Long-Period Building Response to Earthquakes
in the San Francisco Bay Area
by Anna H. Olsen, Brad T. Aagaard, and Thomas H. Heaton
Abstract
This article reports a study of modeled, long-period building responses to
ground-motion simulations of earthquakes in the San Francisco Bay Area. The earth-
quakes include the 1989 magnitude 6.9 Loma Prieta earthquake, a magnitude 7.8 si-
mulation of the 1906 San Francisco earthquake, and two hypothetical magnitude 7.8
northern San Andreas fault earthquakes with hypocenters north and south of San Fran-
cisco. We use the simulated ground motions to excite nonlinear models of 20-story,
steel, welded moment-resisting frame (
MRF
) buildings. We consider
MRF
buildings
designed with two different strengths and modeled with either ductile or brittle welds.
Using peak interstory drift ratio (
IDR
) as a performance measure, the stiffer, higher
strength building models outperform the equivalent more flexible, lower strength de-
signs. The hypothetical magnitude 7.8 earthquake with hypocenter north of San Fran-
cisco produces the most severe ground motions. In this simulation, the responses of
the more flexible, lower strength building model with brittle welds exceed an
IDR
of
2.5% (that is, threaten life safety) on 54% of the urban area, compared to 4.6% of the
urban area for the stiffer, higher strength building with ductile welds. We also use the
simulated ground motions to predict the maximum isolator displacement of base-iso-
lated buildings with linear, single-degree-of-freedom (
SDOF
) models. For two existing
3-sec isolator systems near San Francisco, the design maximum displacement is
0.5 m, and our simulations predict isolator displacements for this type of system
in excess of 0.5 m in many urban areas. This article demonstrates that a large,
1906-like earthquake could cause significant damage to long-period buildings in
the San Francisco Bay Area.
Introduction
The northern San Andreas fault produced the devastat-
ing 1906 San Francisco earthquake. The same fault may pro-
duce a similar earthquake in the future, and the consequences
of a similar earthquake in a modern, urban area are uncertain.
The city of San Francisco and surrounding communities are
significantly different than they were 100 years ago. Speci-
fically, urban areas now include long-period buildings that
have been built only in the last several decades. To better
understand the possible performance of these long-period
buildings in San Francisco
’
s next great earthquake, we study
the response of two examples of long-period buildings: 20-
story, steel, welded moment-resisting frame (
MRF
) and base-
isolated buildings.
Experience provides few examples of steel-frame re-
sponses in great (magnitude
>
7
:
5
) earthquakes. Contempor-
ary and modern reconnaissance reports of the 1906 San
Francisco earthquake conclude that the steel frames existing
in 1906 performed well in the severe ground motions (Soulé,
1907; Tobriner, 2006). However, the steel frames in 1906
were markedly different than modern frames. The
tallest building in 1906 San Francisco was the 18-story
braced-frame Call Building, and Soulé (1907) notes only
nine steel-frame buildings between nine and 12 stories; mod-
ern steel-frame buildings are often much taller. Also, modern
structural engineers can choose longer-period,
MRF
s rather
than the shorter-period, braced frames, which were state-of-
the-art 100 yr ago. (See Hamburger and Nazir (2003) for a
brief discussion of historic and modern steel frames.) Be-
cause modern steel frames may be taller and have modern
designs, it is difficult to infer the performance of modern
steel frames based on the reported response of 1906 build-
ings. Whereas one great earthquake tested steel frames
designed a century ago, there are several examples of modern
steel-frame responses in smaller earthquakes (Hamburger
and Nazir 2003). In particular, tall steel-frame buildings
showed either repairable or no damage following both the
1994 magnitude 6.7 Northridge and 1995 magnitude 6.9
Kobe earthquakes. However, in the Northridge earthquake,
the largest recorded ground displacement near a building
was 0.31 m near the Olive View Hospital. Somerville
et al.
1047
Bulletin of the Seismological Society of America, Vol. 98, No. 2, pp. 1047
–
1065, April 2008, doi: 10.1785/0120060408
(1995) provide recorded and simulated ground motions at
seven building sites; none of these ground motions exceeds
0.3 m. Dynamic ground displacements beneath tall steel
frames were less than 0.5 m in the Kobe earthquake (Build-
ing Research Institute, 1996). Furthermore, investigations of
modern, steel
MRF
buildings after the Northridge earthquake
demonstrated that many welds in existing moment-resisting
joints are brittle (Gilani, 1997). The SAC Steel Project re-
ports document this widespread problem (e.g., Krawinkler,
2000; Roeder, 2000). In this study, we explore the response
of
MRF
buildings with either ductile or brittle welds to strong
ground motions with large displacements.
To predict the response of
MRF
buildings in large and
great earthquakes, many research engineers employ numeri-
cal models. Luco and Cornell (2000) studied the effect of
beam-to-column connection failure on the seismic response
of 3-, 9-, and 20-story building models developed by the
SAC Steel Project. The authors found that some of their lar-
gest ground motions induced interstory drifts in excess of
10% or caused collapse in the 20-story building models with
the brittle connections they studied. However, the authors did
not include internal gravity frames or shear connections in
this study, but they found in a separate study that including
these features typically reduced large drifts like these re-
sponses. Gupta and Krawinkler (2000) used the same
SAC building models with no deterioration mechanisms
to predict the seismic response at several seismic hazard le-
vels. In some of the ground motions that represent the hazard
level of 2% exceedance in 50 yr, the authors noted large drift
demands in the 20-story building model designed for the Los
Angeles area. They conclude that
“
the potential for unaccep-
table performance is not negligible.
”
Lee and Foutch (2006)
designed several alternatives to the SAC building models
with various strengths. From a nonlinear time history analy-
sis, the authors found that some 20-story models exceeded
their interstory drift capacities, including models with higher
strengths. Krishnan
et al.
(2006) studied the response of a
common type of building in the Los Angeles area to simu-
lated ground motions from hypothetical magnitude 7.9 rup-
tures on the southern San Andreas fault. They used a fully
three-dimensional building model to compare the responses
of 18-story steel
MRF
buildings designed to the 1982 and
1997 Uniform Building Codes (
UBCs
). In the simulations,
the
MRF
models showed large drifts, which would threaten
life safety in many areas of Los Angeles. The present study
augments previous work by applying thousands of simu-
lated, large ground motions to different 20-story building
models and by evaluating the building model responses
on a regional level near San Francisco.
Base-isloated buildings are a relatively new type of
long-period structure, characterized by a purposely built
flexible zone in the foundation that supports a superstructure.
By design, the natural frequencies of the superstructure are
high compared to the effective frequency of the isolation
system. Base isolation can significantly reduce high-
frequency vibrations of a building because the isolation level
does not transfer high-frequency motions to the super-
structure. The seismic forces in a building
’
s superstructure
are significantly smaller for a base-isolated building com-
pared to an identical building without isolators. However,
all isolation systems have a limited range of motion. Depend-
ing on the individual system, base-isolated buildings may
experience impacts between foundation walls and the super-
structure (Heaton
et al.
, 1995). Base-isolation systems per-
formed well during the 1994 Northridge and 1995 Kobe
earthquakes (Kelly, 2004) although no base-isolated build-
ings were in the near-source areas for these earthquakes.
Furthermore, these earthquakes produced much smaller
ground displacements than those produced in the 1906 mag-
nitude 7.8 San Francisco earthquake. One goal of this study
is to estimate isolator displacements that might occur in a
large San Andreas fault event like the 1906 San Francisco
earthquake. Detailed models of isolation systems would pro-
vide the best predictions of isolator behavior, but we do not
presently have such models. Instead, we use an equivalent-
linear approximation of the isolator system to estimate the
isolator displacements in our considered earthquakes.
The 2006 international building code (
IBC
) requires a
dynamic analysis to design base-isolation systems in the San
Francisco Bay Area. The designer must perform response
spectrum and response history analyses to determine the
design and maximum displacements of the isolators, among
other design parameters (International Code Council, 2006).
The code also specifies that these displacements must not fall
below minimum values. Thus, the model response to ground
motions may control the design, and the choice of motions
may affect the design. If a design engineer uses ground mo-
tions larger than those required by the building code, then he
or she will call for a system with a larger isolator displace-
ment capacity than an engineer who uses smaller motions.
There has been active discussion on the use of near-source
ground motions
—
characterized by a large displacement
pulse
—
for design (e.g., Hall, 1999; Kelly, 1999). Jangid
and Kelly (2001) stated that base-isolation systems should
be designed primarily to minimize damage to contents (mea-
sured by superstructure acceleration) in moderate earth-
quakes and secondarily to minimize isolator displacement
in large pulse-type ground motions. The authors showed
the existence of an optimum isolator damping that minimizes
superstructure accelerations. This optimum damping does
not minimize isolator displacement because isolator dis-
placement monotonically decreases with increasing damp-
ing. Ryan and Chopra (2004b) compared the results of a
nonlinear analysis of base-isolation systems to the equiva-
lent-linear procedure of the 2000
IBC
used to determine
the design displacement. The authors found that the isolator
displacements from the equivalent-linear procedure under-
estimated those from the nonlinear analysis by 20%
–
50%
on average. Thus, the minimum design displacements re-
quired by the code were not conservative for the strong
ground motions, consistent with moderate earthquakes, used
in their study.
1048
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
Since the ground motions in future earthquakes are un-
certain, structural engineers cannot design a building for the
specific ground motions that it will experience in its lifetime.
Instead, structural engineers rely on building codes to define
the types of ground motions that buildings must safely
survive. The design response spectrum represents ground
motions from events that are deemed likely to excite build-
ings in their assumed lifetime. From the design spectrum,
structural engineers determine the minimum design forces
that buildings must withstand. The building code also ac-
knowledges that unusually large earthquakes occur in or near
urban areas. The 2006
IBC
describes the maximum consid-
ered earthquake (
MCE
)as
“
the most severe earthquake
effects considered in this code
”
and defines the
MCE
as
1.5 times the design response spectrum (International Code
Council, 2006). In an earthquake consistent with the
MCE
,
structural engineers acknowledge that buildings will sustain
significant damage to structural systems and building con-
tents. However, the buildings may collapse
—
either partially
or totally
—
only in ground motions that exceed the
MCE
(Hamburger, 2003). The design of
MRFs
like the ones in this
study is consistent with this philosophy. The design of base-
isolated buildings in the San Francisco Bay Area requires the
use of ground motions consistent with the
MCE
, not the stan-
dard design spectrum (Structural Engneering Institute [
SEI
],
2006). In this way, structural engineers design most build-
ings for likely earthquakes while acknowledging that a large,
unusual event will test the limits of the lateral force resisting
systems.
The purpose of the study described in this article is to
predict the response of some long-period buildings to scenar-
io and hypothetical earthquakes in the San Francisco Bay
Area. We apply ground motions from simulations of the
1989 Loma Prieta and large, northern San Andreas fault
events (including the 1906 San Francisco earthquake) to steel
MRF
and base-isolated building models. We evaluate the
effect on overall building performance of rupture direction,
MRF
building strength and weld state, and base-isolation sys-
tem period and damping. We use response spectra to com-
pare the spectral accelerations predicted in the magnitude 7.8
earthquakes to the 1994
UBC
, 1997
UBC
, and 2006
IBC
de-
sign response spectra. This study continues the work of Hall
and Challa (1995), Heaton
et al.
(1995), and Hall (1998) by
considering long-period building response to recent simula-
tions of ground motions from great earthquakes on a large
geographical region.
Our analysis is limited in several ways. We only consid-
er a few realizations of possible future great earthquakes. The
next great northern San Andreas earthquake will almost
certainly be different than our scenario and hypothesized
events. Nonetheless, these events seem plausible because
they are compatible with the current understanding of the
1906 San Francisco earthquake. Our study is also limited be-
cause we only consider several idealized building models.
The buildings we consider are likely not the most vulnerable
buildings in our considered earthquakes. We choose these
buildings because their analysis is presently accessible to
us. Furthermore, they are examples of very flexible struc-
tures. Increasing a building
’
s flexibility can help to limit
the stress in a building subjected to high-frequency ground
motions, which may result from moderately sized earth-
quakes. However, extremely flexible structures may experi-
ence very large deformations in the large long-period ground
motions produced by unusual great earthquakes.
Description of Ground Motions
Aagaard, Brocher, Dolenc, Dreger, Graves, Harmsen,
Hartzell, Larsen, and Zoback (2008) and Aagaard, Brocher,
Dolenc, Dreger, Graves, Harmsen, Hartzell, Larsen,
McCandless,
et al.
(2008) simulated ground motions for
the 1989 Loma Prieta earthquake, the 1906 San Francisco
earthquake, and seven large hypothetical northern San
Andreas fault events. In this study, we use time histories from
four of these ground-motion simulations: the 1989 Loma
Prieta scenario earthquake with the Wald source model;
the 1906 San Francisco scenario earthquake (with an epi-
center west of San Francisco and 3 km offshore); and two
hypothetical earthquakes with the same slip distribution as
the 1906 San Francisco scenario, SongModHypoC (with
an epicenter near Bodega Bay, north of San Francisco)
and SongModHypoS (with an epicenter near San Juan
Bautista, south of San Francisco). We refer to the last two
simulations as the magnitude 7.8 Bodega Bay and San Juan
Bautista hypothetical earthquakes.
We use ground motions on an area 250 by 110 km that
includes the southern half of the magnitude 7.8 ruptures. We
select ground-velocity time histories at sites on several grids,
with major urban areas gridded at higher resolutions. This
sampling generates ground motions at 4945 sites. Figure 1
shows the simulation domain, regional geography, and urban
areas (as defined in the 2000 census). Because this study con-
siders plausible earthquakes on the northern San Andreas
fault, which may damage long-period buildings, we focus
our attention on currently urbanized locations. The total
urban area is
3266
km
2
, and this includes urban areas at dis-
tances 60 km from the fault, specifically the East Bay, Santa
Rosa, and Concord. Of course, only urban areas on the San
Francisco peninsula or near San Jose are close to the San
Andreas fault, and some sites in our urban areas are not suit-
able for the buildings we consider. When we report results,
we consider damage on the entire urban area, and on urban
areas in the cities of Oakland, San Francisco, San Jose, and
Santa Rosa.
The ground-motion simulations include the local three-
dimensional geologic structure via three-dimensional geo-
logic (Jachens
et al.
, 2006) and seismic velocity (Brocher
et al.
, 2006) models. The bandwidth of the simulations is
limited to periods of 2.0 sec and longer (frequencies of
0.5 Hz and lower). Aagaard limits the minimum shear-wave
speed to
700
m
=
sec, so the simulated ground motions do not
include amplification and nonlinear effects from soft near-
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1049
surface sediments. The thin sedimentary deposits and artifi-
cial fill around the immediate perimeter of the San Francisco
Bay are well recognized as an important issue when asses-
sing seismic hazard (e.g., Lawson, 1908; Hough
et al.
, 1990;
Dobry
et al.
, 2000). Because we do not include the ampli-
fication effect of these sediments (bay mud) or fill on the
ground motions, the building responses in these areas may
be considered a lower bound. To further complicate the in-
terpretation of soft-soil sites, the designs of our buildings
only meet or exceed the 1994
UBC
for zone four and stiff
soils. Buildings designed for very soft soils, such as bay
mud, may be significantly different than the designs used
in our study.
Figures 2 and 3 show the peak ground displacements
and velocities, respectively, for the earthquake simulations.
Aagaard, Brocher, Dolenc, Dreger, Graves, Harmsen, Hart-
zell, Larsen, McCandless,
et al.
(2008) noted that they
believe errors in the seismic velocity model lead to over-
estimation of the ground motions in the La Honda basin
in the Santa Cruz mountains, southwest of San Jose. (The
La Honda basin is not included in our urban area. See Fig. 1.)
Compared to observations in the Loma Prieta earthquake, the
overestimation in the ground motions was 1
–
2 modified
Mercalli intensity units.
Twenty-Story, Steel MRF Buildings
Description of Building Models
A 20-story, steel, welded,
MRF
building is one example
of a long-period building. This building is susceptible to
long-period ground motions (Heaton
et al.
, 1995), but it
is not representative of every tall building. We choose to
study a 20-story building, in part, because steel
MRF
build-
ings shorter and taller than approximately 20 stories tend to
be stronger than this building. Taller buildings use different
lateral-force-resisting systems to withstand larger wind
loads, and so taller buildings require specialized design
and analysis. Shorter buildings tend to have a larger base
shear at yield, making them stronger than 20-story buildings.
Also, the fundamental frequency of shorter buildings is out-
side the frequency content of the simulated ground motions
in this study. The design natural frequency of a 20-story
building (approximately 0.4
–
0.5 Hz) is in the frequency
range of the ground motions. The 20-story steel
MRF
build-
ing models used in this study are regular and rectangular.
Figure 4 shows the dimensions of the models.
To study the effect of building strength and stiffness on
seismic response, we employ models with two different de-
signs. Hall (1997) designed the buildings according to the
1992 Japanese Building Code (
JBC
) and the 1994
UBC
for seismic zone four and stiff soils. There are no soft stories
in the
MRF
models. Both designs have moment-resisting
joints on the perimeter of the floor plan and simple connec-
tions at interior joints, with one exception: on the center line
of the
JBC
design, there are moment-resisting joints at the
interior beam-column connections. Also, some beams and
columns in the
JBC
design are larger than their counterparts
in the
UBC
design, and some columns in the
JBC
design bend
about their weak axis. Because of the strength and drift re-
quirements of each building code, the
JBC
designs are stron-
ger and stiffer than the
UBC
design for the same elevation
and plan dimensions. We do not compare the building codes
themselves nor suggest that existing buildings in the San
Francisco Bay Area were designed according to the 1992
JBC
provisions.
Hall (1998) compares the steel
MRF
building models
’
behavior to the 1997
UBC
static lateral-force requirements.
While the building designed to the 1994
UBC
is conservative,
it does not satisfy the 1997
UBC
. The building designed to
the
JBC
satisfies the 1997
UBC
static lateral-force require-
ments for velocity-based near-source factors,
N
v
, less than
or equal to 1.2. This near-source factor is less than or equal
to 1.2 for sites greater than 10 km from any fault and for sites
5 km from a fault with maximum moment magnitude less
than 7.0 and slip rate less than
5
:
0
mm
=
yr. While the models
designed to the 1992
JBC
satisfy the lateral force provisions
Figure
1.
The San Francisco Bay Area has a large urban area
near the San Andreas fault (heavy black line). Light orange shading
highlights the urban areas that include 97% of the region
’
s popula-
tion. (In subsequent maps, a solid black line bounds the urban
areas.) Red dots indicate the sites with simulated ground motion;
some sites are located under water, but we do not include the results
from these sites in this article. Blue triangles locate sites for re-
sponse spectra presented in Figure 12.
1050
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
of the 1997
UBC
at some sites in the San Francisco Bay Area,
the models with brittle welds do not comply with the 1997
UBC
. Again, the models designed to the 1992
JBC
and 1994
UBC
represent buildings with different combinations of
strength and stiffness, which could have ductile or brit-
tle welds.
Hall (1997) developed a finite-element algorithm speci-
fically for steel-frame building models, and we use this algo-
rithm to calculate
MRF
-model responses. A building model
consists of planar frames coupled at the floors by an approxi-
mately rigid floor diaphragm. Several fibers constitute the
cross section of each beam or column element, and each fiber
behaves according to a nonlinear hysteretic steel model. This
method explicitly models panel-zone behavior with a non-
linear hysteretic moment-shear strain model. The computa-
tional algorithm includes coordinate updating to account for
P
-
Δ
effects. (Under large lateral deformations, the columns
are no longer approximately vertical. Thus, the columns
carry vertical loads
P
at a distance
Δ
from the vertical axis.
P
-
Δ
effects must be included in an analysis of frames with
large deformations because the effect induces an additional
moment carried by the columns.) The models do not capture
out-of-plane or torsional motions. The building models have
uniform mass and stiffness distributions in each story, and we
assume uniform ground motion at the base of the buildings.
Therefore, we do not expect a torsional component in the
Figure
2.
The 1989 Loma Prieta scenario earthquake generates moderate ground motions. The largest peak ground displacement is
0.67 m. The peak ground motions in the magnitude 7.8 scenario and hypothetical earthquakes are significantly larger in amplitude and
geographic extent. The largest peak ground displacements are 4.24 m (
M
7.8 Bodega Bay), 4.26 m (1906 San Francisco), and 4.29 m
(
M
7.8 San Juan Bautista). The black stars locate the epicenters.
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1051
building response. A planar-frame model is adequate for
these simulations because it captures the important behaviors
of these regular building models. In order to estimate the lar-
gest model responses for each ground motion, we apply the
ground motion in the weakest orientation of the building: we
orient the weak bending axis of the building perpendicular to
the direction of the largest peak-to-peak ground velocity. We
resolve the orthogonal horizontal components into this direc-
tion, and we apply the resultant horizontal and vertical time
histories to the planar model.
The only deterioration mechanism in the
MRF
models is
weld fracture. At each moment-resisting joint, a stochastic
procedure assigns a fracture strain to the weld fibers, and
the assignment is the same for all simulation sites. If the
strain in the weld fiber exceeds the fracture strain, the fiber
no longer carries tensile loads for the remainder of the simu-
lation. The
MRF
models may have ductile welds (
P
), which
do not fracture, or brittle welds (
B
), which fracture according
to the following constraints. For the bottom beam flange at a
moment-resisting joint, 20% of weld fibers fracture when the
ratio of strain to yield strain,
ε
=
ε
yield
, is 0.7; 40% fracture
when
ε
=
ε
yield
1
; 20% fracture when
ε
=
ε
yield
10
; 10%
fracture when
ε
=
ε
yield
50
; and 10% fracture when
ε
=
ε
yield
100
. For the top beam flanges, column splices,
and column base-plate welds, 40% of weld fibers fracture
when
ε
=
ε
yield
1
; 30% fracture when
ε
=
ε
yield
10
; and
30% fracture when
ε
=
ε
yield
100
. Hall (1997) chose these
ratios and distributions to approximate observed structural
damage after the 1994 Northridge earthquake. Following
Hall, we abbreviate the building code, height in stories,
Figure
3.
The three magnitude 7.8 earthquake simulations generate significantly larger peak ground velocities than those experienced in
the 1989 Loma Prieta earthquake. The largest peak ground velocities are
0
:
82
m
=
sec (1989 Loma Prieta),
3
:
03
m
=
sec (
M
7.8 Bodega Bay),
3
:
85
m
=
sec (1906 San Francisco), and
3
:
85
m
=
sec (
M
7.8 San Juan Bautista). The black stars locate the epicenters.
1052
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
and weld state of the
MRF
models. For example,
J20B
de-
notes a 20-story, steel
MRF
building model designed to
the
JBC
with brittle welds. The four building types we con-
sider in this article are
J20P
,
J20B
,
U20P
, and
U20B
.
The period and base shear at yield are two commonly
reported building parameters. The elastic first- and second-
mode periods with damping are 3.4 and 0.90 sec, respec-
tively, for the
J20P
model and are 4.1 and 1.1 sec, respec-
tively, for the
U20P
model. Note that the ground motions
’
bandwidth includes the first-mode period but not the sec-
ond-mode period. Engineers often use a pushover analysis
to quantify characteristics of the building models. Important
characteristics include the base shear when the building
yields; the ultimate base shear; and the maximum amount
of displacement supported by the frame as a fraction of
the displacement at yield, or ductility. A pushover analysis
applies an increasing, lateral force to the model until it loses
all lateral-load-carrying capacity and the building collapses
due to
P
-
Δ
instability. The simulation generates a pushover
curve that relates the shear in the first floor columns to the
lateral roof displacement. Figure 5 shows the pushover
curves for the four models. The base shear at yield (as a frac-
tion of the design weight) is approximately 0.12 for the
J20P
model and approximately 0.09 for the
U20P
model. The
JBC
building models are stronger than the
UBC
models, whereas
the
UBC
models have a longer fundamental period.
Response of Building Models
We measure building response with the peak interstory
drift ratio (
IDR
). The
IDR
quantifies the relative displacement
of the two floors forming a story, as a ratio of the story
height. The
IDR
indicates the amount of shear deformation,
averaged over the story, in any story for the duration of the
simulation.
The computational algorithm provides no direct mea-
surement of partial or total collapse of the model. As a proxy
measure, we use
IDR
. As the
IDR
increases, we assume that
the building model
’
s ability to remain standing decreases.
Some ground motions induce forces in the building models
that exceed their ability to resist lateral loads for the remain-
der of the simulation. The simulated responses show that no
MRF
building model retains any lateral-load-carrying capa-
city after registering an
IDR
of 0.15. Thus, we deem the si-
mulation result a simulated collapse when the
IDR
exceeds
0.15. Such large responses violate the small strains and rota-
tions assumptions of the finite elements, so the simulations
do not give valid results for larger deformations.
Figures 6
–
9 show
MRF
model responses to the four sce-
nario earthquakes. Each of the four panels in a figure shows
the response of one building type at all locations in the
simulation domain. Also, ground motions in areas with soft
soils do not include site amplification. The responses of long-
period buildings in those areas may be a lower bound. The
colors on the maps are consistent with the structural
performance levels for steel moment frames defined by
the Federal Emergency Management Agency report 356
(
FEMA
356; American Society of Civil Engineers [
ASCE
],
2000). Areas colored blue indicate elastic or mild inelastic
response (
0
:
007
<
IDR
<
0
:
025
, immediate occupancy).
Areas colored green to yellow indicate moderate to severe
inelastic deformation; a building located in these areas would
Figure
4.
We use four 20-story steel
MRF
building models in
our analysis. The
MRF
models share the same floor plan and dimen-
sions. There are no soft stories in the design. The third story is most
likely to show simulated collapse.
0
50
100
150
200
250
300
0
0.05
0.1
0.15
0.2
0.25
Lateral Roof Displacement [cm]
Base Shear [Fraction of Seismic Design Weight]
J20B
J20P
U20B
U20P
Figure
5.
The pushover curves show the characteristic pushover
responses of the steel
MRF
building models. The
JBC
buildings are
stronger than the equivalent
UBC
buildings, as measured by the base
shear at yield. The buildings with ductile welds have approximately
the same ductility, or ratio between the lateral roof displacements at
failure (approximately 240 cm) to yield (approximately 60 cm).
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1053
need structural repairs for reoccupation following the earth-
quake (
0
:
025
<
IDR
<
0
:
05
, life safety). Areas colored or-
ange to red indicate the potential for partial or total
collapse (
IDR
>
0
:
05
, collapse prevention); a building in
these areas may be a total loss. Areas colored pink indicate
that the building models lost all lateral-load-carrying capa-
city in the simulation (simulated collapse).
Tables 1 and 2 record the
MRF
building model re-
sponses. Table 1 shows the percentage of areas within the
San Francisco Bay region on which the model responses ex-
ceed the
FEMA
356 life safety level (
IDR
>
0
:
025
). Table 2
shows the percentage of areas within the bay region on which
the building models reach simulated collapse (
IDR
>
0
:
15
).
In both tables, we report only model responses within urba-
Figure
6.
The 1989 Loma Prieta scenario earthquake induces small interstory drifts in the
MRF
building models. The largest building
responses would have been observed in the urban area of the Santa Clara Valley, near San Jose. (See text for physical interpretation of map
coloring.)
1054
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
nized areas. Also, we report the areas of percent exceedance
on all urban areas as well as the urban areas of the Oakland,
San Francisco, San Jose, and Santa Rosa subdomains
mapped in Figure 1.
The building responses to the 1989 Loma Prieta scenario
earthquake are small. On most of the San Francisco Bay
Area, the 20-story, steel
MRF
buildings remain elastic in the
simulation. For example, the response of the more flexible,
lower strength building with brittle welds exceeds the life
safety level on
13
km
2
of the entire urban area. Table 1
shows that, of the four major subdomains we consider, only
San Jose has building responses that exceed life safety, and
the areas of exceedance are small. Table 2 shows no simu-
lated collapses of models in the urban areas. The Loma Prieta
Figure
7.
In the magnitude 7.8 Bodega Bay hypothetical earthquake, the rupture propagates north to south into the San Francisco Bay
Area. Urban areas south of San Francisco and Oakland show large steel
MRF
building model responses. As in all scenarios, the stiffer, higher
strength design tends to have responses smaller than those of the more flexible, lower strength design, and buildings with ductile welds
outperform those with brittle welds. (See text for physical interpretation of map coloring.)
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1055
scenario is significantly less damaging in all parts of the
San Francisco Bay Area than any of the three magnitude
7.8 simulations.
The stiffer, higher strength model with ductile welds
performs better than the other three
MRF
models. The geo-
graphic extents of models with inelastic deformations and
simulated collapses are both smaller for the stiffer, higher
strength buildings than for the more flexible, lower strength
buildings. For the magnitude 7.8 Bodega Bay hypothetical
earthquake, the
MRF
building models with ductile welds ex-
ceed the life safety level on 10.% (
U20P
) and 4.6% (
J20P
)of
the urban area; the same models with brittle welds show the
same level of response on 54% (
U20B
) and 36% (
J20B
) of the
urban area. The responses of the stiffer, higher strength mod-
Figure
8.
This simulation predicts
MRF
building responses in the 1906 San Francisco scenario earthquake. The hypocenter is located
offshore and west of San Francisco, and the fault ruptures bilaterally. Most energy travels away from the city of San Francisco. Of the three
magnitude 7.8 scenario and hypothetical earthquakes, this scenario tends to show the smallest responses in San Francisco. (See text for
physical interpretation of map coloring.)
1056
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
els exceed the life safety level on smaller areas than the re-
sponses of the more flexible, lower strength models. Not sur-
prisingly,
MRF
models with ductile welds perform better than
models with brittle welds.
While all of the magnitude 7.8 earthquakes produce
large long-period ground motions, the results from the mag-
nitude 7.8 Bodega Bay and San Juan Bautista hypothetical
earthquakes are different than those of the 1906 San Francis-
co scenario. The Bodega Bay hypothetical earthquake (with
hypocenter north of San Francisco and unilateral rupture to
the south) is generally the most severe of the three magnitude
7.8 earthquakes. For the entire urban area and the Oakland
and San Jose subdomains, the Bodega Bay hypothetical
earthquake causes the largest percent areas of life safety ex-
ceedance for all four
MRF
models. This earthquake also pro-
duces the largest percent areas of simulated collapse for all
Figure
9.
Considering the urban area in the magnitude 7.8 San Juan Bautista hypothetical earthquake, the largest building responses
occur in San Francisco. Urban areas outside San Francisco tend to have the smallest responses of all three magnitude 7.8 earthquakes. This
scenario produces the fewest simulated collapses in the urban areas. (See text for physical interpretation of map coloring.)
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1057
Table
1
Percent of Urban Areas on which Building Response Threatens Life Safety
Domain
Simulation
J20B
J20P
U20B
U20P
Entire urban outline
1989 Loma Prieta
0.079
0
0.40
0
M
7.8 Bodega Bay
36
4.6
54
10.
1906 San Francisco
9.3
0.99
15
3.1
M
7.8 San Juan Bautista
11
0.95
17
2.4
Oakland subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
18
1.6
48
0.20
1906 San Francisco
0
0
0
0
M
7.8 San Juan Bautista
0
0
0
0
San Francisco subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
48
18
69
35
1906 San Francisco
39
12
49
18
M
7.8 San Juan Bautista
63
6.8
92
13
San Jose subdomain
1989 Loma Prieta
0.20
0
1.7
0
M
7.8 Bodega Bay
64
11
83
26
1906 San Francisco
14
0.018
18
3.8
M
7.8 San Juan Bautista
11
0.51
14
1.6
Santa Rosa subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
18
0
38
3.1
1906 San Francisco
25
0
25
8.6
M
7.8 San Juan Bautista
38
5.7
27
11
FEMA 356 defines a level of life safety for steel MRF buildings at an interstory drift ratio of 0.025. At this level, a building would
require structural repairs, but partial or total collapse is unlikely. Depending on the earthquake and building type, the simulated
building responses may exceed the life safety level on a limited or broad portion of the urban area. The Bodega Bay
hypothetical earthquake especially causes damage in large parts of the San Francisco Bay urban area. The entire urban area in
the simulation domain is
3266
km
2
.
Table
2
Percent of Urban Areas on which Building Models Collapse
Domain
Simulation
J20B
J20P
U20B
U20P
Entire urban outline
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
1.7
0.049
6.6
0.24
1906 San Francisco
0.29
0
0.67
0.00092
M
7.8 San Juan Bautista
0.092
0
0.42
0
Oakland subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
0
0
0
0
1906 San Francisco
0
0
0
0
M
7.8 San Juan Bautista
0
0
0
0
San Francisco subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
8.8
0.89
24
4.1
1906 San Francisco
0.11
0
1.7
0
M
7.8 San Juan Bautista
1.3
0
1.6
0
San Jose subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
3.2
0
15
0.033
1906 San Francisco
0.031
0
0.97
0
M
7.8 San Juan Bautista
0
0
0.59
0
Santa Rosa subdomain
1989 Loma Prieta
0
0
0
0
M
7.8 Bodega Bay
0
0
4.0
0
1906 San Francisco
0
0
0
0
M
7.8 San Juan Bautista
0.069
0
0.069
0
Some ground motions cause exceedance of the lateral force resisting ability of the MRF models (simulated collapse). This level is
much greater than the life safety performance level defined by FEMA 356. MRF models with brittle welds show simulated collapses
on a greater area than models with ductile welds. The response of the stiffer, higher strength models exceed the life safety level on
smaller areas than that of the more flexible, lower strength models. The entire urban area in the simulation domain is
3266
km
2
.
1058
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
MRF
building models and all subregions of the San Francisco
Bay Area. The epicenter of the 1906 scenario earthquake is
due west of San Francisco, and the fault ruptures bilaterally.
In this simulation, there are no significant directivity pulses
in the city of San Francisco. Compared to the other magni-
tude 7.8 simulations, the 1906 scenario generates some of the
smallest percent areas of life safety exceedance in San Fran-
cisco. The San Juan Bautista hypothetical earthquake (with
hypocenter south of San Francisco and unilateral rupture to
the north) produces some of the smallest percent areas of life
safety exceedance of the three magnitude 7.8 simulations.
One important exception to this trend is the responses for
the models with brittle welds in the San Francisco sub-
domain: for the
J20B
model, 63% of the San Francisco area
exceeds the life safety level, and for the
U20B
model, 92% of
the San Francisco area exceeds the life safety level. Another
exception to the smaller responses in the San Juan Bautista
hypothetical earthquake is in the Santa Rosa subdomain: this
earthquake generates the largest percent area of life safety
exceedance in Santa Rosa for all building models except
U20B
. Of the magnitude 7.8 simulations, the Bodega Bay
hypothetical earthquake causes the most damage to 20-story,
steel
MRF
buildings in the modern urban area, whereas the
San Juan Bautista hypothetical earthquake tends to cause the
least damage.
Base-Isolated Buildings
In the past several decades, engineers have accepted the
method of inserting a thin, flexible layer between a building
’
s
superstructure and foundation as a good means to reduce
seismic risk in historic and modern buildings. During an
earthquake, much of the isolated building
’
s deformation con-
centrates in the isolation layer, and the isolated superstruc-
ture experiences smaller deformations than an unisolated
building. However, near-source ground motions from earth-
quakes with large slips may cause isolator displacements that
exceed the isolator capacity (Heaton
et al.
, 1995). In this
study, we estimate the differential displacement between
the building superstructure and its foundation.
A well-isolated base-isolated building can be modeled
adequately by a single-degree-of-freedom (
SDOF
) system.
For a nonisolated building, a modal analysis accurately pre-
dicts linear deformations as a superposition of the individual
modal responses. Adding base isolators to a building intro-
duces three new modes (two translational and one rotational)
to the structural analysis. If the isolators are sufficiently flex-
ible, then these new modes have significantly longer periods
than the fundamental period of the superstructure alone. In
this case, Chopra (2001) shows that the fundamental mode of
the isolated system can be modeled by an
SDOF
oscillator.
The mass of the oscillator is that of the superstructure, and
the restoring force and damping of the oscillator models the
behavior of the isolators. In this way, an
SDOF
system esti-
mates the response of a base-isolated system.
The behavior of base isolators in large earthquakes is
nonlinear. The two most common types of isolators are rub-
ber isolators, which are typically short, cylindrical pads with
alternating layers of rubber and steel plates; and friction-
pendulum isolators, which typically consist of a small steel
bearing that slides on a concave, hemispherical surface. An
accurate and precise model of isolator behavior must include
nonlinear dependencies on isolator deformation (for exam-
ple, displacement) and/or deformation velocity. In rubber
isolators, either fillers in the rubber or a small lead cylinder,
which yields with ductility at significant strains, provides
large amounts of viscous damping. For small strains, the rub-
ber in laminated bearings strain softens, but at very large
strains, the rubber strain hardens. Whereas high-hysteresis
rubber bearings have approximately linear viscous damping,
lead-core laminated bearings undergo plastic deformation
similar to that from dry friction damping. In contrast, the dy-
namic friction of sliding surfaces provides the damping in
friction-pendulum isolators. These isolators induce a build-
ing motion like that of a gravitational pendulum. Hence, the
restoring force is approximately linear for the range of mo-
tions encountered in building applications. The bearing fric-
tion, however, is nonlinear, dry friction: the isolator is very
stiff until it overcomes the static friction, and then friction is
a constant force in the opposite direction of sliding veloc-
ity. The damping in isolation systems is nonlinear and sig-
nificantly larger than damping in traditional steel-frame
buildings.
Description of Building Models
Despite the nonlinear behavior of the isolators, a base-
isolated system can be modeled as an equivalent-linear
system with the equivalent-linear period and damping depen-
dent on the amplitude of isolator displacement. The funda-
mental period of an equivalent-linear system is usually
T
2
π
m=k
e
p
, where
m
is the superstructure mass, and
k
e
is the effective stiffness defined as the ratio of the bearing
restoring force to the displacement, measured at the
maximum expected displacement. Ryan and Chopra
(2004b) studied the difference between maximum isolator
displacements predicted by nonlinear and equivalent-linear
SDOF
models for 20 near-source ground motions recorded
from four earthquakes. To approximate the characteristics
of lead-rubber isolators, they used a bilinear strain-softening
model. They found that the median ratio of maximum iso-
lator displacement from the bilinear model to that of the
equivalent-linear model monotonically increased from 1.3
at 2 sec to 1.5 at 4 sec. Also, the equivalent-linear displace-
ments were log-normally distributed about the median values
with a standard deviation of about 46% of the nominal value.
In a related study, Ryan and Chopra (2004a) investigated the
difference between bilinear models of friction-pendulum iso-
lators and equivalent-linear
SDOF
models. Here, they found
that the median ratio of maximum isolator displacement from
the bilinear model to that of the equivalent-linear model
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1059
monotonically increased from 1.1 at 2 sec to 1.4 at 4 sec. The
standard deviation in this case was about 50% of the nominal
value. Based on these results, equivalent-linear models of
base-isolation systems tend to underestimate the maximum
isolator displacements predicted from more complete, non-
linear models of the same buildings.
In order to estimate the isolator displacements during
our scenario and hypothetical earthquakes, we compute
the maximum displacement of an equivalent-linear
SDOF
system. We assume that the fundamental period of the
base-isolation system is a few times larger than the funda-
mental period of the superstructure. We further assume that
the isolator behavior can be approximated by linear springs
and viscous damping. Under these assumptions, the maxi-
mum isolator displacement is the displacement response
spectral value at the appropriate period and damping. We
simulate systems with periods of 2 and 3 sec with isolator
damping at 10% and 20% of critical.
Base-isolation systems with 3-sec equivalent-linear
periods and 10% damping seem to be common for existing
systems in the San Francisco Bay Area. For example, the 18-
story Oakland City Hall was retrofit with lead-rubber bear-
ings, and its fundamental period is 2.85 sec (Walters, 2003).
If the isolator displacement exceeds 0.51 m, the superstruc-
ture would impact the foundation (Kelly, 1998). The interna-
tional terminal of the San Francisco Airport was constructed
with friction-pendulum isolators. This system
’
s fundamental
period is 3 sec, and the maximum displacement of each bear-
ing is 0.51 m (
EPS
, 2003).
Response of Building Models
Figures 10 and 11 show the isolator displacements of
our 2- and 3-sec linear base-isolation building models for
the magnitude 7.8 Bodega Bay and San Juan Bautista hy-
pothetical earthquakes. For each period, the model has vis-
cous damping at 10% or 20% of critical. Although equivalent
damping values are typically less than 10%, we include 20%
damping to provide insight into the effect of including sup-
plemental damping devices. The map colorings indicate the
maximum isolator displacement. (Note that we choose dif-
ferent color scales for the 2- and 3-sec isolators: longer-
period systems typically have larger design maximum displa-
cements.) Based on the work of Ryan and Chopra (2004a,b)
we expect that the maximum displacements of nonlinear iso-
lators are 10%
–
30% larger than the maximum displacements
of our linear isolators.
The equivalent-linear base-isolation simulations predict
large isolator displacements in the hypothetical earthquakes.
The simulated ground motions are particularly rich in long
periods; the 3-sec model shows larger isolator displacements
than the 2 sec, and the larger responses span a greater area.
As expected, the 20% damped models show smaller isolator
displacements than the 10% damped. East of the San Fran-
cisco Bay, the Bodega Bay earthquake induces isolator dis-
placements two to five times larger than those from the San
Juan Bautista earthquake, whereas north of the Bay, the San
Juan Bautista earthquake generates isolator displacements up
to twice those from the Bodega Bay earthquake.
Using spectral displacement to model base-isolated
building response has the advantage that it is a simple
analysis. A more advanced analysis requires an appropriate
nonlinear model of isolator behavior under large lateral de-
formation and a multi-degree-of-freedom model of the super-
structure. The spectral displacement results are expected to
provide a conservative estimate of isolator displacements; we
expect greater isolator displacements in the areas of large
ground motions. Nonetheless, as recorded in Table 3, the
spectral displacement analysis shows large displacements
in the urban areas. The large ground motions used in this
study should be applied to more sophisticated base-isolation
models to quantify how much the spectral displacements
underestimate more sophisticated predictions of isolator
displacements.
Response Spectra
Ground motions like those from the three magnitude 7.8
simulations are severe and represent part of the seismic
hazard in the San Francisco Bay Area. Because the San
Andreas fault produced the 1906 San Francisco earthquake,
it is capable of producing an earthquake at least as large as
magnitude 7.9. The two hypothetical earthquakes included in
this article are variations on the 1906 scenario. Thus, we be-
lieve these hypothetical earthquakes are reasonable simula-
tions to consider for the San Francisco Bay Area. The next
great earthquake clearly will be none of these three, but given
the uncertainty of the future, we believe simulations like
these should be considered in the seismic hazard of the
San Francisco Bay Area.
The large number of sites in this study precludes dis-
playing the response spectrum for each site. Figure 12 shows
standard response spectra for selected sites in the urban area.
We also present nonstandard response spectra to capture the
character of the spectra at all sites in the urban area of our
simulation domain. We calculate the pseudospectral accel-
eration (
PSA
) for 2
–
5-sec periods at all sites. Then, at each
period, we select the
PSA
values that are exceeded on 1%,
3%, 10%, and 25% of the urban area. For example, the
PSA
at 2
–
4 sec exceeds 0.5
g
on 3% of the urban area in
the Bodega Bay hypothetical earthquake. In this case, 3%
of the urban area experiences a
PSA
greater than 0.5
g
. Fig-
ure 13 shows our response spectra for the three magnitude
7.8 earthquakes, the 1994
UBC
design spectrum, and the
upper and lower bounds on the 1997
UBC
and 2006
IBC
de-
sign spectra for sites in the Bay Area. The response spectra
have damping at 5% of critical, and the site class for the de-
sign spectra is D, or stiff soil.
The design response spectra defined by the 2006
IBC
are
similar to our spectra of the three magnitude 7.8 earthquakes.
The design spectrum for a particular site depends on the
spectral acceleration at long periods assigned to that site
1060
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
by the building code. In this article, we consider the upper
bound (sites on the San Andreas fault) and lower bound (sites
with minimum spectral acceleration values in the area); the
design spectrum at a specific site would lie between these
bounds. The peak spectral accelerations at almost all periods
from the three magnitude 7.8 earthquake ground motions ex-
ceed the upper bound for the 2006
IBC
design spectrum. In
the three earthquakes, the spectral accelerations exceed the
2006
IBC
upper bound on approximately 1% or less of
the urban area. In the Bodega Bay hypothetical earthquake,
the spectral accelerations exceed the 2006
IBC
lower bound
on approximately 25% of the urban area, and in the other two
magnitude 7.8 earthquakes, the ground motions exceed the
lower bound on approximately 10% of the urban area.
Our response spectra from these three simulations
should not be compared too closely with the design spectra.
Figure
10.
For linear isolators with a fundamental period of 2 sec, the predicted maximum isolator displacements are large on most of the
San Francisco Bay Area in the two hypothetical earthquakes. Existing base-isolation systems may not be designed for isolator displacements
this large. Also, we expect peak isolator displacements from nonlinear models to be larger than those predicted by these linear models. (Note
coloring scheme is different than Fig. 11.)
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1061
Our response spectra for the three magnitude 7.8 earthquakes
are similar, with the Bodega Bay hypothetical earthquake
producing higher spectral accelerations. The three earth-
quakes share the same assumed slip distribution, but the hy-
pocenter and rupture direction are different for the
simulations. A different slip distribution could give much lar-
ger or smaller ground motions within the urban area. Further-
more, many Bay Area sites are on soils with shear-wave
velocities of several hundred meters per second, and some
important sites are on very soft soils with shear-wave velo-
cities less than a hundred meters per second. At the resonant
periods of these sites, the ground motions may be amplified
Figure
11.
The magnitude 7.8 Bodega Bay and San Juan Bautista hypothetical earthquakes induce large isolator displacements in linear,
3-sec isolators, and the 3-sec isolators give large displacements on a greater area than the 2-sec isolators. We expect peak isolator displace-
ments from nonlinear models to be even larger (see text). Large damping reduces the peak isolator displacements. (Note coloring scheme is
different than Fig. 10. The white contour is a peak isolator displacement of 1 m, compared to the design maximum displacement of 0.5 m for
two 3-sec systems in the San Francisco Bay Area.)
1062
A. H. Olsen, B. T. Aagaard, and T. H. Heaton
five times that of sites with seismic velocities greater than our
minimum permitted value of
700
m
=
sec.
Conclusions
The long-period buildings we study show simulated re-
sponses that approach and reach failure on a large area in
magnitude 7.8 earthquake simulations. The steel
MRF
mod-
els
’
responses in the 1989 Loma Prieta scenario are consid-
erably smaller than those of the large magnitude simulations.
A long-period building that withstood the ground motions in
the 1989 Loma Prieta earthquake may experience significant
damage in an event like the 1906 San Francisco earthquake.
For all steel
MRF
building models, the building responses in
the 1989 Loma Prieta scenario exceed the life safety level on
no more than 0.4% of the urban area, compared to 1%
–
54%
for the magnitude 7.8 simulations. Additionally, the variabil-
ity in the long-period building models
’
responses at a single
site to the three magnitude 7.8 simulations implies a large
range of possible building responses, depending on the rup-
ture direction.
The design of the steel
MRF
building affects the like-
lihood of large response. For the same ground motion, the
stiffer, higher strength design tends to have a smaller
IDR
than the more flexible, lower strength design. Changing
the design from the weaker to the stronger decreases the area
of simulated collapse by a factor of 3
–
6. Our modeling sug-
gests that buildings with brittle welds are vulnerable to long-
period ground motions from large earthquakes on the San
Andreas fault. Fixing brittle welds decreases the area of
imulated collapse by a factor of 20
–
30 or increases the like-
lihood of no simulated collapses.
Our assumptions influence the results in the following
ways. We assume the most damaging building orientation
by choosing the resultant of the horizontal ground motions
that maximizes peak-to-peak velocity. An existing building
may not have this damaging orientation, and thus the re-
sponse would be smaller. Since the ground-motion simula-
tions limit the minimum shear-wave speed to
700
m
=
sec,
the ground motions do not include the amplification asso-
ciated with soft near-surface sediments and artificial fill.
Consequently, the predictions of building response for these
models in areas of soft sediments or poorly compacted fill
may be considered lower bounds. We cannot comment on
buildings designed specifically for soft-soil sites. Consider-
ing these opposing influences, we believe our estimates are
reasonable.
In these ground motions, equivalent-linear models of
base-isolated buildings show estimates of isolator displace-
ments of 0.4
–
1 m. Local site amplification and nonlinear iso-
lator behavior
—
not included in the ground motion and
building models, respectively
—
would tend to increase the
predicted estimates of base-isolator displacements. Since im-
portant government buildings, hospitals, and communication
centers must remain functional after a large earthquake, en-
gineers should continue to consider large ground motions,
like the ones used in the study, when designing isolation sys-
tems for these types of buildings.
The peak spectral accelerations from the three magni-
tude 7.8 earthquakes exceed the upper bound of the 2006
IBC
design spectra. On 10%
–
25% of the urban area, the spec-
tral accelerations from the simulations exceed the lower
bound of the current design spectrum. We conclude that
some ground motions from these earthquakes exceed the
current design-spectra values. This conclusion is limited be-
cause we do not account for the uncertainties in generating
the ground motions, and we do not compare the response
spectrum at each site to the corresponding site-specific de-
sign spectrum. Future simulations that account for the varia-
bility in slip distribution, hypocenter location, etc., and make
site-specific comparisons, can provide additional informa-
tion on the adequacy of the design spectra and
MCE
level
defined in the 2006
IBC
.
Modern long-period buildings have not experienced
ground motions as large as those predicted by simulations
of earthquakes like the one in 1906. Large long-period
ground motions are a direct consequence of large fault slips
in a great earthquake, and these large fault slips are the pro-
ducts of plate tectonics. Although great earthquakes are unu-
sual, they are inevitable. These simulations provide insight
Table
3
Percent of Urban Areas that Exceed Given Spectral Displacement (SD)
SD Cutoff
Simulation
2-sec Period 10% Damping
2-sec Period 20% Damping
3-sec Period 10% Damping
3-sec Period 20% Damping
0.4 m
M
7.8 Bodega Bay
2.8
1.0
51
23
M
7.8 San Juan Bautista
0.98
0.49
14
6.1
0.6 m
M
7.8 Bodega Bay
0.70
0.34
19
4.8
M
7.8 San Juan Bautista
0.28
0.031
3.9
0.92
1.0 m
M
7.8 Bodega Bay
0.031
0
1.7
0.58
M
7.8 San Juan Bautista
0
0
0.24
0.031
Linear spectral displacement is expected to be a conservative estimate of base-isolated building response. In these magnitude 7.8 hypothetical
earthquakes, the models with a 2-sec period show large displacements on a smaller area compared to the models with a 3-sec period. Higher
damping levels reduce the area where we expect to see large displacements. We expect more sites would exceed these levels in large ground
motions and on soft-soil sites.
Long-Period Building Response to Earthquakes in the San Francisco Bay Area
1063