Geothermics. Vol. 23, No. 2, pp.

111-126, 1994
Pergamon CNR
Elsevier ScienceLtd
Printed in Great Britain
0375-6505/94 $7.00 + 0.00

0375-6505(94)E0003-T

SEISMIC VELOCITY AND AT-I'ENUATION STRUCTURE

OF THE GEYSERS G E O T H E R M A L FIELD, CALIFORNIA

J. J. ZUCCA, L. J. HUTCHINGS and P. W. KASAMEYER

Lawrence Livermore National Laboratory, P.O. Box 808, Livermore, CA 94551, U.S.A.

(Received June 1993; accepted for publication December 1993)

Abstract--The Geysers geothermal field is located in northern California and is one of the world's largest
producers of electricity from geothermal energy. A key resource management issue at this field is the
distribution of fluid in the matrix of the reservoir rock. In this paper, we interpret seismic compressional-
wave velocity and quality quotient (Q) data at The Geysers in terms of the geologic structure and fluid
saturation in the reservoir. Our data consist of waveforms from approximately 300 earthquakes. Using
compressional-wave arrival times, we invert for earthquake location, origin time, and velocity within a
three-dimensional grid. Using the pulse width of the compressional wave, we invert for the pulse width
contribution associated with the source and the one-dimensional Q structure. We find that the velocity
structure correlates with known mapped geologic units. The dry steam reservoir, which is known from
steam well drilling, is mostly correlated with low velocity. The correlation is best for those areas where the
steam pressure has been reduced by production. The Q increases with depth to the top of the dry steam
reservoir and decreases with depth within the reservoir. This decrease of Q with depth suggests that the
liquid saturation of the reservoir rock matrix increases with depth.

Key words: The Geysers geothermal field, seismic wave velocity, reservoir.

INTRODUCTION
The Geysers geothermal field accounts for approximately 8% of Pacific Gas and Electric
Company's power production in Northern California. It has become an area of focused study
since the steam pressure was found to be declining at an increased rate in late 1984 (Barker et al.,
1992). A better understanding of the physical processes at work within the field could provide a
basis for strategies to moderate that trend. Specifically, methods to remotely determine the
boundaries of the reservoir, the distribution of steam and two-phase fluid within the field, and
the changes in that distribution as a result of production and injection are needed. Toward this
end, we have begun to investigate means of computing images of the seismic velocity and quality
quotient (Q; inversely proportional to seismic attenuation) structure of the region and jointly
interpreting these images to determine the in situ phase state of water in the reservoir.
In this study, we use the velocity and Q data to interpret the structure of the target geothermal
region, in this case the central portion of The Geysers, in terms of the existence of pore water
and its phase state. We base our interpretation primarily on the laboratory data of Ito et al.
(1979), who carried out velocity and attenuation measurements on Berea sandstone samples at
elevated temperatures and varying degrees of liquid saturation to approximate reservoir
conditions. However, these results were for frequencies near 10,000 Hz, raising the question of
their applicability to field measurements at lower frequencies. However, results from Evans and
Zucca (1988) and Zucca and Evans (1992) suggest that P-wave attenuation and seismic velocity
structure contain complementary information at Medicine Lake (California) and Newberry
(Oregon) volcanoes and may be used to predict the location of geothermal drilling targets. They
111
112 .1, .I. Z u c c a ct al.

found that regions with low Q and normal-to-high P-wavc vclocity coincide with areas
independently identified as good geothermal prospects by other researchers using other
geophysical techniques.

P R E V I O U S G E O L O G I C A L AND G E O P H Y S I C A L W O R K A T T H E G E Y S E R S
The stratigraphy and geology of The Geysers have been summarized by McNitt et al. (1989),
who attempted to correlate the earlier mapping efforts by Bailey (1946), McNitt (1968), and
McLaughlin (1981). For our target volume, we summarize the results of McLaughlin (1981) in
Fig. 1.
McNitt et al. (1989) also compiled the results of several hundred lithologic logs for The
Geysers region to develop a model of the stratigraphy. They concluded that the steam reservoir
is contained in a highly indurated and fractured graywacke that is capped with a more ductile,
unfractured argillaceous graywacke. The graywacke sequence has been intruded at its base by
silicic magmas to form a felsite batholith whose axis trends northwest and is roughly coincident
with the steam field (Thompson and Gunderson, 1992). The elevations of the top of the steam
reservoir before production and of the top of the felsite intrusive body have recently been made
public by a consortium of Geysers operators (Field Operators, 1992). These contour maps have
been reproduced for our target volume in Fig. 1.
Extensive geophysical studies have been carried out at The Geysers. The most recent results
that are relevant to our study area include a three-dimensional block inversion for P-wave
structure for The Geysers and its surrounding area, which was performed by Eberhart-Phillips
(1986) using the Thurber (1983) method. She used the USGS network to record P-wave arrivals
from local earthquakes and refraction shots, achieving a resolution of 2-3 km. O'Connell and
Johnson (1991 ) performed a progressive inversion for hypocenters and one-dimensional P-wave
and S-wave velocity structure of The Geysers in much of the same area as the current study.
Their resolution was near 1 kin. In this study, we increase this resolution by solving for the three-
dimensional structure and decreasing the size of the model elements to approximately 0.6 km.
Major and McEvilly (1979) have conducted seismicity and seismic refraction studies at The
Geysers. They noted that the upper part of the reservoir is characterized by high velocities and
low attenuation, but did not attempt a three-dimensional interpretation of their data.

DATA
U N O C A L Corporation has monitored the seismicity of portions of The Geysers geothermal
field since 1985 (Stark, 1992). Their network consists of mostly vertical 4.5 Hz electromagnetic
seismometers. The data are telemetered to a central site where they are digitized at 100 samples
per second and waveforms by event are archived. The network records 15,000 to 20,000 events
per year that have a mean magnitude of approximately Mr, = 0.7. Elevations for the events are
fairly evenly distributed to about 2.5 km below sea level (b.s.l.). Events used in this study
occurred from May 1988 through December 1989.
U N O C A L provided us with waveforms and first arrival picks (Debbie Wheeler, pers. comm.,
1990). We used only arrivals with at least 10:1 signal-to-noise ratio of the first pulse observed at 8
or more stations. Our final data set consisted of 300 events for a total of just over 3100 P-wave
arrival times. Twenty-eight stations reported arrivals and, out of these, 17 stations reported
more than 50 arrivals. Figure 2 shows the distribution of stations and events. Table 1 lists all
stations available for this study. Because of the limited number of horizontal recordings
available, only compressional waves are used in this study.
We used the time between the first arrival and the first zero crossing to measure the pulse
 Seismic Velocity and Attenuation Structure of The Geysers Geothermal Field 113

a)
-% K/v'/./'/A

Zone
] Ultramalic Rocks

creek "[hrus _ Zone

I~l Volcanic Rocks

] Franciscan Rocks

Y Generalized Geology

I I
5 km

b) • Powerplant
0.9 Elevation (kin b.s.L)

Top of dry steam reservoir Inset

Geyse~--~ California
iII
c) • I \

// \ N
I \\

Top of felsite

d)

5 km
Inversion grid

Fig. l, Geologic setting of inversion target volume. (a) Generalized surface geology (after McLaughlin, 1981). (b)
Elevation of the top of dry steam reservoir in km below sea level (after Field Operators, 1992). (c) Top of felsite intrusive
body (km b.s.I.) (after Field Operators, 1992). (d) Map view of inversion grid. Spacing is - 0 . 6 km (2000 ft) in all three
directions. The numbers indicate the location of power production units. Contour depths are given in kilometers. Note
that north is to the upper left. Inset at bottom right shows location of The Geysers in the state of California and the target
volume with respect to the field boundary.

widths. We visually examined each pulse for evidence of muitipathing and included only those
pulses that were smooth from the initial arrival to the first zero crossing. The first arrival pick was
also examined to verify the accuracy of the pick. The estimated error in the arrival time reading
was _ 0.05 s. Because of the shape of the signals, the m e a s u r e m e n t errors in the first zero
crossings were small c o m p a r e d to the errors in the first arrival picks and did not contribute to the
uncertainty of the pulse width.
1 14 ,I. ,I. Z u c c a e t al.

Boundary

i 5km~~11
Fig. 2. Location of seismograph stations (triangles) and earthquakes (stars) used in this study.

Table 1. Station locations used in this study

Latitude Longitude
Station Elevation Sta. Cor. * Sta. Cor.-~-
ID (deg.) (min.) (deg.) (min.) (m a.s.l.) (s) (s)

01 38 50.2(1 122 45.51 800 (/.028 -0.002

02 38 48.17 122 45.04 1326 (1.035 -0.004
(14 38 48.49 122 49.37 570 0.000 (t.000
05 38 48.57 122 49.72 551 0. 000 0. 000
06 38 48.57 122 49,72 521 0.012 (l.006
09 38 50.76 122 48.46 862 0.035 0.007
10 38 48.49 122 48,21 711 -0.037 0.000
12 38 47.31 122 48,15 747 -0.036 0.002
14 38 49.39 122 50.05 884 0.032 0.004
15 38 49.41 122 48.53 683 0.012 0.000
19 38 49.37 122 46.30 1006 0,020 -0.001
20 38 50.32 122 47.35 991 -0.012 0.001
21 38 47.13 122 46.25 848 -0.027 -0.002
22 38 46.22 122 45.86 855 -(/.078 -0.002
27 38 47.66 122 46.90 777 0. 000 0. 000
28 38 47.59 122 47.20 646 -(/.013 0.1102
34 38 48.65 122 47.43 1029 0.000 0.000
35 38 48.71 122 46.92 1052 -0.034 0.000
38 38 45.75 122 44.34 859 -0.061 -0.008
39 38 47.03 122 44.10 966 -0.045 -0.0(/6
40 38 49.41 122 48.53 646 -0.014 0.006
41 38 44.93 122 44.47 994 -0.069 -(t.0(/5
42 38 44.41 122 42.60 1073 - 0. 023 - 0. 006
43 38 46.58 122 42.92 715 -0.033 0.001
44 38 45.77 122 43.36 991 -0,056 0.003
45 38 45.97 122 41.87 552 -0.017 -0.002
46 38 45.76 122 44.18 832 -0,084 -0.004
47 38 48.57 122 49.72 401 0.031 0.009

*From the velocity inversion,

+From the Q inversion.
 Seismic Velocity and Attenuation Structure of The Geysers Geothermal Field 115
The picking errors are small compared to travel times, and their contribution to errors in
event locations and velocity structure determinations is small. However, as will be discussed
later, the picking error is a substantial percentage of the pulse widths.

PULSE WIDTH AS A MEANS OF DETERMINING Q

In the idealized case of linear propagation of a seismic signal through a medium with intrinsic
attenuation, the frequency content of the signal is modified by the attenuation operator:
-oR

where A(~o) is the Fourier amplitude, R is ray travel distance, V is seismic velocity, and Q is the
quality quotient (assumed to be independent of frequency). An empirical mathematical model
for pulse broadening in an inhomogeneous medium has been suggested by Gladwin and Stacey
(1974) and Stacey et al. (1975) and supported theoretically by Kjartansson (1979). Stacey et al.
(1975) showed that a pulse propagating in an elastic medium has a pulse width that follows the
relationship:

r = ro + C I ds
J
Vo(r)Q(r ) (2)
ray

where C is a constant (Stacey et al. used 0.5), ds is distance along ray path, and r 0 is the source
contribution to the original pulse width (Stacey et al. used zero for this term), and Vo(r)Q(r) are
the velocity and Q structures, respectively, that vary as position r; r is the time from the onset of
the initial arrival to the initial peak on displacement records, or the first zero crossing on velocity
records.
Our data suggest that pulse broadening is a reasonable model for estimating Q at The
Geysers, because we observe that on average pulse widths broaden at increasing distance from
the source. Figure 3 shows pulse shapes from a single earthquake recorded at several Geysers
stations. The pulse widths vary by up to a factor of 3 between stations, with more distant ones
having broader pulse widths. The variation of pulse width with hypocentral distance is shown
more completely in Fig. 4. The data are clearly quite noisy; however, the slope of the least-
squares line in the figure is significantly different from zero at the 3 c~level (slope = 0.0013 s/km,
O~lopc = 0.00016 s/kin), indicating that pulse width increases with distance. A similar result is
obtained for the subset of data in the distance range of 0 to 7 km indicating that the non-zero
slope is not caused by a few points beyond 7 km.
Inverting pulse widths for Q structure involves several issues that have not been fully resolved
in the literature. Issues that need to be addressed are: (1) the relative contribution of scattering
and intrinsic Q to pulse broadening, (2) the contribution of the source to the observed pulse
widths, and (3) the fmax effect on pulse width. These issues are discussed in the rest of this
section.
Intrinsic Q and scattering affect the seismic signals by different mechanisms (Richards and
Menke, 1983). Intrinsic Q removes high-frequency energy from the entire wavetrain, with the
arrivals that travel along the longest ray paths having the greatest loss of high frequencies, while
scattering delays the arrival of the high-frequency energy, which tends to broaden only the initial
pulse. Richards and Menke (1983) proposed a test to see if the signal is dominated by intrinsic Q
or scattering. If scattering is dominating pulse width, then the spectral corner frequency of the
early parts of the wavetrain would be at a lower frequency than in the later parts. If intrinsic Q is
dominant, no increase in corner frequency would be observed later in the wavetrain. We
116 ./. ,l. Zucca ct al.

sl9

S35

SI2

0 0.1 0.2
S28 Time, s

sOl

]
s20

I I
0 0.1 0.2 I I
0 5 km
Time, s

Fig. 3. Pulse width variations for a single event recorded at several stations. The map in the lower right shows the
location of the stations (triangles) and the event (star) with respect to the target volume. Note that north is up.

examined several events for this effect by computing corner frequencies for two adjacent 0.5 s
time windows following the first arrival. Care was taken not to include the S-wave arrival. We
found no systematic increase in corner frequency in the later time window and conclude that
scattering is not the mechanism of the observed pulse broadening.
The finite duration of the seismic source can contribute to the observed pulse width. Figure 4
shows the pulse width data plotted as a function of distance from the hypocenter with a least-
squares fit line. The non-zero y-intercept of the line indicates that a c o m p o n e n t of the pulse
width is generated at zero distance (i.e. at the source). The pulse widths we measure on velocity
records correspond to the rise time of the displacement pulse and roughly half the source
duration. There may also be some contribution to pulse width due to directivity, but we
deliberately chose earthquakes that are small enough (ML -- 1.1) to minimize the source effect.
This choice can be justified by considering a Brune (1972) source model for a magnitude 1.1
earthquake, which would predict a 20-m fault radius. Therefore, with a rupture velocity of near
2.5 kin/s, source duration could range from 5 to 15 ms. However, we observe much longer source
 Seismic Velocity and Attenuation Structure of The Geysers Geothermal Field 117
0.09

0.08

0.07 • • "" I • •l •! °

° . •

0.06
• •• . . ..'.". >-.....-:. ~•
•.,••. . ...~ . .•- .~, .. .

i • • ". • ~ • r •~ ,I .,••. .I,! ***°a • " * • .,• • "*

0.05 • . • .......- -. -....Li.c,....:.~-..,_~-..-

~. . . . ° • . . , • • . ° . . . ° . °
..: : .,. :'. . . ~
• -... , . . . ~ " .~_-~. ~ * . , ~ . ~ ' ~ • . . .
.'.* ~@" ~-$ . . . . . .a~...* ° * *-" ."
0.04
[3_ •.. ~ *.-¢. : ~ - ~ ' ~ ¢ , . ~- :.. • .
0.03 . . . .
• . .... ":;
• 'b*l*
. ' . .1,1d1 ' ¢i,l y:I r. , , * :, . ' ,., ' " • •
" .
•
•
. •
." "
• • • .*• • ••• h .• • •**•
0.02

0.01

I ! I I I I

2 4 6 8 10 12
Distance from hypocenter, km

Fig. 4. P u l s e w i d t h s p l o t t e d as a f u n c t i o n o f d i s t a n c e f r o m h y p o c e n t e r t o r e c e i v e r . T h e s t r a i g h t line is a l e a s t - s q u a r e s fit

to t h e d a t a . See t h e t e x t f o r a d i s c u s s i o n o f t h e a c c u r a c y o f t h e fit.

durations than would be predicted. This discrepancy could imply larger source dimensions could
contribute to greater variations in pulse widths due to directivity. This may be a significant
contribution to the scatter shown in Fig. 4. Since focal mechanism affects amplitude and not
spectral content, we expect no pulse width variation from this effect.
When Q is very low along a ray path, the shape of the spectrum of an earthquake at a
particular receiver can be altered so that the source corner frequency is masked by an apparent
lower corner frequency due to Q. This effect is particularly important in alluvial basins where
the last few kilometers of propagation can be through material with very low Q. This
phenomenon is referred to as the fmax effect (Hanks, 1982; Frankel, 1982; Hutchings, 1990).
Majer and McEvilly (1979) apparently observed this effect at The Geysers. They found that
earthquakes (their data set included events up to M L = 2.0) at The Geysers had corner
frequencies that were independent of seismic moment. If near-receiver structure is the
dominant factor in our pulse width data, station corrections should dominate the Q inversion
(see discussion below), and a plot of pulse width vs. station would show a strong correlation with
station (see Fig. 5). Instead, we observe that pulse widths spanning the complete range of values
are observed at all stations with more than a few data points. Our results are in apparent conflict
with Majer and McEvilly (1979) and may be the result of differences in recording locations.

INVERSION APPROACH
We used a modified Thurber inversion method (Thurber, 1983) to compute a three-
dimensional model of velocity and a one-dimensional model of Q. The nonlinear inversion
technique solves for both earthquake location and velocity structure by minimizing the residuals
of travel times computed as:

t = to + ) (3)
ray

where to is origin time and V(r) is the velocity structure.

The major advantage of using the Thurber program for inversion of pulse widths is its direct
analog with the travel time inversion. Recognizing that the relationship between QV and r in
118 .I. ,1. Zuccaet al.

0.09

0.08 []
n O

ji
[3
0.07
u~ D
I] [] FI °

J~ 0.06 D
.-g_
0.05
B o O
0.04 Q
n
0.03

0.02 I1 g
D O

0.01

0 I I ¿ l l l l l l l t t l l l l l l

~ m m m m m m ~ m ~ m ~

Seismograph Station Name

Fig. 5. O b s c r v e d p u l s e w i d t h s f o r s e v c r a l s t a t i o n s .

Equation (2) is the same as the relationship between V and t in Equation (3), we inverted pulse
width data for the initial pulse width associated with each event and for Q structure by modifying
the Thurber routine to accept pulse widths as input data, using the inverted velocity structure as
V0(r ) in Equation (2).
The Thurber inversion code uses a node system to specify inversion parameters. Velocity or Q
values are specified at the node intersections and are interpolated to compute the parameter
between the node points. Choice of node spacing can affect the results. Too coarse a grid can
cause averaging over a larger volume than is required by the data, which can result in entirely
missing some parameter variations (Ellsworth, 1977; T o o m e y and Foulger, 1989; Evans and
Achauer, 1993). At a minimum, the node spacing should not be less than the Fresnel zone
(Nolet, 1987), which is the maximum spatial deviation between two ray paths that allows them to
constructively interfere (i.e. a 1/4 wave length). The maximum Fresnel spacing for the current
study is approximately 0.5 km. Furthermore, the node spacing must not be so small that there
are more parameters than observations.
A further consideration is the structure of the resolution matrix. Evans and Achauer (1993)
consider the effect of the progressive decrease of node spacing on the resolution matrix and the
inversion results. They point out that, in inversion problems in which the rays are sub-vertical,
the individual rows of the resolution matrix tend to have positive side lobes above and below the
parameter in question and negative side lobes horizontally around the parameter in question.
Too fine a parameterization can cause artifacts to develop in the spaces between parameters
when the node spacing is decreased below the minimum horizontal ray spacing. They suggest an
offset and average procedure to horizontally smooth the results. We adapted this procedure for
inversions by averaging the results from the base grid system and three variants that were
obtained from the base system by offsetting it along the x- and y-axes one-half grid spacing. Our
base grid system is shown in Fig. 1. It has a spacing of about 0.6 km in all three directions. The
datum is sea level.
Nonlinear inversions can converge to different solutions depending on the starting model.
T o o m e y and Foulger (1989) investigated the dependence of their results on the starting model
and concluded that although their results did indeed depend on the starting model, the
differences in the results were not significantly different given the accuracy of their arrival time
data. Fortunately, at The Geysers there is ample previous work to establish a realistic velocity
 Seismic Velocity and Attenuation Structure of The Geysers Geothermal Field 119
model, which should mitigate the effects of starting model dependence. To locate microearth-
quakes at The Geysers, UNOCAL uses a one-dimensional velocity model that is based on the
results of Eberhart-Phillips (1986) and Mitch Stark, pers. comm. (1990, Table 2). However,
more local detailed information is available. Majer et al. (1988) conducted vertical seismic
profiling to a depth of 1.6 km in the northern extreme of our study area near station 09 (Fig. 2).
O'Connell and Johnson (1991) used these results to derive a one-dimensional model in the
upper 1.2 km, having a P-velocity of - 3 . 0 km/s compared to the UNOCAL model, which is
- 4 . 0 km/s at this depth; the difference illustrates the variability of the near-surface structure.
For our study region, we find that using the UNOCAL starting model results in the lowest
overall travel time residuals. However, widely varying near-surface geology may make the
UNOCAL model inappropriate for the entire study region. To accommodate these variations,
station corrections for the structure directly under the receiver were included in the velocity and
Q inversions.
The inversion for Q also requires a starting model, which we arbitrarily chose as the average Q
for the target volume. This value was obtained by finding the slope of a straight line fit to plots of
pulse width as a function of travel time. The slope is related to Q through Equation (1). The
average value of Q was found to be 60, the same average Q as found by Majer and McEvilly
(1979) in their study.

INVERSION RESULTS
The velocity inversion resolution and results are shown in Figs 6 and 7, respectively. The
results are displayed as horizontal slices. The three-dimensional inversion for velocity resulted
in a 75% weighted variance reduction over the one-dimensional starting model. Three different
weighting criteria were used. Arrivals were weighted according to quality determined during
hand picking. Paths longer than 15 km were down-weighted because they have a substantial
portion of the ray path outside of the target volume. Travel time residuals greater than 0.1 s were
also down-weighted because they indicate a problem with the data (such as an incorrect pick).
The final weighted root-mean-square travel time residual was 0.012 s, which is slightly larger
than the picking error of 0.01 s. The damping parameter was chosen subjectively based on
tradeoffs between data variance, solution length, diagonal resolution value, and standard error
size. The damping was initially picked so that small changes in data variance would not result in
large changes in solution length. We then fine-tuned the selection of the damping parameter so
that a balance was achieved between the resolution matrix diagonals and the standard errors.
We only interpret the four layers shown because resolution drops off dramatically below an
elevation of 1.5 km b.s.1. The standard error in the layers shown in Fig. 7 averages approxi-
mately 0.08 km/s with a few nodes over 0.10 km/s. In general, the velocity increases with depth.
The central portion of the model tends to have the highest relative velocities down to at least a
depth of 0.9 km. At the deepest level shown, the lower (i.e. southwest) part of the image has the
highest overall velocities.
The Q inversion results are shown adjacent to the velocity results in Fig. 7. Unfortunately, we
were only able to achieve significant data variance reduction with a one-dimensional model.
This model achieved a variance reduction of 76%; however most of this is due to solving for to,
the source contribution to the pulse width. Only about 15% of the variance reduction is due to
the structure. The source pulse width (%) ranges from 0.018 to 0.048 s.
The station corrections for both the velocity and Q inversions are listed in Table 1. The
standard errors and resolution for the Q inversion are shown in Table 2. In the case of the Q-
results, the station corrections are less than the picking error and are therefore insignificant. The
small station corrections for Q-inversion further support the conclusion drawn from Fig. 5 that
120 ,1..I. Z u c c a et al.

Elevation: -0.3 km b.s.I.

\
North

.6

Elevation: 0.3 km b.s.I.

.4

: . :.~i:i i
.2

Elevation: 0.9 km b.s.I.

0
Resolution

Elevation: 1.5 km b.s.I.

Fig. 6. Distribution of the diagonals of the resolution matrix for the compressional-wavc velocity inversion. The data
are displayed as horizontal slices.
 Seismic Veloci~ and Attenuation Structure of The Geysers Geothermal Field 121

Cobb Mtn
_Q_

65 . . . . . . : . . . ~ . . . . . .... Elevation: -0.3 km b.s.I.

:i~ ........~:i~:~:::::~: ~:~ Overlay: Surface Geology

\ - .:.....

Layer elevation
Nodh (km, b.s.I.)
-0.3 10.3 ).9 1.5
I
5.1 5.6 5.6

Elevation: 0.3 km
132 b.s.I.
Overlay: Top of
steam reservoir

4.,= 5.0 5.2

Elevation: 0.9 km
138 b.s.I.
Overlay: Top of
steam reservoir

v.v 3A 4.4 4.8

Velocity,
km/s
Elevation: 1.5 km b.s.I.
56
Overlay: Top of felsite

Fig. 7. Velocity variations from the U N O C A L starting model displayed as horizontal slices. Slice elevation is given at
the corner of the images in kilometers below sea level. See Fig. 1 for a description of the overlays. Q values obtained by
inverting for one-dimensional structure for each layer are given on the left. The elevation contour shown in white is the
one closest to the layer elevation. The range of the data for each layer is given next to the gray-scale bar at the right.

pulse width data are not dominated by the fm,x effect. In the case of the velocity results, the
largest of the corrections correspond to stations that are outside of the target volume to the
southeast (i.e.stations 39, 38, 46, 44, 43). Since velocity outside of the target volume is
constrained, the only way for the inversion to compensate for a change in the velocity in this
region is through the station correction. The rest of the station corrections are small and
122 ./, ,/. Z'.t/cc. Cl <il.
"J ;.iblc -"~. ( ) I I C - d i l l / ~ . ! l l s i t l l l a l ~,Hlllin~. iIl~.~dtJ [OF [llC ]J \~,LI~,~_' '~L:JOClT\ IIIkCl~,hl!i
The last l w o co]{111111~~,]lOW lhc sT{illtltll-d errors and rcsohllil)ll \ alucs 1i()111 [hc
[ll'+Cl~lt)ll f o r a layer

Starting Node cxaluation Q inversion Q inversion

model (kin/s) (kin b.s.I ) sid. error rcsolution

4.1) -0.30 15 II.2t~

4.4 O.3O 24 IL3~
5.1 II.c)l 30 IL69
5.1 1.52 [3 0.70
5.4 2.13
5.4 2.74
5.6 3.35
5.6 3.%

correspond to minor ( < 1 0 % ) velocity variations near the station; we do not discuss them
further.

INTERPRETATION
Our uppermost layer at - 0 . 3 km b.s.I. (or 0.3 km a.s.I.) is completely above the reservoir and
should be influenced mostly by the near surface geology. The high-velocity feature in the center
of this layer correlates at its south end with the mapped surface expression of the ultramafic
stringer and continues to the northeast, which implies a dip to the north of the body, which
agrees with McLaughlin's (1981) map and cross section. The apparent pinchout of the high-
velocity body to the southeast may be the result of the loss of resolution in that corner of the
model (see Fig. 6).
The next layer is at 0.3 km b.s.1. It intersects the reservoir at three cupolas, one each in the
north, west, and south of the layer (Fig. 7, near units 17, 3-4, and 20 respectively). The northern
cupola correlates with a low-velocity zone, the western one is only partially correlated with low
velocity, and the southern one is not correlated with any change in velocity.
The next layer down at 0.9 km depth is contained more within the reservoir, and most of the
layer within the reservoir is again correlated with low velocity. The low-velocity zones are even
more strongly correlated with regions of the reservoir that have low-pressure superheat
conditions. Figure 8a shows the same image plane plotted with contours of observed pressure
from 1987 (Williamson, 1992) in addition to a contour enclosing those areas within the reservoir
where >20% of the steam comes from flashed injectate (Gambill, 1992). Both contours should
enclose the driest portions of the reservoir and consequently should have the lowest velocities,
The contours enclose the low velocity anomalies near unit 7-8 and near units 11 and 17 in
addition to the anomaly between units 14 and 20. Note in particular how the contours avoid the
high-velocity region between units 1-2, 12, and 14.
The deepest layer is at 1.5 km depth and is completely within the reservoir. At this level, we
have overlaid the depth to the top of the felsite intrusion. The highest velocities in this layer are
within the felsite. This is a somewhat surprising result since the felsite is known to have lower
density than the reservoir graywacke (Blakely and Stanley, 1993), so it is unlikely to have a
higher matrix velocity. However, the higher velocity could indicate that the felsite is less
fractured or more saturated than the graywacke. Stark (1992) suggests that injected water can
cause microearthquakes to occur. The concomitant increase in liquid saturation would tend to
raise the velocity of the rock. Therefore, microearthquake hypocenters should be spatially
correlated with high-velocity anomalies. We find some evidence for this correlation in our data.
Figure 8b shows Stark's (1992) epicentral locations for events deeper than 1.2 km b.s.i, from
 Seismic Velocity and Attenuation Structure of The Geysers Geothermal Field 123

Image plane
a)
elevation = 0.9
km b.s.I.

• -7~ • t

~ ",:-~,...

b) Image plane
elevation = 1.5
km b.s.I.

%~.. 4 • :

Fig. 8. Further correlation with reservoir data. (a) Velocity variations at 0.9 km below sea level plotted with reservoir
data from Williamson (1992) and Gambill (1992). T h e solid white contours are in MPa. The dashed white contour
encloses those areas within the reservoir where > 2 0 % of the steam comes from flashed injectate. (b) Velocity variations
at 0.9 km below sea level plotted with seismicity and injector data from Stark (1992). Dots are epicenters below 1.2 km
below sea level. Circles are centered on injection wells (circle area is proportional to volume injected).

November 1988 to August 1989, or roughly the same period as the data for our inversion. In
general the microearthquakes are correlated with high-velocity zones in the figure except for the
notable exception of the cluster of seismicity in the vicinity of unit 17.
There is little direct information on the state of pore fluids at The Geysers since any water
present in the formation flashes to steam upon entering the borehole or feeder fracture (Pruess
and Narasimhan, 1982). Estimates for the liquid saturation range from 30 to 50% to well above
50% (Pruess and Narasimhan, 1982). The layer at 0.9 km b.s.l. (Fig. 7) is mostly within the
reservoir and tends to have high Q relative to the next layer down, which is completely within the
reservoir. The high Q in the upper part of the reservoir is consistent with the earlier results of
Majer and McEvilly (1979), who also found relatively high Q in this region. The low Q in the
lower part of the reservoir suggests that the saturation is in the 30 to 70% range, whereas
saturation at the top of the reservoir could increase or decrease and still agree with the lab results
for Q alone obtained by Ito et al. (1979). A drop in saturation at the top of the reservoir below
about 30% seems the most likely since the velocity is lower in the reservoir compared to the
country rocks. This conclusion is further supported by the observation in Fig. 8 that low velocity
is correlated with low pressure (i.e. drier) parts of the reservoir, since low porosity rocks (i.e.
<10%) tend to have lower velocities when dry than when liquid saturated (Gregory, 1976).
Above the reservoir in layers 1 and 2, the Q increases with depth. The low Q at the surface
124 J. ,/, Z u c c a ct al.

Water
v

Graywacke
[-V-Q Saturation

r
Cap rock

I f
J

I
I
/ ,

Fig. 9. Schematicrepresentation of thc results of this study. On the left side is a cartoon of simplified reservoir geology
after McNitt et al. (1989). The right side shows schematically the variation in the two parameters that we calculate,
compressional-wave velocity and Q. Water saturation is shown with a dashed line to emphasize that it is an
interpretation.

may be indicative of the small overburden that tends to leave microcracks open. As depth
increases, the lithostatic pressure increases and tends to close cracks causing Q to increase. This
interpretation assumes that the lithology within the reservoir is primarily graywacke as
described by McNitt et al. (1989) so that we can attribute the variation in the Q structure to a
change in saturation conditions alone.
Figure 9 is a schematic diagram which qualitatively summarizes our results. On the left side of
the figure is a cartoon showing a simplified view of the reservoir geology based on the description
by McNitt et al. (1989). The right side of the figure shows schematically the variation in the two
parameters that we calculate, compressional-wave velocity and Q. Velocity increases to the top
of the reservoir, drops upon entering the reservoir and gradually increases with depth within the
reservoir. The Q increases to the top of the reservoir and then decreases within the reservoir.
We use these data to infer saturation conditions. We believe that conditions above the reservoir
are mostly liquid saturated. At the top of the reservoir the saturation drops dramatically and
then increases with depth. These conclusions are shown with a dashed line to stress that they are
an interpretation.

CONCLUSIONS

We have calculated the three-dimensional compressional-wave velocity and one-dimensional

compressional-wave Q structure of The Geysers region using local earthquakes. Our data for
the inversion consisted of P-wave arrival times and pulse widths. The calculated velocity
structure correlates well with the surface geology and published studies on the geologic structure
of the reservoir. The reservoir appears to exhibit low velocity compared to the surrounding
country rock. The Q decreases with depth within the reservoir, which we infer to indicate partial
liquid saturation (30 to 70%) at depth with drier conditions near the top of the reservoir.
Unfortunately, the Q inversion was limited to computing one-dimensional variations because
of the very noisy nature of the pulse width data. We feel that the noise primarily results from the
large errors in the arrival times relative to the pulse widths.

Acknowledgements--The authors would like to thank Debbie Wheeler of UNOCAL for her great effort in preparing the
data for us to use. Mitch Stark, Jens Pederson and Bill Cummins, also of UNOCAL, helped form our ideas during the
course of the work with extensive technical discussions. The authors would also like to thank two anonymousreviewers
 S e i s m i c V e l o c i t y a n d A t t e n u a t i o n S t r u c t u r e o f T h e G e y s e r s G e o t h e r m a l Field 125
and Ernie Majer of Lawrence Berkeley Laboratory for helpful comments. This work was performed at the Lawrence
Livermore National Laboratory and was supported under the Geothermal Division and the Assistant Secretary for
Energy Efficiency and Renewable Energy of the U.S. Department of Energy under contract W-7405-ENG-48.

REFERENCES

Bailey, E. H. (1946) Quicksilver deposits of the western Mayacamas district, Sonoma county, California. Cal. J. Mines
and Geology 42, 192-230.
Barker, B. J., Gulati, M. S., Bryan, M. A. and Riedel, K. L. (1992) Geysers Reservoir Performance. In Monograph on
The Geysers Geothermal Field (Edited by Stone, C.). Geotherm. Resour. Counc. Special Report No. 17,167-177.
Blakely, R. J. and Stanley, W. D. (1993) The Geysers magma chamber, California: constraints from gravity data,
density measurements, and well information. Geotherm. Resour. Counc. Trans. 17,227-233.
Brune, J. N, (1972) Tectonic stress and the spectra of seismic shear wavcs from earthquakes. J. geophys. Res. 75, 4997-
5010.
Eberhart-Phillips, D. (1986) Three-dimensional velocity structure in northern California Coast Ranges from inversion
of local earthquake arrival times. Bull. Seis. Soc. Am. 76, 1025-1052.
Ellsworth, W. L. (1977) Three-dimensional structure of the crust and upper mantle beneath the island of Hawaii. Ph.D.
thesis, Mass. Inst. Tech., Cambridge, U.S.A.
Evans, J. R. and Zucca, J. J. (1988) Active high-resolution tomography of compressional-wave velocity and attenuation
structure at Medicine Lake Volcano, Northern California Cascade Range. J. geophys. Res. 93, 15,016-15,036.
Evans, J. R. and Achauer, U. (1993) Teleseismic tomography using the ACH method, theory and application to
continental-scale studies. In Seismic Tomography, Theory and Practice (Edited by lyer, H. M. and Hirahara, K.).
Chapman and Hall, London.
Field Operators (1992) Maps of the top of the reservoir and the top of the felsite at The Geysers. Geotherm. Resour.
Counc. Special Report No. 17, back cover pocket.
Frankel, A. (1982) The effects of attenuation and site response on the spcctra of microearthquakes in the northeastern
Caribbean. Bull. Seis. Soc. Am. 72, 1379-1402.
Gambill, D. T. (1992) The recovery of injected water as steam in The Geysers. In Monograph on The Geysers (Edited by
Stone, C.). Geotherm. Resour. Counc. Special Report No. 17,159-163.
Gladwin, M. T. and Stacey, F. D. (1974) Anelastic degradation of acoustic pulses in rock. Phys. Earth Planet. Interiors
8332-8336.
Gregory, A. R. (1976) Fluid saturation effects on dynamic elastic properties of sedimentary rocks. Geophysics 41,895-
921.
Hanks, T. C. (1982) fn, ax. Bull. Seis. Soc. Am. 72, 1867-1879.
Hutchings, L. (1990) Empirical Green's functions from small earthquakes: A waveform study of locally recorded
aftershocks of the 1971 San Francisco earthquake. J. geophys. Res. 95, 1187-1214.
lto, H., DeVilbiss, J, and Nur, A. (1979) Compressional and shear waves in saturated rock during water-steam
transition. J. geophys. Res. 84, 4731-4735.
Kjartansson, E. (1979) Constant Q wave propagation and attenuation. J. geophys. Res. 84, 4737-4748.
Majer, E. L. and McEvilly, T. V. (1979) Seismological investigations of The Geysers geothermal field. Geophysics 44,
246-269.
Majer, E. L., McEvilly, T. V., Eastwood, F. S. and Myer, L. R. (1988) Fracture detection using P-wave and S-wave
vertical seismic profiling at The Geysers. Geophysics 53, 76-84.
McLaughlin, R. J. (1981) Tectonic setting of pre-tertiary rocks and its relation to geothermal resources in The Geysers-
Clear Lake area. In U.S. Geol. Surv. Prof. Pap. No. 1141, pp. 3-24.
MeNitt, J. R. (1968) Geology of the Kelseyville quadrangle, Map sheet 9, Cal. Div. Mines and Geology.
McNitt, J. R., Henneberger, R. C., Koenig, J. B. and Robertson-Tait, A. (1989) Stratigraphic and structural controls of
the occurrence of steam at The Geysers. Geotherm. Resour. Counc. Trans. 13,461-465.
Nolet, G. (1987) Seismic wave propagation and seismic tomography. In Seismic Tomography with Applications in
Global Seismology and Exploration Geophysics (Edited by Nolet, G.). D. Reidel, Dordrecht.
O'Connell, D. and Johnson, L. R. (1991) Progressive inversion for hypocenters and P wave and S wave velocity
structure: application to The Geysers, California, Geothermal Field. J. geophys. Res. 96, 6223-6236.
Pruess, K. and Narasimhan, T. N. (1982) On fluid reserves and the production of superheated steam from fractured,
vapor-dominated geothermal reservoirs. J. geophys. Res. 87, 9329-9339.
Richards, P. G. and Menke, W. (1983) The apparent attenuation of a scattering medium. Bull. Seis. Soc. Am. 73, 1005-
1022.
Stacey, F. D., Gladwin, M. T., McKavanagh, B., Linde, A. T. and Hastic, L. M. (1975) Anelastic damping of acoustic
and seismic pulses. Geophys. Survey 2,133-157.
Stark, M. A. (1992) Microearthquakes---a tool to track injected water in The Geysers reservoir. In Monograph on The
Geysers (Edited by Stone, C.). Geotherm. Res. Counc. Special Report No. 17, 111-117.
Thompson, R. C. and Gunderson, R. P. (1992) The orientation of steam-bearing fractures at The Geysers Geothermal
Field. In Monograph on The Geysers (Edited by Stone, C.). Geotherm. Resour. Counc. Special Report No. 17, 65-
68.
Thurber, C. H. (1983) Earthquake locations and three-dimensional crustal structure in the Coyote Lake area, central
California. J. geophys. Res. 88, 8226-8236.
126 J.J. Zucca ctal.
Toomey, D. R. and Foulger, G. R. (1989)Tomographic inversion of local earthquake data from the Hengill
Grcnsdalur central volcano complex, Iceland. ,l. geophys. Res. 94, 17,497-17,510.
Williamson, K. H. (1992) Development of a reservoir model for The Geysers geothermal field. In Mono,~,raph ~m 77~~
Geysers (Edited by Stone, ('.). Geothermal Rcsour. ('ounc. Special Report No. 17, 179-187.
Zucca. J. J. and Ewms, .1. R. (19~;'2) Active fiigh-resolutkm eomprcssional-wavc attclmation tomograph,, al Nex~bet* ,,
w)lcano, central (Tascadc Range .I..geophvv. Re.~. 97, I 1.047-1 1,055.