Professional Documents
Culture Documents
Saar and Manga (PDFDrive)
Saar and Manga (PDFDrive)
com
"e
8CIENCm~DIREC’I
EPSL
ELSEVIER Earth and PlanetaryScienceLetters 214(2003)605-618
www.elsevier.com/locate/epsl
Abstract
Groundwaterrecharge at Mt. Hood, Oregon, is dominated by spring snow melt which provides a natural large-
amplitude and narrow-widthpore-fluid pressure signal. Timedelays betweenthis seasonal groundwaterrecharge and
seismicity triggered by groundwaterrecharge can thus be used to estimate large-scale hydraulic diffusivities and the
state of stress in the crust. Weapproximateseasonal variations in groundwaterrecharge with discharge in runoff-
dominatedstreams at high elevations. Weinterpolate the time series of numberof earthquakes, N, seismic moment,
Mo,and stream discharge, Q, and determine cross-correlation coefficients at equivalent frequency bands between Q
and both N and Mo.Wefind statistically significant correlation coefficients at a meantime lag of about 151 days.
This time lag and a mean earthquake depth of about 4.5 kmare used in the solution to the pressure diffusion
equation, under periodic (1 year) boundaryconditions, to estimate a hydraulic diffusivity of 1¢= -l m2/s, a
hydraulic conductivity of about Kh=107 m]s, and a permeability of about k=10-L5 m2. Periodic boundary
conditions also allow us to determinea critical pore-fluid pressure fraction, P’/Po = 0.1, of the applied near-surface
pore-fluid pressure perturbation, P0 = 0.1 MPa,that has to be reached at the mean earthquake depth to cause
hydroseismicity. The low magnitudeof P’ =0.01 MPais consistent with other studies that propose 0.01 -< P’ -< 0.1
MPaand suggests that the state of stress in the crust near Mt. Hoodcould be near critical for failure. Therefore, we
concludethat, while earthquakesoccur throughoutthe year at Mt. Hood,elevated seismicity levels along pre-existing
faults south of Mt. Hoodduring summermonthsare hydrologically induced by a reduction in effective stress.
© 2003Elsevier B.V. All rights reserved.
Keywords:hydroseismicity;
groundwater;
pore-fluidpressure;permeability;effectivestress; volcano;triggering; stress; earthquake;
recharge
0012-821X/03/$
- see front matter ©2003ElsevierB.V.All rights reserved.
doi: 10.1016/S0012-821
X(03)00418-7
606 M.O. Saar, M. MangalEarthand Planetary Science Letters 214 (2003) 605418
’-121°50 ’-121030
-121040’ [20]. Moregenerally, understanding the causes
of hydroseismicity mayprovide additional insight
into seismic hazards associated with reservoir
impoundment or injection of fluids into the sub-
surface for exampleduring waste fluid injection,
carbon sequestration, and geothermal energy
exploration. Detection of natural groundwater-re-
charge-induced earthquakes also provides a base-
line against whichreservoir- or fluid-injection-in-
duced seismicity can be measuredto evaluate the
actual extent of human-causedhydroseismicity.
Moreover,natural hydrologic triggering of earth-
quakesprovides insight into the state of stress in
the crust. In addition to tectonic information, the
time lag between groundwater recharge and earth-
[]
[] quakes allows us to determine hydraulic proper-
ties of the upper crust on spatial and temporal
C
scales that are relevant for studies of regional hy-
drogeology.
Fig. la shows seismicity at Mt. Hood, Oregon,
nl 2 4 6
7:4.5
8
+ 2kin
10 -- a Quaternary stratovolcano, for the period from
depth(kin) February 4, 1980 through July 11, 2002. Earth-
Fig. 1. (a) Shadedrelief mapof Mt. Hood,Oregon, indicat- quakes occur preferentially on the southern flanks
ing all earthquakes (pink circles and red squares) and se- of the volcano. The off-axis concentration sug-
lected (for criteria see text) earthquakes(red squares), stream gests the possibility that earthquakes are not all
discharge gauging station on SalmonRiver (blue star), sum-
mit of Mt. Hood(large white triangle), and hydrothermal
SwimWarmSprings (western white diamond) and Meadows 1000
Spring (eastern white diamond). Three (small yellow trian-
gles) out of 14 relevant seismometers (AppendixA) are
cated within the mapregion. (b) Seismic moment,Mo,versus
earthquake depth. (c) Histogram of earthquake depths, mean 1 O0
earthquake depth (4.5 km), and range of lc~ standard devia-
tion (2 km). c
o
10
recharge rates [14-16] which can be the result of
seasonal snowmelt or variations in precipitation
[16-19]. In this paper we use signal processing
techniques to investigate whether natural ground- 1
-2 -1 0
,
2 3
, _4
water recharge variations trigger some earth- Mcma~ nitude
quakes in the Cascade volcanic arc, USA.
Fig. 2. Earthquake magnitude, M], binned in half-magnitude
The Cascade arc is located on a tectonically increments, versus logarithmic frequency of occurrence (Gu-
active, and in parts densely populated, convergent tenberg-Richter relationship) for all earthquakesdisplayed in
plate boundary. Thus, for both seismic and vol- Fig. 1. Opencircles, squares, and solid circles indicate seis-
canic hazard assessmentit is important to discern micity during the summer(15-< week-<40,open circles) and
triggers of earthquakes as tectonic, magmatic,or winter (week-41 or week-<14, open squares), and total
years, respectively. Abovethe cut-off magnitudeof M]= 1
hydrologic. Moreover, rapid groundwater re- the decrease in seismicity with increasing magnitudefollows
charge may initiate larger (local magnitude, approximately the expected slope of b =-1. This suggests
M1> 3) earthquakes on critically stressed faults that the data set is completefor M]-> 1.
M.O. Saar, M. MangalEarth and Planetary Science Letters 214 (2003) 605418 607
~Ee- 7-
e-4-
8
E
0-
intervalI intervalI+II
charge at HoodRiver is available until September 1 , ¯
~"’
. .
¯
30, 2001 and is convolvedwith the transfer func-
0.8. "",
tion to extend the discharge data to that date
(Fig. 3). ¯ ¯
o
0g- : ",.:
400
period(days)
600 0 200 400
period(days)
600
ends whendischarge measurements at HoodRiv- Fig. 5. Period content of time series. Ten-day running aver-
er were discontinued on September30, 2001 (Fig. age power spectra, normalized by their respective maximum
3). Therefore, our analysis focuses on interval I, power, of stream discharge, Q (bold solid line), number
earthquakes, N (thin solid line), and seismic moment,
to allow evaluation of the best-constrained data, (thin dashed line). Dominant periods around one year occur
and on the combinedinterval I+II, to investigate for both interval I (a) and interval I+II (b).
the longest possible time series.
Fig. 4a,b shows histograms for stream dis-
charge, Q, numberof earthquakes, N, and seismic N, and Mohave yearly periodicity, we determine
moment,Mo, binned monthly for intervals I and their power spectra which indeed show dominant
I+II, respectively. Elevated levels of stream dis- periods of about one year (Fig. 5) for both inter-
charge exist from Novemberthrough June with vals I and I+II.
a peak in May due to snow melt. Number of Rather than using monthly binned data aver-
earthquakes and total seismic moment show a aged over all years from Fig. 4 to determinecross-
peak during September and October for both in- correlation coefficients, weuse the actual time se-
tervals I and I+II. In addition, interval I+II ries at a resolution of one day. Seismicity is
shows seismicity in January and February. For binned to yield daily numberof earthquakes, N,
interval I, the time lag betweenpeak stream dis- and total daily seismic moment, Mo. Next, we
charge, a proxy for groundwater recharge, and apply movingleast-squares polynomial fits (Ap-
seismicity is about 5 months. To test whether Q, pendix C) of order -< 5 to Q, N, and Mo(Fig.
6), resulting in the interpolated time series de-
°°°la ’ ’-- ’ ,°,erva,’,,---A
21
’-;
Ax O
12° noted Q, N, and Mo, respectively. This type of
~,o0) ,/No ,’ /’\ ~i2 ~1" interpolation ensures that the data are optimally
°t
~0")400
/ \,,
.~3ool-----~
"o N
I
"/\,\’,k_. --’"
i/’ 5-1 ~41o
e=:I,5 ~= 15 matched in a least-squares sense and that later
determinationof cross-correlation coefficients be-
o:
tween time series is based on equivalent frequency
~,oo / / ~~ .... .,’ _ , L__\~i bands. Moreover, interpolation of N and Mopro-
~800 .... 21 61)
~ ’,x, I0
¢~ b
l ~ /~’~~
~"E700r t,3 /~
¯
intervalI+ll ,{- ". o,
~2.5 ~150
/ vides evenly spaced data allowing for standard
--/ ~
o 6oo ~ i ~ ~2 m 40 spectral analysis techniques. For better temporal
comparison, the interpolated data from Fig. 6
..oI ", /\ \ ’/ X (black curves) are shownnormalized by their re-
,oo spective maximum values in Fig. 7.
J F M A M J J A S O N D
1io. 10
o o For a lead channel f and a lag channel g, the
month
normalized cross- (f¢g) and auto- c= g) c orrela-
Fig. 4. Monthly binned histograms showing cumulative mean
tion coefficients, ~fg(t), at a time lag of t, are
daily stream discharge (bold solid line), cumulative number
of earthquakes (thin solid line), and cumulative seismic mo-
given by:
¢fg(t) <1
ment (thin dashed line) for (a) interval I and (b) interval
I+II.
)-l_<<t>/~(t)= ~/v,,77707v/TT~(o (3)
M.O. Saar, M. Mangal Earth and Planetary Science Letters 214 (2003) 605418 609
where: c
o)
~)fg
/2,r) g(t r) dr (4) ~
t-
0.6
0.8
interval
I intervalI+11 where"rs, r0,/1 are the fault’s shear strength, co-
hesion, and coefficient of friction, respectively,
~ 0.7 and O’n is the effective normal stress across the
8¢o 0.6 fault. Throughoutthis paper compressivestresses
8 are positive, dn clampsthe fault requiring larger
shear stresses, z, to induce failure (Fig. 10) and
=~0.s
o
given (Appendix E) by the effective principal
E
¢~ 0.4
stress tensor:
E o’ = o--ote~ij (6)
0.3
0 50 100 150 200 250 0 50 100150 200250
timelag(days) timelag (days) where~ is the (regular) principal stress tensor,
Fig. 9. Timelag versus the meanof the unbiased normalized is the pore-fluid pressure, ~0 is the Kronecker
movingcross-correlation coefficients ~ (bold solid line) delta, and a is the Biot-Willis coefficient, defined
and ~oo (bold dashed line). Thin horizontal lines indicate as:
the respective upper bounds of the 99%(black) and 90%
(white) confidence intervals of over 2000 cross-correlations a : 1--K/Ks (7)
for randomlyassigned phases for each lag channel (Appendix
D). Lower bounds are comparable to the negative of the Here, K and Ks are the bulk moduli (incompres-
upper bounds. Vertical lines at maximum coefficient values sibilities) of the bulk rock matrix and the solid
indicate mean(center line) and mean+lg standard deviation ’grains’, respectively [25]. Deformationcan lead
(outer lines) for the respective time lags determinedfrom to failure, whenthe shear stress, r (Appendix
(a) and 109 (b) moving windowsof 3-year width. The E), exceeds the material’s shear strength, rs, in
meantime lags and respective standard errors are (in days):
(a) 157+2.7(solid line), 152+2.1(dashed line); (b)
(solid line), 139 + 3.9 (dashedline), resulting in a total mean
and standard error of 151 + 6.6 days. a I
I
I
ti__melag (bold lines in Fig. 9). For both ~--g and
~-o a maximumis reached for a time lag of
about F= 151 days.
3. Discussion
Eq. 5, for a given effective normalstress tin. Be- the Earth’s gravity, and h is the height of the
cause dn and ~" dependon both orientation of the water columnabove the point of interest. Reser-
fault within a stress field and on the magnitudeof voirs exhibit both processes and thus typically
the stresses (Eqs. E1 and E2), failure can be trig- cause earthquakes both concurrent with, and de-
gered by changes in both 0 and d (Fig. 10). The layed from, their filling [2,6]. In this paper weare
latter mechanism may induce earthquakes by interested in seismicity induced by groundwater
causing: (1) an increase in if’l, (2) a decrease recharge and are thus focusing on earthquakes
6’3, or (3) an equal decrease in all effective prin- delayed from recharge by pore-fluid pressure dif-
cipal stresses, for exampledue to an increase in P fusion. Here, the load of the additional ground-
(Eq. 6). The first two mechanisms cause failure water is small and thus triggered seismicity due to
increasing the effective differential stress, dl-d3, a (small) increase in d l is neglected. Nonetheless,
and thus shear stress, ~-, in Eq. E2. The third because only the connected pore space is fluid-
mechanisminduces failure by decreasing the effec- filled, relatively high hydrostatic pore-fluid pres-
tive strength of the material caused by a decrease sure changes, P’, may be reached that may be
in Oan which unclamps the fault and moves the sufficient to induce earthquakes.
Mohrcircle closer to the Mohr-Coulomb failure
envelope, while leaving r unchanged(Fig. 10). 3.2. Analytic model
Fluctuations in pore-fluid pressure, P, denoted
P’, on faults due to an increase in dl (e.g., by the We can approximate the effect of seasonal
weightof a filling reservoir that contracts the pore groundwater recharge with periodic yearly (N=
space) occur immediately.In contrast, a local in- 1 year) pore-fluid pressure variations of amplitude
crease in hydrostatic pore-fluid pressure, Ph = P0 at the surface (z = 0 m) as:
pwgh, (e.g., by groundwater recharge, reservoir
impoundment,or injection of fluids at depth), F(t,z=O)= P0cos (~) (8)
maytrigger earthquakes after a time lag, F, that
is related to pressure diffusion to the fault. Here, Pore-fluid pressure evolution below the water
Pwis the density of water, g is acceleration due to table is governedby the (pressure) diffusion equa-
::~ . . tion. In a one-dimensionalhalf-space the diffusion
0.8 equation is given by:
0.6 ~o 02p ’ dP’
(9)
0.4 ’I Oz~ -- dt
0.2
13o ::
"~z=_ wherethe hydraulic diffusivity:
~" 0
’4
s
r -Kh -
Ss
gk
vSs
(i0)
°’I117 /7 IV" is assumedconstant. Here, Kh, Ss, k, and v are the
ti/ // hydraulic conductivity, specific storage, perme-
t/ , i/ , tJ;
-°::t
0 200 400 600 800 1000
ability, and kinematic viscosity, respectively. The
solution to Eq. 9, with the boundary condition
days given by Eq. 8 is [26]:
Fig. 11. Periodicpore-fluidpressurefluctuations,P’/Po,at
variousdepths,z (in km),for hydraulicdiffusivity~=0.3 P00 = exp -z cos -z (11)
mZ/s andsurfacepressure perturbationperiodicityi/t= 1 year.
Withincreasingdepth,the amplitude of the pore-fluidpres- and is graphed in Fig. 1 1 for N=1 year and hy-
sureperturbation decreasesandthe phaselag increases.Ata draulic diffusivity to= 0.3 m2/s as determinedlater
depthof 4.5 kin, approximately 10%of the originalpressure
amplitude,P0, remainsandthe delayof the peakis about in Eqs. 13 and 14 for the study region. The ex-
151days(dashed line), i.e., a phaselag of about0.8 ponential term in Eq. 1 1 describes the decrease in
612 M.O. Saar, M. Manga/ Earth and Planetary Science Letters 214 (2003) 605418
(12)
smaller than maybe expected. Due to the large al fluid pressure variations that may exceed
uncertainty in earthquake depth ( + 2 km)already P0 = 0.1 MPa. Therefore, according to Eq. 15,
considered and due to the cancelling effects of the critical pore-fluid pressure increase at depth
reducing both z and t, we neglect the effects of z=4.5 km may be as low as P’=0.01 MPa,
water table reduction by a possible 0.5 1 kmrel- and possibly lower, at Mt. Hood.
ative to the surface. Whileseveral studies mentionedin Section 1, as
Using N= 1 year, t= 151 days, and z=4.5 km well as this paper, suggest that someearthquakes
in Eq. 13 yields a hydraulic diffusivity of: are triggered by pore-fluid pressure perturbations,
the necessarycritical pore-fluid pressure increase,
t¢ = 0.30 + 0.22 m2/s (14)
P’, is uncertain [1]. However,it is often argued
This value of t¢ agrees with other crustal hydraulic that manyfaults are near critically stressed [36].
diffusivities compiled by Talwani and Acree [1] Therefore, small stress changesinvoked by a vari-
and is in the upper range for fractured igneous ety of different mechanismsmaybe sufficient to
rocks [29]. Gao et al. [30] suggest a value as cause seismicity on somepre-existing faults. Ex-
high as 1 mZ/s for some fractured volcanic sys- amples include earthquake triggering by solid
tems. Earth and ocean tides [37 39], seasonal modula-
tions of seismicity by the load of snow[40], pre-
3.4. Critical pore-fluid pressurechange cipitation- and snow melt-induced seismicity
[16,19] and general fluid-driven seismicity [41].
Thestep-function characteristics of a pore-fluid Lockner and Beeler [42] conduct laboratory stud-
pressure increase due to initial reservoir impound- ies of rock failure along pre-existing faults in-
mentand its error function solution [1,31] do not duced by periodic axial stress changes superim-
allow for evaluation of the critical value of P’/Po posed on a confining pressure of 50 MPa. They
at whichfailure occurs [1]. In contrast, the addi- find a transition from weakto strong correlation
tional temporal information provided here by the between periodic stress and failure at an ampli-
periodicity of the surface pressure perturbation in tude of 0.05 0.1 MPashear stress. Roeloffs [6,29]
the case of seasonal groundwaterrecharge allows suggests that an increase of P’=0.1 MPacan
for estimation of critical P’/Po. Becausefailure is cause reservoir-induced seismicity. Similarly, pos-
likely to occur whenthe cosine term in Eq. 11 is sible triggering of earthquakes by static stress
one, we can determine P’/Po for the above calcu- changes of about 0.1 MPaand less has been pro-
lated value of r=0.3 m2/s at a mean depth of posed by Kinget al. [43] and Stein et al. [4446]
about 4-<z-<4.5 kmand period ~= 1 year as: Harris [47] summarizesthe work of several au-
thors and states that "It appears that static stress
--
P0
z ox( changes as low as 0.01 MPa(0.1 bar) can affect
the locations of aftershocks".
Thus, about 10%of the estimated amplitude of The examples cited previously as well as our
near-surface pressure variations appears to be suf- results suggest that the necessary stress change
ficient to trigger hydroseismicity at Mr. Hood. for earthquake initiation on pre-existing faults
Average annual precipitation in the (Oregon) may be as low as 0.01~0.1 MPawhich is only a
Cascades, and particularly at Mt. Hood, is about fraction of the coseismic stress drop. Therefore,
3 m [32] of which approximately 50%infiltrates Harris [47] points out that Coulomb stress
the ground [33,34] mostly during spring snow changes are said to ’enhance’ the occurrence of
melt. This leads to infiltration rates of about 1.5 an earthquake, as opposed to generating it. We
m/year concentrated during a few months. Lava support this view and suggest that pore-fluid pres-
flows in the region typically have near-surface po- sure changes due to groundwater recharge at Mt.
rosities of about 15%[35] that decrease with Hoodincrease the probability of seasonal earth-
depth. Thus, groundwaterlevels mayfluctuate an- quake occurrences.
nually by approximately10 m resulting in season- Heki [40] states that seasonal seismicity is not
614 M.O.Saar, )l/L MangalEarth and Planetary Science Letters 214 (2003) 605-618
expectedif the rate of secular (long-term)regional draulic diffusivity range given in Eq. 14, the hy-
stress increase is much larger than the annual draulic conductivity is:
superimposed disturbance. For the case of the 8 x 10-8--<Kh= t¢S~-<5x 10-7 m/s (17)
Cascades subduction zone, and assuming the val-
idity of a characteristic earthquakemodel[48], the which, from Eq. 10 and for a kinematic water
annual secular stress increase maybe estimated by viscosity of v~ 10-7 m2/s at 80°C, corresponds
the (approximately constant) coseismic stress to a permeability range of:
drop, Aa~10 MPa,divided by the average recur- 8)< 10-16-<k<-5× 10-15 2m (18)
rence interval, trY300 years, of large earth-
quakes. Estimatedannual pore-fluid pressure fluc- for the upper ,~ 6 kmof the crust at Mt. Hood.
tuations of 0.01 MPa/yearat Mt. Hoodat a mean Thesuggested permeabilities are relatively high
earthquake depth of z = 4.5 kmare comparableto compared with a maximumof k= 10 -16 m2 re-
Ao’/tr =0.03 MPa/year. Therefore, the pore-fluid quired for mostly conductive heat transfer [52]
pressure fluctuations below Mt. Hoodat depths as suggested for the Oregon Cascades at depths
of about 4.5 km caused by groundwater recharge below about 2 km. However, we suggest two pos-
may be large enough to induce seasonal hydro- sible argumentsthat potentially explain whyk is
seismicity. higher than maybe expected.
The concentration of (hydro-)seismicity south First, a reasonable assumption is that perme-
of Mt. Hood may be due to preferential occur- ability is much higher at shallower depths and
rences of active faults in this region [28,49,50]. decreases with depth due to compaction. There-
Jones and Malone [50] performed earthquake re- fore, pore-fluid diffusion is muchfaster in the
locations and determined focal mechanisms of shallower section and most of the observed time
events from the June-July 2002 swarmwhich re- delay probably occurs at deeper portions of the
veal distinct clusters along normal faults (e.g., profile. Reducingz for exampleby a factor of two
White River fault) located south of Mt. Hood in Eq. 13 woulddecrease to, Kh, and k by a factor
as well as events located beneath the summit of four.
that appear to be magma-related. This relation Second, wesuggest that the region south of Mt.
between snow melt, groundwater recharge, and Hood is anomalous with respect to conductive
seismicity on faults that are kept relatively perme- heat transfer because it hosts the only two hot
able mayalso be reflected by the occurrenceof the springs observed on the flanks of Mt. Hood to
only hydrothermal springs (SwimWarmSprings, date [28]. The presence of hot springs suggests
MeadowsSpring) at Mt. Hood[27,28] in the re- that geothermalfluids rise relatively quickly along
gion south of the volcano(Fig. l a). higher-k paths (e.g., faults) so that time scales for
complete conductive thermal equilibration be-
3.5. Hydraulic conductivity and permeability tween hot fluids and colder surrounding rock
are larger than the travel times of fluids. Indeed,
Oncethe hydraulic diffusivity, to, is known,the Forster and Smith [21] suggest that hot springs
hydraulic conductivity, Kh, permeability, k, or can only be expected for a permeability window
specific storage, S~, can be determined by Eq. of 10-17 -<k-10-15 m2. Our permeability values
10. Thespecific storage is given by: from Eq. 18 fall in the upper range suggested for
hot springs by Forster and Smith [21]. Here, high
Ss = pg(a + nil) (16)
permeabilities lead to increased water fluxes that
where a is the bulk aquifer compressibility (at result in lower geothermal spring temperatures
constant vertical stress and zero lateral strain), that are consistent with those (5°C to about
is the pore fraction, and fl=4.8× 10-1° m2/Nis 25°C [28]) observed at MeadowsSpring and
the compressibility of water. If we assumea mean SwimWarmSprings.
-1°
n=0.05 for the study region [33] and a= 10 The magnitudesfor (mostly vertical) permeabil-
m2/Nfor fractured rock [6,51], then, for the hy- ity, k, calculated here, reflect values for a large
M.O. Saar, M. MangalEarthand Planetary Science Letters 214 (2003) 605418 615
spatial scale that includes Mt. Hoodand its near- rice of Science, of the U.S. Departmentof Energy
by active (normal) faults that are kept permeable under Contract No. DE-AC03-76SF00098 and by
by rupture and appear to provide fluid pathways. the Sloan Foundation. Wethank J. Rector for
That the magnitudesof hydraulic diffusivity, ;¢, helpful discussions related to signal processing,
and permeability, k, are plausible further supports J. Kirchner for suggesting the approach to deter-
the hypothesis of seasonal elevated seismicity lev- mine confidence intervals on correlation coeffi-
els due to groundwaterrecharge in this region. cients, M. Singer for helpful insights into Box-
Jenkins methods, and E. Brodsky for discussions
about catalog completeness. Wealso thank C.
4. Conclusions Paffenbarger for efficient computersystem admin-
istration. The reviewers, K. Heki, S. Ingebritsen,
Several arguments support the hypothesis that and L. Wolf, are thanked for their excellent com-
someseismicity at Mt. Hood,Oregon,is triggered ments and suggestions. Earthquake and stream
by pore-fluid pressure diffusion as a result of rap- discharge data were obtained from the Pacific
id groundwater recharge due to seasonal snow Northwest Seismograph Network (PNSN) and
melt. Statistically significant cross-correlation co- from the USGS,respectively. The shaded relief
efficients between groundwaterrecharge and seis- map was made using the Generic MappingTools
micity suggest a time lag of about 151 days. The (GMT)and digital elevation models (DEM)
values of the correlation coefficients are distinct the USGS.[SK]
from those for random temporal earthquake dis-
tributions determined by Monte Carlo simula-
tions. The time lag and mean earthquake depth AppendixA. Seismometerinformation
provide a reasonable (mostly) vertical hydraulic
diffusivity (to= 10-1 m2/s), hydraulic conductivity The information in Table A1 about the short-
(Kh~ 10-7 m]s), and permeability (k~ 10-15 2) m period vertical-motion seismometersin the vicin-
that agree with other studies [1,21,36]. Finally, we ity of Mt. Hood was provided by the Pacific
determinethe critical fraction (P’/Po = 0.1) of the Northwest Seismograph Network. The date col-
periodic near-surface pore-fluid pressure fluctua- umnrefers to the seismometerinstallation date.
tions (P0 = 0.1 MPa)that reach the meanearth-
quake depth, suggesting that P’ =0.01 MPacan
trigger seismicity. This value of P’ =0.01 MPais AppendixB. Box-Jenkins method
at the lower end of the range of critical pore-fluid
pressure changes of 0.01-< P’ -< 0.1 MPasug- The stream discharge of Salmon River is pre-
gested in previous studies [6,29,43,47]. Therefore, dicted beyond its last measurementby employing
while seismicity is distributed throughoutthe year a transfer function between HoodRiver and Sal-
at Mt. Hood, we conclude that some earthquakes monRiver. The transfer function is determined
are hydrologically induced by a reduction in ef- using a Box-Jenkins [23] method. In the case of
fective stress due to a seasonal increase in hydro- a single input channel, Xt, investigated here, the
static pore-fluid pressure, Ph, by a small amount output (or lag) channel, Yt, can depend on both
P’. In fact, elevated seismicity levels due to such current (t = 0) and previous (t < 0) input, Xt, as
small effective stress changessuggestthat the state well as on previous output, Yr. Thus, input and
of stress in the crust at Mt. Hood,Oregon,could output are related by:
be near critical for failure. co(B)
Yt = ~(B)at-b + (B1)
Acknowledgements where:
og(B)_ 090+ o91B+ 0)2 B2 + ... + nog,,B
This work was supported by the Director, Of- ~" (B2)
~(B) ~o -’k 81B + ~zBZ +... + ~nB
616 M.O. Saar, Mi Manga/Earth and Planetary Science Letters 214 (2003) 605418
Table A1
Seismometerlocations and installation dates
Key Latitude Longitude Elevation Date
(km) (mrrdyy)
KMO "45o38’07.80 "-123°29’22.20 0.975 09/82
SSO 44°51’21.60" - 122°27’37.80" 1.242 09/91
TDH 44° 17’ 23.40" - 121047’25.20" 1.541 09/82
VBE 45o03’37.20" -- 121035’ 12.60" 1.544 10/79
VCR 44°58’58.18" ’
- 120059 17.35" 1.015 08/83
VFP 45° 19’05.00" -- 121°27’54.30" 1.716 10/80
VG2 45009’20.00" - 122° 16’ 15.00" 0.823 09/85
VGB 45o30’56.40" - 120o46’39.00" 0.729 04/80
VLL "45o27’48.00 ’-121o40’45.00" 1.195 10/80
VLM 45032 ’ 18.60" - 122002’21.00" 1.150 06/80
PGO "45027’42.60 - 122027 ’ 11.50" 0.253 06/82
AUG ’10.00"
45044 - 121°40’50.00" 0.865 10/81
GUL 45° 55’27.00" - 121°35’44.00" 1.189 07/86
MTM 46001’31.80" - 122° 12’42.00" 1.121 03/80
is the transfer function with coefficients o) and 6. be performed. We apply a moving polynomial
Becausecoefficients in ~o act on current and pre- interpolation, rather than convolution of the
vious input and coefficients in ?’ act on previous data with a Gaussian normal curve, so that no
output, they may be denoted as moving average artificial frequencyis introduced by a convolution
(MA)and auto-regressive (AR) processes and kernel. In addition, the moving polynomial ap-
transfer function as a so-called ARMA model. In proach allows interpolation with low-order (-< 5)
the previous equations, t is a time index, B is the polynomials while ensuring that the data are op-
backshift operator such that l~Xt = ~’~t-b, 17 and 7/ timally matchedin a least-squares sense. For all
are the number of delays for input and output, three series, Q, N, and Mo,the width of the mov-
respectively, and Nt is noise assumedunrelated to ing windowis 1/12th of the total series length and
the input. Therefore,if the noise, Aft, is neglected the step width is 1/10th of the windowwidth. The
and (in our case) 77 = 4 and ?’= 1 is selected then polynomialcoefficients, mLS,are found using the
the output, Yt, at time t is given by: standard least-squares solution from inverse
¢~oYt = ~1Yt-1 + (-DOeSt q- OklXt-1 -~- o)2Xt-2
theory given as mLS=[GTG]-IGTd, where G is
the (up to) fifth-order polynomial model matrix
-~-(-03Xt-3 q.- o)4X’t-4 (B3) and d is the data vector. Multiple interpolation
values for a given time step result from the over-
where ~0 = 1 is commonlyassumed(or else Eq. B3 lap of the moving windowand are averaged.
is divided by ~0). Therefore, in this example,the
current output depends on one (b= 1) previous
output (Salmon River discharge) as well as Appendix
D. Correlationcoefficients
the current (b = 0) and previous four (1- b-<
inputs (HoodRiver discharge). Wedetermine auto- (f= g) and cross- 0C¢g)cor-
relation coefficients for time series f and g as de-
scribed in principle by Eqs. 3 and 4. In practice,
AppendixC. Movingpolynomialinterpolation correlation coefficients, Ofg, for positive time lags,
l, are calculated by:
The daily binned seismic data have to be inter- 1 M-/-1
polated so that continuous time series are ob- ¢)fg(l) -- M-l Z fm+tgm (D1)
tained on which standard spectral analyses can m=0
M.O. Saar, M. Manga/Earthand Planetary Science Letters 214 (2003) 605<618 617
where m and Mare the index and the length of tively, and 0 is the angle between ~1 and the
the (zero-padded)time series, respectively. Divid- normal to the failure plane (Fig. 10a). The shear
ing by (M-/) provides so-called unbiased cross- stress, r, is given by:
correlations where the reduced overlap length of
the series for large time lags is accounted for. r = O’ll~
-- sin(20) (E2)
Correlation coefficients are then normalizedas de-
scribed by Eq. 3.
Wedetermine confidence intervals for the case
where one of the two time series is assigned ran- References
domphases for each frequency so that its auto-
[1] P. Talwani, S. Acree, Pore pressure diffusion and the
correlation is unaffected. Typically 500 iterations mechanismof reservoir-induced seismicity, PAGEOPH
(each assigning new randomphases for each fre- 122 (1984/85) 947-965.
quency)of cross-correlations, Cfg(t), at lag t, are [2] D.W. Simpson, W.S. Leith, C.H. Scholz, Twotypes of
performed by spectral multiplication directly in reservoir-induced seismicity, Bull. Seismol. Soc. Am.78
(1988) 2025-2040.
the frequency domainas: [3] H.K. Gupta, Reservoir-Induced Earthquakes, Elsevier,
e)fg(t) -t {F{f}F*{g) } (D2) NewYork, 1992, 364 pp.
[4] C.B. Raleigh, J.H. Healy, J.D. Bredehoeft, Faulting and
where F, F-1, and F* denote the Fourier trans- crustal stress at Rangely, Colorado, in: Flow and Frac-
form, its inverse, and its complexconjugate, re- ture of Rocks, Geophys.Monogr.Ser., 1972, pp. 275-284.
[5] C.B. Raleigh, J.H. Healy, J.D. Bredehoeft, Anexperiment
spectively. The range of values that contain 90% in earthquake control at Rangely, Colorado, Science 191
of the data for a given time lag is the 90%con- (1976) 1230 1237.
fidence interval for that time lag. [6] E.A. Roeloffs, Fault stability changesinduced beneath a
To reduce the effect of years with unusually reservoir with cyclic variations in water level, J. Geophys.
high seismicity, movingcross-correlation coeffi- Res. 93 (1988) 2107-2124.
[7] J.H. Healy, W.W.Rubey,D.T. Griggs, C.B. Raleigh, The
cients within overlapping subsections (windows) Denverearthquakes, Science 161 (1968) 1301-1310.
of the time series are also determined.Here, based [8] P.A. Hsieh, J.D. Bredehoeft, A reservoir analysis of the
on all windowsthat provide output for a given Denverearthquakes: A case of induced seismicity, J. Geo-
time lag, the meancoefficient per time lag is de- phys. Res. 86 (1981) 903-920.
termined. Each windowhas to be short enough so [9] P. Segall, Stress and subsidenceresulting fromsubsurface
fluid withdrawalin the epicentral region of the 1983Coal-
that the years with dominantseismicity do not fall
inga Earthquake, J. Geophys. Res. 90 (1985) 6801-
in all windowsbut large enough so that cross- 6816.
correlations are performed over several periods [10] P. Segall, Earthquakestriggered by fluid extraction, Geol-
(we chose 3 years). Furthermore, the time lags ogy 17 (1989) 942 946.
of coefficients that are distinct from the confi- [11] M.D. Zoback, J.C. Zinke, Production-induced normal
faulting in the Valhall and Ekofiskoil fields, Pure Appl.
dence interval for random phase distributions Geophys. 159 (2002) 403420.
are used to calculate a meanand a standard de- [12] A. Nur, J.R. Booker, Aftershocks caused by pore fluid
viation for the (positive) time lag aroundthe first flow?, Science 175 (1972) 885-887.
local maximum coefficient. [13] K.W.Hudnut, L. Seeber, J. Pacheco, Cross-fault trigger-
ing in the November1987 Superstition Hills earthquake
sequence, southern California, Geophys. Res. Lett. 16
(1989) 199-202.
AppendixE. Normaland shear stresses [14] J.K. Costain, G.A. Bollinger, J.A. Speer, Hydroseismic-
ity: A hypothesisfor the role of water in the generationof
Theeffective normalstress, if’n, is given by: intraplate seismicity, Seismol. Res. Lett. 58 (1987) 4164.
[15] P. Roth, N. Pavoni, N. Deichmann,Seismotectonics of
O’tl Jr- O’t3 O"1--O"3COS(20)_~ the eastern Swiss Alps and evidence for precipitation-in-
o"n- ÷ (El)
2 duced variations of seismic activity, Tectonophysics207
(1992) 183-197.
wheredl and 0" 3 are the effective principal max- [16] L.W. Wolf, C.A. Rowe,R.B. Homer,Periodic seismicity
imumand minimumcompressive stresses, respec- near Mt. Ogdenon the Alaska-British Columbiaborder:
618 M.O. Saar, M. MangalEarthand Planetary Science Letters 214 (2003) 605418
A case for hydrologically triggered earthquakes?, Bull. charge of Quaternary volcanics in the OregonCascades,
Seismol. Soc. Am.87 (1997) 1473-1483. Water Resour. Res. 33 (1997) 1813-1822.
[17] M.-J. Jimrnez, M. Garcia-Fern~ndez,Occurrenceof shal- [35] S.E. Ingebritsen, R.H. Mariner, D.R. Sherrod, Hydro-
low earthquakes following periods of intense rainfall in thermal Systems of the Cascades Range, North-central
Tenerife, Canary Islands, J. Volcanol. Geotherm. Res. Oregon, U.S. Geol. Surv. Prof. Pap. 1044-L, 1994.
103 (2000) 463468. [36] J. Townend,M.D.Zoback, Howfaulting keeps the crust
[18] H. Ogasawara, K. Fujimori, N. Koizumi, N. Hirano, S. strong, Geology28 (2000) 399402.
Fujiwara, S. Otsuka, S. Nakao, K. Nishigami, K. Tani- [37] S. Tanaka, M. Ohtake, H. Sato, Spatio-temporal varia-
guchi, Y. Iio, R. Nishida, K. Oike, Y. Tanaka, Micro- tion of the tidal triggering effect on earthquakeoccurrence
seismicity induced by heavy rainfall around flooded ver- associated with the 1982 South Tongaearthquake of Mw
tical ore veins, PAGEOPH 159 (2002) 91-109. 7.5, Geophys.Res. Lett. 29 (2002) art. no. 1756.
[19] V. Lronardi, F. Arthaud, A. Tovmassian, A. Karakha- [38] W.S.D.Wilcock,Tidal triggering of micro earthquakes on
nian, Tectonic and seismic conditions for changes in the Juan de Fuca Ridge, Geophys. Res. Lett. 28 (2001)
spring discharge along the Garni right lateral strike slip 39994002.
fault (Armenianupland), Geodin. Acta 11 (1998) 85-103. [39] J. Kasahara, Tides, earthquakes, and volcanoes, Science
[20] M.D.Zoback,J.H. Healy, Friction, faulting, and in-situ 297 (2002) 348-349.
stress, Ann. Geophys.2 (1984) 689 698. [40] K. Heki, Snowload and seasonal variation of earthquake
[21] C. Forster, L. Smith, The influence of groundwaterflow occurrence in Japan, Earth Planet. Sci. Lett. 207 (2003)
on thermal regimes in mountainous terrain: A model 159-164.
study, J. Geophys.Res. 94 (1989) 9439-9451. [41] L. Audin,J.-P. Avouac,M. Flouzat, J.-L. Plantet, Fluid-
[22] M. Manga,Hydrologyof spring-dominated streams in the driven seismicity in a stable tectonic context: TheRemir-
Oregon Cascades, Water Resour. Res. 32 (1996) 2435- emontfault zone, Vosges,France, Geophys.Res. Lett. 29
2439. (2002)art. no. 1091.
[23] G.E.P Box, G.M. Jenkins, G.C. Reinsel, Time Series [42] D.A. Lockner, N.M. Beeler, Premonitory slip and tidal
Analysis: Forecasting and Control, Prentice-Hall, Engle- triggering of earthquakes, J. Geophys. Res. 104 (1999)
woodCliffs, NJ, 1994, 598 pp. 20133-20151.
[24] T. Lay, T.C. Wallace, ModernGlobal Seismology, Aca- [43] G.C.P. King, R.S. Stein, J. Lin, Static stress changesand
demic Press, London, 1995, 521 pp. the triggering of earthquakes, Bull. Seismol. Soc. Am.84
[25] H.F. Wang,Theory of Linear Poroelasticity, Princeton (1994) 935 953.
University Press, Princeton, NJ, 2000, 287 pp. [44] R.S. Stein, G.C.P. King, J. Lin, Changein failure stress
[26] D.L. Turcotte, G. Schubert, Geodynamics, Cambdridge on the southern San Andreas Fault system caused by the
University Press, Cambridge,2002, 456 pp. 1992 magnitude=7.4 Landers earthquake, Science 258
[27] S. Hurwitz, K.L. Kipp, S.E. Ingebritsen, Groundwater (1992) 1328 1332.
flow, heat transport, and water-table position within vol- [45] R.S. Stein, G.C.P. King, J. Lin, Stress triggering of the
canic edifices: Implications for volcanic processes in the 1994 m=6.7 Northridge, California, earthquake by its
CascadeRange, J. Geophys.Res., 2003 (in press). predecessors, Science 265 (1994) 1432-1435.
[28] M. Nathenson, Springs on and in the vicinity of Mount [46] R.S. Stein, The role of stress transfer in earthquake oc-
Hood, Oregon, USGSOpen-File Report, 2003 (in re- currence, Nature 402 (1999) 605-609.
view). [47] R.A. Harris, Introduction to special section: Stress trig-
[29] E. Roeloffs, Poroelastic techniques in the study of earth- gers, stress shadows,and implications for seismic hazard,
quake-related hydrologic phenomena,Adv. Geophys. 37 J. Geophys.Res. 103 (1998) 24347 24358.
(1996) 135-195. [48] K. Shimazaki, T. Nakata, Time-predictable recurrence
[30] S.S. Gao, P.G. Silver, A.T. Linde, I.S. Sacks, Annual model for large earthquakes, Geophys. Res. Lett. 7
modulation of triggered seismicity following the 1992 (1980) 279-282.
Landers earthquake in California, Nature 406 (2000) [49] C.S. Weaver,R.D. Norris, C. Trisler-Jonientz, Results of
500-504. seismological monitoring in the Cascades Range 1962-
[31] H.S. Carslaw, J.C. Jaeger, Conductionof Heat in Solids, 1989: Earthquakes,eruptions, avalanches, and other curi-
Clarendon Press, Oxford, 1959, 510 pp. osities, Geosci. Can. 17 (1990) 158-162.
[32] Spatial Climate Analysis Service, Averageannual precip- [50] J. Jones, S.D. Malone,MountHoodearthquake activity:
itation of Oregon,OregonState University, online, 2002. Volcanic or tectonic origins (Abstract)?, EOSTrans.
[33] S.E. Ingebritsen, D.R. Sherrod, R.H. Mariner, Rates and AGU83 (Fall Meet. Suppl. 12A) (2002) 1179.
patterns of groundwater flow in the Cascade RangeVol- [51] P.A. Domenico, F.W. Schwartz, Physical and Chemical
canic Arc, and the effect of subsurface temperatures, Hydrogeology, Wiley, NewYork, 1998, 506 pp.
J. Geophys.Res. 97 (1992) 45994627. [52] C.E. Manning,S.E. Ingebritsen, Permeability of the con-
[34] M. Manga, A model for discharge in spring-dominated tinental crust: Implications of geothermaldata and meta-
streams and implications for the transmissivity and re- morphicsystems, Rev. Geophys. 37 (1999) 127-150.