WEBVTT
Kind: captions
Language: en

00:00:02.420 --> 00:00:06.580
[inaudible conversations]

00:00:06.580 --> 00:00:09.240
Good morning, everyone,
and welcome to seminar.

00:00:09.240 --> 00:00:12.240
Next week, our speaker will be
Makoto Otsubo from the

00:00:12.241 --> 00:00:16.230
Geological Survey of Japan, who will
be speaking about stress, pore fluid,

00:00:16.230 --> 00:00:19.930
and rock strength in minor
faults in subduction zones.

00:00:19.930 --> 00:00:23.040
This week, it’s my pleasure
to introduce Eileen Evans.

00:00:23.040 --> 00:00:26.390
Eileen received her bachelor’s
and master’s degree in geophysics

00:00:26.390 --> 00:00:30.150
from Berkeley, followed by
a Ph.D. at Harvard.

00:00:30.150 --> 00:00:34.870
She then joined the USGS as a
Mendenhall postdoc in 2014.

00:00:34.870 --> 00:00:39.440
And today, she’s going to present
the research that she’s done

00:00:39.440 --> 00:00:43.370
as a Mendenhall and – in addition to
some broader thoughts about

00:00:43.370 --> 00:00:46.780
how geodesy can be used for
understanding seismic hazard.

00:00:46.780 --> 00:00:48.000
Eileen?

00:00:51.160 --> 00:00:52.780
- Great. Thanks, Belle.

00:00:54.920 --> 00:00:59.700
All right, so, as Belle said, this is
going to be sort of an overview of the

00:00:59.710 --> 00:01:03.590
work that I’ve done as a Mendenhall
postdoc here in Menlo Park.

00:01:03.590 --> 00:01:06.720
And for most of it, I’m going to talk
about how we use geodesy to

00:01:06.720 --> 00:01:11.570
understand what is happening on faults,
specifically in California.

00:01:11.570 --> 00:01:16.610
And so I know that geologists and
earthquake seismologists and geodesists

00:01:16.610 --> 00:01:20.300
all tend to think about earthquakes
in slightly different ways.

00:01:20.300 --> 00:01:23.790
And I really want to emphasize
what makes geodesy so exciting,

00:01:23.790 --> 00:01:25.370
so to make sure we’re
on the same page,

00:01:25.370 --> 00:01:29.520
I’m going to start out talking
about the earthquake cycle.

00:01:29.520 --> 00:01:32.960
So when talking about interpreting
geodetic observations in the context

00:01:32.960 --> 00:01:36.460
of faults, we do have to talk
about the entire earthquake cycle.

00:01:36.460 --> 00:01:42.200
So the earthquake cycle is
made up of three parts.

00:01:42.200 --> 00:01:44.880
The first part of the earthquake
cycle is what’s happening most of

00:01:44.880 --> 00:01:50.710
the time in between major earthquakes.
And in the western U.S., in between

00:01:50.710 --> 00:01:54.250
major earthquakes,
we have this sort of scenario.

00:01:54.250 --> 00:01:58.320
So in – all right, in the Pacific
Northwest, we have the

00:01:58.320 --> 00:02:02.000
Juan de Fuca Plate subducting
underneath the North American Plate.

00:02:02.000 --> 00:02:08.170
And because this megathrust is locked,
the strain accumulation due to that

00:02:08.170 --> 00:02:13.870
locking pushes GPS velocities
away from the trench, and that is

00:02:13.870 --> 00:02:16.350
exactly what we see sort of
in the Pacific Northwest.

00:02:16.350 --> 00:02:21.030
We see systematic eastward
velocities away from the trench.

00:02:21.030 --> 00:02:24.120
And then, in the rest of the
western U.S., we have a network

00:02:24.120 --> 00:02:28.360
of parallel and sub-parallel strike-slip
faults. And again, in between

00:02:28.360 --> 00:02:31.030
major earthquakes,
these faults are stuck together.

00:02:31.030 --> 00:02:35.310
So you see smooth and gradual
deformation across the boundary where

00:02:35.310 --> 00:02:41.960
you don’t reach the full plate rate until
you move a little bit away from the fault.

00:02:41.960 --> 00:02:45.420
And so all of these faults are
constantly accumulating strain.

00:02:45.430 --> 00:02:48.560
And eventually,
you have an earthquake.

00:02:48.560 --> 00:02:51.500
This is the coseismic part
of the earthquake cycle.

00:02:51.500 --> 00:02:54.459
It’s probably the part of the
earthquake cycle we care about

00:02:54.459 --> 00:02:58.790
the most and are most
interested in understanding.

00:02:58.790 --> 00:03:07.950
And so – right, in the Pacific Northwest,
we would see a systematic motion of

00:03:07.950 --> 00:03:12.510
these GPS velocities back towards the
trench as that elastic strain is released.

00:03:12.510 --> 00:03:16.410
And fortunately, we haven’t had
a major subduction zone earthquake

00:03:16.410 --> 00:03:19.650
in the Pacific Northwest, but we can
get an idea of what that would look like

00:03:19.650 --> 00:03:24.950
in terms of GPS observations from
the Tohoku earthquake in Japan,

00:03:24.950 --> 00:03:29.099
in which there was an
almost instantaneous motion

00:03:29.099 --> 00:03:33.349
of up to 5 meters back
towards the trench.

00:03:33.349 --> 00:03:37.800
And then, on a strike-slip fault,
we would have a similar situation.

00:03:37.800 --> 00:03:41.209
The fault that had been locked
and accumulating strain releases,

00:03:41.209 --> 00:03:46.480
and that displacement allows the fault
to catch up to the rest of the plate.

00:03:46.480 --> 00:03:50.020
So the largest displacements
occur closest to the fault.

00:03:50.020 --> 00:03:54.000
Again, there hasn’t been a major
strike-slip earthquake recently

00:03:54.010 --> 00:03:56.910
in the western U.S., but we can
get a sense of what that would look like

00:03:56.910 --> 00:04:00.950
from the North Anatolian Fault
in Turkey, where the fault is here,

00:04:00.950 --> 00:04:05.040
and the coseismic GPS observations
are shown as these arrows,

00:04:05.040 --> 00:04:08.190
and they are largest
closest to the fault.

00:04:08.190 --> 00:04:11.300
And then, after the earthquake,
we’re not done yet.

00:04:11.300 --> 00:04:13.780
Deformation is still ongoing.

00:04:13.780 --> 00:04:17.889
This is called the postseismic
part of the earthquake cycle.

00:04:17.889 --> 00:04:22.640
And this is another example on the –
on the right is another example

00:04:22.640 --> 00:04:25.839
from the Tohoku earthquake
in which there were an additional

00:04:25.839 --> 00:04:29.599
30 to 40 centimeters of displacement
in the coseismic direction

00:04:29.599 --> 00:04:32.090
in the two weeks
following the earthquake.

00:04:32.090 --> 00:04:35.050
And this is probably the least
well-understood part of the

00:04:35.050 --> 00:04:39.150
earthquake cycle because there is
deformation due to ongoing slip

00:04:39.150 --> 00:04:43.699
on the fault as well as relaxation
of the upper crust – oh, sorry – of the

00:04:43.699 --> 00:04:47.740
lower crust and upper mantle in
response to earthquake stress changes.

00:04:47.740 --> 00:04:49.630
But then eventually,
this postseismic deformation

00:04:49.630 --> 00:04:54.140
decays away, and you start over
again with the interseismic part.

00:04:55.210 --> 00:04:58.100
And so this is what the
entire earthquake cycle looks like

00:04:58.110 --> 00:05:01.210
from the perspective
of a single GPS station.

00:05:01.210 --> 00:05:05.889
This is an example from
California where – so we have

00:05:05.889 --> 00:05:10.969
time on the X axis and position,
or displacement, on the Y axis.

00:05:10.969 --> 00:05:13.750
And so this first part is the
interseismic part of the

00:05:13.750 --> 00:05:18.580
earthquake cycle where you see a very
steady change in position over time.

00:05:18.580 --> 00:05:22.000
And then eventually, we get to
an almost instantaneous change

00:05:22.009 --> 00:05:23.719
in position and time.
That’s the earthquake.

00:05:23.719 --> 00:05:26.419
That is the coseismic displacement.
And then there’s this weird

00:05:26.419 --> 00:05:33.090
nonlinear part that – whoops, sorry –
that eventually decays away.

00:05:33.090 --> 00:05:39.159
And we return to this
steady interseismic strain part.

00:05:39.159 --> 00:05:44.539
And so, at the most basic level,
we hope that if we understand

00:05:44.539 --> 00:05:49.139
this rate of tectonic strain accumulation
on a given fault, this can help us

00:05:49.140 --> 00:05:54.480
better understand the timing
and/or size of a future earthquake.

00:05:56.000 --> 00:05:59.159
And so I think this figure
highlights what is fundamentally

00:05:59.159 --> 00:06:02.649
unique about geodesy.
So in addition to providing

00:06:02.649 --> 00:06:06.569
complementary information to
seismology and earthquake geology

00:06:06.569 --> 00:06:10.410
about the coseismic part
of the earthquake cycle,

00:06:10.410 --> 00:06:13.469
geodesy is really our primary,
if not only, means of observing

00:06:13.469 --> 00:06:17.469
the interseismic and postseismic
parts of the earthquake cycle.

00:06:17.469 --> 00:06:22.240
So in terms of forecasting earthquake
hazards, geodesy is really the only

00:06:22.240 --> 00:06:26.080
technology that can see the development
of an earthquake before it happens.

00:06:27.080 --> 00:06:32.760
My Ph.D. adviser, Brendan Meade,
likes to use the analogy that forecasting

00:06:32.770 --> 00:06:37.789
earthquakes with seismology is like
forecasting the weather with rain gauges.

00:06:37.789 --> 00:06:41.580
By definition, you’re observing
something that has already happened.

00:06:41.580 --> 00:06:44.650
And geodesy is much closer
to using radar that allows us

00:06:44.650 --> 00:06:48.099
to sort of observe a
storm system develop.

00:06:48.099 --> 00:06:51.430
And so I want to talk about
using these types of interseismic

00:06:51.430 --> 00:06:55.620
observations from geodesy
to anticipate seismic hazard.

00:06:57.849 --> 00:07:00.919
And so one of the metrics that’s
especially useful for understanding

00:07:00.919 --> 00:07:06.089
earthquake hazard is the long-term
slip rate on these locked faults.

00:07:06.089 --> 00:07:12.679
And so the goal of most of this talk
will be to use these GPS observations

00:07:12.679 --> 00:07:18.949
in between earthquakes to estimate
long-term slip rates on faults, which we

00:07:18.949 --> 00:07:24.220
expect to be the rate that that fault is sort
of accumulating towards an earthquake.

00:07:25.480 --> 00:07:30.680
However, it’s not quite so simple.
Because these faults are locked in the

00:07:30.680 --> 00:07:34.499
interseismic period between earthquakes,
strain accumulates around them in a way

00:07:34.499 --> 00:07:38.399
that produces a smooth and
continuous velocity field in space.

00:07:38.399 --> 00:07:44.119
So looking at this map of
GPS velocities, it looks like

00:07:44.119 --> 00:07:48.319
there’s a pretty clear boundary
where the San Andreas Fault is.

00:07:48.319 --> 00:07:51.429
But if we actually zoom in to a
profile across some of these

00:07:51.429 --> 00:07:54.759
observations, we see that the situation
is a little bit more complicated.

00:07:54.759 --> 00:08:00.189
So now we’re looking at a profile right
here across the San Jacinto and San

00:08:00.189 --> 00:08:05.320
Andreas Faults in southern California.
This is a figure from Yuri Fialko.

00:08:05.320 --> 00:08:11.360
And so in this profile, the symbols
are GPS observations.

00:08:11.369 --> 00:08:14.649
The dots are
InSAR observations.

00:08:14.649 --> 00:08:19.340
And the red line is a model
that fits this interseismic strain

00:08:19.340 --> 00:08:23.300
accumulation to the faults
shown in these green lines.

00:08:24.880 --> 00:08:30.980
And so Yuri Fialko was lucky to
be working in southern California

00:08:30.990 --> 00:08:34.459
where we do have a pretty good
idea about where the faults are.

00:08:34.459 --> 00:08:42.040
But you can imagine – but if we
remove those faults in the model,

00:08:42.040 --> 00:08:45.070
you can see just how smooth
that geodetic gradient is.

00:08:45.070 --> 00:08:47.840
And so it’s not always
immediately clear,

00:08:47.840 --> 00:08:51.220
given GPS observations alone,
or geodetic observations alone, which

00:08:51.220 --> 00:08:55.560
faults are contributing to interseismic
deformation and by how much.

00:08:55.560 --> 00:08:57.850
And so this is really the
fundamental challenge in

00:08:57.850 --> 00:09:02.160
interpreting geodetic observations
in terms of slip rates on faults.

00:09:02.160 --> 00:09:04.900
And that’s that the
location and geometry

00:09:04.910 --> 00:09:08.940
of faults really has to be
prescribed and modeled.

00:09:08.940 --> 00:09:11.960
And because faults are locked
most of the time, it can be hard to

00:09:11.960 --> 00:09:16.050
know the best way to do that.
And so this is probably the source

00:09:16.050 --> 00:09:20.240
of most of our uncertainty in
determining geodetic slip rates.

00:09:21.220 --> 00:09:23.060
And this is, of course,
different from geologic methods

00:09:23.070 --> 00:09:27.030
where you do definitely know
where the fault is but don’t necessarily

00:09:27.030 --> 00:09:30.030
capture the broader
tectonic context.

00:09:30.030 --> 00:09:33.880
So when I’m talking about geodetic fault
slip rates, I’m talking about a long-term

00:09:33.880 --> 00:09:39.630
fault slip rate that is estimated from the
interseismic geodetic velocity field.

00:09:39.630 --> 00:09:43.110
And we do, in most cases,
expect that to be equal to

00:09:43.110 --> 00:09:45.660
the geologic slip rate
on the fault.

00:09:47.370 --> 00:09:51.160
But that’s not always true.
And when geologic and geodetic

00:09:51.160 --> 00:09:55.210
slip rates don’t agree, that’s where
things get really interesting.

00:09:55.210 --> 00:09:58.550
And so for this first part of the talk,
I’m actually going to focus on a region

00:09:58.550 --> 00:10:02.641
in which many geologically estimated
slip rates do not agree with

00:10:02.641 --> 00:10:08.060
geodetically estimated slip rates
and what we can learn from that.

00:10:08.060 --> 00:10:13.560
And so I’m going to focus on
the eastern California shear zone.

00:10:13.560 --> 00:10:17.670
So the next slide will be
a blow-up of that area.

00:10:17.670 --> 00:10:21.400
So the eastern California shear zone
has been identified as a region

00:10:21.400 --> 00:10:29.030
where geodetic slip rates and geologic
slip rates seem to systematically disagree.

00:10:29.030 --> 00:10:34.320
And this is highlighted in
these figures from Kaj Johnson,

00:10:34.320 --> 00:10:39.090
in which the geodetic slip rate estimates
are shown in the figure on the left,

00:10:39.090 --> 00:10:43.660
and the geologic slip rates estimates
are shown in the figure on the right.

00:10:43.660 --> 00:10:47.450
And in particular to highlight the
eastern Mojave, the geodetic slip rates

00:10:47.450 --> 00:10:51.240
here are systematically higher
than the geologic slip rates with

00:10:51.240 --> 00:10:57.340
up to 18 millimeters per year estimated
across the region geodetically.

00:10:57.340 --> 00:10:59.940
But if you add up all of the
geologic slip rates across –

00:10:59.940 --> 00:11:04.270
on all of the faults across the region,
you can account for no more than,

00:11:04.270 --> 00:11:07.250
optimistically,
8 millimeters per year.

00:11:07.250 --> 00:11:10.200
And this discrepancy has
led to the hypothesis

00:11:10.200 --> 00:11:14.920
that some of this deformation
might be distributed deformation

00:11:14.920 --> 00:11:19.940
that’s being mapped into
the geodetic slip rate estimates.

00:11:19.940 --> 00:11:22.840
And so this idea of distributed
deformation is illustrated in this

00:11:22.840 --> 00:11:26.860
conceptual model of continental
deformation – or this conceptual diagram

00:11:26.860 --> 00:11:30.110
of continental deformation at
the bottom from Wayne, where,

00:11:30.110 --> 00:11:35.410
at one end of the spectrum, perhaps
continental deformation is best described

00:11:35.410 --> 00:11:39.370
as a continuum, in which case we would
need an infinite number of faults.

00:11:39.370 --> 00:11:44.300
And at the other end of the spectrum,
continental deformation is

00:11:44.300 --> 00:11:47.700
best described by plate tectonics,
in which case, we really only

00:11:47.700 --> 00:11:50.060
need one fault
between plates.

00:11:50.060 --> 00:11:57.070
And so sort of understanding how to
turn geodetic observations into fault

00:11:57.070 --> 00:12:01.700
slip rates requires some assessment
of where we fall on that spectrum.

00:12:01.700 --> 00:12:06.220
And if distributed deformation
is really at play in the eastern

00:12:06.230 --> 00:12:09.700
California shear zone, that would
imply a situation much closer

00:12:09.700 --> 00:12:14.980
to this continuum model, in which case,
we need many, many faults.

00:12:15.990 --> 00:12:22.640
And so – and so I’m going to
use an approach to block modeling –

00:12:22.640 --> 00:12:25.200
a kind of novel approach
to block modeling to address

00:12:25.200 --> 00:12:30.260
this question exactly in
the context of this spectrum.

00:12:30.260 --> 00:12:34.800
So a block model – in a block model,
we divide the crust into microplates

00:12:34.800 --> 00:12:38.360
that are bounded by faults.
And then this model is constrained

00:12:38.360 --> 00:12:42.690
by GPS to define our deformation
in terms of the relative motions

00:12:42.690 --> 00:12:48.130
of microplates and the elastic
signal due to locked faults.

00:12:48.130 --> 00:12:51.570
And so this is a block model –
on the left is a block model

00:12:51.570 --> 00:12:54.580
that I built for the
eastern California shear zone.

00:12:54.580 --> 00:12:57.960
You’ll notice that I have defined
a lot of blocks in the eastern

00:12:57.960 --> 00:13:00.960
California shear zone.
There are 64 blocks there.

00:13:00.960 --> 00:13:04.180
And that’s going to let us ask
this question of complexity

00:13:04.180 --> 00:13:09.100
in terms of how many of those
64 blocks, or microplates,

00:13:09.100 --> 00:13:14.180
do we really need to describe eastern
California shear zone deformation?

00:13:15.050 --> 00:13:20.460
And so this concept is illustrated here
in this cartoon, which is not to scale,

00:13:20.470 --> 00:13:23.370
in which we can imagine
a scenario in which

00:13:23.370 --> 00:13:28.410
Block 1 and Block 2 are moving,
or rotating, relative to Block 3.

00:13:28.410 --> 00:13:31.360
They’re moving on a sphere,
so we define their motions in terms of

00:13:31.360 --> 00:13:36.030
Euler poles and rotation rates.
If Block 1 and Block 2 have

00:13:36.030 --> 00:13:39.710
different Euler poles and rotation rates,
these two blocks are behaving

00:13:39.710 --> 00:13:44.600
as separate blocks, and there will be
some slip on the fault between them.

00:13:44.600 --> 00:13:49.430
Alternatively, if we imagine a
situation in which Block 1 and Block 2

00:13:49.430 --> 00:13:53.920
have an identical Euler pole and
rotation rate, these two blocks

00:13:53.920 --> 00:13:57.180
are effectively behaving
as a single larger block,

00:13:57.180 --> 00:14:01.150
and slip on the fault between
them would go to zero.

00:14:01.150 --> 00:14:04.560
And so, in terms of figuring out
how many microplates we need,

00:14:04.560 --> 00:14:09.560
we can start with a lot of blocks, like
64 blocks, and then look for a solution

00:14:09.560 --> 00:14:14.100
in which many of those microplates
have identical rotation vectors.

00:14:14.100 --> 00:14:20.230
In other words, we want to look
for a solution that is grouped.

00:14:20.230 --> 00:14:25.040
And we can do this with a method
that’s borrowed from image processing.

00:14:25.920 --> 00:14:30.980
So total variation regularization
is a method borrowed from

00:14:30.990 --> 00:14:35.450
image processing. It’s a method
for sharpening edges in an image.

00:14:35.450 --> 00:14:39.570
And so the way it works is this.
So if we want to recover a true signal

00:14:39.570 --> 00:14:45.010
from some noisy data, a common way
to influence – to reduce the influence

00:14:45.010 --> 00:14:50.410
of noise is to smooth the solution.
But, of course, in terms of an image,

00:14:50.410 --> 00:14:53.700
smoothing noisy data
results in a blurry image.

00:14:53.700 --> 00:14:57.620
Alternatively, total variation
regularization minimizes the number

00:14:57.620 --> 00:15:02.970
of differences within the solution.
So in terms of an image, this identifies

00:15:02.970 --> 00:15:09.320
the number of unique – of necessary
unique different pixel colors,

00:15:09.320 --> 00:15:13.430
and by doing that, identifies the
most important color boundaries.

00:15:13.430 --> 00:15:16.150
And so the way that –
excuse me – the way this works

00:15:16.150 --> 00:15:19.490
in a block model
looks like this.

00:15:19.490 --> 00:15:25.130
So total variation regularization is
just a minimization of two terms.

00:15:25.130 --> 00:15:27.240
This first term ...

00:15:29.660 --> 00:15:33.980
So we’re minimizing
the L-2 norm of our residuals.

00:15:33.980 --> 00:15:40.260
So lambda would be our block rotations.
That’s the solution that we want.

00:15:40.260 --> 00:15:43.360
And so this first term
is just a least squares.

00:15:43.370 --> 00:15:48.360
So if this was all by itself,
minimizing the L-2 norm of residuals,

00:15:48.360 --> 00:15:55.020
that’s just least squares. We’re tacking
onto that an L-1 norm regularization

00:15:55.020 --> 00:16:00.870
term, which is the L-1 norm on
differences in the rotation vector.

00:16:00.870 --> 00:16:06.280
The L-1 norm is defined as the
sum of absolute values in a vector.

00:16:06.280 --> 00:16:12.040
Minimizing the L-1 norm results
in that vector becoming sparse.

00:16:12.040 --> 00:16:15.770
And so, by applying this to
differences in the rotation vector,

00:16:15.770 --> 00:16:17.360
this gives us
exactly what we want.

00:16:17.360 --> 00:16:22.600
It gives us a solution that is
grouped with many identical values.

00:16:22.600 --> 00:16:25.620
And the strength of that
grouping is controlled with this –

00:16:25.620 --> 00:16:28.980
a scaler tuning
parameter, lambda.

00:16:29.560 --> 00:16:33.740
Unlike traditional least squares,
there’s no exact single-step solution,

00:16:33.740 --> 00:16:38.620
but there are lots of ways to solve it
with convex optimization methods.

00:16:39.490 --> 00:16:47.860
So going back to that model, again,
this total variation regularization

00:16:47.860 --> 00:16:51.360
approach allows us to
start with a lot of blocks.

00:16:51.360 --> 00:16:53.600
That means – because all of
the blocks are bounded by faults,

00:16:53.600 --> 00:16:56.740
this means we get to
start with a lot of faults.

00:16:56.740 --> 00:17:01.980
And actually, because this total
variation regularization approach

00:17:01.990 --> 00:17:06.789
allows us to include so many blocks
and so many faults, this is actually

00:17:06.789 --> 00:17:11.809
the first geodetic model of the
eastern California shear zone

00:17:11.809 --> 00:17:15.800
to include all of the faults on
which we have geologic slip rates.

00:17:18.520 --> 00:17:22.419
So I’m also considering
11 geologic slip rates.

00:17:22.419 --> 00:17:24.640
Those are shown in the
circles that are color-coded

00:17:24.640 --> 00:17:28.049
by their slip rate
and magnitude.

00:17:28.049 --> 00:17:32.360
And I’m actually not going to use
the geologic slip rates as constraints.

00:17:32.360 --> 00:17:37.600
I’m going to use them as a comparison –
as a way to test the model.

00:17:39.200 --> 00:17:43.300
So – okay, so the idea here is that this
dense – this very dense fault geometry,

00:17:43.309 --> 00:17:46.269
where we have many, many faults,
including faults in between

00:17:46.269 --> 00:17:51.769
some of these geologic slip rates,
should be representative of

00:17:51.769 --> 00:17:53.980
something more like
distributed deformation –

00:17:53.980 --> 00:18:00.269
something more like – closer to this left-
side continuum model of deformation.

00:18:00.269 --> 00:18:05.730
And so we should expect that, if the
reason for slip rate discrepancies is this

00:18:05.730 --> 00:18:11.169
distributed deformation, including more
faults should reduce the discrepancy.

00:18:11.169 --> 00:18:17.850
So we’re going to go through this
problem asking the question, which of

00:18:17.850 --> 00:18:22.440
these geodetic models is most
consistent with the geologic slip rates?

00:18:22.440 --> 00:18:26.852
So we’re going to step through
a bunch of these – a bunch of

00:18:26.852 --> 00:18:33.809
block models in which I have
applied total variation regularization

00:18:33.809 --> 00:18:37.960
with different – with different
levels of grouping.

00:18:37.960 --> 00:18:41.720
So on the right, we’re going to
step through an animation

00:18:41.720 --> 00:18:46.269
in which each step represents
increasing that total variation

00:18:46.269 --> 00:18:49.409
regularization parameter
or increasing the grouping.

00:18:49.409 --> 00:18:51.800
And the faults are
color-coded by slip rate.

00:18:51.800 --> 00:18:56.120
Right-lateral is positive, or red.

00:18:56.129 --> 00:19:00.690
And the geologic slip rates are shown in
the same color scale in the filled circles.

00:19:00.690 --> 00:19:03.420
So I’ll play this animation.

00:19:04.720 --> 00:19:10.540
As the blocks are grouped,
the slip rates between them turn off.

00:19:10.540 --> 00:19:12.520
And – did it play?

00:19:18.800 --> 00:19:20.140
Yeah. Okay.

00:19:20.149 --> 00:19:22.450
So it steps through very quickly.
[chuckles]

00:19:22.450 --> 00:19:28.840
And the – as the blocks get grouped,
the slip rates on the faults

00:19:28.840 --> 00:19:32.580
between them go to zero.
That’s what’s shown as the black faults.

00:19:32.580 --> 00:19:36.680
So all of the black faults
have slip rates of exactly zero.

00:19:37.440 --> 00:19:41.500
And so we can look at each of
those models, as we step through

00:19:41.500 --> 00:19:44.659
all of those different block models,
and look for the solution

00:19:44.659 --> 00:19:47.399
that is most consistent
with the geologic slip rates.

00:19:47.400 --> 00:19:50.400
And the geologic slip rates
are shown on the top left.

00:19:52.360 --> 00:20:01.660
This figure on the bottom is showing
the number of blocks on the X axis.

00:20:02.960 --> 00:20:06.980
Sorry – number of
blocks on the X axis.

00:20:06.980 --> 00:20:11.399
And this gray line is the mean
residual velocity, or our misfit.

00:20:11.399 --> 00:20:13.840
So when we have a lot of blocks,
our misfit is lowest.

00:20:13.840 --> 00:20:17.289
When we have the fewest blocks,
our misfit is worst.

00:20:17.289 --> 00:20:23.831
The blue line is the number of these
geodetically estimated block model

00:20:23.831 --> 00:20:29.289
slip rates that agree within
uncertainty with the geologic rates.

00:20:29.289 --> 00:20:37.559
And so there is one model –
one of these block models that is in best

00:20:37.559 --> 00:20:42.740
agreement with the geologic slip rates.
But the best it does is 5.

00:20:42.740 --> 00:20:46.240
And I’m considering –
there are 11 geologic slip rates.

00:20:46.240 --> 00:20:48.980
The best we can do is 5.

00:20:48.980 --> 00:20:53.460
And none of them do even better,
even when we have more active blocks.

00:20:53.470 --> 00:20:56.010
And so that’s how we
selected the reference model.

00:20:56.010 --> 00:20:58.620
So there’s the
reference model.

00:20:58.620 --> 00:21:02.759
This reference model fits
the GPS observations

00:21:02.759 --> 00:21:07.250
with a mean residual velocity
of 1.5 millimeters per year.

00:21:07.250 --> 00:21:12.929
We still end up with 17.6 millimeters
per year cumulative slip across the

00:21:12.929 --> 00:21:16.560
eastern Mojave – there it is –
which certainly doesn’t solve this

00:21:16.560 --> 00:21:20.060
geologic/geodetic discrepancy issue.
And actually, there are very

00:21:20.060 --> 00:21:23.000
high discrepancies on
a couple of the faults.

00:21:23.000 --> 00:21:29.170
There’s a high discrepancy on the Calico
Fault right here, where the block model

00:21:29.170 --> 00:21:34.730
is estimating 7.6 millimeters per year,
and the geologic rate is 1.8.

00:21:34.730 --> 00:21:39.851
And on the Garlock Fault,
we’re estimating less than 1 millimeter

00:21:39.860 --> 00:21:44.040
per year, and the
geologic rate is over 5.

00:21:46.760 --> 00:21:49.280
So that’s interesting.

00:21:49.280 --> 00:21:54.300
Another thing to point out is that this –
so this is the model with the

00:21:54.309 --> 00:21:57.419
best agreement between
geologic and geodetic slip rates.

00:21:57.419 --> 00:22:02.749
But actually, all of these models
fit the GPS velocities pretty well.

00:22:02.749 --> 00:22:04.110
We’re looking at
mean residual velocities

00:22:04.110 --> 00:22:09.019
less than 2 millimeters
per year for all of them.

00:22:09.019 --> 00:22:14.360
So we might also want to think about
this model in the context of the entire

00:22:14.360 --> 00:22:18.150
suite of models and geometries that we
stepped through in that animation.

00:22:18.150 --> 00:22:20.590
All of those are fitting
our GPS velocities

00:22:20.590 --> 00:22:23.380
better than 2 millimeters per year.

00:22:24.440 --> 00:22:30.299
And so that’s what’s shown here.
On the left is this – is a comparison

00:22:30.299 --> 00:22:34.080
of geologic and geodetic
slip rates from this study.

00:22:34.080 --> 00:22:39.990
So we have geologic slip rates on the
X axis, geodetic slip rates on the Y axis.

00:22:39.990 --> 00:22:45.039
The geodetic slip rates
relative to geologic slip rates

00:22:45.039 --> 00:22:49.279
from the preferred model
are shown in these boxes.

00:22:49.280 --> 00:22:55.460
And the horizontal error bars are the
uncertainty in the geologic slip rate.

00:22:56.920 --> 00:23:01.480
And then the vertical distribution
represents the distribution of

00:23:01.480 --> 00:23:03.940
estimated slip rates –
the relative distribution

00:23:03.940 --> 00:23:08.279
of estimated slip rates from that suite
of models considered in this study.

00:23:08.280 --> 00:23:15.160
So for example, in the reference model,
the Camp Rock Fault, which is here,

00:23:15.160 --> 00:23:18.580
has a geodetic slip rate that’s much
higher than the geologic slip rate.

00:23:18.580 --> 00:23:21.159
Geodetically, we’re estimating
over 4-1/2 millimeters per year,

00:23:21.159 --> 00:23:24.190
but the geologic slip
rate is about 1-1/2.

00:23:24.190 --> 00:23:29.519
But in a few of these models,
the slip rate on the Camp Rock Fault

00:23:29.519 --> 00:23:34.250
does agree with geology.
And in some cases, it’s even lower.

00:23:34.250 --> 00:23:37.119
So this is a way to think about
the relative distributions

00:23:37.119 --> 00:23:41.440
of all of these different slip rates
among all of those models.

00:23:41.440 --> 00:23:45.080
And the thing that’s – that I think is
really interesting is that there are

00:23:45.080 --> 00:23:50.059
persistent slip rate discrepancies
regardless of the geometry that we used.

00:23:50.059 --> 00:23:55.879
So across all of the models,
the Calico Fault is higher –

00:23:55.879 --> 00:23:59.269
consistently and significantly
higher than the geologic slip rate.

00:23:59.269 --> 00:24:02.270
We never get lower than
about 5-1/2 millimeters per year.

00:24:02.270 --> 00:24:08.059
Also, on the Garlock Fault,
the geodetically estimated slip rates

00:24:08.060 --> 00:24:12.560
within these block models
are consistently low.

00:24:13.380 --> 00:24:19.540
And so this is maybe suggestive
that these discrepancies between

00:24:19.549 --> 00:24:24.039
geologic and geodetic slip rates
on those two faults are real.

00:24:24.039 --> 00:24:26.980
There’s something interesting
going on there.

00:24:26.980 --> 00:24:32.809
And we’re not the first to suggest –
to at least identify the Calico Fault

00:24:32.809 --> 00:24:37.539
as a major player
in this discrepancy.

00:24:37.539 --> 00:24:44.960
Sally McGill at CSU-San Bernardino
and a bunch of her students

00:24:44.960 --> 00:24:52.009
collected a dense field campaign –
yeah, GPS observations from

00:24:52.009 --> 00:24:55.560
a dense field campaign across a
narrow profile in southern California.

00:24:55.560 --> 00:24:58.220
And the eastern edge
of the profile goes through

00:24:58.220 --> 00:25:00.830
the eastern California
shear zone.

00:25:00.830 --> 00:25:06.809
And so here is the distance along that
profile and their estimated slip rates.

00:25:06.809 --> 00:25:10.830
The eastern California
shear zone – part of it is here.

00:25:10.830 --> 00:25:14.549
And they also identify the
Calico Fault as the fastest fault

00:25:14.549 --> 00:25:18.019
in the eastern Mojave
with a rate even higher than

00:25:18.020 --> 00:25:21.880
we estimated of over
10 millimeters per year.

00:25:23.169 --> 00:25:26.580
And so I think this is really exciting
because it suggests that this

00:25:26.580 --> 00:25:31.260
geologic/geodetic slip rate discrepancy,
at least in the eastern Mojave,

00:25:31.260 --> 00:25:35.639
may be concentrated on or near the
Calico Fault rather than amorphously

00:25:35.639 --> 00:25:38.889
distributed across the region,
which presents a focused way forward

00:25:38.889 --> 00:25:42.000
in terms of understanding
eastern California shear zone

00:25:42.000 --> 00:25:46.000
deformation and identifying
the source of this discrepancy.

00:25:46.000 --> 00:25:50.540
And total variation regularization
allows us to explore the implications

00:25:50.540 --> 00:25:56.500
of model uncertainty of many of these
fault system geometries in a way

00:25:56.500 --> 00:26:00.580
that’s really impossible with
traditional geodetic modeling methods.

00:26:02.340 --> 00:26:05.220
And so we can actually
think about these distributions

00:26:05.220 --> 00:26:10.559
of estimated slip rate as a
proxy for our model uncertainty.

00:26:10.559 --> 00:26:13.539
But of course, these model
uncertainties are still estimated

00:26:13.539 --> 00:26:18.789
within an elastic block model with a
starting geometry that I prescribed.

00:26:18.789 --> 00:26:23.321
And so I haven’t necessarily captured
the entire range of possible slip rate

00:26:23.321 --> 00:26:26.980
values for different
modeling assumptions.

00:26:26.980 --> 00:26:30.059
So for this next part of the talk,
I’m going to talk about how we

00:26:30.059 --> 00:26:36.639
actually might go about calculating a
more realistic modeling uncertainty,

00:26:36.639 --> 00:26:41.710
or model – or epistemic uncertainty
in our geodetic slip rates.

00:26:41.710 --> 00:26:44.779
And I want to do this because,
getting back to this more

00:26:44.779 --> 00:26:47.909
direct question of seismic hazard,
one of the fundamental inputs

00:26:47.909 --> 00:26:52.440
into seismic hazard models and
rupture forecasting is fault slip rate.

00:26:52.440 --> 00:26:56.049
But we need –
in order to include geodesy,

00:26:56.049 --> 00:27:00.149
we need robust estimates of
geodetic slip rate and uncertainty.

00:27:00.149 --> 00:27:04.350
So this flow chart is from UCERF3,
in which community fault models

00:27:04.350 --> 00:27:09.710
are incorporated into deformation
models to estimate fault slip rates.

00:27:09.710 --> 00:27:12.230
Which then may be passed
along to earthquake rate models

00:27:12.230 --> 00:27:13.710
and rupture
probabilities.

00:27:13.710 --> 00:27:18.840
And I’ve highlighted the section
to which geodesy is uniquely capable

00:27:18.840 --> 00:27:21.830
of contributing.
Because, as I mentioned before,

00:27:21.830 --> 00:27:25.300
the advantage of geodesy is that
it records this active elastic strain

00:27:25.300 --> 00:27:31.759
accumulation around faults right now,
which may be – may be an advantage

00:27:31.759 --> 00:27:36.710
over geologic methods, which average
over several earthquake cycles.

00:27:36.710 --> 00:27:42.730
However, as I discussed a few
slides ago, and as is alluded to in

00:27:42.730 --> 00:27:52.680
this flow chart, these geodetic slip
rate models require a fault model.

00:27:53.789 --> 00:27:59.020
And so what I’ve shown on the
bottom are five examples of geodetic

00:27:59.020 --> 00:28:03.570
slip rate models. All of these have
different modeling assumptions.

00:28:03.570 --> 00:28:06.500
They all use different
fault system geometries.

00:28:06.500 --> 00:28:10.780
They all have slightly
different focus regions.

00:28:12.309 --> 00:28:17.740
And so – and they all produce
different slip rate estimates.

00:28:17.750 --> 00:28:21.269
And it isn’t necessarily surprising
that they would give us different

00:28:21.269 --> 00:28:24.039
slip rate estimates because
they’re using very different

00:28:24.039 --> 00:28:26.960
assumptions and
different geometries.

00:28:26.960 --> 00:28:31.610
And in fact, in California,
I’ve identified 33 geodetic slip rate

00:28:31.610 --> 00:28:36.620
models from 32 publications.
These include a variety of study types,

00:28:36.620 --> 00:28:41.029
including elastic block models,
dislocation models, finite element

00:28:41.029 --> 00:28:45.159
models, viscoelastic block models,
and combinations of the above.

00:28:45.159 --> 00:28:48.160
The eastern California
shear zone model that I was

00:28:48.160 --> 00:28:51.460
talking about earlier
is right here.

00:28:54.860 --> 00:29:00.800
Right, so these are all
peer-reviewed published models.

00:29:00.800 --> 00:29:06.340
And so, in some sense, we can
consider that a threshold to validity.

00:29:06.340 --> 00:29:10.960
Therefore, the variability in slip rates
among all of these different studies

00:29:10.960 --> 00:29:14.899
represents variability
in valid model choices.

00:29:14.899 --> 00:29:17.799
And in many cases,
these represent genuinely –

00:29:17.799 --> 00:29:21.149
genuine philosophical differences
that may be important to consider

00:29:21.149 --> 00:29:23.240
in the context of
seismic hazard.

00:29:23.240 --> 00:29:29.429
So this suite of models, arguably –
I will argue – represents the

00:29:29.429 --> 00:29:33.990
range of acceptable slip rates.
And that range itself is what we

00:29:33.990 --> 00:29:39.700
might consider a proxy for our model
uncertainty in these geodetic slip rates.

00:29:43.250 --> 00:29:47.100
And so what I would like to propose is
that we leverage this suite of existing

00:29:47.100 --> 00:29:50.899
models and ongoing research to
summarize the geodetic perspective

00:29:50.899 --> 00:29:54.919
on seismic hazard to provide
to future versions of UCERF

00:29:54.920 --> 00:29:58.060
or the National Seismic
Hazard Model.

00:30:00.049 --> 00:30:02.240
Going about this isn’t
entirely straightforward.

00:30:02.250 --> 00:30:04.820
Because, as I mentioned,
different people may use

00:30:04.820 --> 00:30:07.820
different geometries to
describe the same fault system.

00:30:07.820 --> 00:30:12.979
So this is just a cartoon example where
one author might define the fault system

00:30:12.979 --> 00:30:19.029
in blue, and another author might define
the same cartoon fault system in orange.

00:30:19.029 --> 00:30:21.880
And, in some places,
they might agree perfectly.

00:30:21.880 --> 00:30:25.289
And in those cases,
it’s very easy to compare models.

00:30:25.289 --> 00:30:31.160
But in other cases, they might use very
different geometries for the same fault –

00:30:31.160 --> 00:30:37.120
where did it go – or different
numbers of fault strands.

00:30:37.120 --> 00:30:42.070
And so it may not be immediately
obvious how much these slip rates

00:30:42.070 --> 00:30:46.289
actually agree or disagree,
and that disagreement is what we

00:30:46.289 --> 00:30:49.070
want to understand
in terms of representing

00:30:49.070 --> 00:30:53.729
our inherent uncertainty
in the fault system geometry.

00:30:53.729 --> 00:30:59.729
And so the way that I propose
to address this is to compare slip rates

00:30:59.729 --> 00:31:02.549
on a geographic grid.
And then, within each

00:31:02.549 --> 00:31:06.320
geographic grid cell,
simply ask the question,

00:31:06.320 --> 00:31:11.889
do slip rates from Study 1 and
Study 2 in this grid cell result in

00:31:11.889 --> 00:31:16.420
the same spatially average
deformation of this rectangle?

00:31:19.169 --> 00:31:23.200
And we can answer this question
by decomposing every study

00:31:23.210 --> 00:31:27.440
in every grid cell into a
horizontal velocity gradient tensor.

00:31:27.440 --> 00:31:32.120
So I’m only considering the
horizontal components of fault slip.

00:31:33.460 --> 00:31:36.720
So we decompose every study
in every grid cell into a horizontal

00:31:36.720 --> 00:31:41.879
velocity gradient tensor, which we
can then convert to spatially

00:31:41.879 --> 00:31:46.720
average slip on two perpendicular
structures in every grid cell.

00:31:46.720 --> 00:31:49.399
So we get – for every grid cell,
for every study,

00:31:49.399 --> 00:31:54.039
we get two components of shear
and two components of opening.

00:31:54.039 --> 00:31:59.139
And this gives us four components of
slip for every study, in every grid cell,

00:31:59.139 --> 00:32:03.340
that represents the spatial average of
deformation due to that study.

00:32:03.340 --> 00:32:07.059
And then, once we have this,
we can rotate those four components

00:32:07.059 --> 00:32:11.610
of slip into any
orientation of interest.

00:32:11.610 --> 00:32:17.649
For example, the Pacific North America
plate boundary parallel direction.

00:32:17.649 --> 00:32:20.649
And then we can take statistics
on those components of slip.

00:32:20.649 --> 00:32:29.100
So here I’m showing, at the top,
the average slip rate for each of

00:32:29.100 --> 00:32:32.649
those four components
of slip over all 33 studies.

00:32:32.649 --> 00:32:36.099
And at the bottom are the standard
deviations on those slip rates.

00:32:36.099 --> 00:32:40.960
And unsurprisingly, the most
striking feature is right-lateral

00:32:40.960 --> 00:32:43.500
plate boundary shear along the
San Andreas Fault system.

00:32:43.500 --> 00:32:47.970
So now right-lateral is blue – sorry.

00:32:49.060 --> 00:32:50.380
Okay.

00:32:51.280 --> 00:32:58.380
So just to give a sense of the suite of
models contained within each grid cell,

00:32:58.389 --> 00:33:01.470
here’s a look at a profile along
the central San Andreas Fault.

00:33:01.470 --> 00:33:04.970
So this red line is
showing the profile.

00:33:04.970 --> 00:33:08.950
And here’s the distance
along that profile.

00:33:08.950 --> 00:33:15.169
Each of the – okay, so the black line
shows the average slip rate in this

00:33:15.169 --> 00:33:18.950
orientation over all of those
grid cells along this profile.

00:33:18.950 --> 00:33:28.450
Each of the gray lines is representing the
slip rate due to one of those 33 studies.

00:33:28.450 --> 00:33:32.580
And the gray – shaded gray area
is the standard deviation.

00:33:32.580 --> 00:33:36.769
And so we can think of the mean
in each study as a community average

00:33:36.769 --> 00:33:41.859
slip rate and the standard deviation as
the uncertainty due to these many

00:33:41.860 --> 00:33:47.340
modeling approaches, or the model,
or epistemic, uncertainty.

00:33:48.960 --> 00:33:54.640
So taking a closer look at those
standard deviations, just averaging

00:33:54.650 --> 00:33:59.179
up the standard deviation among
all four components, we get an

00:33:59.179 --> 00:34:02.779
average standard deviation
of 1.5 millimeters per year.

00:34:02.779 --> 00:34:08.510
So if we’re looking for just an average
ballpark model uncertainty number to

00:34:08.510 --> 00:34:14.460
be throwing around when we’re talking
about geodetic slip rates, that’s it.

00:34:15.730 --> 00:34:18.780
But of course, you can see
at the bottom there’s a lot of

00:34:18.790 --> 00:34:21.840
complexity and spatial
variability in these slip rates.

00:34:21.840 --> 00:34:27.179
And that’s really interesting when
compared to the map on the top right

00:34:27.179 --> 00:34:32.150
where the colors represent the
number of models in each grid cell.

00:34:32.150 --> 00:34:40.970
So the densest models, or the most
models, are on the central San Andreas.

00:34:40.970 --> 00:34:46.840
And tend to be lower
sort of in northern California.

00:34:46.840 --> 00:34:50.920
And so this is interesting
because the standard deviations

00:34:50.929 --> 00:34:56.830
tend to be highest where
the fault system is complicated

00:34:56.830 --> 00:34:59.240
but where there are
relatively few models.

00:34:59.240 --> 00:35:09.710
So along the northern San Andreas – if I
can get this – yeah, so along the northern

00:35:09.710 --> 00:35:15.440
San Andreas, there are high standard
deviations and relatively few models.

00:35:15.440 --> 00:35:19.600
And in the plate boundary parallel
opening, we get high standard deviations

00:35:19.600 --> 00:35:24.380
down near the transverse ranges where
there are also relatively few models.

00:35:24.380 --> 00:35:29.720
But then the standard deviations are
lowest where there are very few models.

00:35:31.100 --> 00:35:34.180
I think that’s often because, where there
are very few models, it tends to be the

00:35:34.190 --> 00:35:38.410
same research group doing those models.
[laughter]

00:35:38.410 --> 00:35:41.490
And also relatively low, actually,
along the central San Andreas Fault,

00:35:41.490 --> 00:35:46.570
even though there are –
that’s where the most studies are.

00:35:46.570 --> 00:35:52.860
And so, especially in these locations
where our standard deviation is high,

00:35:52.860 --> 00:35:57.050
but our number of models is low,
this is a nice quantitative way

00:35:57.050 --> 00:36:01.290
to target locations
for future research.

00:36:01.290 --> 00:36:04.620
And so then another way to leverage
standard deviation as a proxy for

00:36:04.620 --> 00:36:09.820
uncertainty is in making comparisons
to geologic slip rates.

00:36:11.020 --> 00:36:14.540
And because we have orientation
information from the four components

00:36:14.540 --> 00:36:19.360
of slip, we can compare each
geodetic slip rate to each geologic

00:36:19.360 --> 00:36:24.570
slip rate in exactly the orientation
of the geologic site location.

00:36:24.570 --> 00:36:28.710
So the figure on the left shows the
locations of geodetic slip rates from

00:36:28.710 --> 00:36:36.780
UCERF3, Appendix B, as filled circles
colored by their geologic slip rate.

00:36:36.780 --> 00:36:39.720
And then the corresponding
grid cell colored by the average

00:36:39.720 --> 00:36:45.250
geodetic slip rate in the
same orientation is shown behind it.

00:36:45.250 --> 00:36:49.800
And then we can compare
these slip rates graphically,

00:36:49.800 --> 00:36:52.900
so that’s what’s done on the right.
This is similar to the geologic/geodetic

00:36:52.910 --> 00:36:56.720
comparison that I showed in the eastern
California shear zone section where we

00:36:56.720 --> 00:37:02.610
have geologic slip rates on the X axis,
geodetic slip rates on the Y axis.

00:37:02.610 --> 00:37:07.410
The geodetic error bars, again,
are coming from UCERF3.

00:37:07.410 --> 00:37:09.750
Sorry – the geologic error bars
are coming from UCERF3.

00:37:09.750 --> 00:37:11.700
The geodetic error bars
are coming from the

00:37:11.700 --> 00:37:17.670
standard deviation in that
grid cell in that orientation.

00:37:17.670 --> 00:37:20.560
The dashed line – the gray
dashed line is a 1-to-1 line.

00:37:20.560 --> 00:37:23.200
So if we were perfectly agreeing –
if geologic and geodetic rates

00:37:23.200 --> 00:37:26.130
perfectly agreed,
they would all fall on that line.

00:37:26.130 --> 00:37:32.610
And the red line is a
weighted linear fit to these rates.

00:37:32.610 --> 00:37:34.830
And the weighted linear fit
gives us a slip rate that’s

00:37:34.830 --> 00:37:37.000
actually very,
very close to 1.

00:37:37.000 --> 00:37:43.760
So at least to first order, there isn’t
a strong – there isn’t an overwhelming

00:37:43.760 --> 00:37:47.720
systematic discrepancy
between these two slip rates.

00:37:48.840 --> 00:37:52.740
Of course, these results do
depend somewhat on grid size.

00:37:52.740 --> 00:38:00.920
And I – here I used the geologic slip
rates to help determine the best grid size.

00:38:02.510 --> 00:38:07.340
So I tested nine grid sizes – that’s
what’s shown here at the bottom –

00:38:07.340 --> 00:38:10.070
where I went through this
comparison with the geologic

00:38:10.070 --> 00:38:14.510
slip rates and identified the
grid size that produced

00:38:14.510 --> 00:38:16.740
the weighted linear fit
that’s closest to 1-to-1.

00:38:16.740 --> 00:38:21.420
So in other words, this grid size that
I am showing is the grid size in which

00:38:21.420 --> 00:38:25.630
the geodetic slip rates summarized
in each grid cell give us slip rates

00:38:25.630 --> 00:38:28.540
that are most consistent
with the geologic slip rates.

00:38:30.620 --> 00:38:35.220
Okay, another thing we
can think about is potency rate.

00:38:35.220 --> 00:38:38.330
Because I’m using the
horizontal components of slip,

00:38:38.330 --> 00:38:42.480
I’m talking about potency rate per unit
depth, or the geometric moment rate per

00:38:42.480 --> 00:38:48.790
unit depth, which is defined as the slip
rate on a fault segment times its length.

00:38:48.790 --> 00:38:52.360
So, in a given grid cell,
you just add that up.

00:38:52.360 --> 00:38:54.840
And this is nice because it’s
independent of orientation,

00:38:54.850 --> 00:38:58.430
and it sums over parallel fault strands,
which makes it more robust to

00:38:58.430 --> 00:39:02.970
small differences in representations
of the same fault system.

00:39:02.970 --> 00:39:05.230
And quantitatively –
or, sorry, qualitatively,

00:39:05.230 --> 00:39:09.430
it looks very similar to the slip rate and
uncertainty figures that we saw earlier.

00:39:09.430 --> 00:39:11.910
There’s high potency along the
San Andreas Fault system and

00:39:11.910 --> 00:39:16.120
high uncertainty – high-ish
uncertainty up near Mendocino.

00:39:18.220 --> 00:39:25.020
And then, if we add up the average
potency and assume a 10-kilometer

00:39:25.020 --> 00:39:31.200
locking depth and a shear modulus, we
can calculate the geodetic perspective –

00:39:31.200 --> 00:39:37.210
the average geodetic perspective on the
potency accumulation rate or – yeah,

00:39:37.210 --> 00:39:40.810
with assumptions – the moment
accumulation rate in California.

00:39:40.810 --> 00:39:45.510
And that comes out to about an
8.1 earthquake every 100 years.

00:39:45.510 --> 00:39:50.460
And I just very quickly looked at
what that – what has been released

00:39:50.460 --> 00:39:56.970
and added that up, and it’s –
so the geodetic potency accumulation

00:39:56.970 --> 00:40:01.290
rate is coming out to be
slightly higher than what has

00:40:01.290 --> 00:40:04.620
been released, but not –
it’s pretty comparable.

00:40:04.620 --> 00:40:09.880
And then the other thing
that’s interesting is that,

00:40:09.890 --> 00:40:16.790
of all of this potency accumulation,
95% of the total potency accumulation

00:40:16.790 --> 00:40:20.220
is occurring in about
half of the active area.

00:40:20.220 --> 00:40:25.700
And so what this means is that most of
the slip is happening on the fastest faults.

00:40:25.700 --> 00:40:27.600
And so that means we
probably aren’t missing

00:40:27.600 --> 00:40:35.390
any major source
of moment accumulation.

00:40:37.320 --> 00:40:40.420
Okay, so finally, given all of
this information about slip rates

00:40:40.430 --> 00:40:45.680
and uncertainty, we can generate a
summary fault model by projecting those

00:40:45.680 --> 00:40:51.310
four components of slip back onto a
fault geometry that we’re interested in.

00:40:51.310 --> 00:40:58.260
And so this gives us a strike-slip rate –
an opening rate on every fault in the cell.

00:40:58.260 --> 00:41:04.500
And if the fault that – the fault
geometry that we’re projecting onto

00:41:04.500 --> 00:41:07.760
isn’t exactly the geometry
used in the slip rate study,

00:41:07.760 --> 00:41:14.400
we also get a strike-slip – an opening rate
on perpendicular structures to that fault,

00:41:14.400 --> 00:41:18.550
which produce an estimate of what I’m
calling off-modeled fault deformation

00:41:18.550 --> 00:41:22.680
because it’s deformation that doesn’t
occur on this particular fault model.

00:41:22.680 --> 00:41:26.820
So what we’re doing – what I’m doing is 
projecting these four components of slip

00:41:26.820 --> 00:41:33.620
back onto a fault geometry. In this case,
I’m using the UCERF3 faults.

00:41:33.620 --> 00:41:37.700
And where it doesn’t line up – where the
rotation – where the projection isn’t

00:41:37.700 --> 00:41:46.040
perfect, you end up with some additional
slip on faults that aren’t in your grid cell.

00:41:47.040 --> 00:41:52.180
And so this gives us a community
average fault model that we’re –

00:41:52.180 --> 00:41:54.560
that we’re sort of
comfortable with.

00:41:54.560 --> 00:41:59.800
It’s a UCERF geometry that we’re –
we’ve seen a lot and are used to.

00:41:59.800 --> 00:42:03.260
And it gives us uncertainties
on each of those faults.

00:42:03.260 --> 00:42:07.830
And then these two slip components
on imaginary perpendicular structures

00:42:07.830 --> 00:42:12.100
produce an estimate of
off-model fault deformation.

00:42:13.390 --> 00:42:17.580
And so when we take the average
slip rates from doing that,

00:42:17.590 --> 00:42:22.680
what we get is a UCERF-style
estimate of strike-slip and

00:42:22.680 --> 00:42:26.270
opening rates on UCERF faults
that takes into account the expertise

00:42:26.270 --> 00:42:29.960
of the entire
geodetic community.

00:42:29.960 --> 00:42:32.320
And then standard
deviations produce uncertainties

00:42:32.320 --> 00:42:35.660
on that community geodetic
slip rate on every fault.

00:42:37.340 --> 00:42:42.900
Finally, the slip rates projected onto
those imaginary perpendicular structures,

00:42:42.900 --> 00:42:46.600
and in grid cells that don’t intersect
UCERF faults, produce a summary

00:42:46.600 --> 00:42:50.680
model of that off-modeled
fault deformation.

00:42:50.680 --> 00:42:55.430
And actually, 28% of the
total summary deformation

00:42:55.430 --> 00:42:59.360
does not project
onto UCERF faults.

00:43:00.620 --> 00:43:04.330
And that’s an interesting number
because the UCERF estimate of

00:43:04.330 --> 00:43:09.500
off-fault deformation was 30%,
and so this is very close to that.

00:43:09.500 --> 00:43:13.800
But this is surprising because
all of the deformation included here

00:43:13.800 --> 00:43:16.620
is deformation from these
geodetic slip rate models.

00:43:16.620 --> 00:43:20.640
So that means that all of this
deformation was attributed by someone

00:43:20.640 --> 00:43:26.560
to be on-fault, and here it’s just not on the
fault geometry that we’re interested in.

00:43:26.560 --> 00:43:30.990
And in fact, 75% of that
off-model fault deformation

00:43:30.990 --> 00:43:35.640
is occurring in grid cells that
overlap with UCERF faults.

00:43:35.640 --> 00:43:39.860
So this is suggestive that
what we often consider

00:43:39.860 --> 00:43:46.040
off-fault deformation might
often be a consequence of our

00:43:46.040 --> 00:43:50.040
epistemic uncertainty in
these geodetic slip rate models.

00:43:52.000 --> 00:43:56.100
And so that – and off-modeled fault
deformation, or off-fault deformation,

00:43:56.110 --> 00:44:00.000
does not necessarily imply true
distributed or plastic deformation,

00:44:00.000 --> 00:44:03.090
which is fundamentally important
in trying to understand

00:44:03.090 --> 00:44:06.850
sources of deformation
and hazard.

00:44:06.850 --> 00:44:12.690
And so, by summarizing and
combining this existing and ongoing

00:44:12.690 --> 00:44:16.530
geodetic research, we can produce a
summary model of geodetic slip rates

00:44:16.530 --> 00:44:21.270
and uncertainties for seismic hazard
models, and we get kind of a surprising

00:44:21.270 --> 00:44:25.620
and potentially provocative perspective
on off-fault deformation.

00:44:25.620 --> 00:44:29.220
So then, in these last couple minutes,
I’m going to really switch gears

00:44:29.220 --> 00:44:34.840
and talk about what I think is the
most exciting thing going on in terms of

00:44:34.840 --> 00:44:41.880
geodesy and characterizing U.S. seismic
hazard, and that is seafloor geodesy.

00:44:43.620 --> 00:44:48.870
And that’s because seafloor geodesy,
I think, can let us solve a problem.

00:44:48.870 --> 00:44:52.690
We can answer a specific yes-or-no
science question that has direct hazard

00:44:52.690 --> 00:44:56.120
implications, which is, is the
megathrust locked at the trench?

00:44:56.120 --> 00:44:59.940
Is the Cascadia Megathrust
locked all the way to the trench?

00:44:59.940 --> 00:45:06.770
And with geodesy, this is answerable.
But the only way – but geodesy is

00:45:06.770 --> 00:45:09.130
really the only way to do it
before there’s another earthquake.

00:45:09.130 --> 00:45:11.470
Because geodesy is the
only way to observe the rate

00:45:11.470 --> 00:45:16.840
of the strain accumulation
before an earthquake.

00:45:16.840 --> 00:45:23.430
And seafloor observations off of Peru –
so these are a couple of seafloor

00:45:23.430 --> 00:45:30.380
observations off of Peru have shown that
the subduction zone locking signal here –

00:45:30.380 --> 00:45:35.340
the red line would be the
prediction for what is fully locked.

00:45:35.340 --> 00:45:39.260
And the blue line is
locked within 2 kilometers.

00:45:41.380 --> 00:45:44.220
So in Peru, the subduction zone
is indistinguishable from

00:45:44.230 --> 00:45:46.740
fully locked all the
way to the trench.

00:45:46.740 --> 00:45:49.400
And observations off of Japan have
shown some really interesting

00:45:49.400 --> 00:45:54.440
temporally variable behavior,
but in Cascadia we just don’t know.

00:45:54.440 --> 00:45:59.010
And Dave Chadwell’s group at Scripps
is really pioneering this technology,

00:45:59.010 --> 00:46:00.210
and they’re actively
collecting data.

00:46:00.210 --> 00:46:05.140
They have a couple of
monuments out on the seafloor.

00:46:05.140 --> 00:46:09.300
But this isn’t – these measurements are
expensive, and they take a long time,

00:46:09.300 --> 00:46:15.860
and so this data isn’t published yet.
And it’s extremely sparse.

00:46:15.860 --> 00:46:20.560
But without them, we are blind
to the behavior at the trench.

00:46:20.560 --> 00:46:26.550
So Fred Pollitz put together this
compilation of estimates of locking

00:46:26.550 --> 00:46:30.010
on the Cascadia subduction zone.
And so it’s sort of a similar approach

00:46:30.010 --> 00:46:34.480
to what I took with California
slip rates where the four models

00:46:34.480 --> 00:46:41.710
on the left are four geodetic
estimates of locking on the

00:46:41.710 --> 00:46:44.460
subduction zone with
different modeling assumptions.

00:46:44.460 --> 00:46:48.690
And they all come up with very different
answers, especially near the trench.

00:46:48.690 --> 00:46:56.160
And then the figure on the right is the
standard deviation among those models.

00:46:56.160 --> 00:47:00.840
And the standard deviation
is highest near the trench,

00:47:00.840 --> 00:47:03.820
and especially where the trench
is farthest from the coast.

00:47:05.810 --> 00:47:10.260
And that’s exactly – so in terms of
understanding hazard, and especially

00:47:10.260 --> 00:47:13.940
tsunami hazard, that’s sort of exactly
where we want to have information.

00:47:13.940 --> 00:47:17.970
And, again, the only way to
do this is with – at least before

00:47:17.970 --> 00:47:22.160
there’s an earthquake,
is with geodesy.

00:47:22.160 --> 00:47:30.150
And so we – geodesy is –
seafloor geodesy, especially,

00:47:30.150 --> 00:47:34.730
is extremely expensive. And we
want to make sure we do it right.

00:47:34.730 --> 00:47:38.410
And then I’ve also shown – sorry –
at the top is this equation,

00:47:38.410 --> 00:47:44.170
which is just a reminder that the
locking rate, or the rate of

00:47:44.170 --> 00:47:48.820
strain accumulation on a
subduction zone or on a fault,

00:47:48.820 --> 00:47:53.360
is the difference between the overall
tectonic convergence and the creep rate.

00:47:53.360 --> 00:47:56.920
And so, in terms of understanding
hazard, we want to know the

00:47:56.920 --> 00:48:01.730
locking rate, but we also really want to
know the tectonic rate because the

00:48:01.730 --> 00:48:05.040
active locking rate is how fast
it’s accumulating strain rate now,

00:48:05.040 --> 00:48:07.030
and the tectonic rate
gives us sort of an upper bound

00:48:07.030 --> 00:48:10.080
on what could have been
accumulating in the past.

00:48:11.180 --> 00:48:17.620
Great. So in terms of wanting to do this,
the best-case scenario, why we should

00:48:17.620 --> 00:48:22.160
do seafloor geodesy and where we
should put them, I’ve been working

00:48:22.160 --> 00:48:24.910
with Sarah Minson to identify
optimal locations for

00:48:24.910 --> 00:48:28.560
seafloor observations to
answer exactly this problem.

00:48:30.080 --> 00:48:34.400
And this is preliminary.

00:48:34.400 --> 00:48:39.460
But the idea is that, in order to
resolve locking, we want to estimate

00:48:39.460 --> 00:48:43.560
the strain accumulation rate
on the subduction zone.

00:48:43.560 --> 00:48:46.500
And so we would want to
do this in a block model

00:48:46.500 --> 00:48:49.740
that has the Cascadia subduction
zone embedded in it.

00:48:49.740 --> 00:48:55.440
And then we can calculate a quantity, H,
which is information entropy.

00:48:55.440 --> 00:49:00.010
If you want to find out – I’m not going
to talk about entropy beyond the fact

00:49:00.010 --> 00:49:03.140
that it gives us some measure of
the information in our problem.

00:49:03.140 --> 00:49:07.200
You can talk to me or
Sarah or Jess about it afterwards.

00:49:07.200 --> 00:49:11.480
So this is a measure of
unpredictability or a measure

00:49:11.480 --> 00:49:15.080
of average information content.
And so what’s showing –

00:49:15.080 --> 00:49:20.280
what I’m showing on the left
is the change in entropy

00:49:20.280 --> 00:49:26.300
by adding a single GPS
observation at any location.

00:49:26.300 --> 00:49:29.480
And so it’s color-coded by the
change in entropy or the

00:49:29.480 --> 00:49:35.550
information gain by adding
one station at any of those locations.

00:49:35.550 --> 00:49:40.440
And so the first thing that
jumps out is that the

00:49:40.440 --> 00:49:48.090
greatest information gain
occurs on the Juan de Fuca Plate.

00:49:48.090 --> 00:49:51.120
And that’s because we
have no geodetic observations.

00:49:51.120 --> 00:49:55.560
There are no islands out there,
so we don’t know geodetically

00:49:55.560 --> 00:49:59.010
how fast this tectonic
convergence rate is.

00:49:59.010 --> 00:50:02.420
We think it’s probably pretty
close to what’s been estimated

00:50:02.420 --> 00:50:09.281
from magnetic reversals, but we don’t
have a geodetic measurement of it,

00:50:09.281 --> 00:50:11.580
and so the entire – anywhere on the
Juan de Fuca Plate would be

00:50:11.580 --> 00:50:18.080
really great to put a seafloor station
is what this figure is showing.

00:50:18.080 --> 00:50:22.380
And the minimum of that –
sort of uniform minimum –

00:50:22.380 --> 00:50:28.760
is shown by the star –
the little white star.

00:50:28.760 --> 00:50:33.320
And then, if we put –
so if that station were to exist,

00:50:33.320 --> 00:50:35.190
where would we put
the second station?

00:50:35.190 --> 00:50:38.961
So if that station exists,
now our information gain

00:50:38.961 --> 00:50:44.680
for adding a GPS station on
the Juan de Fuca Plate goes down,

00:50:44.680 --> 00:50:48.340
especially right near that station,
and the next best place to put it

00:50:48.340 --> 00:50:53.440
is at the trench, really
farthest from the coast.

00:50:53.440 --> 00:50:56.100
And then, if we want to
keep adding them ...

00:50:58.740 --> 00:51:06.300
I went up to 39 because it takes a
long time, and I got bored. [laughter]

00:51:06.310 --> 00:51:10.040
So the – what’s really interesting
about this is that the first onshore

00:51:10.040 --> 00:51:15.680
location doesn’t show up
until there are 23 offshore.

00:51:15.680 --> 00:51:20.010
And so to look at that – to look more
quantitatively at the relative information

00:51:20.010 --> 00:51:23.910
gain, or decrease in entropy, with each
additional station is this figure.

00:51:23.910 --> 00:51:26.720
So we have the number of
additional stations on the X axis,

00:51:26.720 --> 00:51:30.990
the change in entropy
on the Y axis.

00:51:30.990 --> 00:51:34.030
The stations that have been
added onshore are shown in green.

00:51:34.030 --> 00:51:38.500
And then we can compare this
to what this figure would look like

00:51:38.500 --> 00:51:41.340
if we were to only
add onshore stations.

00:51:42.360 --> 00:51:46.300
Oh – oh, geez. Sorry.
There it is.

00:51:46.300 --> 00:51:49.180
If we were to only
add onshore stations.

00:51:49.180 --> 00:51:53.460
And you can see that no quantity
of onshore stations is going to

00:51:53.460 --> 00:51:57.460
provide the information gain
of a single offshore observation.

00:51:57.460 --> 00:52:01.660
So, again, this is preliminary, but this
approach provides a quantitative way to

00:52:01.660 --> 00:52:08.740
describe the value in offshore geodesy,
which is really exciting and cool.

00:52:12.880 --> 00:52:18.880
So finally, to wrap up, my main point
is that geodesy is uniquely capable of

00:52:18.880 --> 00:52:23.000
observing strain accumulation rates
and anticipating earthquake hazards.

00:52:23.000 --> 00:52:27.520
And in the eastern California shear zone,
we can find a block model that fits

00:52:27.520 --> 00:52:33.750
GPS velocities and is most
consistent with geologic slip rates,

00:52:33.750 --> 00:52:38.590
but there are still persistent
discrepancies on specific faults.

00:52:38.590 --> 00:52:42.510
By combining – by leveraging existing
research for seismic hazard models,

00:52:42.510 --> 00:52:49.020
we can summarize the geodetic
perspective on geodetic fault slip rates.

00:52:49.020 --> 00:52:54.180
And what we think as – what we think
of as off-fault deformation may

00:52:54.190 --> 00:52:58.730
in fact be, at least in some cases,
a product of epistemic uncertainty.

00:52:58.730 --> 00:53:06.940
And seafloor geodesy gives us
information that just we cannot get

00:53:06.940 --> 00:53:09.960
with onshore stations.
Thank you.

00:53:09.960 --> 00:53:16.820
[ Applause ]

00:53:21.300 --> 00:53:23.100
- Questions?

00:53:30.460 --> 00:53:33.880
- I have a question way back
on your first or second slide.

00:53:33.890 --> 00:53:35.690
- Sure.
- But you showed a photographer

00:53:35.690 --> 00:53:39.860
taking a picture, and then you
showed how it was fuzzy.

00:53:39.860 --> 00:53:43.440
And we were able to see that you
really had sharpened the image there.

00:53:43.440 --> 00:53:46.980
I suppose it was a maximum
entropy or something like that.

00:53:46.980 --> 00:53:52.420
Why can’t you use that same
criterion on your fault system?

00:53:52.430 --> 00:53:55.300
Isn’t there some way that you
select which is the best

00:53:55.300 --> 00:53:59.890
of your 50-odd [inaudible]?

00:53:59.890 --> 00:54:03.310
- So you’re asking …
- Right here, I can, by eyes,

00:54:03.310 --> 00:54:06.900
you’ve – that you’ve been successful.
- Yes.

00:54:06.900 --> 00:54:10.640
- By your total variation regularization.
- Yes.

00:54:10.640 --> 00:54:14.700
- And I suppose there’s a
quantitative measure of that too.

00:54:14.700 --> 00:54:18.660
- Oh, of the sharpness?
- Yes.

00:54:18.660 --> 00:54:21.940
- That’s a good – that is an
interesting question. Probably.

00:54:21.940 --> 00:54:27.960
I don’t know what that would be.
But I’m sure that that’s calculatable.

00:54:27.960 --> 00:54:31.900
- But then could you use that –
could you use that in the fault system?

00:54:31.900 --> 00:54:35.650
- Right. So – right, so we go
through this with the fault system,

00:54:35.650 --> 00:54:41.880
but we just quantify it with our
fit to the GPS data rather than –

00:54:41.880 --> 00:54:47.340
rather than an entropy or a – or a –
does that make sense?

00:54:47.340 --> 00:54:52.000
- I don’t mean to labor this, but yes,
you find the standard deviations

00:54:52.000 --> 00:54:56.560
decrease the more stations you put in,
the more blocks you put in.

00:54:57.980 --> 00:55:01.890
So then wouldn’t you say that the
best solution is the most blocks?

00:55:01.890 --> 00:55:07.360
Or do you have another
way of saying that you’re –

00:55:07.360 --> 00:55:12.560
they’re no longer significant –
increasing the number of blocks – well.

00:55:12.560 --> 00:55:14.960
- Right. So that’s why
we used the geology.

00:55:14.960 --> 00:55:20.070
That’s why we used the geologic rates as
a test for where we had reached what we

00:55:20.070 --> 00:55:24.230
thought was an appropriate level of –
appropriate number of blocks.

00:55:24.230 --> 00:55:28.890
Alternatively, one could also say,
I don’t believe my GPS observations

00:55:28.890 --> 00:55:33.600
better than some uncertainty –
1.5 millimeters per year, perhaps –

00:55:33.600 --> 00:55:36.980
and use that as the threshold.
- Thank you.

00:55:39.480 --> 00:55:41.940
- I think Ben?

00:55:41.940 --> 00:55:44.300
- Thanks, Eileen.
On behalf of the International

00:55:44.300 --> 00:55:47.870
Association of Seafloor Geodesists,
we would like to thank you for

00:55:47.870 --> 00:55:52.890
your presentation and endorse it.
[laughter]

00:55:52.890 --> 00:55:56.800
But I actually have two questions about
the first two halves, or two thirds.

00:55:56.800 --> 00:55:58.140
- Sure.

00:55:58.140 --> 00:56:01.900
- [laughs] For the Calico Fault,
you showed the – I forget which

00:56:01.910 --> 00:56:04.930
color it was in
the cool color plot.

00:56:04.930 --> 00:56:10.000
How representative do you think
the geologic uncertainty is in that?

00:56:10.000 --> 00:56:13.850
It’s extremely difficult
to get geologic rates.

00:56:13.850 --> 00:56:17.640
You know, how about
along-strike variation, et cetera?

00:56:17.640 --> 00:56:23.850
- Right. So the short answer is that the
geologic slip rates and their uncertainties

00:56:23.850 --> 00:56:29.510
is not my area of expertise, so I just took
them from their respective publications.

00:56:29.510 --> 00:56:36.550
These rates are taken at exactly
the location – so I take the geodetic rate

00:56:36.550 --> 00:56:43.510
at exactly the location on the fault
where the geologic slip rate was taken.

00:56:43.510 --> 00:56:45.829
So that, hopefully,
gets around some of that.

00:56:45.829 --> 00:56:48.180
- Okay. Thanks. Yeah, I didn’t –
I didn’t catch that the first – okay, cool.

00:56:48.180 --> 00:56:54.560
And then – so then the other is you said,
okay, so the standard deviation,

00:56:54.570 --> 00:57:00.220
or the representation of uncertainty
is the sort of summary of our –

00:57:00.220 --> 00:57:04.760
the geodetic community’s
expertise at the moment.

00:57:04.760 --> 00:57:08.820
How about the geodetic
community’s lack of expertise?

00:57:08.820 --> 00:57:12.440
Certainly there are new models
that’ll arise or – do you expect

00:57:12.440 --> 00:57:17.600
those uncertainties to go down or up?
[laughter]

00:57:19.280 --> 00:57:23.080
- I don’t know. So – right,
so I did mention that,

00:57:23.090 --> 00:57:26.470
in the places where the standard
deviations were quite low was

00:57:26.470 --> 00:57:29.720
generally where one group had worked
on a region where nobody else was.

00:57:29.720 --> 00:57:31.140
- Right. [laughs]

00:57:31.140 --> 00:57:36.320
- So in that case, I think if we have
more people working in more parts

00:57:36.320 --> 00:57:41.240
of California, I would expect the
standard deviations to go up.

00:57:43.340 --> 00:57:49.500
- But even – but that’s not – that’s just 
more current expertise models, not …

00:57:49.500 --> 00:57:51.840
- Right.
- … new models that might arise.

00:57:51.840 --> 00:57:54.330
- Right. Yeah.
So that’s a harder problem.

00:57:54.330 --> 00:57:56.950
And so what’s nice about doing this
in California is that we do have

00:57:56.950 --> 00:58:02.720
so many models that this is
a more reasonable thing to do here

00:58:02.720 --> 00:58:05.430
than it might be in
a lot of other places.

00:58:05.430 --> 00:58:12.320
And so certainly it’s not perfect.
Yeah.

00:58:16.240 --> 00:58:19.880
- And that was an awesome talk.
So as someone who only can measure

00:58:19.880 --> 00:58:23.020
the rainfall [laughter], I have two
maybe really naive questions.

00:58:23.020 --> 00:58:27.050
So the first is on this slide.
The geologic rates are some –

00:58:27.050 --> 00:58:29.480
obviously are averaged
over some longer time.

00:58:29.480 --> 00:58:33.540
And what’s to say that the geodetic rate
is not measuring what’s happening right

00:58:33.540 --> 00:58:38.470
now, and so that’s not a discrepancy,
there’s just a time coefficient in there?

00:58:38.470 --> 00:58:41.090
- There is nothing not to say that.
- Okay. [laughter]

00:58:41.090 --> 00:58:44.460
Okay, so maybe my second naive
question is, when you’re comparing the

00:58:44.460 --> 00:58:48.730
UCERF, and you get 28% off-modeled
fault, how does that compare with the

00:58:48.730 --> 00:58:51.540
background rate that UCERF has?
Is the – could that be

00:58:51.540 --> 00:58:53.500
thought of as the same thing?
Or is that different?

00:58:53.500 --> 00:58:55.750
- Oh, that’s a – that’s a really
good question, and I think it

00:58:55.750 --> 00:58:58.870
should be considered much more
analogous to a background rate.

00:58:58.870 --> 00:59:04.320
Because that’s still potentially –
because that’s a summary of

00:59:04.320 --> 00:59:09.950
deformation that somebody considered
on-fault, it’s all potentially seismogenic.

00:59:11.240 --> 00:59:13.800
It’s just not exactly on the
fault that we thought it was on.

00:59:13.800 --> 00:59:16.119
So I think the background rate –
I haven’t looked at it, but that …

00:59:16.119 --> 00:59:18.840
- Okay. So maybe there’s some
estimate of the background rate

00:59:18.840 --> 00:59:23.100
in comparison to the fault rate on –
in UCERF that you could compare?

00:59:23.100 --> 00:59:25.400
- Yes.
- Okay. Thanks.

00:59:26.500 --> 00:59:30.620
- So that was a great talk, Eillen.
I had two questions.

00:59:30.630 --> 00:59:37.120
One is sort of along the same lines
as the previous question, which is,

00:59:37.120 --> 00:59:40.080
on that discussion of
off-fault deformation,

00:59:40.080 --> 00:59:42.790
if I – I want to make sure I
understand your procedure correctly.

00:59:42.790 --> 00:59:48.460
Because in each of those grid cells,
that cell only gets a slip rate if faults

00:59:48.460 --> 00:59:50.860
in somebody’s model
went through that cell, correct?

00:59:50.860 --> 00:59:55.421
So you had made a comment that
most of this off-modeled fault

00:59:55.421 --> 00:59:59.770
deformation was adjacent to faults,
but it kind of has to be that way, right?

00:59:59.770 --> 01:00:01.790
Because there would
be nothing in that cell …

01:00:01.790 --> 01:00:08.180
- Right. Yeah. Right. So what I meant
is that it’s adjacent to UCERF faults.

01:00:08.180 --> 01:00:10.780
- Okay.
- And so …

01:00:13.720 --> 01:00:17.360
So there are some
places where somebody …

01:00:20.420 --> 01:00:25.660
Somebody, for example, modeled
a fault here, but the UCERF fault

01:00:25.660 --> 01:00:27.160
doesn’t go through it.
- Mm-hmm.

01:00:27.160 --> 01:00:29.140
- And so …

01:00:31.700 --> 01:00:39.260
But – or, like, around in here.
But most of this deformation

01:00:39.260 --> 01:00:43.220
is occurring in these same grid cells
with UCERF, just not exactly

01:00:43.220 --> 01:00:47.390
in the same orientation.
- Mm-hmm. So, but indeed,

01:00:47.390 --> 01:00:52.120
the true deformation could be occurring
on a fault that’s parallel to what

01:00:52.120 --> 01:00:55.160
someone called a – well, what was
determined to be a UCERF fault.

01:00:55.160 --> 01:00:59.570
I mean, if it’s in that grid cell, it’s,
like, it’s on-fault if it’s parallel

01:00:59.570 --> 01:01:01.910
to the UCERF fault
that comes through there.

01:01:01.910 --> 01:01:05.310
- It’s on-fault if it is parallel, yes.
- I see. Uh-huh.

01:01:05.310 --> 01:01:12.720
So I guess my broader question is,
in your – for example, in the first part

01:01:12.720 --> 01:01:18.040
of your talk where you looked at the
effect of different configurations

01:01:18.040 --> 01:01:22.480
of faults and sizes of
blocks and things like that,

01:01:22.480 --> 01:01:25.980
that was looking at
one aspect of the problem.

01:01:25.980 --> 01:01:30.720
But I’m curious what your thoughts are
on the appropriateness of elastic block

01:01:30.720 --> 01:01:33.580
modeling versus other kinds of block
modeling that might include

01:01:33.580 --> 01:01:38.240
viscoelastic effects and the best ways
to incorporate dipping faults

01:01:38.250 --> 01:01:42.420
into these kind of models.
Because, you know, if you were

01:01:42.420 --> 01:01:47.740
coming from the perspective of,
I want to do a new model for Region X,

01:01:47.740 --> 01:01:51.320
you know, what do you
think is the best way to do it?

01:01:52.780 --> 01:01:59.450
- I think that – so in terms of
putting together a compilation

01:01:59.450 --> 01:02:10.180
of research that people are doing,
I want – ideally, I want a number of

01:02:10.180 --> 01:02:17.030
different people making different
assumptions that they think are correct.

01:02:17.030 --> 01:02:22.530
So I want somebody to think
that a viscoelastic block model is the

01:02:22.530 --> 01:02:26.640
right way to do it and to build one.
And I want somebody to think

01:02:26.640 --> 01:02:31.150
that a finite element model is the
right way to do it and to build one.

01:02:32.060 --> 01:02:36.400
Assuming that all of those are,
like, appropriate – capture, sort of,

01:02:36.400 --> 01:02:40.180
our uncertainty and
exactly how to do the problem.

01:02:40.180 --> 01:02:46.380
- And if you were one of
those people? [laughter]

01:02:46.900 --> 01:02:49.160
- I would do – I would do
an elastic block model

01:02:49.160 --> 01:02:51.980
with a lot of blocks
and then group them.

01:02:57.920 --> 01:03:01.360
- Thanks for that talk.
Can we go to your last slide?

01:03:01.370 --> 01:03:06.030
I would change your – suggest
changing your conclusion a little bit.

01:03:06.030 --> 01:03:09.660
- Oh. [laughter]
That’s fine. This conclusion?

01:03:09.660 --> 01:03:17.580
- Yeah. I would say geodesy, when
interpreted along with paleoseismology,

01:03:17.580 --> 01:03:21.770
is uniquely capable of observing strain
and anticipating earthquake hazards.

01:03:21.770 --> 01:03:25.970
In New Madrid, you know,
Tish Tuttle has given us a map

01:03:25.970 --> 01:03:30.270
of the number of earthquakes
over the last 2,000 years

01:03:30.270 --> 01:03:35.020
and recurrence intervals,
and the geodesy signal was very low.

01:03:35.020 --> 01:03:37.720
In Alaska, George Plafker
has a recurrence interval

01:03:37.720 --> 01:03:41.090
of 810 years for
great earthquakes.

01:03:41.090 --> 01:03:46.160
So I think the geodesy is best interpreted
in conjunction with paleoseismology.

01:03:46.160 --> 01:03:51.440
- Yes. The point of this bullet is that
what geodesy contributes to observing

01:03:51.440 --> 01:03:55.510
strain accumulation and
earthquake hazards is unique.

01:03:55.510 --> 01:03:59.980
What geodesy contributes, no other
technology, at this point, can contribute.

01:04:02.440 --> 01:04:07.060
- Okay. I read that sentence
differently than you would. [laughter]

01:04:09.100 --> 01:04:14.660
- My question is very similar to Jess’.
Let me be sure that

01:04:14.670 --> 01:04:18.880
I understand your conclusions.
When you talk about off-fault

01:04:18.880 --> 01:04:24.960
deformation in the context of what
you’ve done, it’s very near off-fault.

01:04:24.960 --> 01:04:27.920
- Yes.
- And so that relates to the issue

01:04:27.920 --> 01:04:33.610
of background seismicity,
for example, which may be

01:04:33.610 --> 01:04:42.460
smoothed and spread over a wider
region than your results suggest.

01:04:43.830 --> 01:04:46.800
Is that right?
- Yes. I think so.

01:04:46.800 --> 01:04:49.680
- Which one is better? [laughs]

01:04:50.880 --> 01:04:53.200
- Off-fault seismicity or the –
or, you’re asking …

01:04:53.200 --> 01:04:54.580
- Yeah.
- Between them?

01:04:54.580 --> 01:04:57.810
Well, I think it would be really
interesting to do the comparison.

01:04:57.810 --> 01:05:00.890
And without doing the comparison,
I don’t feel comfortable

01:05:00.890 --> 01:05:04.520
answering that question.
- You’re very careful and circumspect.

01:05:04.520 --> 01:05:08.250
[laughter]
Which is a good thing. Thanks.

01:05:08.250 --> 01:05:13.440
- So my question had to do with,
how did you determine where you

01:05:13.440 --> 01:05:18.700
were going to drape your grid cells?
You had spoken a bit to how you

01:05:18.700 --> 01:05:21.930
determined the optimal grid size,
which makes sense to me.

01:05:21.930 --> 01:05:28.310
But it seems to me that where the grid
was centered would kind of determine,

01:05:28.310 --> 01:05:31.620
or change, the net deformation
that would be in that spatial location.

01:05:31.620 --> 01:05:34.490
So I don’t know. How much
do you think where the grid

01:05:34.490 --> 01:05:37.280
was draped would
affect your results?

01:05:37.280 --> 01:05:40.460
- It would. It does. Yeah.
I didn’t show it, and I was

01:05:40.460 --> 01:05:43.140
hoping nobody would ask about it.
[laughter]

01:05:43.140 --> 01:05:44.100
- Sorry.

01:05:44.100 --> 01:05:48.120
- Actually, so I did the same procedure
where I – I’ll go back to the slide.

01:05:57.140 --> 01:05:59.640
So I – so what I had done –
what I showed is that I

01:05:59.640 --> 01:06:01.619
tested different
grid sizes.

01:06:01.619 --> 01:06:07.890
I also went through the same thing for
each grid size, scooting the grid around.

01:06:07.890 --> 01:06:12.240
And this 0.4 –
approximately 0.4-degree grid size

01:06:12.240 --> 01:06:15.880
came out to be the
best fit in most of them.

01:06:15.880 --> 01:06:20.500
It’s a little less clean of a picture,
but it’s not inconsistent.

01:06:23.380 --> 01:06:29.260
[ Silence ]

01:06:29.960 --> 01:06:35.280
- Yeah. This will be a couple about
geologic issues while we’re on this slide.

01:06:35.290 --> 01:06:40.060
First of all, to go back to one of your
very early slides, when it comes to

01:06:40.060 --> 01:06:43.660
paleoseismology and
faults and geodesy,

01:06:43.660 --> 01:06:49.360
geodesy does not see faults
very well, as you illustrated to us.

01:06:49.360 --> 01:06:54.620
And so you’ve got to know where the
faults are in order to make a go of it.

01:06:54.620 --> 01:07:01.260
So, yeah, so always be sound
reasons for doing earthquake geology.

01:07:01.260 --> 01:07:09.119
The second question is related to that.
When you went to the east California

01:07:09.119 --> 01:07:18.540
shear zone and did the regularization
and eliminated these faults, is being

01:07:18.540 --> 01:07:23.860
important, but it does raise the question
of, well, why the hell is the fault there?

01:07:23.860 --> 01:07:28.720
And it may be much older, or this,
or that, or one thing or another,

01:07:28.720 --> 01:07:33.880
but it does kind of invite some questions
as to, if it’s unimportant from a

01:07:33.880 --> 01:07:39.220
geodetic point of view, why is it
there from a geologic point of view?

01:07:39.220 --> 01:07:45.200
And then, this picture –
I’m not sure that I’m getting all of it,

01:07:45.200 --> 01:07:50.390
but it looks like the uncertainty –
the geologic uncertainties

01:07:50.390 --> 01:07:55.260
are increasing with the
size of the geologic slip rate.

01:07:55.260 --> 01:07:58.040
And that’s not a really a
question for you, but that’s a

01:07:58.059 --> 01:08:01.190
question for the geologists.
Why is that so?

01:08:01.190 --> 01:08:06.740
I would have thought that the –
that the largest geologic slip rates

01:08:06.740 --> 01:08:08.430
would have been those faults
that had been the most –

01:08:08.430 --> 01:08:13.040
the best-studied, so why are
the uncertainties the largest?

01:08:13.960 --> 01:08:15.840
And I’m sorry to ask you that.
[laughter]

01:08:15.859 --> 01:08:18.460
- I agree. I don’t know
the answer to that.

01:08:18.460 --> 01:08:24.260
- Well, we’ll – Belle does. Okay.
- I’ll have to think about it. [laughter]

01:08:24.260 --> 01:08:27.210
- So I’ll preface this by saying
I totally agree with your point

01:08:27.210 --> 01:08:30.900
on seafloor geodesy.
But it seems like your map –

01:08:30.900 --> 01:08:34.279
if you could normalize the
information gained by cost of

01:08:34.280 --> 01:08:37.820
doing business, do you
get the same answer?

01:08:38.940 --> 01:08:42.040
- Yeah, well, so – right.
And that is sort of where we intend

01:08:42.049 --> 01:08:49.250
to go to put that in terms of dollar
signs rather than information gain.

01:08:49.250 --> 01:08:51.880
Because dollar signs mean something,
and information gain doesn’t really.

01:08:51.880 --> 01:08:55.380
[laughter]
So exactly.

01:08:56.580 --> 01:08:59.680
Right, but you did – so, like,
the one point that I will make is that,

01:08:59.690 --> 01:09:01.460
no matter how many –
you could sort of see that,

01:09:01.460 --> 01:09:05.560
no matter how many of those
onshore stations you added, you’re still

01:09:05.560 --> 01:09:12.900
not going to get the information
gain of one offshore location.

01:09:12.900 --> 01:09:18.520
And so that’s sort of infinite cost
for infinite number of onshore stations

01:09:18.520 --> 01:09:22.420
still isn’t going to get you what
you would get for hundreds of –

01:09:22.420 --> 01:09:26.040
millions of dollars for an offshore.
I don’t know how much they cost.

01:09:26.040 --> 01:09:29.710
- Well, you can – back of the envelope,
you can say, well, what’s the cost of

01:09:29.710 --> 01:09:33.250
the continuous GPS stations that
have been installed in Cascadia.

01:09:33.250 --> 01:09:38.040
And I feel pretty confident to say
it’s more expensive than one

01:09:38.040 --> 01:09:43.480
seafloor geodetic observation.
So throw down mic. [laughter]

01:09:46.160 --> 01:09:50.180
[ Silence ]

01:09:51.220 --> 01:09:54.400
- Any other pressing inquiries?

01:09:56.080 --> 01:09:59.360
Okay, thanks, everyone,
for your questions, and we’re

01:09:59.360 --> 01:10:02.660
going to meet to take Eileen
to lunch inside the doors of

01:10:02.660 --> 01:10:08.940
Building 3 in, let’s say, 10 minutes.
Thanks, everyone.

01:10:08.940 --> 01:10:14.480
[ Applause ]

01:10:15.960 --> 01:10:24.860
[ Silence ]