The Parkfield, California earthquake experiment: An update in 2000

Abstract

The US Geological Survey, in cooperation with other institutions, continues to monitor the San Andreas Fault (SAF) near Parkfield, California, hoping to capture high resolution records of continuous de-formation before, during and after a magnitude 6 earthquake, as well as the details of its rupture initia-tion and strong ground motion. Despite the failure of the prediction that the next M 6 Parkfield earthquake would occur before 1993, Parkfield still has a higher known probability (1 to 10% per year) than anywhere else in the US of a M 6 or greater earthquake. Park-field instrumentation is still largely in place, although there have been losses due to attrition as well as improvements made possible by new technology. Most Parkfield data sets are now available via the Internet, and all others may be obtained upon request from individual investigators. Detailed seismic monitoring has shown that events with identical seismograms, recurring in exactly the same locations, account for a high proportion of the background seismicity at Park-field. Geophysical studies have revealed that fault zone seismic and electrical properties are consistent with high fluid content. The rate of interseismic slip on the SAF changed significantly in late 1992 or early 1993, during a period of relatively high seismic acti-vity. The strain-rate change, measured by borehole tensor strainmeters and the two-colour electronic dis-tance-measuring network, was also manifested as shortened recurrence intervals of repeating micro-earthquakes. Whether or not the accelerated defor-mation turns out to be an intermediate-term precursor to the next M 6 Parkfield earthquake, docu-menting the variation of interseismic strain rates with time has important implications for fault dynamics and seismic hazard estimation. Two possible instances of pre-earthquake signals have been recorded at Parkfield: water-level and strain changes over a period of three days prior to the nearby 1985 M w 6.1 Kettleman Hills, California, earthquake and anoma-lous electromagnetic signals prior to the M 5 earth-quake near Parkfield on 20 December 1994. Future work planned at Parkfield includes a National Science Foundation proposal to construct an SAF Observa-tory at Depth (SAFOD), as part of the Earthscope initiative. The Observatory will consist of a 4-km-deep borehole to penetrate the SAF and a shallow micro-earthquake cluster on Middle Mountain, directly above the hypocenter of the 1966 Parkfield earthquake.

Full text

Page 1

SPECIAL SECTION: SEISMOLOGY 2000
The Parkfield, California earthquake
experiment: An update in 2000
CURRENT SCIENCE, VOL. 79, NO. 9, 10 NOVEMBER 2000
1226
Evelyn Roeloffs
US Geological Survey, 5400 MacArthur Blvd., Vancouver, WA 98661, USA
The US Geological Survey, in cooperation with other
institutions, continues to monitor the San Andreas
Fault (SAF) near Parkfield, California, hoping to
capture high resolution records of continuous de-
formation before, during and after a magnitude 6
earthquake, as well as the details of its rupture initia-
tion and strong ground motion. Despite the failure of
the prediction that the next M 6 Parkfield earthquake
would occur before 1993, Parkfield still has a higher
known probability (1 to 10% per year) than anywhere
else in the US of a M 6 or greater earthquake. Park-
field instrumentation is still largely in place, although
there have been losses due to attrition as well as
improvements made possible by new technology. Most
Parkfield data sets are now available via the Internet,
and all others may be obtained upon request from
individual investigators. Detailed seismic monitoring
has shown that events with identical seismograms,
recurring in exactly the same locations, account for a
high proportion of the background seismicity at Park-
field. Geophysical studies have revealed that fault
zone seismic and electrical properties are consistent
with high fluid content. The rate of interseismic slip
on the SAF changed significantly in late 1992 or early
1993, during a period of relatively high seismic acti-
vity. The strain-rate change, measured by borehole
tensor strainmeters and the two-colour electronic dis-
tance-measuring network, was also manifested as
shortened recurrence intervals of repeating micro-
earthquakes. Whether or not the accelerated defor-
mation turns out to be an intermediate-term
precursor to the next M 6 Parkfield earthquake, docu-
menting the variation of interseismic strain rates with
time has important implications for fault dynamics
and seismic hazard estimation. Two possible instances
of pre-earthquake signals have been recorded at
Parkfield: water-level and strain changes over a
period of three days prior to the nearby 1985 Mw 6.1
Kettleman Hills, California, earthquake and anoma-
lous electromagnetic signals prior to the M 5 earth-
quake near Parkfield on 20 December 1994. Future
work planned at Parkfield includes a National Science
Foundation proposal to construct an SAF Observa-
tory at Depth (SAFOD), as part of the Earthscope
initiative. The Observatory will consist of a 4-km-deep
borehole to penetrate the SAF and a shallow micro-
earthquake cluster on Middle Mountain, directly
above the hypocenter of the 1966 Parkfield earthquake.
1. Introduction
The US Geological Survey (USGS), in partnership with
the state of California and other institutions, has inten-
sively monitored the San Andreas Fault (SAF) near the
town of Parkfield since 1985 (Figure 1). One goal is to
obtain a detailed long-term record of fault behaviour
believed likely to culminate in a moderate earthquake.
Another goal is to record the details of seismic rupture,
strong ground motion, and earthquake effects. Roeloffs
and Langbein1 described the experiment’s instrumenta-
tion and findings as of 1993. Currently (April 2000),
the anticipated M 6 earthquake that the experiment was
designed to record has not yet occurred. Seven years
have elapsed since expiration of the original prediction,
which stated that the next M 6 Parkfield earthquake would
take place with 95% confidence before 1993 (ref. 2).
Although many still regard the Parkfield experiment as a
critical opportunity to observe an active fault as it loads to
failure in a moderate earthquake, there is controversy as
to how long the experiment should continue, and at what
expense.
This article, an update of Roeloffs and Langbein1,
summarizes current thinking about the likelihood of
future Parkfield earthquakes and describes changes to the
field instrumentation as technology has improved and
funding has fallen. There is strong evidence for a change
in the rate of aseismic fault slip at Parkfield in 1992 or
1993. Recognition of clustering in background micro-
seismicity and geophysical indicators of fault zone fluids
are important new findings about the SAF. Funding has
been requested for a 4-km-deep drillhole to penetrate the
SAF near Parkfield and sample fault zone rocks and flu-
ids, as well as provide an observatory for fault zone
observations at depth.
2. Parkfield earthquake probabilities
Since the expected M 6 Parkfield earthquake did not
occur prior to 1993, revised annual probabilities for the
e-mail: evelynr@usgs.gov

Page 2

SPECIAL SECTION: SEISMOLOGY 2000
CURRENT SCIENCE, VOL. 79, NO. 9, 10 NOVEMBER 2000
1227
next such event have been suggested, most of which are
summarized in Roeloffs and Langbein1. In a more recent
study, Kagan3 computed probabilities of sequences like
those of the historic Parkfield earthquakes, and con-
cluded that the observed sequence was not inconsistent
with the assumption that earthquakes occur according to
a Poisson, rather than quasi-periodic, model. This analy-
sis puts the annual probability of a Parkfield mainshock
at about 1%. A review of all available studies4 concluded
that 10% was a concensus value for the annual probabil-
ity, and judged this probability high enough to recom-
mend continuing the experiment.
The Parkfield experiment includes a computerized sys-
tem for detecting unusual signals in the monitoring data,
allowing the USGS to alert the California Office of
Emergency Services if tectonic events indicate that the
next M 6 Parkfield earthquake might be imminent. Four
levels of alert exist, the highest (level ‘A’) reachable only
by rapid aseismic creep or a potential large foreshock,
corresponding to an estimated 37% probability of the
mainshock occurring within three days.
Michael and Jones5 provide new estimates of Parkfield
seismic alert probabilities, based on lower estimates of
the annual mainshock probability (4 to 10%), a better
data set for estimating background seismicity, and a revi-
sed methodology6. They also propose a new definition of
the Parkfield mainshock as any event with Mw ≥ 5.7,
accompanied by surface rupture, whose epicenter is
within 5 km of the mapped fault trace between the lati-
tudes of 35°45′ N and 36° N. They suggest that Parkfield
foreshocks should be considered capable of occurring
anywhere within a ‘Parkfield box’, which extends along
the same length of the SAF as the zone specified to con-
tain the mainshock, but is twice as wide. With these
definitions, the probability of a mainshock with epicenter
in the specified area, given the occurrence of a M > 5
potential foreshock, is between 2.2 and 19%, depending
upon possible variations in annual mainshock probability
and form of the dependence of foreshock probabilities on
magnitude. A potential foreshock at Parkfield could now
only produce a level ‘A’ alert if special studies of the
waveform and hypocenter were to show it to be a recur-
rence of the immediate foreshock to the 1934 and 1966
Parkfield earthquakes. A M 5 foreshock preceded each of
those events by 17 minutes, and these 1934 and 1966
foreshocks had essentially identical seismograms7. Rec-
ognizing a future M 5 event as a repeat of one of these
foreshocks would currently require more time than 17
minutes, implying that a level ‘A’ alert is unlikely to be
issued.
Although the probability that the Parkfield earthquake
will occur in any given year has been revised downward,
it is still the location of highest known probability for a
moderate earthquake anywhere in California, and has one
of the shortest recurrence intervals, based on the same
methods used to estimate earthquake probabilities else-
where in the state. In 1857, M 6 earthquakes on the SAF
near Parkfield were followed within 2 hours by the great
1857 Fort Tejon, California, earthquake (M 7.9)8, sug-
gesting that a future Parkfield event might herald a dam-
aging earthquake in now densely-populated southern
California. On the other hand, the anticipated Parkfield
earthquake is not believed likely to cause fatalities or
significant financial losses by itself. Level funding for
earthquake studies in the US, increased emphasis since
the 1994 Northridge, California earthquake on research
with short-term hazard mitigation payoffs, and uncer-
tainty about the likelihood of the next Parkfield earth-
quake have made the issues of how long the experiment
should continue, and at what cost, important ones to cri-
tically review.
3. Instrumentation status
The instrumentation at Parkfield includes seismic net-
works, crustal deformation sensors, and several arrays of
strong-motion instruments. The strong-motion instrumenta-
tion remains essentially as described in Roeloffs and Lang-
bein1, but other networks have undergone some change.
Two borehole volumetric strainmeters (dilatometers)
have ceased to operate and have not been replaced, leav-
ing five functional dilatometers. The EDT borehole
tensor strainmeter (BTSM) still operates close to the
location of one of the non-operational dilatometers.
The southernmost creepmeter, X461, has been de-
commissioned. Groundwater radon, soil hydrogen, shal-
low borehole tilt, and borehole microtemperature are no
longer measured. Two groundwater-level monitoring
Figure 1. Map of the Parkfield area.

Page 3

SPECIAL SECTION: SEISMOLOGY 2000
CURRENT SCIENCE, VOL. 79, NO. 9, 10 NOVEMBER 2000
1228
wells were abandoned, and the former radon-monitoring
well at Carr Hill was converted to a groundwater-level
observation site in 1994. With these changes, eight
groundwater-level monitoring sites are still operating,
including one well in which the water-level is monitored
at two depths.
The two-colour electronic distance-measuring (EDM)
network continues to be measured at a reduced frequency
of six times per month. It is currently supplemented by
continuous GPS observations along several of its base-
lines, and the continuous GPS is envisioned to replace
the two-colour EDM network eventually.
The downhole High Resolution Seismic Network
(HRSN) is being upgraded with new electronics, and
three new stations will be added near the proposed deep
drillhole (see Section 6). The controlled-waveform-
monitoring experiment has been discontinued for lack of
funds. Two new broadband seismic stations have been
installed in the Parkfield area.
The resistivity and magnetic monitoring networks have
been upgraded, but funding difficulties make their future
uncertain. There are two magnetic monitoring sites. A
vertical coil is monitored at one site, and at the other site
three orthogonal coils and two orthogonal dipoles allow
all non-zero components of the magnetic and electrical
fields to be monitored in the frequency range from 10–4 to
20 Hz. Data are digitized with 20-bit precision at 40 Hz
and 1 Hz and are sent by telemetry to the Northern Cali-
fornia Earthquake Data Center (NCEDC), in contrast to
the previous practice in which only average values of the
magnetic fields in certain frequency bands were retained.
Improved data-processing techniques were implemented
so that long-period magnetic signals of upper atmosphe-
ric origin could be identified by comparison with a simi-
lar station near Hollister9. The electrical dipoles would
permit detection of any possible ‘Seismic Electrical Sig-
nals’ similar to those reported by Varotsos et al.10, and
the magnetic data would help clarify the mechanisms of
such signals. The magnetic coils are also designed to
detect signals similar to those recorded by Fraser-Smith
et al.11 prior to the Loma Prieta earthquake. Data from
the Parkfield resistivity network12 are now also digitized
at the same rates as the magnetic data.
Many types of data from Parkfield can now be viewed
or downloaded via Internet. The USGS Parkfield home
page (http://quake.wr.usgs.gov/QUAKES/Parkfield/) con-
tains links to plots of crustal deformation data. Parkfield
earthquakes recorded by the surface short-period network
are part of the Northern California Earthquake Catalogue,
accessible at the NCEDC, operated by the USGS and the
University of California at Berkeley. Data from broad-
band seismometers and the downhole HRSN are also
available at the NCEDC. Work is currently underway to
place all Parkfield deformation data at the NCEDC,
where it will be available to the entire scientific commu-
nity. In the meantime, most Parkfield data may be obtained
on request by contacting individual investigators.
4. Discoveries about the SAF at Parkfield
The large amount of detailed data collected at Parkfield
has yielded new insights about microseismicity patterns
and fault structure. Many of these findings are consistent
with a role for fluids in controlling fault behaviour.
4.1 Repeating earthquakes
Independent analyses by two groups have shown that, at
Parkfield, earthquakes with M > 4, as well as microearth-
quakes with Mw < 1.3, occur largely in earthquake
families, or clusters, of recurring events with identical
seismograms.
It has long been known that foreshocks 17 minutes
before each of the 1934 and 1966 M 6 Parkfield earth-
quakes produced nearly identical Wood–Anderson seis-
mograms near Berkeley, California7. This phenomenon is
the basis for the current level ‘A’ alert criterion, which
requires an earthquake whose seismogram matches that
of the 1934 and 1966 foreshocks. Ellsworth and
others13,14 have extended this intriguing observation to
show that almost all M > 4 earthquakes near Parkfield
can be grouped into classes, each characterized by a dis-
tinctive seismogram and presumably repeatedly rupturing
the same small fault patch despite the occurrence of two
M 6 earthquakes. Two of the four M > 4 earthquakes
since 1992 (EQ1 and EQ3 in Table 1) have fit formal
criteria for potential Parkfield foreshocks (and prompted
public warnings that a M 6 event was temporarily more
likely). Closer scrutiny showed seismograms of EQ3, in
Table 1. Earthquakes M > 4 in Parkfield since 1986, from the University of California at Berkeley/US
Geological Survey Northern California Earthquake Data Center. Coda magnitudes were obtained
from the NCEDC. Other studies have assigned different, usually higher magnitudes to these
events, so to avoid confusion they will be referred to as EQ1 to EQ4



Date (UT) Latitude


EQ1 1992/10/20 05:28:08.90 35.9285°N
EQ2 1993/04/04 05:21:25.27 35.9413°N
EQ3 1993/11/14 12:25:34.87 35.9527°N
EQ4 1994/12/20 10:27:47.17 35.9175°N





Earthquake








Time
Longitude
Depth
(km)
Magnitude
(coda)






120.4728°W
120.4925°W
120.4968°W
120.4643°W
10.21
7.65
11.70
9.10
4.3
4.2
4.6
4.7

Page 4

SPECIAL SECTION: SEISMOLOGY 2000
CURRENT SCIENCE, VOL. 79, NO. 9, 10 NOVEMBER 2000
1229
November 1993, to be identical to those from a foreshock
three days prior to the 1966 Parkfield mainshock. None
of the events in Table 1 has been a repeat of the 17-
minute foreshocks, however.
The downhole HRSN at Parkfield has recorded micro-
seismicity on this stretch of the fault for 11 years at a
level of detail unsurpassed anywhere worldwide. Nadeau
et al.15 showed that about half of the events recorded by
the HRSN with 0.2 < MW < 1.3 occur in about 300 clus-
ters of microearthquakes with highly similar waveforms.
Within clusters, relative event locations based on wave-
form cross-correlation can be accurate to within 5 m (ref.
16). Many of these clusters have characteristic recurrence
times that scale with the magnitude of the repeating
events. Changes in this recurrence time have been ex-
ploited, as described later, to infer that slip rates over
portions of the fault vary with time. Since the clusters
cannot be detected without low-noise data and special
data-processing techniques, it remains unknown whether
other parts of the SAF also exhibit the clustering beha-
viour. Tullis17 suggests that the microearthquake clusters
may represent lithologically distinct sites that have
velocity-weakening frictional behaviour. Boatwright and
Cocco18 point out that the fault planes can be divided into
zones based on their tendency for either seismic rupture,
aftershocks, or background seismicity, suggesting that a
mix of velocity-weakening and velocity-strengthening
behaviour is distributed within the part of the fault zone
that is macroscopically creeping with localized patches of
background seismicity. The picture emerging from all of
these studies is that background seismicity at Parkfield
repeatedly ruptures the same tiny areas of the fault,
which seem to have unique properties that control the
SAF’s behaviour.

4.2 Role of fluids
Fault-zone fluids are widely believed key to understand-
ing earthquake generation. Specifically, it has been hypo-
thesized that high fluid pressure in the fault zone is the
mechanism that reduces the frictional strength of the fault
zone, and that time variations in fluid pressure control the
timing of earthquakes (e.g. Miller et al.19). Some of the
data collected at Parkfield now allow the feasibility of
these hypotheses to be tested.
Several studies of seismic wave velocities around
Parkfield have identified bodies at seismogenic depths
with velocities or attenuation that could be caused by
high fluid pressure. Between depths of 6 and 10 km,
there is a 3-km-wide zone of low Vp (ref. 20), relatively
lower Vs, and therefore high Vp/Vs (1.9–2.0) (ref. 21) im-
mediately north-east of the active fault surface as
defined by microseismicity. Along strike of the SAF, this
body extends from the 1966 hypocentre about 5 km to the
south-east21,22. This zone is also characterized by high
attenuation. The seismic wave propagation characteristics
of this body are consistent with high fluid content, but do
not require pressure in the fluid to be elevated. High fluid
pressure in a 1-km-deep borehole on the NE side of the
SAF in this area, however, shows that there is at least
localized overpressure at depth.
Unsworth et al.23 conducted a magnetotelluric transect
across the SAF on Middle Mountain, directly above the
north-eastern limit of the high Vp/Vs body. The prominent
finding is a vertical zone of low resistivity along the fault
trace, about 500 m wide, extending to about 4000 m
depth, with higher resistivities, or narrower width, at
greater depth. This zone contains significant areas where
the resistivity is about 1 Ω – m, a range too low to be
reached through the presence of clays or serpentinite
alone, and therefore strongly indicates a network of
interconnected pore space. Unsworth et al.23 estimated
the amount of fluid present as either 9 to 30% fluid-filled
porosity, or a total width of 30 m of fluid-filled macro-
scopic cracks, assuming that the fluid is a brine of
30,000 ppm chloride similar to that found in the deepest
drillhole to date near Parkfield. Li et al.24 had previously
inferred the existence of a seismic low-velocity zone of
similar width to model seismic trapped waves in the fault
zone. The seismic low-velocity zone would require at
least some of the fault-zone fluid to be distributed in
pores, rather than localized in cracks.
Studies of gas flux have been made at Parkfield to
investigate the hypothesis that CO2 outgassing from deep
sources could provide the fluid pressurization needed to
account for low frictional stress across the SAF. Lewicki
and Brantley25 measured CO2 fluxes and concentrations
along 16 fault-crossing transects, and found high CO2
flux anomalies on 12 transects within about 40 m of the
fault trace. The high-flux locations, however, were not
always the sites of highest CO2 concentration, and iso-
topic studies indicate the CO2 is of biogenic origin. Thus
the fault zone represents a high permeability or diffusi-
vity conduit for the escape of biogenic CO2, with no evi-
dence for deeper degassing. This study, which included
transects on Middle Mountain close to the MT profile
and directly above the 1966 epicentre, suggests that the
fault at shallow depth is a conduit for vertical flow, rather
than a low-permeability zone that would help maintain
high fluid pressure at depth.
The previous studies show that bodies that might con-
tain fluids are likely present, and that the hydrologic
properties of the fault zone are distinct from those of its
surroundings. Johnson and McEvilly22 considered how
microseismicity features might reveal fluid involvement.
The clusters in which similar microearthquakes repeat at
fairly regular intervals could be sites of unique fluid
pressure, rather than lithologic conditions. Some bursts
of localized microseismicity include sequences of hypo-
centres successively further from the initial event over
periods of hours and distances of 1 to 2 km. Although

Page 5

SPECIAL SECTION: SEISMOLOGY 2000
CURRENT SCIENCE, VOL. 79, NO. 9, 10 NOVEMBER 2000
1230
moment tensor decomposition is problematic for such
small events, there are some events with non-double-
couple components that indicate flattening of the source.
Both features are consistent with hydrofracture of high-
fluid-pressure ‘pods’, and less consistent with failure of
isolated impermeable asperities surrounded by a gener-
ally high-fluid-pressure fault plane.
Evidence to date for fluid-driven seismicity at Park-
field includes seismic velocities consistent with the pre-
sence of fluids in the fault zone, including the 1966
hypocentre, and with additional corroboration from the
resistivity structure shallower than 4 km. Seismicity pat-
terns that resemble those expected from diffusion of
localized pore pressure have occurred. Gas studies show
that at shallow depths, the fault is a zone of relatively
high vertical permeability. The sources of fault-zone flu-
ids and the level of in situ pressures remain unknown,
and are important scientific goals of proposed deep drill-
ing at Parkfield.
5. Departures from steady-state strain
accumulation
Roeloffs and Langbein1 noted few departures from
steady-state strain accumulation between 1985 and 1993.
This picture has now changed: geodetic and seismic
observations show that aseismic fault slip at Parkfield
accelerated in late 1992 or early 1993. This tectonic epi-
sode began with increased seismicity, which has since
subsided, and continued until 1997 with faster strain rates
measured by BTSM’s, the 2-colour EDM, and at least
one reliable creepmeter. Fault slip rates at depth, inferred
using an innovative technique based on the recurrence
intervals of repeating microearthquakes, agree strikingly
well with the zones of accelerated slip inferred from sur-
face geodetic measurements.
5.1 Increased seismicity
Starting with a M 4.6 earthquake in October 1992, seis-
micity began to increase near the hypocentre of the 1966
Parkfield earthquake. Seismicity rate increased almost
ten-fold, primarily at depths greater than 5 km, and by
December 1994 there had been four earthquakes with
M > 4 in the nucleation zone (Table 1). Ruptures in the
three largest events propagated toward the 1966 main-
shock hypocentre, and their slip zones surround the 1966
hypocentre on three sides26,27. The M > 4 earthquakes in
this sequence have added at least one bar of stress
favouring right-lateral fault slip in the nucleation zone,
without triggering or developing into another M 6 Park-
field event. Fletcher and Guatteri27 point out that all of
these events ruptured either updip or to the NW, whereas
numerical simulations28 show that right-lateral ruptures
should propagate more easily to the SE because the
Figure 2. Map of the most densely instrumented segment of the SAF near Parkfield, showing locations of creepmeters, BTSMs and the 2-colour
EDM network.