A mechanism for sustained groundwater pressure changes

induced by distant earthquakes

Emily E. Brodsky,1 Evelyn Roeloffs,2 Douglas Woodcock,3 Ivan Gall,4

and Michael Manga5

Received 25 November 2002; revised 8 March 2003; accepted 3 April 2003; published 22 August 2003.

[1] Large, sustained well water level changes (>10 cm) in response to distant (more than

hundreds of kilometers) earthquakes have proven enigmatic for over 30 years. Here we

use high sampling rates at a well near Grants Pass, Oregon, to perform the first

simultaneous analysis of both the dynamic response of water level and sustained changes,

or steps. We observe a factor of 40 increase in the ratio of water level amplitude to seismic

wave ground velocity during a sudden coseismic step. On the basis of this observation

we propose a new model for coseismic pore pressure steps in which a temporary barrier

deposited by groundwater flow is entrained and removed by the more rapid flow induced

by the seismic waves. In hydrothermal areas, this mechanism could lead to 4 102 MPa

pressure changes and triggered seismicity. INDEX TERMS: 1829 Hydrology: Groundwater

hydrology; 7209 Seismology: Earthquake dynamics and mechanics; 7212 Seismology: Earthquake ground

motions and engineering; 7260 Seismology: Theory and modeling; 7294 Seismology: Instruments and

techniques; KEYWORDS: earthquakes, triggering, time-dependent hydrology, fractures

Citation: Brodsky, E. E., E. Roeloffs, D. Woodcock, I. Gall, and M. Manga, A mechanism for sustained groundwater pressure

changes induced by distant earthquakes, J. Geophys. Res., 108(B8), 2390, doi:10.1029/2002JB002321, 2003.

1. Introduction

[2] Earthquakes can produce sustained water level

changes in certain distant wells [Coble, 1965; Bower and

Heaton, 1978; Matsumoto, 1992; Roeloffs, 1998; King et

al., 1999] that are often orders of magnitude larger than can

be explained by static stress changes [Bower and Heaton,

1978]. Many researchers suggest that seismic waves interacting

with aquifers produce the sustained changes in pore

pressure, or steps, hundreds of kilometers from an earthquake

[Bower and Heaton, 1978; Roeloffs, 1998; King et

al., 1999]. The redistribution of pore pressure can generate

crustal deformation [Johnston et al., 1995] and perhaps

even trigger seismicity [Hill et al., 1993; Brodsky et al.,

2000; U.S. Geological Survey (USGS), 2000]. However, the

mechanism by which small cyclic stresses induce persistent

pore pressure changes has remained uncertain.

[3] Here we constrain the mechanism for coseismic steps

in a well near Grants Pass, Oregon, by using both high

sample rate water level data from the well and seismic data

from the broadband Berkeley Digital Seismic Network

station Yreka Blue Horn Mine (YBH) in Yreka, California

(Figure 1). The water level in a well penetrating a confined

aquifer is a manometer measuring the pore pressure at a

point. During 1993–2001, several seismic water level

oscillations and two coseismic steps were recorded digitally.

The 1 September 1994 Mw = 7.2 Petrolia, California

(epicentral distance = 2.71), earthquake generated a

15 cm decrease in water level over 2.5 days, and the

30 September 1999 Mw = 7.4 Oaxaca, Mexico ( =

34.65) earthquake generated an immediate 11 cm decrease

in water level. We show that (1) the coseismic steps are

related to the passage of seismic waves, (2) the amplitude

of the water level oscillations relative to the seismic ground

velocity increased abruptly at the time of the step induced

by the Oaxaca earthquake, and (3) gradual water level

steps are consistent with pore pressure changes diffusing to

the well from within the aquifer. These observations

motivate a new model for distant water level changes.

Seismic waves remove a temporary barrier of sediment or

solid precipitate resulting in both an increase in the seismic

wave amplification and a persistent water level change. We

then test the model with a new observation during the

3 November 2002 Mw = 7.9 Alaska earthquake.

2. Observations

[4] The 91.4 m deep NVIP-3 well near Grants Pass,

Oregon, has been monitored continuously since 1984

[Woodcock and Roeloffs, 1996]. The well is drilled into a

fractured granodiorite confined aquifer and a float measures

the water level. The chart recorder installed in 1984 was

replaced in November 1993 with a digital data logger

recording at 1.7 103 or 1.1 103 Hz. If the water

level changed more than 0.6 mm, the sampling rate increased up to a maximum of 1 Hz. In October 1998 a pressure

transducer was added sampling at 1.7 103 Hz. Since

March 2001, 1 Hz data from both the float and the transducer

have been collected to verify that no instrumental delay is

introduced by the float. The well geometry and hydrological

properties are given in Table 1. The dynamic response

cannot be modeled for the eight earthquake-related water

level drops before 1994 [Woodcock and Roeloffs, 1996]

because the chart records lack sufficient resolution.

[5] Below we first discuss the hydrological and seismological

observations pertaining to the oscillatory response of

the well in the seismic frequency band (0.02–0.2 Hz). We

then present direct observations of steps in water level.

2.1. Oscillatory Well Response to Shaking

[6] During 1993–2001, several earthquakes produced

ground shaking on the order of mm s1 at the site and water well level oscillations with amplitudes 10 cm. These

large responses imply a large amplification in the wellaquifer

system (Figure 2). The water level displacement in

the well measures the head change in the aquifer induced by

the strain of the seismic waves. Hydraulic head h is defined

as h p/rg z, where p is the pore pressure, r is the density

of water, g is gravitational acceleration, and z is the elevation.

For waves in an elastic medium, strain is proportional to

particle velocity [e.g., Love, 1927, equation XIII.17]. Therefore

the amplification of the seismic waves in the well is

measured by the ratio c of the amplitude of the water level

oscillations to the particle velocity in the seismic waves. The

units of c are m/(m/s).

[7] The amplification factor c is computed by dividing

the observed well spectra from NVIP-3 by the seismically

observed vertical ground velocity spectra from YBH for the

records (Figure 3b). Amplitude corrections are applied to

the seismograms to account for differences in geometric

spreading and radiation pattern between YBH and the well.

These corrections are small (<15%) for all of the events

discussed in this paper. A 2002 seismic installation showed

that both YBH and NVIP-3 are hard rock sites and no site or soil correction is necessary. Regional earthquakes generate

relatively high-frequency signals (>0.05 Hz) while teleseismic

waves have lower frequencies. We use the regional

events to determine the high-frequency c and the teleseismic

events for the low-frequency c (Figure 3b). During

four regional earthquakes the average observed value of the

amplification factor c is 280 m/(m/s) at 0.06 Hz with a

standard deviation of 49 m/(m/s).

[8] The large value of c quantifies the large amplification

in the well and is primarily determined by the small specific

storage Ss. Specific storage is defined as the volume of fluid

a unit volume of aquifer releases under a unit decrease in

head. The definition can be shown to be equivalent to

Ss ¼ rgða þ fbÞ; ð1Þ

where a and b are the compressibility of the aquifer and

fluid, respectively, and f is the porosity [Freeze and Cherry,

1979]. Small Ss, indicative of low porosity, is expected for

crystalline rock. The small strains of the Earth tides (107)

which produce large amplitude signals (>6 cm) in the well

(Figures 1b, 1d, and 1f ) constrain Ss (Table 1). The hydraulic

conductivity and local structure also affect c. The flow

through the porous medium and the resonance of the water

column in the well both introduce a frequency dependence as

discussed below.

[9] We use a pumping test to constrain the geometry of the

aquifer near the well. Water was pumped from the well at a

rate of 20 GPM (20 GPM = 1.26 103 m3 s1) for 52 hours

(1 sample per 10 min or 1 sample per 15 min, respectively).

Figure 3 shows that the form of the drawdown curve is well

modeled by flow through a single, infinitesimally thin square

planar fracture embedded in an unbounded, homogeneous

and isotropic confined aquifer. Flux is constant over the

surface area of the fracture. The solution for the drawdown

curve in this geometry can be derived by integrating point

sources or using the Green’s function method of Gringarten

and Ramey [1973] to derive

h t ð Þ h t ¼ 0 ð Þ ¼

1

Ss

Z t

0

QðtÞ

L2

1

2

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

pKðt tÞ=Ss

p

erf

L

4

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Kðt tÞ=Ss

p

!2

dt; ð2Þ

where h(t) is the head in the well, Q(t) is the volumetric

withdrawal rate, K is the hydraulic conductivity, and L is the

fracture length. The relatively small-radius well intersects

and samples the fracture but has no effect on the modeled

flow. The shape of both the pumping test curve and the

seismic wave response are well-fit by a single fracture

model (Figure 3). A simpler, cylindrical flow model (Theis

solution) does not fit the data. The geometry is also

consistent with the drilling log which recorded a marked

increase in flow in the final 2 m of the well. Note that the

slope of 1/2 at early times on the log-log drawdown curve

precludes significant storage of water in the fracture and is

consistent with the small fracture aperture (1 mm) used in

the mechanism discussion below.

[10] The linearized response function of the well to

seismic waves is calculated following the method of Cooper

et al. [1965] with the geometry modified to be consistent

with the pumping test data in Figure 3. Cooper et al. [1965]

assumed cylindrical flow into the well; we assume strictly

linear flow into the fracture that intersects the well. The pore

pressure changes in the aquifer are assumed to be proportional

to the dilatational strain of the seismic waves. The

most significant dilatation is generated by the Rayleigh

wave, rather than the body waves; therefore we only

consider the Rayleigh waves as sources of pressure oscillations.

The flow directly into the well is assumed negligible

because the surface area of the fracture is much greater than

the well. The amplification c( f ) is

cð f Þ ¼ Aðn; cÞ 1

4p2H f2

g

þ

p

2

r2

w

L2

ffiffiffiffiffiffiffiffi

p f

KSs

s

ð1 þ iÞ

1

; ð3Þ

where A(n, c) is the ratio of the dilatational strain to the

vertical ground velocity which is a function of the Poisson’s

ratio n and the seismic phase velocity c, is the ratio of the

confined aquifer fluid pressure to changes in the dilatational

static strain and is inferred from the tidal response

[Rojstaczer and Agnew, 1989], f is frequency, H is the

water column height, and rw is the well bore radius. The

quantity A(n, c) is 0.46/c for a Rayleigh wave in a Poisson

solid [Ben-Menahem and Singh, 1981]. We use c =

3.7 km s1 as a representative value for the Rayleigh wave

phase velocity at the site. On the basis of records of the

2002 Denali earthquake, the actual frequency-dependent

phase velocity could possibly vary by as much as 3.3–

4.1 km s1 for 20–35 s Rayleigh waves at this site. The

variability in c introduces an error of as much as 10% in c.

More significantly, we expect some coupling between the

fluid pressure and the other seismic phases. In fractured

rock, shear stresses can produce oscillations in some

orientations and Love waves are occasionally observed in

this well. However, full prediction of the effects of all the

seismic phases requires that we know the full threedimensional

strain tensor at the site. Such an inversion

would require a local seismic array, which was not

emplaced at the time of the earthquakes studied here. We

show below that the overall spectral response and relative

changes can be satisfactorily explained with a simplified

model in which only the large dilatation of the Rayleigh

waves affects the hydrological system. The specific

parameters of the response is also consistent with

independent measurements from a pumping test. Therefore

we accept the simplified model of equation (3) as an

adequate representation of the coupling.

2.2. Water Level Steps

[11] The rapid step in water level during the Oaxaca

earthquake occurred during the seismic shaking (Figure 4).

The water level is recorded independently on both the float

and the pressure transducer. Therefore the step in the

record was not caused by instrumental error. Even given

a 10 min ambiguity in arrival time due to a station clock

error (Figure 4), the step must have occurred during the

passage of the seismic waves. We use the frequency

content of the dispersed Rayleigh wave to align the records

more precisely.

[12] The well record for the Oaxaca event shows only

small oscillations due to surface waves prior to the step (Figure 4). The float was clearly free to track the water level

as tides and random small fluctuations were recorded, yet

significant amplification of the Rayleigh wave train did not

occur until after the step. More specifically, before the step

the amplification factor c is less than 10 m/(m/s) at 0.06 Hz,

whereas after the step c at 0.06 Hz resumes a nearly normal

value of 380 ± 19 m/(m/s) (Figure 4b). Since neither the

tidal amplification nor the well geometry changes with time,

the change in c must be due to a local change in K or, more

plausibly, the geometry of the fracture. Equation (3) shows

that much larger changes in K than L are required to achieve

a significant effect. The reduction in c before the step by a

factor of nearly 40 requires at least a 75% reduction in

fracture length L if all other parameters in equation (3) are

constant (Figure 4). As will be detailed below, a simultaneous

change in c and a step can occur if the seismic waves

remove a temporary low permeability barrier in the fracture

near the well. Removing the barrier both returns the

amplification to normal and allows the well to drain. A

local blockage in the fracture has only small effects on the

amplitude of the aquifer tidal signal as the long-period tidal

stresses sample the average wall rock (matrix) properties.

[13] Drops in water level can be sudden, as in the case of

Oaxaca, or gradual, as in the case of Petrolia (Figure 1). The

different drop durations correspond to differing distances

from the well to localized sources of pore pressure change

[Roeloffs, 1998]. For times much greater than the duration

of the seismic wave train, we model the effect of the

pressure step on the well as the one-dimensional solution

to a diffusion equation in an unbounded, homogeneous

aquifer [Crank, 1975]

W ¼ W0

p

rg

erfc d=

ffiffiffiffiffiffiffiffi

4Dt

p

; ð4Þ

where W is the water level, W0 is the initial water level, p

is the amplitude of the pressure drop at the source, d is the

distance from the source of the pore pressure change to the

fracture, t is the time since the drop at the source, and D

K/Ss is the hydraulic diffusivity [Freeze and Cherry, 1979].

On the basis of the pumping test and the seismic wave response, D in the granodiorite matrix is 0.2 m2 s1

(Table 1). A least squares fit of the Petrolia water level

record to equation (4) yields d = 70 m and p/rg = 22 cm

(Figure 5). The misfit at large times may be due to unrelated

seasonal trends or the flow encountering aquifer boundaries

not sampled in the 52 hour pumping test. Unfortunately,

there is no other monitored well sampling the same deep,

confined aquifer within a 1 km radius; therefore we cannot

triangulate to determine a more accurate location for the

pressure source.

[14] The process generating the step did not affect c in

the well during the Petrolia earthquake since d greatly

exceeds the pressure diffusion length scale for the seismic

waves (

ffiffiffiffiffiffi

Dt

p

= 2 m, where t is the dominant wave period of

21 s). For Oaxaca, the drop is a step function to within the

data resolution and the solution to the diffusion equation is

the trivial one, i.e., d = 0 and p/rg = 11 cm. Both steps

are consistent with a localized, instantaneous source of

pressure. The consistency suggests that the same mechanism

is active both near and far from the well, i.e., the steps

are not generated by the well bore itself.

3. Mechanism

[15] The static stress change at the well directly generated

by the 3850 km distant ( = 34.650) Mw = 7.4 Oaxaca

earthquake is less than 0.2 Pa. A static stress change of

103 Pa is required to explain the observed 11 cm water level

drop. Therefore we can eliminate static stress as the cause of

the drop and limit our investigations to a detailed study of

the dynamic stresses (seismic waves).

[16] On the basis of the observations we propose a new

model for water level changes in wells far from an earthquake

(Figure 6). The permeability structure is dominated

by highly conductive fractures. From time to time, these

fractures become clogged with weathering products and low

permeability flocs of colloidal material. The fluid pressure

on the upgradient (or upstream) side of the low-permeability

barrier in steady state increases relative to the unclogged

state (Figure 6b). In addition, the effective fracture size is

smaller and the value of c measured by an intersecting

well is therefore reduced. When a seismic wave passes, it

induces rapid flow between the formation and fracture

which removes the barrier by loosening particles and

entraining them. Once the barrier is removed, water drains

from the well to produce a step in pressure as the permeability

structure returns to its normal state. If the barrier

forms at some distance from the fracture intersected by the

well, the pressure change diffuses gradually to the fracture,

but if the barrier is immediately adjacent to the fracture,

then the observed drop is very rapid.

[17] Water pumped from deep in the aquifer contains

4 107 micron-size aluminosilicate particles per liter.

These suspended weathering products could aggregate to

form the requisite blockages. Dense clay flocs have conductivities

comparable to the wall rock value 7 108 m s1

[Freeze and Cherry, 1979] and could thus effectively block

the fracture. For simplicity, we assume that the floc and wall

rock conductivities are identical and the water flows

through both according to Darcy’s law,

u ¼ K=ðfrgÞrp; ð5Þ

where u is the interstitial fluid velocity and f is the porosity.

The Darcy (volumetrically average) velocity is fu. In order

to deposit a 0.3 mm thick barriers of densely packed

colloids in two years, we require a Darcy velocity of 5 m d1.

In steady state, the head drop across this barrier is 25 cm

(2.5 kPa of pressure). When the barrier is removed, the

upgradient side returns to the initial pressure P0 by dropping

by half the head difference, or 13 cm.

[18] The fluid velocity at the edge of the flocs adjacent to

the wall is of the same order as that draining the matrix by

continuity. The seismic waves induce a flow rate Qclog that

is observed during Oaxaca to be 5 105 m3 s1 during the

Rayleigh waves before the step. If the fracture thickness w is

1 mm and the clogged fracture length L is at most 30 m as

inferred from the response (Figure 4), the fluid velocity is

Qclog/wL = 1 mm s1. Viscous stress on the particles is

approximately hu/R, where h is the viscosity of water

(103 Pa s) and R is the radius of the particle (1 mm). The

interstitial fluid velocity u is at least the Darcy velocity, fu,

and can be substantially greater if f is small. The resulting

viscous shear is 1 Pa which is sufficient to initiate motion

and disaggregate flocs [Kessler, 1993]. Low permeability

clay flocs can have very high porosity (>99%), therefore

when the flocs are disaggregated, the separated micronscale

clay particles no longer have a significant effect on the

permeability [Kessler, 1993]. It is possible that the small

amplitude solid strains of the shaking also contribute

directly to loosening the barriers, but there is no need to

invoke such a difficult mechanism as the observed induced

flow velocities are sufficient for entrainment.

[19] The gradual Petrolia drop suggests that seismic

waves can induce floc-entraining flow velocities outside

the immediate vicinity of the well bore. Seismic waves

induce pressure gradients between any zones with different

values of , i.e., different compressibilities and porosities.

The head difference between two geological units can be

80% of the head difference between one of the units and the

well bore. The pressure difference p12 between two units

with tidal amplifications 1 and 2 in response to a

dilatational strain q is (1 2/1)1q. If a well is drilled

into the 1 unit, the pressure difference pw between the

unit and the well is 1q. Therefore the ratio p12/pw is

(1 2/1). The range of observed in nature is at least a

factor of 5 [Roeloffs, 1998]. In open fractures, is 1. As

long as these contrasts are sustained over sharp boundaries,

such as a fracture wall, flow will be on the order of that

observed in the well. In a fracture that does not intersect a

well, there will be no water column resonance, i.e., the

second term of equation (3) will be negligible. The absence

of water column resonance far from the well reduces flow

velocities by only 40% relative to those observed here.

[20] The Oaxaca earthquake was followed 2 weeks later

by shaking with 20% greater vertical ground velocity from

the Hector Mine earthquake (Figure 2). No rainfall occurred

between the events and all other observables were indistinguishable,

yet the second event produced no step. The

above model predicts that a barrier could not reform within

2 weeks as the particle concentration is too low. Therefore

no step is expected with the second earthquake despite its

size.

[21] One observation not directly addressed by the model

is that the pressure changes at the well are always drops.

An abundant source of weathering material or disequilibrium

precipitation may exist downgradient of the well.

Alternatively, the fracture may have an easily blocked

constriction downgradient. In either case, the downgradient

location of the blockage would then favor drops at the

well.

[22] Other mechanisms that have been suggested for farfield

coseismic pore pressure changes include mobilization of

gas bubbles [Linde et al., 1994; Roeloffs, 1998; Sturtevant et

al., 1996], shaking-induced dilatancy [Bower and Heaton,

1978] and fracture of an impermeable fault [King et al.,

1999]. The high tidal amplification implies that a compressible

gas phase comprises <104% of the aquifer; therefore

bubbles cannot account for the observed water level drop.

The other two mechanisms fail to explain the observed drop

in effective permeability followed by the return to the original

value. Both dilatancy and fracture models predict that the

effective permeability, and hence c, should increase immediately

following the earthquake. We observe that the poststep

response of Oaxaca is of the same order as the long-term

average.

4. The 3 November 2002 Mw = 7.9 Denali

Earthquake

[23] While this manuscript was in preparation, another

large earthquake generated a coseismic step at NVIP-3. The 3 November 2002 Denali earthquake (Figure 1) allowed us

to independently test the model already developed based on

the Oaxaca data.

[24] The Denali earthquake ( = 25.2) generated a

sudden drop in water level during the shaking (Figure 7).

The ambiguity in aligning the seismic and hydrologic

records is <5 s. Uncertainties in site effects were eliminated

since there was a Streickeisen STS-2 broadband seismometer

at the well site beginning in December 2001. Like

Oaxaca, the Denali record shows very little response to

shaking before the step and a normal response afterward.

However, unlike the previous case, the step occurred at the

beginning of the Rayleigh wave, therefore the responses

before and after cannot be directly compared as in Figure 4c.

For Denali, the water level is responding first to the S and

then to the Rayleigh wave. Shear phase coupling is observed

for other earthquakes and is probably due to an

anisotropic poroelastic response in the fractured rock

[Wang, 2000]. The Oaxaca earthquake appears to have been

very convenient in that it had a long, dispersed Rayleigh

wave which generated a constant excitation with a varying

response.

[25] Before the 12 cm step, the peak flow rate was 4

105 m3 s1, i.e., nearly identical to the 5 105m3 s1

value of Qclog observed for Oaxaca. The entrainment

threshold for Denali is consistent with that proposed for

Oaxaca.

[26] The 4 mm s1 shaking of the Denali earthquake is

some of the strongest shaking recorded on site. The last

earthquake with a normal oscillatory record at the well prior

to the Denali event was the 23 June 2001 Mw = 8.4 Peru

earthquake that produced 17 cm oscillatory motion in the

well in response to 0.7 mm s1 shaking (Figure 8). The

observation is consistent with the predicted amplitude of

19 ± 2 cm calculated from equation (3) using the unblocked

fracture parameters in Table 1. In contrast, the 23 October

2002 Mw = 6.7 foreshock to the Denali event shook the

Oregon site with an amplitude of 0.04 mm s1. Although

this shaking is smaller than most earthquakes we studied,

according to equation (3) with the values in Table 1, the

water level should have oscillated with an amplitude of 1 cm.

The observed water level was less than the 0.6 mm trigger

for high sample rate recording. This order of magnitude

suppression of water level shaking is consistent with the barrier model. If the barrier formed over the 1.4 year interval

between the Peru and Denali earthquake and all other

parameters are the same as in the above model, we predict

a 9 cm head drop could occur during barrier removal. This

result from our order-of-magnitude model is reasonably near

the observed value of 12 cm.

5. Implications for Earthquakes

[27] We have constrained here a naturally occurring

process that suddenly redistributes pore pressure on fractures.

Faults, like fractures, form hydrological boundaries

that have contrasting porosity (storage) with the surrounding

rock and can accumulate sediment. Seismically induced

pore pressure steps can occur by the mechanism proposed

here in any hydrogeological systems that has (1) low matrix

specific storage, (2) fractures or faults, and (3) a source of

material for clogging. At least one other well-studied site of

coseismic steps is in a granitoid pluton with conditions 1–2

documented and condition 3 likely [King et al., 1999].

[28] Recent studies of regional-scale seismic triggering

suggest that seismic waves generate pore pressure

changes in geothermal areas that in turn generate seismicity

[Hill et al., 1993; Brodsky et al., 2000]. Geothermal

systems satisfy all three conditions above. In

particular, rapid precipitation of minerals is more common

in geothermal areas than in ordinary hydrogeologic environments

because of the large temperature and chemical

gradients [Lowell et al., 1993]. We speculate that high

incidences of triggered seismicity have been observed in

geothermal systems [Hill et al., 1993; Brodsky et al.,

2000; USGS, 2000] because of unstable permeability

structures generated by hot, circulating fluids. Shaking

by seismic waves loosens new precipitate and readjusts

the pore pressure on faults and fractures as is observed in

the Grants Pass well. The rapid redistribution of pore

pressure may promote earthquakes by quickly reducing

the effective stress on faults locally. Although the NVIP-3

well records only water level drops, the barrier removal

model requires pore pressure increases on the downgradient

end of the fracture or fault.

[29] In geothermal areas under typical conditions, precipitation

rates can be a factor of 15 greater than modeled for

Grants Pass [Lowell et al., 1993] resulting in barrier thicknesses

and step amplitudes proportionally larger. If all other

factors are equal to what we observe and model above, the

pressure change on a fault in a geothermal area will be 4

102 MPa which is sufficient to trigger an earthquake

according to static stress studies [Hardebeck et al., 1998].

6. Conclusions

[30] The water well record from the Oaxaca earthquake

presented here is the first high sample rate recording of a

rapid far-field coseismic well step ever published to the best

of our knowledge. Such records are rare because (1) only a

small fraction of wells show far-field coseismic steps and

(2) hydrogeologists normally record water level at rates no

greater than 2 103 Hz. We analyzed this unique water

well record in conjunction with seismic data to show that during a rapid drop in water level, as the well drains, the

response c increases from the unusually low value of 9 ±

2 m/(m/s) to a more normal value of 380 ± 19 m/(m/s). We

interpret the change as the removal of a temporary blockage

in a fracture. Whether or not a step occurs depends on the

preexisting hydrogeology as well as the seismic input.

[31] Acknowledgments. This work was supported in part by a NSF

Earth Science Postdoctoral Fellowship and the Miller Institute for Basic

Research. We thank N. Beeler, D. Hill and S. Rojstaczer for reviews of an

early form of this manuscript. J. Barker gave insights into interpreting the

pumping test data, P. Hsieh advised on flow in fractured rocks and C. Chen

contributed SEM imaging of the particulates. Data from Yreka Blue Hill are

courtesy Berkeley Digital Seismic Network. PASSCAL instrument center

provided the STS-2 seismometer for the 2002 installation.

The equations need to be corrected before an audio can be made. Otherwise it will not make sense.