C H

2A

Engineering Seismology
P T E R

This chapter presents a brief introduction to the basics of seismology and engineering
seismology. The fundamental principles, definitions, and terminologies are crucial
to the seismic risk analysis of nuclear power plants (NPPs) and are used throughout
this book. More details on seismology and engineering seismology can be found in
standard textbooks, such as Kramer (1996).

2.1 Earthquakes
2.1.1 Internal Structure of the Earth
The earth is nearly spherical, with an average radius of approximately 6,400 km. The
earth consists of three different structural layers from its surface inward: crust, mantle,
and core (Figure 2.1). The crust is the outermost layer of the earth. Its thickness ranges
from about 5 km (beneath the oceans) to about 80 km (under some young mountain
ranges). The average thickness of the crust ranges from 30 km to 40 km under the
continents. The crust is only a very small fraction of the earth’s diameter but most
earthquakes occur within the crust.
The layer beneath the crust is the mantle. The boundary between the crust and the
underlying mantle is known as the Mohorovicic Discontinuity, or the Moho, named
after the seismologist who discovered it in 1909. Within the range of 40 km to 70 km
beneath the Moho is a layer of solid rock. This layer of solid rock together with the
crust forms the so-called lithosphere with a thickness of about 150 km. Underneath the
lithosphere, there is a layer of partially molten rock with a thickness of about 450 km,
named asthenosphere. The layer of the solid rock together with the asthenosphere is
called the upper mantle. Beneath the upper mantle is the lower mantle, which appears
to be structurally and chemically homogeneous. No earthquakes have been recorded
in the lower mantle.
17
18

Figure 2.1 Internal structure of the earth.

The core consists of the outer core (or liquid core) and the inner core (or solid core).
The outer core is liquid, which cannot transmit shear waves, while the inner core is a
very dense solid, made up of nickel–iron material. With the depth of the earth increases
from its surface to its inside, in general, the densities, temperatures, and pressures of
the earth layers increase significantly.

2.1.2 Plate Tectonics and Boundaries

Theory of Plate Tectonics
The theory of continental drift was proposed based on the observations of similarity
between the coastlines and geology of eastern South America and western Africa,
and the southern part of India and northern part of Australia, as shown in Figure
2.2. Wegener, a German geophysicist, believed that a supercontinent (Pangea) existed
about 250 million years ago and that Pangea broke apart into the continents about 200
million years ago; the broken pieces of the Pangea then slowly drifted into the present
configuration of the continents.
Based on the theory of continental drift, the modern theory of plate tectonics
started to evolve, which became one of the foundations of modern geology. The basic
hypothesis of plate tectonics is that the surface of the earth consists of a number of large
and intact blocks, called plates, and that these plates move with respect to each other.
Two major pieces of evidence support the theory of plate tectonics: paleomagnetism
and the map of earthquake epicentres (more details in Kramer, 1996).
2.1 earthquakes 19

Figure 2.2 Pangea and the evolution of the continents.

20

Figure 2.3 Plate boundaries.

While the theory of plate tectonics assumes movements of large continental plates,
the sources of such movements have been of great interest to the researchers in this area.
The most widely recognized explanation of the sources of the plate movements is based
on the thermomechanical equilibrium of the materials of the earth.
The upper portion of the mantle is in contact with the cool crust and the lower
portion of the mantle is in contact with the hot outer core, which creates significant
temperature gradient within the mantle. As the denser magma rises, it becomes cooler.
Due to the unstable situation that the denser (cooler) material rests on top of the less
dense (warmer) material, the cooler and denser material begins to sink under the action
of gravity. The sinking material gradually becomes warmer and less dense; eventually,
it moves laterally and begins to rise. Such convection currents in the semimolten rock
of the mantle impose shear stresses on the bottom of the plates, “dragging” them in
various directions across the surface of the earth, as shown in Figure 2.3.
Plate Boundaries
Depending on the direction in which the plates move, three types of plate boundaries
have been identified: divergent boundary, convergent boundary, and transform fault
boundary. The characteristics of the plate boundaries lead to different natures of the
earthquakes that occur along them.
When the plates tend to separate, the divergent boundary, also called spreading
ridge boundary, forms. As shown in Figure 2.3, the molten rock from the underlying
mantle rises to the surface, cools in the gap formed by spreading plates, and becomes
part of the spreading plates.
When the plates tend to come together, the convergent boundary, or subduction
zone boundary, is formed. As shown in Figure 2.3, at the contact points of the plates,
one plate subducts beneath the other plate. Because the size of the earth remains
constant, the consumption of the plate material at the subduction zone boundary is
considered to balance the creation of new plate material at the spreading ridge boundary.
2.1 earthquakes 21

Figure 2.4 San Andreas Fault.

Figure 2.5 An example of a fault exposed to the ground surface.

Figure 2.6 Accumulation and release of elastic strain energy in the vicinity of boundary.

When the plates slide horizontally past each other, the transform fault boundary is
formed. As illustrated in Figure 2.3, the transform faults are usually found in offsetting
spreading ridges.
Fault is defined as a break in earth’s crust along which displacement of rock occurs.
The well-known San Andreas Fault, shown in Figure 2.4, is a continental transform
fault that extends roughly 1,300 km through California and forms the tectonic boundary
between the Pacific Plate and the North American Plate. An example of a fault exposed
to the ground surface is shown in Figure 2.5.
22

2.1.3 Seismic Source Mechanism

Elastic Rebound Theory
The occurrence of earthquakes can be explained by the plate tectonic movements and
the elastic rebound theory. Relative movement of the plates causes stresses to build
up on their boundaries. As illustrated in Figure 2.6, when relative movement occurs,
elastic strain energy accumulates in the vicinity of the boundaries. When the shear
stress reaches the shear strength of the rock along the fault, the rock fails, and the
accumulated strain energy is released.
The effects of the failure depend on the nature of the rock along the fault. If the
rock is weak and ductile, strain energy stored will be released relatively slowly, and
the movement will occur aseismically. If the rock is strong and brittle, the failure will
be rapid. Rupture of the rock will release the stored energy explosively, producing
earthquakes. The theory of elastic rebound describes this process of the successive
buildup and release of strain energy in the rock adjacent to faults. The size of an
earthquake depends on the amount of energy released.
Faults
At a particular location, a fault is assumed to be planar with an orientation described
by strike and dip. As shown in Figure 2.7, the strike is the compass direction relative to
north, and the dip is the angle of inclination or slope relative to horizontal plane. The
focus is the point on the fault plane where an earthquake originates, and the epicentre
is the point on the earth’s surface located directly above the focus of an earthquake.
Figure 2.7 also shows the hanging wall, which is the block positioned over the fault
plane, and the foot wall, which is the block positioned under the fault plane. Various
fault movements are illustrated in Figure 2.8.
Fault movement occurring primarily in the direction of the dip (or perpendicular to
the strike) is referred to as dip-slip movement. Normal faults occur when the ground is
stretched and when the hanging wall moves downward relative to the foot wall. Reverse
faults occur when the ground is compressed and when the hanging wall moves upward
relative to the foot wall.
Fault movement occurring parallel to the strike is referred to as strike-slip movement.
Right-lateral strike-slip fault occurs when an observer would observe the material on
the other side of the fault moving to the right. The San Andreas Fault shown in
Figure 2.4 is a right-lateral strike-slip fault. Left-lateral strike-slip fault occurs when an
observer would observe the material on the other side of the fault moving to the left. In
most cases, fault slip is a mixture of strike-slip and dip-slip and is called oblique fault.
Interplate Earthquakes
The seismic source mechanism of the interplate earthquakes, which occur at the plate
boundaries, has been well explained by the theories of plate tectonics and elastic
2.1 earthquakes 23

Figure 2.7 Definition of fault orientation.

Figure 2.8 Fault movements.

24

rebound. Because the deformation occurs predominantly at the boundaries between

the plates, the locations of earthquakes have been concentrated near plate boundaries.
Earthquakes occurring at convergent plate boundaries contribute more than 90 % of the
world’s release of seismic energy. Most of the largest earthquakes have originated in the
subduction regions a result of the thrusting of one plate under another. About 10 % of
the world’s earthquakes occur along the divergent ocean-ridge system, and contribute
only about 5 % of the world’s total seismic energy.
The (Pacific) Ring of Fire is an area where large numbers of earthquakes and volcanic
eruptions occur in the basin of the Pacific Ocean (Figure 2.9). In a 40,000-km horseshoe
shape, it is associated with a nearly continuous series of oceanic trenches, volcanic arcs,
volcanic belts, and plate movements. It contains 452 volcanoes (over 75 % of the world’s
active and dormant volcanoes). About 90 % of the world’s earthquakes and 80 % of the
world’s largest earthquakes occur along the Ring of Fire.
Intraplate Earthquakes
Some earthquakes do not necessarily occur at the plate boundaries; instead, they
occur in the interior of the tectonic plates and are called intraplate earthquakes. The
recurrence frequency of the intraplate earthquakes is much lower than the interplate
earthquakes. The seismic damage of the intraplate earthquakes could be tremendous
because the human population is more concentrated within the plates. Moreover,
the interior of the tectonic plates generally has thick layers and high strength, which
may produce large and shallow-focus earthquakes. Intraplate earthquakes arise from
more localized systems of forces in the crust, perhaps associated with ancient geological
structural complexity, or with anomalies in temperature and strength of the lithosphere.

2.1.4 Seismic Waves

When an earthquake occurs, different types of seismic waves are produced: body waves
and surface waves (Figure 2.10).
❧ Body Waves (P-waves and S-waves) can travel through the interior of the earth
along paths influenced by the material properties (density and stiffness), resembling
the reflection and refraction of light waves (Figure 2.11).
• P-waves (primary, compressional, or longitudinal waves) involve successive com-
pression and dilation of the materials through which they pass. They are analogous
to sound waves; the motion of an individual particle is parallel to the direction of
the P-wave. Like sound waves, P-waves can travel through any type of material. The
speed of body waves varies with the stiffness of the materials they travel through.
Because geologic materials are stiffest in compression, P-waves travel faster than
other seismic waves (at nearly twice the speed of S-waves) and are therefore the first
to arrive at a particular site.
 2.1 earthquakes

Figure 2.9 Ring of Fire.

25
26

P-Wave
Compression Compression
Dilation Dilation Dilation
Undisturbed material

Direction of wave propagation

S-Wave
Undisturbed material

Direction of wave propagation

Rayleigh Wave
Undisturbed material

Direction of wave propagation

Love Wave
Undisturbed material

Direction of wave propagation

Figure 2.10 Seismic waves.
2.1 earthquakes 27

Figure 2.11 Light beams.

Figure 2.12 Seismic waves travelling through the earth.

• S-waves (secondary, shear, or transverse waves) cause shearing deformations as they

travel through a material. The motion of an individual particle is perpendicular to
the direction of the S-wave. The direction of particle movement divides S-waves
into two components: SV (vertical plane movement) and SH (horizontal plane
movement). S-waves are slower than P-waves, with speeds typically around 60 % of
that of P-waves in any given material. An S-wave shakes the ground surface vertically
28

and horizontally, which is particularly damaging to structures. Fluids, which have

no shearing stiffness, cannot sustain S-waves.
❧ Surface waves (Rayleigh waves and Love waves) result from the interaction between
body waves at the surface and surficial layers of the earth. They travel along the
earth’s surface with amplitudes that decrease roughly exponentially with depth. Due
to the nature of the interactions required to produce them, surface waves are more
prominent at distances farther from the source of the earthquake. At distances
greater than about twice the thickness of the earth’s crust, surface waves, rather than
body waves, will produce peak ground motions. The most important surface waves,
for engineering purposes, are Rayleigh waves and Love waves.
• Rayleigh waves are produced by the interaction of P-waves and SV-waves with the
earth’s surface. They involve both vertical and horizontal particle motions. Rayleigh
waves are also called “ground rolls”, similar to water waves produced by a rock
thrown into a pond. They are slower than body waves, roughly 90 % of the velocity
of S-waves for typical homogeneous elastic media.
• Love waves result from the interaction of SH-waves within a soft surficial layer.
They have no vertical component of particle motion. They usually travel slightly
faster than Rayleigh waves, about 90 % of the S-wave velocity, and have the largest
amplitude. The horizontal shaking of Love waves is particularly damaging to the
foundations of structures. Love waves do not propagate through liquids.
As the different types of seismic waves travel through the earth, they are refracted and
reflected at boundaries between different layers, as light beams (Figure 2.11). Whenever
a wave is reflected or refracted, some of the energy is converted to waves of the other
types (as shown in Figure 2.12). Seismic waves reach different points on the earth’s
surface by different paths. In strong ground shaking on land, after the first few shakes,
a combination of the two kinds of waves, body waves and surface waves, is usually
felt. It is noted that studies of the refractions and reflections of the seismic waves
actually brought to light the layered structure of the earth and provided insight into the
characteristics of each layer.

2.1.5 Size of Earthquakes

The size of an earthquake is a crucial parameter and has been described in different
ways. The oldest measure of earthquake size is the earthquake intensity. The intensity
is a qualitative description of the effects of the earthquake at a particular location, as
evidenced by observed damage and human reactions. The most famous earthquake
intensity is the Modified Mercalli intensity (MMI) scale originally developed by the
Italian seismologist Mercalli and modified later by Richter (1958). The MMI scale is
listed in Table 2.1.
2.1 earthquakes 29
Table 2.1 Modified Mercalli intensity (MMI) scale.
I Not felt
II Felt only by persons at rest
III–IV Felt by persons indoors only
V–VI Felt by all; some damage to plaster, chimneys
VII People run outdoors, damage to poorly built structures
VIII Well-built structures slightly damaged;
poorly built structures suffer major damage
IX Buildings shifted off foundations
X Some well-built structures destroyed
XI Few masonry structures remain standing; bridges destroyed
XII Damage total; waves seen on ground; objects thrown into air

Seismic instruments have allowed an objective, quantitative measurement of earth-

quake size in terms of earthquake magnitude. Most magnitude scales are based on
measured ground-motion characteristics. The concept of earthquake magnitude was
first introduced by Richter (1935), named Richter local magnitude ML , i.e.,

ML = log10 A − log10 A0 (δ), (2.1.1)

where the empirical function A0 (δ) depends only on the epicentral distance of the
station δ. The base-10 logarithm of the maximum amplitude of waves (as shown in
Figure 2.13) recorded by seismographs, is adjusted to compensate for the variation in
the distance between the various seismographs and the epicentre of the earthquake.
There are other instrumental magnitude scales: surface wave magnitude Ms , which
is based on the amplitude of Rayleigh waves; body wave magnitude mB , which is based
on the amplitude of P-waves that is not affected by the focal depth of the source.
For strong earthquakes, the measured ground-shaking characteristics become less
sensitive to the size of the earthquake than for smaller earthquakes. This phenomenon
is known as the magnitude saturation. Body wave magnitude mB and Richter local
magnitudes ML saturate at magnitudes of 6 to 7, and surface wave magnitude Ms
saturates at Ms = 8, as shown in Figure 2.14.
Because these amplitudes tend to reach limiting values, they may not accurately
reflect the size of very large earthquakes. The moment magnitude (denoted as Mw or
M ), introduced by Hanks and Kanamori (1979) to describe the size of any earthquake,
is defined as
Mw = M = 23 log10 M0 − 10.7, (2.1.2)
which is not obtained from ground-motion characteristics. M0 is the seismic moment
(in dyne-cm) of an earthquake given by

M0 = μA D̄, (2.1.3)
30

Maximum
Amplitude A
t

Figure 2.13 Maximum amplitude of ground motion.

9
Mw (M ) Ms
8
ML
7
mb
Magnitude

3
3 4 5 6 7 8 9 10
Moment Magnitude Mw (M )
Figure 2.14 Earthquake magnitudes.

Table 2.2 Energy released in an earthquake.

ML e (Joules) Number of Nuclear Bombs
6 6.310× 1013 0.8
7 1.995× 1015 25
8 6.310× 1016 794
9 1.995× 1018 25,099

where μ is the rupture strength of the material along the fault, A is the rupture area,
and D̄ is the average amount of slip. Because the seismic moment is a measure of the
work done by the earthquake, the moment magnitude is the only magnitude scale not
subject to saturation. Figure 2.14 shows a comparison between different magnitude
scales. Because earthquake magnitude scales are base-10 logarithmic, a unit change
in magnitude corresponds to a 10-fold change in the magnitude parameter (ground-
motion characteristic or seismic moment).
The total seismic energy (in ergs = 10−7 joules) released during an earthquake is
often estimated as
log10 e = 11.8 + 1.5 ML . (2.1.4)
A unit change in magnitude corresponds to a 101.5 = 32-fold change in energy released
by an earthquake. The energy released in an atomic bomb of the size used at Hiroshima
is 19,000-ton TNT equivalent (1 ton of TNT = 4.184×109 joules). Table 2.2 lists the
relationship between energy released and the Richter magnitude ML in an earthquake.
2.1 earthquakes 31

Before Earthquake

Lock
ed

Tsunami Tsunami
Water
pushed up
Subside During Earthquake
Spring upward

Slippe
d

Figure 2.15 Generation of tsunami.

Wave Length 10.6 km

213 km 23 km

10 m
50 m

4,000 m
Depth (m) Velocity (km/hr) Wave Length (km)
7,000 943 282
4,000 713 213
2,000 504 151
200 159 48
50 79 23
10 36 10.6

Figure 2.16 Propagation of tsunami.

2.1.6 Generation of Tsunami

As relative movement of the tectonic plates occurs, the plates are locked together at
the boundary, and tremendous elastic strain energy is built up in the materials near
the boundary, as illustrated in Figure 2.15. When the shear stress reaches the shear
strength of the rock along the fault, the rock fails, and the accumulated strain energy is
released. If the rock is strong and brittle, rupture of the rock will release the lock at the
boundary, and the top plate springs upward explosively. The sudden displacement in
the seabed is sufficient to cause the sudden raising or lowering of a large body of water.
Following the initial disturbance to the sea surface, water waves spread in all direc-
tions. Figure 2.16 illustrates the propagation of tsunami waves. Their speed of travel in

deep water is given by gH, where H is the sea depth.
In the open sea, tsunami waves are characterized by very large wavelength, reaching
tens to hundreds of kilometers. Their velocities, depending on the water depth, can
32

reach as high as 800 to 900 km/hour. They can propagate for thousands of kilometers
without dissipating much of their energy and can travel long distances. Tsunami waves
are usually unnoticed because of their heights commonly not exceeding 1 m.
When tsunamic waves approach the coast, their velocities decrease directly propor-
tionally to the water depth; their heights increase and can reach tens of meters.

2.2 Case Study − The Great East Japan Earthquake

The Great East Japan Earthquake (東日本大震災, Higashi Nihon Daishinsai), also
known as the 2011 Tōhoku Earthquake, occurred at 14:46 JST (05:46 UTC) on Friday,
March 11, 2011. It was an undersea mega thrust earthquake with a moment magnitude
Mw = 9.0. The epicentre was at 38.322◦ N, 142.369◦ E, near the east coast of the island
of Honshu, with the hypocenter at an underwater depth of approximately 32 km. The
nearest major city to the epicentre is Sendai, on the main island of Honshu, 130 km
away. The duration of the quake was six minutes.
It was the most powerful known earthquake to have hit Japan; the earthquake and
the accompanying tsunami provoked the largest crisis that Japan has encountered in
the 65 years since the end of World War II. It is the fourth most powerful earthquake in
the world, since modern record keeping began in 1900.
The largest PGA (three components vector sum) among K-NET and KiK-net sites
was recorded at MYG004 K-NET station (latitude: 38.7292◦ , longitude: 141.0217◦ ,
distance to epicentre: 173.3 km) reaching 2.99g.
Japan lies at the crossing of four tectonic plates: Eurasian, North American, Philip-
pine, and Pacific plates, as shown in Figure 2.17. Details of the Kanton Triple Junction
are shown in Figure 2.18. It is clearly seen that the Philippine plate subducts under
the Eurasian plate at the Nankai Trench, and the Pacific plate subducts under both the
Philippine and Eurasian plates at the Japan Trench and Izu-Bonin Trench. As a result,
Japan endures 20 % of the world’s powerful earthquakes. The northern part of Japan
belongs to the North American plate but the relative displacement between the Eurasian
and the North American plates in this region is relatively low.
The earthquake was an extremely massive event; the fault moved upwards of 30−40
m and slipped over an area approximately 300 km long (along-strike, N-S) by 150 km
wide (in the down-dip direction, E-W) caused multiple ruptures of seismic sources.
The rupture zone is roughly centred on the earthquake epicentre along strike, while
peak slips were up-dip of the hypocentre, towards the Japan Trench axis. As shown
in Figures 2.19 and 2.20, the main earthquake was preceded by a number of large
foreshocks and followed by hundreds of aftershocks. The first major foreshock was
a 7.2 Mw event on March 9, approximately 40 km from the location of the March 11
quake, with another three on the same day in excess of 6.0 Mw .
2.2 case study − the great east japan earthquake 33

Figure 2.17 Tectonic plates around Japan.

Pa
Japa

c ific
e

n
lat

Tren

Pla
nP

ch

te
ia

Japan
ras

Iz
Eu

u-

Sagami
Bo

Trench
n
Iz

in
u-

Tr
B
on

en
in

ch
Ar
c

nch
re
aiT te
N
ank
e Pla
pin
ilip
Ph

Figure 2.18 Kanton triple junction.

34

9.0

8.0

Magnitude (Mw) 7.0

6.0

5.0
Date (MM/DD)
03/09 03/11 03/13 03/15 03/17
Figure 2.19 Time distribution of earthquake magnitudes.

Figure 2.20 Location distribution of earthquake magnitudes.

An upthrust of 6 to 8 m along the rupture fault resulted in a major tsunami that

brought destruction along the Pacific coastline of Japan’s northern islands. Figure 2.21
shows the distribution of height of tsunamic waves along the Pacific coastline of Japan.

Fukushima Nuclear Power Stations

Table 2.3 lists the PGA values at Fukushima Nuclear Power Stations (NPS) and the
corresponding design basis earthquake (DBE) accelerations. It is seen that PGA in E-W
direction at Daiichi Units 2, 3, and 5 exceeded the design tolerances.
Ground acceleration thresholds for automatic reactor shutdown are 1.35−1.50 m/sec2
in the horizontal directions or 1.00 m/sec2 in the vertical direction. When the earth-
quake occurred, the reactors of Units 1, 2, and 3 were operating. They were successfully
2.2 case study − the great east japan earthquake 35

Figure 2.21 Height of tsunami waves.

Table 2.3 Ground acceleration at Fukushima Nuclear Power Stations

Peak Ground Accel. Design Basis Accel.
Fukushima (m/sec2 ) (m/sec2 )
Horizontal Horizontal
Vertical Vertical
Unit N-S E-W N-S E-W
1 4.60 4.47 2.58 4.87 4.89 4.12
2 3.48 5.50 3.02 4.41 4.39 4.20
3 3.22 5.07 2.31 4.49 4.41 4.29
Daiichi
4 2.81 3.19 2.00 4.47 4.45 4.22
5 3.11 5.48 2.56 4.52 4.52 4.27
6 2.98 4.44 2.44 4.45 4.48 4.15
1 2.54 2.30 3.05 4.34 4.34 5.12
2 2.43 1.96 2.32 4.28 4.29 5.04
Daini
3 2.77 2.16 2.08 4.28 4.30 5.04
4 2.10 2.05 2.88 4.15 4.15 5.04

shut down by the automatic systems installed as part of the design of the NPS to detect
earthquakes (called SCRAM).Although all off-site power was lost when the earthquake
occurred, all available emergency diesel generator power systems were in operation, as
designed. Units 4, 5, and 6 had already been shut down for periodic inspection.
To protect Fukushima Daiichi NPS again tsunami (Figure 2.22), the design basis was
5.7 m (height of seawall). However, the first of a series of large tsunami waves reached
the site about 46 min after the earthquake. The maximum height of tsunami was 14 to
15 m. The ground level is 10 m at Units 1−4, and the units were inundated by about 4
36
Reactor Building O.P. — Onahama Port Construction Base Level
Inundation Height
Turbine Building Approximately O.P. +14–15 m
Site Level O.P. +4 m Seawall
Height
Site Level O.P. +5.7 m
O.P. +10 m Base Level
(Units 1–4) O.P. 0 m
Site Level of Units 5 & 6 O.P. +13 m Water Intake Seawall
Figure 2.22 Elevations of Fukushima Daiichi NPS.

to 5 m of water. The ground level is 13 m at Units 5 and 6, and the units were inundated
by up to 1 m of water.
For Fukushima Daini NPS, the design basis was 5.2 m. The height of tsunami was
about 6.5 to 7 m. Although the ground level at the site is 12 m above sea level, the
run-up height of the wave in the main building area reached about 14 to 15 m on the
south side of the Unit 1 buildings, with flooding of about 2 to 3 m. The area on the
ocean side, where seawater pumps were located, was inundated.
Although the plants withstood the effects of the earthquake, the tsunami caused
the loss of all power sources, except for one emergency diesel generator providing
emergency power shared by Units 5 and 6. The station blackout rendered the loss of
all instrumentation and control systems at Units 1−4: safety systems for cooling of the
reactor cores and control of containment pressures. The tsunami and associated large
debris caused widespread destruction of many buildings, doors, roads, tanks, and
other site infrastructure, including loss of heat sinks. The operators had to work in
darkness, with almost no instrumentation and control systems, to secure the safety of
reactors and associated fuel pools, a common fuel pool, and dry cask storage facilities.
In summary, at Units 2, 3, and 5 of the Fukushima Daiichi NPS, the acceleration
response spectra of seismic ground motion observed on the basemat exceeded the
design basis in a part of the period-band. Although damage to the external power
supply was caused by the earthquake, no damage caused by the earthquake to systems,
equipment, or devices that are important for nuclear reactor safety, at nuclear reactors,
has been confirmed.
On the other hand, the tsunamis that hit the Fukushima Daiichi NPS were 14−15
m high, substantially exceeding the height assumed under the design of construction
permit or the subsequent evaluation. The tsunamis severely damaged seawater pumps
and other equipment, causing the failure to secure the emergency diesel power supply
and reactor cooling function. The procedural manual did not assume flooding from a
tsunami, but rather only stipulated measures against a backrush. The design against
tsunamis has been based on tsunami folklore and indelible traces of tsunami, not on
adequate consideration of the recurrence of large-scale earthquakes in relation to a
safety goal to be attained. The assumption on the frequency and height of tsunamis
was insufficient; therefore, measures against large-scale tsunamis were not prepared
adequately.
2.3 strong ground motion 37

2.3 Strong Ground Motion

2.3.1 Strong-Motion Measurement
Ground motion, also known as ground shaking, is a vibration near or at the ground
surface of the earth induced by the seismic waves released from the seismic sources.
Although the world’s first seismic instrument (seismoscope) was invented in 132 B.C.,
the first measurements of ground motions, which are important from an engineering
perspective, were made only several decades ago.
The current widely used instruments for ground-motion measurements are gener-
ally divided into two categories: seismographs and accelerographs. The seismographs
are used to measure relatively weak ground motions, and the records they produce are
called seismograms. Seismologists usually use seismograms to investigate and better
understand geophysics or seismology-related areas, such as seismic source mecha-
nisms, earth mediums along the seismic wave paths, and seismic wave propagations.
The accelerographs are used to measure strong ground motions and to produce
the records in the form of accelerograms in most cases. The strong-motion records
are of great interest to engineers who work on engineering seismology, earthquake
engineering, or structural engineering. They may be used as a direct seismic input
to analyze engineering structures or used as inputs for seismic hazard analysis and for
determining the design earthquakes (as discussed in Chapters 4 and 5).
The main purpose of an accelerograph is to measure the entire strong ground-
motion process near or at the ground surface of the earth during an earthquake.
Ground-motion acceleration is usually selected as the physical quantity to describe
the ground-motion process with time based on several reasons:
1. Acceleration is closely relevant to the seismic inertia forces, which are the main
parameter considered in earthquake and structural engineering.
2. Recording apparatus for acceleration-based accelerographs can be readily manu-
factured.
3. Ground-motion velocity and displacement processes can be easily obtained
mathematically from the recorded acceleration process.
As a result, most accelerographs record strong ground motions in terms of acceleration.
Early strong-motion recording instruments are analog accelerographs, which trans-
form the ground motions to the motion of a physical mechanism using a stylus on
paper. Although the strong-motion records produced by these analog accelerographs
have played an extremely important role in the development of earthquake engineering,
several deficiencies of analog accelerographs have been widely recognized:
1. Because the analog accelerographs are triggered to start recording by the exceedance
of a small threshold acceleration at the beginning of the earthquake motion, they
are unable to record any vibrations that have preceded the triggering.
38

Figure 2.23 An example of ground motion recorded in Manitoba, Canada.

2. The accuracy and capacity of the analog accelerographs are not adequate to record
strong motions with long-period (low-frequency) components and high amplitudes
of acceleration.
3. The physical recording mechanism of the analog accelerographs results in consid-
erable errors in the digitization of the records (including corrections for instrument
response and baseline), which may require significant effort on processing the
strong-motion records.
Compared to analog accelerographs, modern digital accelerographs are able to over-
come these deficiencies. Digital accelerographs are capable of recording strong-
motions with much wider ranges of frequency and amplitude and have significantly
improved the accuracy. Digital accelerographs are able to record a process with several
seconds prior to the arrival of the seismic waves, such that a complete strong-motion
record (starting with P-wave, containing the maximum amplitude of the P-wave, and
having complete processes of S-wave and surface waves) can be obtained. An example
of a complete ground motion recorded in Manitoba, Canada, is shown in Figure 2.23.

2.3.2 Ground-Motion Parameters

Earthquake engineers are interested in strong ground motion, i.e., motion of sufficient
strength to affect people and their environment. At a given point, ground motions
produced by earthquakes can be completely described by three components of transla-
tion (two orthogonal horizontal directions and one vertical direction, e.g., East–West,
North–South, Vertical) and three components of rotation, which are usually neglected
in engineering practice due to their complexity and minor effect.
For engineering purposes, three characteristics of earthquake motion are of primary
significance: amplitude, frequency content, and duration of the motion. It is important
2.3 strong ground motion 39

to identify a number of ground-motion parameters that reflect these characteristics in

a compact and quantitative form by analyzing the strong-motion records.
Figure 2.24 shows the East–West component of ground acceleration, velocity, and
displacement of the 1940 El Centro Earthquake. The acceleration time-history shows a
significant proportion of relatively high frequencies. Because velocity is the integration
of acceleration and integration produces a smoothing or filtering effect, the veloc-
ity time-history shows substantially less high-frequency motion than the acceleration
time-history. Displacement time-history is dominated by relatively low-frequency mo-
tion.
Peak ground acceleration (PGA), as illustrated in Figure 2.24, is the most commonly
used parameter for describing ground motion, because of the relationship between
acceleration and inertial force. Ground motion has two orthogonal horizontal and one
vertical components. In seismic design and analysis practice, it is usually assumed that
the peak ground accelerations in the two horizontal directions are equal. The maximum
vertical acceleration is assumed to be a factor (e.g., two-thirds) of the peak horizontal
component. The largest dynamic forces induced in certain types of structures (very
stiff structures) are closely related to the PGA.
Because the velocity is less sensitive to the higher-frequency components of the
ground motion, peak ground velocity (PGV) can characterize ground-motion ampli-
tude more accurately at intermediate frequencies. For structural systems with dominant
intermediate frequencies, such as tall or flexible buildings, PGV provides a more accu-
rate indication of the potential for damage than PGA.
Peak ground displacement (PGD) can characterize ground-motion amplitude more
accurately at low frequencies. For flexible structural systems with dominant low fre-
quencies, PGD provides a more accurate indication of the potential for damage.
Because the PGA is considered as a random number of a strong-motion process, the
sum of all the squared acceleration values from a strong-motion record can be used as
a more reliable parameter to represent the statistical characteristics of the amplitude of
a strong-motion record, i.e., the Arias intensity defined as
 T
π  
IA = A(t) 2 dt, (2.3.1)
2g 0

where T is the duration of the earthquake.

A plot of the buildup of Arias intensity with time, i.e,
 t
π  
IA (t) = A(s) 2 ds, 0  t  T, (2.3.2)
2g 0

or the cumulative normalized Arias intensity IA (t)/IA is known as a Husid plot, and it
serves to identify the interval over which the majority of the energy is imparted.
The duration of a strong-motion is related to the time required for release of accu-
mulated strain energy by rupture along the fault. As the length, or area, of fault rupture
40
0.2
0.1 Time Interval Δt = 0.01s
A (g) 0
−0.1 T = 53.45s
−0.2 Ground Acceleration
PGA= 0.2107g
−0.3
40
20
V (cm/s)

0
−20 Ground Velocity
PGV =31.31 cm/s
−40
30
PGD = 24.15 cm
20 Ground Displacement
D (cm)

10
0
Time (s)
−10
0 10 20 30 40 50 53.45
Figure 2.24 East–West component of the 1940 El Centro Earthquake.
Strong-Motion Duration t5-95
0.2
0.1
0
A (g)

−0.1 T= 53.45 s
Ground Acceleration
−0.2 PGA 0.2107g Time (s)
−0.3
0 2.14 10 20 26.29 30 40 50
Strong-Motion Duration t5-75
100
Cumulative Normalized Arias Intensity IA(t) / IA (%)

95 IA = IA(53.45 s) = 1.1687 m/s

80
75

60

40

20

5 Time (s)
19.87
0
0 2.14 10 20 26.29 30 40 50 53.45
Figure 2.25 Arias intensity and strong-motion duration of the El Centro Earthquake.

increases, the time required for rupture increases. The duration is proportional to the
cubic root of the seismic moment. For engineering purposes, only the strong-motion
portion of the accelerogram is of interest. A popular definition of a strong-motion
duration tm used in nuclear industry is t5−75 (the time span between 5 % and 75 % of
the Arias intensity) or t5−95 (the time span between 5 % and 95 % of the Arias intensity).
2.3 strong ground motion 41
3.5

Fourier Amplitude Spectrum (m/s) 3 (a) Frequency Resolution = Δf = 0.0122 Hz

2.5
2
1.5
1
0.5
0
0 5 10 15 20 25
Frequency (Hz)
2.5
Fourier Amplitude Spectrum (m/s)

2 (b) Frequency Resolution = 4 Δf = 0.0488 Hz

1.5

0.5

0
0 5 10 15 20 25
Frequency (Hz)
2
Fourier Amplitude Spectrum (m/s)

1.5 (c) Frequency Resolution = 8 Δf = 0.0976 Hz

0.5

0
0 5 10 15 20 25
Frequency (Hz)
Figure 2.26 FAS of the El Centro Earthquake.

Figure 2.25 presents the cumulative normalized Arias intensity IA (t)/IA of the
East–West component of the El Centro Earthquake and the strong-motion duration
   
t5−75 = 2.14, 19.87 ⇒ 17.73 s or t5−95 = 2.14, 26.29 ⇒ 24.15 s.
The amplitudes of the ground motions, such as PGA, PGV, PGD, or Arias inten-
sity, are individual parameters to represent the strength of the ground motion. The
frequency contents, which have been widely recognized as important parameters to
engineering structures, are usually described by ground response spectra (GRS, see
Chapter 4) or Fourier amplitude spectra (FAS, see Section 3.1.4 and Appendix B, in
particular, Sections B.2 and B.6).
42

The duration of the E-W component of the El Centro Earthquake is T = 53.45 s with
time interval s = 0.01 s, resulting in 53.45/0.01 = 5345 data points. To perform fast
Fourier transform (FFT) to obtain FAS, the number of data points must be 2N ; hence,
0s are padded at the end of the earthquake to extend the duration to T = 81.92 s so that
there are 213 = 8192 data points.
FAS of the E-W component of the El Centro Earthquake is shown in Figure 2.26(a)
with frequency resolution F = 1/T =1/81.92 = 0.0122 Hz. Because F = 1/T, the
longer the duration T of the earthquake time series, the higher the frequency resolution
(the smaller the value of F ). It is seen in Figure 2.26(a) that, at such a high-frequency
resolution, there are clusters of sharp spikes in the FAS, because it is able to differentiate
harmonics at 0.0122 Hz. As discussed in Section B.6, for engineering applications, it
is necessary to increase the value of frequency interval F to obtain a smoother FAS;
Figure 2.26(b) and (c) shows FAS with frequency resolutions 4F and 8F .
As discussed in Section 3.1.3, power spectral density (PSD) function SXX (ω) is
very important in characterizing a stationary random process X(t). For a transient
nonstationary random process, FAS is often used (Section 3.1.4). Nevertheless, PSD is
frequently employed in earthquake engineering to characterize the frequency contents
of strong-motion portions of earthquake ground motions because the strong-motion
portion of an earthquake ground motion can be reasonably modelled as a stationary
Gaussian process (Section 3.5.2). For an acceleration time-history A(t), the one-sided
PSD is defined by equation (2-1) in ASCE/SEI 4-16 (ASCE/SEI, 2017) or equation (1) in
Appendix B of SRP 3.7.1 (USNRC, 2012b):
 2
2A(ω)
SAA (ω) = , (2.3.3)
2πtm
 
F
 
where A(ω) = A(t) is the Fourier transform of A(t) over the duration tm , A(ω)
is the one-sided FAS, and tm = t5−75 is the strong-motion duration required for the
Arias intensity to rise from 5 % to 75 %. For the E-W component of the El Centro
Earthquake, the one-sided PSD and the average one-sided PSD are shown in Figure
2.27(a) and (b). To obtain average PSD, at any frequency F, the average PSD is
computed over a frequency bandwidth of ±20 % centred on the frequency F, i.e.,
 
0.8 F, 1.2 F .
GRS are used extensively in earthquake engineering practice. A plot of the peak value
of a response quantity, such as acceleration, velocity, or displacement, as a function
of the natural vibration period T or natural frequency F of a linear single degree-of-
freedom (SDOF) system is called the earthquake response spectrum for that quantity.
Each such plot is for SDOF systems having a range of natural frequencies, such as
0  F  100 Hz, and a fixed damping ratio ζ ; several such plots for different values
of ζ are included to cover the range of damping values encountered in engineering
structures. A detailed study of GRS is presented in Chapter 4.
2.4 probabilistic ground-motion parameters 43
0.05

Power Spectral Density (m2/s3) 0.04

(a) One-Sided PSD

0.03

0.02

0.01

0
0 5 10 15 20 25
Frequency (Hz)
0.016

0.014
Power Spectral Density (m2/s3)

(b) Average One-Sided PSD

0.012

0.010

0.008

0.006

0.004

0.002

0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 2 3 4 5 6 7 8 9 10 20 25
Frequency (Hz)
Figure 2.27 PSD of the El Centro Earthquake.

Figure 2.28 Commonly used source–site distance measures.

2.4 Probabilistic Ground-Motion Parameters

2.4.1 Magnitude and Distance Effects
Much of the energy released in an earthquake takes the form of stress waves. Because
the amount of energy released is strongly related to the earthquake magnitude, the
characteristics of the stress waves are strongly related to magnitude. The characteristics
44

of the stress waves are also strongly related to source–site distance. As stress waves travel
away from the source of an earthquake, they spread out and are partially absorbed by
the materials they travel through. Energy per unit volume decreases with increasing
distance from the source. A number of commonly used source–site distance measures
are illustrated in Figure 2.28.
Observations of the relationships between the stress waves and the earthquake mag-
nitude and source–site distance have provided a strong link between the ground-motion
parameters and the earthquake magnitude and source–site distance. This link is the
basis for deriving ground-motion prediction equations (GMPE) for seismic hazard
analysis.

2.4.2 Ground-Motion Prediction Equations

GMPEs for ground-motion intensity measures (i.e., ground-motion parameters), such
as PGA, and spectral acceleration at individual vibration period, as functions of earth-
quake magnitude, source–site distance, and some other variables, are important tools
in seismic hazard analysis. In this section, for illustration purposes, spectral accelera-
tions for horizontal components are used as the intensity measures.
GMPEs are typically developed empirically by regression analyses of recorded
strong-motion amplitude data versus magnitude, distance, and possibly other predic-
tor variables. Some GMPEs were also determined based on simulated ground motions
(Atkinson and Boore, 2006). GMPEs provide a connection between the intensity mea-
sures of earthquake-induced ground motions (e.g., spectral accelerations), which are
directly correlated to seismic analysis and design, and the parameters of earthquakes
(e.g., magnitude and distance).
A typical GMPE for spectral acceleration SA(Tj ) at vibration period Tj can be ex-
pressed as
ln SA(Tj ) = F(m, r, Tj , θ ) + σ ε(Tj ), (2.4.1)
where F(m, r, Tj , θ) is the expected value of the prediction equation, in which, besides
the essential parameters (earthquake magnitude m and source–site distance r), various
predictor variables θ, such as fault type, hanging wall effect, seismic wave propagation
path, and local site condition, may be considered. σ is the total standard deviation of
the prediction equation model, which could be a combination of intra-event (ground
motions within the same earthquake event) and inter-event (ground motions between
different earthquake events) aleatory uncertainties and a function of m, Tj , and other
predictor variables. ε(Tj ) is the number of standard deviations σ by which the loga-
rithmic spectral acceleration ln SA(Tj ) deviates from the expected value F(m, r, Tj , θ)
(Abrahamson and Silva, 1997; Atkinson and Boore, 2006; Boore and Atkinson, 2008).
Given a set of predictor variables m, r, and θ, ε(Tj ) has been verified to follow
standard normal distribution; consequently, ln SA(Tj ) follows normal distribution
2.4 probabilistic ground-motion parameters 45

conditional on m, r, and θ (Baker and Jayaram, 2008). Detailed discussion on the

probability distribution of spectral accelerations is presented in Section 2.4.4.
The functional form of the prediction equation F(m, r, Tj , θ) in equation (2.4.1) is
usually selected to reflect the mechanisms of the ground-motion process as closely
as possible. However, ground motions are complicated; they are influenced by, and
consequently reflect, characteristics of the seismic source, the rupture process, the
source–site travel path, and local site conditions (Kramer, 1996). Various functional
forms of the prediction equation have been adopted to account for different ground-
motion mechanisms, based on the availability of regional recorded ground motions
and engineering practice (Douglas, 2011). To introduce the basic concept of GMPE,
three examples of typical GMPEs are presented.
☞ As discussed in Section 2.1.5, there are many definitions to quantify earthquake
magnitude, such as Richter local magnitude ML and moment magnitude M . A
specific set of GMPEs are usually developed in terms of a particular definition of
earthquake magnitude; moment magnitude is the most popular. When a set of
GMPEs are developed specifically in terms of moment magnitude, the notation
M is used for moment magnitude.
☞ In general, earthquake magnitude is a random variable in probabilistic seismic
hazard analysis (PSHA). The formulations in PSHA are general and are appli-
cable to any earthquake magnitude definition as long as it is consistent with that
for which GMPEs were developed. Following the conventions in probability, the
uppercase M is used to denote earthquake magnitude as a random variable, and
the lowercase m is used to denote a specific magnitude value that M takes.

GMPEs by Abrahamson and Silva

Abrahamson and Silva (1997) developed GMPEs for spectral accelerations. The GM-
PEs are derived for the geometric average of two horizontal components and vertical
component for shallow earthquakes in active tectonic regions. A database of 655 sets of
tridirectional recorded ground motions from 58 earthquakes are used in the regression
analysis. The functional form F(m, r, Tj , θ ) in equation (2.4.1) of the GMPE is given by

ln SA = F 1 (M , rrup ) + F · F 3 (M ) + HW · F 4 (M , rrup ) + S · F 5 (PGArock ), (2.4.2)

where SA is spectral acceleration at a specific vibration period, M is moment magni-

tude, rrup is the closest distance to the rupture plane, F is a coefficient for the fault type
(1 for reverse, 0.5 for reverse/oblique, and 0 otherwise), HW is a coefficient for hanging
wall sites (1 for sites over the hanging wall and 0 otherwise), and S is a coefficient for
local site conditions (0 for rock or shallow soil and 1 for deep soil).
46

Function F 1 (m, rrup ) in equation (2.4.2), which is the basic functional form of the
GMPE for strike-slip events recorded at rock sites, is given by
  
F 1 (M , rrup ) = a1 + a(M −6.4) + a12 (8.5− M )n + a3 + a13 (M −6.4) ln r2rup + c24 ,
(2.4.3)
where a = a2 for M  6.4 and a = a4 for M > 6.4.
Function F 3 (M ) in equation (2.4.2), which allows for magnitude and period depen-
dence of the type of fault, is given by
⎧
⎪
⎪ a , for M  5.8,
⎨ 5
a6 −a5
F 3 (M ) = a5 + , for 5.8 < M < 6.4, (2.4.4)
⎪
⎪ 6.4−5.8
⎩
a6 , for M  6.4.
Function F 4 (M , rrup ) in equation (2.4.2) that allows for magnitude and distance
dependence of the effect of hanging wall is given by

F 4 (M , rrup ) = F HW (M ) · F HW (rrup ), (2.4.5)

where
⎧
⎨ 0, for M  5.5,
F HW (M ) = M −5.5, for 5.5 < M < 6.5,
⎩
1, for M  6.5,
⎧
⎪
⎪ 0, for rrup  4 km,
⎪
⎪
⎪
⎪ 1
⎨ 4 a9 (rrup −4),
⎪ for 4 km < rrup  8 km,
F HW (rrup ) = a9 , for 8 km < rrup  18 km,
⎪
⎪  
⎪
⎪ a9 1− 1 (rrup −18) , for 18 km < rrup  24 km,
⎪
⎪ 7
⎪
⎩ 0, for r > 24 km.
rup

The nonlinear soil response term F 5 (PGArock ) in equation (2.4.2) is modelled by


F 5 (PGArock ) = a10 + a11 ln PGArock +c5 , (2.4.6)

where PGArock is the expected value of peak ground acceleration (PGA) on rock, as
predicted by the expected prediction equation (2.4.2) with S = 0.
The total standard deviation σ in equation (2.4.1) of the GMPE, including intra-
event and inter-event aleatory uncertainties, is magnitude-dependent modelled by
⎧
⎨ B5 ,
⎪ for M  5.0,
σtotal (M ) = B5 − B6 (M −5), for 5.0 < M < 7.0, (2.4.7)
⎪
⎩
B5 − 2 B6 , for M  7.0.
In equation (2.4.7), the total standard deviation σtotal (M ) for the horizontal ground
motions is determined by calculating the geometric average of two horizontal com-
ponents. The total standard deviation can also be obtained for arbitrary horizontal
 2.4 probabilistic ground-motion parameters 47

components of ground motions by inflating σtotal (M ) using a functional fit developed

by Baker and Cornell (2006c). Consequently, the standard deviation for arbitrary
horizontal components, in terms of magnitude M and vibration period T, is given by

2
σtotal, arb (M , T) = σtotal (M ) . (2.4.8)
1.78 − 0.039 lnT
In equations (2.4.2) to (2.4.7), a1 , . . . , a6 , a9 , . . . , a13 , c4 , c5 , n, B5 , and B6 are period-
dependent parameters of regression analysis.

GMPEs by Atkinson and Boore

Atkinson and Boore (2006) developed GMPEs for horizontal ground motions at hard
rock and soil sites in eastern North America (ENA). The GMPEs are obtained by
regression analysis of a database of simulated ground motions using a stochastic finite-
fault model. The GMPE model incorporates information obtained from new ENA
seismographic data, including tridirectional broadband ground motions that provide
new information on ENA source and path effects. The functional form of the GMPE is
given by

log10 SA = c1 + c2 M + c3 M 2 + (c4 +c5 M ) · F 1 + (c6 +c7 M ) · F 2

+ (c8 +c9 M ) · F 0 + c10 rcd + S, (2.4.9)
    R 
10
F0 = max log10 r , 0 , F1 = min log10 rcd , log10 70 , F2 = max log10 cd , 0 ,
cd 140
where SA is spectral acceleration at a specific vibration period, M is the moment
magnitude, and rcd is the closest distance to the fault rupture.
To account for soil amplification in both linear and nonlinear ranges, a piecewise
soil amplification factor S in equation (2.4.9), as a function of shear wave velocity in
the upper 30 m (V30 ), is modelled as
⎧
⎪
⎪ 0, for hard-rock sites,
⎪
⎪   
⎪
⎨ V 60 2
log10 exp Blin · ln 30 + Bnl · ln , for pgaBC  60 cm/s ,
. S= 760 100 (2.4.10)
⎪
⎪
⎪
⎪   
⎪
⎩ log exp B · ln V30 + B · ln pgaBC , for pgaBC > 60 cm/s2 ,
10 lin nl
760 100
where pgaBC is the predicted value of PGA for V30 = 760 m/s, Blin is the linear factor
of soil amplification, and the nonlinear soil amplification factor Bnl is given by
⎧
⎪
⎪ B , for V30  180 m/s,
⎪ 1
⎪
⎪
⎪ lnV30 − ln300
⎪
⎪ (B1 −B2 ) + B2 , for 180 m/s < V30  300 m/s,
⎨
ln180 − ln300
Bnl = (2.4.11)
⎪
⎪ lnV − ln760
⎪B
⎪
30
, for 300 m/s < V30  760 m/s,
⎪
⎪ 2
−
⎪
⎪ ln300 ln760
⎩
0, for V30 > 760 m/s.
48

In equation (2.4.9), the spectral acceleration is predicted for an event with fault
stress of 140 bars. For stress values other than 140 bars (within the tested range from
35 to 560 bars), the spectral acceleration log10 SA, adj is obtained by adjusting log10 SA
in equation (2.4.9) as

log10 SA, adj = log10 SA + SF1 · log10 SF2 , (2.4.12a)

log10 Stress − log10 140
SF1 = , (2.4.12b)
log10 2
   
 · max (M −M1 ), 0
log10 SF2 = min +0.05, 0.05+ . (2.4.12c)
Mh −M1

The standard deviations of log10 SA for all vibration periods are 0.3. In equations
(2.4.9) to (2.4.12), c1 , . . . , c10 , Blin , B1 , B2 , , M1 , and Mh are period dependent param-
eters of regression analysis.

GMPEs by Boore and Atkinson

Boore and Atkinson (2008) developed GMPEs for the geometric average values of two
horizontal components of ground motions as a function of earthquake magnitude,
source–site distance, local average shear wave velocity, and fault type. The geometric
average is determined from the 50th percentile value of the geometric mean values
computed for all non-redundant rotation angles, which is approximately identical to
the geometric mean in most cases (Boore et al., 2006).
The GMPEs are derived by regression analysis of an extensive strong-motion database
compiled by the Next Generation Attenuation (NGA) project of the Pacific Earthquake
Engineering Research (PEER) Center. For periods less than 1 s, the analysis uses 1574
recorded ground motions from 58 main-shocks in the distance range from 0 km to 400
km (the number of available data decreases with increasing vibration period).
The functional form F(m, r, Tj , θ) in equation (2.4.1) of the GMPE is given by

ln SA = FM (M ) + FD (rJB , M ) + FS (V30 , rJB , M ), (2.4.13)

where FM , FD , and FS represent the magnitude scaling, distance, and site amplifi-
cation functions, respectively. M is the moment magnitude, rJB is the Joyner-Boore
distance (defined as the closest distance to the surface projection of the fault, which
is approximately equal to the epicentral distance for events of M < 6), and V30 is the
average shear wave velocity from the surface to a depth of 30 m.
The magnitude scaling function FM in equation (2.4.13) is given by

e1 · U+e2 · SS+e3 · NS+e4 · RS+e5 (M −MH )+e6 (M −MH )2, for M  MH ,
FM (M ) =
e1 · U + e2 · SS+e3 · NS+e4 · RS+e7 (M −MH ), for M > MH ,
(2.4.14)
2.4 probabilistic ground-motion parameters 49

where U, SS, NS, and RS are coefficients (1 for unspecified, strike-slip, normal-
slip, and reverse-slip fault type, respectively, and 0 otherwise) and MH , the “hinge
magnitude” for the shape of the magnitude scaling, is a coefficient to be determined
during the analysis.
The distance function FD in equation (2.4.13) is given by
 
  
FD (rJB , M ) = c1 + c2 (M −4.5) ln rJB 2 + H2 + c
3 r 2 + H2 − 1 .
JB (2.4.15)

The site amplification function FS in equation (2.4.13) is given by

FS (V30 , rJB , M ) = FLIN + FNL , (2.4.16)

where the linear term FLIN is

FLIN = Blin ( lnV30 − ln760), (2.4.17)

and the nonlinear term FNL is

⎧  
⎪ 0.06
B
⎪ nl ln , for pga4nl  0.03g,
⎪
⎪ 0.1
⎪
⎪     2   pga4nl 3
⎪
⎪
⎨ B ln 0.06 + c ln pga4nl + D ln ,
nl
FNL = 0.1 0.03 0.03 (2.4.18)
⎪
⎪
⎪
⎪ for 0.03g < pga4nl  0.09g,
⎪
⎪  pga4nl 
⎪
⎪
⎩ B ln , for pga4nl > 0.09g,
nl 0.1
1  
c= 3Bnl ( ln0.09− ln0.06) − Bnl ( ln0.09− ln0.03) ,
( ln0.09− ln0.03) 2

−1  
D= 2Bnl ( ln0.09− ln0.06) − Bnl ( ln0.09− ln0.03) ,
( ln0.09− ln0.03)3
⎧
⎪
⎪ B , for V30  180 m/s,
⎪ 1
⎪
⎪
⎪
⎪
⎪ (B1 −B2 )( lnV30 − ln300)
⎪
⎨ + B2 , for 180 m/s < V30  300 m/s,
ln180− ln300
Bnl =
⎪ B ( lnV30 − ln760)
⎪
⎪
⎪ 2
for 300 m/s < V30 < 760 m/s,
⎪
⎪ ln300− ln760 ,
⎪
⎪
⎪
⎩
0, for V30  760 m/s.

The total standard deviation σ for ln SA in equation (2.4.13), including intra-event

and inter-event aleatory uncertainties, has been tabulated for fault type specified and
unspecified (Boore and Atkinson, 2008).
In equation (2.4.18), pga4nl is the predicted PGA for V30 = 760 m/s using equation
(2.4.13) with the site amplification function FS = 0. In equations (2.4.14) to (2.4.18),
e1 , . . . , e7 , c1 , . . . , c3 , B1 , B2 , MH , H, and Blin are period-dependent parameters of
regression analysis.
50

As can be seen in these three examples of GMPE, various functional forms have
been used to characterize the ground-motion mechanisms based on the availability of
recorded ground motions and the corresponding geographical locations, over which
the recorded ground motions were collected. Historically, the functional forms of
GMPE used in regression analysis have been revised significantly with the increase of
recorded ground motions since 1960s (Douglas, 2011). Most GMPEs are updated in the
literature every 3 to 5 years or shortly after the occurrences of major earthquakes in
well-instrumented regions (Kramer, 1996).
In this situation, it is not feasible to have the “correct” functional form of a GMPE
with 100 % confidence. In other words, each GMPE contains epistemic uncertainty to a
varying extent. It is therefore important to include multiple GMPEs in seismic hazard
analysis or risk analysis with weights (McGuire, 2004). One useful procedure is
1. Determine the magnitude-distance range most critical to seismic hazard analysis.
2. Collect a set of recorded ground motions relevant to the site of interest.
3. Compare the selected GMPEs with the ground-motion data in the magnitude-
distance range. GMPEs that fit the ground-motion data better are given high
weights, and GMPEs that fit less well are given low weights. The weight can be
calculated proportionally to the inverse of the residual of the data around each
GMPE (McGuire, 2004).
On the other hand, when applying any GMPE, it is very important to ensure that pa-
rameters, such as spectral acceleration, magnitude, and distance, are defined and used
consistently with the application. For example, the horizontal spectral acceleration
of a GMPE can be predicted for the geometric average of two horizontal components,
arbitrary horizontal component, or the 50th percentile values of the geometric means of
two horizontal components computed for all non-redundant rotation angles. Typically,
GMPE for arbitrary horizontal component should be used for horizontal ground mo-
tion when structural responses are predicted by engineers (Baker and Cornell, 2006c).

2.4.3 Correlation of Ground-Motion Parameters

GMPEs describe the probability distributions of spectral accelerations at individual
vibration periods, given a set of predictor variables, such as magnitude and distance.
However, the statistical correlations between spectral accelerations at multiple periods
are not addressed by GMPEs. Because the correlations of spectral accelerations at mul-
tiple periods are required in determining the joint distribution functions of spectral
accelerations in vector-valued PSHA, the Pearson product-moment correlation coeffi-
cients of spectral accelerations have been obtained empirically based on a large number
of recorded ground motions (Baker and Cornell, 2006a; Baker and Jayaram, 2008).
In equation (2.4.1), only ε(Tj ) and ln SA(Tj ) are random variables, which follow
standard normal distribution and normal distribution conditional on predictor vari-
2.4 probabilistic ground-motion parameters 51

ables (m, r, and θ), respectively. Hence, the determination of correlation coefficients
of spectral accelerations is equivalent to that of correlation coefficients of ε.
By rearranging equation (2.4.1), ε(Tj ) can be expressed as
 
ε(Tj ) = σ ln SA(Tj ) − F(m, r, Tj , θ ) .
1
(2.4.19)

For a given sample (a recorded ground motion with known SA(Tj ), m, r, and θ), ε(Tj )
in equation (2.4.19) is a known number. Equation (2.4.19) then eliminates the impact
of predictor variables (m, r, and θ ) on the variability of ε(Tj ), i.e., the variability of
ε(Tj ) is independent of the predictor variables. In practice, it is thus convenient to
study ε(Tj ), instead of SA(Tj ), in statistical and regression analysis (Baker and Cornell,
2006a). It is noted that, because the normalized residual ε(Tj ) is used in the analysis,
any outcomes are in principle dependent on the GMPEs selected.
The Pearson product-moment correlation coefficient between ε(Tu ) and ε(Tv ), at
vibration periods Tu and Tv , is given by


N   
ε(I) (Tu )−ε(Tu ) ε(I) (Tv )−ε(Tv )
I=1
ρε(T = , (2.4.20)
u ), ε(Tv )

N   
N  
ε(I) (Tu )−ε(Tu ) 2 ε(I) (Tv )−ε(Tv ) 2
I=1 I=1

where ε (I) (Tu ) is the number of standard deviations departing from F(m, r, Tu , θ) of
the Ith recorded ground motion at period Tu using equation (2.4.19), ε(Tu ) is the
mean value of ε of all N recorded ground motions at Tu , and N is the total number of
recorded ground motions used in the statistical and regression analysis. Using equation
(2.4.20), a symmetric positive semi-definite matrix of the correlation coefficients can
be determined from all the combinations of ε at different periods.
Given a set of predictor variables (m, r, and θ ) in equation (2.4.19), by substituting
equation (2.4.19) into equation (2.4.20), the correlation coefficient between spectral
accelerations ln SA(Tu ) and ln SA(Tv ) at any two periods Tu and Tv is obtained as
N 
  
ln SA (Tu )− ln SA(Tu )
(I)
ln SA (Tv )− ln SA(Tv )
(I)
I=1
ρ ln S ln SA(Tv )
= ,
A(Tu ), N  2 
N  2

ln S
(I)
ln SA(Tu ) ln S
(I)
A (Tu )− A (Tv )− ln SA(Tv )
I=1 I=1
(2.4.21)
where ln S
(I)
A (Tu )
is the natural logarithmic spectral acceleration of the Ith recorded
ground motion at period Tu , and ln SA(Tu ) is the mean value of logarithmic spectral
acceleration of all N recorded ground motions at Tu . Hence, the correlation coefficients
of ε and those of spectral accelerations conditional on a given scenario earthquake
(known predictor variables m, r, and θ) are identical.
52

Using equations (2.4.19) and (2.4.20), a number of models of correlation coefficients

of spectral accelerations (abbreviated as spectral correlation models) have been devel-
oped based on different databases of recorded ground motions (Inoue and Cornell,
1990; Baker and Cornell, 2006a; Baker and Jayaram, 2008). Two recent correlation
models are described below.

Spectral Correlation Model by Baker and Cornell

Baker and Cornell (2006a) developed a spectral correlation model empirically based
on 267 sets of recorded ground motions. An approximate analytical equation for the
correlation coefficients between horizontal spectral accelerations at any two vibration
periods is determined using nonlinear regression. The spectral correlation model is
equally valid for both arbitrary component and geometric mean of two horizontal
components of spectral accelerations. The valid range of vibration period of this
correlation model is between 0.05 s and 5 s. It is observed that the resulting spectral
correlations do not vary significantly when different GMPEs are used, which suggests
that the correlation model are applicable regardless of the GMPEs chosen.
The correlation coefficient between the horizontal spectral accelerations ln SA(Tu )
and at any two periods Tu and Tv (Tu  Tv ) is estimated as
  
π   
− 0.359 + 0.163 i (Tu < 0.189) · ln
Tu T
ρ ln S (T ), ln S (T ) = 1 − cos · ln v ,
A u A v 2 0.189 Tu
(2.4.22)
where i (Tu < 0.189) is an indicator function equal to 1 if Tu < 0.189 s and equal to 0
otherwise. The matrix of correlation coefficients of spectral accelerations of this model
is shown as contours in Figure 2.29.
As can be seen in Figure 2.29, the spectral correlations decrease with increasing
separation of vibration periods when the periods are larger than around 0.2 s, which
cover the range of period of general engineering interest.

Spectral Correlation Model by Baker and Jayaram

Baker and Jayaram (2008) developed a spectral correlation model empirically using
the NGA ground-motion library and new NGA GMPEs (Abrahamson and Silva, 2008;
Boore and Atkinson, 2008; Campbell and Bozorgnia, 2008; Chiou and Youngs, 2008).
The spectral correlation model is valid for a variety of definitions of horizontal spectral
acceleration, including spectral acceleration of arbitrary component, the geometric
average of spectral accelerations from two orthogonal horizontal components, and the
50th percentile value of the geometric means computed for all non-redundant rotation
angles. The valid range of vibration period of this model is between 0.01 s and 10 s.
The correlation model of spectral correlations has the following properties:
1. It is not sensitive to the choice of accompanying GMPE models.
2.4 probabilistic ground-motion parameters 53

5 0.0
0.1
0.2
0.3
1

Period (sec)
0.4
0.5
0.6
0.7
0.8
0.1
0.9
0.05 1.0
0.05 0.1 1 5
Period (sec)
Figure 2.29 Correlation coefficients of spectral accelerations by Baker and Cornell (2006).

10 0.0
0.1
0.2

1 0.3
Period (sec)

0.4
0.5
0.6
0.1
0.7
0.8
0.9
0.01 1.0
0.01 0.1 1 10
Period (sec)
Figure 2.30 Correlation coefficients of spectral accelerations by Baker and Jayaram (2008).

2. Intra-event (ground motions in the same earthquake event) error, inter-event

(ground motions between different earthquake events) error, and total error in
the regression analysis of the GMPEs all exhibit similar correlation structure.
The correlation coefficient between the horizontal spectral accelerations ln SA(Tu ) and
ln SA(Tv ) at any two periods Tu and Tv (Tu  Tv ) is estimated as
⎧
⎪
⎪ C2 , if Tv < 0.109,
⎪
⎪
⎨C , else if Tu > 0.109,
1
ρ ln S (T ), ln S (T ) = (2.4.23)
A u A v ⎪
⎪ min(C2 , C4 ), else if Tv < 0.2,
⎪
⎪
⎩
C4 , else,

where
  
π Tv
C1 = 1 − cos − 0.366 ln  ,
2 max Tu , 0.109
54

⎧   
⎨ 1 − 0.105 1 − 1 Tv −Tu
, if Tv < 0.2,
C2 = 1+ exp(100Tv −5) Tv −0.0099
⎩
0, otherwise,

C2 , if Tv < 0.109,
C3 =
C1 , otherwise,
  π T 
 u
C4 = C1 − 0.5 C3 −C3 1 + cos .
0.109

The matrix of correlation coefficients of spectral accelerations of this model is shown

as contours in Figure 2.30.

2.4.4 Probability Distribution of Ground-Motion Parameters

In Sections 2.4.2 and 2.4.3, the GMPEs and correlation coefficients of spectral accelera-
tions are described. In this section, they are used to determine probability distributions
of spectral accelerations, which are the core components in probabilistic seismic hazard
analysis (PSHA).
For a given set of GMPEs for spectral accelerations at different vibration periods as
in equation (2.4.1), based on a large number of recorded ground motions, the same
number of samples is obtained for ε(Tj ) at each period Tj . Using normal Q-Q plots, in
which “Q” stands for quartile, univariate normality of the samples of ε(Tj ) (including
inter-events and intra-events) at individual period Tj are tested. The normal Q-Q
plots show strong linearity, indicating that the residuals ε(Tj ) are well represented
by a standard normal distribution marginally. Thus, spectral acceleration SA(Tj ) at
individual period Tj follows lognormal distribution marginally, for a given scenario
earthquake in terms of m, r, θ (Jayaram and Baker, 2008),
 lns − μ  
   j ln S A(Tj )
 m,r
P SA(Tj ) > sj  m, r = 1 −
, (2.4.24)
σ ln S 
m
A(Tj )

where
(·) is the standard normal distribution function, sj is the threshold value, and
the mean and standard deviation of ln SA(Tj ) are

μ ln S 
 m,r = F(m, r, Tj , θ ), σ ln S 
m = σ, (2.4.25)
A(Tj ) A(Tj )

where F(m, r, Tj , θ) and σ are obtained from the GMPE in equation (2.4.1). It is
noted that the marginal lognormal distribution in equation (2.4.24) is conditional
on a scenario earthquake, which may be represented by magnitude m, source–site
distance r, and other predictor variables θ. Only m and r are expressed explicitly
in equation (2.4.24) to represent a scenario earthquake, because m and r are treated
as random variables in the PSHA in most cases. In this context, m and r are used
2.4 probabilistic ground-motion parameters 55

to represent a scenario earthquake explicitly, while other predictor variables θ are

considered implicitly.
Based on the samples of ε(Tj ) (including inter-events and intra-events) at different
periods Tj (j = 1, 2, . . . , K), using Henze-Zirkler test, Mardia’s tests of skewness and
kurtosis, it is shown that the residuals ε(Tj ) at different periods Tj follow a multivariate
standard normal distribution. Consequently, spectral accelerations SA(Tj ) at differ-
ent periods Tj conditional on a scenario earthquake follow multivariate lognormal
distribution (Jayaram and Baker, 2008)
  
P SA(T1 ) > s1 , . . . , SA(TK ) > sK  m, r
 ∞  ∞  
= ... F SA(T S s1 , . . . , sK  m, r ds1 · · · dsK , (2.4.26)
1 ), ..., A(TK )
sK s1

where s1 , . . . , sK are threshold values of spectral accelerations at different periods and

 
F SA(T ), ..., SA(T ) s1 , . . . , sK  m, r is the PDF of multivariate lognormal distribution of K
1 K
spectral accelerations conditional on m and r.
In the conditional multivariate lognormal distribution of spectral accelerations in
equation (2.4.26), the mean and standard deviation values of ln SA(Tj ) in terms of m
and r are obtained from the GMPE for each vibration period using equation (2.4.25).
The correlation coefficient between the natural logarithmic spectral accelerations at any
two periods given a scenario earthquake (m and r), as in equation (2.4.21), has been
empirically obtained as discussed in Section 2.4.3.
The marginal and joint probability distributions of spectral accelerations governed
by equations (2.4.24) and (2.4.26), respectively, will be used in the PSHA.

❧ ❧
In this chapter, the fundamentals of engineering seismology are presented.
❧ Tectonic plates drift due to the convective motion of magma underneath the earth’s
crust. When two tectonic plates move towards each other, one plate subducts under
the other. The plates grind against each other, causing stresses to build up at the
boundary. When the shear stresses accumulated reach the shear strength of the
rock at the interface, the rock fails. If the rock is strong and brittle, rupture of the
rock releases the stored energy explosively, resulting in earthquake. The theory of
tectonic plates and the theory of elastic rebound theory explain most of the major
earthquakes in the world.
❧ When earthquake occurs, body waves (P-waves and S-waves) and surface waves
(Rayleigh waves and Love waves) are produced. When seismic waves propagate,
they refract and reflect at boundaries between different layers of the earth and
convert to waves of the other types. S-waves shake the ground surface vertically
and horizontally and are particularly damaging to structures. Love waves shake the
56

ground surface horizontally and are particularly damaging to the foundations of

structures.
❧ The size of an earthquake is measured by its magnitude. Most definitions of earth-
quake magnitude are based on amplitudes of seismic waves. The moment magni-
tude Mw or M is independent of seismic wave (ground-motion) characteristics and
is not subjected to magnitude saturation; it has been widely used in establishing
ground-motion prediction equations (GMPEs).
❧ When an earthquake occurs due to an explosive release of lock at the boundary of
tectonic plates, the upward spring motion of the top plate displaces a large body
of water, producing tsunami waves. The speed of water waves is proportional to
square-root of the sea depth. When tsunami waves approach the coast, their speeds
decrease and their heights increase dramatically, capable of reaching tens of meters.
❧ Strong ground motions are measured in terms of PGA, PGV, and PGD. Arias in-
tensity, which is the sum of squared acceleration values, is a more reliable statistical
measure of the amplitude of a strong motion record. The time span between 5 % and
75 % (or 95 %) of the Arias intensity is usually defined as a strong-motion duration.
❧ Fourier amplitude spectra (FAS) and power spectral density (PSD) functions of
earthquake time-histories have been used to characterize the frequency contents of
earthquake ground motions. Ground response spectra (GRS) are used extensively
in seismic analysis and design and will be studied in detail in Chapter 4.
❧ GMPEs provide empirical relations between ground motion intensity measure (e.g.,
PGA or spectral acceleration at individual vibration frequency) and earthquake
magnitude, source–site distance, and possibly other predictive variables, based
on statistical analysis of real earthquake ground-motion records and sometimes
simulated ground-motion time-histories. Statistical correlation between spectral
accelerations at two different frequencies are also obtained from statistical analysis.
Spectral accelerations at different frequencies conditional on a scenario earthquake
(given earthquake magnitude and source–site distance) follow multivariate lognor-
mal distribution. These results are used in probabilistic seismic hazard analysis
(PSHA) in Chapter 5.