Patience Pays Off:

1
Patience Pays Off Electromagnetic Signals
and the 2004 Parkfield Earthquake
Stephen Park Institute of Geophysics and
Planetary Physics Department of Earth
Sciences UC Riverside January 20, 2006
2
First, a little background
3
We knew a lot more about predicting earthquakes
in the 1970s than we do now.
Laboratory measurements showed that electrical
resistivity changed by up to 10 in response to
both compression and shear stresses close and
the changes increased as the yield strength of
the rock was approached.
Field measurements in Japan and China showed
measurable changes in resistivity associated
with Earth tides and distant earthquakes.
4
Monitoring by Madden in the Palmdale had shown
changes of 1-2 over the period from 1985-1989
that could be located on the lower crustal
portion of the San Andreas fault.
All of this led to the conclusion in the 1980s
that monitoring resistivity was a useful
component of any earthquake prediction experiment.
5
In 1984, Bakun and McEvilly noticed that
Parkfield had experienced a M6 earthquake an
average of once every 22 years
and they predicted the next one between 1983 and
1993.
6

Earthquakes were SUPPOSED to occur regularly he
re!
Parkfield
7
Goal Measure premonitory changes in resistivity
of lt 1 that may develop over
periods of days-months-years.
Use a telluric array because of the
demonstrated accuracy of such measurements.
Terminology telluric natural electric field
magnetic natural magnetic
field magnetotelluric
natural electric and magnetic fields
8
Telluric monitoring array was installed in 1987
and has operated since 1988
9
You may be wondering
10
Voltage signals are monitored with fancy grounds
Each PbCl2 electrode is a 1 m long tube filled
with saturated solution and buried in the ground
to minimize thermal variations
Depending on soil conditions, the fluid lasts
from 3 months to 5 years (and counting).
11
Electrodes are connected via telephone lines back
to a central recording site.
-

Essentially, the telephone lines act as long
leads on a digital volt meter!
12
Voltages from electrodes are conditioned by
precision analog circuits matched to 0.1 and
then differenced to create the dipoles
13
.. resulting in time series for voltages from
1988 to the present.
8000 mV
Raw signals are amplified with gains of 20-40,
filtered with a dc-300s analog bandpass filter,
and sampled every 30 seconds.
Sept 29, 2005
Sept 28, 2005
14
Electric field can be derived from the voltage by
dividing the voltage difference between
two electrodes by the distance between them
can be rewritten
assumes that E does not vary over path (i.e,
wavelength of field is large compared to dipole
length). This is important later!
15
Now, what do we do with these signals?
Di
D7
D8
Because all of the dipoles are measuring
components of the same electric field, any
dipole can be constructed from a linear
combination of two others
Robust time domain processing techniques (Park,
1991 1997) are used to estimate the telluric
coefficients x and y for dipoles 1-6, using
dipoles 7 and 8 as references.
16
However, telluric coefficients are not stable
enough to be used directly for monitoring because
they fluctuate over a range of 10 due to slight
daily changes in field orientation. Madden
(1980) proposed projecting the daily variations
of x and y onto eigenvectors representing the
two polarizations of the electric field. These
eigenvectors are determined using singular value
decomposition
Vi are eigenvectors representing the geologic
structure of the site
17
Compute daily fluctuations of x and y
Large, stable component of field
Observed signal
Daily fluctuation (residual)
-
Projections are defined
18
In Parkfield, the major eigenvector is oriented
perpendicular to the coast (parallel to D8), and
the minor one is parallel to the San Andreas
fault (and parallel to D7).
19
Projections vary less than 1 over the duration
of the experiment
Dipole 2 is the noisiest one. A-E are
earthquakes with Mgt4.0. Monthly averages of
projections daily averages are similar.
Note that the projection on the major eigenvector
is more stable than the one on the minor
eigenvector (noise has a relatively larger
contribution for the minor polarization).
20
and so we sat.
1993 rolled past
No earthquake
1995 rolled past
No earthquake
2000 rolled past.
No earthquake!
-- Gary Larsen
21
and we monitored signals for over 17 years (well
beyond 1993!!)
During this time, we had only 5 earthquakes with
magnitude greater than 4
IF variations prior to the earthquakes are
identified as precursors, THEN the minimum
resistivity changes at depths of 3-6 km needed to
produce them are shown.
22
However, NONE of these fluctuations are
statistically significant (several fluctuations
with these magnitudes occur annually).
(Park, 1997, 1999)
23
It appeared that we had created an earthquake
prevention network!
Finally on Sept. 28th, a M6.0 earthquake occurred
in Parkfield followed by 3 M4 aftershocks and 2
M5 aftershocks.
Better late than never!!
24
Yay.
25
Again, none of the instruments recorded any
precursory signals.
M6.0
M6.0
Variations of electrical resistivity parallel
(P1) and perpendicular (P2) to San Andreas fault
26
We also examined the residual signal by
subtracting from the observed dipole the telluric
component as predicted from the telluric
coefficients
. and we found a clear coseismic signal!
We also found that using dipoles 2 and 5 as
references produced clearer anomalous signals ..
27
Sept. 29, 2004
Sept. 28, 2004
28
We then began the search for anomalous
transients at other times.
and found LOTS of them! However, the
residual signal is where all of the noise
(i.e., the part of the signal not predicted by
the telluric relation), so it is reasonable to
find many transients.
1. A comparison of all records from September
2003 and September 2004 revealed that both months
had approximately the same number of transient
signals ( 25/month).
?There was NO anomalous activity prior to the
M6.0 earthquake.
transient here is defined to be a signal with a
sharp rise time and a slow decay of minutes to
hours.
29
2. There were NO anomalous coseismic transients
with any of the Mgt4 earthquakes prior to 2004.
30
Because the dipole signals are the result of
differencing electrode voltages, one can only
estimate the relative changes at each electrode
from the observed results
Note polarity difference between main shock and
aftershock and differences in magnitude
between the signals from each earthquake.
31
MAIN SHOCK
The largest potential changes were seen on
electrodes to the southwest of the San Andreas
fault.
Transients lasted for at least 3 ½
hours (exponential decay time constant of 3000
s).
M6.0
Rise time comparable to response of 300s analog
filter to impulsive input.
Dots at electrodes reflect potential changes
from main shock relative to Tf. Red is negative,
blue is positive. Largest dot is -3.8mV.
size of uncertainty in determination of
potentials
32
AFTERSHOCK
M 5 aftershock
The largest potential changes were seen on
electrodes to the southwest of the San Andreas
fault.
Transients lasted for at least 17
minutes (exponential decay time constant of 180
s).
Dots at electrodes reflect potential changes
from M5.0 aftershock relative to Tf. Red is
negative, blue is positive. Largest dot is 2.55
mV.
size of uncertainty in determination of
potentials
33
Potential causes
Electrode noise?
Why seen on multiple sensors and at time of
earthquake?
Physical shaking of electrode does not produce
potential spike.
? Probably not!
34
Fluid flow
Electrokinetic potentials arise because of
fluid flow through porous media in response to a
pressure difference
where ? dielectric constant of water (80 ?0 ),
? Zeta potential of mineral surface
(-60 mV), ? fluid viscosity, and
? fluid conductivity.
(Typical signals in rocks like those in Parkfield
are 1-10mV/bar.)
35
But fluid flow where?
Flow generated at the hypocenter?
Should be almost 10X difference between Hr
and Ff signal.
? Probably not!
36
Flow generated at the fault?
Signal at Hq should be largest and signal at Ff
should be much smaller than Hr.
? Probably not!
37
Localized flow generated in ground water aquifers
in valley adjacent to fault?
Rapid rise of transient and 3000 s time constant
are consistent with source of fluid flow within
100 m of electrode - similar to changes in
streamflow observed by Wang et al. (2004) in Paso
Robles after the San Simeon earthquake.
Signal does not scale with energy of earthquake.
Nonlinear process?
? Maybe!
38
Local response to what, however?
Static Stress Change?
The sign change between the two electrode
sites is inconsistent with the observations.
39
Lets add another nail to the static stress
coffin.
The voltage ratios at Ff or Hr between the
aftershock and main shock are much too large, AND
the change in voltage polarity between the main
shock and aftershock is inconsistent with the
dilations.
40
Well, what COULD explain a polarity change in the
electric potentials between the main shock and
aftershock?
41
So, may be related to dynamic stresses associated
with onset of ground motion.
But not simply!
Brodsky has proposed a mechanism of
permeability enhancement by dislodging clay
particles that block pore throats, but shaking in
any direction should produce increased flow and
therefore signals from the main shock and
aftershock with the same polarity.
An additional complication Voltage is sampled
once every 30 seconds, so probably looking at
integrated effect of wave motion (much like
liquefaction dependence on number of cycles of
shaking).
Next step get ground motion records from nearby
strong motion instruments (none near Hr or Ff)
and compare.
42
CONCLUSIONS
?
No significant resistivity changes recorded
1988-2005.
Coseismic electrical transients were recorded
with the September 28, 2004 M6.0 Parkfield
earthquake.
?
?
NO identifiable precursors prior to earthquake.
?
Signals appear to have threshhold of Mgt5 to be
observed.
Signals have little predictive power at this
time many similar transients are observed (about
1 per day on average).
?
43
CAVEAT
Transients were about 100X smaller than
background telluric signals, so search in other
experiments will need to have GOOD cancellation
of naturally occurring signal fluctuations AND
BETTER ways of discriminating between noise and
tectonically significant transient signals.
Standard MT analyses will likely NOT be
sufficient coherency between E, H not high
enough.