Kind: captions
Language: en-US

00:00:00.679 --> 00:00:02.560
[silence]

00:00:02.560 --> 00:00:05.520
Okay. Welcome, everyone,
to the Earthquake Science seminar

00:00:05.520 --> 00:00:08.376
for March 23rd.

00:00:08.400 --> 00:00:11.840
As a reminder, please remember to
mute your microphone and turn off

00:00:11.840 --> 00:00:17.200
your video while the talk is on.
If you want closed-captioning,

00:00:17.200 --> 00:00:21.423
you can click the three buttons on the
top of your screen, then go to down to,

00:00:21.423 --> 00:00:26.536
if I remember correctly, captioning,
and then just turn that on.

00:00:26.560 --> 00:00:30.292
So, before we start, we have
a few announcements.

00:00:30.320 --> 00:00:32.720
First is, tomorrow at 10:00 a.m.,
we’re going to have a virtual

00:00:32.720 --> 00:00:38.160
coffee hour farewell to Jessie Saunders,
who is going to be moving on to a

00:00:38.160 --> 00:00:43.976
research scientist position at Caltech.
So give a farewell to Jessie at 10:00.

00:00:44.000 --> 00:00:48.776
Tomorrow at 11:00 a.m., we have the
all-hands. That’s going to be on Teams.

00:00:48.800 --> 00:00:53.920
And then next Wednesday at noon,
we have the USGS Town Hall.

00:00:53.920 --> 00:00:56.464
That’s also going to
be on Teams.

00:00:57.760 --> 00:00:59.576
Those are our
announcements.

00:00:59.600 --> 00:01:02.936
And today we have
Wenbo Wu from Woods Hole.

00:01:02.960 --> 00:01:07.840
So, Wenbo received a B.S. and M.S.
in geophysics from the University of

00:01:07.840 --> 00:01:12.560
Science and Technology of China.
He then completed a Ph.D. in

00:01:12.560 --> 00:01:16.456
geophysics at Princeton
working with Jessica Irving.

00:01:16.480 --> 00:01:20.560
And he did work on the composition
and structure of the lower mantle

00:01:20.560 --> 00:01:24.720
and the outer and inner core.
He then moved on to a Director’s

00:01:24.720 --> 00:01:29.560
Postdoctoral Fellowship at Caltech in
the seismology laboratory, where he

00:01:29.560 --> 00:01:33.016
worked on seismic ocean thermometry,
which he’s going to talk about today.

00:01:33.040 --> 00:01:37.176
But he also did some work
on fiber optic seismology.

00:01:37.200 --> 00:01:42.400
Wenbo has since moved on to the
Woods Hole Oceanographic Institute

00:01:42.400 --> 00:01:48.432
in Cape Cod, where he is
now an assistant scientist.

00:01:48.432 --> 00:01:52.056
The title of his talk today is
Seismic Ocean Thermometry.

00:01:52.080 --> 00:01:54.500
So, with that, I’ll just give it
over to you, Wenbo.

00:01:54.500 --> 00:01:59.280
- Okay. Thanks for the introduction.
It’s a great pleasure to introduce

00:01:59.280 --> 00:02:03.920
our work at USGS.
And today I’m going to talk about

00:02:03.920 --> 00:02:07.520
how to use seismic waves to monitor
ocean temperature change.

00:02:07.520 --> 00:02:10.640
And this work is done
with my collaborators.

00:02:10.640 --> 00:02:16.000
So me, Zhongwen, Sidao,
and Zhichao, we are seismologists.

00:02:16.000 --> 00:02:22.320
We also have oceanographers, so Shirui
and Joem, they are oceanographers.

00:02:22.320 --> 00:02:26.400
And, just for fun, as you can see this
picture, this picture was taken by a

00:02:26.400 --> 00:02:32.880
tourist two years ago, or three years
ago, in the ocean near Los Angeles.

00:02:32.880 --> 00:02:36.720
As you can see, these two whales
jumped out of the water, and,

00:02:36.720 --> 00:02:42.240
at that time, a local magnitude 4.7
earthquake occurred.

00:02:42.240 --> 00:02:48.160
So just suspected these whales
might be disturbed by the earthquake.

00:02:48.160 --> 00:02:52.056
Of course, it could be
just a coincidence.

00:02:52.080 --> 00:02:55.496
Just show this picture again,
as you can see these two whales.

00:02:55.520 --> 00:03:05.520
And so this is a outline of my talk.
And so I will first introduce some

00:03:05.520 --> 00:03:13.600
results from Indian Ocean.
And so, in that results, I will talk about

00:03:13.600 --> 00:03:18.000
just how we can use one frequency band
to monitor ocean temperature change.

00:03:18.000 --> 00:03:20.800
And then I want to talk
about recent progress.

00:03:20.800 --> 00:03:26.400
It’s how to use different frequencies
of acoustic waves – the acoustic

00:03:26.400 --> 00:03:30.936
wave generated by earthquakes.
And different modes to do some sort of

00:03:30.960 --> 00:03:35.680
depth tomography or depth information
of the ocean temperature change.

00:03:35.680 --> 00:03:38.880
And finally is
my conclusions.

00:03:40.720 --> 00:03:45.120
So climate change is one of the major
global crisis faced by human beings.

00:03:45.120 --> 00:03:50.000
And, in the climate system,
the ocean plays a key role in regulating

00:03:50.000 --> 00:03:55.040
how the global warming evolves.
Because it has huge heat capacity

00:03:55.040 --> 00:03:59.120
and absorbs almost all the excess
energy due to increasingly

00:03:59.120 --> 00:04:03.520
abundant greenhouse gases.
As you can see in the right figure here,

00:04:03.520 --> 00:04:06.800
almost all the energy that
goes into the ocean.

00:04:06.800 --> 00:04:14.936
And only a small portion of them, it
goes into the ice, land, and atmosphere.

00:04:14.960 --> 00:04:18.800
So the ocean really acts as
a buffer of the warming.

00:04:18.800 --> 00:04:22.480
Without this oceanic buffer,
the global temperatures would rise

00:04:22.480 --> 00:04:32.136
much more rapidly than
the case without ocean.

00:04:32.160 --> 00:04:35.440
However, accurate monitoring of
ocean temperature change is

00:04:35.440 --> 00:04:38.800
a challenging problem.
And there have been many debates

00:04:38.800 --> 00:04:44.080
about how much the ocean, at different
depths, the upper ocean and the deep

00:04:44.080 --> 00:04:48.480
ocean, how much of them –
how much temperature change

00:04:48.480 --> 00:04:50.480
are in these deep
depths of the ocean.

00:04:50.480 --> 00:04:57.736
It’s quite a controversial problem,
or there are large uncertainties there.

00:04:57.760 --> 00:05:01.360
And, to better constrain the ocean
temperature change, various types of

00:05:01.360 --> 00:05:06.480
methods, including sea surface
temperature monitored by satellites

00:05:06.480 --> 00:05:10.160
have been developed.
However, monitoring temperature

00:05:10.160 --> 00:05:13.256
below sea surface is
much more difficult.

00:05:13.280 --> 00:05:16.480
Currently, the most important
constraints on deep ocean

00:05:16.480 --> 00:05:20.960
come from the Argo floats.
So Argo array is composed of a few

00:05:20.960 --> 00:05:25.656
thousands of drifting buoys to help us
monitor ocean temperature change.

00:05:25.680 --> 00:05:29.656
Although Argo is excellent,
some limitations are present.

00:05:29.680 --> 00:05:35.736
For example, first of all, most Argo
floats drift in the top 2,000 meters.

00:05:35.760 --> 00:05:38.240
So they cannot monitor
temperature below that.

00:05:38.240 --> 00:05:41.440
And we know the ocean
below 2,000 meters takes about

00:05:41.440 --> 00:05:43.040
half of the
total volume.

00:05:43.040 --> 00:05:46.640
So half of the – half of the
water, we just have no data.

00:05:46.640 --> 00:05:51.416
And secondly, the lateral spatial
resolution is another issue.

00:05:51.440 --> 00:05:56.880
Even we already have thousands
of Argo floats currently, Argo data

00:05:56.880 --> 00:06:02.456
product could be still biased
due to the aliasing effects.

00:06:02.480 --> 00:06:07.040
So there are lots of small-scale or
mesoscale dynamic processing in the

00:06:07.040 --> 00:06:13.040
ocean, and they are not resolvable
using even these thousands of floats.

00:06:13.040 --> 00:06:17.280
And finally, Argo started in 2004,
and we don’t have data before that.

00:06:17.280 --> 00:06:21.600
Because of these limitations, we really
need more of the original data to

00:06:21.600 --> 00:06:25.336
better constrain the deep
ocean temperature change.

00:06:25.360 --> 00:06:30.080
Among different thermometry methods,
taking ocean temperature acoustically has

00:06:30.080 --> 00:06:33.496
the advantage of low cost
and high accuracy.

00:06:33.520 --> 00:06:37.120
The principle of acoustic ocean
thermometry is quite simple.

00:06:37.120 --> 00:06:42.616
It takes advantage of high sensitivity
of ocean sound speed to temperature.

00:06:42.640 --> 00:06:46.416
And this right figure just
shows us this relationship.

00:06:46.440 --> 00:06:50.856
Basically, sound waves
travel faster in warmer ocean.

00:06:50.880 --> 00:06:56.720
So, for a given salinity and pressure,
the derivative of sound speed, alpha,

00:06:56.720 --> 00:07:00.000
with respect to temperature, T,
is roughly about

00:07:00.000 --> 00:07:03.336
4 to 5 meter per second
per degree.

00:07:03.360 --> 00:07:07.840
That means we can use this relationship
to derive temperature change by

00:07:07.840 --> 00:07:13.256
observing sound speed or
sound wave travel time change.

00:07:13.280 --> 00:07:17.176
This is the principle of
acoustic thermometry.

00:07:17.200 --> 00:07:19.680
High sensitivity to
temperature is only one reason

00:07:19.680 --> 00:07:22.240
to make acoustic
thermometry feasible.

00:07:22.240 --> 00:07:26.640
Another equally important reason
is the long-distance acoustic

00:07:26.640 --> 00:07:29.496
transmission in the
ocean SOFAR channel.

00:07:29.520 --> 00:07:34.376
So SOFAR here stands for
Sound Fixing and Ranging channel.

00:07:34.400 --> 00:07:38.160
This long distance currently
is guaranteed by what we call

00:07:38.160 --> 00:07:40.136
waveguide effects.

00:07:40.160 --> 00:07:44.640
As I said, sound speed increases
with temperature, so if we go from

00:07:44.640 --> 00:07:49.440
sea surface here to the deep ocean,
we will see the sound speed gets

00:07:49.440 --> 00:07:57.360
reduced at first because temperature is –
because temperature is becoming

00:07:57.360 --> 00:08:01.976
colder. And then it comes back
due to the higher pressure.

00:08:02.000 --> 00:08:04.720
That gives us this
low-velocity channel.

00:08:04.720 --> 00:08:09.520
So this low-velocity channel makes the
SOFAR channel act as a waveguide to

00:08:09.520 --> 00:08:14.000
protect sound from the complicated
interactions and energy loss

00:08:14.000 --> 00:08:17.760
with the seafloor.
Similarly, in seismology –

00:08:17.760 --> 00:08:23.040
in seismology, we have lots of
other kind of waveguide effects,

00:08:23.040 --> 00:08:27.600
such as the fault zone,
or the subducted oceanic crust.

00:08:27.600 --> 00:08:30.856
And we can see the
waveguide effects.

00:08:30.880 --> 00:08:35.840
This long-distance transmission carries
out a spatial integration inherently,

00:08:35.840 --> 00:08:41.680
and therefore average out the effects of
mesoscale and small-scale variabilities,

00:08:41.680 --> 00:08:46.560
such as eddies in the ocean.
So accumulated travel time change

00:08:46.560 --> 00:08:50.160
is ideal for measuring ocean
temperature change averaged

00:08:50.160 --> 00:08:53.760
along their traveling paths.
And another advantage of this

00:08:53.760 --> 00:08:57.600
long-distance transmission is that
we don’t need to deploy thousands

00:08:57.600 --> 00:09:00.880
of receivers like Argo
floats in the ocean.

00:09:00.880 --> 00:09:04.480
Instead, probably tens
of receivers would be enough

00:09:04.480 --> 00:09:08.080
for a global
application.

00:09:08.080 --> 00:09:11.120
The feasibility and advantage
of acoustic thermometry have

00:09:11.120 --> 00:09:14.856
been demonstrated in
a few experiments.

00:09:14.880 --> 00:09:18.320
So here I just list three
of these experiments.

00:09:18.320 --> 00:09:25.920
The first experiment was conducted at
Perth of Australia, as early as 1960.

00:09:25.920 --> 00:09:31.840
Then the Heard Island experiment in
1991 tested the feasibility of long-range

00:09:31.840 --> 00:09:36.696
acoustic transmission and latitudes
and [inaudible] project ATOC.

00:09:36.720 --> 00:09:41.920
The ATOC is a joint program proposed
by Walter Munk from UCSD Scripps

00:09:41.920 --> 00:09:47.440
and Carl Wunsch from MIT.
What they did in ATOC is put into

00:09:47.440 --> 00:09:52.560
repeatable acoustic sources in the ocean.
One near central California here,

00:09:52.560 --> 00:09:56.240
and the other acoustic
source at Hawaii.

00:09:56.240 --> 00:10:00.296
And a bunch of hydrophones
in North Pacific.

00:10:00.320 --> 00:10:07.336
These two sources just repeatedly send
out 75 hertz signals every four days.

00:10:07.360 --> 00:10:09.520
And the sound waves’
travel time change

00:10:09.520 --> 00:10:12.536
and these hydrophones
are tracked.

00:10:12.560 --> 00:10:16.080
Finally, they found very good
consistency between the temperature

00:10:16.080 --> 00:10:21.360
change derived from ATOC and
other oceanography measurements,

00:10:21.360 --> 00:10:25.256
as well as ocean circulation
model predictions.

00:10:25.280 --> 00:10:29.520
So, for the first time,
ATOC demonstrates the excellence

00:10:29.520 --> 00:10:34.080
of acoustic thermometry in terms of
low cost and high accuracy.

00:10:34.080 --> 00:10:39.920
In the 10 years, ATOC spent about
$35 million and is able to measure

00:10:39.920 --> 00:10:46.240
temperature change with accuracy
as high as 20 millidegrees.

00:10:46.240 --> 00:10:51.600
However, ATOC stopped in 2006.
An important reason for that is its

00:10:51.600 --> 00:10:56.880
potential impact on the environment.
ATOC got troubled in lots of media

00:10:56.880 --> 00:11:02.320
reports and concern from the public
because these manmade sources might

00:11:02.320 --> 00:11:07.600
disorient whales or even kill them.
The ATOC group spent $6 million

00:11:07.600 --> 00:11:13.336
to investigate this issue and concluded
no significant biological impact.

00:11:13.360 --> 00:11:18.456
So far, I would say it’s not easy to
get a conclusion regarding whether

00:11:18.480 --> 00:11:23.040
or how much these active sources
could affect marine animals.

00:11:23.040 --> 00:11:27.047
But it did raise the permitting
issues for these experiments.

00:11:27.047 --> 00:11:32.720
And here is some recent news and
a review paper published last year.

00:11:32.720 --> 00:11:38.320
You can see the increasing concerns
about the impact of ocean noises,

00:11:38.320 --> 00:11:42.320
acoustic noises, made by
different human activities,

00:11:42.320 --> 00:11:44.776
for example,
these wind farms.

00:11:44.800 --> 00:11:50.429
We should seriously consider how to
better manage these sound sources.

00:11:52.000 --> 00:11:56.720
We were not allowed to use active
sources. We have other choices.

00:11:56.720 --> 00:12:01.360
You know the ocean is a world of sound.
This figure shows different kinds of

00:12:01.360 --> 00:12:10.000
sound sources in the ocean. So green
colors are sources from marine animals.

00:12:10.000 --> 00:12:15.600
Orange represents human activities,
where ATOC is sitting here,

00:12:15.600 --> 00:12:19.896
and the blue colors,
the natural sources.

00:12:19.920 --> 00:12:24.240
Since the concept of acoustic
thermometry proposed in 1980s,

00:12:24.240 --> 00:12:28.880
various types of naturally occurring
sources including volcanoes,

00:12:28.880 --> 00:12:31.760
whales, and ambient noises, even.

00:12:31.760 --> 00:12:35.896
They have been suggested
to be used for thermometry.

00:12:35.920 --> 00:12:40.136
However, none of them get
successful at basin scale,

00:12:40.160 --> 00:12:43.016
hundreds or thousands
kilometer scale.

00:12:43.040 --> 00:12:48.160
One important reason for that
is the energy of these sources.

00:12:48.160 --> 00:12:53.760
As you can see in this figure, so the
X axis is frequency, and the Y axis

00:12:53.760 --> 00:12:59.896
is power spectral density. Basically,
how powerful these sources are.

00:12:59.920 --> 00:13:04.880
Many of these sources cannot produce
the strong signals to be observed

00:13:04.880 --> 00:13:09.416
at basin scale because they
are not sufficiently powerful.

00:13:09.440 --> 00:13:14.160
In contrast, earthquakes are the
most powerful source to produce

00:13:14.160 --> 00:13:19.360
strong low-frequency
sound waves on our planet.

00:13:19.360 --> 00:13:21.736
So earthquake could
be a good choice.

00:13:21.760 --> 00:13:26.720
However, some obstacles are
present if we want to use earthquake

00:13:26.720 --> 00:13:29.920
to do thermometry.
So next, let’s see how we

00:13:29.920 --> 00:13:34.974
overcome these difficulties to
make the earthquakes work.

00:13:35.760 --> 00:13:40.240
Here is a example of sound
waves generated by earthquakes.

00:13:40.240 --> 00:13:43.600
And the bottom here is
a typical seismogram.

00:13:43.600 --> 00:13:49.888
We can see the first arrival P wave
and the barely visible S wave.

00:13:49.920 --> 00:13:54.400
Because it’s filtered at this
frequency band, relatively high

00:13:54.400 --> 00:13:57.736
frequency band, so S wave
just disappears.

00:13:57.760 --> 00:14:04.320
After half hour – after half hour
of the earthquake occurrence,

00:14:04.320 --> 00:14:09.600
you can see this very big signal.
This signal is the sound waves

00:14:09.600 --> 00:14:12.880
generated by the earthquake.
So the earthquake generated P and

00:14:12.880 --> 00:14:17.920
S waves, and then they get coupled into
the ocean, propagating as sound waves.

00:14:17.920 --> 00:14:20.720
Finally, they are detected
by seismometers.

00:14:20.720 --> 00:14:27.760
So this wave is the acoustic waves
generated by earthquake, and we call it

00:14:27.760 --> 00:14:34.856
a T wave, or tertiary, because it
arrives after P and S waves.

00:14:34.880 --> 00:14:40.480
Generally, T wave can be readily
observed at this frequency band.

00:14:40.480 --> 00:14:46.960
And, in the next part, as I said, I will
first show our results just using one

00:14:46.960 --> 00:14:52.085
frequency band, this frequency band,
and then I’ll talk about how to use

00:14:52.085 --> 00:14:57.736
multiple frequency bands to do depth
tomography of ocean [inaudible].

00:14:57.760 --> 00:15:01.120
Also submarine earthquakes
usually produce strong T waves.

00:15:01.120 --> 00:15:05.840
Not all of them they are usable for SOT,
or seismic ocean thermometry.

00:15:05.840 --> 00:15:11.760
Because T wave is quite complicated,
and the locations and origin times of

00:15:11.760 --> 00:15:17.200
earthquakes are not accurate
enough to make the absolute times

00:15:17.200 --> 00:15:21.256
of T wave useful for
ocean thermometry.

00:15:21.280 --> 00:15:26.320
And our solution for this problem
is using repeating earthquakes.

00:15:26.320 --> 00:15:32.160
So basically, we follow the same idea
of repeating sources as previous active

00:15:32.160 --> 00:15:37.040
acoustic thermometry, but these
sources, they are natural earthquakes.

00:15:37.040 --> 00:15:41.576
They have no impact
on marine animals.

00:15:41.600 --> 00:15:48.640
And, as you can see, I put this quotation
mark for the word repeating because

00:15:48.640 --> 00:15:51.680
that definition of repeating
earthquakes is quite tricky.

00:15:51.680 --> 00:15:57.123
If only use waveform similarity
information, which we will use later,

00:15:57.147 --> 00:16:05.336
it’s quite hard to identify the overlapping
rupture of two earthquakes.

00:16:05.360 --> 00:16:11.200
So, if we want to get a strict
criteria, the two earthquakes, we want to

00:16:11.200 --> 00:16:14.880
identify they are repeating events,
they should be overlapping –

00:16:14.880 --> 00:16:19.360
they should be overlapping
rupture areas for this [inaudible].

00:16:19.360 --> 00:16:24.880
But what I want to say here is, for our
purpose, as long as the error

00:16:24.880 --> 00:16:29.440
due to repeating event location
difference is much smaller than that

00:16:29.440 --> 00:16:33.760
from ocean temperature change
is safe to use them for SOT.

00:16:33.760 --> 00:16:37.920
That means even there’s
no overlapping rupture area,

00:16:37.920 --> 00:16:43.976
it’s still, they are still
useful for our purpose.

00:16:44.000 --> 00:16:47.920
And, as Part I, I will show our results
from Sumatra subduction zone,

00:16:47.920 --> 00:16:52.696
which we think is the best location
for SOT for two reasons.

00:16:52.720 --> 00:16:56.240
First, it’s seismically active.
We know a few big earthquakes

00:16:56.240 --> 00:17:02.000
occurred in this region in the last two
decades, including the 2004 big Sumatra

00:17:02.000 --> 00:17:08.160
earthquake and the 2005 magnitude 8.6
Nias earthquake, which is shown by

00:17:08.160 --> 00:17:13.656
this orange star here and in
this big map, this orange star.

00:17:13.680 --> 00:17:18.560
Secondly, we have very good T wave
station DGAR located on this island,

00:17:18.560 --> 00:17:24.296
Diego Garcia, which is about 3,000
kilometers away from the earthquakes.

00:17:24.320 --> 00:17:30.560
So what we did here is collecting the
T wave data from DGAR as well as

00:17:30.560 --> 00:17:39.816
P and S waves at these three reference
stations, which we are using for identify

00:17:39.840 --> 00:17:45.120
earthquakes and deriving their relative
origin times between repeaters.

00:17:45.120 --> 00:17:49.896
So we will see this
in the next slide.

00:17:49.920 --> 00:17:55.733
And the time we are starting
is from 2005 to 2016.

00:17:55.733 --> 00:18:00.536
This red line just shows
the T wave path.

00:18:00.560 --> 00:18:03.280
And it samples the
northeastern Indian Ocean.

00:18:03.280 --> 00:18:07.520
This big map is just a zoom-in this
small source region here

00:18:07.520 --> 00:18:11.680
to show the earthquakes.
So the black stars are all the 4,000

00:18:11.680 --> 00:18:17.336
earthquakes we used, and the red stars
are repeating earthquakes we found.

00:18:17.360 --> 00:18:20.720
Many of these earthquakes,
they are aftershocks

00:18:20.720 --> 00:18:24.776
of the 2005 big
Nias earthquake.

00:18:24.800 --> 00:18:27.680
So let’s first look at one
repeating earthquake example.

00:18:27.680 --> 00:18:31.200
These two earthquakes
occurred in 2006 and 2008.

00:18:31.200 --> 00:18:35.680
At the top, what I’m showing you
are P and S waves recorded

00:18:35.680 --> 00:18:41.120
at this reference station PSI.
It’s very close to the earthquake.

00:18:41.120 --> 00:18:44.880
So this station is in Indonesia.
And, as you can see, the P wave

00:18:44.880 --> 00:18:50.456
takes only about 35 seconds
to reach this station.

00:18:50.480 --> 00:18:55.336
At the bottom, what I’m showing
you is the T wave from DGAR.

00:18:55.360 --> 00:19:01.040
As you can see, these two seismograms,
they are very similar to each other.

00:19:01.040 --> 00:19:06.560
So both in the P and S waves,
as well as in the T wave.

00:19:06.560 --> 00:19:13.200
But, for the T wave, for the T wave,
this plotting even after origin time

00:19:13.200 --> 00:19:17.920
correction. So this correction
is from the P and S waves,

00:19:17.920 --> 00:19:20.400
we do the cross-correlation
to get the correction.

00:19:20.400 --> 00:19:24.640
And, for the P wave, even after that
origin time correction, we can still

00:19:24.640 --> 00:19:29.920
see apparent time shifting
between these two seismograms.

00:19:29.920 --> 00:19:34.080
This travel time change,
it must be due to ocean change.

00:19:34.080 --> 00:19:37.360
And, if we use waveform
cross-correlation to measure it,

00:19:37.360 --> 00:19:41.120
it gives us this number,
negative 0.27 second.

00:19:41.120 --> 00:19:46.640
So a negative number here means T
wave in 2008 arrives earlier than 2006.

00:19:46.640 --> 00:19:52.936
And the inference is,
the ocean in 2008 gets warmer.

00:19:52.960 --> 00:19:58.800
Then we just apply our method to all the
4,000 earthquakes at the nearest region.

00:19:58.800 --> 00:20:04.080
And finally we get about 2,000
repeating pairs, which are composed

00:20:04.080 --> 00:20:08.480
of 900 individual earthquakes.
So, in this figure, each dot –

00:20:08.480 --> 00:20:15.120
each gray dot is one earthquake.
And we have 900 such kind of dots.

00:20:15.120 --> 00:20:20.456
You may notice that we have lots
of data in 2005 and 2006 here.

00:20:20.480 --> 00:20:24.640
Most of them, they are aftershocks
of the 2005 big Nias earthquake.

00:20:24.640 --> 00:20:30.056
And then the aftershock sequence
just decays as time goes on.

00:20:30.080 --> 00:20:34.376
So here we have about
2,000 pair measurements.

00:20:34.400 --> 00:20:39.680
But these pairs, they only tell us
the relative travel time change between

00:20:39.680 --> 00:20:46.296
repeaters. In other words, these pairs,
they are not linked to each other.

00:20:46.320 --> 00:20:53.440
So, in order to get the time series of
change, we need to do an optimization,

00:20:53.440 --> 00:20:57.521
or inversion, to make these pairs
connected with each other.

00:20:57.521 --> 00:21:02.376
And here we use a relatively simple
scheme to do the optimization.

00:21:02.400 --> 00:21:07.920
This scheme is minimizing the
cost function, L, in this equation.

00:21:07.920 --> 00:21:12.800
So, in this equation, tau is the travel
time anomaly of each earthquake,

00:21:12.800 --> 00:21:18.216
and you can see the index, i and j,
the represented earthquake.

00:21:18.240 --> 00:21:21.440
And delta-tau is the
measured travel time change.

00:21:21.440 --> 00:21:28.056
And this index, k,
represents the repeating pair.

00:21:28.080 --> 00:21:30.640
So this cost function is
composed of two terms,

00:21:30.640 --> 00:21:34.720
the first term of data misfit and
the second term of curvature.

00:21:34.720 --> 00:21:38.880
So the second term, the curvature term,
can be called the regularization term,

00:21:38.880 --> 00:21:43.120
if you want, that represents some
sort of smoothing we apply

00:21:43.120 --> 00:21:47.760
to the inverted results.
And, by this, we can get the inverted

00:21:47.760 --> 00:21:53.096
time series, which has an arbitrary but
common reference for all the pairs.

00:21:53.120 --> 00:21:57.096
And this blue line is
the final time series.

00:21:57.120 --> 00:22:02.480
As you can see, this blue line contains
many interesting features or variations.

00:22:02.480 --> 00:22:05.656
And we will discuss these
features in more detail later.

00:22:05.680 --> 00:22:10.400
Before we get into the detailed
oceanography interpretation for these

00:22:10.400 --> 00:22:14.960
variations, we need to work out
a critical tactic question.

00:22:14.960 --> 00:22:18.800
That, of course, is how to
convert this travel time change

00:22:18.800 --> 00:22:21.576
to ocean
temperature change.

00:22:21.600 --> 00:22:26.080
Basically, here we want to know
which part of the ocean our T waves

00:22:26.080 --> 00:22:30.776
are sensitive to.
And here, we choose SPECFEM2D

00:22:30.800 --> 00:22:33.920
to solve this problem of
wave propagation.

00:22:33.920 --> 00:22:39.360
I guess some of us are familiar 
with SPECFEM, but this is

00:22:39.360 --> 00:22:43.200
a spectral element method.
It’s a full numerical method.

00:22:43.200 --> 00:22:49.336
So bathymetry and other structure
complexities can be fully considered

00:22:49.360 --> 00:22:54.536
in the numerical simulation of
wave propagation by this tool.

00:22:54.560 --> 00:22:57.440
3D modeling is [inaudible]
computation are expensive.

00:22:57.440 --> 00:23:02.616
So here we simply buy the program to a
2D simulation along our T wave path.

00:23:02.640 --> 00:23:09.920
And the – sorry, and the 2D sound speed
profile in the ocean is derived from

00:23:09.920 --> 00:23:14.936
ECCO. So let me just briefly
introduce what ECCO is.

00:23:14.960 --> 00:23:19.680
ECCO is a 3D global ocean
state data product developed

00:23:19.680 --> 00:23:24.536
by JPL at Caltech and
many other collaborators.

00:23:24.560 --> 00:23:29.760
And what they do is they use all the
possible data, include Argo, satellites,

00:23:29.760 --> 00:23:33.360
and others – all the possible
oceanography data to do data

00:23:33.360 --> 00:23:37.760
simulation, and it gets the best
possible estimates of ocean state,

00:23:37.760 --> 00:23:41.840
including temperature.
Here we just use ECCO’s average

00:23:41.840 --> 00:23:47.200
temperature and salinity from
2005 to 2016 to build up our

00:23:47.200 --> 00:23:52.296
2D sound speed profile
and then run SPECFEM2D.

00:23:52.320 --> 00:23:58.696
So here, this figure shows our final
result of travel time sensitivity kernels.

00:23:58.720 --> 00:24:02.800
So these sensitivity kernels
that tell us which part of the ocean

00:24:02.800 --> 00:24:05.336
our T waves
are sensitive to.

00:24:05.360 --> 00:24:10.400
As we expect, T waves are very
sensitive to the ocean SOFAR channel,

00:24:10.400 --> 00:24:16.536
this part and these kernels that get
decayed to shallow and deep ocean.

00:24:16.560 --> 00:24:22.400
Almost 40% of these kernels, they are
located below the depth of 2 centimeters

00:24:22.400 --> 00:24:26.400
where Argo has no data.
That means all results are nicely

00:24:26.400 --> 00:24:33.656
complementary to data from Argo
and satellites, which is very nice.

00:24:33.680 --> 00:24:39.440
And then let’s look at what ECCO
predicts for this particular repeating

00:24:39.440 --> 00:24:43.736
earthquake pair.
And, in this middle figure here,

00:24:43.760 --> 00:24:49.600
what it shows is the ECCO temperature
change from this date to this date.

00:24:49.600 --> 00:24:56.616
As you can see, they are strong
temperature change near the sea surface.

00:24:56.640 --> 00:25:02.240
However, our T waves are not
sensitive to this shallow ocean at all.

00:25:02.240 --> 00:25:09.120
What T wave really cares is just the
product of sensitivity kernels and the

00:25:09.120 --> 00:25:16.056
temperature change. And I plot
this product in this right figure.

00:25:16.080 --> 00:25:21.200
So this is the ECCO prediction.
And, in this figure, if we do

00:25:21.200 --> 00:25:26.720
2D integration to sum up all the
contributions from each point in this –

00:25:26.720 --> 00:25:33.040
in this figure, and finally ECCO
predicts negative 0.14 second.

00:25:33.040 --> 00:25:36.640
And that number here
goes the same direction as our

00:25:36.640 --> 00:25:40.400
T wave observation or measurement,
which is encouraging.

00:25:40.400 --> 00:25:45.520
However, the amount of 0.14 second
is much lower than the observed

00:25:45.520 --> 00:25:52.056
0.27 second. In other words,
ECCO underestimates the warming.

00:25:52.080 --> 00:25:56.960
In order to relate temperature to
travel time change, here we define

00:25:56.960 --> 00:26:02.320
a parameter of weighted average
temperature change, which is this term.

00:26:02.320 --> 00:26:04.840
And the weightings –
the weightings are just the

00:26:04.840 --> 00:26:08.536
sensitivity kernels, K,
shown in this figure.

00:26:08.560 --> 00:26:14.160
And if we plot – sorry, if we plug the
sensitivity kernels into the denominator

00:26:14.160 --> 00:26:20.560
here, this term turns out to be 5.4.
And finally, we get this very simple

00:26:20.560 --> 00:26:24.000
relationship between
travel time change, delta-tau,

00:26:24.000 --> 00:26:27.176
and temperature change,
delta-big T.

00:26:27.200 --> 00:26:31.120
So, for example, for this particular
example, the travel time change,

00:26:31.120 --> 00:26:37.816
0.14 second, is converted to 27
millikelvin temperature change.

00:26:37.854 --> 00:26:43.520
I need to point out that our T wave here
can only tell us the temperature change

00:26:43.520 --> 00:26:49.656
averaged along the T wave path, both
on horizontal and depth directions.

00:26:49.680 --> 00:26:55.360
But we were not able to know how these
anomalies are distributed in the ocean.

00:26:55.360 --> 00:26:58.640
And here, this figure, it’s
the distribution from ECCO.

00:26:58.640 --> 00:27:03.388
It’s not from our T wave.
We don’t know this distribution.

00:27:04.640 --> 00:27:10.480
So we only know the integrated
number, this number, with T wave.

00:27:10.480 --> 00:27:13.680
So, with this equation, we know
how to convert travel time change

00:27:13.680 --> 00:27:16.960
to temperature change.
And let’s look at what we can

00:27:16.960 --> 00:27:21.280
learn from this time series
we obtained previously.

00:27:21.280 --> 00:27:25.496
So, for comparison, here,
three lines are plotted.

00:27:25.520 --> 00:27:32.240
Our T wave with blue line, Argo
with orange, and ECCO with green.

00:27:32.240 --> 00:27:37.840
The Argo and ECCO results here are
calculated in the same way we have

00:27:37.840 --> 00:27:42.296
used for this particular repeating
earthquake pair, this one.

00:27:42.320 --> 00:27:46.800
So we just repeated this
calculation for all the repeating pairs

00:27:46.800 --> 00:27:48.856
and get their
time series.

00:27:48.880 --> 00:27:53.120
So, in this figure the Y axis on
the left is travel time anomaly,

00:27:53.120 --> 00:27:57.280
and then we just use this equation
to convert it to average temperature

00:27:57.280 --> 00:28:01.336
anomaly, which is
Y axis on the right here.

00:28:01.360 --> 00:28:06.560
Note that the negative
travel time anomaly goes up here

00:28:06.560 --> 00:28:09.496
because that’s corresponding
to warmer temperature.

00:28:09.520 --> 00:28:14.720
And, as you can see, the travel
time anomaly is up to 0.4 second,

00:28:14.720 --> 00:28:20.216
and temperature change range
is about 80 millikelvins.

00:28:20.240 --> 00:28:23.920
Generally, if we look at these time
series, three time series, they are

00:28:23.920 --> 00:28:28.473
generally consistent with each other.
Our T wave result has a high

00:28:28.473 --> 00:28:35.440
cross-correlation number for 0.84 – 
or, sorry, 0.89 with ECCO and a relatively

00:28:35.440 --> 00:28:42.240
low number of 0.74 with Argo.
These good consistencies give us –

00:28:42.240 --> 00:28:45.896
give us more confidence that
what we are doing is correct.

00:28:45.920 --> 00:28:50.560
For example, we can see these features.
They are generally consistent with

00:28:50.560 --> 00:28:54.000
each other. However, some
discrepancies between them

00:28:54.000 --> 00:28:57.600
are obvious. For example,
these peaks – for example,

00:28:57.600 --> 00:29:03.600
this one, this one, and this one,
they are more or less underestimated

00:29:03.600 --> 00:29:08.616
in ECCO and Argo,
the green and orange lines.

00:29:08.640 --> 00:29:13.920
And we think this is due to the
applied smoothing and limited

00:29:13.920 --> 00:29:17.896
spatial and temporal resolutions
in Argo and ECCO.

00:29:17.920 --> 00:29:23.360
And our T wave here, we have
2,000 repeating pairs that give us

00:29:23.360 --> 00:29:27.576
a very good
temporal resolution.

00:29:27.600 --> 00:29:32.480
And if we take linear trends – take
linear trends of these three time series,

00:29:32.480 --> 00:29:38.240
T wave result gives a much larger
warming trend than Argo and ECCO.

00:29:38.240 --> 00:29:45.360
So, if we keep this change derived
from T wave in the next 100 years, the

00:29:45.360 --> 00:29:51.520
eastern Indian Ocean would be warmed
up by 0.4 degree, which is quite large.

00:29:51.520 --> 00:29:54.160
So, after 100 year,
0.4 degree.

00:29:54.160 --> 00:30:00.160
For Argo and ECCO, they are
about 0.3 degree after 100 years.

00:30:00.160 --> 00:30:05.490
So, for this region, our T wave gives
a much larger warming trend.

00:30:06.560 --> 00:30:10.800
And then we have many, many
earthquakes in 2005 and 2006 here

00:30:10.800 --> 00:30:15.280
after the big Nias earthquake.
That allows us to zoom in this black

00:30:15.280 --> 00:30:19.656
window and look at the variations
with the higher temporal resolution.

00:30:19.680 --> 00:30:23.200
So here this is about
a one-year time series.

00:30:23.200 --> 00:30:31.200
And, in this time series,
as you can see, these three time series,

00:30:31.200 --> 00:30:35.440
the six-month periodicity.
From here to here is six months.

00:30:35.440 --> 00:30:40.296
Such kind of cycle is quite clear
in all of these three time series.

00:30:40.320 --> 00:30:45.760
Because our T wave path is close to
the equator, and you can imagine the

00:30:45.760 --> 00:30:51.896
wind-driven ocean temperature contours
just go up and down twice a year.

00:30:51.920 --> 00:30:58.856
This half-a-year periodicity is as
expected from oceanography view.

00:30:58.880 --> 00:31:04.160
Similar as the previous result,
our T wave result – as the previous

00:31:04.160 --> 00:31:09.280
figure, our T wave result here is more
consistent with the Argo than –

00:31:09.280 --> 00:31:12.160
sorry, more consistent
with ECCO’s than Argo.

00:31:12.160 --> 00:31:15.680
We think the reason is that ECCO
using more improved 2D data

00:31:15.680 --> 00:31:20.011
simulation, which seems to
work quite well for this region.

00:31:22.800 --> 00:31:29.200
And, because they have lots of data.
And so next we can even zoom in this –

00:31:29.200 --> 00:31:34.720
we can even zoom in this about 
one month, or less than one month,

00:31:34.720 --> 00:31:39.040
time window and look at
the 10 days time variation.

00:31:39.040 --> 00:31:44.560
So even at this very short time scale,
as you can see, our T wave result

00:31:44.560 --> 00:31:48.400
is still in phase
with ECCO generally.

00:31:48.400 --> 00:31:52.160
ECCO has very weak trends because
ECCO floats can only provide

00:31:52.160 --> 00:31:57.040
a 10 days’ resolution and therefore
misses short time scale features.

00:31:57.040 --> 00:32:01.760
So overall, our method looks quite 
reliable. And we do see some signals not

00:32:01.760 --> 00:32:07.016
captured by ECCO and Argo,
which is quite encouraging.

00:32:08.094 --> 00:32:12.856
I was showing the results from
Sumatra, and it works not bad.

00:32:12.880 --> 00:32:18.616
But what we really want
is a real global application.

00:32:18.640 --> 00:32:25.896
A global T wave analysis by
Buehler and Shearer in 2003,

00:32:25.920 --> 00:32:31.840
this study does indicate that T wave
is readily observable, or is readily

00:32:31.840 --> 00:32:40.080
observable globally, not only in Indian
Ocean, this part, but in global ocean –

00:32:40.080 --> 00:32:45.200
Pacific and Atlantic Ocean.
But we want to confirm that SOT

00:32:45.200 --> 00:32:50.160
should also work in this – in these
regions in the global ocean.

00:32:50.160 --> 00:32:53.920
And we also know the
limitations of our method.

00:32:53.920 --> 00:33:00.271
The two biggest limitations of
our method are listed here.

00:33:01.520 --> 00:33:07.280
The first one is we rely on repeating
earthquakes, which we know

00:33:07.280 --> 00:33:10.960
the number – the number of
repeating events is limited.

00:33:10.960 --> 00:33:14.640
And the second one is
the lack of depth distribution.

00:33:14.640 --> 00:33:16.720
We don’t know
how the anomalies –

00:33:16.720 --> 00:33:20.696
the temperature anomalies
are distributed in the ocean.

00:33:20.720 --> 00:33:26.720
So next part I will try to convince you
that SOT can be applied in global ocean.

00:33:26.720 --> 00:33:29.760
And I will also talk about
the possible solutions

00:33:29.760 --> 00:33:33.786
to overcome these
two difficulties.

00:33:35.040 --> 00:33:40.560
I will use the CTBTO hydrophones to
answer the previous three questions.

00:33:40.560 --> 00:33:45.816
CTBTO stands for Comprehensive
Nuclear Test Ban Treaty Organization.

00:33:45.840 --> 00:33:50.480
So CTBTO have built up a global
network, including hydrophones and

00:33:50.480 --> 00:33:56.720
T wave stations in the ocean to
monitor nuclear explosions globally.

00:33:56.720 --> 00:34:01.176
And here we just use their
hydrophone data to do SOT.

00:34:01.200 --> 00:34:09.760
Presumably, all of these orange triangles
and squares, all of these stations, they

00:34:09.760 --> 00:34:17.576
can be used for SOT, and so far we
have done these three hydrophones.

00:34:17.600 --> 00:34:21.680
Today I will talk about these two
hydrophones – the results from these

00:34:21.680 --> 00:34:27.656
two hydrophones because this one –
this one the result is quite complicated

00:34:27.680 --> 00:34:34.160
from oceanography view, also from the
repeating earthquakes in this region.

00:34:34.160 --> 00:34:39.927
And, due to the limited time, let’s just
focus on these two hydrophones.

00:34:41.600 --> 00:34:47.256
So let’s first look at this
hydrophone, the H08 hydrophone.

00:34:47.280 --> 00:34:53.040
A great advantage of hydrophone data
is that they usually have high T wave

00:34:53.040 --> 00:34:58.616
signal-to-noise ratios, which makes
small repeating events usable.

00:34:58.640 --> 00:35:04.800
It’s quite nice that this hydrophone
H08 is almost co-located with DGAR.

00:35:04.800 --> 00:35:08.216
That allows us to do
a comparison between them.

00:35:08.240 --> 00:35:14.160
It turns out – it turns out that this
hydrophone H08 can record T waves

00:35:14.160 --> 00:35:19.600
from earthquakes at Sumatra
subduction zones with magnitude

00:35:19.600 --> 00:35:22.616
as small as
magnitude 3.

00:35:22.640 --> 00:35:26.320
However, many of these small
earthquakes, they are missing

00:35:26.320 --> 00:35:30.480
in the current catalogs.
So what we can do is, using the

00:35:30.480 --> 00:35:35.040
template match method to find out
these missing events – these missing

00:35:35.040 --> 00:35:39.520
small events, which are
represented by these orange bars.

00:35:39.520 --> 00:35:44.880
So finally, as you can see, these blue
bars, they are the earthquakes in the –

00:35:44.880 --> 00:35:50.320
in the existing catalogs from USGS.
And the orange bars are the new

00:35:50.320 --> 00:35:54.456
earthquakes that we find
using template matching.

00:35:54.480 --> 00:35:59.840
And, with this more complete
earthquake catalog, we apply our

00:35:59.840 --> 00:36:05.896
method to the hydrophone,
and finally we get about 3,000 repeating

00:36:05.920 --> 00:36:12.080
events from this hydrophone H08.
In contrast, only 900 repeating events

00:36:12.080 --> 00:36:18.240
from DGAR, so with the high
signal-to-noise ratio of hydrophone,

00:36:18.240 --> 00:36:23.576
we do increase the repeating
events by a factor of 3.

00:36:23.600 --> 00:36:27.680
So, using hydrophone does help us
make small repeating events usable,

00:36:27.680 --> 00:36:32.776
which we know occur much more
frequently than large earthquakes.

00:36:32.800 --> 00:36:38.720
And, as expected, if we look at the time
series of these two stations, so orange is

00:36:38.720 --> 00:36:43.600
from hydrophone and blue is for DGAR,
as you can see, they are generally

00:36:43.600 --> 00:36:49.336
consistent with each other except
this part – this period when –

00:36:49.360 --> 00:36:51.600
sorry, when the
hydrophone has no data.

00:36:51.600 --> 00:36:55.466
So this is the data gap
for the hydrophone.

00:36:57.120 --> 00:37:00.880
Next, let’s switch to
another path – this path.

00:37:00.880 --> 00:37:09.096
So, in this path, we are using the
hydrophone at the west Australia coast.

00:37:09.120 --> 00:37:14.696
And the earthquakes we are using
are the same earthquakes in this

00:37:14.720 --> 00:37:18.640
nearest region. Again,
in the time series plotted here,

00:37:18.640 --> 00:37:25.520
we have three time series –
T waves, Argo, and ECCO.

00:37:25.520 --> 00:37:29.520
And, in this time series, as you
can see, the one-year periodicity.

00:37:29.520 --> 00:37:32.936
For example, from here to here,
it’s one year.

00:37:32.960 --> 00:37:39.120
This kind of one-year cycle, or seasonal
change, is quite clear in all of these

00:37:39.120 --> 00:37:45.474
three lines, which is consistent
with our oceanography view.

00:37:46.480 --> 00:37:53.440
But, if we – but, if we look at the early
part of this time series, for our T waves,

00:37:53.440 --> 00:37:57.600
the blue line, you can
see there are some spikes –

00:37:57.600 --> 00:38:02.616
this one, this one, this one –
these spikes on the blue curve.

00:38:02.640 --> 00:38:06.776
And these spikes, they are
missing in ECCO and Argo.

00:38:06.800 --> 00:38:11.896
Actually, these spikes, they are
associated with mesoscale eddies.

00:38:11.920 --> 00:38:15.920
To confirm this interpretation,
we have checked out the sea surface

00:38:15.920 --> 00:38:23.600
elevation data, and we do find that
each of these spikes, is associated

00:38:23.600 --> 00:38:29.920
with sea surface elevation anomalies.
So here I have a animation of

00:38:29.920 --> 00:38:34.640
sea surface elevation –
sorry, sea surface anomalies.

00:38:34.640 --> 00:38:41.096
And let me –
let me play this animation.

00:38:41.120 --> 00:38:43.185
Wait one second.

00:38:45.760 --> 00:38:50.800
Yeah. So I want to draw your attention
to this part, the western Australia coast.

00:38:50.800 --> 00:38:55.440
And this part, in the summer, we will
see the hotspots. For example, here.

00:38:55.440 --> 00:39:00.160
This is the summer in 2005, and in the
summer, you can see these two hotspots.

00:39:00.160 --> 00:39:03.520
These hotspots, they are
corresponding to eddy fields.

00:39:03.520 --> 00:39:10.560
When these eddies heat up – heat our
T wave path, we will see these spikes.

00:39:10.560 --> 00:39:19.068
This is 2005. And let’s speed it up and
get into 2006 – the summer of 2006.

00:39:20.080 --> 00:39:25.840
Again, you can see these hotspots.
So here we have another evidence

00:39:25.840 --> 00:39:30.960
coming from sea surface elevation
data to confirm our interpretation.

00:39:30.960 --> 00:39:34.880
And that give us more confidence to say
whatever our measurements, they are

00:39:34.880 --> 00:39:40.419
two signals from the ocean – very
interesting signals from eddy fields.

00:39:41.600 --> 00:39:45.280
So these results, the previous results
I talked about, they are consistent

00:39:45.280 --> 00:39:51.570
with our understanding of ocean
dynamic processes in many ways.

00:39:53.209 --> 00:39:56.375
[silence]

00:39:56.400 --> 00:40:03.016
And these spikes, they are not seen
in the previous T wave path from

00:40:03.040 --> 00:40:06.640
H08 here because that path
is on the equator,

00:40:06.640 --> 00:40:10.936
and we don’t expect
eddy fields on the equator.

00:40:10.960 --> 00:40:16.720
This is previous – these previous results,
we are working on one frequency band

00:40:16.720 --> 00:40:21.176
and using that environment
to get the time series.

00:40:21.200 --> 00:40:25.440
In the next part, I will talk
about recent work about

00:40:25.440 --> 00:40:28.936
using multiple
frequency band.

00:40:28.960 --> 00:40:33.520
As I said, one frequency band can
only tell us average temperature change

00:40:33.520 --> 00:40:39.416
and without any spatial
distribution information.

00:40:39.440 --> 00:40:43.520
We cannot get that information.
Theoretically, different frequencies

00:40:43.520 --> 00:40:47.760
of T wave can help us
resolve the depth distribution.

00:40:47.760 --> 00:40:53.200
Because they have different sensitivity
kernels along the depth, which are

00:40:53.200 --> 00:40:58.960
shown in this right figure here.
As you can see, we have three

00:40:58.960 --> 00:41:05.176
frequencies, and they have different
frequencies along the depth.

00:41:05.200 --> 00:41:09.920
This is exactly the same idea as
we do for surface wave tomography

00:41:09.920 --> 00:41:13.600
in seismology. So, in seismology,
we are working on different periods

00:41:13.600 --> 00:41:21.680
of surface waves and get the seismic
speed distribution along the depth.

00:41:21.680 --> 00:41:25.256
And here we are
trying to do same thing.

00:41:25.280 --> 00:41:30.720
And we do see this frequency
dependency trend in the observed data.

00:41:30.720 --> 00:41:34.560
For example, the right figure here,
it shows one repeating earthquake

00:41:34.560 --> 00:41:44.936
example. So the Y axis is frequency, and
X axis is the T wave travel time delays.

00:41:44.960 --> 00:41:47.840
And I’ll just draw your
attention to this black line.

00:41:47.840 --> 00:41:53.120
This black line is corresponding to the
maximum cross-correlation number.

00:41:53.120 --> 00:41:59.120
So that tell us the measured travel
time delays as a function of frequency.

00:41:59.120 --> 00:42:05.816
And, as you can see, there’s clear
trend up to 4 hertz, so this trend.

00:42:05.840 --> 00:42:12.560
And such [inaudible] trend should
tell us something about the ocean

00:42:12.560 --> 00:42:17.496
temperature change
distribution along the depth.

00:42:17.520 --> 00:42:22.880
So we first try this idea with
hydrophone and get some –

00:42:22.880 --> 00:42:25.336
and do get some
interesting results.

00:42:25.360 --> 00:42:31.120
A quick way to check that frequency
dependency is looking at their

00:42:31.120 --> 00:42:40.296
differences, right? The differences –
the travel time change at, for example,

00:42:40.320 --> 00:42:46.000
in this figure here, I’m going to show the
frequency – the two frequency results.

00:42:46.000 --> 00:42:49.360
At the top is the travel
time change at 2 hertz.

00:42:49.360 --> 00:42:57.256
And, in the middle figure here, it’s the
difference between 4 hertz and 2 hertz.

00:42:57.280 --> 00:43:03.680
As you can see, these features in these
two figures, they’re generally in phase.

00:43:03.680 --> 00:43:09.840
So they go the same direction.
That means 4 hertz anomalies

00:43:09.840 --> 00:43:18.640
are larger than 2 hertz, or amplified.
And, to do it better and get the depth

00:43:18.640 --> 00:43:22.560
slice tomography, here we first
measure travel time change

00:43:22.560 --> 00:43:28.000
not only at these two frequencies.
We also measure at 3 hertz.

00:43:28.000 --> 00:43:31.760
So we have three frequencies –
2 hertz, 3 hertz, and 4 hertz.

00:43:31.760 --> 00:43:36.320
And then we conduct a singular
vector decomposition with these

00:43:36.320 --> 00:43:42.456
three frequency results for
some sort of depth tomography.

00:43:42.480 --> 00:43:46.800
And here, at the bottom here,
I’m showing you the SVD,

00:43:46.800 --> 00:43:52.240
singular vector decomposition, result.
So, in this result, we only keep the first

00:43:52.240 --> 00:43:58.720
two singular vectors because the third
singular vector has larger uncertainty.

00:43:58.720 --> 00:44:06.296
It has the smallest eigenvalue,
and so we just discard the third one.

00:44:06.320 --> 00:44:11.280
And so, in this result, it gives us the
temperature evolution at long depth,

00:44:11.280 --> 00:44:16.560
so vertical axis is depth.
And, as you can see – red is warm

00:44:16.560 --> 00:44:21.200
and blue is cold – as you can see,
there is a clear migration trend

00:44:21.200 --> 00:44:27.520
along the depth. So this migration is
expected from oceanography view

00:44:27.520 --> 00:44:33.576
if we interpreted them as long surface
generated equatorial waves.

00:44:33.600 --> 00:44:38.160
So our T wave path, H08,
is on the equator.

00:44:38.160 --> 00:44:42.640
And we have such [inaudible]
for equatorial waves in the ocean

00:44:42.640 --> 00:44:46.195
that perturbs the ocean
temperature change.

00:44:47.693 --> 00:44:51.095
[silence]

00:44:51.120 --> 00:44:56.320
And if we look at the – if we
look at the long-term change,

00:44:56.320 --> 00:45:00.341
the long-term change
as a function of – sorry.

00:45:01.865 --> 00:45:04.441
[silence]

00:45:04.466 --> 00:45:08.800
The long-term change, which is
plotted in the upper figure here,

00:45:08.800 --> 00:45:14.856
so this is long-term change along
the depth at different ocean depths.

00:45:14.880 --> 00:45:23.600
And, as you can see, the left figure here,
this is the measurement from Argo,

00:45:23.600 --> 00:45:27.280
ECCO, and hydrographic measurement.
Hydrographic measurement is the

00:45:27.280 --> 00:45:31.520
red line. This is a repeating
measurement. The first one is

00:45:31.520 --> 00:45:37.336
conducted in 2006, and the second
one in 2016, if I remember correctly.

00:45:37.360 --> 00:45:48.160
And then we just project these
trends using our two singular vectors.

00:45:48.160 --> 00:45:51.920
So the projected results are
plotted in the right figure here.

00:45:51.920 --> 00:45:56.000
So, in the right figure here,
as you can see, the T wave result

00:45:56.000 --> 00:46:02.536
is a warming trend all the way
down to depth of 4 kilometers.

00:46:02.560 --> 00:46:08.800
But, for – sorry, for ECCO and Argo,
below the depths of –

00:46:08.800 --> 00:46:14.776
sorry, below the depth of
2.5 kilometers, it’s not a warming.

00:46:14.800 --> 00:46:20.720
They indicate a colder temperature,
which is the opposite result of T wave.

00:46:20.720 --> 00:46:24.696
And, if we look at the hydrographic
measurements, the red line,

00:46:24.720 --> 00:46:28.856
it does show positive trend,
which means warming.

00:46:28.880 --> 00:46:35.095
So probably here, it means
our T wave result, it captures

00:46:35.120 --> 00:46:40.136
the warming trend in the
deeper – in the deep ocean.

00:46:40.160 --> 00:46:44.720
But, of course, both our T wave results
and the hydrographic environments

00:46:44.720 --> 00:46:53.570
have large uncertainties.
So this conclusion is quite preliminary.

00:46:57.200 --> 00:47:02.640
And then next, let’s look at this
T wave path that’s hydrophone H01.

00:47:02.640 --> 00:47:07.120
The same thing, we have these
two frequency results that

00:47:07.120 --> 00:47:11.256
go the same direction.
The features in this figure and

00:47:11.280 --> 00:47:16.376
this one go the same direction.
And if we look at the SVD solution,

00:47:16.400 --> 00:47:20.880
there is no migration.
If we go back here, this is migration.

00:47:20.880 --> 00:47:26.160
And there is no obvious migration
for this path, which is also expected.

00:47:26.160 --> 00:47:30.000
Because this path is not –
is not on the equator.

00:47:30.000 --> 00:47:34.400
Most part of this path is not on
the equator, so we don’t expect

00:47:34.400 --> 00:47:39.496
the equatorial waves – the effects
from the equatorial waves.

00:47:39.520 --> 00:47:45.736
And if we look at
the long-term change,

00:47:45.761 --> 00:47:50.233
all of them show some
sort of [inaudible].

00:47:51.360 --> 00:47:55.280
Okay, we have tried the idea of
frequency dependency, and it does

00:47:55.280 --> 00:48:02.216
give us some depth information.
However, this method fails at

00:48:02.240 --> 00:48:07.920
the frequency above 4 hertz.
The reason is the T wave coherence

00:48:07.920 --> 00:48:13.816
rapidly drops about 4 hertz.
As you can see in this figure,

00:48:13.840 --> 00:48:19.200
above 4 hertz, so above 4 hertz,
the CC number is so low that we

00:48:19.200 --> 00:48:23.440
cannot get reliable measurements.
This is a bit disappointing, and we

00:48:23.440 --> 00:48:27.256
are trying to sort out, why is this,
it drops above 4 hertz.

00:48:27.280 --> 00:48:32.936
One possibility is the
mode-dependent travel time change.

00:48:32.960 --> 00:48:37.520
Because different modes sample
different part of the ocean,

00:48:37.520 --> 00:48:43.040
as shown in the eigenfunctions
in the right figure here.

00:48:43.040 --> 00:48:47.440
So this figure shows two frequencies –
2 hertz and 5 hertz.

00:48:47.440 --> 00:48:51.360
So each frequency, we have two lines –
the solid line, dashed lined.

00:48:51.360 --> 00:48:56.800
The solid line is the fundamental
mode and dashed line is the Mode 2.

00:48:56.800 --> 00:49:02.960
There more higher modes, but here just
plotted – I just plot the first two modes.

00:49:02.960 --> 00:49:08.456
As you can see, the eigenfunctions,
they sample different oceans.

00:49:08.480 --> 00:49:15.976
And then, when the ocean temperature
change, it’s not uniform along the depth.

00:49:16.000 --> 00:49:20.560
The consequent travel time change of
these two modes, it would be different,

00:49:20.560 --> 00:49:26.080
right? And then, when we add up
these two modes together, and we will

00:49:26.080 --> 00:49:31.176
get different waveforms –
the waveform change.

00:49:31.200 --> 00:49:33.920
So this is – mm-hm?
- Wenbo, just jumping in to say

00:49:33.920 --> 00:49:36.684
we have about
10 minutes left.

00:49:36.708 --> 00:49:39.176
- Okay, yeah.
Thank you.

00:49:39.200 --> 00:49:46.936
Yeah, so this is the possible reason.
And it decreases the CC number.

00:49:46.960 --> 00:49:51.840
So, to understand this process,
we did some numerical experiment.

00:49:51.840 --> 00:49:56.480
And, in this numerical experiment,
what I did is we perturbed the ocean –

00:49:56.480 --> 00:50:03.487
the top 2 kilometer ocean, and then –
and then run the SPECFEM.

00:50:03.487 --> 00:50:08.800
So, on the right here, this is the
sound speed profile, and the

00:50:08.800 --> 00:50:13.816
color represents sound speed.
And on the right here, I plot –

00:50:13.840 --> 00:50:18.560
what I’m showing you is,
for these black lines, they are two

00:50:18.560 --> 00:50:24.960
seismograms from Event 1 and Event 2.
In Event 2, we changed the ocean

00:50:24.960 --> 00:50:29.200
temperature change in the – sorry,
we changed the ocean temperature

00:50:29.200 --> 00:50:32.640
in the top 2 kilometers.
As you can see, because temperature

00:50:32.640 --> 00:50:37.976
change and the waveforms change,
the T waves change.

00:50:38.000 --> 00:50:42.640
And however, if we look at the
individual mode, so here we have

00:50:42.640 --> 00:50:46.960
two modes – Mode 1 and Mode 2.
If we look at the individual mode,

00:50:46.960 --> 00:50:51.520
there is a travel time shift, right?
There’s travel time shift here as well as

00:50:51.520 --> 00:50:55.440
here. But the waveforms are
very similar to each other.

00:50:55.440 --> 00:51:00.560
If we use the – if we use the waveform
cross-correlation to [inaudible],

00:51:00.560 --> 00:51:04.320
as you can see, the Y axis is
the cross-correlation number.

00:51:04.320 --> 00:51:08.080
For individual mode,
the CC number, both of them,

00:51:08.080 --> 00:51:14.216
they are above 0.9.
But, for the black lines, it’s below 0.6.

00:51:14.240 --> 00:51:21.496
So this interprets the
drops of CC above the 4 hertz.

00:51:21.520 --> 00:51:25.840
Of course, this experiment is very
preliminary experiment, and we need

00:51:25.840 --> 00:51:32.560
to do more analysis about that.
But, if this interpreting is true, that

00:51:32.560 --> 00:51:38.800
means we can do mode tomography.
[inaudible] measured travel time change

00:51:38.800 --> 00:51:44.720
of each mode. This concept has been
proposed in active ocean acoustic

00:51:44.720 --> 00:51:51.576
community decades ago, but it was
not widely used for a bunch of reasons.

00:51:51.600 --> 00:51:59.120
One reason – one reason is that it needs
a vertical hydrophone array to separate

00:51:59.120 --> 00:52:02.080
the modes, which is
usually not available.

00:52:02.080 --> 00:52:06.960
Usually, what we have is just, you know,
one hydrophone data, for example, the

00:52:06.960 --> 00:52:14.677
previous CTBTO data, the hydrophone,
it’s one hydrophone at one site.

00:52:16.000 --> 00:52:21.176
And another difficulty
there is the mode coupling.

00:52:21.200 --> 00:52:26.616
So these modes coupling effects, they
are strong at high frequency, which is

00:52:26.640 --> 00:52:32.640
confirmed in active ocean acoustics.
However, the coupling effects might

00:52:32.640 --> 00:52:38.640
be weaker for T waves because T waves
are lower frequencies and would be

00:52:38.640 --> 00:52:44.056
not affected too much by the sound
speed heterogeneities in the ocean.

00:52:44.080 --> 00:52:49.280
So the same idea as we do in
seismology – in seismology for

00:52:49.280 --> 00:52:54.000
surface wave, you know, the
multiple modes of surface waves,

00:52:54.000 --> 00:52:58.080
and there are coupling effects there.
These coupling effects,

00:52:58.080 --> 00:53:02.856
they become smaller for longer
periods. The same thing here.

00:53:02.880 --> 00:53:10.000
If this is true, and then what we
need to – we only need to focus on

00:53:10.000 --> 00:53:14.720
is how to get vertical hydrophone
array data, right?

00:53:14.720 --> 00:53:19.440
And especially, can we get –
can we get a scalable and cost-efficient

00:53:19.440 --> 00:53:24.856
way to install hydrophone array –
vertical hydrophone array that helps us

00:53:24.880 --> 00:53:28.696
separate these modes and
do mode tomography.

00:53:28.720 --> 00:53:33.040
So briefly summarize.
We think SOT can be applied globally.

00:53:33.040 --> 00:53:37.360
And the hydrophone arrays could
help us make these more repeating

00:53:37.360 --> 00:53:41.520
events usable for SOT.
And we are still working on how to

00:53:41.520 --> 00:53:45.040
get depth information of
ocean change using different

00:53:45.040 --> 00:53:49.052
frequencies and
using different modes.

00:53:50.000 --> 00:53:54.000
And this is my conclusion.
SOT is feasible and accurate.

00:53:54.000 --> 00:53:59.638
It has no anthropogenic
interference with marine animals.

00:54:00.240 --> 00:54:06.320
Because our T wave samples the
deep ocean, it desirably complements

00:54:06.320 --> 00:54:10.000
other point measurements.
And, for the depth information,

00:54:10.000 --> 00:54:14.800
we are trying to work on multiple
frequencies and modes, which could

00:54:14.800 --> 00:54:20.216
give us the depth distribution
of ocean temperature anomalies.

00:54:20.240 --> 00:54:26.000
And I want to say is, eventually, the
optimal solution would be combining

00:54:26.000 --> 00:54:29.680
all the possible data, including
Argo, SOT, and satellites to do

00:54:29.680 --> 00:54:34.320
data simulation, and get the best possible
constraints on ocean temperature

00:54:34.320 --> 00:54:40.453
change. I think I will just stop here
and open for your questions.

00:54:42.047 --> 00:54:44.056
- Thank you, Wenbo.
That was a great talk.

00:54:44.080 --> 00:54:47.840
So, before we do questions, want to give
a chance for everyone to thank Wenbo.

00:54:47.840 --> 00:54:50.775
It was awesome.

00:54:50.800 --> 00:54:52.240
So now we’ll
open it up to questions.

00:54:52.240 --> 00:54:56.936
If anyone has a question, they can
raise their hand or type it in the chat.

00:54:56.960 --> 00:55:01.040
I guess I can start off with a question.
So I’m wondering, so seeing that Argo

00:55:01.040 --> 00:55:05.520
only has data to about 2004,
do you have any plans to extend this

00:55:05.520 --> 00:55:11.016
SOT to previous seismic events going
back to – I know we have data to, like,

00:55:11.040 --> 00:55:13.396
’90s, maybe ’80s, with digital data.
- Yeah.

00:55:13.396 --> 00:55:16.800
- And then maybe do you think
you could do this on analog records

00:55:16.800 --> 00:55:19.157
that go back to
maybe the ‘30s?

00:55:19.157 --> 00:55:23.040
- Yeah, that’s good question.
We did – we did try to do that.

00:55:23.040 --> 00:55:28.240
And, yeah, I took some
work to look at that.

00:55:28.240 --> 00:55:35.360
As you said, it’s analog data.
And I think currently, what we are

00:55:35.360 --> 00:55:42.880
show is we can do this back to 1990s.
And, before 1990s, most of the –

00:55:42.880 --> 00:55:49.656
most of the data, they are analog data.
And I think USGS had a project

00:55:49.680 --> 00:55:56.180
few years ago, which is – which is
trying to, you know, digitize this

00:55:56.180 --> 00:56:02.480
analog data. I look at some of this data.
It turns out it is quite challenging

00:56:02.480 --> 00:56:09.040
because these instruments,
their response is more sensitive

00:56:09.040 --> 00:56:14.960
to the long periods for our purpose –
long periods, like tens of seconds.

00:56:14.960 --> 00:56:18.720
So it’s quite hard to get the
1 hertz or a few hertz signals

00:56:18.720 --> 00:56:23.680
there from the analog data.
But, for – yeah, back to 1990, or after

00:56:23.680 --> 00:56:29.840
1990, yes. We are working on that.
We have the digital data.

00:56:30.849 --> 00:56:35.271
- Awesome. Any other
questions for Wenbo?

00:56:36.138 --> 00:56:38.731
Raise your hands,
in the chat.

00:56:41.440 --> 00:56:46.240
I guess one other question I had was,
so when you’re making the delta-T

00:56:46.240 --> 00:56:51.815
measurement on the T waves,
do you see an increase in the

00:56:51.840 --> 00:56:55.440
phase lag between the two repeating
events as you go into the coda?

00:56:55.440 --> 00:56:57.040
Is there some kind of –
is there a scattering effect?

00:56:57.040 --> 00:57:04.576
Or is it just like a similar phase, like,
throughout the entire T wave?

00:57:05.193 --> 00:57:09.818
- Yeah. It depends on –
you mean for T wave, right?

00:57:09.818 --> 00:57:13.200
- Yes. Yeah.
- Yeah. For T wave, the same is

00:57:13.200 --> 00:57:19.016
for the entire seismogram.
Duration is usually is, like,

00:57:19.040 --> 00:57:24.136
20 or 50 seconds.
It depends on which region

00:57:24.160 --> 00:57:29.520
the earthquakes are.
And, in this 20 or 50 seconds, it turns

00:57:29.520 --> 00:57:38.000
out the time lag is roughly similar.
There is no trend of time lag changing

00:57:38.000 --> 00:57:43.976
with the arrival time of T wave.
Usually we see this for

00:57:44.000 --> 00:57:48.320
P or S coda waves, right?
And there’s a crust change,

00:57:48.320 --> 00:57:53.760
and we see the change in the – in the
coda, the travel time lag is different.

00:57:53.760 --> 00:58:01.280
But, for T wave, it’s roughly the same.
Yeah, I think that also tells us is for

00:58:01.280 --> 00:58:03.280
the different parts
of these T waves.

00:58:03.280 --> 00:58:07.320
And think their sensitivity
kernels roughly the same.

00:58:09.255 --> 00:58:12.640
- That’s good to hear.
Any other questions?

00:58:12.640 --> 00:58:14.640
We’re getting towards the
end of the hour, so if anyone

00:58:14.640 --> 00:58:18.945
has a last question,
feel free to chime in.

00:58:20.974 --> 00:58:26.320
[silence]

00:58:26.320 --> 00:58:32.720
Okay. If not, let’s thank Wenbo again.
He’s agreed to stick around just to have

00:58:32.720 --> 00:58:38.560
an informal chat, so before you head
out, give him our thanks for a great talk.

00:58:38.560 --> 00:58:40.404
- Thank you.

00:58:42.466 --> 00:58:44.656
- Thank you, Wenbo.

00:58:45.794 --> 00:58:48.367
- Thank you.

00:58:51.120 --> 00:58:51.609
So I just …