WEBVTT
Kind: captions
Language: en-US

00:00:00.099 --> 00:00:02.740
We’re very excited to have four
speakers today who are willing to

00:00:02.740 --> 00:00:11.000
give their AGU talks a week early.
So today we will have Fred Pollitz,

00:00:11.000 --> 00:00:15.860
Evan Hirakawa, Jean Soubestre –
who is visiting us from France,

00:00:15.860 --> 00:00:21.070
and then Sara McBride.
And before we start, I would just like to

00:00:21.070 --> 00:00:26.690
announce that we will not be having a
seminar next week because of AGU.

00:00:26.690 --> 00:00:32.879
And then, on the 18th, we will have
an AGU poster showcase during our

00:00:32.879 --> 00:00:36.199
normal ESC Coffee Hour
starting at 10:00.

00:00:36.199 --> 00:00:41.600
So, if you are presenting a poster or
have a poster from a recent conference

00:00:41.600 --> 00:00:47.269
that you would like to just, like,
put up and show off, please bring that.

00:00:47.269 --> 00:00:52.210
And also let us know so we can
contact you about exactly the rules

00:00:52.210 --> 00:00:55.060
about putting them up
here at Moffett.

00:00:55.060 --> 00:00:59.960
And then, other than that,
we won’t have a regular seminar

00:00:59.980 --> 00:01:04.020
until January 8, I believe,
so not until the new year.

00:01:05.400 --> 00:01:07.560
And then – let’s see.
For today, we’re going to

00:01:07.560 --> 00:01:13.310
try to stick to the 15-minute timing.
So I will be sitting over in that chair,

00:01:13.310 --> 00:01:16.580
and I will be timing it.
And, at 15 minutes, I will stand up.

00:01:16.580 --> 00:01:19.040
So if you’re speaking,
and you see me stand up,

00:01:19.040 --> 00:01:22.680
that’s time for you to wrap up
as quickly as possible.

00:01:22.689 --> 00:01:29.119
And then also, just to keep things
moving quickly, we probably

00:01:29.119 --> 00:01:32.659
won’t be passing around the
speaker that we have in this room.

00:01:32.659 --> 00:01:36.219
So, if you get a question,
can you just repeat the question

00:01:36.220 --> 00:01:38.660
before you answer it?

00:01:38.660 --> 00:01:42.060
Okay. And with that,
we have Fred.

00:01:46.880 --> 00:01:50.340
- Okay. Thanks for coming.
I’m going to talk about the postseismic

00:01:50.350 --> 00:01:53.750
deformation after the Ridgecrest
earthquakes and what they imply

00:01:53.750 --> 00:01:57.340
for the mechanics of
postseismic deformation.

00:01:57.860 --> 00:02:01.100
[inaudible]
[beeping]

00:02:03.920 --> 00:02:11.320
Okay. So the sequence is well-known,
but on July 3rd, local time, it was

00:02:11.330 --> 00:02:15.220
a magnitude 6.4 on a southwest-
northeast-trending structure.

00:02:15.220 --> 00:02:19.980
And that was followed 34 hours
later by a magnitude 7.1

00:02:19.980 --> 00:02:26.730
on a northwest-southeast-trending
structure, which ruptured about

00:02:26.730 --> 00:02:32.740
50 kilometers and produced
a spectacular surface rupture.

00:02:33.020 --> 00:02:35.820
Maximum surface slip
was about 5 meters.

00:02:35.820 --> 00:02:40.080
And the photo on the right,
provided courtesy of Josie Nevitt,

00:02:40.080 --> 00:02:47.440
was taken close to the magnitude 7.1
epicenter near the high-slip region.

00:02:49.780 --> 00:02:55.500
To do geophysical modeling of this
sequence of the foreshock and the

00:02:55.500 --> 00:02:58.260
main shock, we need to define
rupture planes, so I used these

00:02:58.260 --> 00:03:03.480
two black planes for the
magnitude 6.4 and three planes

00:03:03.480 --> 00:03:08.380
in the heavy white line
for the magnitude 7.1.

00:03:08.380 --> 00:03:14.540
And everything has assumed
right-lateral slip except for the

00:03:14.540 --> 00:03:18.370
southwest-northeast-trending
structure of the magnitude 6.4,

00:03:18.370 --> 00:03:21.960
which is assumed to
have left-lateral slip.

00:03:21.960 --> 00:03:25.590
So it’s of interest to compare the
postseismic motions that we infer

00:03:25.590 --> 00:03:29.220
from the Ridgecrest earthquake
to what happened 20 years ago

00:03:29.220 --> 00:03:36.300
after the Hector Mine earthquake.
So the processed transient motions

00:03:36.300 --> 00:03:39.470
after the Ridgecrest earthquake
are shown on the right.

00:03:39.470 --> 00:03:47.870
And, even if you take away some of the
large transient velocities that are due to

00:03:47.870 --> 00:03:53.620
some of the campaign stations on here,
the postseismic motions are 2-1/2 to

00:03:53.620 --> 00:03:57.060
3 times greater after the Ridgecrest
earthquake than they are after

00:03:57.060 --> 00:04:01.680
the Hector Mine earthquake.
And that could reflect more vigorous

00:04:01.680 --> 00:04:06.840
afterslip, or it might be viscoelastic
relaxation, or a combination of both.

00:04:08.240 --> 00:04:13.520
So our objectives are to identify
some of the usual postseismic processes

00:04:13.520 --> 00:04:17.250
for elastic rebound, afterslip,
viscoelastic relaxation of the

00:04:17.250 --> 00:04:21.060
lower crust and upper mantle.
Try to characterize the spatial

00:04:21.060 --> 00:04:24.780
and temporal distribution
of afterslip.

00:04:24.780 --> 00:04:29.250
Constrain the regional viscoelastic
structure of the ductile lower crust

00:04:29.250 --> 00:04:33.090
and upper mantle.
And all of this can be used to

00:04:33.090 --> 00:04:38.200
infer time-dependent crustal stresses,
which is of great interest,

00:04:38.200 --> 00:04:41.680
although I won’t be talking
about that in this presentation.

00:04:41.680 --> 00:04:45.560
So the postseismic
deformation data set.

00:04:45.560 --> 00:04:50.440
This is predominantly a
three-component time series

00:04:50.440 --> 00:04:56.920
from continuous GPS sites and a
number of survey mode GPS sites.

00:04:56.920 --> 00:05:03.970
I used a total of one year’s worth
of data to estimate the transient

00:05:03.970 --> 00:05:09.560
postseismic motions, and that’s
done to minimize seasonal effects.

00:05:09.560 --> 00:05:14.220
So the purpose of the – of the
processing is to isolate the transient

00:05:14.220 --> 00:05:18.440
postseismic motions, which are
over and above the background,

00:05:18.440 --> 00:05:21.670
which was going on before the
earthquakes actually happened.

00:05:21.670 --> 00:05:25.370
There is additional information from
InSAR and strainmeter time series,

00:05:25.370 --> 00:05:29.720
which also do show substantial
transients, but I haven’t folded that

00:05:29.720 --> 00:05:35.660
into the modeling just yet, but we can
compare with those kind of data.

00:05:35.660 --> 00:05:42.280
So, looking at just the continuous GPS
stations, over the first third of a year,

00:05:42.280 --> 00:05:48.180
the transient velocity field is shown
on the left for the horizontal motions,

00:05:48.180 --> 00:05:52.610
and this looks pretty typical for
what I’ve seen countless times for

00:05:52.610 --> 00:05:56.680
viscoelastic relaxation of the lower
crust and upper mantle, but it could

00:05:56.680 --> 00:06:00.040
be partially shaped
by afterslip as well.

00:06:00.040 --> 00:06:03.550
And a lot of the gaps in the
near field we will be filling in

00:06:03.550 --> 00:06:07.620
with the campaign GPS.
The vertical motions – there are

00:06:07.620 --> 00:06:16.740
some patterns that seem to jump out,
but the errors are very high, and I think

00:06:16.740 --> 00:06:19.980
we need longer time series before
they should be used in the model,

00:06:19.980 --> 00:06:24.600
so I’m not using those – just the
horizontals. So example time series.

00:06:24.600 --> 00:06:31.050
One of my favorite stations is P595,
and that time series for the three

00:06:31.050 --> 00:06:36.710
components shows offsets due to
the foreshock and the main shock,

00:06:36.710 --> 00:06:41.750
as well as a change in the slope of the
time series, which is hard to see in

00:06:41.750 --> 00:06:45.880
this figure, but if we zoom in on the
postseismic interval, this is referenced

00:06:45.880 --> 00:06:51.180
to the pre-seiscmic velocity,
so this is definitely a change in the

00:06:51.180 --> 00:06:55.210
character of the postseismic
crustal deformation field.

00:06:55.210 --> 00:07:00.360
And the average change in slope
is derived by fitting a simple

00:07:00.360 --> 00:07:05.780
quadratic function to each of
the postseismic time series.

00:07:06.240 --> 00:07:08.639
Getting a message
on this screen. Okay.

00:07:09.940 --> 00:07:13.560
I fit in a quadratic function,
and just using that to derive

00:07:13.560 --> 00:07:19.980
the average change in slope
over the first 125 days.

00:07:19.980 --> 00:07:27.139
And we can go through the same
exercise for another continuous GPS

00:07:27.140 --> 00:07:33.860
site, where we see very nice
decaying postseismic transients.

00:07:34.700 --> 00:07:39.220
And similarly, another site
west of the fault, which shows

00:07:39.220 --> 00:07:42.900
nice decaying transients.

00:07:45.840 --> 00:07:48.060
There’s also a wealth
of campaign data,

00:07:48.060 --> 00:07:52.910
and this is due to the very diligent
efforts of field crews from the USGS

00:07:52.910 --> 00:08:00.700
as well as the Southern California
Earthquake Center, particularly given

00:08:00.700 --> 00:08:07.860
the logistics needed to access the China
Lake Naval Air Weapons Station.

00:08:07.860 --> 00:08:17.020
So this field work was vigorous within
hours following the foreshock and

00:08:17.020 --> 00:08:21.430
throughout the month of July and has
continued off and on until the present

00:08:21.430 --> 00:08:24.740
time, and it yields a very valuable
data set for studying the near-field

00:08:24.740 --> 00:08:28.729
deformation. And, in combination
with data from continuous time series,

00:08:28.729 --> 00:08:33.829
it helps us resolve the different
postseismic processes.

00:08:33.829 --> 00:08:40.340
So this one time series that I show
here from CLR1, that’s located …

00:08:44.900 --> 00:08:48.339
It’s located right there on the fault.
This shows probably the largest

00:08:48.339 --> 00:08:51.529
displacements for many of the
campaign sites, and it’s due very likely

00:08:51.529 --> 00:08:56.540
to some kind of poroelastic rebound,
and that’s corroborated by

00:08:56.540 --> 00:08:59.220
interferogram analysis that
our colleagues from

00:08:59.220 --> 00:09:02.280
UC-Berkeley have
been carrying out.

00:09:02.280 --> 00:09:08.120
And we can do the same kind of
analysis to estimate the postseismic

00:09:08.120 --> 00:09:11.320
velocities at the campaign sites
that we do for the continuous sites.

00:09:11.320 --> 00:09:15.300
So this is an example
for site GS26.

00:09:16.360 --> 00:09:20.319
Here’s another example
for site GS19.

00:09:20.319 --> 00:09:26.499
Here you can see this one’s a bit
noisier, but the signal is still fairly

00:09:26.499 --> 00:09:33.399
robust for decaying – for detecting
a decaying postseismic signal.

00:09:33.399 --> 00:09:39.369
The verticals – this is probably
a testimonial – the same kind of

00:09:39.369 --> 00:09:42.050
analysis we’re doing for the
horizontals probably shouldn’t be

00:09:42.050 --> 00:09:46.520
done for the verticals, which I’m
not including in the model.

00:09:49.060 --> 00:09:54.760
There are postseismic interferograms,
so this one is from July 10 to July 16.

00:09:54.760 --> 00:09:58.540
It shows on the order of 10 millimeters’
displacement across the southern part

00:09:58.559 --> 00:10:02.790
of the fault, which is much smaller than
other signals that we’re looking at.

00:10:02.790 --> 00:10:09.639
And that signal nearly vanishes
in the July 16 to 22 interval.

00:10:09.639 --> 00:10:15.660
So this is suggestive that afterslip,
if it is going on, has a fairly short time

00:10:15.660 --> 00:10:18.600
constant on the order
of about two weeks.

00:10:21.100 --> 00:10:24.700
So the goal in the remainder of
this presentation is to interpret the

00:10:24.709 --> 00:10:29.240
transient postseismic velocity field
with models of either kinematic

00:10:29.240 --> 00:10:35.020
afterslip alone or combine
the afterslip and viscoelastic

00:10:35.020 --> 00:10:37.980
relaxation of the
lower crust and mantle.

00:10:40.020 --> 00:10:47.050
So I’ve defined potential afterslip to
be on the main shock and foreshock

00:10:47.050 --> 00:10:52.540
rupture planes, extended downward
from 17 kilometers all the way to the

00:10:52.540 --> 00:10:58.920
cross-mantle boundary at 30 kilometers.
And, when we insist that afterslip

00:10:58.920 --> 00:11:04.080
is all that we need to explain the data,
we get this fit that we see on the right.

00:11:04.080 --> 00:11:09.300
And it’s a somewhat good fit,
but there are systematic misfits,

00:11:09.310 --> 00:11:13.990
namely the model velocities
east of the fault are too small,

00:11:13.990 --> 00:11:17.620
and the model velocities
west of the fault are too large.

00:11:17.620 --> 00:11:20.120
And that’s basically because
there is an asymmetry in the

00:11:20.120 --> 00:11:23.600
observed velocity field.
And that is something that

00:11:23.600 --> 00:11:27.639
we can address with the
viscoelastic relaxation process.

00:11:27.639 --> 00:11:33.889
So, to embark on joint afterslip in
viscoelastic relaxation modeling,

00:11:33.889 --> 00:11:38.179
we need source models of the
foreshock and the main shock.

00:11:38.180 --> 00:11:42.660
We’re going to look at
two-dimensional viscoelastic structures,

00:11:42.660 --> 00:11:47.949
and we’re going to take all that
and try to fit the available data

00:11:47.949 --> 00:11:53.559
with a combination of afterslip
and the viscoelastic relaxation.

00:11:53.559 --> 00:11:58.060
So this is one of the source models
that we’ve derived, and we’re going to

00:11:58.060 --> 00:12:07.550
be looking at this in greater detail,
but essentially, we have up to

00:12:07.550 --> 00:12:14.670
5 meters of slip in the – in the main
shock at a fairly shallow depth.

00:12:14.670 --> 00:12:20.340
And on the order of 1 to 2 meters’
maximum slip in the foreshock.

00:12:20.840 --> 00:12:26.020
These models are going to be explored
in greater detail in a manuscript and –

00:12:26.040 --> 00:12:31.760
manuscript and preparation and
an AGU poster by Jessica Murray.

00:12:32.860 --> 00:12:37.059
So here I want to make the case
that we need very strongly laterally

00:12:37.059 --> 00:12:42.939
variable viscoelastic structure
to explain the data.

00:12:42.939 --> 00:12:49.869
And essentially, if we – if we look
at pairs of stations, for example,

00:12:49.869 --> 00:12:55.029
CCCC on the west side of the
fault and P594 on the east side,

00:12:55.029 --> 00:13:01.029
the displacements are much lower
on the west side of the fault.

00:13:01.029 --> 00:13:04.839
And we can look at other
stations here at similar distances

00:13:04.840 --> 00:13:07.100
and draw the
same conclusion.

00:13:07.100 --> 00:13:18.279
So, west of this boundary, the
viscoelastic structure seems to involve

00:13:18.279 --> 00:13:23.559
much longer relaxation times, much
longer – much greater viscosities.

00:13:23.559 --> 00:13:28.010
And that seems to be corroborated
by the pattern of heat flow that is

00:13:28.010 --> 00:13:32.869
shown here – much higher heat flow
in the Basin and Range side

00:13:32.869 --> 00:13:35.620
compared with the
Central Valley side.

00:13:37.260 --> 00:13:41.489
So we can define a viscoelastic
model on the Basin and Range side

00:13:41.489 --> 00:13:44.620
and Central Valley side with
a number of parameters.

00:13:44.620 --> 00:13:47.470
And that’s more parameters
than we would like to work with,

00:13:47.470 --> 00:13:56.439
so I simplify that by focusing on the
transient rheology, which is basically

00:13:56.440 --> 00:14:01.860
the Kelvin component of this analog
rheology that I showed on the bottom.

00:14:01.860 --> 00:14:09.100
And I assume that the ratios of the
transient viscosities are 100 in the

00:14:09.100 --> 00:14:12.540
lower crust to 5 in the mantle lid
to 1 in the mantle asthenosphere.

00:14:12.540 --> 00:14:15.449
So the weakest component is
where the temperature is greatest

00:14:15.449 --> 00:14:17.499
in the mantle asthenosphere.

00:14:17.499 --> 00:14:20.260
The lower crust is assumed to
be pretty strong, which is a result

00:14:20.260 --> 00:14:23.100
we obtain from other
postseismic studies.

00:14:23.100 --> 00:14:27.989
So there’s really only two
parameters to deal with

00:14:27.989 --> 00:14:33.779
when we make this simplification.
So, after running 60 different models

00:14:33.779 --> 00:14:40.939
and doing a joint fit for afterslip
and relaxation, in all cases,

00:14:40.940 --> 00:14:49.460
we arrive at an optimal viscoelastic
structure, which is shown on the left.

00:14:49.460 --> 00:14:57.160
And that involves really low viscosities
on the Basin and Range side.

00:14:57.160 --> 00:15:01.299
The lower crust viscosity is still
fairly high, but we have viscosities

00:15:01.299 --> 00:15:05.639
less than 10 to the 18 pascal-seconds
on the Basin and Range side, which,

00:15:05.639 --> 00:15:11.480
in this kind of work, is pretty low.
The contrast in viscosity is nearly

00:15:11.480 --> 00:15:15.309
10-to-1 between the Central Valley
side and the Basin and Range side.

00:15:15.309 --> 00:15:20.529
And we really need that in order to
account for the asymmetry in

00:15:20.529 --> 00:15:26.989
the postseismic crustal motion.
So, on the left, is the afterslip,

00:15:26.989 --> 00:15:30.960
which is associated with that
particular viscoelastic model.

00:15:30.960 --> 00:15:37.779
And on the right is the fit to the data.
And we are accounting for the

00:15:37.779 --> 00:15:46.029
asymmetry across the fault in this case.
So the picture on the left is the

00:15:46.029 --> 00:15:50.180
best picture that I have
so far of the – of the afterslip.

00:15:50.180 --> 00:15:53.649
It’s not as illuminating as
I would like it to be.

00:15:53.649 --> 00:15:56.689
One thing worth pointing out is
the shallow afterslip on the

00:15:56.689 --> 00:16:02.470
middle section seems to coincide
with a dip in the coseismic slip.

00:16:02.470 --> 00:16:08.029
So the coseismic slip contours of 3
and 4 meters are superimposed here.

00:16:08.029 --> 00:16:13.060
And there’s a bit of a dip in
the coseismic slip where we’re

00:16:13.060 --> 00:16:17.300
putting afterslip. So a tentative
conclusion would be that afterslip

00:16:17.310 --> 00:16:22.199
is filling in a gap in the coseismic
slip there, as well as at depth –

00:16:22.200 --> 00:16:25.700
well below where the
coseismic slip happened.

00:16:25.700 --> 00:16:35.040
And we can finally compare time
series based on this best viscoelastic

00:16:35.040 --> 00:16:38.779
and afterslip model, and we can
look at the separate contributions

00:16:38.779 --> 00:16:44.190
of afterslip and relaxation.
So the combined processes

00:16:44.190 --> 00:16:49.060
fit the time series as
well as we might expect.

00:16:49.060 --> 00:16:55.400
Afterslip is
insignificant at this site.

00:16:55.400 --> 00:17:01.340
We can do the same
exercise for other sites.

00:17:01.900 --> 00:17:05.320
This is the next to the last slide.
[laughs]

00:17:05.320 --> 00:17:11.779
So the time series that we predict
at the campaign sites gives us

00:17:11.779 --> 00:17:17.769
a handle on the time scale of relaxation.
The assumed time scale is too slow

00:17:17.769 --> 00:17:20.689
for this campaign site,
but it’s too rapid for this site.

00:17:20.689 --> 00:17:27.270
So that indicates different time scales
of afterslip in different regions.

00:17:27.270 --> 00:17:36.149
So, just to conclude, the main
conclusion is that the viscoelastic

00:17:36.149 --> 00:17:39.100
mantle relaxation explains
most of the postseismic velocities,

00:17:39.100 --> 00:17:45.149
but we also need afterslip.
And with continued observations,

00:17:45.149 --> 00:17:49.900
we can refine our models of
post processes. Thank you.

00:17:49.900 --> 00:17:54.540
[Applause]

00:17:55.620 --> 00:18:01.580
- We have one question [inaudible].

00:18:03.040 --> 00:18:09.980
- Is the viscoelastic structure that you
assume consistent with, say, Hector Mine

00:18:09.980 --> 00:18:16.880
and the SCEC [inaudible]
or rheology model?

00:18:16.880 --> 00:18:19.720
- And then, Fred, can you just repeat
the question for the [inaudible]?

00:18:19.720 --> 00:18:26.700
- Yeah. So, is the viscoelastic
structure consistent with the …

00:18:27.600 --> 00:18:31.800
- If you were to extend the Basin and
Range over towards Hector Mine,

00:18:31.800 --> 00:18:35.540
are you getting comparable values?
Or you get more lateral variation?

00:18:35.540 --> 00:18:39.080
- Oh, we would need
more lateral variations.

00:18:40.140 --> 00:18:47.059
I showed initially that the transient
deformation is about three times higher

00:18:47.059 --> 00:18:51.150
than it was following Hector Mine.
So the viscosities have to

00:18:51.150 --> 00:18:56.230
be correspondingly lower.
And so they may be jumping to

00:18:56.230 --> 00:19:00.440
lower values across the Garlock
Fault, by all appearances.

00:19:02.680 --> 00:19:04.620
- Okay. Thank you.

00:19:04.620 --> 00:19:07.600
Our next speaker will be
Evan Hirakawa.

00:19:10.040 --> 00:19:12.560
- Okay, thank you.

00:19:12.560 --> 00:19:15.260
[inaudible] green box.
It’s going to be really important.

00:19:15.260 --> 00:19:17.480
[laughter]
- Tried and failed to get it …

00:19:17.490 --> 00:19:24.059
- Okay. So I’m going to show some
results from some wave propagation

00:19:24.060 --> 00:19:28.580
finite difference simulations I’ve been
doing for the magnitude 7.1 Ridgecrest

00:19:28.580 --> 00:19:35.020
earthquake using this kinematic
rupture model that I’ve developed.

00:19:35.020 --> 00:19:38.550
So not much background
needed if you’re a seismologist.

00:19:38.550 --> 00:19:43.960
This has been a pretty famous event.
We were kind of startled by the

00:19:43.960 --> 00:19:48.610
6.4 earthquake, and then, before we
could gather ourselves, the biggest

00:19:48.610 --> 00:19:53.380
earthquake in 20 years in southern
California happened the next day.

00:19:54.850 --> 00:19:59.740
It takes place – the tectonic setting is
the Eastern California Shear Zone.

00:19:59.759 --> 00:20:03.409
The specific fault is called the
Little Lake Fault, and this is

00:20:03.409 --> 00:20:08.620
kind of a immature fault zone.
And so we have these very

00:20:08.620 --> 00:20:12.019
geometrically complex faults.
You don’t have these long,

00:20:12.019 --> 00:20:14.960
straight faults like we have
more in the western part

00:20:14.960 --> 00:20:17.620
of California where
it’s more mature.

00:20:17.620 --> 00:20:23.240
And we’ve started to see now
through a lot of the surface ruptures

00:20:23.240 --> 00:20:26.360
and a lot of new studies that are
coming out, that there’s a

00:20:26.360 --> 00:20:33.040
very complex pattern of rupture.
It’s likely that multiple fault surfaces

00:20:33.040 --> 00:20:36.480
slipped during both events –
during the magnitude 6.4

00:20:36.480 --> 00:20:39.080
and during the magnitude 7.1.

00:20:39.080 --> 00:20:44.140
And we’ve also, interestingly, heard,
right after the earthquake, that the –

00:20:44.140 --> 00:20:49.200
some of the real-time estimates at first
underestimated the magnitude of the

00:20:49.200 --> 00:20:55.020
large event, which, to me, suggests
that maybe there’s some temporally

00:20:55.020 --> 00:20:57.940
complex or, you know, slowly
developing rupture going on.

00:20:57.940 --> 00:21:05.919
So it’s turned into a very scientifically
interesting topic – earthquake to study.

00:21:07.920 --> 00:21:11.860
In addition, we’ve got pretty good
station coverage right around the fault.

00:21:11.860 --> 00:21:17.000
So these blue lines here is –
this is the surface trace of the –

00:21:17.000 --> 00:21:21.780
mainly from the 7.1 earthquake.
Here’s the left-lateral fault

00:21:21.780 --> 00:21:27.000
from the 6.4 earthquake.
The star here is the epicenter of the 7.1.

00:21:27.000 --> 00:21:30.660
And we see all these different kinds of
stations are right around this area.

00:21:30.660 --> 00:21:35.880
We have a large network of these
Caltech strong motion accelerometers,

00:21:35.880 --> 00:21:41.240
which are the red triangles.
I’m going to be showing some data

00:21:41.240 --> 00:21:47.880
from these GNSS high-rate GPS
stations that Jeff McGuire,

00:21:47.880 --> 00:21:52.779
Diego Melgar, and Brendan Crowell
helped me obtain and work with

00:21:52.779 --> 00:21:56.850
a little bit. And then there’s
three PBO strainmeters

00:21:56.850 --> 00:22:03.180
that Andy Barbour helped
me work with and process.

00:22:07.100 --> 00:22:12.300
So what I’m going to do is a finite 
difference simulation I used – SW4,

00:22:12.300 --> 00:22:15.180
which is a 4th order elastic
wave propagation code

00:22:15.180 --> 00:22:18.100
developed by the people
at Lawrence Livermore.

00:22:18.100 --> 00:22:24.240
I’m going to have a 200-by-200-
kilometer domain with 250-meter grid

00:22:24.240 --> 00:22:28.400
spacing, so relatively coarse.
And that’s because I’m only looking

00:22:28.400 --> 00:22:33.880
at frequencies between 10 and
30 seconds. I’m simulating

00:22:33.880 --> 00:22:38.700
80 seconds of wave propagation.
And I’m using a 3D velocity structure

00:22:38.700 --> 00:22:40.920
from SCEC CVM-H.

00:22:41.160 --> 00:22:43.600
And these – and so that’s
the color scale here.

00:22:43.600 --> 00:22:46.790
So you can see – and this is
the shear wave velocity

00:22:46.790 --> 00:22:51.260
at 2 kilometers’ depth
from the SCEC CVM.

00:22:51.260 --> 00:22:56.130
I’m not using any attenuation here,
which I think is relatively valid

00:22:56.130 --> 00:22:59.639
at this frequency band.
And the kinematic rupture model

00:22:59.640 --> 00:23:04.820
I’m going to use I developed
from Gavin Hayes’ solution

00:23:04.820 --> 00:23:08.259
and kind of modified
a little bit through trial and error.

00:23:08.259 --> 00:23:12.710
So here is a picture of
my fault slip model.

00:23:12.710 --> 00:23:18.660
So here’s the Hayes solution that
I pulled right off of the USGS event page.

00:23:18.660 --> 00:23:23.210
And, through some trial and error,
I kind of found slip patches that

00:23:23.210 --> 00:23:26.230
needed to be added to
match the GPS displacements.

00:23:26.230 --> 00:23:31.649
So the black arrows here are the
observed displacements from the

00:23:31.649 --> 00:23:36.330
coseismic rupture from the 7.1,
and the green arrows are my

00:23:36.330 --> 00:23:41.090
final displacements based on
the finite difference simulation.

00:23:41.090 --> 00:23:47.169
And so, yeah, I’d say I, more or less,
modified this by just amplifying

00:23:47.169 --> 00:23:51.260
different patches of this to find which
would make the arrows most hit the –

00:23:51.260 --> 00:23:55.340
green arrows to the black arrows.
And so the main feature here is that

00:23:55.340 --> 00:23:59.600
we have a pretty large slip patch
near the hypocenter, which has

00:23:59.600 --> 00:24:03.360
been well-documented now
by a number of authors.

00:24:04.780 --> 00:24:07.000
Another very interesting
part of this earthquake

00:24:07.000 --> 00:24:10.280
is the time evolution
of the rupture.

00:24:10.280 --> 00:24:17.400
So, through the first couple models
that I ran, I very quickly found that,

00:24:17.409 --> 00:24:20.669
if the whole – if the whole fault
ruptures at once, it’s very hard

00:24:20.669 --> 00:24:24.370
to match the waveforms.
And if you look at the coseismic

00:24:24.370 --> 00:24:28.480
GPS displacement at, you know,
one of these stations that’s – this is

00:24:28.480 --> 00:24:33.940
station CCCC, you can see that there’s
these different – what I would think

00:24:33.940 --> 00:24:37.920
are maybe multiple subevents.
So you have the initial big event,

00:24:37.929 --> 00:24:40.370
but then you have other little
ruptures that might be happening.

00:24:40.370 --> 00:24:44.260
So what I did was I broke the fault
into four different subevents.

00:24:44.260 --> 00:24:48.330
So I have Rupture A, Rupture B,
Rupture C, and Rupture D.

00:24:48.330 --> 00:24:50.950
So the initial rupture
is Rupture A.

00:24:50.950 --> 00:24:53.410
This is the hypocenter,
and it ruptures toward the north.

00:24:53.410 --> 00:24:57.780
This color is the rupture
time of each sub-fault.

00:24:57.780 --> 00:25:01.929
And this – so this ruptures at –
initiates at zero seconds.

00:25:01.929 --> 00:25:08.780
This is a magnitude 6.69 rupture.
Ruptures at 3 kilometers per second.

00:25:08.780 --> 00:25:13.840
And unilaterally to the north.
Then B is – has that

00:25:13.840 --> 00:25:18.149
big slip patch that I showed.
This is a 6.7 rupture.

00:25:18.149 --> 00:25:23.000
It only takes 3 seconds after the
initial rupture, but it ruptures slowly

00:25:23.000 --> 00:25:27.419
at 2 kilometers per second.
Rupture C is pretty small – 10-second

00:25:27.419 --> 00:25:32.629
delay time, 2 kilometers per second,
and ruptures unilaterally to the south.

00:25:32.629 --> 00:25:39.139
And then Rupture D ruptures at
2.5 kilometers per second and

00:25:39.140 --> 00:25:43.700
starts after 15 seconds, and it’s –
this rupture is bilateral,

00:25:43.700 --> 00:25:47.120
but it’s mostly toward the south.
So here’s – this is going to be

00:25:47.129 --> 00:25:52.460
a little animation of
the rupture process.

00:25:52.460 --> 00:25:55.950
So this is the fault, and it’s
going to show the slip rate.

00:25:55.950 --> 00:25:58.870
It’s going to kind of light up when it
shows which part of the fault is

00:25:58.870 --> 00:26:01.260
rupturing, and then over here,
we’ll see the ground motions.

00:26:01.260 --> 00:26:06.379
And I just kind of drew the fault
traces on here for reference.

00:26:06.379 --> 00:26:09.600
So the first part of the
rupture is Rupture A.

00:26:09.600 --> 00:26:13.540
This is going towards the north,
and you see it shoots out this

00:26:13.540 --> 00:26:16.679
wave pack here.
And we had Rupture B.

00:26:16.679 --> 00:26:21.340
It’s going to fast for me to explain.
[laughter] I’ll show it again.

00:26:21.340 --> 00:26:24.020
So we have Rupture A, Rupture B,
Rupture C, and Rupture D

00:26:24.020 --> 00:26:28.280
as this kind of higher-frequency
[inaudible]. So let’s start over.

00:26:34.120 --> 00:26:38.060
So Rupture A goes to the north,
then Rupture B suddenly starts up at

00:26:38.060 --> 00:26:44.040
3 seconds and has this big motion here.
Rupture C is pretty small and has

00:26:44.049 --> 00:26:47.290
this little boop of ground motion.
And then Rupture D ruptures

00:26:47.290 --> 00:26:53.440
bilaterally and has this sharp wave
field of higher frequency motion.

00:26:57.580 --> 00:27:01.539
This slide I kind of put here in case
my movie didn’t work, but this is

00:27:01.539 --> 00:27:06.760
another nice look at what I just showed.
So, at about 10-1/2 seconds, you see

00:27:06.760 --> 00:27:10.019
Rupture A is almost finished.
Rupture B is actually also

00:27:10.019 --> 00:27:13.830
almost finished. You have quite
a bit of ground motion going.

00:27:13.830 --> 00:27:15.950
Rupture C was small.
It was this little part here.

00:27:15.950 --> 00:27:22.240
And it just shot out a little bit of
motion compared to the rest.

00:27:22.240 --> 00:27:27.400
Here’s the wave front of Rupture D,
and you see it’s starting to show

00:27:27.409 --> 00:27:30.960
the ground motion part.
And then here is after the fault

00:27:30.960 --> 00:27:34.590
has finished rupturing completely
that you see this strong motion

00:27:34.590 --> 00:27:38.980
coming from the last
southernmost rupture.

00:27:38.980 --> 00:27:41.960
Just to look at some of the time series,
I wanted to break them in and show

00:27:41.960 --> 00:27:45.610
you the relative contribution
from each one of the subevents.

00:27:45.610 --> 00:27:52.140
So this is at the GPS station cccc.
This is at p463.

00:27:52.140 --> 00:27:57.470
And you can see, as I showed earlier,
these very distinct subevents

00:27:57.470 --> 00:28:00.580
come from these different –
Subevent A, B, C, and D.

00:28:00.580 --> 00:28:05.700
So, from Subevent A, we have
very little displacement here,

00:28:05.700 --> 00:28:09.899
but Subevent B is much bigger.
Then this spike here comes from

00:28:09.900 --> 00:28:14.940
part C, and then Subevent D
has kind of the final lobe there.

00:28:14.940 --> 00:28:17.600
And then similar
story on this side.

00:28:20.440 --> 00:28:24.600
Okay. And so now I’m just going to
kind of go around the whole domain

00:28:24.610 --> 00:28:27.059
and show different
parts of the wavefield.

00:28:27.059 --> 00:28:30.499
So this is showing kind of
a move-out towards the north.

00:28:30.500 --> 00:28:34.520
So most of these stations up here
are mainly affected by Rupture A,

00:28:34.520 --> 00:28:37.019
which ruptured
towards the north.

00:28:37.019 --> 00:28:42.210
The red curve – well, the black curves
are all the data, but the red curve is

00:28:42.210 --> 00:28:45.000
my synthetic strong motion,
and the blue curve here is one

00:28:45.000 --> 00:28:49.090
of these PBO strainmeters.
And so you can see the first arrival,

00:28:49.090 --> 00:28:51.960
at least in these north stations,
look pretty good.

00:28:51.960 --> 00:28:57.240
And some of these coda waves
are coming from the southern –

00:28:57.240 --> 00:29:00.180
the more southern parts of the rupture.
And you can see that we’re able to

00:29:00.180 --> 00:29:03.220
match the strain as well,
which is not typically done

00:29:03.220 --> 00:29:05.920
in these ground
motion simulations.

00:29:07.440 --> 00:29:12.080
Further off to the west,
the first arrivals look very nice and

00:29:12.090 --> 00:29:15.610
are dominated by that Rupture A.
But Rupture B comes in

00:29:15.610 --> 00:29:19.169
a little bit too strong.
You can see the sharp spike here.

00:29:19.169 --> 00:29:22.450
So something’s going on there.

00:29:22.450 --> 00:29:26.320
And later arrivals are coming
from the southernmost ruptures.

00:29:29.140 --> 00:29:32.840
Going down this way, we see, again,
the first arrivals look nice, but you

00:29:32.840 --> 00:29:35.419
see these big motions
that are coming in later.

00:29:35.419 --> 00:29:40.070
And so I would attribute that either to –
I would likely think it’s most likely

00:29:40.070 --> 00:29:44.130
attributed to some kind of basin effect
going on in this low-velocity area.

00:29:44.130 --> 00:29:48.100
But it could be possible that
there’s a re-rupture at some

00:29:48.100 --> 00:29:51.640
other part of the fault up
here that I’m not capturing.

00:29:53.020 --> 00:29:58.580
In this line further south, you see
that it’s mostly dominated by

00:29:58.580 --> 00:30:02.140
Ruptures B, C, and D, which
ruptured towards the south with

00:30:02.150 --> 00:30:07.270
Rupture A giving a little energy
because it’s directed off to the north.

00:30:07.270 --> 00:30:10.120
But you can very clearly
separate the beginning phases

00:30:10.120 --> 00:30:12.020
at these closer stations.

00:30:12.020 --> 00:30:16.340
And then, off further out here, you start
to see that it gets a little bit jumbled.

00:30:16.340 --> 00:30:20.080
But we still see that the
first arrivals look decent.

00:30:23.820 --> 00:30:26.940
To the south, this is where it gets a little
more complicated, because I had those

00:30:26.940 --> 00:30:29.680
three different ruptures
all directed down this way.

00:30:29.690 --> 00:30:32.029
And so you have a lot
of interference going on.

00:30:32.029 --> 00:30:34.679
But we see the main features still.
We see the closest stations.

00:30:34.679 --> 00:30:38.529
We see the first arrivals match.
We see these very sharp spikes

00:30:38.529 --> 00:30:43.419
coming from Rupture D in both
the strainmeters and in the

00:30:43.419 --> 00:30:46.679
strong motion stations and
also in this GPS station here.

00:30:46.679 --> 00:30:51.120
And if – and even farther out,
you start to see that these synthetic

00:30:51.120 --> 00:30:54.940
amplitudes are much larger.
So there might be some kind of

00:30:54.940 --> 00:31:00.879
attenuation or dissipation that I’m
not capturing with the elastic model.

00:31:02.840 --> 00:31:07.640
And then finally, off to the east here,
we see – actually, there’s a very clean –

00:31:07.640 --> 00:31:11.820
you can very cleanly pick out
the separation between the

00:31:11.820 --> 00:31:16.240
different subevents. This is one
of the stations I showed earlier.

00:31:16.240 --> 00:31:20.010
But it’s very nice that you can –
that we can make use of the

00:31:20.010 --> 00:31:25.140
GPS stations here in addition
to those ground motion stations.

00:31:26.500 --> 00:31:28.279
Definitely helps on this
side where there’s

00:31:28.280 --> 00:31:31.840
very low coverage of
the accelerometers.

00:31:33.380 --> 00:31:38.779
Just to look a little bit at the amplitudes.
This is the synthetic PGV – the log

00:31:38.779 --> 00:31:43.059
of the synthetic PGV on this axis.
This is the log of the observed PGV.

00:31:43.059 --> 00:31:47.490
And we see that it – on the dashed
line here is a – is a 1-to-1 line,

00:31:47.490 --> 00:31:51.629
so we see that it at least
follows a decent trend.

00:31:51.629 --> 00:31:54.740
Some of the stations out here,
though, we are overestimating

00:31:54.740 --> 00:31:57.870
the synthetic PGVs.
And those are the stations

00:31:57.870 --> 00:32:01.879
down here that I mentioned.
So maybe there could be some

00:32:01.880 --> 00:32:05.880
effects of attenuation down here, or
there could be scattering attenuation.

00:32:05.880 --> 00:32:09.560
So that could be a topic
for a future model.

00:32:09.560 --> 00:32:12.320
And then up here, again,
you see that the observed motions

00:32:12.320 --> 00:32:16.700
are much larger than the synthetics.
And so these stations here are these

00:32:16.700 --> 00:32:20.500
stations in these low-velocity areas
where it’s possible that we’re

00:32:20.500 --> 00:32:24.500
kind of underestimating
the [inaudible].

00:32:26.620 --> 00:32:31.539
Okay, and so, in summary,
the magnitude 7.1 Ridgecrest

00:32:31.540 --> 00:32:35.060
earthquake was well-recorded by
many different types of data,

00:32:35.060 --> 00:32:39.940
and we’re able to match them with a
simple elastic wave propagation code,

00:32:39.940 --> 00:32:43.420
at least in – at these
relatively low frequencies.

00:32:44.860 --> 00:32:49.560
These models are best fit by
a roughly 50-kilometer rupture

00:32:49.560 --> 00:32:54.620
with most slip concentrated
by the hypocenter.

00:32:54.620 --> 00:32:59.429
But the time series are best fit if
we split the rupture into four

00:32:59.429 --> 00:33:03.289
different subevents in this case.
The first subevent nucleated to –

00:33:03.289 --> 00:33:06.340
nucleated and ruptured to the north,
and the subsequent events

00:33:06.340 --> 00:33:09.300
all cascaded towards the south.

00:33:09.760 --> 00:33:15.140
And the synthetic PGVs increased
relatively well with the observed,

00:33:15.140 --> 00:33:20.040
however future models might want
to include more physics and things

00:33:20.040 --> 00:33:23.850
that could break up
these sharp wave fields.

00:33:25.680 --> 00:33:26.840
Thank you.

00:33:26.840 --> 00:33:30.860
[Applause]

00:33:30.860 --> 00:33:38.020
- [inaudible] questions if
anyone has one? [inaudible]

00:33:38.020 --> 00:33:41.640
- Hey, [inaudible] never went back to
the data [inaudible] magnitudes.

00:33:41.640 --> 00:33:43.980
And I don’t know which magnitude
you meant or whether those

00:33:43.980 --> 00:33:48.560
magnitudes [inaudible] are related …
- Even the real-time – the real-time …

00:33:48.560 --> 00:33:52.920
- The real-time [inaudible].
There was multiple

00:33:52.920 --> 00:33:54.660
[inaudible] magnitudes.
- Yeah.

00:33:54.660 --> 00:33:57.060
- But you never go back to it.
[inaudible] it or …

00:33:57.060 --> 00:34:01.940
- Okay. That’s a good – okay.
- [inaudible]

00:34:01.940 --> 00:34:06.180
- I meant to have a talk around
the office about how the first

00:34:06.180 --> 00:34:09.260
[inaudible] something,
but maybe I should …

00:34:09.260 --> 00:34:11.540
- Was that [inaudible] or was it …
- Yeah. The [inaudible].

00:34:11.540 --> 00:34:14.000
- Okay. And [inaudible] …
- So maybe I should talk to you guys …

00:34:14.010 --> 00:34:16.600
- [inaudible]
- I don’t know that much about those,

00:34:16.600 --> 00:34:21.760
so yeah. I’d like to chat with someone
after this to either pull that out or …

00:34:21.760 --> 00:34:25.659
- [inaudible]
- It’s a complicated [inaudible].

00:34:25.660 --> 00:34:28.720
- Okay.
- [inaudible] a network that’s …

00:34:28.720 --> 00:34:32.860
- So that’s not a physical – okay.
- [inaudible]

00:34:32.860 --> 00:34:36.960
- [inaudible] do a
postmortem on [inaudible].

00:34:36.960 --> 00:34:39.500
- Yeah. Jeff was the one I talked to,
but I haven’t talked to you

00:34:39.500 --> 00:34:43.360
since a few months ago, so …
[laughter]

00:34:43.360 --> 00:34:47.800
- Did we [inaudible]?
- I really like how you match

00:34:47.800 --> 00:34:54.960
all the different [inaudible] data.
And then also that you used [inaudible].

00:34:54.960 --> 00:34:59.280
So do you think, in places where
there’s not quite a match between

00:34:59.280 --> 00:35:03.140
the [inaudible] and the synthetic,
[inaudible] part of the velocity model?

00:35:03.140 --> 00:35:08.120
Or do you think it’s more …
- I think so, yeah.

00:35:08.120 --> 00:35:13.400
Actually, there was some of these –
some of these very distant stations

00:35:13.400 --> 00:35:16.710
out here actually looked better
if I just used the 1D velocity model.

00:35:16.710 --> 00:35:23.030
I didn’t show that, but overall, I think,
when I look at the whole domain,

00:35:23.030 --> 00:35:25.941
the 3D velocity model
looked the best.

00:35:25.941 --> 00:35:29.260
But there’s definitely
other velocity models to use.

00:35:29.260 --> 00:35:33.260
There’s also the CDMS model.
- So you think that they need

00:35:33.260 --> 00:35:37.600
[inaudible] to update the velocity
model based on your findings?

00:35:37.600 --> 00:35:44.180
Or do you think that [inaudible] …
- I wouldn’t go as bold as to say that.

00:35:44.180 --> 00:35:50.980
I’d rather do something like I think
Rob Graves did, where he simulated

00:35:50.980 --> 00:35:53.940
a lot of the small aftershocks.
So we get rid of all the source

00:35:53.940 --> 00:35:59.100
effects and just see [inaudible]
of each and [inaudible].

00:35:59.100 --> 00:36:01.800
Yeah.
- Okay. Thanks.

00:36:06.580 --> 00:36:13.400
- Okay. Our next speaker will be
Jean Soubestre from INVOLCAN.

00:36:15.640 --> 00:36:29.100
[Silence]

00:36:29.100 --> 00:36:32.020
- Okay. Thank you very much
for giving me this opportunity

00:36:32.020 --> 00:36:33.760
to present this work.

00:36:33.760 --> 00:36:38.160
So I will tell you about a
matter that we call CovNet,

00:36:38.160 --> 00:36:40.620
for Covariance Network.

00:36:40.620 --> 00:36:43.580
It is a method for the
automatic detection and

00:36:43.580 --> 00:36:46.560
location of the
seismo-volcanic signals.

00:36:46.560 --> 00:36:50.190
I’ve been working on this
during my present postdoc

00:36:50.190 --> 00:36:54.070
in Paris with Nikolai Shapiro
and Leonard Seydoux.

00:36:54.070 --> 00:36:58.970
And now I continue to work on
this during my current postdoc

00:36:58.970 --> 00:37:04.280
in Canary Island at the volcano
observatory and trying to do

00:37:04.280 --> 00:37:09.980
collaborations with many people.
And that’s why I am here this week

00:37:09.980 --> 00:37:15.840
to work with Phillip Dawson to try
to apply this network on this date.

00:37:15.840 --> 00:37:20.020
So this is a method designed
for volcanic monitoring.

00:37:20.020 --> 00:37:25.280
So volcanic monitoring – new
methods face different challenges.

00:37:25.280 --> 00:37:32.120
The first one is the fact to be able to
be adapted to the different kind of

00:37:32.120 --> 00:37:35.610
signals that we have on volcanoes
from volcano-tectonic earthquakes

00:37:35.610 --> 00:37:42.090
to the long-period seismicity.
Another challenge is to be –

00:37:42.090 --> 00:37:46.980
for this method is to be adapted
to very low energy events

00:37:46.980 --> 00:37:50.420
that cannot be picked
or detected manually.

00:37:50.420 --> 00:37:55.980
And, with the continuous growing
monitoring networks that we have,

00:37:55.980 --> 00:38:00.760
it’s important for new method to be
able to deal with large amount of data

00:38:00.760 --> 00:38:07.500
and, if possible, to do it automatically.
So the method we developed

00:38:07.500 --> 00:38:11.820
[inaudible] was –
all those requirements.

00:38:11.820 --> 00:38:18.900
So I will present you the method
first and then I will focus on results

00:38:18.900 --> 00:38:25.530
on four different volcanic contexts.
The one cluster of volcanoes in

00:38:25.530 --> 00:38:31.080
Kamchatka. The Teide Volcano in
Canary Island. The French Mayotte

00:38:31.080 --> 00:38:36.780
Island – coast of Madagascar.
And the Mammoth Mountain volcano.

00:38:38.420 --> 00:38:45.120
Okay, so first, we – so this is a
network method, so first we get the –

00:38:45.120 --> 00:38:48.870
we only use the vertical component.
And we filter and downsample

00:38:48.870 --> 00:38:54.360
our data. And then we divided –
we divide the data into windows.

00:38:54.360 --> 00:38:58.460
So here you have an example
of the blue one and a red one.

00:38:58.460 --> 00:39:02.080
And, on those window,
we apply a spectral normalization.

00:39:03.120 --> 00:39:09.200
Then we divide – so each normalized
window is divided into sub-windows.

00:39:09.200 --> 00:39:11.690
So here you have the example.

00:39:11.690 --> 00:39:15.680
And we calculate the cross-spectra
matrices between all the different

00:39:15.680 --> 00:39:20.040
stations in Paris of our network.
And we obtain finally the

00:39:20.040 --> 00:39:26.020
covariance matrix of the average
of those cross-spectra matrices.

00:39:26.020 --> 00:39:29.010
So it’s the covariance –
we call it the covariance matrix.

00:39:29.010 --> 00:39:32.640
It’s in the frequency domain.
But it’s the equivalent in the

00:39:32.640 --> 00:39:36.340
frequency domain of the cross-
correlation in the [inaudible].

00:39:36.340 --> 00:39:42.940
And the interesting thing about
this covariance matrix is that,

00:39:42.950 --> 00:39:46.110
being Hermitian, it can be
decomposed on the basis of

00:39:46.110 --> 00:39:52.600
real positive eigenvalues associated
with the complex eigenvectors.

00:39:52.600 --> 00:39:57.910
And we use the eigenvalues to detect
sources. And then, once we have

00:39:57.910 --> 00:40:03.160
the detected sources, we can use the
eigenvectors to clusterize them or to

00:40:03.160 --> 00:40:10.550
locate them. Today we’ll just tell
you about the detection and location.

00:40:10.550 --> 00:40:16.340
So how we – how we detect sources,
here it’s an example – you have

00:40:16.340 --> 00:40:21.130
different seismic traces at the top.
And you have a period of noise in

00:40:21.130 --> 00:40:25.300
blue and a period with
an earthquake in red.

00:40:25.300 --> 00:40:29.180
And, at the bottom, you see
the distribution of the eigenvalues

00:40:29.180 --> 00:40:32.530
of the covariance matrix.
And you see that, when you have

00:40:32.530 --> 00:40:38.600
a period of months, the distribution
is more spread than when you

00:40:38.600 --> 00:40:41.280
have a dominating source.

00:40:41.280 --> 00:40:46.440
So we defined a scalar parameter
that we call the spectral width.

00:40:46.450 --> 00:40:51.240
And that measure the width of the
distribution of the eigenvalues.

00:40:51.240 --> 00:40:53.840
And we use it as a proxy
of the number of

00:40:53.840 --> 00:40:57.810
dominating sources
in our wavefield.

00:40:57.810 --> 00:41:05.240
And once we have detected the source,
we can locate it using the eigenvectors.

00:41:05.240 --> 00:41:11.100
Our main hypothesis is that we do not
use the full – the full covariance matrix,

00:41:11.100 --> 00:41:16.560
but we rebuild one from its first
eigenvector, with the hypothesis that

00:41:16.560 --> 00:41:21.400
the first eigenvector contains
the information about the

00:41:21.410 --> 00:41:25.600
dominant source in the wavefield.
So we rebuild a covariance

00:41:25.600 --> 00:41:29.610
matrix from the first eigenvector.
And then we go back in the time

00:41:29.610 --> 00:41:34.500
domain by an inverse [inaudible],
and we do a cross-correlation

00:41:34.500 --> 00:41:39.760
based the location record.
So we do it with a 1D velocity model

00:41:39.760 --> 00:41:43.680
in order to be able to
have sources at depth.

00:41:43.680 --> 00:41:48.580
And then we perform a 3D grid search.
So for each point of the grid, we can

00:41:48.580 --> 00:41:52.251
calculate the travel times and the
different travel times, and so we

00:41:52.251 --> 00:41:57.500
can shift our cross-correlations.
And if we – then, if we stack the

00:41:57.500 --> 00:42:02.950
envelope of the cross-correlation at
each point – each point of the 3D grid,

00:42:02.950 --> 00:42:09.760
the point where the stack is maximum
is the location of the source.

00:42:14.020 --> 00:42:18.270
So let’s see results first on the –
on the Klyuchevskoy volcanic group

00:42:18.270 --> 00:42:23.520
in Kamchatka. So it’s in this
peninsula at the east tip of Russia.

00:42:25.040 --> 00:42:28.510
This is the largest cluster of
subduction volcanoes in the

00:42:28.510 --> 00:42:34.660
world with a lot of – a lot of
composition and eruption styles.

00:42:35.380 --> 00:42:40.220
And this is the network we are
using to monitor those volcanoes.

00:42:40.220 --> 00:42:45.120
So you have the five volcanoes –
Kizimen, Tolbachik –

00:42:45.120 --> 00:42:51.680
those are the white triangles.
And we have 19 seismic stations.

00:42:53.300 --> 00:42:56.280
And this is the results of
the calculation of the

00:42:56.290 --> 00:43:00.120
[inaudible] for
4-1/2 years of data.

00:43:00.120 --> 00:43:02.670
So this is a graph in
frequency and time domain.

00:43:02.670 --> 00:43:06.480
This is not a spectrogram.
This is – each pick cell of this

00:43:06.480 --> 00:43:11.200
graph represent the width
of the distribution of the eigenvalues

00:43:11.200 --> 00:43:14.990
of the covariance matrix at
that time and that frequency.

00:43:14.990 --> 00:43:17.750
And when you have a warm color,
it means that you have a dominating

00:43:17.750 --> 00:43:21.780
source, and a cold color,
you have noise.

00:43:21.780 --> 00:43:27.720
So we detect many different
kind of signals in those graphs.

00:43:27.720 --> 00:43:34.680
But I would just focus on one.
Those two very long signals in red

00:43:34.690 --> 00:43:39.510
are tremors. So we know from a priori
knowledge that there are tremors from

00:43:39.510 --> 00:43:44.020
two different volcanoes that are
Klyuchevskoy and Tolbachik.

00:43:44.020 --> 00:43:49.220
But for sure, with just this analysis,
using a scalar parameter, we cannot

00:43:49.220 --> 00:43:51.880
distinguish if they are
from two different source.

00:43:51.880 --> 00:43:55.300
So that’s why we
used the eigenvectors.

00:43:55.300 --> 00:44:03.020
And what we did is that we did a
location for every day of this data set.

00:44:04.280 --> 00:44:11.440
So, when you have a day with noise,
you end up with a unpure location.

00:44:11.440 --> 00:44:15.220
But you have some day when
you have an interesting location.

00:44:15.220 --> 00:44:19.670
So this is an example.
In 2009, we have – the tremor

00:44:19.670 --> 00:44:23.140
is located below the
Shiveluch Volcano at the north.

00:44:25.000 --> 00:44:32.000
If we go in time, and just to the
beginning of – the end of 2009,

00:44:32.010 --> 00:44:37.170
you have a quite deep source of
tremor below the Klyuchevskoy.

00:44:37.170 --> 00:44:44.940
And then, if you – if you continue
in time and go at the end of –

00:44:44.940 --> 00:44:48.430
the tremor at the Klyuchevskoy
was about one year.

00:44:48.430 --> 00:44:53.700
And if you go in time, you see that the
source of the tremor is more shallow.

00:44:53.700 --> 00:44:58.450
This was – this was at the beginning of
the – of the tremor. It was quite deep.

00:44:58.450 --> 00:45:03.600
And then it goes very shallow
when you go close to the eruption.

00:45:03.600 --> 00:45:07.480
And, yeah, we did it every day.
So we also located a tremor in

00:45:07.480 --> 00:45:13.520
the Kizimen. And finally, at the end,
it’s a tremor at the other volcano.

00:45:15.260 --> 00:45:20.440
And this is the same thing but
representing the – for the third –

00:45:20.440 --> 00:45:22.240
for four different volcanoes.

00:45:22.240 --> 00:45:26.040
You have the depths of the
source as a function of time.

00:45:26.040 --> 00:45:30.600
And if you look at the second one
from the top, it’s the Klyuchevskoy.

00:45:30.600 --> 00:45:35.450
And you see that we can track
the vertical migration of the –

00:45:35.450 --> 00:45:40.380
of the tremor source, and it was –
ended up very shallow close to

00:45:40.380 --> 00:45:44.380
the eruption that
was the red part.

00:45:46.720 --> 00:45:51.060
Let’s skip to another
data set that are different.

00:45:51.060 --> 00:45:55.180
So the Teide Volcano
is in Canary Island.

00:45:55.180 --> 00:45:58.660
It’s an intraplate volcano,
so very different,

00:45:58.670 --> 00:46:03.640
with also different composition
and eruption styles.

00:46:03.640 --> 00:46:08.330
This is the network we
used to monitor it.

00:46:08.330 --> 00:46:12.980
We have about
12 stations – broadband.

00:46:14.270 --> 00:46:20.320
And here you have the result of the
spectral reach for about two weeks.

00:46:20.320 --> 00:46:24.810
So here, again, we detect very –
a lot of different sources.

00:46:24.810 --> 00:46:28.770
Being on an island, for sure,
we detect the oceanic sources –

00:46:28.770 --> 00:46:33.710
so the primary microseism
and the secondary one.

00:46:33.710 --> 00:46:38.630
But if we focus on the volcanic
source, just an example of a LP event.

00:46:38.630 --> 00:46:42.570
So here, you cannot see the –
if you zoom at that time,

00:46:42.570 --> 00:46:47.210
you have a an LP event –
very small one.

00:46:47.210 --> 00:46:51.720
This event was not in our
catalog because it’s impossible

00:46:51.720 --> 00:46:55.570
to detect it manually.
So here, this is the detection

00:46:55.570 --> 00:46:58.160
applying a spectral
widening that I show you.

00:46:58.160 --> 00:47:03.420
If you apply no pre-processing,
you cannot detect this event.

00:47:04.240 --> 00:47:06.540
And this is with a
one-bit normalization.

00:47:06.540 --> 00:47:08.380
You never
detect the event.

00:47:08.380 --> 00:47:12.850
So you clearly need
a way to enhance

00:47:12.850 --> 00:47:15.940
the signal in all
these frequency.

00:47:17.380 --> 00:47:22.610
And I performed the location of this
event, so it’s about 10 kilometer

00:47:22.610 --> 00:47:29.040
below the cone of the data.
And it’s quite coherent with

00:47:29.040 --> 00:47:32.640
another location that we
have from this [inaudible].

00:47:32.640 --> 00:47:38.800
So it’s quite interesting because the
method is able to locate those LPs.

00:47:41.760 --> 00:47:44.640
Other volcanic context.
The Mayotte Island

00:47:44.640 --> 00:47:50.160
in the French Comores.
So between Madagascar and Africa.

00:47:52.400 --> 00:48:00.660
I’m sure you know that there is a new –
a newborn volcano about 50 kilometers

00:48:00.670 --> 00:48:06.230
east of Mayotte that was revealed
with bathymetry campaign in 2019.

00:48:06.230 --> 00:48:11.950
It’s about 800 meters high
with 5 cubic kilometer of magma

00:48:11.950 --> 00:48:15.840
in less than one year.
It’s a monster.

00:48:15.840 --> 00:48:25.190
And we had a very – a very highly
energetic VLP – very long period

00:48:25.190 --> 00:48:32.080
event on November 11, 2018,
that was recorded all over the world.

00:48:32.080 --> 00:48:38.420
And we did the analysis of the data
during one year between April 2018

00:48:38.420 --> 00:48:45.380
and May 2019. And, in reality,
we detect a lot of self-similar VLPs.

00:48:46.620 --> 00:48:53.700
This is the data I used –
regional network from IRIS.

00:48:55.820 --> 00:49:02.520
Here you have the VLP signal
of the – of the 11 of November.

00:49:02.580 --> 00:49:06.320
As you can see,
it’s quite monochromatic.

00:49:07.480 --> 00:49:15.080
And this is the detection for – so,
for this day, the VLP detected here.

00:49:15.080 --> 00:49:21.290
So we have large detection windows
because this network is very big,

00:49:21.290 --> 00:49:27.160
but we detect clearly the event.
And we can locate it with the –

00:49:27.160 --> 00:49:32.420
with the method.
So it’s – we obtained a first location,

00:49:32.420 --> 00:49:36.600
let’s say, that is east of –
east of Mayotte.

00:49:38.440 --> 00:49:44.320
And then I did the exercise
using one year of data to try to

00:49:44.320 --> 00:49:48.140
see if we detect
self-similar events.

00:49:48.140 --> 00:49:49.810
So here you have the
stakeholders at the top.

00:49:49.810 --> 00:49:54.340
And, at the bottom, you have three
stacks in different frequency bands.

00:49:54.340 --> 00:49:58.090
And this interesting one are
the blue one at the bottom

00:49:58.090 --> 00:50:06.600
is between 0.02 and 0.08.
And the green one is centered around

00:50:06.600 --> 00:50:13.820
the – from [inaudible] frequency
of the VLP that is 0.065 hertz.

00:50:13.830 --> 00:50:17.110
So we can – if we do the
difference between the green

00:50:17.110 --> 00:50:22.170
and the – and the blue line, we can –
we can have a clear detection of this –

00:50:22.170 --> 00:50:25.950
of those self-similar VLPs.

00:50:25.950 --> 00:50:31.760
And just have a look – a zoom at the
beginning of the – of the sequence.

00:50:31.760 --> 00:50:36.780
So we – at the beginning, we had
a seismic crisis with a lot of

00:50:36.780 --> 00:50:43.100
volcano-tectonic earthquakes.
And if we zoom later in November,

00:50:43.100 --> 00:50:48.900
we have this kind of things where you
can see that we have all those VLP.

00:50:48.900 --> 00:50:56.480
Very monochromatic. With only
[inaudible] at the 0.065 hertz.

00:50:58.780 --> 00:51:03.100
So we are working on
this VLP and its repeaters.

00:51:03.100 --> 00:51:08.540
Here you have just the evolution of
the fundamental frequency of this

00:51:08.540 --> 00:51:13.460
self-similar events. And you see
it’s decreasing at the beginning.

00:51:13.460 --> 00:51:18.600
You have the major event
in November – the yellow circle.

00:51:18.600 --> 00:51:21.440
And then it increases.

00:51:22.740 --> 00:51:28.580
Okay, and just to finish, just one
result about Mammoth Mountain.

00:51:28.580 --> 00:51:33.700
So you know Mammoth
Mountain better than me.

00:51:34.910 --> 00:51:39.560
So we used – we’re using
two different network.

00:51:39.560 --> 00:51:41.630
The blue one is the
permanent network,

00:51:41.630 --> 00:51:47.100
and the orange was a
temporary network in 2013.

00:51:47.100 --> 00:51:52.890
And here you have the
results for 10 days of data.

00:51:52.890 --> 00:51:55.870
At the top, the spectral width,
and at the bottom, stacks in

00:51:55.870 --> 00:52:00.460
different frequency range.
But just – the interesting thing is the –

00:52:00.460 --> 00:52:03.850
if you compare – so this is the
result with the temporary network,

00:52:03.850 --> 00:52:06.530
and this is the result
with the permanent network.

00:52:06.530 --> 00:52:10.800
So the temporary network being
very close to the summit.

00:52:10.800 --> 00:52:16.180
Clearly see the chair lifts of
the ski station. [laughter]

00:52:16.180 --> 00:52:20.760
But it also see – quite interesting
[inaudible] here that we are

00:52:20.760 --> 00:52:26.930
interested in now with Phillip –
potential long period events

00:52:26.930 --> 00:52:30.300
or maybe very
small tremors.

00:52:31.200 --> 00:52:35.700
So we are working on
that with Phillip right now.

00:52:35.700 --> 00:52:40.090
Okay, so, yeah, I presented you
this method that is interesting

00:52:40.090 --> 00:52:46.850
for volcano monitoring.
This is running – currently running

00:52:46.850 --> 00:52:51.270
as a routine at the volcano
observatory in Canary Islands.

00:52:51.270 --> 00:52:54.050
And I’m trying to apply
the method on different data sets

00:52:54.050 --> 00:52:58.500
now to [inaudible]
other volcanic contexts.

00:52:58.500 --> 00:53:00.020
Thank you.

00:53:00.020 --> 00:53:04.060
[Applause]

00:53:04.060 --> 00:53:06.660
- [inaudible] one question?

00:53:11.840 --> 00:53:15.280
- I was – I’m actually interested
in sort of your noise sources.

00:53:15.280 --> 00:53:18.280
Have you tried looking at the
noise sources to see if there’s

00:53:18.280 --> 00:53:22.380
any variability in where
the noise is coming from?

00:53:22.380 --> 00:53:30.260
- Yeah. In [inaudible] in Canary Island,
I’m trying to understand where are

00:53:30.260 --> 00:53:36.520
the noise sources.
And doing other kind of analysis

00:53:36.520 --> 00:53:40.300
with polarization analysis and –
because we also do ambient

00:53:40.300 --> 00:53:43.030
noise tomography.
So we have to know where

00:53:43.030 --> 00:53:47.270
the noise sources are from.
But, for the monitoring point of view,

00:53:47.270 --> 00:53:51.160
I’m more interested in the
volcanic sources. [laughs]

00:53:52.820 --> 00:53:55.460
- Okay. Thank you.

00:53:57.260 --> 00:54:13.500
[Silence]

00:54:13.500 --> 00:54:17.260
Okay. And our last speaker
will be Sara McBride.

00:54:19.880 --> 00:54:22.800
- Good morning, everyone.
Thank you so much for inviting me

00:54:22.810 --> 00:54:25.730
to give this talk. Thanks,
Grace and Jesse, for setting it up.

00:54:25.730 --> 00:54:31.180
So this talk is less a research
talk as it is a talk about research.

00:54:31.180 --> 00:54:34.340
And what we’re doing here
at the USGS about social science

00:54:34.340 --> 00:54:35.470
and ShakeAlert.

00:54:35.470 --> 00:54:40.030
So recently, Elizabeth Cochran and
Allen Husker wrote a – sort of an op-ed

00:54:40.030 --> 00:54:44.320
in Science called How Low Should We
Go When Warning for Earthquakes?

00:54:44.320 --> 00:54:48.250
And one of the calls the paper made
was, we really need more social

00:54:48.250 --> 00:54:51.660
science to understand what people
want and expect from warnings.

00:54:51.660 --> 00:54:56.200
And, we at the USGS, we agree.
So about a year and a half ago,

00:54:56.200 --> 00:54:57.920
we started out on
a social science program,

00:54:57.920 --> 00:55:01.420
and I’m the social science
coordinator for ShakeAlert.

00:55:01.420 --> 00:55:05.020
So we’re trying to do two things
with our social science program.

00:55:05.020 --> 00:55:09.600
One is to develop an understanding
where populations are in Oregon,

00:55:09.600 --> 00:55:13.030
Washington, and California
in terms of – in terms of our

00:55:13.030 --> 00:55:16.280
understanding around earthquake
risk perception – protective action

00:55:16.280 --> 00:55:20.080
knowledge. So do they know to drop,
cover, and hold on, or what to do

00:55:20.080 --> 00:55:22.990
if they’re in their car and there’s an
earthquake? And basic earthquake

00:55:22.990 --> 00:55:27.280
preparedness across the three states
of where ShakeAlert operates.

00:55:27.280 --> 00:55:30.640
And we also want to establish
a monitoring evaluation plan

00:55:30.640 --> 00:55:32.640
for ShakeAlert.
And what that means is,

00:55:32.640 --> 00:55:36.170
we want to say – we want to see,
does ShakeAlert make any difference

00:55:36.170 --> 00:55:38.970
over time to what people
understand about earthquakes

00:55:38.970 --> 00:55:42.530
and what kind of protective
actions they take?

00:55:42.530 --> 00:55:46.150
So we actually have seven active
projects across nine universities,

00:55:46.150 --> 00:55:50.200
and that includes five different
states and two countries.

00:55:50.200 --> 00:55:53.880
We might be expanding that
further in the coming months.

00:55:53.880 --> 00:55:58.420
And here, the first one I’m going to
talk about is the most immediate

00:55:58.420 --> 00:56:02.060
project we’ve got going on, which is
the Ridgecrest earthquake sequence

00:56:02.060 --> 00:56:05.891
research, which is Responses to
the 2019 Ridgecrest Earthquake

00:56:05.891 --> 00:56:09.100
Sequence and ShakeAlert.
And this is being worked on by

00:56:09.100 --> 00:56:11.690
Michele Wood, who is
a longtime health communication

00:56:11.690 --> 00:56:15.870
researcher at Cal State-Fullerton,
and Jeannette Sutton.

00:56:15.870 --> 00:56:18.610
So Michele Wood is going to be
doing the quantitative component of it.

00:56:18.610 --> 00:56:21.530
So we’re doing a survey, which
means she’s going to be looking at

00:56:21.530 --> 00:56:24.400
the numbers which people express.
And Jeannette’s going to be

00:56:24.400 --> 00:56:26.260
the qualitative lead,
which means she’s going to be

00:56:26.260 --> 00:56:29.180
looking at the words,
which is the content.

00:56:29.180 --> 00:56:32.860
And really what we want to understand
in this purpose is to understand how

00:56:32.860 --> 00:56:37.800
the public responded in southern
California, particularly in the

00:56:37.800 --> 00:56:41.120
Ridgecrest area and outside
the Ridgecrest area with an

00:56:41.120 --> 00:56:45.020
explicit focus on what people
perceived about ShakeAlert.

00:56:45.020 --> 00:56:49.050
We understand from some of the
public feedback that people thought

00:56:49.050 --> 00:56:53.440
that ShakeAlert was a failure – at least
the ShakeAlertLA app was a failure.

00:56:53.440 --> 00:56:56.280
Well, what about it was
a failure, and how can we

00:56:56.280 --> 00:56:58.960
make improvements
going forward?

00:56:58.960 --> 00:57:02.040
So the study’s aim is really to
examine the perceptions

00:57:02.040 --> 00:57:05.830
and self-protective actions.
So if people felt the earthquake,

00:57:05.830 --> 00:57:09.330
what kind of actions they took.
As well as baseline knowledge

00:57:09.330 --> 00:57:14.150
and attitudes – so where they’re
at today in southern California.

00:57:14.150 --> 00:57:19.090
And then the public reactions to
ShakeAlertLA during this event.

00:57:19.090 --> 00:57:23.090
So we’re taking two approaches,
like I said, which is the main part of

00:57:23.090 --> 00:57:25.700
this is going to be the quantitative
research, which is going to be an

00:57:25.700 --> 00:57:30.450
online survey, which Cal State-
Fullerton is going to administrate.

00:57:30.450 --> 00:57:36.260
And the sample’s going to be obtained
by Qualtrics, which is a panel.

00:57:36.260 --> 00:57:39.320
Basically, they have different
panels that they access of people

00:57:39.320 --> 00:57:41.930
who volunteer to be on
these panels to be surveyed.

00:57:41.930 --> 00:57:46.230
And so the first geographic region is
going to be southern California, where

00:57:46.230 --> 00:57:51.060
we have a sample size of 1,000 people.
And then the second sample is going to

00:57:51.060 --> 00:57:57.180
be within the Ridgecrest area, which is
a sample size of 384 – thereabouts –

00:57:57.180 --> 00:58:00.920
because we want to see what the
different populations have to say.

00:58:00.920 --> 00:58:04.100
The reason why the Ridgecrest sample
is so much smaller is because the

00:58:04.100 --> 00:58:08.000
population in Ridgecrest is
so much smaller than in L.A.

00:58:08.000 --> 00:58:09.590
And then we also have some
open-ended questions, where is

00:58:09.590 --> 00:58:12.730
where the qualitative research –
the content analysis, looking at

00:58:12.730 --> 00:58:16.460
themes and meaning
and construction come in.

00:58:16.460 --> 00:58:20.910
So hopefully, through this process
that Jeannette and Michele are

00:58:20.910 --> 00:58:24.350
coming through for us, is a better
understanding of the cognitive and

00:58:24.350 --> 00:58:28.750
behavioral responses of earthquakes
and the immediate impact of alerts.

00:58:28.750 --> 00:58:30.830
And they’re going to be
doing some theory testing –

00:58:30.830 --> 00:58:34.330
some social science theory testing.
As well as, when they finish this,

00:58:34.330 --> 00:58:39.160
they’re going to make some conclusions
and suggestions at how to improve our

00:58:39.160 --> 00:58:42.620
understanding of policymaking and
organizational response to public

00:58:42.620 --> 00:58:45.500
perceptions of warning limitations,
particularly around things

00:58:45.500 --> 00:58:47.540
like the threshold.

00:58:48.840 --> 00:58:51.020
And then they’re going to be
doing a baseline for us so that

00:58:51.020 --> 00:58:53.780
we can track it over time.
They’re also going to be

00:58:53.780 --> 00:58:58.020
cross-training geoscience students.
What we’ve noted in the past few

00:58:58.030 --> 00:59:00.060
years is a lot of geoscience
students are becoming

00:59:00.060 --> 00:59:03.950
very interested in social science.
And they want to have opportunities

00:59:03.950 --> 00:59:07.030
to cross-train in social science
as well as geoscience.

00:59:07.030 --> 00:59:10.190
So we’re going to be
achieving that with this.

00:59:10.190 --> 00:59:13.620
So that’s that piece.
Now we have the baseline survey,

00:59:13.620 --> 00:59:17.770
which I’m a part of that research team.
I’m the lead of that team.

00:59:17.770 --> 00:59:21.200
And then we have Dr. Julia Becker
at Massey University and Ann Bostrom

00:59:21.200 --> 00:59:24.620
at University of Washington.
Ann Bostrom is sort of – I hate to

00:59:24.620 --> 00:59:27.860
use this term, but she’s kind of the
grandmother of risk communication.

00:59:27.870 --> 00:59:31.990
So we’re so lucky to have her
on this project to help us out.

00:59:31.990 --> 00:59:34.630
And really, the baseline survey
is for Washington and Oregon.

00:59:34.630 --> 00:59:38.940
So you know how we’re doing this
for southern California and

00:59:38.940 --> 00:59:41.580
a little bit of central California?
Well, this is going to look at

00:59:41.580 --> 00:59:45.790
Washington and Oregon in particular.
And there was a survey done in Japan,

00:59:45.790 --> 00:59:49.810
which was replicated in New Zealand,
which is now going to be modified and

00:59:49.810 --> 00:59:53.990
used in Washington state, so we can
do a three-country analysis of how –

00:59:53.990 --> 00:59:56.960
compare and contrast how these
three countries – the populations in

00:59:56.960 --> 00:59:59.860
these three countries understand
earthquake early warning.

00:59:59.860 --> 01:00:03.040
So first, we’re working on
the survey instrument design.

01:00:03.040 --> 01:00:05.270
Then we’re going to do survey
implementation in Washington

01:00:05.270 --> 01:00:07.700
and Oregon.
And then we’re going to

01:00:07.700 --> 01:00:11.920
have the first round of descriptive
reports in April and June.

01:00:11.920 --> 01:00:14.500
It’s a very aggressive timeline.

01:00:14.500 --> 01:00:17.760
Then we have – we’ve invited
an anthropologist to be a part of

01:00:17.760 --> 01:00:21.110
our team – Dr. Beth Reddy, who is
with Colorado School of Mines.

01:00:21.110 --> 01:00:25.520
And she is looking at how
the ShakeAlert community,

01:00:25.520 --> 01:00:29.550
both internally within the USGS and
externally with our university partners,

01:00:29.550 --> 01:00:33.740
perceive and respond and
understand how ShakeAlert works.

01:00:33.740 --> 01:00:37.660
What we’ve come to understand is,
each of us, as scientists and as

01:00:37.660 --> 01:00:40.400
individuals, kind of perceive
ShakeAlert and earthquake

01:00:40.400 --> 01:00:42.260
early warning a
little bit differently.

01:00:42.260 --> 01:00:46.360
And so, what Beth’s role is going to be
looking at the community of practice.

01:00:46.360 --> 01:00:52.440
So how we come together,
how we share ideas, and how

01:00:52.440 --> 01:00:56.710
we understand ShakeAlert in terms of
the different purposes of our

01:00:56.710 --> 01:00:59.820
perception as a – as a group.
And hopefully what she will come

01:00:59.820 --> 01:01:04.230
together with is to create a shared
understanding and language –

01:01:04.230 --> 01:01:07.890
how to talk about ShakeAlert as
a community internally within

01:01:07.890 --> 01:01:11.060
the USGS and with
our ShakeAlert partners.

01:01:11.920 --> 01:01:15.060
Beth also completed this
kind of work in Mexico City.

01:01:15.060 --> 01:01:19.800
So this is not her first go at it.
She also has worked really hard

01:01:19.800 --> 01:01:22.160
on the Mexico City side
as well and has published

01:01:22.160 --> 01:01:25.020
a couple of articles about that.

01:01:25.020 --> 01:01:29.190
So we also have earthquake early
warning in schools, which is being

01:01:29.190 --> 01:01:32.050
run by Lori Peek through
the National Hazards Center.

01:01:32.050 --> 01:01:35.120
And if anyone knows about Lori Peek,
she worked with Dennis Mileti,

01:01:35.120 --> 01:01:38.580
who some of you might know.
Dennis is sort of the founder of

01:01:38.580 --> 01:01:42.730
disaster sociology in a lot of ways.
And she is one of his students.

01:01:42.730 --> 01:01:46.000
And she focused on schools.
She is now the director of the

01:01:46.000 --> 01:01:49.420
Natural Hazards Center at
University of Colorado-Boulder.

01:01:49.420 --> 01:01:52.810
And we’ve hired her and –
her and her team to just look at

01:01:52.810 --> 01:01:56.450
schools because that is
Lori’s expertise.

01:01:56.450 --> 01:02:00.820
So right now, Lori is – Lori and
her team are going to go up to

01:02:00.820 --> 01:02:05.960
Anchorage, Alaska, and then go down
to Ridgecrest to interview people

01:02:05.960 --> 01:02:10.310
who are working for the school district
to understand – so they’re looking at

01:02:10.310 --> 01:02:13.561
school leaders as well as
emergency managers for the

01:02:13.561 --> 01:02:16.580
recent Anchorage earthquake.
And they’re using a particular

01:02:16.580 --> 01:02:19.130
framework, which Lori developed,
called Ready, Willing, and Able

01:02:19.130 --> 01:02:24.070
framework, to develop their survey
design for California, Oregon,

01:02:24.070 --> 01:02:27.920
Washington, and Alaska after they
complete their initial interviews in

01:02:27.920 --> 01:02:33.540
Alaska and California. Phase I is
their qualitative interview research.

01:02:33.540 --> 01:02:37.620
So they’re going to be doing that in
January and February of next year.

01:02:37.620 --> 01:02:42.360
And they’re really looking, like I said,
at school district representatives and

01:02:42.370 --> 01:02:45.910
emergency managers in these areas.
They’re going to really focus on

01:02:45.910 --> 01:02:48.390
pre-disaster mitigation and
preparedness activities.

01:02:48.390 --> 01:02:53.060
So what people did in Anchorage
and Anchorage surrounds about the

01:02:53.060 --> 01:02:56.440
earthquake as well as Ridgecrest.
You know, did they do ShakeOut?

01:02:56.440 --> 01:02:58.460
Did they have
ShakeOut in schools?

01:02:58.460 --> 01:03:02.700
How did they talk about it
within the school’s community?

01:03:02.710 --> 01:03:06.440
And as well as what people understood
about the impacts of earthquakes.

01:03:06.440 --> 01:03:09.830
And then how, given that Anchorage
and Ridgecrest have had these

01:03:09.830 --> 01:03:14.010
earthquakes, how this has changed
their ideas about what earthquakes

01:03:14.010 --> 01:03:17.510
can do and about their
own preparedness within

01:03:17.510 --> 01:03:20.210
the school environment.
Then they’re going to be doing

01:03:20.210 --> 01:03:24.260
that survey, like I talked about, which
is the quantitative survey research.

01:03:24.260 --> 01:03:27.580
And that’s going to be starting
April 2020. And they’re going to be

01:03:27.580 --> 01:03:31.780
just focusing on K through 12
public schools at a district level.

01:03:31.780 --> 01:03:36.430
And so here is some of their sample
sizes that they’ve shared with us

01:03:36.430 --> 01:03:40.560
in terms of the schools and what
they’re looking at responses.

01:03:40.560 --> 01:03:43.850
So usually with surveys, you get
between 8 to 10% of responses.

01:03:43.850 --> 01:03:48.690
So, you know, for California,
we might get 1,000 responses.

01:03:48.690 --> 01:03:50.870
For Oregon,
we might get 120.

01:03:50.870 --> 01:03:53.490
So that will be enough for us
to have a representative sample size

01:03:53.490 --> 01:03:55.870
in most of these states.

01:03:55.870 --> 01:03:59.780
So here is the project team that
are working with us here.

01:03:59.780 --> 01:04:03.780
Then we’re also working on
analyzing CCTV footage.

01:04:03.790 --> 01:04:06.800
This is a project I started out with Dare
Baldwin at University of Oregon.

01:04:06.800 --> 01:04:12.340
Dare is a psychologist.
And she has a very prestigious career

01:04:12.340 --> 01:04:18.420
analyzing CCTV footage of people’s
behavior in different environments.

01:04:18.420 --> 01:04:23.220
So we brought in Dare because
we wanted her to assist us in analyzing

01:04:23.220 --> 01:04:27.040
people’s behavior that we could
record on Twitter and on Facebook

01:04:27.040 --> 01:04:28.830
and on Instagram.
You know how everyone shares,

01:04:28.830 --> 01:04:31.910
like, here’s me in the earthquake.
Well, we want to analyze that.

01:04:31.910 --> 01:04:35.180
And that has been done before in
Christchurch with Emily Lambie’s

01:04:35.180 --> 01:04:40.030
work where she created a coding
schema to analyze CCTV footage.

01:04:40.030 --> 01:04:42.460
And the research value here is,
we want to see what people did in

01:04:42.460 --> 01:04:46.300
Anchorage and in Ridgecrest without
getting an alert so that, when we

01:04:46.300 --> 01:04:50.420
do have an alert, and we do have
CCTV footage, we can take a look and

01:04:50.420 --> 01:04:55.140
see what the differences are, just by
people – just by watching people.

01:04:55.140 --> 01:04:59.510
Oftentimes, the problem with surveys
and self-reporting is people will tell you

01:04:59.510 --> 01:05:03.740
what they think you want them
to think they think. [laughter]

01:05:03.740 --> 01:05:09.000
I’ll let you think about that for a second.
It’s a bit of a mind twist, but it works.

01:05:09.010 --> 01:05:12.640
So with the self-reporting aspect,
people want you to kind of think that

01:05:12.640 --> 01:05:17.330
they did the right thing, when actually
they might have freaked out and not

01:05:17.330 --> 01:05:20.430
done what they said that they did.
So by analyzing human behavior,

01:05:20.430 --> 01:05:25.010
we actually get to see what
people’s actual behavior was.

01:05:25.010 --> 01:05:29.570
So we’re working with Dare Baldwin’s
lab up at University of Oregon.

01:05:29.570 --> 01:05:33.180
Dare is going to be the North
American repository for CCTV

01:05:33.180 --> 01:05:36.350
footage for earthquake behavior.
So we’re helping her set up that lab

01:05:36.350 --> 01:05:41.240
and the studies for the students there.
This is the final project – Analyzing

01:05:41.240 --> 01:05:43.190
Ridgecrest – Media and Social
Media Analysis for the

01:05:43.190 --> 01:05:47.110
Ridgecrest Earthquakes.
And this – I’m leading this research.

01:05:47.110 --> 01:05:51.380
And so far, I’m doing this alone.
So what I’ve done so far is I’ve

01:05:51.380 --> 01:05:55.030
analyzed 70 media stories about
Ridgecrest and about 2,000 tweets.

01:05:55.030 --> 01:05:59.050
I’m using content analysis using
an NVivo software, which means

01:05:59.050 --> 01:06:05.000
I highlight and code different content,
and then I see the connections between

01:06:05.000 --> 01:06:09.410
frequency and who’s saying
what, where, and when.

01:06:09.410 --> 01:06:13.450
So if you want to know more about that
particular presentation for ShakeAlert,

01:06:13.450 --> 01:06:20.600
I’m presenting the preliminary results of
that next Thursday from 4:15 to 4:20 –

01:06:20.600 --> 01:06:25.120
sorry, that should be 4:30 –
at Moscone South 157, upper –

01:06:25.120 --> 01:06:28.580
at the upper mezzanine.
And so I’ll be talking about that more.

01:06:28.590 --> 01:06:31.140
I also have a couple of other
presentations. I have two

01:06:31.140 --> 01:06:36.180
invited presentations at AGU.
One is the Failure to Alert, which is –

01:06:36.180 --> 01:06:40.520
which is talking about ShakeAlert itself,
and that’s the media analysis and

01:06:40.520 --> 01:06:44.640
social media analysis that
I just talked about here.

01:06:44.640 --> 01:06:47.260
And then there’s another one –
Stop the Presses, which is another

01:06:47.260 --> 01:06:50.720
invited talk on aftershock forecasts.
I’m going after Jeanne Hardebeck.

01:06:50.720 --> 01:06:54.540
We’re going to do sort of a two-punch
there, where Jeanne is going to present

01:06:54.550 --> 01:06:58.780
the aftershock forecast, and I’m going to
present how the media understood and

01:06:58.780 --> 01:07:03.200
communicated the aftershock forecast,
which is the same 70 media stories

01:07:03.200 --> 01:07:08.780
I’ve already analyzed for ShakeAlert.
And that’s on Thursday at 8:15 to

01:07:08.780 --> 01:07:13.020
8:30 a.m., so it’s an early one,
and it’s in Moscone South 157,

01:07:13.020 --> 01:07:16.160
upper mezzanine.
Same place as the other one.

01:07:16.160 --> 01:07:18.850
And then this presentation –
Social Science and ShakeAlert –

01:07:18.850 --> 01:07:23.050
is going to be at a session called
Automatic for the People, which I

01:07:23.050 --> 01:07:29.070
got to use an REM album to title.
And that’s going to be from

01:07:29.070 --> 01:07:33.920
11:00 to 11:15 on Wednesday
at Moscone South 301 to 302.

01:07:33.920 --> 01:07:36.640
So does anyone
have any questions?

01:07:37.640 --> 01:07:43.020
[Applause]

01:07:44.340 --> 01:07:48.060
- [inaudible] questions for Sara?
- Hey, Jeff.

01:07:48.060 --> 01:07:51.940
- Sara, is there anything going on
that’s going to specifically target

01:07:51.940 --> 01:07:58.140
monitoring the difference between
[inaudible] versus [inaudible]?

01:07:58.800 --> 01:08:05.040
[inaudible] populations.

01:08:05.040 --> 01:08:08.400
- It’s a – that’s a really good question.
We’re looking at some specific

01:08:08.410 --> 01:08:12.630
questions on the surveys to ask,
did you download the ShakeAlert app,

01:08:12.630 --> 01:08:15.360
versus those who didn’t.
Because we actually haven’t had

01:08:15.360 --> 01:08:20.660
a WEA sent out yet, it’s a little bit
more complex, whereas we have had

01:08:20.660 --> 01:08:25.920
the opportunity for an app to have
sent out an alert, and it didn’t

01:08:25.920 --> 01:08:28.839
for threshold issues.
So we actually haven’t

01:08:28.839 --> 01:08:32.100
been able to send that –
to focus on that yet.

01:08:32.100 --> 01:08:35.040
But there are plans
to look at them.

01:08:38.140 --> 01:08:40.300
- [inaudible] questions?

01:08:40.860 --> 01:08:43.000
All right. Let’s thank all
of our speakers again.

01:08:43.000 --> 01:08:46.500
[Applause]

01:08:46.500 --> 01:08:51.000
- We will see you at the poster
showcase during Coffee Hour.

01:08:52.860 --> 01:09:12.000
[inaudible background conversations]

01:09:12.000 --> 01:09:15.300
- Thank you.
- Yeah. Thank you. That was great.

01:09:15.300 --> 01:09:17.300
Thanks so much.

01:09:18.000 --> 01:09:36.200
[inaudible background conversations]

01:09:36.200 --> 01:10:05.360
[Silence]