Beruflich Dokumente
Kultur Dokumente
A
B
Fig. 3. Water level response relative to semidiurnal tidal dilatation
strain. (A) Phase lag; (B) amplitude response. Values were calculated
using a Bayesian Monte Carlo Markov chain inversion method in the time
domain (13). The inversion was applied by 29.6-day segments overlap-
ping by 80%, respectively. The error bars represent the 95% confidence
interval.
www.sciencemag.org SCIENCE VOL 340 28 JUNE 2013 1557
REPORTS
imposes a dilatational strain on the surrounding
rock formation that pumps water cyclically in and
out of the well through the rock around the un-
cased portion of the borehole below800 mdepth.
The clear oscillations indicate that the aquifer is
well-confined. The transmissivity and storage co-
efficient determine the phase and amplitude re-
sponse of the water level to the tidal loading (11).
To first order, phase lag is inversely related to
transmissivity, and amplitude response is propor-
tional to storage coefficient. Using tidal response
to measure hydrogeologic properties has two
distinct advantages: (i) Tidal response is passive
and records the in situ properties undisturbed by
repeated pump tests or water injections, and (ii)
tidal response provides a continuous record of
temporal changes of hydrogeologic properties in
the rocks below the main rupture zone.
We translated the phase and amplitude re-
sponses into transmissivity and storage coefficient
values on the basis of the analytical solution for
a two-dimensional isotropic, homogeneous, and
laterally extensive aquifer (12, 13). The observed
phase lag ranged from 20 to 30, where
negative values indicate that the water level
oscillations lag behind the imposed dilatational
strain; the amplitude response ranged from
5.5 10
7
to 6.3 10
7
m
1
(Fig. 3). The cor-
responding transmissivity T varied systematically
over a range of 3.6 10
6
to 6.8 10
6
m
2
s
1
,
with an average value of 5.1 10
6
m
2
s
1
(Fig. 4).
In contrast, the storage coefficient S did not
evolve systematically, having an average value of
2.2 10
4
with small fluctuations about this mean
(standard deviation = 5.7 10
6
). The different
behavior for storage and transmissivity is the result
of the weak sensitivity of the solution to variations
in storage coefficient (12) and little real variation
of the storage coefficient. Accordingly, we fixed S
to the average value to more robustly solve for T
and found that the resulting values of transmissiv-
ity were unchanged from the original inversion
(Fig. 4). Because of the geometrical idealizations
of the model, the absolute values are lower bounds
(13), although relative variations over time are
more robust.
For the average values of T and S over the
observation period, the average hydraulic diffu-
sivity, D = T/S, is 2.4 10
2
m
2
s
1
. On the basis
of this observed hydraulic diffusivity, the tidal
observations sense a zone extending ~40 m from
the well and thus are sensitive to mesoscale
fractures (13). The most transmissive units in the
damage zone control the tidally driven flow. The
estimate of D is two orders of magnitude larger
than that of the most directly comparable post-
earthquake fault zone study [7 10
5
m
2
s
1
on
the Chelungpu fault slip zone after the 1999 Chi-
Chi earthquake from a cross-hole experiment
(14)]. Our diffusivity value implies that as soon
as the currently observed level of damage
developed during the earthquake, coseismic
drainage was important. However, the diffusivity
may be small enough that advective flowthrough
the fault zone does not have a major impact on
the postseismic temperature measurements that
201001 201004 201007 201010 201101 201104 201107
1
1.2
1.4
1.6
1.8
P
e
r
m
e
a
b
i
l
i
t
y
(
1
0
-
1
5
m
2
)
Date
201001 201004 201007 201010 201101 201104 201107
2.1
2.15
2.2
2.25
2.3
2.35
2.4
Date
S
t
o
r
a
g
e
c
o
e
f
f
i
c
i
e
n
t
(
1
0
-
4
)
3.9
4.7
5.5
6.3
7.1
T
r
a
n
s
m
i
s
s
i
v
i
t
y
(
1
0
-
6
m
-
2
s
-
1
)
A
B
Fig. 4. Hydrogeologic properties of the well-aquifer systemover time.
(A) Permeability and transmissivity; (B) storage coefficient. Values were in-
verted from the phase and amplitude of each 29.6-day segment based on
the analytical model (9). Segments that overlap the remote earthquakes
(vertical dashed lines) were not inverted [see (13) for inversion results
including these times]. The black dots denote an unconstrained inversion;
the red dots are the results of inversion with the storage coefficient fixed to
a single value. Because the two separate inversions have identical results
for transmissivity, the red dots cover the black dots in (A). The vertical
dashed lines show the time of the selected teleseismic events, which
correspond to sudden increases in permeability. The best-fit linear trends
between each set of permeability increases are shown as light gray dashed
lines. Permeability errors are estimated by propagating the range of phase
errors.
28 JUNE 2013 VOL 340 SCIENCE www.sciencemag.org 1558
REPORTS
are another objective of post-earthquake studies
(15). Previous modeling work has shown that
hydraulic diffusivities comparable to the ob-
served effective D could suppress the temper-
ature anomaly by at most a factor of 2 relative to
a conductively cooled model (13, 16), and there-
fore thermal anomalies in the fault zone could be
observable if the fault friction is comparable to
laboratory values.
The effective permeability k is related to trans-
missivity T by
k
m
rgd
T 1
where m is the fluid dynamic viscosity, d is the
thickness of the open interval of the well, and r is
the density of fluid. Using m = 10
3
Pas at 20C,
r = 10
3
kg m
3
, g = 9.8 ms
2
, and d = 400 m, the
average value of the effective permeability is 1.4
10
15
m
2
(Fig. 4). Permeability errors are esti-
mated by propagating the range of phase errors.
This approach is appropriate for measuring the
precision of the inversion, and these errors are
useful for assessing time variability. The absolute
value of the permeability is more strongly affected
by the limitations of the flow model (13). The ob-
servation constrains the effective permeability aver-
aged over the entire open interval and is therefore
a lower bound for the effective permeability of
the highly fractured regions.
Our observed fault zone permeability is much
larger than core-scale laboratory measurements
of permeability from active-fault core samples
(17), which range from 10
19
to 10
18
m
2
; it is
also larger than the previously measured average
permeability of 1.9 10
16
m
2
for the intact
upper Triassic rock near the Wenchuan drilling
site (18). The difference is likely due to mesoscale
fractures and highlights the importance of damage
in determining the field-scale behavior (6).
There are also substantial temporal changes in
transmissivity, which we interpret as permeability
changes because the formation thickness and fluid
properties are unlikely to vary during the obser-
vation period. During most of the study period, the
permeability trends downward and is most easily
interpreted as a reduction in fracture aperture and
connectivity during the continuous evolution
since the original earthquake. Seismic studies
in Wenchuan suggest that damage healed over
a protracted time after the earthquake (19), and
permeability does not generally evolve in time
in the absence of a disturbance (20). The only
candidate perturbation besides the earthquake
is the drilling itself, which couldpotentially produce
transient damage. However, drilling-induced fac-
tures are expected to extend at most a few bore-
hole radii away fromthe hole (21) (i.e., <0.3 m) and
cannot account for the phase change of the long-
period tidal response that senses average proper-
ties up to ~40 m from the borehole.
This decreasing permeability may reflect
the healing process of the fault zone after the
Wenchuan earthquake due to a combination of
fracture closure, sealing, precipitation, biogenic
growth, and pressure solution (7). The heal-
ing rates range from 4.1 10
16
m
2
year
1
to 2.1
10
17
m
2
year
1
, using linear fits to each interval
between perturbation events (Fig. 4) (13). Pre-
vious work (22, 23) modeled fault zone healing as
an exponential recovery process with decay times
on the order of decades or longer. However, our
data are best fit with much shorter exponential
decay times of 0.6 to 2.5 years, indicating a much
more rapid process than anticipated (table S2). The
short exponential decay times might indicate a fast
healing process, such as removal of props trapped
in fractures, or crack sealing with a strongly dis-
equilibrated fluid to allow mass transfer with the
observed characteristic times.
Fault zone healing has been documented in
seismic velocity changes (24, 25) and has been
suspected on the basis of discrete repeated ac-
tive formation tests (26). After the 1995 Kobe
(Hyogoken-Nanbu) earthquake, water injection
experiments in 1997 and 2000 tracked fluid flow
in the hanging wall 50 m from the Nojima fault
core and found that the permeability in 2000 had
decreased to 50% of the value in 1997 (26). Seis-
mic studies document seismic velocity decreases
around the fault after an earthquake continuing
for years, which can also be interpreted as a
consequence of fracture closure (19, 24, 25). In
Wenchuan, the repeated seismic velocity measure-
ments made in the first year are consistent with
such healing (19).
The sudden increases in permeability result in
an overall rate of decrease that is more gradual
than the short-term trends by a factor of 1.5 to
7.5. Previous work suggests that permeability
might be enhanced by remote or regional earth-
quakes (20, 27). Plausible mechanisms include
fracture unclogging due to the rapid, oscillatory
flow driven by the seismic waves as they pass
through the fault zone (27, 28). The times of the
four permeability increases in Fig. 4 are cor-
related with the four teleseismic earthquakes
that produced the largest integrated seismic
shaking at the drilling site during the observa-
tion period (table S2). However, like many
hydrogeologic observations, the magnitudes of
the perturbations are not simply proportional to
that of the peak amplitude of the seismic wave
(27). Most important, our observations imply
that any physical modeling of precipitation,
fracture closure, or any other healing process of
a fault zone needs to match the much more rapid
healing rate that is only visible in the continu-
ously recorded data.
An interplay between permeability evolution
and fault strength has previously been suggested
on geological and theoretical grounds (29). The
Wenchuan earthquake Fault Scientific Drilling
Project captured the permeability evolution in the
critical post-earthquake period, when damage
heals and the stage is set for the next earthquake.
The unexpectedly high average hydraulic diffusiv-
ity (2.4 10
2
m
2
s
1
) measured here also implies
substantial fluid circulation in the evolving fault
zone. If this value represents the hydrogeologic
properties during the earthquake, fluid flowshould
take place during the earthquake rupture.
References and Notes
1. M. K. Hubbert, W. W. Rubey, Geol. Soc. Am. Bull. 70, 115
(1959).
2. D. Andrews, J. Geophys. Res. 107, 2363 (2002).
3. J. R. Rice, J. Geophys. Res. 111, B05311 (2006).
4. A. W. Rempel, J. R. Rice, J. Geophys. Res. 111, B09314
(2006).
5. C. Marone, D. M. Saffer, in The Seismogenic Zone of
Subduction Thrust Faults, T. H. Dixon, J. C. Moore, Eds.
(Columbia Univ. Press, New York, 2007), pp. 346369.
6. J. S. Caine, J. P. Evans, C. B. Forster, Geology 24, 1025 (1996).
7. J. P. Gratier, Oil Gas Sci. Technol. Rev. IFP Energies
Nouvelles 66, 491 (2011).
8. H. Li et al., Tectonophysics 584, 23 (2013).
9. Z. Xu et al., Episodes 31, 291 (2008).
10. H. M. Savage, E. E. Brodsky, J. Geophys. Res. 116,
B03405 (2011).
11. Transmissivity is a measure of the rate of volumetric flow
through a unit width of aquifer under a unit hydraulic
gradient and is directly proportional to permeability. Storage
coefficient is the volume of water released from storage per
unit surface area of aquifer per unit imposed head (30).
12. P. A. Hsieh, J. D. Bredehoeft, J. M. Farr, Water Resour.
Res. 23, 1824 (1987).
13. See supplementary materials on Science Online.
14. M. Doan, E. Brodsky, Y. Kano, K. Ma, Geophys. Res. Lett.
33, L16317 (2006).
15. E. E. Brodsky, J. Mori, P. M. Fulton, Eos 91, 237 (2010).
16. P. M. Fulton, R. N. Harris, D. M. Saffer, E. E. Brodsky,
J. Geophys. Res. 115, B09402 (2010).
17. D. Lockner, H. Naka, H. Tanaka, H. Ito, R. Ikeda, in
Proceedings of the International Workshop on the
Nojima Fault Core and Borehole Data Analysis Tsukuba,
Japan, H. Ito, K. Fujimoto, H. Tanaka, D. A. Lockner, Eds.
(U.S. Geol. Surv. Open File Rep., 1999), pp. 2223.
18. Z. Rukai et al., Petrol. Explor. Dev. 36, 46 (2009).
19. Y. Li, Imaging, Modeling and Assimilation in Seismology
(China High Education Press, Beijing, 2012), vol. 1,
pp. 151198.
20. J. E. Elkhoury, E. E. Brodsky, D. C. Agnew, Nature 441,
1135 (2006).
21. M. Brudy, M. Zoback, Int. J. Rock Mech. Min. Sci. 36, 191
(1999).
22. F. Renard, J. P. Gratier, B. Jamtveit, J. Struct. Geol. 22,
1395 (2000).
23. J. P. Gratier, P. Favreau, F. Renard, J. Geophys. Res. 108,
2104 (2003).
24. F. Brenguier et al., Science 321, 1478 (2008).
25. Y. G. Li, J. E. Vidale, K. Aki, F. Xu, T. Burdette, Science
279, 217 (1998).
26. Y. Kitagawa, K. Fujimori, N. Koizumi, Geophys. Res. Lett.
29, 1483 (2002).
27. M. Manga et al., Rev. Geophys. 50, RG2004 (2012).
28. J. E. Elkhoury, A. Niemeijer, E. E. Brodsky, C. Marone,
J. Geophys. Res. 116, B02311 (2011).
29. R. Sibson, Tectonophysics 211, 283 (1992).
30. R. A. Freeze, J. A. Cherry, Groundwater (Prentice-Hall,
Upper Saddle River, NJ, 1977).
Acknowledgments: Supported by the National Science and
Technology Planning Project in China (H.-B.L.) and NSF grant
EAR1220642 (E.E.B.). Seismic data from the Chinese national
network are archived and distributed by Incorporated Research
Institutions for Seismology (IRIS) Data Management System.
Supplementary Materials
www.sciencemag.org/cgi/content/full/340/6140/1555/DC1
Materials and Methods
Supplementary Text
Figs. S1 and S2
Tables S1 and S2
References (3134)
1 March 2013; accepted 17 May 2013
10.1126/science.1237237
www.sciencemag.org SCIENCE VOL 340 28 JUNE 2013 1559
REPORTS