WEBVTT
Kind: captions
Language: en

00:00:00.800 --> 00:00:03.980
[inaudible conversations]

00:00:04.980 --> 00:00:08.160
Good morning, everyone.
Welcome to seminar.

00:00:08.160 --> 00:00:11.840
Next week, our speaker will be Nana
Yoshimitsu from Stanford University,

00:00:11.840 --> 00:00:16.530
and she’ll be talking about Towards
Improved Stress Drop Accuracy.

00:00:16.530 --> 00:00:20.100
This week, Ruth Harris
will introduce our speaker.

00:00:23.800 --> 00:00:24.860
- Thank you.

00:00:26.060 --> 00:00:30.000
Hello. Happy to welcome
Sunny Park, who comes from

00:00:30.000 --> 00:00:32.890
freezing cold
[laughter] Harvard –

00:00:32.890 --> 00:00:38.199
no, Harvard University, where she
is a grad student with Miaki Ishii.

00:00:38.199 --> 00:00:42.070
And right now she is doing a lecture
tour of prominent institutions in the

00:00:42.070 --> 00:00:46.800
West Coast and East Coast, and we
got there. No, we got on the list.

00:00:46.800 --> 00:00:51.860
So Sunny was an undergrad at
Seoul National University, where she

00:00:51.860 --> 00:00:56.480
did two bachelor’s degrees – one in
economics and one in engineering.

00:00:56.480 --> 00:00:59.989
She did a master’s in geophysics,
and now she’s at Harvard,

00:00:59.989 --> 00:01:02.679
and she’s been working on
a bunch of really cool projects.

00:01:02.679 --> 00:01:05.640
And included among these
are a topic that’s really

00:01:05.640 --> 00:01:08.830
different from her talk today. So if you
are scheduled to talk with her this

00:01:08.830 --> 00:01:12.670
afternoon or join us for lunch, you know,
definitely ask her more about these.

00:01:12.670 --> 00:01:16.500
She’s done source properties
of very deep earthquakes –

00:01:16.500 --> 00:01:18.680
the deepest earthquakes
that we know about.

00:01:18.680 --> 00:01:22.020
And she’s also looked at
inversion studies of

00:01:22.020 --> 00:01:26.720
tomographic inversion to see what
we really can resolve and can’t resolve.

00:01:26.720 --> 00:01:31.100
Then today she’s going to talk about
surface – or near – closer to the

00:01:31.100 --> 00:01:35.740
Earth’s surface looking at velocity 
structures. So I welcome Sunny.

00:01:36.840 --> 00:01:39.960
[ Applause ]

00:01:39.970 --> 00:01:41.930
- Thank you.

00:01:41.930 --> 00:01:44.300
Is the good – is volume good?
Yeah.

00:01:45.120 --> 00:01:49.640
So, yeah, today I’m going to
tell you about the new approach

00:01:49.640 --> 00:01:53.320
to constrain the
near-surface seismic velocity.

00:01:54.780 --> 00:01:59.320
So first, I’ll tell you why
we care about near surface.

00:01:59.320 --> 00:02:03.240
And there are some conventional
methodologies doing this, so I’m

00:02:03.250 --> 00:02:08.679
going to tell you about that. And then
I’m going to talk about this work.

00:02:08.679 --> 00:02:13.670
So I’ll tell you how I’m doing this,
what’s the methodology.

00:02:13.670 --> 00:02:17.360
And then I apply this to
the real data, and I’ll show you

00:02:17.360 --> 00:02:20.600
some interesting results.
And then I’ll summarize.

00:02:22.300 --> 00:02:23.940
So, yeah, this is from USGS.

00:02:23.950 --> 00:02:30.060
So we have a lot of surface observations
which are direct observations.

00:02:30.060 --> 00:02:34.840
Here you’re seeing not just
geology, which are in color,

00:02:34.840 --> 00:02:38.110
also topography
you can actually see.

00:02:38.110 --> 00:02:43.390
The other surface observation would be,
for example, geochemical observation.

00:02:43.390 --> 00:02:48.000
We have all this direct measurement
at the surface, and we want to know

00:02:48.000 --> 00:02:53.100
how actually they propagate
or extrapolate into depth.

00:02:53.100 --> 00:02:58.790
So if you think about this Menlo Park,
and we want to know how the surface

00:02:58.790 --> 00:03:08.240
observation propagates into depth, the
simplest way to do this is to drill a hole.

00:03:08.240 --> 00:03:13.400
And what you can do is,
here we are drilling, and we can

00:03:13.410 --> 00:03:16.770
actually take this sample –
the rock from the depth,

00:03:16.770 --> 00:03:21.930
and measure from the lab whatever
properties you want to measure.

00:03:21.930 --> 00:03:23.959
For example, it could be
Young’s modulus,

00:03:23.959 --> 00:03:27.910
bulk modulus,
or porosity – whatever.

00:03:27.910 --> 00:03:36.160
You can also have source at
the top of the borehole,

00:03:36.160 --> 00:03:39.300
and you can have some
geophones actually at depth.

00:03:39.900 --> 00:03:45.120
Then you can actually measure
seismic velocities – P and S velocity,

00:03:45.120 --> 00:03:51.560
like vertical seismic profiling. And you
can actually measure the seismic speed.

00:03:51.560 --> 00:03:55.250
So this is the methodology
that would give us basically

00:03:55.250 --> 00:03:58.560
the most detailed
image of the subsurface.

00:03:58.560 --> 00:04:05.330
But on the other hand, as you know, it’s
also the most expensive methodology.

00:04:05.330 --> 00:04:07.270
Not only it’s expensive,
it’s also invasive.

00:04:07.270 --> 00:04:12.020
You might not want to dig
a hole in your backyard, so other way

00:04:12.020 --> 00:04:18.190
to do this is to do local seismic survey.
We have vibrator truck here,

00:04:18.190 --> 00:04:22.380
but you can also have explosive
or dynamite there as a source.

00:04:22.380 --> 00:04:29.430
You see the receivers in the red dots.
So you’re recording the waves that are

00:04:29.430 --> 00:04:34.190
traveling through the near surface.
So you are recordings reflections

00:04:34.190 --> 00:04:38.220
and refractions, and basically,
you can interpret them.

00:04:39.030 --> 00:04:41.220
You can also be
a source yourself.

00:04:41.220 --> 00:04:48.020
So here I am doing an experiment just
right across the department building.

00:04:48.020 --> 00:04:53.280
So I’m actually having the steel plate
there and hammering the steel plate.

00:04:53.280 --> 00:04:57.500
And there are these
geophones that are in red.

00:04:58.320 --> 00:05:03.740
This was the only field work I’ve
done through my Ph.D. [laughter]

00:05:03.740 --> 00:05:06.680
And actually it was
just part of the class.

00:05:06.690 --> 00:05:13.590
And we actually got the image of this
heat pump going through the –

00:05:13.590 --> 00:05:20.460
just below there, which was
represented as a low-velocity feature.

00:05:20.460 --> 00:05:24.150
So why am I spending all these
muscles and energy doing this

00:05:24.150 --> 00:05:27.270
and/or why are people
spending money to do this?

00:05:27.270 --> 00:05:33.220
As you all know, not only it’s
important for – in a scientific reason,

00:05:33.220 --> 00:05:39.220
extrapolating the surface measurement,
we also care about seismic hazards.

00:05:40.240 --> 00:05:43.160
As you know, especially
the shear wave speed

00:05:43.160 --> 00:05:48.260
has been known to be
controlling this ground shaking.

00:05:48.260 --> 00:05:52.860
And, as you see in the cartoon,
if we are having same earthquake,

00:05:52.870 --> 00:05:58.520
but it’s going to the high-velocity region,
the – usually amplitude is smaller.

00:05:58.520 --> 00:06:03.070
And if it’s propagating into low-velocity
region, the amplitude gets bigger.

00:06:03.070 --> 00:06:06.840
Not only that, additionally,
wave gets trapped in the layer,

00:06:06.840 --> 00:06:09.320
and we have
ringing effect.

00:06:09.320 --> 00:06:15.700
So we really want to know what
velocity we are actually sitting here.

00:06:15.700 --> 00:06:19.700
And one of the most widely
used parameter is Vs30,

00:06:19.700 --> 00:06:22.669
as many of you know.
And it’s used as an input to the

00:06:22.669 --> 00:06:29.260
ground motion prediction equations.
So here is a simple way to look at this.

00:06:29.260 --> 00:06:34.860
So here on the right axis, you’re
seeing the Vs30 as meter per second.

00:06:34.860 --> 00:06:37.900
So we’re looking at
really small values.

00:06:37.900 --> 00:06:42.460
And on the vertical axis, we’re
looking at the amplification factor.

00:06:42.460 --> 00:06:50.380
So as I’m pointing there,
around 1.1 kilometer per second, or 12 –

00:06:50.380 --> 00:06:54.979
yeah, 100 meter per second,
let’s say that’s amplitude factor of 1.

00:06:54.979 --> 00:06:57.580
Then compared to that,
what you have if you have

00:06:57.580 --> 00:07:03.540
much slower velocity is 5 times
or even more amplification.

00:07:03.540 --> 00:07:08.030
So today I want to talk about
how I’m getting to this Vs30,

00:07:08.030 --> 00:07:13.320
or this shallow speeds estimates but
without spending too much money,

00:07:13.320 --> 00:07:19.060
or actually no money [laughs],
and/or hammering yourself.

00:07:19.060 --> 00:07:24.289
And I’m going to do that using
polarization of body waves.

00:07:24.289 --> 00:07:27.800
And the easiest way to think
about polarization of body wave

00:07:27.800 --> 00:07:32.820
is when the earthquake is
happening just right under you.

00:07:32.820 --> 00:07:38.310
So this is the real data example
that happened in Japan 1978.

00:07:38.310 --> 00:07:42.610
This is a deep earthquake,
around to 80 kilometer.

00:07:42.610 --> 00:07:48.980
And the distance is basically zero
degree. So it’s really just under you.

00:07:48.980 --> 00:07:53.840
We have three component
recordings, so here it’s vertical.

00:07:53.840 --> 00:07:57.600
Here is
north-south component.

00:07:57.600 --> 00:08:00.860
And here is the
east-west component.

00:08:00.860 --> 00:08:04.400
As we all know,
P wave, we’re expecting

00:08:04.410 --> 00:08:08.150
the particle motion to
be parallel to the ray path.

00:08:08.150 --> 00:08:13.240
So in this case, we are looking at – we
are expecting the energy to be shown at

00:08:13.240 --> 00:08:19.110
more or less in the vertical component
rather than the horizontal component.

00:08:19.110 --> 00:08:22.300
So here we can actually
identify P wave here.

00:08:22.300 --> 00:08:26.830
In contrast, for S wave,
we know that the particle motion

00:08:26.830 --> 00:08:30.669
should be perpendicular
to the ray path.

00:08:30.669 --> 00:08:36.310
So in this case, we’re expecting all the
motion on the horizontal component.

00:08:36.310 --> 00:08:38.979
So we can actually
identify S here.

00:08:38.980 --> 00:08:42.680
We actually see almost
no energy at the vertical.

00:08:45.540 --> 00:08:51.830
And if – sorry – if earthquake is far
away, or it’s not right under you,

00:08:51.830 --> 00:08:55.709
it would basically just come in
at a certain angle, and we can actually

00:08:55.709 --> 00:09:00.720
use this directional measurement,
which I call the polarization.

00:09:01.850 --> 00:09:07.700
This measurement is actually
rarely exploited relatively.

00:09:07.710 --> 00:09:12.140
I mean, compared to, for example,
travel time information.

00:09:12.140 --> 00:09:16.080
In global seismology, we use travel
time or – yeah, even regional scale,

00:09:16.080 --> 00:09:20.640
we use travel time or waveform
information, which has time information

00:09:20.640 --> 00:09:26.470
in it, to image this near surface,
or in – even in global scale.

00:09:26.470 --> 00:09:33.240
But what if we had error in earthquake
origin time? Like, 5-second error?

00:09:33.240 --> 00:09:39.180
Then that would just 100% propagate
into this travel time parameter.

00:09:39.180 --> 00:09:44.860
However, if you actually had
5 seconds error in origin time,

00:09:44.860 --> 00:09:48.580
but if you’re actually just measuring
polarization direction, and as you –

00:09:48.580 --> 00:09:52.700
as long as you know where is the
P wave or where is the S wave,

00:09:52.700 --> 00:09:57.670
which you can actually identify from
polarization, this direction measurement,

00:09:57.670 --> 00:10:00.080
or angle measurement,
doesn’t change at all.

00:10:00.080 --> 00:10:06.850
And it’s not at all – it’s independent,
basically, from the origin time error.

00:10:06.850 --> 00:10:11.160
The other error we can think of is the
location error of the earthquake.

00:10:11.160 --> 00:10:16.230
Many times that is more uncertain
than the horizontal dimension.

00:10:16.230 --> 00:10:22.220
If there’s, say, 50 kilometer of error in
earthquake location, it would actually

00:10:22.220 --> 00:10:29.830
only make an error in polarization of
about 0.1%, which is really small.

00:10:29.830 --> 00:10:33.670
In contrast, for travel time,
it would be 5% error,

00:10:33.670 --> 00:10:37.500
which would be
50 times even more error.

00:10:38.390 --> 00:10:42.880
So what I’m all saying here is that – I’m
not saying we shouldn’t use travel time,

00:10:42.890 --> 00:10:46.921
but what I’m saying here is that
the polarization measurement

00:10:46.921 --> 00:10:51.160
is actually really a robust measurement
that we can actually use.

00:10:52.220 --> 00:10:57.220
So I’m going to use this to constrain
this near-surface layer, which I call

00:10:57.220 --> 00:11:04.360
having velocity V. And I have the
angle from the vertical, which I call i.

00:11:05.660 --> 00:11:12.640
And we can get ray parameter from 1D 
model – well-known 1D model IASP91.

00:11:12.640 --> 00:11:16.440
And we can just think about it
as simple Snell’s law where

00:11:16.440 --> 00:11:19.380
sine-i over V
is the constant.

00:11:20.540 --> 00:11:24.160
If we have low-velocity structure,
then basically the direction

00:11:24.160 --> 00:11:28.700
becomes much steeper,
and the angle becomes smaller.

00:11:28.709 --> 00:11:34.570
If we have higher velocity, it becomes
the opposite. We have a bigger angle.

00:11:34.570 --> 00:11:37.770
So that’s the basic intuition
how I would like to get to

00:11:37.770 --> 00:11:40.740
this near-surface
seismic wave speed.

00:11:43.130 --> 00:11:46.240
In the next slide, I would like to
show you – I would like to

00:11:46.240 --> 00:11:51.380
turn it around and see, what if we
know the seismic velocity here?

00:11:51.380 --> 00:11:56.660
What’s the relationship of the
angle to the velocity is how we

00:11:56.670 --> 00:12:02.240
can actually get to velocity –
how we can actually get to the angle.

00:12:02.240 --> 00:12:08.540
So it turns out, basically it’s not
as simple as what I just told you.

00:12:10.120 --> 00:12:14.840
So let’s first look at
P wave incident case.

00:12:18.940 --> 00:12:23.400
So this derivation –
you might feel it’s boring.

00:12:23.410 --> 00:12:26.720
But actually, the result I’m
going to show you from the

00:12:26.720 --> 00:12:32.920
derivation is very surprising,
so I would like you to follow me.

00:12:32.920 --> 00:12:37.370
So as soon as P hits
this free surface,

00:12:37.370 --> 00:12:43.190
actually there is reflected P and S
wave recorded at this station too.

00:12:43.190 --> 00:12:47.089
So let’s say P is coming in
at the angle i.

00:12:47.089 --> 00:12:53.870
And let’s say the reflected angle of S
is i-b. And the reflected angle of P,

00:12:53.870 --> 00:12:58.340
as you know, will be just i –
same as the incident angle.

00:12:58.340 --> 00:13:01.830
And I call what I’m measuring
at this station i-bar,

00:13:01.830 --> 00:13:07.500
which is – I call
apparent incident angle.

00:13:09.180 --> 00:13:13.190
So what we are measuring at this
station is not just incident P wave,

00:13:13.190 --> 00:13:16.899
but actually the summation
of all three different phases –

00:13:16.899 --> 00:13:22.040
incident P wave, reflected P wave,
reflected S wave.

00:13:22.040 --> 00:13:25.450
We just said we know the
direction of all these three vectors,

00:13:25.450 --> 00:13:31.340
which are i – angle i, angle i,
and angle i-b for S.

00:13:32.400 --> 00:13:34.399
What about
the amplitude?

00:13:34.399 --> 00:13:39.779
So let’s say incident wave
amplitude is 1 – just normalize it.

00:13:39.779 --> 00:13:44.040
And then the reflected P wave
has amplitude of reflection coefficient

00:13:44.040 --> 00:13:48.600
at the free surface,
which is – which I say R-pp.

00:13:50.070 --> 00:13:55.360
And that you can calculate just solving
the free surface boundary condition.

00:13:55.360 --> 00:13:59.120
And it is a function of
both Vp and Vs.

00:14:00.200 --> 00:14:03.280
Same thing for S.
So the amplitude of S would be

00:14:03.290 --> 00:14:06.800
reflection coefficient
of P-to-S conversion.

00:14:06.800 --> 00:14:10.720
And it is also a function
of both Vp and Vs.

00:14:11.940 --> 00:14:16.260
In terms of the – in terms of these
angles, they are – given the ray

00:14:16.260 --> 00:14:23.720
parameter, they are only a function of
Vp for i’s, only a function of Vs for i-b.

00:14:24.520 --> 00:14:28.960
So now we have direction information
and amplitude information

00:14:28.960 --> 00:14:32.360
of all these three different vectors,
so we can actually do the

00:14:32.360 --> 00:14:36.620
vector summation to get to
see what we are observing.

00:14:39.320 --> 00:14:43.410
So this angle we are observing
at the surface – i-bar –

00:14:43.410 --> 00:14:48.320
I want to say is not necessarily
the incident angle i anymore.

00:14:49.080 --> 00:14:54.920
We can actually get to it by
taking the tangent of i-bar,

00:14:54.930 --> 00:14:59.209
and we can calculate this horizontal
component – total horizontal component

00:14:59.209 --> 00:15:06.010
by using the sine of this angle, amplitude
times sine angle, and sum them all up.

00:15:06.010 --> 00:15:10.300
You can do the same thing
for vertical using cosine.

00:15:10.300 --> 00:15:15.120
So you can actually derive this
horizontal and vertical component.

00:15:15.120 --> 00:15:22.220
And if you actually do this derivation,
the result is actually tangent-2i-b.

00:15:25.080 --> 00:15:30.000
So what this means is,
the observed angle, i-bar,

00:15:30.000 --> 00:15:34.980
is just 2 times S ray path angle.

00:15:36.980 --> 00:15:40.920
You’re not surprised by this?
[laughs]

00:15:40.920 --> 00:15:46.560
So the reason why this is surprising
is that I’m saying i-bar is not at all

00:15:46.570 --> 00:15:55.220
related to i. Or, as I showed you
here, i-b is just a function of Vs.

00:15:55.220 --> 00:15:59.480
So all I’m saying here is that
i-bar we are observing is

00:15:59.480 --> 00:16:04.399
only a function of
Vs and not at all Vp.

00:16:04.399 --> 00:16:07.480
So I was actually
surprised by this because

00:16:07.480 --> 00:16:12.589
you would expect this value
to be Vp- and Vs-dependent.

00:16:12.589 --> 00:16:16.080
Because both components
are Vp- and Vs-dependent.

00:16:16.080 --> 00:16:20.899
But actually they cancel out.
Vp cancels out somehow.

00:16:20.900 --> 00:16:24.480
And we actually just
end up with tangent-2i-b.

00:16:28.280 --> 00:16:32.660
So P wave is only
sensitive to Vs only.

00:16:32.670 --> 00:16:36.279
So I actually did this derivation
multiple times to confirm this

00:16:36.279 --> 00:16:41.220
because I couldn’t really
convince myself about that.

00:16:41.220 --> 00:16:43.360
But even if – yeah,
whatever I do,

00:16:43.360 --> 00:16:47.790
in a slightly different way,
I get to the same result.

00:16:47.790 --> 00:16:49.470
So I was really
surprised by this.

00:16:49.470 --> 00:16:54.300
This was the most surprising result
I have actually today, for me.

00:16:54.300 --> 00:17:00.970
And actually, it turns out this was known
more than 100 years ago by Wiechert.

00:17:01.600 --> 00:17:04.720
Wiechert, actually –
he is German.

00:17:04.720 --> 00:17:09.419
So he is actually the Wiechert
who built the seismogram.

00:17:09.420 --> 00:17:14.440
This is Wiechert’s seismogram that
I actually took picture of yesterday.

00:17:14.440 --> 00:17:18.020
It’s in the basement
of a Berkeley department.

00:17:19.440 --> 00:17:24.359
So he was actually interested in what we
are actually observing at these stations.

00:17:24.359 --> 00:17:29.509
So he did all this derivation and see
what are the direction we are looking at.

00:17:29.509 --> 00:17:36.720
And he did get this result saying
this i-bar equals just 2 times i-b.

00:17:37.680 --> 00:17:44.720
Well, he didn’t get a – get to this step
further and use this to constrain this Vs,

00:17:44.730 --> 00:17:48.570
but he did know that what
we are looking at at the surface

00:17:48.570 --> 00:17:51.660
is not related
to the angle i.

00:17:53.380 --> 00:18:02.799
So first implication here is that P is
not sensitive to Vp or bulk modulus.

00:18:02.799 --> 00:18:06.489
And the other reason why
this is interesting is that,

00:18:06.489 --> 00:18:12.950
if you think about it from a
Vs perspective, in global seismology,

00:18:12.950 --> 00:18:16.970
we might – we sometimes
constrain this using surface waves.

00:18:16.970 --> 00:18:21.550
Or one of the high-frequency
information we use to constrain

00:18:21.550 --> 00:18:25.600
Vs would be
S body waves.

00:18:25.600 --> 00:18:30.379
But here I am saying this is P body
wave, but it is sensitive to Vs.

00:18:30.379 --> 00:18:34.389
So Vs can be constrained
by P body wave, which is usually

00:18:34.389 --> 00:18:40.000
higher frequency than S body wave.
So I think of it as one of the

00:18:40.000 --> 00:18:44.480
highest-frequency information
we can get for constraining Vs.

00:18:45.680 --> 00:18:52.540
Then what about S incident case?
Would it be just sensitive to Vp only?

00:18:52.540 --> 00:18:57.360
It doesn’t turn out that way.
So S wave – this is the same thing.

00:18:57.369 --> 00:19:00.649
Now S wave is coming in
at an angle i-b.

00:19:00.649 --> 00:19:03.869
And P wave is coming
out at an angle i.

00:19:03.869 --> 00:19:10.989
And I say the apparent angle,
or the observed angle, as i-b-bar.

00:19:10.989 --> 00:19:15.120
We can actually do the same
derivation here, but in the interest of

00:19:15.120 --> 00:19:25.520
time, I’ll just say this angle i-bar is, as
we expect, a function of both Vp and Vs.

00:19:29.280 --> 00:19:33.520
So if we combine both
data from S and P,

00:19:33.520 --> 00:19:38.960
then we can constrain
both Vp and Vs at the surface.

00:19:38.960 --> 00:19:44.880
As I have just told you,
P data is just sensitive to Vs only.

00:19:44.880 --> 00:19:49.340
S data is sensitive
to both Vp and Vs.

00:19:50.340 --> 00:19:54.739
From Vp point of view, it is only
constrained by S wave data.

00:19:54.740 --> 00:19:58.600
Vs is constrained
by both P and S data.

00:19:58.600 --> 00:20:03.320
So, as you can imagine, if we
have more data to constrain Vs,

00:20:03.320 --> 00:20:07.359
plus P data is usually
cleaner than S wave data,

00:20:07.360 --> 00:20:12.460
so we actually have much
better constraint for Vs than Vp.

00:20:13.540 --> 00:20:18.480
So next I will show you the
actual application of this.

00:20:20.440 --> 00:20:25.580
So this is one station
in Japan – AGMH station.

00:20:27.800 --> 00:20:33.039
And there are about 250 events that
I’m using, which are in green dots.

00:20:34.640 --> 00:20:38.429
These are all teleseismic.
The reason why I’m using teleseismic

00:20:38.429 --> 00:20:42.809
is because, if we are getting into
regional or local events, then the

00:20:42.809 --> 00:20:48.649
earthquake error propagates more into
my measurements, and I don’t want that.

00:20:48.649 --> 00:20:52.450
And then I’m using
intermediate and deep events,

00:20:52.450 --> 00:20:56.629
meaning depth that’s
deeper than 60 kilometer.

00:20:56.629 --> 00:20:58.700
That’s because I don’t
want to deal with

00:20:58.700 --> 00:21:02.440
the depth phases coming in
in a closer time.

00:21:03.320 --> 00:21:09.300
And I’m using events from
2004 January to 2016 April.

00:21:09.309 --> 00:21:12.519
These events are
magnitude above 6,

00:21:12.520 --> 00:21:17.100
and I’m not using the ones that
are signal-to-noise ratio under 2.

00:21:19.480 --> 00:21:23.980
Let’s look at the one –
actual data from one of these event.

00:21:23.980 --> 00:21:27.880
And here is the
three-component data.

00:21:29.100 --> 00:21:34.659
This is an event from 2014
of magnitude 6.9,

00:21:34.659 --> 00:21:38.619
and distance is
about 70 degrees away.

00:21:38.619 --> 00:21:42.830
As you can see, we have, more or less,
all the energy in vertical compared to

00:21:42.830 --> 00:21:47.559
horizontal, so you can
already tell this is P wave.

00:21:47.559 --> 00:21:52.229
If you look at this five-second
time window and translate that

00:21:52.229 --> 00:21:57.770
into particle motion in 3D,
this is the particle motion.

00:21:57.770 --> 00:22:00.260
So basically, as time goes …

00:22:05.040 --> 00:22:10.720
As times goes, basically it’s
doing like this – this movement.

00:22:12.780 --> 00:22:16.620
You can actually already see this
dominant direction is in this direction.

00:22:16.629 --> 00:22:22.549
But we can actually get to it by using
principal component analysis, which can

00:22:22.549 --> 00:22:28.549
give me, then, the dominant direction,
which is the first principal component.

00:22:28.549 --> 00:22:31.769
And then, after getting this component,
I can just measure this angle

00:22:31.769 --> 00:22:36.549
from the vertical, which is
the i-bar I was talking about.

00:22:36.549 --> 00:22:43.639
If we measure this for all the events,
as I’ve shown you here –

00:22:43.639 --> 00:22:48.590
so we have events from 30 to 90 degrees
because they are teleseismic,

00:22:48.590 --> 00:22:52.960
so here is
the distance – 30 to 90.

00:22:55.300 --> 00:22:58.880
And each dot is
from each event.

00:23:00.020 --> 00:23:03.909
There are, for P wave,
220 measurements.

00:23:03.909 --> 00:23:07.450
I don’t have all the 250 measurement,
and that’s because about

00:23:07.450 --> 00:23:12.160
30 measurements are
signal-to-noise ratio under 2.

00:23:13.340 --> 00:23:19.320
And the color of each dot is basically
the robustness of the measurement.

00:23:19.320 --> 00:23:24.700
So if it’s – if it’s 1, that means
all the particle motion is ideally –

00:23:24.700 --> 00:23:30.139
it’s really linear motion.
But if it’s getting smaller values,

00:23:30.139 --> 00:23:33.930
that means it’s not – as I just
showed you, the real data, it’s not

00:23:33.930 --> 00:23:39.660
just one line, but it might have some –
an average elliptical motion.

00:23:41.520 --> 00:23:44.780
But, as you can see,
for P wave, actually most of the data

00:23:44.789 --> 00:23:50.039
are having 95-plus
percent robustness.

00:23:50.039 --> 00:23:54.729
For S wave, though – here are
the measurements of S wave.

00:23:54.729 --> 00:23:59.040
First of all, the number of measurement
is much smaller than P wave.

00:23:59.040 --> 00:24:04.000
It’s 130 measurements.
And you can also see the color

00:24:04.000 --> 00:24:09.580
of its dot has some red ones, which
are much less robust measurement.

00:24:09.580 --> 00:24:14.640
I’m basically using this color of dot,
or robustness of measurement,

00:24:14.640 --> 00:24:20.120
as a weight for each data point
when I’m getting to the velocities.

00:24:20.120 --> 00:24:27.480
So remember we are using this – both P
and S data to constrain both Vp and Vs.

00:24:27.480 --> 00:24:33.139
So what I can do is take all this
data for this station and basically

00:24:33.140 --> 00:24:38.120
do the grid search and find
the best-matching Vp and Vs.

00:24:39.330 --> 00:24:44.120
So here’s the research for all the
data I had in the previous slides.

00:24:44.130 --> 00:24:47.820
On the horizontal axis,
we have Vp going from

00:24:47.820 --> 00:24:51.019
zero to 7 kilometer per second.

00:24:51.019 --> 00:24:57.350
On the vertical axis, we have Vs going
from zero to 5 kilometer per second.

00:24:57.350 --> 00:25:02.900
The colored region is the region
where I’m doing this research.

00:25:02.900 --> 00:25:08.820
And the color contour is the
misfit where the red is the high misfit,

00:25:08.820 --> 00:25:12.970
blue is the ideally
smaller misfit.

00:25:12.970 --> 00:25:18.460
I was pretty happy about this.
Because first, the misfit contour

00:25:18.460 --> 00:25:22.419
is really well-defined, meaning
the minimum is really well-defined.

00:25:22.419 --> 00:25:25.730
So I don’t have to worry
about local minimas.

00:25:25.730 --> 00:25:29.489
The other thing is it only took
a few seconds to calculate this,

00:25:29.489 --> 00:25:32.420
so it was really
fast calculation.

00:25:33.720 --> 00:25:38.600
So now what I can do is
randomly re-sample these events,

00:25:38.600 --> 00:25:42.999
or data, to actually get to this
uncertainty of Vp and Vs

00:25:42.999 --> 00:25:47.580
by doing this research
many, many times.

00:25:47.580 --> 00:25:52.559
So in the next slide, I’m going to
show you the distribution of this

00:25:52.560 --> 00:25:58.210
grid search value from 500 times
re-sampled data sets.

00:25:59.060 --> 00:26:05.139
Here’s the uncertainty estimation by
doing bootstrapping of re-sampling

00:26:05.139 --> 00:26:07.820
of 500 times.

00:26:07.820 --> 00:26:14.800
So these are the research
results of 500 times of Vp and Vs.

00:26:16.860 --> 00:26:24.739
So, as you can already see, Vp is –
Vs is much more contracted than Vp.

00:26:24.739 --> 00:26:28.070
Here the dashed line is the
mean value, which I actually take

00:26:28.070 --> 00:26:34.849
as my final estimate value. So for Vp,
that is 4.2 kilometer per second.

00:26:34.849 --> 00:26:39.499
And I take the standard deviation as an
uncertainty because they follow nice

00:26:39.499 --> 00:26:45.440
Gaussian distribution. So in this case,
it’s 0.3 kilometer per second.

00:26:45.440 --> 00:26:54.320
And this case, for S, it’s 2.4 kilometer
per second with uncertainty of 0.1.

00:26:54.320 --> 00:26:59.789
So we do have much
smaller uncertainty for Vs.

00:26:59.789 --> 00:27:06.009
Now I would like to expand this
analysis to whole array of data.

00:27:06.009 --> 00:27:09.509
So here we are
looking at the Hi-net.

00:27:09.509 --> 00:27:14.859
There are more than 700 stations here,
so it’s really dense coverage.

00:27:14.859 --> 00:27:17.529
So if I actually get
Vs and Vp values

00:27:17.529 --> 00:27:23.779
for each station, basically
I can get a map of Vp or Vs.

00:27:23.779 --> 00:27:28.259
So that’s first good thing. And second
good thing about Hi-net is that it has

00:27:28.259 --> 00:27:33.999
benchmark data that I can compare my
result to, which I’ll come back to it.

00:27:33.999 --> 00:27:38.250
Since this is the new methodology,
I would like to know if the estimates

00:27:38.250 --> 00:27:42.760
I’m getting is actually comparable
to the benchmark data.

00:27:42.760 --> 00:27:45.220
For the earthquakes,
I’m using basically the same

00:27:45.220 --> 00:27:49.100
sets of earthquakes
for all these stations.

00:27:50.080 --> 00:27:54.700
Previously, we did this
analysis for this one station.

00:27:54.700 --> 00:27:58.969
So what I’m going to show you
in the next slide is this final value

00:27:58.969 --> 00:28:05.330
I get from bootstrapping –
in this case, 2.4 kilometer per second –

00:28:05.330 --> 00:28:08.869
as a colored dot
in this map.

00:28:08.869 --> 00:28:13.369
If I do that,
this is the result I get for Vs.

00:28:13.369 --> 00:28:15.571
From now, I’m going to
show you Vs first

00:28:15.580 --> 00:28:19.400
because they’re much
better constrained than Vp.

00:28:19.400 --> 00:28:26.800
And these are the values with the red
being slow and the blue being fast.

00:28:26.809 --> 00:28:32.960
The actual distribution of these values
look like this yellow histogram.

00:28:32.960 --> 00:28:36.519
Here the black dashed line
is the mean value.

00:28:36.520 --> 00:28:40.880
And the gray dashed line
is the IASP91 value.

00:28:42.429 --> 00:28:47.899
So this IASP91 value
is 3.36 for Vs.

00:28:47.900 --> 00:28:54.500
And that’s just one representative value
from surface to 10-plus kilometer.

00:28:55.820 --> 00:29:01.999
So this is actually giving me some
confidence that the value I’m getting

00:29:02.000 --> 00:29:06.999
are much shallower than what we
were looking at in this 1D model.

00:29:07.840 --> 00:29:12.399
We can do the same thing for Vp.
So this is the Vp map.

00:29:12.399 --> 00:29:17.840
Now color bar goes to as fast as
7 kilometer per second.

00:29:17.840 --> 00:29:19.919
And the distribution
looks like this.

00:29:19.919 --> 00:29:25.989
And again, most of the measurements
are much slower than IASP91.

00:29:25.989 --> 00:29:30.560
And the mean value is about
3 kilometer per second for Vp.

00:29:32.320 --> 00:29:35.120
And I mentioned that Hi-net
has this benchmark data.

00:29:35.129 --> 00:29:38.960
So what they actually have
is this borehole data.

00:29:38.960 --> 00:29:43.570
So they actually – this measurement
of velocities by having the source

00:29:43.570 --> 00:29:46.529
at the top and the
receivers at the depth,

00:29:46.529 --> 00:29:50.540
and they actually have these
velocity values of Vp and Vs.

00:29:51.700 --> 00:29:55.479
So I’m going to compare
my result to that.

00:29:55.480 --> 00:30:01.020
So here’s the comparison for Vs
with my estimate and the well.

00:30:01.020 --> 00:30:04.559
So, as you can see,
they’re in good correlation.

00:30:04.559 --> 00:30:09.779
Where it’s fast, I’m also
having the fast values.

00:30:09.779 --> 00:30:15.349
And the – here it’s relatively fast,
which is also true in well values.

00:30:15.349 --> 00:30:18.920
And the other regions
are relatively slower.

00:30:20.260 --> 00:30:26.560
I can do the same comparison for Vp.
They’re also in the good correlation

00:30:26.560 --> 00:30:34.559
where it’s similar parts of Japan is
fast and slow for both Vp and Vs.

00:30:34.559 --> 00:30:39.659
Now I can look at the map of residual,
meaning I can subtract these

00:30:39.659 --> 00:30:44.799
benchmark values from my
estimates and plot it on the map.

00:30:44.799 --> 00:30:49.460
So they look like this.
So this is the residual map,

00:30:49.460 --> 00:30:55.859
and now I’m plotting residual map for
both Vs and Vp in same color scale.

00:30:55.859 --> 00:31:01.749
So you can see that they distribute – they
are distributed around zero, whereas –

00:31:01.749 --> 00:31:06.000
where P wave has much
more scattered value.

00:31:07.000 --> 00:31:12.100
These are the actual histogram
of the residual for Vs and Vp,

00:31:12.110 --> 00:31:16.299
with the zero line marked
as the gray dashed line.

00:31:16.300 --> 00:31:20.460
And the red dashed line
is the mean residual.

00:31:20.460 --> 00:31:24.779
So first thing is that, yes,
my methodology works.

00:31:24.779 --> 00:31:29.409
It is very comparable
to the well data.

00:31:29.409 --> 00:31:34.239
And the second thing is that
you might be seeing that, for P wave,

00:31:34.239 --> 00:31:41.619
it’s really zero line. For S wave,
there’s slight positive residual here.

00:31:41.619 --> 00:31:46.509
So we are kind of actually
struggling interpreting this.

00:31:46.509 --> 00:31:50.629
First thing is that actually
it could be within the –

00:31:50.629 --> 00:31:55.739
it is comparable to our uncertainty.
And the second thing I found out is that

00:31:55.740 --> 00:32:03.580
actually the benchmark data we were
using is benchmark, but they also have

00:32:03.580 --> 00:32:09.040
quite amplitude of uncertainty there too,
which I don’t have the number of.

00:32:10.600 --> 00:32:18.320
But if this is actually significant result,
the S wave is faster than the reference,

00:32:18.320 --> 00:32:23.009
and P wave is more comparable than the
reference, one possibility that could be

00:32:23.009 --> 00:32:28.610
happening is that, as I told you,
Vs is more or less constrained by Vp,

00:32:28.610 --> 00:32:31.759
which is more or less
vertical motion in this case.

00:32:31.759 --> 00:32:37.379
And Vp, on the other hand,
is only constrained by S wave,

00:32:37.379 --> 00:32:40.849
which are, in this case,
more or less horizontal motion.

00:32:40.849 --> 00:32:45.889
So it is possible that P wave is
actually averaging over more depth,

00:32:45.889 --> 00:32:50.789
whereas S is averaging over just
this horizontal shallower layer.

00:32:50.789 --> 00:32:53.799
So that’s one explanation
we have currently.

00:32:53.800 --> 00:32:58.140
But if you have other ideas, yeah,
I would be interested to hear about it.

00:32:59.080 --> 00:33:06.539
Now actually look at all these
interesting patterns in lateral variability.

00:33:06.539 --> 00:33:12.580
First of all, interestingly, we see these
abrupt changes – not only there –

00:33:12.580 --> 00:33:17.139
in many other places because these
are really densely distributed stations,

00:33:17.139 --> 00:33:21.259
but there are these
abrupt changes in values.

00:33:21.260 --> 00:33:26.940
So that’s giving me some idea
about spatial resolution that I have.

00:33:28.769 --> 00:33:32.040
And then I mentioned
surface observation.

00:33:32.049 --> 00:33:38.320
So this is the geotectonic map of Japan.
So different color is from different

00:33:38.320 --> 00:33:45.309
time of – different geologic time,
where blue is the oldest.

00:33:45.309 --> 00:33:52.460
So what you can think of is, usually,
older the more consolidated and faster.

00:33:52.460 --> 00:33:59.080
So here it’s faster velocity,
which we also see here.

00:34:00.240 --> 00:34:04.220
This small streak here
we also see there.

00:34:05.190 --> 00:34:11.040
This part is relatively old,
and we see this part being faster.

00:34:11.040 --> 00:34:16.980
Only this part is actually
quite a low value here,

00:34:16.980 --> 00:34:20.070
except this small
streak of values.

00:34:20.070 --> 00:34:25.220
It turns out geology is not the
only thing we should compare to.

00:34:25.220 --> 00:34:29.359
For example, topography
is very important.

00:34:29.359 --> 00:34:37.250
So most of the cases, mountainous
region is represented as high velocity.

00:34:37.250 --> 00:34:41.500
And those basins,
or low-topography regions here

00:34:41.500 --> 00:34:44.750
are represented
as low velocity.

00:34:44.750 --> 00:34:49.700
So the part we just saw
actually is a basin structure.

00:34:49.700 --> 00:34:52.950
And these basins actually have
lots of populated cities,

00:34:52.950 --> 00:34:57.880
and they do care about
the seismic hazard as well.

00:34:59.130 --> 00:35:05.420
And you do see these low-velocity
regions here and here,

00:35:05.420 --> 00:35:09.789
which is still
high topography.

00:35:09.789 --> 00:35:14.910
And it turns out they are
correlated with volcanoes.

00:35:14.910 --> 00:35:19.269
So this is interesting.
So first, I was only thinking about

00:35:19.269 --> 00:35:23.279
seismic hazard when I was
looking into this methodology.

00:35:23.279 --> 00:35:29.170
But this is encouraging in a sense that
I might be sensitive to volcano as well.

00:35:29.170 --> 00:35:35.430
So I can expand this analysis to
look into volcanic hazard as well.

00:35:35.430 --> 00:35:39.780
So far, I was just looking at all
the earthquakes as a chunk.

00:35:39.780 --> 00:35:42.900
But if I divide this earthquake
as different times,

00:35:42.900 --> 00:35:47.580
I can actually see – maybe see
the change in the velocity.

00:35:48.829 --> 00:35:54.700
And the second thing is that, as you –
many of you might know,

00:35:54.700 --> 00:35:58.630
Vs30 is approximated many
times using topography –

00:35:58.630 --> 00:36:05.210
or actually topography gradient.
But here you saw these volcano regions,

00:36:05.210 --> 00:36:10.620
which are actually high gradient in
topography having low velocities.

00:36:12.220 --> 00:36:19.990
So we would like to know
actual velocities just by actually

00:36:19.990 --> 00:36:22.260
measuring them
and estimating them.

00:36:22.260 --> 00:36:30.039
Maybe there are some limitations just
approximating Vs30 using topography.

00:36:30.039 --> 00:36:35.859
The other thing for Vs30 is that that’s
one of the most widely used parameter,

00:36:35.859 --> 00:36:39.079
but people have been realizing
we need more information

00:36:39.079 --> 00:36:42.210
to characterize this site.
For example, if you’re sitting at

00:36:42.210 --> 00:36:45.980
the basin, you want to know
how deep this basin is.

00:36:45.980 --> 00:36:48.650
And you sometimes want to
know even deeper Vs,

00:36:48.650 --> 00:36:52.279
for example at
the 100-meter depth.

00:36:52.279 --> 00:36:59.140
So I would really like to know what all
these values that I get are sensitive to.

00:36:59.140 --> 00:37:01.819
What are – what are the
depths I am looking at?

00:37:01.819 --> 00:37:03.869
So in order to get
to depth information,

00:37:03.869 --> 00:37:09.420
I’m trying to introduce another
factor here, which is the frequency.

00:37:10.550 --> 00:37:15.080
So far, we have looked at Hi-net data,
which is really narrow band data.

00:37:15.089 --> 00:37:18.580
But if I have broadband data,
then I can actually play around with

00:37:18.580 --> 00:37:24.640
frequency by filtering them
into different frequency band.

00:37:24.640 --> 00:37:29.700
So basic intuition here is that,
if we have high frequency wave

00:37:29.700 --> 00:37:34.599
with the small wavelength,
then in this structure, we would be

00:37:34.599 --> 00:37:40.829
looking at this first layer, and I
get the estimate of this first layer.

00:37:40.829 --> 00:37:44.660
If you go into lower frequency
or longer wavelength,

00:37:44.660 --> 00:37:48.090
we are actually looking at
two of these layers.

00:37:48.090 --> 00:37:53.060
And the estimate I’m getting should
be the average of these two layers.

00:37:54.150 --> 00:37:58.440
If I go even lower frequency,
I’m looking at this – all the layer,

00:37:58.440 --> 00:38:01.880
and the average value
I get would be like green.

00:38:02.580 --> 00:38:06.640
Since we are usually expecting the
velocity to increase as a function of

00:38:06.650 --> 00:38:14.579
depth, if I lower the frequency, the value
will basically get higher and higher.

00:38:14.579 --> 00:38:16.400
So I actually tested this.

00:38:16.400 --> 00:38:24.250
So this is one station, also in Japan,
but it’s now a broadband station.

00:38:24.250 --> 00:38:30.380
So here you have the velocity
estimate as a function of frequency.

00:38:30.380 --> 00:38:34.240
I’m using all the same analysis,
but just filtering them into different

00:38:34.240 --> 00:38:41.440
frequency and get this Vs value.
So here, for 0.25 frequency hertz case,

00:38:41.440 --> 00:38:46.560
then it’s around – it’s bigger
than 3 kilometer per second.

00:38:46.560 --> 00:38:52.640
But if I do 0.5 hertz,
it’s getting lower.

00:38:52.640 --> 00:38:56.120
1 hertz,
it’s much lower.

00:38:56.130 --> 00:39:03.309
If I compare that to 1D model, actually –
IASP91 model values plot here,

00:39:03.309 --> 00:39:08.000
meaning this value is almost
approaching this 1D model.

00:39:08.000 --> 00:39:12.450
And here, this 1 hertz value is actually
approaching average Hi-net value,

00:39:12.450 --> 00:39:15.380
which is much
higher frequency.

00:39:15.380 --> 00:39:17.119
So that is encouraging.

00:39:17.119 --> 00:39:22.930
And I am currently working on how to
translate this velocity estimate as a

00:39:22.930 --> 00:39:28.750
function of frequency into velocity
as a function of depth.

00:39:28.750 --> 00:39:31.690
Turns out it’s not
that straightforward,

00:39:31.690 --> 00:39:35.839
but if you also have good idea,
I would like to hear about it.

00:39:35.839 --> 00:39:40.120
So we can also extend
this to whole F-net data.

00:39:40.120 --> 00:39:43.829
F-net is the
broadband data in Japan.

00:39:43.829 --> 00:39:47.519
This is not as dense as Hi-net,
but I can actually do the same thing

00:39:47.519 --> 00:39:50.559
for whole station.

00:39:50.559 --> 00:39:54.859
And I will plot this same way
as I did for Hi-net for Vp.

00:39:54.859 --> 00:40:03.339
So this is the 1 hertz case. Again,
the low value is in red. Blue is the fast.

00:40:03.339 --> 00:40:07.140
And the histogram of
these values look like this

00:40:07.140 --> 00:40:11.300
with the mean value around
1.7 kilometer per second.

00:40:12.460 --> 00:40:15.819
If I compare again to
Hi-net and IASP91,

00:40:15.819 --> 00:40:20.539
they plot in between
Hi-net and IASP91.

00:40:20.539 --> 00:40:26.349
I can change the frequency
to 1.5 hertz – lower frequency.

00:40:26.349 --> 00:40:30.260
And as you can see,
you see more blue color here.

00:40:30.260 --> 00:40:33.289
And the distribution of
values look like this.

00:40:33.289 --> 00:40:36.580
And the mean value
is shifting to the right.

00:40:36.580 --> 00:40:42.720
Again, if I even lower the frequency,
you get even more blue.

00:40:43.840 --> 00:40:47.820
And the distribution of
value look like this green.

00:40:47.820 --> 00:40:50.700
And as you can see,
all these mean value lines

00:40:50.700 --> 00:40:55.609
are shifting to the right
as I lower the frequency.

00:40:55.609 --> 00:41:02.480
If I compare this map to Hi-net map,
for sure, the 1 hertz map is

00:41:02.480 --> 00:41:07.260
the most comparable
map to Hi-net.

00:41:09.700 --> 00:41:13.079
So that concludes
my presentation.

00:41:13.079 --> 00:41:18.039
So I showed you the new approach
to look into near-surface velocity.

00:41:18.039 --> 00:41:22.829
And, as I told you, it’s actually
applicable to single station.

00:41:22.829 --> 00:41:26.349
It is non-invasive, and it’s
very computationally efficient,

00:41:26.349 --> 00:41:29.440
for example, compared to
noise tomography.

00:41:29.440 --> 00:41:34.910
And interesting thing was that
P wave is not sensitive to Vp,

00:41:34.910 --> 00:41:40.480
and it’s actually giving us highest
frequency information for Vs.

00:41:40.480 --> 00:41:44.440
We applied this to Hi-net
where we saw this methodology

00:41:44.440 --> 00:41:48.580
is working by comparing
to benchmark data.

00:41:48.580 --> 00:41:55.140
And we see some correlation with
geology, topography, and volcanoes.

00:41:55.140 --> 00:42:01.039
When we applied this to broadband data,
we saw that interesting trend

00:42:01.039 --> 00:42:04.930
as a function of frequency,
which I’m hoping to translate it

00:42:04.930 --> 00:42:08.329
into velocity profile
at each station.

00:42:08.329 --> 00:42:09.840
Thank you very much.

00:42:09.840 --> 00:42:15.980
[ Applause ]

00:42:17.180 --> 00:42:20.840
[ Silence ]

00:42:21.940 --> 00:42:23.260
- Questions?

00:42:24.360 --> 00:42:31.640
[ Silence ]

00:42:32.360 --> 00:42:37.269
- Yeah, thanks, Sunny. Since you
mentioned ambient noise tomography,

00:42:37.269 --> 00:42:42.120
how does your method – how do your
results compare with those results?

00:42:43.260 --> 00:42:51.440
[ Silence ]

00:42:53.760 --> 00:42:57.380
- Yeah. I thought I had
a slide for that, but …

00:43:00.340 --> 00:43:05.329
So – yeah,
I don’t think I have it.

00:43:05.329 --> 00:43:07.370
But anyway, yeah,
it is comparable.

00:43:07.370 --> 00:43:15.100
There’s a study using noise –
actually using borehole station

00:43:15.100 --> 00:43:18.670
and the surface station and trying
to get the velocity in between.

00:43:18.670 --> 00:43:23.480
So I compared to the values there,
and it is basically very comparable.

00:43:23.480 --> 00:43:27.690
They also get the values that
are comparable to the benchmark data

00:43:27.690 --> 00:43:31.319
I just showed you.
So, yeah, they’re in good agreement.

00:43:31.319 --> 00:43:36.019
But then, that is only specific
case of noise tomography,

00:43:36.019 --> 00:43:41.259
where you have the station at the
surface and at the borehole available.

00:43:41.259 --> 00:43:43.730
But actually that’s not the case –
most of the case.

00:43:43.730 --> 00:43:47.160
Meaning, like, U.S. array,
you just have surface at the station,

00:43:47.160 --> 00:43:49.660
so you can’t
actually do that.

00:43:49.660 --> 00:43:55.420
And if you just do usual
noise tomography, usually there’s –

00:43:55.420 --> 00:44:00.759
one of the current work using the noise.
And they have much better sensitivity

00:44:00.759 --> 00:44:04.690
from really deeper depth,
like 75-kilometer depth.

00:44:04.690 --> 00:44:08.160
So they are also getting
shallower and shallower

00:44:08.160 --> 00:44:11.859
by having denser
array like that.

00:44:11.859 --> 00:44:20.440
But, yeah, there is limitation to it
because you don’t have this surface

00:44:20.440 --> 00:44:23.030
and the borehole
stations everywhere.

00:44:23.030 --> 00:44:28.450
So if you just rely on this –
only surface station, I think there’s

00:44:28.450 --> 00:44:34.880
still some limitation in how shallower
you can get to using the noise, yeah.

00:44:36.100 --> 00:44:38.760
- One other question.
As you know very well,

00:44:38.760 --> 00:44:42.100
the P wave you’re observing
is a very high-frequency phase,

00:44:42.100 --> 00:44:44.180
like, let’s say, 10 hertz.
- Mm-hmm.

00:44:44.180 --> 00:44:47.000
- Whereas, the S wave is
perhaps 1 hertz or …

00:44:47.000 --> 00:44:50.319
- Mm-hmm.
- So you have different frequency

00:44:50.320 --> 00:44:52.920
content between the P and S wave.
- Right.

00:44:52.920 --> 00:44:56.160
- Does that introduce a
problem into your analysis?

00:44:56.170 --> 00:44:57.930
- So …
- You hinted at it, but I …

00:44:57.930 --> 00:44:59.740
- Right.
- Maybe you could just summarize.

00:44:59.740 --> 00:45:02.600
- Right.
So that is possible.

00:45:02.609 --> 00:45:07.619
For example, P wave is only
constrained by – P wave –

00:45:07.620 --> 00:45:15.680
Vp is only constrained by S.
So I am expecting P value to be –

00:45:15.680 --> 00:45:19.920
maybe having lower frequency
information than Vs.

00:45:19.920 --> 00:45:25.760
But then it is complicated because S
wave, as I told you, is horizontal motion.

00:45:25.760 --> 00:45:30.119
Whereas, P wave is vertical motion.
So it can actually average in vertical

00:45:30.119 --> 00:45:34.609
direction more than S wave.
So there is some complications like that.

00:45:34.609 --> 00:45:38.869
And I’m hoping to resolve that
while I’m looking at this velocity

00:45:38.869 --> 00:45:42.589
as a function of frequency.
And maybe hopefully I can get

00:45:42.589 --> 00:45:50.800
some sensitivity of Vs and Vp for
both different type of S and P phase.

00:45:50.800 --> 00:45:54.260
But, yeah, I don’t have
quantitative answer, but that’s –

00:45:54.270 --> 00:45:57.780
yeah, what I’m
thinking about it.

00:45:57.780 --> 00:46:00.009
- One last quick question.
Do you think you can do

00:46:00.009 --> 00:46:02.220
temporal measurements – you know, 
measurement this over time …

00:46:02.220 --> 00:46:03.660
- Yeah.
- … and watch the P wave –

00:46:03.660 --> 00:46:08.580
S wave change before a volcano erupts?
- Right. That’s actually what I said.

00:46:08.580 --> 00:46:11.480
[laughs]
But, yeah, so basically here I’m

00:46:11.490 --> 00:46:16.000
showing you the result from using
all the earthquakes that’s available.

00:46:16.000 --> 00:46:19.970
But if I actually divide them
into different time ranges,

00:46:19.970 --> 00:46:24.970
then I think that’s possible to,
yeah, monitor, for example, volcano –

00:46:24.970 --> 00:46:28.880
around a volcano, the change
of the velocity, yeah.

00:46:30.740 --> 00:46:33.720
[ Silence ]

00:46:34.700 --> 00:46:38.880
- Thank you, Sunny, for that
really interesting presentation.

00:46:38.880 --> 00:46:42.030
I may have missed it, but I –
what do you mean by – when you

00:46:42.030 --> 00:46:45.099
say near-surface velocity?
- Right.

00:46:45.099 --> 00:46:48.059
- Do you have a – have a sense of that?
And how do – how do the depth

00:46:48.059 --> 00:46:52.660
of the boreholes at Hi-net …
- Affect – yeah.

00:46:52.660 --> 00:46:56.760
- … affect that comparison?
- Right, right.

00:46:56.760 --> 00:47:07.920
So most of the benchmark data
I showed you, which are comparable

00:47:07.920 --> 00:47:12.579
to my result, are – these
benchmark values are basically

00:47:12.579 --> 00:47:16.079
measured from surface
to the borehole depth.

00:47:16.079 --> 00:47:19.720
And typical borehole
depth are 100 meter.

00:47:19.720 --> 00:47:23.200
So this is average value
from zero to 100 depth.

00:47:23.200 --> 00:47:28.470
So I’m still working on this
velocity as a function of frequency

00:47:28.470 --> 00:47:31.769
into converting into depth.
But actually, this is already

00:47:31.769 --> 00:47:35.109
giving me some idea that these
values are sensitive from

00:47:35.109 --> 00:47:39.190
surface to 100-meter depth.
So that’s one thing.

00:47:39.190 --> 00:47:44.490
And the second question about
Hi-net at the borehole is –

00:47:44.490 --> 00:47:48.280
yeah, that’s a really good question.
And I did think about that.

00:47:48.280 --> 00:47:52.210
So, for a surface station,
as I showed you, what we are

00:47:52.210 --> 00:47:56.480
looking at is this interaction
of three different waves.

00:47:56.480 --> 00:48:01.539
And what’s different for a borehole
station is that it’s slightly located

00:48:01.539 --> 00:48:07.000
at depth, and this reflection –
reflective P and S waves are

00:48:07.000 --> 00:48:10.599
actually arriving slightly
later than the incident wave.

00:48:10.599 --> 00:48:12.920
So that’s the only
difference here.

00:48:12.920 --> 00:48:17.549
So here I actually did a test where –
so just look at here.

00:48:17.549 --> 00:48:21.030
So this is when the
station is at the surface.

00:48:21.030 --> 00:48:24.740
And blue line is the
vertical particle motion.

00:48:24.740 --> 00:48:27.320
And the green line is the
horizontal particle motion.

00:48:27.320 --> 00:48:31.280
And this is particle
motion in 3D or 2D.

00:48:31.280 --> 00:48:35.160
And then I’m basically
measuring this angle from vertical.

00:48:35.160 --> 00:48:39.049
So this test is showing me –
I did the same test using the

00:48:39.049 --> 00:48:43.079
different borehole at different depths.
But actually, most of the Hi-net

00:48:43.079 --> 00:48:47.430
are at 100-meter depth, which is
showing me basically the same angle.

00:48:47.430 --> 00:48:51.809
So the uncertainty – slight difference
here is much smaller than uncertainty,

00:48:51.809 --> 00:48:57.200
so it actually doesn’t matter. To treat
them all as a surface station. Yeah.

00:48:58.140 --> 00:49:02.440
[ Silence ]

00:49:03.400 --> 00:49:05.079
- Thanks, Sunny.
I really enjoyed your talk.

00:49:05.079 --> 00:49:07.460
- Thank you.
- So I have a simple question.

00:49:07.460 --> 00:49:11.420
What happens if you have topography,
either at the Earth’s surface or at –

00:49:11.420 --> 00:49:16.200
on some horizon at depth?
Does that interfere with the converted

00:49:16.200 --> 00:49:21.660
phase, or does that work okay?
- So that is very possible.

00:49:25.200 --> 00:49:32.539
So, so far, basically I’m
assuming really simple model,

00:49:32.539 --> 00:49:35.609
just like layered model,
at each station.

00:49:35.609 --> 00:49:39.190
But there are some
interesting observations.

00:49:39.190 --> 00:49:44.009
For example, here –
this is F-net result –

00:49:44.009 --> 00:49:52.780
if you look at this station specifically,
it’s greener here, and it’s getting redder

00:49:52.780 --> 00:49:57.019
and redder, which is the opposite
from what we would expect.

00:49:57.019 --> 00:50:02.690
And this is actually one of the island
stations that’s covered by water,

00:50:02.690 --> 00:50:06.730
and it has this interesting
topography because it’s island.

00:50:06.730 --> 00:50:10.210
So there might be interesting
happening because of these

00:50:10.210 --> 00:50:15.360
reflections because of this
topography or water surrounding it.

00:50:15.360 --> 00:50:18.920
I haven’t thought about it after that,
but yeah, I’m sure there

00:50:18.930 --> 00:50:21.349
should be some effect.

00:50:21.349 --> 00:50:25.799
And I usually see more uncertainty
around the edge of the island –

00:50:25.799 --> 00:50:30.080
of the whole Japan island –
than in the very inland stations.

00:50:30.080 --> 00:50:34.060
So there might be actually that
topography gradient effect

00:50:34.060 --> 00:50:37.260
affecting those
reflections. Yeah.

00:50:40.520 --> 00:50:42.720
- Any more questions?

00:50:44.920 --> 00:50:48.400
Okay, we’re going to take
Sunny to lunch at the patio.

00:50:48.400 --> 00:50:54.360
We’ll meet at 11:45 at the flagpole.
Let’s thank our speaker again.

00:50:54.360 --> 00:50:55.380
- Thank you.

00:50:55.380 --> 00:50:59.960
[ Applause ]

00:51:00.800 --> 00:51:15.440
[ Silence ]