WEBVTT
Kind: captions
Language: en-US

00:00:00.000 --> 00:00:03.040
Good morning. I’d like to thank you
for inviting me to give a talk

00:00:03.040 --> 00:00:05.600
at the Northern California
Earthquake Hazards Workshop.

00:00:05.600 --> 00:00:08.320
Today I’d like to talk about
moment tensors, state of stress,

00:00:08.320 --> 00:00:11.280
finite-source scaling, and
characterizing a pseudo fracture

00:00:11.280 --> 00:00:14.616
network from microearthquakes
at the Geysers, California.

00:00:14.640 --> 00:00:18.000
This is work that I’ve done with
Ronald Gritto from Array Information

00:00:18.000 --> 00:00:21.600
Technology and Sierra Boyd and
Taka’aki Taira from the Berkeley

00:00:21.600 --> 00:00:25.896
Seismological Laboratory
over the last several years.

00:00:25.920 --> 00:00:29.840
What I’m going to focus on today
is the seismicity swarm in the

00:00:29.840 --> 00:00:33.360
Northwest Geysers Enhanced
Geothermal System Experiment.

00:00:33.360 --> 00:00:37.520
This red outline is showing the
northwestern extent of the Geysers

00:00:37.520 --> 00:00:42.320
geothermal field. And this
1-kilometer-by-2-kilometer rectangle

00:00:42.320 --> 00:00:45.360
shows the study area
for today’s talk.

00:00:45.360 --> 00:00:51.096
And this study area is centered
on the Prati-32/Prati-31 wells.

00:00:51.120 --> 00:00:54.400
The purpose of the enhanced
geothermal system experiment was to

00:00:54.400 --> 00:00:59.520
deepen the Prati State 31 and Prati-32
wells into the high-temperature

00:00:59.520 --> 00:01:07.896
reservoir and introduce water – inject
water and stimulate fracture generation.

00:01:07.920 --> 00:01:13.360
This plot is showing the average
monthly water injection rate

00:01:13.360 --> 00:01:18.136
beginning on October 6th, 2011 –
the blue curve.

00:01:18.160 --> 00:01:21.280
And the asterisks are showing
a very strong correlation with

00:01:21.280 --> 00:01:26.320
monthly seismicity rates
within our study area.

00:01:26.320 --> 00:01:31.200
So I’d like to give an overview
of our seismic source analysis

00:01:31.200 --> 00:01:34.456
at the Prati-32 EGS
injection site.

00:01:34.480 --> 00:01:37.840
Give some examples of seismic
moment tensor analysis to obtain

00:01:37.840 --> 00:01:42.696
waveform-constrained estimates
of Mw and fault plane orientations.

00:01:42.720 --> 00:01:46.240
We use that information to
invert for the stress tensor

00:01:46.240 --> 00:01:50.936
to investigate temporal variation
in stress during injection.

00:01:50.960 --> 00:01:54.480
We’ve done some finite-source
analysis to determine the causative

00:01:54.480 --> 00:01:57.680
plane and the scaling of rupture
parameters such as average slip,

00:01:57.680 --> 00:01:59.976
rupture area,
and stress drop.

00:02:00.000 --> 00:02:03.520
And I’d like to finish up by showing
how you can integrate all of this

00:02:03.520 --> 00:02:08.240
information to characterize seismicity
in terms of a pseudo fracture network,

00:02:08.240 --> 00:02:12.560
which is illustrated here, where,
instead of a distribution of dots

00:02:12.560 --> 00:02:18.080
in a box, we can actually start to
think about the dimensions of the

00:02:18.080 --> 00:02:22.400
microearthquakes in the study area
and the degree of interconnectivity

00:02:22.400 --> 00:02:26.376
of the – of the
stimulated fractures.

00:02:26.400 --> 00:02:29.760
So, for the moment tensor analysis,
we used instrument-corrected

00:02:29.760 --> 00:02:34.880
waveforms filtered between 0.7 to
1.7 hertz. We used a grid search

00:02:34.880 --> 00:02:41.176
to optimally align the waveforms
with the Green’s functions.

00:02:41.200 --> 00:02:44.480
And, for some of the events,
we perform a joint waveform

00:02:44.480 --> 00:02:49.280
first-motion source-type inversion
based on the work of Sean Ford et al.

00:02:49.280 --> 00:02:55.520
and Avinash Nayak, in which we
search for the optimal moment tensor

00:02:55.520 --> 00:02:58.640
solution in the source-type space.
So this takes into the account the

00:02:58.640 --> 00:03:01.760
full moment tensor,
including volumetric terms.

00:03:01.760 --> 00:03:06.296
We’ve analyzed 160 events
inside the 1-by-2-kilometer

00:03:06.320 --> 00:03:11.656
study area, and the smallest event
was moment magnitude 0.6.

00:03:11.680 --> 00:03:15.280
So this shows the results,
spanning approximately

00:03:15.280 --> 00:03:19.280
a five-year period of time.
So about a year, year and a half,

00:03:19.280 --> 00:03:25.040
before the stimulation experiment began,
we collected – well, we collected

00:03:25.040 --> 00:03:29.920
about 20, 25 moment tensor solutions.
The rate in the study area was

00:03:29.920 --> 00:03:34.880
very low prior to injection.
The red curve is showing the

00:03:34.880 --> 00:03:39.760
injection rate in gallons per minute.
And you can see, in late 2011 –

00:03:39.760 --> 00:03:44.240
in October 2011, there’s a marked
increase in the numbers of earthquakes

00:03:44.240 --> 00:03:46.696
and – as we saw in
the previous slide.

00:03:46.720 --> 00:03:50.616
And then also the magnitudes
of these microearthquakes.

00:03:50.640 --> 00:03:54.616
The color scaling is showing
the percent CLVD.

00:03:54.640 --> 00:03:58.240
So there are non-double-coupled
components as well as double-coupled

00:03:58.240 --> 00:04:02.160
moment tensor solutions.
And we see a variety of mechanisms

00:04:02.160 --> 00:04:05.600
which are, you know, more or less
distributed between strike-slip

00:04:05.600 --> 00:04:08.560
and normal-type faulting.
In fact, when we look at the distribution

00:04:08.560 --> 00:04:13.920
with source depth, this red bar is
showing the interval over which the

00:04:13.920 --> 00:04:18.080
well casing is perforated, allowing
water to enter the surrounding rock.

00:04:18.080 --> 00:04:23.200
The seismicity is peaked at this depth
range. And you can see that there’s a –

00:04:23.200 --> 00:04:28.240
roughly a – almost a 50/50 mix between
strike-slip and normal types of events.

00:04:28.240 --> 00:04:31.680
The brackets down here are showing
overlapping time windows that we

00:04:31.680 --> 00:04:36.560
used to sample the moment tensor
catalog for stress inversion.

00:04:36.560 --> 00:04:43.840
We used a stress inverse methodology
of Vaclav Vavryčuk 2014, and this

00:04:43.840 --> 00:04:47.040
shows the results as a function
of time for each of those

00:04:47.040 --> 00:04:50.000
staggered time windows.
So pre-injection,

00:04:50.000 --> 00:04:53.200
we find the maximum
compressive stress, red,

00:04:53.200 --> 00:04:57.096
and the minimum compressive
stress, blue, are in the horizontal.

00:04:57.120 --> 00:05:00.720
So consistent with
a strike-slip regime.

00:05:00.720 --> 00:05:04.080
And then, during the injection,
this remains the case, but you can

00:05:04.080 --> 00:05:09.520
see evidence of a slight rotation
of the minimum compressive

00:05:09.520 --> 00:05:13.200
stress orientation.
So it’s almost east-west.

00:05:13.200 --> 00:05:20.560
And, mid-2012, injection was halted,
and there was a dramatic rotation

00:05:20.560 --> 00:05:25.360
in the stress direction.
So, for example, the maximum

00:05:25.360 --> 00:05:31.256
compressive stress, red, is horizontal.
It rotates 90 degrees into the vertical.

00:05:31.280 --> 00:05:36.776
So this stress regime is most
consistent with normal-type faulting.

00:05:36.800 --> 00:05:41.600
When injection restarted,
the maximum principal stress

00:05:41.600 --> 00:05:45.520
rotated back into the horizontal.
So we see a very profound change

00:05:45.520 --> 00:05:50.000
in the state of stress at
the initiation of injection

00:05:50.000 --> 00:05:53.760
and during changes
in injection rate.

00:05:53.760 --> 00:05:56.960
For finite-source analysis,
we use a moment rate function

00:05:56.960 --> 00:05:59.896
finite-source
inverse method.

00:05:59.920 --> 00:06:04.376
This is based on the method
of Mori and Hartzell 1990.

00:06:04.400 --> 00:06:08.776
We use empirical Green’s function
deconvolution term in the seismic

00:06:08.800 --> 00:06:11.280
moment rate functions
at multiple stations.

00:06:11.280 --> 00:06:13.200
Here I show an
example for Parkfield.

00:06:13.200 --> 00:06:15.976
This is for a
magnitude 2.1 earthquake.

00:06:16.000 --> 00:06:20.536
There’s a nearby moment
magnitude 0.68 earthquake.

00:06:20.560 --> 00:06:25.520
We can perform Fourier transform
on both of these traces, take the ratio

00:06:25.520 --> 00:06:29.440
of these, and common terms cancel
as long as the two events are

00:06:29.440 --> 00:06:31.520
co-located and have
the same mechanism.

00:06:31.520 --> 00:06:34.080
Of course, recorded
on the same instrument.

00:06:34.080 --> 00:06:37.760
Then G and I are going to cancel out.
They’re in common.

00:06:37.760 --> 00:06:42.400
And what’s left is the – an estimate
of the main shock, or the target event,

00:06:42.400 --> 00:06:49.600
source spectrum. We can inverse FFT
and obtain a pulse in the time domain,

00:06:49.600 --> 00:06:54.000
which represents the moment rate.
So what Mori and Hartzell did was

00:06:54.000 --> 00:06:57.760
take these moment rate functions –
here’s a cartoon depicting one at

00:06:57.760 --> 00:07:04.000
one station – and devised a
finite-source model in which these

00:07:04.000 --> 00:07:09.440
moment rate functions are modeled
in terms of the superposition of

00:07:09.440 --> 00:07:16.880
slip-distributive or finite fault rupture
plane and with timing controlled by a –

00:07:16.880 --> 00:07:20.240
by a rupture velocity. So I’m going
to give you several examples.

00:07:20.240 --> 00:07:22.880
This first one is for
a magnitude 1.5 earthquake.

00:07:22.880 --> 00:07:27.816
It’s located in the center
of the study area.

00:07:27.840 --> 00:07:31.976
The moment rate
functions are shown in black.

00:07:32.000 --> 00:07:35.040
Since we’re using a spectral
domain deconvolution procedure,

00:07:35.040 --> 00:07:40.000
we interpret the moment rate
functions from the zero crossing –

00:07:40.000 --> 00:07:42.960
zero to positive, positive
back down to zero crossing.

00:07:42.960 --> 00:07:45.280
And the red traces are
showing the model fit.

00:07:45.280 --> 00:07:51.200
This is a very small earthquake.
And we find a very compact source.

00:07:51.200 --> 00:07:56.400
But, in order to explain the impulsive
nature of these moment rate functions,

00:07:56.400 --> 00:07:58.880
the slip is non-uniform.
So there’s a peak slip at the

00:07:58.880 --> 00:08:03.200
center of the fault, and it tapers
out to the edges, not surprisingly.

00:08:03.200 --> 00:08:06.400
The dimensions here
is about 100 meters.

00:08:06.400 --> 00:08:14.056
So roughly a 50-meter radius for
this magnitude 1.5 earthquake.

00:08:14.080 --> 00:08:18.696
We can take this slip distribution,
using the method of Ripperger and Mai,

00:08:18.720 --> 00:08:20.800
estimate the
coseismic stress change.

00:08:20.800 --> 00:08:23.040
And that’s what’s shown
in this diagram here.

00:08:23.040 --> 00:08:27.416
Red indicates a stress drop.
Blue indicates a stress increase.

00:08:27.440 --> 00:08:30.880
And what we find is an average
stress drop for this event, which is

00:08:30.880 --> 00:08:37.120
0.4 megapascals, and a peak stress
drop of about 1.8 megapascals.

00:08:37.120 --> 00:08:42.240
So a little bit on the low side,
but squarely within the observed

00:08:42.240 --> 00:08:45.016
range of earthquake
stress drops.

00:08:45.040 --> 00:08:52.640
The next event is a magnitude 2.9
located at the southeastern corner

00:08:52.640 --> 00:08:57.760
of the study area. The moment rate
functions have slightly larger duration,

00:08:57.760 --> 00:09:01.520
perhaps a little bit more complexity –
evidence of multiple sub-events.

00:09:01.520 --> 00:09:06.136
Two peaks here at MCL.
Two peaks here at AL5.

00:09:06.160 --> 00:09:10.880
There does seem to be a directivity
effect in terms of the overall shape

00:09:10.880 --> 00:09:16.776
of these moment rate functions.
The slip has a slightly larger dimension.

00:09:16.800 --> 00:09:19.120
The long dimension
is about 200 meters.

00:09:19.120 --> 00:09:25.280
The short dimension is quite narrow –
only about, say, 100 meters at the most.

00:09:25.280 --> 00:09:28.560
There’s evidence of two primary
asperities, perhaps a third.

00:09:28.560 --> 00:09:32.320
We take this slip distribution and,
using the Ripperger and Mai approach,

00:09:32.320 --> 00:09:36.480
we can map it to stress change.
And what we find is that the average

00:09:36.480 --> 00:09:41.440
stress drop is 11.4 megapascals,
or 110 bars.

00:09:41.440 --> 00:09:46.720
Peak stress drop is 69.1 megapascals.
I guess the point to make here is that

00:09:46.720 --> 00:09:52.480
these microearthquakes are not simple,
uniform slip ruptures. They’re complex.

00:09:52.480 --> 00:10:00.400
They do have sub-events and various
variations in slip amplitude, and

00:10:00.400 --> 00:10:04.160
therefore, variations in the
stress drop across the fault surface.

00:10:04.160 --> 00:10:08.000
The next event is a 4.4 event.
This one’s actually outside

00:10:08.000 --> 00:10:11.336
the study area but within
the Geysers geothermal area.

00:10:11.360 --> 00:10:15.360
And, for this case, we used
stations recorded by the

00:10:15.360 --> 00:10:18.776
Berkeley
Seismological Network.

00:10:18.800 --> 00:10:22.400
And the dimensions for the model
space is considerably higher.

00:10:22.400 --> 00:10:26.616
It’s about 9 square kilometers –
so 3 kilometers on a side.

00:10:26.640 --> 00:10:29.040
We see evidence of
two primary asperities.

00:10:29.040 --> 00:10:32.800
At the Hopland station,
you can see the dual pulses here.

00:10:32.800 --> 00:10:36.880
There’s a very pronounced directivity
effect in the moment rate functions.

00:10:36.880 --> 00:10:44.160
When we – when we convert the
slip distribution to stress change,

00:10:44.160 --> 00:10:48.240
we find an average stress
drop of 11.4 megapascals.

00:10:48.240 --> 00:10:54.160
The peak is about 46.9 megapascals.
So the average for this 4.4 is very

00:10:54.160 --> 00:10:58.480
similar to what we just saw for the 2.9.
All right. So we put all of this together.

00:10:58.480 --> 00:11:01.200
Of course, we’ve looked at
many more earthquakes.

00:11:01.200 --> 00:11:05.120
We also looked at a magnitude 5,
which was just outside the study area –

00:11:05.120 --> 00:11:09.176
about 600 meters south-southwest
of the study area.

00:11:09.200 --> 00:11:12.080
And, for this case, we had to use
theoretical Green’s functions.

00:11:12.080 --> 00:11:15.600
This was a very complicated rupture
for a magnitude 5 earthquake.

00:11:15.600 --> 00:11:20.400
But what we’re after here is to try to
get an idea of the rupture area scaling

00:11:20.400 --> 00:11:25.256
and the average slip scaling –
so there’s a typo here.

00:11:25.280 --> 00:11:28.720
So we have moment magnitude on
the X axis, and we have rupture area.

00:11:28.720 --> 00:11:31.680
This is 1 square kilometer.
That corresponds to about

00:11:31.680 --> 00:11:38.400
a magnitude 4.0 earthquake.
The red curve is the Wells

00:11:38.400 --> 00:11:42.880
and Coppersmith relation.
The blue curve is the Leonard 2010

00:11:42.880 --> 00:11:49.176
relationship. And this dashed curve is
the two-sigma error for Leonard 2010.

00:11:49.200 --> 00:11:53.520
And the bold dashed lines
are showing the 0.1 megapascal

00:11:53.520 --> 00:11:55.360
to 100 megapascal range.

00:11:55.360 --> 00:11:58.480
We see stress drops that are consistent
with what we’ve seen for larger

00:11:58.480 --> 00:12:02.720
earthquakes on a global scale.
Maybe there’s some evidence

00:12:02.720 --> 00:12:06.240
of lowering of stress drop
towards the smaller magnitudes,

00:12:06.240 --> 00:12:08.640
but a lot more work –
many more events are needed

00:12:08.640 --> 00:12:12.696
to see if this is actually
borne out by the data.

00:12:12.720 --> 00:12:17.680
So, given high-resolution event
locations, the seismic moment tensors

00:12:17.680 --> 00:12:22.160
are used for insight into the mechanism
type and to determine the stress tensor.

00:12:22.160 --> 00:12:25.200
Finite-source scaling is used
to determine rupture area

00:12:25.200 --> 00:12:27.896
and fault dimension
for a given Mw.

00:12:27.920 --> 00:12:31.520
And the stress tensor is used to
randomly sample fault orientations

00:12:31.520 --> 00:12:35.016
that are optimally aligned
with the stress field.

00:12:35.040 --> 00:12:38.160
The fault orientation and rupture
dimensions are assigned to the

00:12:38.160 --> 00:12:42.080
seismicity cloud to characterize
a pseudo fracture network.

00:12:42.080 --> 00:12:46.160
Or we sometimes call it
a statistical fracture network.

00:12:46.160 --> 00:12:51.920
So what’s meant by that is we
can actually plot the dimensions

00:12:51.920 --> 00:12:54.216
of these microearthquakes.

00:12:54.240 --> 00:12:57.816
This blow-up region is shown
by the bold square here.

00:12:57.840 --> 00:13:03.663
This is the Prati-32 injection well.
This is the Prati-31 recovery well.

00:13:03.689 --> 00:13:10.000
And what we can start to do with
the finite-source scaling is look at

00:13:10.000 --> 00:13:13.895
the degree of interconnectivity
of these microearthquakes.

00:13:15.600 --> 00:13:20.320
So we refer to this as a pseudo fracture
network because it’s not a direct

00:13:20.320 --> 00:13:24.936
estimation of parameters for
the many thousands of events.

00:13:24.960 --> 00:13:30.240
But the orientation of the faults and
the – and the rupture dimension

00:13:30.240 --> 00:13:37.596
scaling is constrained by analysis of
earthquakes in the – in the study region.

00:13:38.560 --> 00:13:41.040
What can we do with this?
Well, we can – we can show

00:13:41.040 --> 00:13:44.640
what the seismicity cloud looks like.
So this is the many thousands –

00:13:44.640 --> 00:13:48.936
I think it’s over 7,000 earthquakes that
were produced by the stimulation.

00:13:48.960 --> 00:13:53.120
If we look at 100-meter bins,
we can bin up the number of

00:13:53.120 --> 00:13:58.480
earthquakes and we find that,
you know, the greatest density

00:13:58.480 --> 00:14:02.080
of earthquakes lies
below the injection well.

00:14:02.080 --> 00:14:05.600
And, with the scaling parameters,
we can start to say, okay, well,

00:14:05.600 --> 00:14:12.080
what is the distribution of fracture area?
It’s basically a way of getting towards

00:14:12.080 --> 00:14:15.360
fracture density from
these seismic observations.

00:14:15.360 --> 00:14:17.760
All right. To summarize,
I’ll leave you with a video –

00:14:17.760 --> 00:14:21.582
an animation of this
pseudo fracture network.

00:14:23.040 --> 00:14:26.240
Small earthquakes at the Geysers
and elsewhere display complex

00:14:26.240 --> 00:14:32.216
rupture like larger counterparts.
Slip and stress drop are non-uniform.

00:14:32.240 --> 00:14:36.400
The 2016 magnitude 5 event is
very complex and suggests discrete

00:14:36.400 --> 00:14:39.896
sub-events rather than a more
typical connected rupture.

00:14:39.920 --> 00:14:43.040
We have been able to obtain seismic
waveform-based moment tensor

00:14:43.040 --> 00:14:46.776
solutions for the Geysers events
moment magnitude 0.6.

00:14:46.800 --> 00:14:50.480
This gives insight into activated
faults, assisting in calibrating the

00:14:50.480 --> 00:14:54.320
LBL P wave spectrum-based
Mw estimates and enabling

00:14:54.320 --> 00:14:56.720
inversion for the
moment tensor.

00:14:56.720 --> 00:15:00.640
The finite-source models demonstrate
complexity, scaling consistent with

00:15:00.640 --> 00:15:05.576
Leonard 2010, and stress drop
between 1 and 10 megapascals.

00:15:05.600 --> 00:15:10.080
We’ve illustrated a pseudo discrete
fracture network approach for the

00:15:10.080 --> 00:15:14.880
Prati-32 EGS site, utilizing the focal
mechanism-based in situ stress,

00:15:14.880 --> 00:15:17.760
moment tensor,
and finite-source scaling results.

00:15:17.760 --> 00:15:21.280
This can be used to infer
the degree of interconnectivity

00:15:21.280 --> 00:15:23.920
of microearthquake fractures.

00:15:24.880 --> 00:15:29.440
Thank you for your attention,
and I’m happy to take any questions.