Bulletin of the Seismological Society of America, Vol. 92, No. 5, pp.

1923–1932, June 2002

Ground-Motion Attenuation Relationships for Cascadia Subduction Zone

Megathrust Earthquakes Based on a Stochastic Finite-Fault Model
by Nicholas J. Gregor, Walter J. Silva, Ivan G. Wong, and Robert R. Youngs

Abstract The number of strong ground motion recordings available for regression
analysis in developing empirical attenuation relationships has rapidly grown in the
last 10 years. However, the dearth of strong-motion data from the Cascadia subduc-
tion zone has limited this development of relationships for the Cascadia subduction
zone megathrust, which can be used in the calculation of design spectra for engi-
neered structures. A stochastic finite-fault ground-motion model has been used to
simulate ground motions for moment magnitude (M) 8.0, 8.5, and 9.0 megathrust
earthquakes along the Cascadia subduction zone for both rock-and soil-site condi-
tions. The stochastic finite-fault model was validated against the 1985 M 8.0 Mi-
choacan, Mexico, and the 1985 M 8.0 Valpariso, Chile, earthquakes. These two
subduction zone megathrust earthquakes were recorded at several rock sites located
near the fault rupture. For the Cascadia megathrust earthquakes, three different rup-
ture geometries were used to model the M 8.0, 8.5, and 9.0 events. The geometries
only differ in their respective fault lengths. A fault dip of 9⬚ to the east with a rupture
width of 90 km was selected to represent average properties of the Cascadia sub-
duction zone geometry. A regional crustal damping and velocity model was used
with the stochastic finite-fault model simulations. Ground motions were computed
for 16 site locations. The parametric uncertainties associated with the variation in
source, path, and site effects were included in the development of the ground motions.
A functional form was fit to the ground-motion model simulations to develop region-
specific attenuation relationships for the Cascadia megathrust rupture zone for both
rock and soil site conditions. The total uncertainty was based on a combination of
the modeling and parametric uncertainties (sigmas). These newly developed atten-
uation relationships for Cascadia subduction zone megathrust earthquakes can be
used in both the probabilistic and deterministic seismic-hazard studies for engineer-
ing design for the Pacific Northwest.

Introduction
The prediction of strong ground shaking in the Pacific Youngs et al., 1997), which have been traditionally used to
Northwest from a great megathrust (interface) earthquake estimate both deterministic and probabilistic ground-motion
rupturing the Cascadia subduction zone is hampered by the hazard in the Pacific Northwest for interface events have, by
absence of empirical data. Until recently, the only significant necessity, been derived from strong-motion recordings from
strong-motion data for the region were the recordings of the other subduction zones (e.g., Japan, Chile, and Mexico).
1949 surface magnitude (Ms) 7.1 Olympia and the 1965 Furthermore, the largest earthquake in these data is only a
body-wave magnitude (mb) 6.5 Seattle–Tacoma intraplate M 8.2 earthquake (Crouse, 1991). Whether these relation-
earthquakes and the 1992 moment magnitude (M) 7.2 Cape ships are appropriate for the Pacific Northwest in terms of
Mendocino, California, earthquake at the south end of the rupture plane geometry, earthquake stress drops, crustal at-
Cascadia subduction zone. The recent 28 February 2001 M tenuation (Q), velocity structure, and other source and path
6.8 Nisqually intraplate earthquake has increased the total parameters is uncertain. The validity of extrapolating these
number of strong ground motion recordings from the Pacific relationships up to M 9, the magnitude of the most recent
Northwest region by more than 10-fold. Thus, empirical earthquake in 1700 (Satake et al., 1996), is also a significant
attenuation relationships (Crouse, 1991; Atkinson, 1997; issue, especially for engineering design purposes.

1923
1924 N. J. Gregor, W. J. Silva, I. G. Wong, and R. R. Youngs

Existing Attenuation Relationships ation relationships are based on Cascadia megathrust strong-
motion records and because they are extrapolated above
For the probabilistic National Seismic Hazard Maps M 8.2, we have developed a set of attenuation relationships
(Frankel et al., 1996), the Youngs et al. (1997) empirical appropriate for megathrust earthquakes from M 8.0 to 9.0
attenuation relationship was the only subduction zone rela- in the Cascadia subduction zone. In seismic-hazard analysis,
tionship used for the modeling of ground motion from the it is often desired to use multiple attenuation relationships
Cascadia subduction zone megathrust source. Youngs et al. to incorporate the epistemic uncertainty in estimating ground
(1997) stated, however, that their relationships are poorly motions. Our relationships provide an alternative approach
constrained at small distances (⬍50 km) because of a lack to empirically based ground-motion estimation. For exam-
of data from more than a few earthquakes. To aid in esti- ple, in the development of probabilistic microzonation
mating ground motions at these short distances from large ground shaking maps for the Portland, Oregon, metropolitan
subduction events, Youngs et al. (1997) utilized the same area, ground motions from the Cascadia megathrust were
stochastic finite-fault model that was used in this study to estimated using the Youngs et al. (1997) relationships and
simulate ground motions for M 7.0, 8.0, and 8.5 megathrust the stochastic finite-fault model-based relationship described
earthquakes. Their simulations predicted higher near-field in this article (Wong et al., 2000).
ground motions than those extrapolated from empirical data. The attenuation relationships developed in this study are
The simulations predicted ground motions that were com- for 5%-damped, horizontal response spectral acceleration.
parable in amplitude to the ground motions predicted by Because these relationships were originally developed for
empirical attenuation relationships for shallow-crustal use in preparing the Portland microzonation hazard maps,
reverse-faulting earthquakes (e.g., Sadigh et al., 1993). the input parameters described subsequently are most appro-
Thus, for short distances, Youngs et al. (1997) suggested priate for northwestern Oregon and southwestern Washing-
that higher ground motions be considered than those empir- ton. However, we believe they are also applicable to many
ically predicted. Based on this recommendation, the 1996 other areas along the Cascadia subduction zone.
National Hazard Maps also included the Sadigh et al. (1993)
empirical attenuation relationship for reverse faulting for
modeling ground motions from Cascadia subduction zone
interface events. Approach
Estimating ground motions at short distances (⬍50 km) The computed ground motions used in the attenuation
for a Cascadia megathrust rupture is important in the Pacific relationships are based on a stochastic finite-fault ground-
Northwest. Based on a model of megathrust rupture extend- motion model developed by Silva et al. (1990). The target
ing halfway into the transition zone of Flück et al. (1997), earthquake ground motions are modeled as a combination of
horizontal distances from the eastern edge of a Cascadia smaller subevent earthquakes occurring over the target rup-
megathrust rupture to the coast range from zero at Cape ture plane. This is the same model used by Youngs et al.
Mendocino, California; Cape Blanco, Oregon; and along the (1997) in their development of the empirical attenuation re-
western Washington coast to about 40–50 km along the lationship for subduction-zone earthquakes. A revised ver-
northern Oregon coast (Wong and Silva, 1998). The heavily sion of this model is described in detail by Schneider et al.
urbanized areas of Puget Sound region in Washington and (1993) and Silva et al. (1998).
the Portland metropolitan region and the Willamette Valley This stochastic finite-fault ground-motion model has
in Oregon are at approximate horizontal distances of 100– been used to simulate site-specific ground motions for en-
160 km from the Cascadia megathrust. gineering design for M 8.5–9.0 Cascadia megathrust events
The attenuation relationships of Crouse (1991), which in the Pacific Northwest (Silva et al., 1998; Wong and Silva,
are appropriate for firm soil conditions, do not differentiate 1998). The advantage of employing finite-fault numerical
between the ground motions from interface and intraplate simulations is that unlike empirical attenuation relationships,
earthquakes. This is a significant consideration because it which contain an uneven sampling of site and source ge-
appears that ground motions from these two types of sub- ometries based on the available strong-motion dataset, finite-
duction zone earthquakes differ in amplitude for a given fault effects such as rupture propagation, directivity, and
magnitude and distance (Youngs et al., 1988, 1997). source-to-site geometry can be systematically accounted for.
Atkinson (1997) and Atkinson and Boore (1997) de- For this study the stochastic finite-fault model was used
veloped attenuation relationships for the Cascadia region, to simulate subduction zone megathrust events of M 8.0, 8.5,
but they did not distinguish among crustal, intraplate, and and 9.0 on the Cascadia subduction zone. Uncertainties in
megathrust events. The Atkinson (1997) relationships are source, path, and site parameters are included in the com-
empirically derived, whereas those of Atkinson and Boore putations through parametric variations. Uncertainties in the
(1997) are based on a stochastic point-source numerical regression of the simulated data are added to the modeling
ground-motion model. uncertainty to produce 16th, 50th (median), and 84th per-
Because none of the aforementioned empirical attenu- centile attenuation relationships.
Ground-Motion Attenuation Relationships for Cascadia Subduction Zone Megathrust Earthquakes 1925

Model Validation
The stochastic finite-fault model has been validated for
crustal earthquakes using 15 earthquakes recorded at about
500 sites (W. J. Silva et al., unpublished report, 1997). For
subduction zone earthquakes, the stochastic finite-fault
model has been validated using the recorded ground motions
from two large subduction zone megathrust earthquakes
(Humphrey et al., 1993). The first large subduction earth-
quake used in the validation of the finite-fault modeling was
the 19 September 1985 M 8.0 Michoacan, Mexico, earth-
quake. The slip model of Mendez and Anderson (1991) was
used to model the 5%-damped response spectral accelera-
tions at 14 rock sites. Epicentral distances to the sites ranged
from 20 to 400 km, and shortest distances to the rupture
surface range from 15 to 250 km. Site-specific kappa values
(Humphrey and Anderson, 1993; Silva and Darragh, 1995)
were used in the modeling process for each site. A local
crustal velocity model (Somerville et al., 1991) and a local
crustal damping model of Humphrey and Anderson (1993)
were used for the modeling of the Michoacan data. In gen-
eral the stochastic finite-fault ground motions agree favora-
bly with the recorded ground motions at periods of 0.02 to
10–20 sec. For the four closest sites, the model predicts
higher ground motions than were recorded. This type of
overprediction of ground motion from megathrust events at
short distances has been previously noted by Youngs et al.
(1997). A quantitative measure of the goodness-of-fit will
be discussed subsequently for the combined Michoacan and
Valpariso earthquakes.
Compared to the Michoacan earthquake, the 3 March Figure 1. (a) Modeling uncertainty (solid) and
1985 M 8.0 Valpariso, Chile, earthquake has substantially bias-corrected modeling uncertainty (dashed) (both
rln) and (b) model bias with 90% confidence limits
fewer recorded ground motions and less information on the (dotted line) computed for the combined data from
recording stations for the validation of the model. There the 1985 Michoacan, Mexico, and Valpariso, Chile,
were six rock strong ground motion stations ranging in rup- earthquakes.
ture distance from 39 to 120 km that were modeled from
this earthquake. The slip model of Somerville et al. (1991)
was used in the modeling. The Michoacan regional crustal 1 Hz, the total modeling uncertainty (rln) is low, generally
damping model (Humphrey and Anderson, 1993) and the below about 0.4. The modeling bias (Abrahamson et al.,
crustal amplification factors of Boore (1986) were used for 1990) is shown in the lower plot in Figure 1 for the combined
the Valpariso earthquake because of a lack of an appropriate data along with 90% confidence intervals (dotted lines).
crustal-damping model for the region. Site-specific kappa Above 0.6 Hz, the bias is essentially zero. The negative bias
values were estimated based on the comparison of response values for longer periods (i.e., frequencies less than about
spectral shapes with the spectral shapes for known kappa 0.5 Hz) indicate an overprediction of the ground motions by
values (Silva and Darragh, 1995). The agreement between the model compared to the recorded ground motions. The
the recorded ground motions and the model ground motions bias-corrected modeling uncertainty (Abrahamson et al.,
are similar to the agreement for the Michoacan data. 1990) is shown in the top part of Figure 1 as the dashed line.
To provide a quantitative measure of the uncertainties This uncertainty will be included with the parametric re-
in the ground-motion predictions, a simple goodness-of-fit gression uncertainty to give a total uncertainty in the atten-
was performed at each spectral period. The modeling un- uation relationship.
certainty is the average at each frequency of the difference
of the natural logarithms of the observed ground motions Input Parameters
and the model-predicted ground motions. The uncertainty The seismic source, path, and site parameters used in
for the combined Michoacan and Valpariso data is shown in the stochastic finite-fault ground-motion simulations are de-
the top part of Figure 1. For frequencies greater than about scribed in the following sections.
1926 N. J. Gregor, W. J. Silva, I. G. Wong, and R. R. Youngs

Seismic Source. The rupture planes for the M 8.0, 8.5, and the east, which is an average value for the central portion of
9.0 events were modeled as being 90 km wide and 150, 450, the subduction zone. The top of the rupture plane was placed
and 1100 km long, respectively (Fig. 2). The width of the at a depth of 5 km, and the bottom was placed at about 20
potential megathrust rupture varies along the length of the km based on the model of Wong and Silva (1998) for north-
subduction zone from approximately 50 km in southern western Oregon and southwestern Washington.
Oregon and the northwest corner of California to about 150 For each simulation, a different randomized slip model
km in the Olympic Peninsula (Hyndman and Wang, 1995; was used. These models were generated using a scheme that
Flück et al., 1997). The 90-km width is an average value preserves the number and size of asperities given a rupture
assuming that rupture involves the locked zone and extends area to observed slip models from past earthquakes. The
halfway into the transition zone of the Flück et al. (1997) amplitude of slip was also varied over the fault plane. This
model of the Cascadia subduction zone (Wong and Silva, slip generation scheme is based on a suite of actual slip dis-
1998). This width is the same width used in the National tributions determined from inversions (Abrahamson et al.,
Seismic Hazard Map modeling for the Cascadia subduction 1994). Reverse slip was assumed with a variation in the rake
zone (Frankel et al., 1996). The extent to which the transition angle of 30⬚.
zone is involved in coseismic rupture is highly uncertain, In addition to randomized slip models, the point of rup-
and the validity of the Cascadia subduction zone model of ture initiation was also randomized, although it was con-
Flück et al. (1995) and its predecessors (e.g., Hyndman et strained to be located in an area occupying the lower half of
al., 1997) is still being evaluated (Wong and Silva, 2000). the rupture and within 10% from the ends (dashed box in
The rupture planes were modeled as rectangular areas Fig. 2). The average of the suite of random hypocenters was
(Fig. 2). To define the lengths of the ruptures, the Wells and selected to be approximately in the center of the north–south
Coppersmith (1994) empirical relationship between rupture extent of the fault plane and three-fourths the width down
area and magnitude was used. Using an empirical relation- the fault plane.
ship for rupture area and magnitude derived from subduction
zone earthquakes (Abe, 1981, 1984) yields very similar Crustal Model and Attenuation. A frequency-dependent
earthquake fault areas to that of Wells and Coppersmith Q(f ) relationship for the Cascadia subduction zone region
(1994). The rupture planes were modeled as dipping 9⬚ to was used in the simulations where Q(f ) ⳱ Qo f g. Values of
Qo of 380 and g of 0.39 were adopted from Atkinson’s
(1995) estimate for western Washington and southwestern
Canada. The parameter Qo was randomly varied assuming a
rln of 0.40. The parameter g was held fixed at 0.39 in the
ground-motion simulations.
A P-wave crustal model was developed in this study
based on the models of Trehu et al. (1994), Cohee et al.
(1991), and Ludwin et al. (1991). Based on this model, S-
wave velocities were calculated assuming a Poisson’s ratio
of 0.25. These models are listed in Table 1, along with the
densities from Couch and Riddihough (1989).

Site Model. Ground motions for both rock-and soil-site

Table 1
Crustal Model for Northwest Oregon–Southwest Washington
Depth* Vp Vs q†
(km) (km/sec) (km/sec) (g/cm3)

0.0 0.29–3.03 0.17–1.75 2.00–2.20

0.3 3.46 2.00 2.30
1.5 4.50 2.60 2.50
3.5 5.10 2.94 2.60
5.5 5.70 3.29 2.65
8.0 6.30 3.64 2.70
11.5 6.60 3.80 2.80
Figure 2. Megathrust rupture and site geometry 16.0 6.73 3.89 2.85
used to calculate finite-fault stochastic ground mo- 20.0 6.86 3.96 2.90
tions used in regressing attenuation relationships. The 25.0 6.95 4.01 2.90
rupture initiation location is constrained to be within 41.0 7.80 4.50 3.00
the dashed box. Ground motions are computed for the
16 sites shown in as triangles along the two limbs of *Depth to top of layer.
†
stations. Densities from Couch and Riddihough (1989).
Ground-Motion Attenuation Relationships for Cascadia Subduction Zone Megathrust Earthquakes 1927

conditions were generated. We assumed that the top 0.3 km This oversaturation of ground motion is observed in the
of the crustal velocity model (Table 1) for rock-site condi- preferred nonparametric empirical attenuation model of An-
tions corresponded to Columbia River basalt (CRB), which derson (1997) for subduction events recorded in the Guer-
occurs throughout a large portion of southwestern and south- rero array in Mexico. The third model presented by Ander-
central Washington and northwestern Oregon. The CRB was son (1997), which is not significantly different statistically
modeled with an S-wave velocity gradient based on Mabey than the preferred model and contains greater smoothing of
and Madin (1993). The 300-m-thick soil profile was devel- the data, has magnitude saturation but not oversaturation.
oped based on an analysis of geologic borehole profiles in Other published attenuation models for both subduction
the Portland region for Quaternary alluvium surface soil con- (e.g., Youngs et al., 1997) and crustal (e.g., Abrahamson
ditions. The average shear-wave velocities for the rock and and Silva, 1997) earthquakes do not have an overstaturation
soil profiles in the upper 30 m are 363 m/sec and 182 m/sec, of ground motion. From an engineering design viewpoint,
respectively. These profiles were randomized for the ground- any attenuation model that produces lower ground motions
motion simulations using a correlation model for seismic for larger events at the same distances is difficult to justify,
velocities and layer thicknesses developed by G. Toro (W. J. given our current understanding of earthquake ground mo-
Silva et al., unpublished report, 1997). A total low-strain tions. Although limited in the empirical data from large
kappa value of 0.04 sec was used for both the rock and soil earthquakes at short distances, the Youngs et al. (1997) at-
sites. This base kappa value was also randomized assuming tenuation relationship, as well as other crustal attenuation
a rln of 0.30 (Electric Power Research Institute [EPRI], 1993) relationships (e.g., Abrahamson and Silva, 1997), were de-
for the ground-motion simulations. veloped with the constraint of ground-motion saturation but
The top 150 m of both the rock and soil profiles were not oversaturation.
modeled with nonlinear rock and soil site properties (EPRI, For the M 9.0 event, the fault length is 1100 km, which
1993). The relatively low S-wave velocities (170 m/sec) near means that the two limbs of stations used in the simulations
the surface for the CRB suggest that this profile will produce are separated by 550 km. For the limb of stations located at
ground motions that are similar to shallow soil site condi- the southern end of the rupture plane, the ground motions
tions, typical of weathered CRB. As a result, high-frequency tend to be lower in amplitude and longer in duration than
motions, such as peak acceleration, may be expected to be for the stations located along the middle limb because of the
higher on CRB than stiffer rock-site conditions, all else being large separation distance from the southern end of the fault
equal (EPRI, 1993; W. J. Silva et al., unpublished report, plane to the rest of the rupture plane. This type of variation
1999). in ground motions is also observed in large crustal earth-
quakes (e.g., 1992 Landers, California) due to the propagat-
Results ing rupture along the finite-rupture plane. Hence, for the
shorter rupture plane of the M 8.0 (150 km long) event, these
Simulations for M 8.0, 8.5, and 9.0 events were per- effect are not as pronounced. The effect is also reduced as
formed at eight sites along each of two east–west profiles the stations are located further to the east of the rupture plane
(Fig. 2). One profile was placed at the southern end of the and the rupture distance increases. Based on these factors
rupture plane, and the other was placed at the center of the and the constraint of the model for staturation and not over-
rupture plane. Sites along each limb were placed directly saturation, the modeled ground motions from the two limbs
along the centerline of the rupture plane (at approximately of the M 8.0 event were weighted equally. For the M 8.5
45-km distance) and above its eastern edge. The remaining event, weights of 0.7 and 0.3 were assigned to the central
six sites for each limb were placed at horizontal distances of and southern limb of stations, respectively, and 0.86 and
10, 20, 50, 100, 200, and 500 km (Fig. 2). 0.14 to the M 9.0 event.
A total of 20 simulations were made for each magni- The simulation results were fit to a functional form of
tude, distance, and site location, and the resulting 5%-
damped acceleration response spectra were fitted with an LnY ⳱ C1 Ⳮ C2*M Ⳮ (C3 Ⳮ C4*M)*Ln[R
attenuation functional form that accommodates magnitude-
dependent ground-motion saturation and far-field distance- Ⳮ exp(C5)] Ⳮ C6*(M ⳮ 10)3, (1)
dependent fall-off. A weighting scheme was applied to the
data from each limb of stations to control the magnitude where Y is the peak ground-motion parameter, R is closest
saturation at short distances, thereby preventing the regres- distance to the rupture plane, and C1–C6 are coefficients fit
sion model from predicting lower ground motions for larger to the data for rock (Table 2) and soil site conditions (Table
magnitude events at the same short distances. If this weight- 3). The total uncertainty in the regression results are also
ing scheme constraint was not used in the regression pro- listed in Tables 2 and 3, where these values include the vec-
cedure, the results based on the simulations of the M 8.5 tor sum of the modeling and parametric uncertainty. The
and 9.0 events would produce lower ground-motion esti- regressions were performed out to a period of 5.0 sec.
mates at short distances (i.e., for sites located over the fault Figure 3 illustrates the range of simulated peak hori-
plane) than those from the M 8.0 event. zontal ground accelerations (PGA) for rock-site conditions
1928 N. J. Gregor, W. J. Silva, I. G. Wong, and R. R. Youngs

Table 2
Coefficients and Standard Errors for Rock-Site Conditions
Period (sec) C1 C2 C3 C4 C5 C6 Param. Sigma Model Sigma Total Sigma

PGA 21.0686 ⳮ1.7712 ⳮ5.0631 0.4153 4.2 0.0017 0.6083 0.3926 0.7240
0.010 20.9932 ⳮ1.7658 ⳮ5.0404 0.4132 4.2 0.0226 0.6031 0.3926 0.7195
0.020 21.072 ⳮ1.772 ⳮ5.0529 0.4142 4.2 0.0025 0.6036 0.3926 0.7195
0.025 21.152 ⳮ1.779 ⳮ5.0663 0.4154 4.2 0.0023 0.6042 0.3983 0.7235
0.032 21.366 ⳮ1.797 ⳮ5.1036 0.4187 4.2 0.0017 0.6062 0.3926 0.7221
0.040 17.525 ⳮ1.339 ⳮ4.8602 0.3868 4.2 ⳮ0.0318 0.5836 0.3818 0.6969
0.050 19.347 ⳮ1.519 ⳮ4.9731 0.3960 4.2 ⳮ0.0155 0.5908 0.3925 0.7086
0.056 20.774 ⳮ1.625 ⳮ5.1875 0.4118 4.3 ⳮ0.0155 0.5974 0.4052 0.7215
0.063 21.331 ⳮ1.672 ⳮ5.2561 0.4173 4.3 ⳮ0.0146 0.6028 0.4132 0.7302
0.071 24.221 ⳮ1.924 ⳮ5.6250 0.4478 4.4 ⳮ0.0071 0.6116 0.4042 0.7326
0.083 24.950 ⳮ1.979 ⳮ5.6696 0.4493 4.4 ⳮ0.0018 0.6337 0.4584 0.7815
0.100 30.005 ⳮ2.349 ⳮ6.3862 0.5009 4.7 ⳮ0.0019 0.6448 0.4668 0.7954
0.125 39.719 ⳮ3.090 ⳮ7.8541 0.6161 5.1 ⳮ0.0064 0.6654 0.5461 0.8605
0.143 43.414 ⳮ3.385 ⳮ8.3122 0.6513 5.2 ⳮ0.0001 0.6769 0.5225 0.8544
0.167 39.579 ⳮ2.957 ⳮ7.9723 0.6139 5.2 ⳮ0.0264 0.6810 0.5050 0.8478
0.200 39.345 ⳮ3.087 ⳮ7.6002 0.5972 5.1 0.0060 0.7034 0.5089 0.8679
0.250 37.690 ⳮ2.960 ⳮ7.3790 0.5842 5.1 ⳮ0.0023 0.7121 0.4539 0.8444
0.333 34.787 ⳮ2.899 ⳮ6.7855 0.5616 4.9 0.0256 0.7372 0.4764 0.8776
0.400 33.393 ⳮ2.776 ⳮ6.9595 0.5863 4.9 ⳮ0.0039 0.7110 0.5187 0.8801
0.500 29.159 ⳮ2.424 ⳮ6.2114 0.5216 4.7 0.0161 0.6745 0.4382 0.8039
0.769 15.279 ⳮ1.220 ⳮ4.3240 0.3618 3.9 ⳮ0.0011 0.6111 0.5611 0.8295
1.000 6.528 ⳮ0.406 ⳮ3.1991 0.2589 3.2 ⳮ0.0225 0.5898 0.4751 0.7567
1.667 7.467 ⳮ0.676 ⳮ2.6465 0.2193 2.8 0.0416 0.4931 0.4889 0.6943
2.000 8.657 ⳮ0.851 ⳮ2.7398 0.2339 2.8 0.0370 0.4666 0.4247 0.6305
2.500 6.637 ⳮ0.651 ⳮ2.3124 0.1879 2.8 0.0364 0.4163 0.5198 0.6657
5.000 8.013 ⳮ0.943 ⳮ2.4087 0.2154 2.3 0.0647 0.3931 0.6656 0.7730

Table 3
Coefficients and Standard Errors for Soil-Site Conditions
Period (sec) C1 C2 C3 C4 C5 C6 Param. Sigma Model Sigma Total Sigma

PGA 23.8613 ⳮ2.2742 ⳮ4.8803 0.4399 4.7 0.0366 0.3760 0.3926 0.5436
0.010 25.4516 ⳮ2.4206 ⳮ5.1071 0.4605 4.8 0.0372 0.3742 0.3926 0.5422
0.020 25.4339 ⳮ2.4185 ⳮ5.1044 0.4602 4.8 0.0370 0.3742 0.3926 0.5422
0.025 25.4200 ⳮ2.4168 ⳮ5.1026 0.4600 4.8 0.0369 0.3743 0.3983 0.5464
0.032 25.3849 ⳮ2.4127 ⳮ5.0977 0.4594 4.8 0.0366 0.3743 0.3926 0.5422
0.040 22.7042 ⳮ2.1004 ⳮ4.9006 0.4353 4.8 0.0164 0.3590 0.3818 0.5241
0.050 23.2948 ⳮ2.1619 ⳮ4.8855 0.4332 4.8 0.0263 0.3592 0.3925 0.5319
0.056 23.2165 ⳮ2.1528 ⳮ4.8744 0.4319 4.8 0.0255 0.3598 0.4052 0.5413
0.0625 24.7067 ⳮ2.2814 ⳮ5.0947 0.4509 4.9 0.0245 0.3607 0.4132 0.5480
0.070 24.9425 ⳮ2.3045 ⳮ5.0672 0.4476 4.9 0.0295 0.3609 0.4042 0.5413
0.083 26.5395 ⳮ2.4402 ⳮ5.3025 0.4677 5.0 0.0276 0.3617 0.4584 0.5835
0.100 29.9693 ⳮ2.7254 ⳮ5.8054 0.5098 5.2 0.0226 0.3654 0.4668 0.5926
0.125 35.6660 ⳮ3.1853 ⳮ6.6251 0.5769 5.5 0.0123 0.3821 0.5461 0.6665
0.143 50.7368 ⳮ4.5292 ⳮ8.7213 0.7649 5.9 0.0108 0.3923 0.5225 0.6532
0.167 55.6402 ⳮ4.9662 ⳮ9.5555 0.8435 6.0 ⳮ0.0070 0.3927 0.5050 0.6393
0.200 75.8218 ⳮ6.8396 ⳮ12.0687 1.0753 6.3 0.0096 0.4231 0.5089 0.6618
0.250 100.3357 ⳮ9.0324 ⳮ15.3511 1.3731 6.6 ⳮ0.0043 0.4472 0.4539 0.6371
0.330 71.7967 ⳮ6.4990 ⳮ11.6056 1.0415 6.2 0.0102 0.4324 0.4764 0.6431
0.400 67.3720 ⳮ6.1755 ⳮ11.1567 1.0167 6.1 0.0035 0.4243 0.5187 0.6699
0.500 56.0088 ⳮ5.1176 ⳮ9.5083 0.8632 5.9 0.0164 0.4305 0.4382 0.6139
0.770 26.3013 ⳮ2.4482 ⳮ5.3818 0.4957 4.8 0.0259 0.4601 0.5611 0.7256
1.000 17.2330 ⳮ1.5506 ⳮ4.3287 0.3930 4.2 0.0133 0.4599 0.4751 0.6606
1.670 11.9971 ⳮ1.1180 ⳮ2.9451 0.2639 3.7 0.0538 0.4781 0.4889 0.6837
2.000 17.9124 ⳮ1.7505 ⳮ3.8150 0.3574 4.1 0.0583 0.4628 0.4247 0.6276
2.500 16.1666 ⳮ1.5091 ⳮ3.7101 0.3344 4.1 0.0473 0.4193 0.5198 0.6676
5.000 7.4856 ⳮ0.8360 ⳮ2.0627 0.1779 ⳮ0.2 0.0821 0.4802 0.6656 0.8207
Ground-Motion Attenuation Relationships for Cascadia Subduction Zone Megathrust Earthquakes 1929

Figure 4. Comparison between Youngs et al.

(1997) model and this model for peak horizontal ac-
celeration for rock-site conditions for M 8.0 and 9.0.

The median PGA attenuation relationships for M 8.0

and 9.0 are shown on Figure 4. Due to the magnitude satu-
ration constraint, the attenuation model predicts about 0.5g
for both magnitudes at a distance of 10 km for rock-site
conditions. This level of magnitude saturation is similar to
the magnitude saturation observed in the Youngs et al.
(1997) empirical attenuation curves, which are also plotted
in Figure 4 for M 8.0 and 9.0. The median Youngs et al.
(1997) curves are shown with a hypocentral depth of 16 km,
which represents an average depth of rupture initiation used
in the generation of the model ground motions. As noted by
Youngs et al. (1997), the simulations exceed the empirical
predictions by a factor of approximately 2 at distances less
than 50 km. An approach using the equally weighted Youngs
Figure 3. Regression of finite-fault simulations et al. (1997) and Sadigh et al. (1997) relationships, as rec-
for (a) M 8.0 and (b) 9.0 earthquakes for rock-site ommended by R. Youngs (Geomatrix Consultants, personal
conditions. Triangles represent results for the central comm., 1999), would result in PGAs of about 0.5g at 10 km,
limb and pluses for the southern limb indicating the which is approximately equal to the value predicted by our
lower simulated ground motions for the southern limb relationship (Fig. 4). The simulations exceed the Youngs et
of stations.
al. (1997) predictions at longer distances.
Figure 5 is a plot of the PGA attenuation with rupture
distance for soil-site conditions. Also shown are the Youngs
for each site location profile and the best-fitting relationship et al. (1997) relationships for M 8.0 and 9.0 events. Both
for M 8.0 and 9.0 events. The M 9.0 simulations show a relationships predict similar PGA values at distance out to
greater variability in PGA than do the M 8.0 events based 100–150 km. At greater distances, the current relationship
on the larger separation distance of the two station limbs has a slower rate of ground-motion attenuation and conse-
relative to the total length of the rupture plane, as discussed quently predicts larger PGA values. This difference in far-
previously. The soil ground-motion simulations show the field attenuation can be attributable to the softer soil profile
same variation in ground motion with magnitude and site used in our ground-motion simulations when compared to
locations. the average soil profiles of the empirical data recorded at
1930 N. J. Gregor, W. J. Silva, I. G. Wong, and R. R. Youngs

Figure 5. Comparison between Youngs et al. Figure 6. Horizontal median response spectra
(1997) model and this model for peak horizontal ac- (5%-damped) for rock-and soil-site conditions for a
celeration for soil-site conditions for M 8.0 and 9.0. M 8.5 earthquake at distances of 50 and 100 km es-
timated by the current model.

large distances used in the Youngs et al. (1997) attenuation

relationship.
Comparisons between the PGA attenuation relation-
ships for both rock and soil from this study and the relation-
ships from the previous finite-fault studies for the Cascadia
megathrust (Youngs et al., 1997) indicates a lower rate of
attenuation and higher PGA values at longer distances (Fig.
4, 5). Youngs et al. (1997) used the Cohee et al. (1991)
crustal velocity and the Humphrey and Anderson (1993)
crustal attenuation model. These differ slightly from the
models used in the current ground-motion simulations. Also,
in the previous modeling, crustal amplification factors were
applied based on the Cohee et al. (1991) velocity model,
which has faster velocities than the model used in this study
(Table 1). This faster velocity structure leads to smaller
ground-motion amplification than the slower velocity struc-
ture model used in the present study. Analysis of subduction
zone earthquake strong-motion data recorded on shallow
stiff soil sites indicated approximately 50% higher peak ac- Figure 7. Comparison of horizontal median re-
celerations than data recorded on rock sites (Geomatrix Con- sponse spectra (5%-damped) between Youngs et al.
sultants, unpublished report, 1993). (1997) model and this model for rock-site conditions
for a M 8.5 earthquake at distances of 50 and 100 km.
Median (5%-damped) response spectra for both rock
and soil are shown in Figure 6 for a M 8.5 event at rupture
distances of 50 and 100 km. The response spectra contain
small peaks and valleys because the regression coefficients ues (Fig. 7). Our model for soil-site conditions has a spectral
have not been smoothed. A comparison between the current peak at a longer spectral period (i.e., 0.7 sec) than either the
rock spectra and the predicted rock spectra from the Youngs Youngs et al. (1997) or Crouse (1991) attenuation models.
et al. (1997) attenuation model indicate a similar spectral This is an expected result based on the relatively soft soil
shape, with our model predicting higher ground-motion val- profile used in the ground-motion simulations.
Ground-Motion Attenuation Relationships for Cascadia Subduction Zone Megathrust Earthquakes 1931

Acknowledgments Mabey, M. A., and I. P. Madin (1993). Ground motion amplification map,
Portland quadrangle, Oregon, in Earthquake Hazard Maps of the
This article was greatly improved based on the comments and sug- Portland Quadrangle, Multnomah and Washington Counties,
gestions by two anonymous reviewers. This study was supported by the Oregon, and Clark County, Washington, Oregon Department of
U.S. Geological Survey under NEHRP Award 1434-HQ-96GR-02727. Our Geology and Mineral Industries Geological Map Series GMS-79.
thanks to Rachel Griener, Fumiko Goss, and Melinda Lee for assisting in Mendez, A. J., and J. G. Anderson (1991). The temporal and spatial evo-
the preparation of this article. lution of the 10 September 1985 Michoacan earthquake as inferred
from near-source ground-motion records, Bull. Seism. Soc. Am. 81,
844–861.
References Sadigh, K., C.-Y. Chang, N. A. Abrahamson, S. J. Chiou, and M. S. Power
(1993). Specification of long-period ground motions: updated atten-
Abe, K. (1981). Magnitude of large shallow earthquakes from 1904–1980, uation relationships for rock site conditions and adjustment factors
Phys. Earth Planet. Interiors 27, 72–92. for near-fault effects, in Proceedings ATC-17-1 Seminar on Seismic
Abe, K. (1984). Complements to “Magnitudes of large shallow earthquakes Isolation, Passive Energy Dissipation, and Active Control, 59–70.
from 1904–1980,” Phys. Earth Planet. Interiors 34, 17–23. Sadigh, K., C.-Y. Chang, J. A. Egan, F. Makdisi, and R. R. Youngs (1997).
Abrahamson, N. A., W. J. Silva, and J. Schneider (1994). Variability of Attenuation relationships for shallow crustal earthquakes based on
strong ground motion due to the variability of seismic slip distribution California strong motion data, Seism. Res. Lett. 68, 180–189.
(abstract), Seism. Res. Lett. 65, 35. Satake, K., K. Shimazaki, Y. Tsuji, and K. Ueda (1996). Time and size of
Abrahamson, N. A., P. G. Somerville, and C. A. Cornell (1990). Uncer- a giant earthquake in Cascadia inferred from Japanese tsunami records
tainty in numerical strong motion proceedings, in Proc. Fourth U.S. of January 1700, Nature 179, 246–249.
National Conference on Earthquake Engineering, Vol. 1, p. 407–416. Schneider, J. F., W. J. Silva, and C. L. Stark (1993). Ground motion model
Anderson, J. G. (1997). Nonparametric description of peak acceleration for the 1989 M6.9 Loma Prieta earthquake including effects of source,
above a subduction thrust, Seism. Res. Lett. 68, 86–93. path, and site, Earthquake Spectra 9, 251–287.
Atkinson, G. M. (1995). Attenuation and source parameters of earthquakes Silva, W. J., and R. B. Darragh (1995). Engineering characterization of
in the Cascadia region, Bull. Seism. Soc. Am. 85, 1327–1342. strong ground motion recorded at rock sites, Electric Power Research
Atkinson, G. M. (1997). Empirical ground motion relations for earthquakes Institute, Rept. EPRI TR-102262.
in the Cascadia region, Can. J. Civ. Eng. 24, 64–77 Silva, W., R. Darragh, C. Stark, I. Wong, J. Stepp, J. Schneider, and S.
Atkinson, G. M., and D. M. Boore (1997). Stochastic point-source mod- Chiou (1990). A methodology to estimate design response spectra in
eling of ground motions in the Cascadia region, Seism. Res. Lett. 68, the near-source region of large earthquakes using the Band-Limited-
74–85. White-Noise ground motion model, in Proc. Fourth U.S. National
Boore, D. M. (1986). Short-period P- and S- wave radiation from large Conference on Earthquake Engineering, Vol. 1, 487–494.
earthquakes: implications for spectral scaling relations, Bull. Seism. Silva, W. J., I. G. Wong, and R. B. Darragh (1998). Engineering charac-
Soc. Am. 76, 43–64. teristics of earthquake strong ground motions in the Pacific Northwest,
Cohee, B. R., P. G. Somerville, and N. A. Abrahamson (1991). Simulated in Assessing Earthquake Hazards and Reducing Risk in the Pacific
ground motions for hypothesized Mw ⳱ 8 subduction earthquakes in Northwest, A. M. Rogers, T. J. Walsh, W. J. Kockelman, and G. R.
Washington and Oregon, Bull. Seism. Soc. Am. 81, 28–56. Priest (Editors), U.S. Geol. Surv. Profess. Pap. 1560, Vol. 2, 313–
Couch, R. W., and R. P. Riddihough (1989). The crustal structure of the 324.
western continental margin of North America, in Geophysical Frame- Somerville, P. G., M. Sen, and B. Cohee (1991). Simulation of strong
work of the Continental United States, L. C. Pakiser and W. D. Moo- ground motions recorded during the 1985 Michoacan, Mexico, and
ney (Editors), Geological Society of America Memoir 172, 103–128. Valparaiso, Chile, earthquakes, Bull. Seism. Soc. Am. 81, 1–27.
Crouse, C. B. (1991). Ground motion attenuation equations for earthquakes Trehu, A. M., I. Asudeh, T. M. Brocher, J. H. Luetgert, W. D. Mooney,
on the Cascadia subduction zone, Earthquake Spectra 7, 201–236. J. L. Nabelek, and Y. Nakamura (1994). Crustal architecture of the
Electric Power Research Institute (1993). Guidelines for determining design Cascadia forearc, Science 266, 237–242.
basis ground motions, Electric Power Research Institute, Rept. EPRI Wells, D. L., and K. J. Coppersmith (1994). New empirical relationships
TR-102293, 1–5 among magnitude, rupture length, rupture width, and surface displace-
Flück, P., R. D. Hyndman., and K. Wang (1997). 3-D dislocation model ment, Bull. Seism. Soc. Am. 84, 974–1002.
for great earthquakes of the Cascadia subduction zone, J. Geophys. Wong, I. G., and W. J. Silva (1998). Earthquake ground shaking hazards
Res. 102, 20,539–20,550. in the Portland and Seattle metropolitan areas, in Geotechnical Earth-
Frankel, A., C. Mueller, T. Barnard, D. Perkins, E. V. Leyendecker, N. quake Engineering and Soil Dynamics III, P. Dakoulas, M. Yegian,
Dickman, S. Hanson, and M. Hopper (1996). National seismic-hazard and R. D. Holtz (Editors), American Society of Civil Engineering
maps; documentation June 1996, U.S. Geol. Surv. Open-File Rept Geotechnical Special Publication ASCE, no. 75, Vol. 1, 66–78.
96-532, 110 pp. Wong, I. G., and W. J. Silva (2000). Predicting great earthquake ground
Humphrey, J. R., and J. G. Anderson (1993). Shear wave attenuation and shaking in the Pacific Northwest from the Cascadia subduction zone,
site response in Guerrero, Bull. Seism. Soc. Am. 82, 1622–1645. in Penrose Conference 2000: Great Cascadia Earthquake Tricenten-
Humphrey, J. R., W. J. Silva, and R. R. Youngs (1993). Factors influencing nial, Program Summary and Abstracts, Oregon Department of Ge-
site-specific ground motion estimates for the 1985 M 8.1 Michoacan ology and Mineral Industries Special Paper 33, 141–143.
earthquake (abstract), Seism. Res. Lett. 64, 17. Wong, I. G., W. J. Silva, J. Bott, D. Wright, P. Thomas, N. Gregor, S. Li,
Hyndman, R. D., and K. Wang (1995). The rupture zone of Cascadia great M. Mabey, A. Sojourner, and Y. Wang (2000). Earthquake scenario
earthquakes from current deformation data and thermal regime, J. and probabilistic ground shaking maps for the Portland, Oregon met-
Geophys. Res. 100, 22,133–22,154. ropolitan area, Oregon Department of Geology and Mineral Industries
Ludwin, R. S., C. S. Weaver, and R. S. Crosson (1991). Seismicity of Interpretive Map Series IMS-16, scale 1:62,500, 11 sheets with 16
Washington and Oregon, in Neotectonics of North America: Geolog- pp. text.
ical Society of America Decade of North American Geology, D. B. Youngs, R. R., S. J. Chiou, W. J. Silva, and J. R. Humphrey (1997). Strong
Slemmons, E. R. Engdahl, M. Zoback, and D. Blackwell (Editors) ground motion attenuation relationships for subduction zone earth-
Decade Map, Vol. 1, 77–98, quakes, Seism. Res. Lett. 68, 58–73.
1932 N. J. Gregor, W. J. Silva, I. G. Wong, and R. R. Youngs

Youngs, R. R., S. M. Day, and J. P. Stevens (1988). Near-field motions on URS Corporation
rock for large subduction zone earthquakes, in Earthquake Engineer- 500 12th Street, Suite 200
ing and Soil Dynamics II: Recent Advances in Ground Motion Eval- Oakland, CA 94607
uation, American Society of Civil Engineering Geotechnical Special (I.G.W.)
Publication No. 20, ASCE, 445–462.
Geomatrix Consultants
2101 Webster Street
Pacific Engineering and Analysis Oakland, CA 94612
311 Pomona Avenue (R.R.Y.)
El Cerrito, CA 94530
(N.J.G., W.J.S.) Manuscript received 25 February 2002.