Earthquake Spectra Volume 29

Application of Supplemental Damping
Characteristics to Response Spectrum

Analyses of Nonproportionally Damped
Multistory Asymmetric-Plan Buildings
Jui-Liang Lina) and Keh-Chyuan Tsaib)
For practicing engineers, knowledge of the characteristics of supplemental

damping in buildings is essential to understand the resultant effects of added
damping. Characterizing the overall system parameters representing the amount
and the plan-wise distribution of the supplemental damping in a single-story
asymmetrical building is straightforward. However, this becomes a difficult
task for multistory asymmetrical buildings. For this reason, this paper first
develops the effective one-story building (EOSB), which retains the character-
istics of the two dominant vibration modes of the original nonproportionally
damped multistory asymmetrical building. By using the EOSB, it becomes
convenient to characterize the supplemental damping in the original multistory
asymmetrical building. The effectiveness of this approach is verified by using
three numerical examples, which include one one-story, three eight-story, and
one 20-story asymmetrical buildings. Next, the relationships between the roof
displacements of the original building and those of the EOSB are established.
This enables the application of the response spectra constructed from the
EOSBs to estimate the peak roof displacements of the original multistory
asymmetrical building. [DOI: 10.1193/1.4000105]
INTRODUCTION
Various kinds of dampers have been developed for earthquake engineering in order to
suppress structural responses due to seismic loads. Therefore, in order to conduct the
response spectrum analysis on buildings with viscous dampers as an example, model build-
ing codes (ICC 1997, ICC 2000, FEMA 1997, FEMA 2000) stipulate the equivalent damping
ratio arising from the supplemental dampers. Obviously, both the amount of supplemental
damping and the plan-wise placement of the viscous dampers, such as the dispersion of the
viscous dampers and the distance between the center of supplemental damping (CSD) and the
center of mass (CM), influence the seismic responses of asymmetric-plan buildings. In fact,
the damping effects of the viscous dampers on the seismic responses of elastic single-story
asymmetric-plan buildings have been extensively studied (Goel 1998, Lin and Chopra 2001).
The reasons for the effects of the plan-wise distribution of viscous dampers on the seismic
a)
National Center for Research on Earthquake Engineering, 200, Section 3, Xinhai Rd., Taipei 106, Taiwan,
R.O.C. E-mail: jllin@ncree.narl.org.tw
b)
Department of Civil Engineering, National Taiwan University, Taipei, Taiwan, R.O.C. E-mail: kctsai@ntu.
edu.tw
207
Earthquake Spectra, Volume 29, No. 1, pages 207–232, February 2013; © 2013, Earthquake Engineering Research Institute
208 J. L. LIN AND K.-C. TSAI
responses of elastic single-story asymmetric-plan buildings have also been discussed

(Lin and Chopra 2001, Goel 2000). Although the complex eigenvalue analyses provide
the exact information about the modal properties of the nonproportionally damped buildings,
the overall system parameters representing the amount and the plan-wise distribution of the
added viscous dampers cannot be obtained from conducting this type of analysis. The overall
system parameters are useful in developing a fundamental understanding of the resultant
effects of the added viscous dampers. Hence, Goel (1998) characterized the supplemental
damping in the single-story asymmetric-plan buildings with linear viscous dampers by
using three overall system parameters: (1) the supplemental damping ratio ξsd ; (2) the
normalized supplemental damping eccentricity ēsd ; and (3) the normalized supplemental
damping radius of gyration ρ̄sd . Goel (1998) also conducted a parametric study on these
three parameters and provided insights into the effects of the amount, the resultant location,
and the dispersion of the added dampers. This gives practicing engineers an effective physical
interpretation of the total installed viscous dampers in terms of these three overall system
parameters for a single-story asymmetrical building. Goel (1998) concluded that further
investigation of multistory asymmetrical buildings with supplemental viscous dampers is
needed. Nevertheless, to the best of the authors’ knowledge, there is not yet a way to
characterize the overall system parameters representing the resultant effects of viscous
dampers installed in multistory asymmetrical buildings.
The response spectra constructed from the nonproportionally damped single-story
asymmetrical buildings have been extensively used in investigating the effects of the supple-
mental damping on their seismic responses (Goel 1998, Lin and Chopra 2001). The abscissa
and the ordinate of these response spectra, respectively, represent the uncoupled vibration
periods—that is, the purely translational period—and the peak responses of the single-story
asymmetrical buildings. However, there is no quantitative relationship between the seismic
responses of these single-story buildings and those of the general multistory nonproportion-
ally damped buildings. Thus, even though many researchers have investigated this type of
response spectra (Goel 1997, Goel 2001, Goel and Booker 2001, Riddell and Santa-Maria
1999, Stathopoulos and Anagnostopoulos 2003, Tso and Sadek 1985), it remains difficult to
apply this type of response spectra to estimate the seismic responses of the general multistory
nonproportionally damped buildings.
It has been found that each vibration mode of the multistory one-way asymmetrical build-
ing can be effectively represented by using a 2-degree-of-freedom (2DOF) modal system
(Lin and Tsai 2007a, Lin and Tsai 2007b). Because the characterization of the supplemental
damping in the single-story asymmetrical buildings is straightforward (Goel 1998), this
study first develops the effective one-story building (EOSB) associated with the original
multistory asymmetrical building. The ESOB uses the subspace spanned by the first
undamped translational-dominant mode and the first rotational-dominant mode, accounting
for damping coupling. One can then use the EOSB to estimate peak response. The concept
and the application of the EOSB are schematically shown in Figure 1. The EOSB is con-
structed from the two 2DOF modal systems, which represent the first translational-dominant
vibration mode and the first rotational-dominant vibration mode of the original building.
By means of the EOSB, it becomes convenient to characterize the supplemental damping
in the original multistory asymmetrical building. In addition, the relationships between
the roof displacement responses of the original building and those of the EOSB are
APPLICATION OF SUPPLEMENTAL DAMPING CHARACTERISTICS TO RESPONSE SPECTRUM ANALYSES 209
Figure 1. The sketch of the concept and the application of the EOSB.
established in this study. This enables the application of the response spectra constructed
from the EOSBs to estimate the peak roof responses of the original multistory buildings.
CHARACTERIZATION OF THE SUPPLEMENTAL DAMPING

IN MULTISTORY ASYMMETRICAL BUILDINGS
The existing model building codes only consider the first vibration mode’s contribution
in computing the equivalent damping ratio. This paper uses the first translational-dominant
and the first rotational-dominant vibration modes of the multistory asymmetrical building to
construct the EOSB. Then the EOSB is used as a vehicle to identify the overall system
normalized damping eccentricity, damping radius of gyration, and equivalent damping
ratio of the original multistory building.
THE EFFECTIVE ONE-STORY BUILDING

The two horizontal axes of the coordinate system used in this paper are the X- and the
Z- axes. The direction of the Y-axis is opposite to the direction of gravity. For the sake
of simplicity, the original N-story asymmetrical buildings are only symmetric about
the X-axis and the ground motion is applied along the Z-axis. The subscripts z and θ
used in the following content refer to quantities related to the Z-translational and the
Y-rotational components, respectively. The first translational-dominant and the first
rotational-dominant vibration modes of the original building are chosen as the two tar-
geted vibration modes to form the EOSB. That is to say, the two 2DOF modal systems
representing the two vibration modes of the EOSB are the same as those representing the
two targeted vibration modes of the original building. The mass, damping, and stiffness
matrices of the EOSB are respectively denoted as
! ! " ! ! " ! ! "
m 0 czz c!zθ kzz k!zθ
M ¼
!
; C ¼ !
!
; K ¼ !
!
(1)
0 I! cθz c!θθ kθz k !θθ
EQ-TARGET;temp:intralink-;e1;41;603
# $
and its mode shapes are denoted as ϕ!zi ϕ!θi T , i ¼ 1; 2. The superscript * used in Equa-
tion 1 denotes the quantities belonging to the EOSB in order to differentiate those notations
from the commonly used notations for the original building. Because the EOSB and the
original multistory building have two 2DOF modal systems in common, the following equal-
ities should exist:
ϕ!zi m! ϕ!zi ∶ϕ!θi I ! ϕ!θi ¼ mi ∶I i ; i ¼ 1; 2 (2)
where mi , I i , i ¼ 1; 2 are the properties of the two 2DOF modal systems representing the two
targeted vibration modes of the original multistory building and computed as follows (Lin
and Tsai 2007a):
mi ¼ φTzi mφzi ; I i ¼ φTθi Iφθi ; i ¼ 1; 2 (3)
and m and I are the N × N mass and mass moment of inertia matrices of the original N-story
one-way asymmetrical building; φzi and φθi are the N × 1 subvectors of the ith undamped
mode shape of the original building. Note that the subscript i equal to 1 and 2 (Equation 2) of
the quantities related to the original building represent the lower and the higher vibration
modes, respectively, among the two targeted vibration modes. On the other hand, the sub-
script i equal to 1 and 2 (Equation 2) of the quantities related to the EOSB represent the first
and the second vibration modes, respectively, of the EOSB. Equation 2 indicates that
sffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffi
ϕ!z1 m1 I ! ϕ!z2 m2 I !
¼ # ; ¼ # (4)
ϕ!θ1 I 1 m! ϕ!θ2 I 2 m!
From Equation 4, we have the mode shape matrix of the EOSB as:
2 3
! ! !
" 1 ffi
sffiffiffiffiffiffiffiffiffi sαffiffiffiffiffiffiffiffiffiffi
ϕz1 ϕz2 I m! I m! 5
Φ! ¼ ¼4 (5)
ϕ!θ1 ϕ!θ2 s1 1 ! αs2 2 !
m1 I m2 I
where α is an arbitrary constant, and s1 and s2 may equal 1 or −1. The transformed mass
matrix of the EOSB is computed as
2 3
sffiffiffiffiffiffiffiffiffiffiffi
6 m I1
! I 1I 2 7
6 m! þ αm! þ s1 s2 αm! 7
6 m1 m1 m2 7
M ¼Φ M Φ ¼6
! !T ! ! 6 s ffiffiffiffiffiffiffiffiffiffiffi 7 (6)
7
6 ! I I
1 2 m !I
2 7
4αm þ s1 s2 αm !
αm þ α
2 ! 2 5
m1 m2 m2
Since the off-diagonal terms of the transformed mass matrix (Equation 6) should be equal
to zero, it yields
EQ-TARGET;temp:intralink-;e7;62;615 s1 s2 ¼ −1; I 1 I 2 ¼ m1 m2 (7)
and Equation 6 becomes
2 I1 3
1þ 0
6 m1 & '7
M! ¼ m ! 4
I2 5 (8)
2
0 α 1þ
m2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Since m1 þ I 1 ¼ m2 þ I 2 ¼ 1 (Lin and Tsai 2009) and choose α ¼ m2 ∕m1 , Equation 8
can be further simplified as
! "
m! 1 0
M! ¼ (9)
m1 0 1
and Equation 7 yields m1 ¼ I 2 and m2 ¼ I 1 . Thus, the stiffness matrix of the EOSB is
obtained as
2 3
m1 ω21(þ m2 ω22 ) ( sym: )
K! ¼ ðΦ!T Þ−1 ΛM̄! ðΦ! Þ−1 ¼ m! 4 ! pffiffiffiffiffiffiffiffiffiffiffi 5 (10)
−r m1 m2 s2 ω21 þ s1 ω22 r !2 I 1 ω21 þ I 2 ω22
where
! "
ω21 0
Λ¼ (11)
0 ω22
pffiffiffiffiffiffiffiffiffiffiffiffi
and r ! ¼ I ! ∕m! ; ω1 and ω2 are the undamped vibration frequencies of the two targeted
vibration modes of the original building. The values of s1 and s2 , which may be 1 or −1, are
determined by investigating the two targeted undamped mode shapes of the original build-
ing. For example, if the signs of φzn and φθn, n¼ 1; 2 are the same (schematically shown as
Figure 2a), sn is equal to 1. Conversely, if the signs of φzn and φθn, n¼ 1; 2 are opposite
(schematically shown as Figure 2b), sn is equal to −1. The damping matrix of the EOSB is
assumed to be positively related to the 2DOF modal damping matrix of the lower vibration
mode among the two targeted vibration modes of the original building. That is,
ϕ!z1 c!zz ϕ!z1 cz1
ϕ!z1 c!zz ϕ!z1 ∶ϕ!θ1 c!θθ ϕ!θ1 ∶ϕ!z1 c!zθ ϕ!θ1 ¼ cz1 ∶cθ1 ∶czθ1 ; ¼ (12a)
ϕ!z1 m! ϕ!z1 m1
EQ-TARGET;temp:intralink-;e12a;62;194
where
EQ-TARGET;temp:intralink-;e12b;62;140
cz1 ¼ φTz1 czz φz1 ; cθ1 ¼ φTθ1 cθθ φθ1 ; czθ1 ¼ φTz1 czθ φθ1 (12b)
and czz , cθθ , and czθ are the N × N damping submatrices of the original multistory building.
Equation 12 indicates
Figure 2. The schematic drawing of the mode shapes for cases in which sn equals (a) 1 and (b) −1.
rffiffiffiffiffiffiffiffiffi
ffi
m! I! m! I !
c!zz ¼ cz1 ; c!θθ ¼ cθ1 ; c!zθ ¼ s1 c (13)
m1 I1 m1 I 1 zθ1
Therefore, the damping matrix of the EOSB can be expressed as:

2 3
cz1 s1 r ! czθ1
! ! " 6 pffiffiffiffiffiffiffiffiffiffi 7
czz c!zθ 6 m1 m1 I 1 7
C ¼ !
! ¼m 6 !
! (14)
cθz c!θθ s r c
4 1 zθ1 r !2
cθ1 7 5
pffiffiffiffiffiffiffiffiffiffi
m1 I 1 I1
It is well known that the damping matrix of an arbitrary single-story asymmetric-plan

building with supplemental damping can be expressed in the following format (Goel 1998):
! "
1 esd
C ¼ C̄0 þ Csd ¼ C̄0 þ 2m ω ξsd
! ! ! ! ! !
(15)
esd e2sd þ ρ2sd
pffiffiffiffiffiffiffiffiffiffiffiffiffi
where ω! is the uncoupled vibration frequency equal to k!zz ∕m! and C̄!0 is the inherent
damping matrix. From Equations 1, 10, and 14, the equation of motion of the EOSB becomes
2 c s czθ1 3!
z1
! "! ! " p1ffiffiffiffiffiffiffiffiffi
ffi "
1 0 u¨ z 6 m1 m1 I 1 7 u_ !z
þ 4 s1 czθ1 cθ1 5 r ! u_ !
0 1 r ! u¨ !θ pffiffiffiffiffiffiffiffiffiffi θ
" m1 I 1
I1 # (16a)
m1 ω(21 þ m2 ω22 sym: ! ! " ! "
) uz 1
þ pffiffiffiffiffiffiffiffiffiffiffi ¼− u¨ ðtÞ
− m1 m2 s2 ω21 þ s1 ω22 I 1 ω21 þ I 2 ω22 r ! u !θ 0 g
or in terms of Equation 15, the equation of motion of the EOSB becomes

! "! ! " & ! "'! ! "
1 0 u¨ z 1 sym: u_ z
! ¨! þ C0 þ 2ω ξsd
! !
0 1 r uθ ēsd ēsd þ ρ̄sd
2 2
r ! u_ !θ
" # ! ! " ! " (16b)
m1 ω21 þ m2 ω22 sym: uz
1
þ pffiffiffiffiffiffiffiffiffiffiffi( 2 )
¼ − u¨ ðtÞ
− m1 m2 s2 ω1 þ s1 ω22 I 1 ω21 þ I 2 ω22 r ! u !θ 0 g
where
esd ρsd
ēsd ¼ ; ρ̄sd ¼ (16c)
r! r!
EQ-TARGET;temp:intralink-;e16c;62;627
and C!0 is the inherent damping matrix when the mass matrix is presented as an identity
matrix and the corresponding displacement vector is expressed as ½ u !z r ! u !θ (T . By compar-
ing Equation 16a with Equation 16b, the supplemental damping properties of the EOSB can
be characterized by using the three associated parameters, that is, ξsd , ēsd , and ρ̄sd . Particu-
larly, if the inherent damping matrix C!0 is neglected, the equivalent supplemental damping
ratio, ξsd , the normalized equivalent supplemental damping eccentricity, ēsd , and the nor-
malized equivalent supplemental damping radius of gyration, ρ̄sd , can be expressed as
(Equations 16a and 16b):
cz1 cz1
ξsd ¼ ! ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2m1 ω 2m m1 ω21 þ m2 ω22
rffiffiffiffiffiffi 1
sc m1
ēsd ¼ 1 zθ1
cz1 I1 (17)
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
& 'ffi
m1 c2zθ1
ρ̄sd ¼ c −
I 1 cz1 θ1 cz1
Because the EOSB is associated with the original multistory building in terms of the two
common 2DOF modal systems (Figure 1), the supplemental damping properties of the ori-
ginal building may be characterized by using the parameters ξsd , ēsd , and ρ̄sd of the EOSB. As
noted earlier, Goel (1998) investigated the effects of these parameters on the responses of
general single-story asymmetrical buildings. Thus it is reasonable to expect that the results of
these parametric study are also applicable to the EOSBs.
NUMERICAL VALIDATION
In order to validate the effectiveness of the EOSB on retaining the supplemental damping
properties of the original building, the two 2DOF modal response histories of the EOSB are
first computed and then respectively multiplied by the undamped mode shapes of the two
targeted vibration modes of the original building. These two response histories are mathe-
matically summed up as the estimated total response histories of the original building. The
mentioned two undamped mode shapes are the same as those of the corresponding propor-
tionally damped building, which can be considered as all of the added dampers of the original
multistory building being removed. Thus, if the estimated total response histories are satis-
factory, it implies that the supplemental damping properties of the original multistory build-
ing are effectively retained in the EOSB.
The first example is the single-story example building (Lin and Tsai 2007b) shown in
Figure 3. In the special case where the original building happens to be a single-story
asymmetric-plan building, the proposed EOSB (Equations 10, 14, and 16) should be exactly
the same as the original building. The second example is of three
24 m ðXÞ × 15 m ðZÞ × eight-story buildings. These three eight-story buildings are varia-
tions of the numerical examples used in other studies (Ahlawat and Ramaswamy 2003,
Kan and Chopra 1977). The main purpose of the second example is to verify the applicability
Figure 3. (a) The single-story example building and (b) the floor plan.
of the proposed approach on the three well-known types of asymmetric-plan buildings: tor-
sionally stiff (TS), torsionally similarly stiff (TSS), and torsionally flexible (TF). The third
example is the variation of the SAC 20-story building located in Los Angeles (Krawinkler
2000). The main purpose of the third example is to see the applicability of the proposed
approach to high-rise buildings, which may have significantly higher mode effects. Detailed
properties of these three example buildings can be found in the related literature (Lin and Tsai
2007b, Ahlawat and Ramaswamy 2003, Kan and Chopra 1977, Krawinkler 2000).
Example 1
Table 1a shows the mass, damping, and stiffness matrices of the single-story example
building (Figure 3). The damping coefficient C z of the linear viscous damper installed
along the Z-axis (Figure 3) is equal to 2mωz ξsd , in which ωz is the transverse vibration circular
frequency. The supplemental damping ratio, ξsd , used in this building is equal to 30%.
Table 1b shows the modal properties and the transformed damping matrix, which has
nonzero off-diagonal elements. The properties of the two 2DOF modal systems for this single-
story asymmetrical building are shown in Table 1c. Table 2 shows the properties of the EOSB
computed by using Equations 10, 14, and 16. Note that Equation 16 is the equation of motion
corresponding to the displacement vector defined as ½ u !z r ! u !θ (T . It was found that the prop-
erties shown in Table 2 are the same as those shown in Table 1a, except the latter is 9.45 times
the former and the latter is presented corresponding to the displacement vector defined as
½ u z u θ (T . It indicates that the EOSB is identical to the original building when the original
building happens to be single-story. Therefore this confirms the accuracy of the derivation for
Equations 10 and 14. Moreover, Table 2 shows that the supplemental damping properties of
this example building are ξsd ¼ 0.277, ēsd ¼ 2.16, and ρ̄sd ¼ 0.
Example 2
The translational and rotational stiffness of each floor of the corresponding symmetrical
eight-story building are 4.503 × 105 kN=m and 3.84 × 107 kN-m=rad, respectively. The cen-
ter of rigidity (CR) is at the geometric center of the floor plan and the center of mass (CM) is
6 m away from the CR in the positive direction along the X-axis, yielding an eccentricity ratio
equal to 25%. The 8 × 8 stiffness submatrices (kzz , kz! , and k!! ) of these three example
Table 1. (a) The properties of the single-story example building (units: kN, m, rad, sec).
Damping matrix
Supplemental
Mass matrix Inherent damping damping Stiffness matrix
9.45 0 8.10 3.09 115.30 389.1 4594 5176
0 23.03 3.09 42.02 389.1 1313.3 5176 48547
(b) The mode shapes, vibration frequencies, and transformed damping matrix of the single-
story example building.
! "
ω21 0 Transformed damping
Mode shapes 0 ω22 matrix
−0.319 −0.066 413.49 0 4.383 15.309
0.042 −0.204 0 2180 15.309 67.525
(c) The properties of the two 2DOF modal systems for the single-story example building.
! " ! " ! "
mn 0 czn czθn k zn k zθn
nth mode 0 In cθzn cθn k θzn k θn
0.9589 0 12.5206 −5.2786 466.2 −69.7

1st
0 0.0411 −5.2786 2.4193 −69.7 86.7
0.0411 0 0.5368 5.2786 20 69.7
2nd
0 0.9589 5.2786 56.4307 69.7 2021
Table 2. The properties of the EOSB (units: kN, m, rad, sec).
M! C! K! ξsd ēsd ρ̄sd

1 0 13.06 26.59 486.09 350.69
0.277 2.16 0
0 1 26.59 58.86 350.69 2107.4
buildings are shown in Table 3a. The story mass of the third to the eighth stories of these three
example buildings is 3.456 × 105 kg. The story mass of the first and the second stories is
twice the 3.456 × 105 kg story mass of the third to eighth stories, that is, 6.912 × 105 kg. The
only difference between these three example buildings is the mass moment of inertia of each
story, which is equal to one-third, one, and five times 2.37 × 107 kg-m2 . It was found that the
frequency ratios of these three example buildings, which are defined as the ratio of the rota-
tional frequency to the translational frequency of the corresponding symmetrical building, are
equal to 1.40, 1.07, and 0.71, respectively. Thus, these three example buildings respectively
represent torsionally stiff (TS), torsionally similarly stiff (TSS), and torsionally flexible (TF)
Table 3. (a) The stiffness matrices of the eight-story example buildings (units: kN, m, rad).
kzz ¼ 105 ×
4.50
−4.50 9.01 symm.
0.00 −4.50 9.01
0.00 0.00 −4.50 9.01
0.00 0.00 0.00 −4.50 9.01
0.00 0.00 0.00 0.00 −4.50 9.01
0.00 0.00 0.00 0.00 0.00 −4.50 9.01
0.00 0.00 0.00 0.00 0.00 0.00 −4.50 9.01
kz! ¼ k!z ¼ 105 ×
27.02
−27.02 54.04 symm.
0.00 −27.02 54.04
0.00 0.00 −27.02 54.04
0.00 0.00 0.00 −27.02 54.04
0.00 0.00 0.00 0.00 −27.02 54.04
0.00 0.00 0.00 0.00 0.00 −27.02 54.04
0.00 0.00 0.00 0.00 0.00 0.00 −27.02 54.04
k!! ¼ 105 ×
546.11
−546.11 1092.22 symm.
0.00 −546.11 1092.22
0.00 0.00 −546.11 1092.22
0.00 0.00 0.00 −546.11 1092.22
0.00 0.00 0.00 0.00 −546.11 1092.22
0.00 0.00 0.00 0.00 0.00 −546.11 1092.22
0.00 0.00 0.00 0.00 0.00 0.00 −546.11 1092.22
(b) The supplemental damping matrices of the eight-story example buildings (units: kN, m, rad, sec).
½czz (sd ¼ 103 ×
4.99
−4.99 9.98 symm.
0 −4.99 9.98
0 0 −4.99 9.98
0 0 0 −4.99 9.98
0 0 0 0 −4.99 9.98
0 0 0 0 0 −4.99 9.98
0 0 0 0 0 0 −4.99 9.98
(continued )
Table 3. (continued )
½cz! (sd ¼ ½c!z (sd ¼ 103 ×

79.84
−79.84 159.68 symm.
0 −79.84 159.68
0 0 −79.84 159.68
0 0 0 −79.84 159.68
0 0 0 0 −79.84 159.68
0 0 0 0 0 −79.84 159.68
0 0 0 0 0 0 −79.84 159.68
½c!! (sd ¼ 103 ×
1277
−1277 2555 symm.
0 −1277 2555
0 0 −1277 2555
0 0 0 −1277 2555
0 0 0 0 −1277 2555
0 0 0 0 0 −1277 2555
0 0 0 0 0 0 −1277 2555
(c) The properties of the two 2DOF modal systems representing the two targeted vibration modes of
Buildings TS, TSS, and TF.
! " ! " ! "
Targeted mn 0 czn czθn k zn k zθn
Building mode 0 In cθzn cθn k θzn k θn
0.9287 0 0.6685 0.4455 41.552 14.750

1
0 0.0713 0.4455 0.4386 14.750 16.807
TS
0.0849 0 0.1135 0.4116 7.761 13.615
2
0 0.9151 0.4116 5.6598 13.615 216.910
0.7274 0 0.5273 0.4545 32.545 14.732
1
0 0.2727 0.4545 0.5944 14.732 21.410
TSS
0.2737 0 0.2043 0.4402 12.677 14.286
2
0 0.7263 0.4402 1.5893 14.286 57.258
0.1495 0 0.1098 0.1692 6.695 5.276
1
0 0.8505 0.1692 0.4377 5.276 13.348
TF
0.8487 0 0.6250 0.1604 38.129 5.005
2
0 0.1513 0.1604 0.0864 5.005 2.686
(continued )
(d) The vibration frequencies, effective modal participation mass, and mass moment of inertia of the
two targeted vibration modes of Buildings TS, TSS, and TF.
*X
2N *X
2N
Γ2znM n Γ2zi M i Γ2θnM n Γ2θi M i
Targeted ωn i i Mode No. (dominant
Building mode (rad./sec) (%) (%) motion)
1 5.37 75.3 6.2 1st (translation)
TS
2 15.87 3.9 79.8 3rd (rotation)
1 4.95 58.9 23.5 1st (translation)
TSS
2 9.93 23.6 61.1 2nd (rotation)
1 3.08 12.0 73.0 1st (rotation)
TF
2 7.13 70.5 11.6 2nd (translation)
buildings and are respectively denoted as Building TS, Building TSS, and Building TF. Ray-
leigh damping with a 2% damping ratio for the two targeted vibration modes was assumed for
representing the inherent damping. The CSD is located 10 m away from and on the other side
of the CR, opposite to that of the CM. That is, the CM is 16 m away from the CSD. The added
supplemental damping at each story of these three example buildings is equal to
2mωz ξsd ¼ 2 × 3:456 × 105 ×ð4:503 × 103 =3:456Þ0:5 × 20%=103 ¼ 4990 kN × sec=m. This
results in the 8 × 8 supplemental damping submatrices, that is, ½czz (sd , ½cz! (sd , and ½c!! (sd ,
shown as Table 3b. Table 3c lists the 2DOF modal properties of the two targeted vibration
modes, that is, the first translational-dominant and the first rotational-dominant vibration
modes, for these three example buildings. Table 3d shows the vibration frequencies, the
effective modal participation mass ratio, and the effective modal participation mass moment
of inertia ratio for the two targeted vibration modes. In addition, Table 3d shows the original
building’s mode numbers corresponding to the targeted vibration modes and their dominant
motions. The ground motion record used for the elastic response history analysis was the
1940 El Centro earthquake, which was denoted as LA02 with the scaled peak ground accel-
eration (PGA) being 0.66 g in the SAC steel research project (Krawinkler 2000). The transla-
tional and rotational response histories of the eighth and the fourth floors of Buildings TS,
TSS, and TF are respectively shown in Figures 4 to 6. Figures 4 to 6 indicate that the seismic
responses of these three example buildings belonging to different types of asymmetrical
buildings were all satisfactorily estimated. The properties of the EOSBs corresponding to
these three example buildings are shown in Table 4. Table 4 indicates that the values of
ξsd and ρ̄sd of these three example buildings are almost the same. However, the value of
ēsd significantly decreases as the mass moment of inertia of the example building increases.
Furthermore,pTable
ffiffiffi pffiffiffi that the ratio of the values of ēsd between Buildings TS, TSS, and
4 shows
TF is about 3∶1∶1∕ 5, which is the ratio of the inverse square root of the mass moment of
inertia between these three example buildings. Thus, it implies that the proposed EOSB can
appropriately present the properties of the original buildings.
Figure 4. The translational responses at the (a) eighth and the (b) fourth floors for Building TS;
the rotational responses at the (c) eighth and the (d) fourth floors for Building TS.
Example 3
The 20-story example building (Figure 7), denoted as Building ASY20, is a variation of the
symmetrical 20-story building located in Los Angeles investigated in the SAC steel research
project. Each floor is simulated as a rigid diaphragm with a lumped mass at the CM. The CR is
at the geometric center. The CM was moved away from the CR in the negative X-direction,
Figure 5. The translational responses at the (a) eighth and the (b) fourth floors for Building TSS;
the rotational responses at the (c) eighth and the (d) fourth floors for Building TSS.
yielding an eccentricity ratio of −20%. The flexible side (FS) is defined as one of the two
Z-directional sides closer to the CM. The other side opposite to the FS is the stiff side
(SS). The first translational-dominant and the first rotational-dominant vibration modes are
the first and the second vibration modes of the 20-story example building, respectively. Ray-
leigh damping with 2% damping ratios for these two vibration modes was assumed to represent
Figure 6. The translational responses at the (a) eighth and the (b) fourth floors for Building TF;
the rotational responses at the (c) eighth and the (d) fourth floors for Building TF.
the inherent damping. The same ground motion record as that used in Example 2 was applied to
this 20-story example building in the Z-direction. Elastic response history analysis was per-
formed. There were 20 linear viscous dampers with a damping coefficient equal to
5 × 104 kN × sec=m arbitrarily installed in the Z-directional bays shown as Figure 7b. In
detail, at each of the 1st, 5th, 8th, 10th, 13th, 15th, and 17th floors, there is a viscous damper
Table 4. The properties of the EOSBs for Buildings TS, TSS, and TF (units: kN, m, rad, sec).
Building M! C! K! ξsd ēsd ρ̄sd

1 0 0.7198 1.7312 48.179 62.620
TS 0.034 3.444 0.0
0 1 1.7312 6.1523 62.620 236.410
1 0 0.7250 1.0210 44.777 33.027
TSS 0.035 1.97 0.0
0 1 1.0210 2.1800 33.027 78.355
1 0 0.7344 0.4746 44.553 14.723
TF 0.036 0.88 0.0
0 1 0.4746 0.5147 14.723 15.653
Figure 7. (a) The typical floor plan and (b) the elevation of Building ASY20.
Table 5. The properties of the two 2DOF modal systems representing the two targeted
vibration modes of Building ASY20.
! " ! " ! "
mn 0 czn czθn k zn k zθn
Building Targeted mode 0 In cθzn cθn k θzn k θn
0.9611 0 0.4224 −0.0672 2.7810 −0.2991

1
0 0.0389 −0.0672 0.0662 −0.3000 0.4004
ASY20
0.0387 0 0.0174 0.0685 0.1128 0.2979
2
0 0.9613 0.0685 1.6477 0.2989 9.8889
installed on the FS. For each of the other floors, there is a viscous damper installed on the SS.
The vibration frequencies of the two targeted vibration modes are 1.61 rad./sec and 3.26 rad./
sec. The effective modal participation mass ratios of the two targeted vibration modes are
78.21% and 3.1%. Moreover, the ratios of the effective modal participation mass moment
of inertia for the two targeted vibration modes are 3.17% and 78.5%. Thus, the two targeted
Table 6. The properties of the EOSB for Building ASY20 (units: kN, mm, rad, sec).
M! C! K! ξsd ēsd ρ̄sd

1 0 0.439 −0.348 2.892 −1.547
0.110 −0.899 1.849
0 1 −0.348 1.703 −1.547 10.286
Figure 8. The translational responses at the (a) 20th and the (b) 10th floors for Building ASY20;
the rotational responses at the (c) 20th and the (d) 10th floors for Building ASY20.
Figure 9. The translational responses on the (a) SS and the (b) FS at the 20th floor for Building
ASY20.
vibration modes take about 81% of the total mass and total mass moment of inertia of Building
ASY20. The properties of the two 2DOF modal systems representing the two targeted
vibration modes of Building ASY20 are shown in Table 5. Table 6 shows the properties
of the EOSB for Building ASY20. The translational and the rotational response histories
at the 20th and the 10th floor of Building ASY20 are shown in Figure 8. In general, Figure 8
shows the estimated responses were satisfactory, except the amplitudes of the few cycles
shown in Figure 8c. Furthermore, Figure 9 shows that the estimated translational response
on the two sides at the 20th floor, which includes the influence of the estimated rotational
response (Figure 8c), is still acceptable.
Table 6 shows the value of ρ̄sd for Building ASY20 not equal to zero because the vis-
cous dampers were arbitrarily distributed on the FS as well as the SS. In comparison, the
values of ρ̄sd for the buildings discussed in Examples 1 and 2 are equal to zero because the
viscous dampers were concentrated on one side of the buildings. Furthermore, Table 6
shows that the sign of ēsd for Building ASY20 is negative, which is the opposite of Exam-
ples 1 and 2 (Tables 2 and 4). Because the number of viscous dampers on the SS is greater
than that on the FS for Building ASY20, it may intutively know that the CSD is closer to the
SS. Thus, the relative position between CM, CR, and CSD for Building ASY20 is opposite
to those for the buildings discussed in Examples 1 and 2. This justifies the negative sign of
ēsd shown in Table 6. Note that the damping coefficient of each viscous damper installed in
Building ASY20 is equal to 5 × 104 kN × sec=m, which is about ten times that ð4990 kN ×
sec=mÞ of the viscous damper installed in Buildings TS, TSS, and TF. Moreover, the num-
ber of viscous dampers installed in Building ASY20 is obviously greater than those in
Buildings TS, TSS, and TF. This explains why the value of ξsd for Building ASY20
(0.11, shown in Table 6) is significantly larger than those for Buildings TS, TSS, and TF
(Table 4). In addition to the satisfactory results shown in Figures 4, 5, 6, 8, and 9, the above-
stated comparisons between the values of ρ̄sd , ēsd , and ξsd for the buildings used in Examples
1 to 3 confirm that the supplemental damping properties in the multistory asymmetrical
buildings can be effectively characterized by using those of the EOSB. That is to say,
by means of ξsd , ēsd , and ρ̄sd of the EOSB, engineers can understand and concretely describe
the resultant effects of the added viscous dampers in the multistory asymmetrical buildings.
RESPONSE SPECTRUM ANALYSES TO NONPROPORTIONALLY

DAMPED MULTISTORY ASYMMETRICAL BUILDINGS
THE RELATIONSHIPS BETWEEN THE SEISMIC RESPONSES OF THE EOSB

AND THOSE OF THE ORIGINAL MULTISTORY BUILDING
In order to apply the response spectra constructed from the EOSBs to predict the peak
responses of the original multistory buildings, it is necessary to find the relationships between
the peak responses of the EOSB and those of the original multistory building. From Equa-
tions 5 and 16a and the previous study (Lin and Tsai 2007b), the responses of the EOSB can
be approximately expressed as:
+ !
u z ≈ Γ!z1 ϕ!z1 D!z1 þ Γ!z2 ϕ!z2 D!z2 ¼ m1 D!z1 mffi2 D!z2
pþffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi (18)
r u θ ≈ Γz1 ϕθ1 Dθ1 þ Γz2 ϕθ2 Dθ2 ¼ s1 m1 I 1 D!θ1 þ s2 m2 I 2 D!θ2
! ! ! ! ! ! ! !
where Γ!zn, n¼1~2, are the modal participation factors of the nth vibration mode of the
EOSB; D!zn, D!θn, n¼1~2, are the 2DOF modal responses of the nth vibration mode of
the EOSB. Remember that the EOSB is constructed from the two targeted 2DOF modal
systems of the original multistory building. Hence, the two modal responses of the
EOSB are the same as the modal responses of the first translational-dominant and the
first rotational-dominant vibration modes of the original multistory building, that is, D!zn ¼
Dzn D!θn ¼ Dθn, n¼1~2. Assume that the mode shapes of the first translational-dominant and
the first rotational-dominant vibration modes of the original multistory building are triangu-
lar, that is
! "
N −1 1 T
’zn ≈ ϕzn;r h; ’θn ≈ ϕθn;r h ; h ¼ 1 ··· ; n¼ 1e2 (19)
N N
where the subscript r (Equation 19) represents the roof component. Note that the nth mode
shape shown in Equation 19 is normalized to have the nth modal mass equal to one. Thus, the
mn and I n, n¼ 1e2, defined in Equation 3, can be approximated as:
mn ≈ ϕ2zn;r hT mh; I n ≈ ϕ2θn;r hT Ih ; n¼ 1e2 (20)
Substituting Equation 20 and D!zn ¼ Dzn, D!θn ¼ Dθn, n¼ 1e2, into Equation 18 results in
8 ( )
< u !z ≈ ϕ2z1;r Dz1 þ ϕ2z2;r Dz2 hT mh
( )pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (21)
: r ! u ! ≈ ϕz1;r ϕθ1;r Dθ1 þ ϕz2;r ϕθ2;r Dθ2 hT mhhT Ih
θ
If the radius of gyration r of the floors of the original multistory building is constant,
Equation 21 can be further simplified as:
( ! ( 2 )
u z ≈ ϕz1;r Dz1 þ ϕ2z2;r Dz2 hT mh
( ) (22)
r! u !θ ≈ r ϕz1;r ϕθ1;r Dθ1 þ ϕz2;r ϕθ2;r Dθ2 hT mh
On the other hand, from Equation 19 and the previous study (Lin and Tsai 2007b), the
roof displacements of the original multistory building can be approximately expressed as:
( )
+ u ≈ Γ ϕ D þ Γ ϕ D ¼ ϕ2 D þ ϕ2 D hT m1
z;r z1 z1;r z1 z2 z2;r z2 z1 z2;r z2
( z1;r ) (23)
u θ;r ≈ Γz1 ϕθ1;r Dθ1 þ Γz2 ϕθ2;r Dθ2 ¼ ϕz1;r ϕθ1;r Dθ1 þ ϕz2;r ϕθ2;r Dθ2 hT m1
where Γzn, n¼ 1e2, are the modal participation factors of the nth vibration mode of the
original multistory building; 1 is the N × 1 unit vector. Comparing Equation 23 with Equa-
tion 21 results in
hT m1 ! hT m1
u z;r ≈ u ; u θ;r ≈ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r ! u !θ (24)
hT mh z hT mhhT Ih
Equation 24 indicates that the roof displacements of the original multistory building can
be predicted by using those of the EOSB. Therefore, it clearly implies that the displacement
spectra, denoted as S!dz and S!dθ , constructed from the EOSBs can be used to predict the peak
roof displacements, denoted as Sdz;r and Sdθ;r , of the original nonproportionally damped mul-
tistory asymmetrical buildings, that is
Sdz;r ≈ ηz S!dz ; Sdθ;r ≈ ηθ S!dθ (25a)
where
hT m1 hT m1
ηz ¼ T ; ηθ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (25b)
h mh hT mhhT Ih
, - ., , - ., , - ., , - .,
EQ-TARGET;temp:intralink-;e25c;41;264 Sdz;r ¼ ,u z;r t ,max; Sdθ;r ¼ ,u θ;r t ,max; S!dz ¼ ,u !z t ,max; S!dθ ¼ ,r ! u !θ t ,max (25c)
Note that the ηz has no unit and the unit of ηθ is the inverse of length. Obviously, the use
of the EOSB in predicting the roof displacements of the original multistory building—that is,
only considering the two targeted vibration modes of the original building—may be insuffi-
cient for buildings with significantly higher modes’ effect. Note that the higher modes’ effect
also depends on the frequency contents of the ground motions. In addition, it is well known
that the higher modes have more effect on the base shear than the floor displacements. Thus,
the application of the EOSB in predicting the structural responses, where the higher modes
play an important role, needs further investigation.
NUMERICAL EXAMPLES
Buildings TS, TSS, and TF were used as the example buildings to verify the accuracy of
applying the response spectra constructed from the EOSB to estimate their peak roof
Table 7. Earthquake data.
Designation Record information Duration (sec.) Magnitude Mw R (km) Scale PGA (g)
LA01 Imperial Valley, 1940 39.38 6.9 10.0 2.01 0.46
LA07 Landers, 1992 79.98 7.3 36.0 3.20 0.42
LA08 Landers, 1992 79.98 7.3 36.0 3.20 0.43
LA09 Landers, 1992 79.98 7.3 25.0 2.17 0.52
LA10 Landers, 1992 79.98 7.3 25.0 2.17 0.36
LA11 Loma Prieta, 1989 39.98 7.0 12.4 1.79 0.67
LA12 Loma Prieta, 1989 39.98 7.0 12.4 1.79 0.97
LA13 Northridge, 1994, Newhall 59.98 6.7 6.7 1.03 0.68
LA14 Northridge, 1994, Newhall 59.98 6.7 6.7 1.03 0.66
LA15 Northridge, 1994, Rinaldi 14.95 6.7 7.5 0.79 0.53
LA16 Northridge, 1994, Rinaldi 14.95 6.7 7.5 0.79 0.58
LA17 Northridge, 1994, Sylmar 59.98 6.7 6.4 0.99 0.57
LA18 Northridge, 1994, Sylmar 59.98 6.7 6.4 0.99 0.82
LA19 North Palm Springs, 1986 59.98 6.0 6.7 2.97 1.02
LA20 North Palm Springs, 1986 59.98 6.0 6.7 2.97 0.99
Table 8. The translational vibration periods, damping ratio, eccentricity, frequency ratio,
and the values of ηz and ηθ of the EOSBs for Buildings TS, TSS, and TF.
Building T z (sec.) ξ ē Ω ηz ηθ (1/m)

TS 0.905 0.02 1.300 1.794 1.493 0.316
TSS 0.939 0.02 0.738 1.098 1.493 0.183
TF 0.941 0.02 0.331 0.492 1.493 0.082
responses, including the peak roof translational and rotational displacements. The selected
ground motion records are the set of 20 ground motion records (Table 7) with a return period
of 475 years, which was used in the SAC steel research project for buildings located in Los
Angeles (Krawinkler 2000). By using the contents given in Table 4, the properties of the
EOSBs for Buildings TS, TSS, and TF are shown in Table 8. These properties include
the translational vibration period T z , the inherent damping ratio ξ, the normalized eccentricity
ē, and the frequency ratio Ω ¼ T z ∕T θ . Table 8 also shows the values of ηz and ηθ . Using the
values of ξ, ē, Ω (Table 8) and the values of the supplemental damping properties ξsd , ēsd , and
ρ̄sd (Table 4), the response spectra of the EOSBs were constructed shown in Figure 10. The
spectral values S!dz and S!dθ (Figure 10), corresponding to the translational vibration periods of
Figure 10. The displacement response spectra for Building (a) and (b) TS (ξ =
0.02; ē ¼ 1.3; Ω ¼ 1.794Þ; (c) and (d) TSS ðξ ¼ 0.02; ē ¼ 0.738; Ω ¼ 1.098Þ ; (e) and (f) TF
ðξ ¼ 0.02; ē ¼ 0.331; Ω ¼ 0.492Þ subjected to the selected set of 20 seismic ground motions.
Buildings TS, TSS, and TF, multiplied by ηz and ηθ , respectively, are shown in Table 9.
Table 9 also shows the exact peak roof responses Sdz and Sdθ of the original multistory build-
ings and the corresponding estimation errors. Table 9 shows that the means of the estimation
errors are between 3.8% and 13.4%; the coefficients of variation (CV) for the estimation
errors are between 0.23 and 1.61. The positive and the negative estimation errors (Table 9)
indicate that the estimated peak roof responses are on the conservative side and the uncon-
servative side, respectively. Table 9 shows that most of the estimations are on the conser-
vative side. Furthermore, Figure 11 shows the comparison of the estimated and the exact peak
roof responses. Specifically, Figure 12 shows the estimated and the exact roof response his-
tories of these three example buildings subjected to the excitation of the LA02 (Table 7)
ground motion (1940 El Centro earthquake). Table 9 and Figures 11 and 12 confirm the
Table 9. The estimated and the exact peak roof responses for Buildings TS, TSS, and TF subjected to the selected ground motions.
Building TS Building TSS Building TF
ηz S!dz ηθ S!dθ Sdz Sdθ Error z Error θ ηz S!dz ηθ S!dθ Sdz Sdθ Error z Error θ ηz S!dz ηθ S!dθ Sdz Sdθ Error z Error θ
EQs (m) (rad.) (m) (rad.) (%) (%) (m) (rad.) (m) (rad.) (%) (%) (m) (rad.) (m) (rad.) (%) (%)
LA01 0.210 0.014 0.209 0.013 0.4 9.9 0.177 0.014 0.172 0.014 2.8 3.7 0.083 0.005 0.073 0.005 14.1 1.9
LA02 0.175 0.012 0.160 0.011 9.2 13.3 0.151 0.013 0.123 0.012 22.4 4.9 0.068 0.003 0.061 0.003 12.4 7.4
LA03 0.301 0.019 0.265 0.015 14.0 21.7 0.213 0.016 0.195 0.014 9.7 11.2 0.269 0.014 0.237 0.013 13.2 7.1
LA04 0.198 0.013 0.176 0.010 12.5 21.2 0.142 0.012 0.129 0.010 9.8 11.7 0.124 0.007 0.105 0.007 18.1 1.4
LA05 0.246 0.016 0.227 0.013 8.5 16.5 0.231 0.018 0.211 0.016 9.2 9.3 0.176 0.010 0.156 0.009 13.3 7.9
LA06 0.111 0.007 0.101 0.006 9.6 16.7 0.097 0.007 0.083 0.006 17.7 11.7 0.150 0.009 0.132 0.008 12.9 10.0
LA07 0.122 0.008 0.103 0.006 19.3 29.5 0.094 0.007 0.084 0.006 12.4 9.8 0.080 0.009 0.073 0.008 10.2 7.3
LA08 0.118 0.008 0.111 0.007 6.4 13.6 0.121 0.009 0.110 0.008 10.1 11.4 0.077 0.004 0.068 0.004 12.9 8.3
LA09 0.199 0.013 0.182 0.011 9.2 18.5 0.168 0.014 0.155 0.013 8.3 7.8 0.142 0.007 0.126 0.006 12.5 6.5
LA10 0.076 0.006 0.084 0.006 −9.0 0.0 0.108 0.009 0.098 0.008 10.7 2.4 0.089 0.005 0.083 0.005 6.6 3.8
LA11 0.222 0.014 0.206 0.012 7.8 16.4 0.179 0.013 0.164 0.012 9.0 9.5 0.131 0.006 0.112 0.005 16.7 1.9
LA12 0.065 0.004 0.066 0.004 −1.2 0.0 0.068 0.005 0.068 0.005 −0.3 −3.9 0.044 0.003 0.040 0.003 9.8 −3.3
LA13 0.134 0.008 0.119 0.007 12.2 18.3 0.093 0.008 0.087 0.007 7.3 9.7 0.086 0.005 0.076 0.005 12.6 0.0
LA14 0.135 0.010 0.120 0.008 12.8 20.7 0.108 0.011 0.104 0.010 4.0 4.9 0.103 0.005 0.089 0.005 15.6 4.0
LA15 0.546 0.033 0.494 0.028 10.4 18.6 0.594 0.039 0.521 0.035 14.0 11.4 0.376 0.036 0.331 0.032 13.9 9.9
LA16 0.626 0.038 0.571 0.033 9.5 17.2 0.650 0.046 0.555 0.042 17.0 10.8 0.388 0.038 0.345 0.035 12.5 10.0
LA17 0.297 0.019 0.281 0.017 5.7 13.0 0.231 0.019 0.223 0.018 3.5 5.4 0.094 0.005 0.085 0.004 10.9 11.4
LA18 0.233 0.015 0.216 0.013 7.8 15.4 0.205 0.017 0.188 0.016 8.7 7.6 0.138 0.006 0.121 0.006 14.4 0.0
LA19 0.054 0.003 0.063 0.004 −13.9 −17.1 0.053 0.005 0.059 0.005 −10.4 0.0 0.051 0.002 0.042 0.002 21.1 −9.5
LA20 0.184 0.012 0.189 0.011 −3.0 3.5 0.147 0.014 0.145 0.012 1.5 13.9 0.117 0.005 0.103 0.005 13.8 −10.0
APPLICATION OF SUPPLEMENTAL DAMPING CHARACTERISTICS TO RESPONSE SPECTRUM ANALYSES
mean 0.213 0.014 0.197 0.012 6.4 13.4 0.191 0.015 0.174 0.013 8.4 7.7 0.139 0.009 0.123 0.009 13.4 3.8
CV 0.69 0.64 0.67 0.63 1.25 0.76 0.82 0.70 0.77 0.70 0.85 0.59 0.70 1.08 0.70 1.05 0.23 1.61
229
Figure 11. The comparison of the estimated (a) roof translations and (b) roof rotations with the
corresponding exact responses for Buildings TS, TSS, and TF subjected to the selected set of 20
seismic ground motions.
Figure 12. The roof translational and rotational responses: (a) and (b) for Building TS; (c) and
(d) for Building TSS; (e) and (f) for Building TF under the excitation of the 1940 El Centro
earthquake.
accuracy of Equation 25. That is to say, the peak roof displacements of these three eight-story
example buildings can be satisfactorily estimated by using the proposed approach. It is inter-
esting to note that the code-prescribed equivalent damping ratio for a nonproportionally
damped asymmetrical building is commonly computed for only one single mode. Thus,
the analysis not presented in this paper indicates that the roof translation and rotation of
such a building cannot be satisfactorily predicted from using the traditional response spec-
trum with the code-prescribed damping ratio.
SUMMARY AND CONCLUSIONS

Response spectrum analyses are widely applied on earthquake engineering. Several cur-
rent building codes stipulate the equivalent damping ratio resulting from the added viscous
dampers in order to apply response spectrum analyses to nonproportionally damped build-
ings. Evidently, besides the equivalent damping ratio, there should be other overall system
parameters for characterizing the plan-wise distribution of added viscous dampers in mul-
tistory asymmetrical buildings. In this study, there were three proposed supplemental damp-
ing characteristics for multistory asymmetrical buildings with linear viscous dampers, which
are the same as those stated in other literature focusing on single-story asymmetrical build-
ings. These three supplemental damping characteristics are the equivalent supplemental
damping ratio ξsd , the normalized equivalent supplemental damping eccentricity ēsd , and
the normalized equivalent supplemental damping radius of gyration ρ̄sd . Nevertheless, the
approach of characterizing these three overall system parameters for a multistory asymme-
trical building is not as straightforward as that for a single-story asymmetrical building. This
study first constructed the EOSB as the vehicle for characterizing the supplemental damping
in multistory asymmetrical buildings. The EOSB was constructed from the 2DOF modal
properties of the first translational-dominant and the first rotational-dominant vibration
modes of the original multistory building. These overall system parameters enable practicing
engineers to understanding and quantitatively describe the resultant effects of the added vis-
cous dampers in multistory asymmetrical buildings. Second, this study showed that the roof
displacement responses of the original multistory building can be estimated by mapping the
displacement responses of the EOSB. The response spectra constructed from the EOSBs
were satisfactorily applied to estimate the peak roof displacements of three nonproportionally
damped eight-story example buildings subjected to 20 seismic ground motions. However, it
is worth noting that in the EOSB only two vibration modes of the original building are con-
sidered in estimating the seismic responses of the original building. The method of including
the higher modes’ effect needs further research.
As indicated above, the EOSB can be used in describing the overall supplemental damp-
ing properties, and conveniently applied in constructing the response spectrum for nonpro-
portionally damped multistory asymmetrical buildings. Other applications of the EOSB have
been studied. For example, the optimal locations of the added dampers in a multistory asym-
metrical building can be determined from analyzing the frequency response functions of the
corresponding EOSBs (Bui 2011). Extending the applications of the EOSB for the inelastic
multistory asymmetrical buildings is currently under investigation.
ACKNOWLEDGMENTS
This research was financially supported by the National Science Council of Taiwan, R.O.C.,
under grant no. NSC 98-2221-E-002-107-MY2, and is hereby gratefully acknowledged.
REFERENCES
Ahlawat, A.S., and Ramaswamy, A., 2003. Multiobjective optimal absorber system for torsion-
ally coupled seismically excited structures, Engineering Structures 25, 941–950.
Bui, M. T., 2011. Optimal Placement of Viscous Dampers in Two-Way Asymmetric-Plan Buildings
by Using the Effective One-Story Buildings, Master’s Thesis, National Taiwan University.
Federal Emergency Management Agency (FEMA), 1997. NEHRP Recommended Provisions for
Seismic Regulations for New Buildings, Washington, DC.
Federal Emergency Management Agency (FEMA), 2000. Prestandard and Commentary for the
Seismic Rehabilitation of Buildings, Tech. Rep. FEMA-356, Washington, DC.
Goel, R. K., 1997. Seismic response of asymmetric systems: Energy-based approach, Journal of
Structural Engineering 123, 1444–1453.
Goel, R. K., 1998. Effects of supplemental viscous damping on seismic response of asymmetric-
plan systems, Earthquake Engineering and Structural Dynamics 27, 125–141.
Goel, R. K., 2000. Seismic behavior of asymmetric buildings with supplemental damping, Earth-
quake Engineering and Structural Dynamics 29, 461–480.
Goel, R. K., 2001. Simplified analysis of asymmetric structures with supplemental damping,
Earthquake Engineering and Structural Dynamics 30, 1399–1416.
Goel, R. K., and Booker, C. A., 2001. Effects of supplemental viscous damping on inelastic
seismic response of asymmetric systems, Earthquake Engineering and Structural Dynamics
30, 411–430.
International Code Council (ICC), 1997. Uniform Building Code, International Conference of
Building Officials, Whittier, CA.
International Code Council (ICC), 2000. International Building Code, International Conference
of Building Officials Whittier, CA.
Kan, L. C., and Chopra, A. K., 1977. Elastic earthquake analysis of the torsionally coupled mul-
tistory buildings, Earthquake Engineering and Structural Dynamics 5, 395–412.
Krawinkler, H., 2000. State of the Art Report on Systems Performance of Steel Moment Frames
Subject to Earthquake Ground Shaking, Report No. FEMA-355C, SAC Joint Venture.
Lin, J. L., and Tsai, K. C., 2007a. Simplified seismic analysis of asymmetric building systems,
Lin, J. L., and Tsai, K. C., 2007b. Simplified seismic analysis of one-way asymmetric elastic
systems with supplemental damping, Earthquake Engineering and Structural Dynamics
36, 783–800.
Lin, J. L., and Tsai, K. C., 2009. Modal parameters for the analysis of inelastic asymmetric-plan
structures, Earthquake Spectra 25, 821–849.
Lin, W. H., and Chopra, A. K., 2001. Understanding and predicting effects of supplemental vis-
cous damping on seismic response of asymmetric one-story systems, Earthquake Engineering
and Structural Dynamics 30, 1475–1494.
Riddell, R., and Santa-Maria, H., 1999. Inelastic response of one-story asymmetric-plan systems
subjected to bi-directional earthquake motions, Earthquake Engineering and Structural
Dynamics 28, 273–285
Stathopoulos, K. G., and Anagnostopoulos, S. A., 2003. Inelastic earthquake response of single-
story asymmetric buildings: An assessment of simplified shear-beam models, Earthquake
Engineering and Structural Dynamics 32, 1813–1831.
Tso, W. K., and Sadek, A. W., 1985. Inelastic seismic response of simple eccentric structures,
(Received 9 March 2011; accepted 4 December 2011)
Life-Cycle Risk Assessment of Spatially
Distributed Aging Bridges under Seismic
and Traffic Hazards
Alberto Decò,a) and Dan M. Frangopol,b) M.EERI
This paper presents a rational framework for the quantitative assessment of

life-cycle risk of bridges within a transportation network. Bridge vulnerability
is evaluated with respect to seismic and abnormal traffic hazards. The effects
induced by seismic hazard are investigated by means of fragility analysis.
Random earthquakes are generated using Latin Hypercube sampling technique,
and probabilities of exceeding specific structural damage states are computed
for each specific seismic scenario. Traffic hazard is assessed considering Weibull
distributed time-to-failure of the bridge superstructure. Epistemic and aleatory
uncertainties are accounted for in order to provide a rational assessment of
life-cycle risk. Consequence analysis includes different levels of bridge
serviceability (fully serviceable, partly serviceable, closed, and collapsed), and
monetary values are used to evaluate direct and indirect consequences. Seismic
risk, traffic-induced risk, and total risk are evaluated for a group of existing bridges
located north of the San Diego metropolitan area. [DOI: 10.1193/1.4000094]
INTRODUCTION
Life-cycle management of single structures or groups of structures has become an issue of
great relevance for many countries, including the United States. Maintenance and retrofit are
crucial in preventing excessive deterioration of structural performance over time. Efficient
and optimal maintenance strategies must be included in a general framework able to capture
the social effects produced by the gained benefit. In this context, life-cycle risk assessment
provides a tool that directly relates the effects of potential hazards with the associated
consequences. During their designed lifetime, structures and infrastructures are intended
to withstand the effects produced by a spectrum of hazards that could generate major safety
and serviceability concerns. Transportation networks rely on the serviceability of their single
bridges in order to allow users to reach their destinations in a reasonable time.
Risk assessment for single bridges has been studied with respect to single or multiple
hazards. Stein et al. (1999) developed a practical approach to estimate risk due to the effects
a)
Graduate Research Assistant, Department of Civil & Environmental Engineering, Engineering Research Center
for Advanced Technology for Large Structural Systems (ATLSS), Lehigh University, Bethlehem, PA 18015-
4729, USA
b)
Professor and the Fazlur R. Khan Endowed Chair of Structural Engineering and Architecture, Department of
Civil & Environmental Engineering, Engineering Research Center for Advanced Technology for Large
Structural Systems (ATLSS), Lehigh University, Bethlehem, PA 18015-4729, USA, email address: dan
.frangopol@lehigh.edu, phone: 610-758-6103 or 610-758-6123, fax: 610-758-4115 or 610-758-5553
127
128 A. DECÒ AND D. M. FRANGOPOL
induced by scour (single hazard), whereas Adey et al. (2003) investigated the potential con-
sequences due to traffic hazard and scour in order to schedule optimal interventions over time
(multiple hazards).
Recently, different aspects concerning performance of transportation networks have
been extensively investigated (Golroo et al. 2010, Frangopol 2011, Frangopol and
Bocchini 2012, Bocchini and Frangopol 2012a). For instance, life-cycle analyses of a
bridge network located within the Denver metropolitan area have been performed, mostly
focusing on the assessment of network performance and on the optimization of
maintenance interventions (Liu and Frangopol 2005, 2006, Frangopol and Liu 2007,
Bocchini and Frangopol 2011).
A predominant part of the research activity focused on the performance of transportation
networks subjected to seismic ground motion considering bridge spatial distribution. Risk
has often been included in such analyses (Chang et al. 2000, Jones et al. 2005, Padgett
et al. 2010). Although these studies fully cover seismic-related aspects, most of the analyses
focus on risk induced only by a single hazard. So far, few studies focused on the inclusion of
traffic hazards as part of the identified predominant hazards for risk assessment. Furthermore,
except for a few studies (such as Adey et al. 2003, Decò and Frangopol 2011), the research
lacks studies assessing time-dependent risk of bridges. The time-dependency of risk in a life-
cycle context is a quite new area to be explored. In seismic analysis, aging consideration has
started to be included in seismic performance prediction models (Ghosh and Padgett 2010,
Akiyama et al. 2011).
The aim of this paper is to provide a probabilistic approach for the assessment of life-
cycle risk of a group of aging bridges within a transportation network. Both structural
vulnerability and potential consequences induced by hazards are time-dependent. Bridge
vulnerabilities are evaluated with respect to hazards induced by earthquakes and abnormal
traffic. Based on the original approach proposed by Cornell (1968) and on the method used
by Jones et al. (2005), the adopted probabilistic seismic hazard analysis is based upon the
generation of random earthquakes spatially distributed over a region and having a random
magnitude that follows a truncated-exponential relationship. Then, according to Basöz and
Mander (1999) and including aging considerations, time-dependent fragility analysis pro-
vides the probability of exceeding specific damage states (slight, moderate, extensive,
and complete) over time. Time-dependent vulnerability due to traffic, mainly affecting
the bridge superstructure, is modeled assuming for each bridge a Weibull distribution of
the time-to-failure (Maunsell Ltd. 1999). Risk is assessed for different levels of bridge
serviceability and, therefore, different consequences are associated with each serviceability
level. Direct and indirect consequence analyses include the probabilistic evaluation of the
monetary values associated with rebuilding costs, rehabilitation costs, material damage
costs, operating costs, time-loss costs, loss of human life costs, and injury costs. Both epis-
temic and aleatory uncertainties are considered. The profiles of time-dependent direct, indir-
ect, and total risk are developed for each bridge within the network. A simplified method is
used to compute the risk associated with a small transportation network. A group of existing
bridges located north of the San Diego metropolitan area is investigated.
LIFE-CYCLE RISK ASSESSMENT OF SPATIALLY DISTRIBUTED AGING BRIDGES 129
BRIDGE VULNERABILITY UNDER MULTIPLE HAZARDS

The estimation of time-dependent bridge vulnerability is affected by the level of detail
required by the type of analysis. When dealing with analyses at the network scale, simpli-
fications must be introduced, and, consequently, the accuracy of the analysis is affected by
approximations (Frangopol and Bocchini 2012).
Seismic and traffic hazards affect the whole bridge structure. However, different bridge
components are more likely to fail due to different effects that can be caused by specific
hazards. For instance, vertical loads (due to traffic hazard) mainly affect the bridge super-
structure that mostly involves flexural failure mode (Akgül 2002), while seismic fragility of
the bridge is largely dependent on bridge substructure components (columns, abutments, and
bearings, according to Padgett and DesRoches (2007) and Ghosh and Padgett (2010)). There-
fore, load effects can be analyzed separately for bridge superstructure and substructure,
respectively. Moreover, load effects induced by seismic and traffic hazards are considered
to be statistically independent since their occurrences are independent.
SEISMIC VULNERABILITY
During the last decade, seismic vulnerability of bridges was a topic extensively studied
(Basöz and Mander 1999, Nielson and DesRoches 2007). In this paper, fragility analysis of
bridges is conducted based on a simulated set of ground motions caused by earthquakes
having random magnitudes and random spatial epicenter coordinates over a region
(Chang et al. 2000, Sánchez-Silva and Rackwitz 2004, Jones et al. 2005).
Probabilistic ground motion
In this study, the assessment of risk relies on simulations of seismic events striking within
a considered region. The following steps summarize the adopted procedure to simulate the
seismic scenarios:
• Definition of a set of random seismic events. A specific number of random earth-
quake events, EQ, is generated within the region under investigation.
• Assignment of a random Richter magnitude M. According to Cosentino et al.
(1977), depending on the defined seismic region, and during a given time period,
the frequency-magnitude relationship can be modeled by adopting a truncated
exponential probability density function (PDF) defined as:
! βexpð−βðM − M 0 ÞÞ
for M 0 ≤ M ≤ M p
f TE ðMÞ ¼
1 − expð−βðM p − M 0 ÞÞ (1)
0 for M < M 0 and M > M p
where M 0 is the threshold magnitude value, M p is the maximum regional finite

magnitude value, and β is a distribution parameter.
• Assignment of a random location and temporal occurrence for each seismic event.
Event coordinates (latitude Lat and longitude Long) are randomly generated within
the considered area. The epicenter coordinates (x and y) of the i-th random earth-
quake, over the considered square seismic region, are assumed to be random vari-
ables triangularly distributed. Moreover, the temporal occurrence of the seismic
events is assumed to be uniform distributed over the considered life span.
• Definition of the characteristics of each generated seismic event. Information

regarding earthquake characteristics must be assigned, including rupture geometry
parameters (depth, width, and dip angle). Furthermore, earthquake depth is ran-
domly generated following a triangular distribution. The rupture length Rl and
width Rw of each seismic event depends on the random magnitude (Wells and
Coppersmith 1994):
logðRl Þ ¼ al þ bl M (2) EQ-TARGET;temp:intralink-;e2;41;566
logðRw Þ ¼ aw þ bw M
EQ-TARGET;temp:intralink-;e3;41;548 (3)
where al , aw , bl , and bw are coefficients depending on the earthquake characteristics
(their values are given in Wells and Coppersmith 1994). Furthermore, the orienta-
tion angle of the rupture trace is also considered random for each generated event.
Based on historical data, correlation among epicenter coordinates, depths, and magni-
tudes is obtained and implemented while generating the random events.
Bridge fragility analysis

The distance between the rupture traces of the i-th earthquake and the locations of the j-th
distributed bridge influences the level of expected structural damage. In this paper, the attenua-
tion relationship proposed by Campbell and Bozorgnia (2007) is implemented while modeling
the effect of seismic hazard. The median ground motion model is described by:
lnŶ j ¼ f mag þ f jdis þ f f lt þ f hng þ f jsite þ f jsed
where InŶ is the natural logarithm of the median value of the peak ground acceleration (PGA)
or of the response spectra accelerations (PSA) for several oscillator periods, f mag is the mag-
nitude term, f dis is the distance term, f f lt is the style-of-faulting term, f hng is the hanging-wall
term, f site is the shallow site response term, and f sed is the deep site response term. Epistemic
uncertainties are accounted for by (Campbell and Bozorgnia 2007):
j
lnY junc ¼ lnŶ % ΔlnŶ
where lnY unc is the predicted value of InŶ accounting for epistemic uncertainties and ΔlnŶ is
the incremental value of the median ground motion due to epistemic uncertainties. All the
terms introduced (in Equations 4 and 5) and their relevant evaluations are discussed in detail
in Campbell and Bozorgnia (2007).
According to Mander (1999), by means of fragility analysis, four curves describing the
probability of exceeding specific damage states (slight, moderate, extensive, and complete)
can be developed for each seismic scenario and for each considered bridge. Fragility curves
are defined by log-normal cumulative probability functions:
" # $%
j j 1 PSAj
P½DS jPSA ' ¼ Φ ln (6)
βc AjDS
where P½DSjPSA' is the probability of exceedance of a specific damage state DS given a

specific PSA (period of 1 sec), βc is the normalized composite log-normal standard deviation,
ADS is the median spectral acceleration associated with the occurrence of the damage state
DS, and Φ is the standard normal cumulative distribution function. The approach
necessitates bridge data (bridge class, number of spans, skew angle, site class, and year
of construction), which can be obtained from the National Bridge Inventory (NBI)
(FHWA 2010).
Fragility time-dependency
The effect of corrosion on the substructure due to exposure to chloride ions may influence
the seismic response of a bridge (Ghosh and Padgett 2010). Life-cycle analysis must account
for the potential increase of seismic effects on aged structures. In this context, the bridge
fragility depends on the time t at which the simulated earthquake strikes. Thus, Equation 6
changes as: " # $%
j j 1 PSAj
P½DS jPSA 'ðtÞ ¼ Φ ln j (7)
βc ADS ðtÞ
where ADS ðtÞ represents the time-dependency of the fragility parameters. General methods
accounting for the deterioration process of seismic-sensitive components for different bridges
and deterioration typologies are usually missing in the existing literature. Although this is a
formidable task, if further studies are developed and validated (for instance, by means of
collecting field data, which are usually unavailable for long-term predictions), the proposed
general framework would be able to accommodate such considerations. In the
proposed approach, the time-dependent median spectral acceleration ADS ðtÞ associated
with the occurrence of a specific damage state of the j-th bridge is assumed to vary in
time by following a linear degradation:
AjDS ðtÞ ¼ AjDS ½0' ⋅ ð1 − t ⋅ γÞ

where ADS ½0' is the original fragility parameter (Basöz and Mander 1999), t is the time, and γ
is the aging coefficient. Indications concerning possible values of the aging coefficient can be
derived from Ghosh and Padgett (2010), where a reduction of about 30%–40% of the initial
fragility parameter was reported in 100 years. The use of a linear degradation law captures the
total reduction of the fragility parameters over the bridge’s lifetime. However, the adopted
degradation law may lead to some approximations that are usually unavoidable for analyses
of groups of bridges at the network scale, mainly due to the lack of detail and/or insufficient
data (Frangopol and Bocchini 2012).
TRAFFIC-INDUCED VULNERABILITY
In the past, lifetime functions have been developed for bridge life-cycle-related topics
(Maunsell Ltd. 1999). For the bridge superstructure, the effects caused by corrosion pro-
cesses or general aging phenomena can be accounted for by means of lifetime functions
(van Noortwijk and Klatter 2004, Yang et al. 2004). Accordingly, time-dependent bridge
performance with respect to vertical load may be modeled by considering the time-to-failure
of the superstructure as a random variable (Decò and Frangopol 2011). Weibull distribution
can be used for the assessment of single bridge superstructures due to the effects of traffic and
aging. The PDF and the associated cumulative distribution function (CDF) of the time-to-failure
of the superstructure (f ttf and F ttf , respectively) for the j-th bridge are modeled by a Weibull
distribution as follows (Leemis 1995):
κj
f jttf ðtÞ ¼ λj κj ðλj tÞðκ −1Þ e−ðλ tÞ
j j
t≥0 (9)
κj
F jttf ðtÞ ¼ 1 − e−ðλ tÞ
j
t≥0 (10)
where κ and λ are the shape and scale parameter, respectively, that can be found by statistical
inference of observations as shown in previous studies (van Noortwijk and Klatter 2004), or
by analyzing bridge failure modes considering the failure of one or more components as
described in Yang et al. (2004).
The annual failure rate of the superstructure due to traffic hazard, which coincides with its
annual failure probability PT; f , given that the structure has survived until time t and given
that the time interval is one year, is obtained by (Bocchini and Frangopol 2011b, Decò and
Frangopol 2011):
f jttf ðtÞ
PjT; f ðtÞ ¼ (11)
1 − F jttf ðtÞ
CONSEQUENCE ANALYSIS
This section presents the consequence analysis of specific risks that are associated with
seismic and traffic hazards. The quantification of risk requires the evaluation of bridge vul-
nerability for each identified hazard and the assessment of the negative consequences. More-
over, while performing life-cycle analyses, time is a variable of extreme importance to be
incorporated. In fact, some types of consequences may vary with time, such as those asso-
ciated with the average daily traffic (ADT) of a carried route. In this paper, consequence
analysis uses an economic approach in order to associate regional economic impacts
with a monetary value expressed in United States dollars (USD).
Direct consequences include rebuilding costs, rehabilitation costs, and material damage
costs, whereas indirect consequences include operating costs, time-loss costs, and accident
costs (loss of human lives and injuries). Material damage costs and accident costs depend on
the number of accidents associated with bridge failure.
When consequences are estimated in terms of monetary values, the prediction of future
risk must account for the difference between present and future values of monetary losses.
Thus, future consequences in terms of monetary values are given by:
FVðtÞ ¼ PVð1 þ rÞt
where FVðtÞ is the future monetary value with respect to year t, PV is the present monetary
value, and r represents the annual discount rate of money.
Time-dependent direct C D ðtÞ and indirect C I ðtÞ consequences associated with a bridge
damage/failure for a specific year t are defined in Table 1, where E Reb, E Reh , E MD , E Op , E TL ,
E LHL , and E Inj are the factors accounting for epistemic uncertainties associated with rebuild-
ing costs, rehabilitation costs, material damaged in a vehicular accident, operating costs,
Table 1. Summary of the considered direct and indirect consequences.
Consequence type Eq. no.
* EQ-TARGET;temp:intralink-;e13;264;19
Rebuilding CjRebðtÞ ¼ E RebcjRebW j Lj ð1 þ rÞt (13)
† j j j j cjRebW j Lj (14)
Rehabilitation EQ-TARGET;temp:intralink-;e14;264;7
δj
CReh ðtÞ ¼ E Reh ⋅ min 0.6cRebW L ; c0 þ ð1 þ rÞt
" %
* τj
Direct costs CD ðtÞ
Material damage j
EQ-TARGET;temp:intralink-;e15;264;-36
j (15)
cS;car ð1 − 100 Þþ Aj ðtÞLj
τ
C MD ðtÞ ¼ E MD ð1 þ rÞt
" %
þðcS;truck þ cS;cargo Þ 100 24000Sj ðtÞ
*
Operating τj (16)
cOp ;car ð1 − 100 Þþ j j
τj
C jOp ðtÞ ¼ E Op Dl A ðtÞd j ð1 þ rÞt
" %
þcOp ;truck 100

*
Time loss
τj (17)
cW OCar ð1 − 100 Þþ
τj
C jTL ðtÞ ¼ E TL T jm ðtÞAj ðtÞd j ð1 þ rÞt
" %
þðcTC OTruck þ cInv Þ 100

Indirect costs CI ðtÞ
τj
* α c (18)
Loss of human lives EQ-TARGET;temp:intralink-;e18;264;-156
j
LIFE-CYCLE RISK ASSESSMENT OF SPATIALLY DISTRIBUTED AGING BRIDGES
OCar ð1 − 100 Þþ Aj ðtÞLj

100 τ
C jLHL ðtÞ ¼ E LHL LHL VSL ð1 þ rÞt
! &
þOTruck 100 24000Sj ðtÞ
j
τ
αInj cInj (19)
*Injuries
j
OCar ð1 − 100 Þþ Aj ðtÞLj
100 τ
C jInj ðtÞ ¼ E Inj ð1 þ rÞt
! &
þOTruck 100 24000Sj ðtÞ
*Based on Stein et al. (1999)

†
Based on Sánchez-Silva and Rackwitz (2004)
133
time-loss costs, loss of human lives, and injury costs, respectively; cReb is the rebuilding cost
per square meter ðUSD∕m2 Þ; c0 are costs that do not depend on the level of damage (USD);
cS;car is the car salvage value (USD); cS;truck is the truck salvage value (USD); cS;cargo is the
value of the cargo (USD); cOp ;car and cOp ;truck are the operating costs of cars and trucks per
kilometer (USD/km), respectively; cW is the wage per hour (USD/h); cTC is the total com-
pensation per hour (USD/h); cInv is the inventory cost (USD/h); cVSL is the value of a sta-
tistical life (VSL) (USD); cInj is the cost associated with injuries (USD); W and L are the
bridge width (m) and length (m), respectively; δ is a coefficient accounting for the level of
damage depending on bridge residual serviceability; AðtÞ is the ADT referred to year t; τ
represents the average daily truck traffic as percentage of ADT; SðtÞ is the traveling
speed (km/h); Dl is the bypass detour length (km); d is the duration of the detour (days);
OCar and OTruck are the vehicle occupancies for cars and for trucks, respectively; T m ðtÞ
is the marginal travel time (h) required to cover the traveled route or the detour depending
on bridge serviceability and traffic speed; and αLHL and αInj are the percentages of expected
human losses and injuries, respectively.
In summary, rebuilding costs are based on the cost per square meter of deck surface
(based on Stein et al. 1999), while rehabilitation costs depend on the level of damage affect-
ing the bridge. It has been assumed that rehabilitation costs should not exceed 60% of the
construction costs, otherwise the bridge would be replaced (based on Sánchez-Silva and
Rackwitz 2004). Material damage costs are associated with material losses in terms of
the residual values of cars and trucks involved in potential accidents when a sudden failure
occurs. Operating costs are generated when users travel through a bypass detour that is longer
than the original route while time-loss costs are associated with the marginal time loss for
users and goods while traveling through a partially serviceable bridge or through the bypass
detour (based on Stein et al. 1999). Costs due to the loss of human lives and injuries
are associated with the number of expected fatalities and injuries due to a potential sudden
failure, respectively.
The time needed for the restoration of bridge serviceability is assumed as: 36 months for
ADT ≤ 100; 24 months for 100 < ADT ≤ 500; 18 months for 500 < ADT ≤ 1,000; 12
months for 1,000 < ADT ≤ 5,000; and 6 months for ADT > 5,000 (Stein et al. 1999). Regard-
ing the costs due to the loss of human life, the VSL includes out-of-pocket costs, lost work
and housework-related costs, human capital costs, and the expenses people are willing to face
to prevent injury, according to Caltrans (2009) and USDOT (2011). Meanwhile, as per Spicer
and Miller (2010), costs associated with injuries are a fraction of VSL depending on the
injury severity, based on the concept of Quality Adjusted Life Years (QALYs). Most of
the parameters included in the evaluation of costs are affected by uncertainties, hence
they are treated as random variables.
The estimation of the traffic volume of a route containing a bridge is provided by the
ADT information of the bridge, which can be obtained from the National Bridge Inventory
(NBI) data file (FHWA 2010). The ADT value varies over time. The magnitude of this var-
iation can be assessed by investigating NBI data files referring to different years. The pro-
posed model for the prediction of ADT, associated with the j-th bridge, is assumed based on a
linear estimator function and on a time interval of 10 years as follows:
$ #
ρj t
¼ ⋅ 1þ Aj ðtÞ Aj ½k' (20)
1000
# j $
j A ½k þ 10'
ρ ¼ − 1 ⋅ 100 (21)
Aj ½k'
where A½k' is the ADT value referred to year k, ρ is the percentage of traffic increment
between two available ADTs spaced 10 years apart, and t is the time (years). If ADT is
expected to decrease over time, ρ is negative.
The speed SðtÞ required to cover a certain route distance depends on the route capacity
and traffic conditions. According to Martin and McGuskin (1998), for each considered route
or detour associated with the j-th bridge, such speed is defined as:
Sj0
Sj ðtÞ ¼
# $4 (22)
F j ðtÞ
1 þ 0.15 F j nj
C L
where S0 is the free-flow traffic speed, FðtÞ is the total traffic flow, F C is the bridge traffic
flow capacity (usually considered to be 2,000 and 1,000 vehicles per hour per lane for high-
ways and minor arterials, respectively), and nL is the number of lanes carried by the bridge.
The total traffic flow can be obtained by the ADT associated with each bridge. Based on TRB
(2000) and Caltrans (2009), the relationship between traffic flow and ADT accounting for
peak and off-peak hours is defined as:
!
j K 1 DAj ðtÞ peak hours
F ðtÞ ¼ (23)
K 2 DAj ðtÞ off − peak hours
where K 1 and K 2 are the percentages of traffic associated with the portion of ADT during the
peak and off-peaks hours, respectively, and D is the directional factor. Accordingly, the con-
sequence analysis is performed for these two time periods considering every single day.
LIFE-CYCLE RISK ASSESSMENT

The identification of the most dangerous hazards depends on the level of detail that a
specific risk assessment requires. The required scale of specific approaches determines
the level of the analysis. Moreover, the identification of critical scenarios to be considered
depends on the desired accuracy level. In this paper, the effects of hazards induced by earth-
quakes and traffic are investigated for a group of bridges belonging to a transportation net-
work. Since structural vulnerabilities and consequences are time-dependent, lifetime
considerations are necessary.
SERVICEABILITY
Risk assessment is based on the level of serviceability of the single bridges within a
transportation network. Bridge serviceability can be defined as the ability to provide adequate
service to users. Although risk is affected when failure occurs, other circumstances may put
the bridge at risk. For instance, the closure of one traffic lane causes indirect consequences in
terms of traffic delay. The vehicular traffic capacity and the speed required to cover the link
associated with the j-th bridge are affected by the level of serviceability of the bridge. Dif-
ferent service levels have already been investigated in previous studies (Adey et al. 2003,
Bocchini and Frangopol 2011). Accordingly, a bridge may be subjected to interventions in
order to improve its performance with respect to the considered hazard-induced effects. Ser-
viceability may vary over time depending on the performance level and damage level experi-
enced by the bridge during its lifetime. In this paper, the bridge serviceability is classified into
four main service states:
• Fully serviceable (S1): The bridge is open and fully available to users.
• Partially serviceable (S2): The bridge is only partially available to users due to
minor rehabilitation work. Some of the lanes are closed to traffic and these condi-
tions potentially increase delay for travelers.
• Closed (S3): The bridge is closed due to major rehabilitation work. This restriction
causes travelers to drive through a bypass detour, thus generating loss of time and
longer travel distance.
• Collapsed (S4): The bridge needs to be rebuilt. The vehicular traffic must follow the
bypass detour.
It is assumed that a fully serviceable bridge does not generate any kind of negative con-
sequences, and users traveling on the associated link reach their destination without being
affected by any loss of time.
RISK DEFINITION
Depending on the level of performance, the bridge serviceability will be associated with
probabilities of being in one of the four service states(S1 to S4) defined previously. For both
seismic and traffic hazards, the probabilities of being in a specific service state for the j-th
bridge at time t are summarized in Table 2, where PE;S1 ðtÞ and PT;S1 ðtÞ, PE;S2 ðtÞ and PT;S2 ðtÞ,
PE;S3 ðtÞ and PT;S3 ðtÞ, and PE;S4 ðtÞ and PT;S4 ðtÞ are the probabilities of the bridge being in
service state S1, S2, S3, and S4, respectively, due to seismic (subscript E) and traffic (sub-
script T) hazards; P½ND'ðtÞ, P½SL'ðtÞ, P½MOD'ðtÞ, P½EXT'ðtÞ, and P½COM'ðtÞ are the prob-
abilities of having no damage ND, slight damage SD, moderate MOD, extensive damage
EXT, and complete damage COM, respectively; P½S1jSurv'ðtÞ, P½S2jSurv'ðtÞ, and
P½S3jSurv'ðtÞ are the probability of being in service states S1, S2, and S3, respectively,
given that the bridge does not fail; PT; f ðtÞ is the probability of failure; and t is the time.
For a bridge, the events of being in specific service states are mutually exclusive events
within the single hazard.
The event tree of Figure 1 shows how serviceability states, traffic capacity, bridge vul-
nerability, and consequences are considered. The values of the conditional probabilities
P½S1jSurv', P½S2jSurv', and P½S3jSurv', of Equations 28–30 in Table 2, are based on the
percentage of time for which a bridge experiences full service, partial service, or closure,
respectively. Such values can be obtained by considering the yearly duration of maintenance
works (such as preventive or essential maintenance).
Table 2. Probabilities for a bridge of being in specific damage states in case of seismic
(subscript E) and traffic (subscript T) hazards.
Hazard type Service state Probability Eq. no.
S1 PjE;S1 ðtÞ ¼ P½NDj 'ðtÞ

(24)
S2 PjE;S2 ðtÞ ¼ P½SLj 'ðtÞ (25)
Seismic
S3 PjE;S3 ðtÞ ¼ P½MODj 'ðtÞ þ P½EXT j 'ðtÞ

(26)
S4 PjE;S4 ðtÞ ¼ P½COM j 'ðtÞ

(27)
S1 PjT;S1 ðtÞ ¼ P½S1j jSurvj 'ðtÞ ⋅ ½1 − PjT;f ðtÞ'

(28)
S2 PjT;S2 ðtÞ
EQ-TARGET;temp:intralink-;e29;214;449 ¼ P½S2j jSurvj 'ðtÞ ⋅ ½1 − PjT;f ðtÞ' (29)
Traffic
S3 PjT;S3 ðtÞ ¼ P½S3j jSurvj 'ðtÞ ⋅ ½1 − PjT;f ðtÞ'
(30)
S4 PjT;S4 ðtÞ
EQ-TARGET;temp:intralink-;e31;214;425 ¼ PjT;f ðtÞ (31)
Finally, recalling that sevice state S1 does not cause any consequences, the proposed
life-cycle risk definitions for seismic and traffic hazards are, respectively, as follows:
S4 !
X " %&
RjE ðtÞ ¼
EQ-TARGET;temp:intralink-;e32;62;351 PjE;Si ðtÞ ⋅ C jD;Si ðtÞ þ CjI;Si ðtÞ (32)
Si¼ S1
S4 !
X " %&
RjT ðtÞ ¼
EQ-TARGET;temp:intralink-;e33;62;303 PjT;Si ðtÞ ⋅ C jD;Si jðtÞ þ CjI;Si ðtÞ (33)
Si¼ S1
where CD;Si ðtÞ and CI;Si ðtÞ are the time-dependent direct and indirect consequences,
respectively, given a specific serviceable state Si.
In a bridge, although a specific service state due to seismic hazard and a specific service
state due to traffic hazard are statistically independent events, their potential conjunct occur-
rence cannot be excluded, even though it is unlikely to happen. If this occurs, the conse-
quences are determined by those associated with the worse service state. Therefore, the
life-cycle total risk RTot ðtÞ for the j-th bridge is:
X S4 !" % &
j j j j
RTot ðtÞ ¼ PE;Si ðtÞ þ PT;Si ðtÞ ⋅ C Tot;Si ðtÞ −
Si¼ S1
S4 S4 ! " %& (34)
X X
j j j j
PE;Sk ðtÞ ⋅ PT;Sn ðtÞ ⋅ max CTot;Sk ðtÞ; CTot;Sn ðtÞ
Sk¼ S1 Sn¼ S1
Figure 1. Event tree summarizing bridge serviceability under seismic and traffic hazards.
Different consequences (Equations 13–19 of Table 1) are associated with the serviceability states
(S1 to S4).
where C Tot;Si ðtÞ, CTot;Sk ðtÞ, and C Tot;Sn ðtÞ are the time-dependent total consequences for each
service state Si, Sk, and Sn, respectively. The introduced formula accounts for the cases in
which the bridge is affected by a specific service state due to a single hazard or to the con-
junction of both.
BRIDGE NETWORK
The approach for the evaluation of risk of a complex transportation network requires the
study of traffic flow distribution within it. For instance, graph theory may be used to assign an
origin-destination matrix for each vertex included in the model, and bidirectional links con-
necting the vertex represent the traffic flow directions (Gibbons 1985). Although such a
method is powerful, the assumption that all the travelers remain within the considered net-
work may lead to approximations. Bocchini and Frangopol (2011) enhanced graph theory
applied to transportation networks, including the effect on traffic flows through the quanti-
fication of the level of damage of the bridges carrying the links.
For some small networks, a good approximation for the evaluation of network risk consists
in considering the actual traffic flow traveling on each bridge, which can be obtained by Equa-
tion 23. In this way, and according to Cesare et al. (1993), life-cycle total risk for the network
RNet may be evaluated as the sum of the risks generated by all the single bridges as:
B
X
RNet ðtÞ ¼
EQ-TARGET;temp:intralink-;e35;41;214 RjTot ðtÞ (35)
j¼ 1
where B is the total number of bridges within the network. Equation 35 can provide
indications on the risk associated with some small transportation networks, whereas for com-
plex ones, the use of more sophisticated methods is required, including those presented for
network reliability assessment (Liu and Frangopol 2005, 2006).
Figure 2 shows the developed flowchart for the assessment of the life-cycle network risk.
Generate samples for all the considered
random variables
Generate a seismic event associating a
Probabilistic Ground Motion

random magnitude according to Eq. 1
Define the random location, depth, rupture

orientation angle, rupture length, and width
(Eqs. 2 and 3) for each seismic event
Evaluation of the shortest distance between

the considered bridge and the rupture trace
Traffic Hazard
Evaluation of the median values of PGA and
Fragility Analysis
PSA by using the attenuation function Evaluation of the
contained in Eqs. 4 and 5 probability of failure
of the superstructure
due to traffic hazard
Evaluation of the probability of occurrence (Eq. 11)
of specific damage states Eq. 7
Evaluation of traffic conditions and detour

Consequence
length
Analysis
Evaluation of direct and indirect

consequences (Eqs. 13 to 19 of Table 1)
Serviceability
Analysis
Evaluation of the probability of a bridge in

specific serviceability state (S1 to S4)
Evaluation of the bridge risk induced by

seismic and traffic hazards depending on its
serviceability state (Eqs. 32 to 34)
NO Is the total number of

Risk Analysis
bridges reached?
YES
NO Is the total number of
sample reached?
YES
Evaluation of the total risk of the bridge
network associated with seismic and traffic
hazards depending on bridge serviceability
(Eq. 35)
Figure 2. Flowchart for the computation of the life-cycle risk for a network under seismic and
traffic hazards.
CASE STUDY: GROUP OF EXISTING BRIDGES

The proposed approach is applied to a group of existing bridges in order to investigate the
life-cycle risk caused by seismic and traffic hazards. Figure 3 shows the schematic layout of
the considered group of highway and secondary bridges within a specific network. A segment
of the northbound of Interstate 15 belonging to the city of Escondido, California (north of the
San Diego metropolitan area), and a segment of the northbound of the Centre City Parkway
form the considered one-way network. There are two origin points, A1 and A2 , and one des-
tination point, E. The intersection of the two routes is in point C. Five bridges denoted as 1, 2,
3, 4, and 5 are included in this analysis.
A Monte Carlo simulation is the core of the developed program performed by the soft-
ware MATLAB (MathWorks 2009). Based on simulated samples generated by the Latin
Hypercube technique (McKay et al. 1979), histograms, expected values, and specific
percentiles of life-cycle risk are evaluated for a group of bridges. According to McKay
et al. (1979) the generation of random numbers by using the Latin Hypercube technique
is a valid alternative to crude Monte Carlo simulation, providing a better represented sample
space with a limited number of samples.
BRIDGE VULNERABILITIES
Time-dependent structural performances by means of structural vulnerability are evalu-
ated for seismic and traffic-induced hazards. Over time, structural vulnerability increases due
to aging considerations. All the obtained profiles span over 70 years, starting from 1978 (year
ESCONDIDO, CA
E
5 0 1
Mile
0 1
D Kilometer
4
Legend
C
Road links
2
Points connecting
the links
Interstate
3
Bridge positions
N A2
A1
Figure 3. Schematic layout of the transportation network.

50 50
42 40
Shortest
30 distance
y-axis
y-axis - Triangular distribution
5 projection
20
10 4 Orientation
y-axis (km)
2
0
3
-10 x-axis
1
-20 Transportation
Legend
Network
-30 Rupture traces
Epicenter positions
-40
PDF Rupture extreme points
-50 -50
0.02 -50 -40 -30 -20 -10 0 10 20 30 40 50
-50 x-axis (km) 50
PDF
0.02
x-axis - Triangular distribution
Figure 4. Graphical representation of random epicenters, rupture traces, and bridge network over
the considered region. The triangular distributions of the x−y axes and the representation of the
projection of the shortest distance between a bridge and the rupture trace for different rupture
locations and orientations are also reported.
in which the last bridge of the network has been built according to the collected NBI data
(FHWA 2010)).
The effect of seismic hazard is modeled herein by means of fragility analysis (Basöz and
Mander 1999). The ground motion at each bridge location is simulated starting from 20,000
earthquakes randomly generated over a square area having a width of 100 km. From a parallel
analysis performed by using 10,000 samples it was found that the difference in terms of mean
total network risk is about 1%; therefore, it was deduced that 20,000 samples satisfy an
acceptable level of convergence. The center of gravity of this region has the following coor-
dinates: Lat: ¼ 33.16° and Long: ¼ −117.10°. As shown in Figure 4, the epicenter coordi-
nates (x and y) of the random seismic events are triangularly distributed between −50 km and
50 km with mode of 50 km, for the x-axis, and −50 km and 50 km with mode of 42 km for the
y-axis, respectively (see Figure 4). The magnitude of the random seismic events is provided
by a sample obtained from a truncated exponential PDF (β in Equation 1 has been set to be
equal to 1.1). The depth of the seismic source is triangularly distributed between 1 km and
30 km with mode of 8 km. The parameter β and the parameters of the triangular distributions
(x-y axes and depth) are obtained by using the method of moments (Ang and Tang 2007)
based on the analysis of the historic earthquakes database for the region under investigation
(USGS 2011). The minimum and maximum magnitudes (Richter scale) considered are set as
4 and 8, respectively. In this paper, strike-slip earthquakes are randomly generated causing
rupture lengths and widths evaluated according to Equations 2 and 3, respectively. Consider-
ing the predominant direction (southeast to northwest) of the faults in the area, the orientation
of the rupture traces are uniformly distributed with angles between 90° and 180°. For
instance, Figure 4 show the graphical representation of rupture traces over the considered
region for 50 samples. The rupture traces are represented by solid lines in the plane x-y
(inspired by Jones et al. 2005). Epicenters and extreme points of the traces are shown as
well. The projection of the shortest distance between a bridge and the random rupture
trace is evaluated depending on rupture location and orientation.
Each bridge is affected by ground motion depending on many factors accounted in the
attenuation functions proposed by Campbell and Bozorgnia (2007). Once a set of PSA values
is evaluated for the j-th bridge and for the i-th sample, time-dependent fragility analysis can
be performed and exceedance probabilities of various damage states are calculated. The
initial values of the fragility parameters are those calibrated by Basöz and Mander
(1999). Further needed information and parameters concerning the bridges are found in
the NBI database (FHWA 2010). By using Equation 8, the dependency of the median spectral
acceleration ADS ðtÞ associated with the occurrence of a specific damage state is evaluated by
introducing the aging coefficient γ. The value of γ is set in order to obtain an assumed reduc-
tion of about 25% of the fragility parameter after 70 years (consistent with the results reported
in Ghosh and Padgett 2010). The result of this seismic analysis is a set of time-dependent
probabilities of the bridges being in specific damage states: no damage P½ND'ðtÞ, slight
damage P½SL'ðtÞ, moderate damage P½MOD'ðtÞ, extensive damage P½EXT'ðtÞ, and complete
damage P½COM'ðtÞ.
The vulnerability due to traffic hazard is evaluated by assuming that the PDF of the time-
to-failure follows a Weibull distribution with parameters κ and λ. For this network analysis, κ
and λ are assumed to be equal to 3 and 0.01 for bridges 1 and 2, respectively, and 5 and 0.01
for bridges 3, 4 and 5, respectively. These values are consistent with those reported in
previous studies (van Noortwijk and Klatter 2004, Yang et al. 2004). The time-dependent
probability of failure of the superstructure is obtained by applying Equation 11.
EVALUATION OF THE CONSEQUENCES

The time-dependent monetary values associated with direct and indirect consequences
are evaluated at this stage. Total costs depend on the knowledge of unitary costs (such as cost
per square meter, cost per hour, etc.). Direct costs are those associated with structural and
material losses (Equations 13–15 of Table 1). Rehabilitation costs are also defined as direct
because they correspond to the amount of money that is needed towards the renewal of
damaged structures. Indirect costs are associated with the loss of time, the marginal traveled
distance, and with the losses due to the number of potential fatalities and injuries (Equa-
tions 16–19 of Table 1). The cost, for the years preceding 2011, are provided by Equation 12
by setting a negative time interval t, while the cost values for the years following 2011 are
obtained by considering the future cost, using a positive time interval t in Equation 12.
Some of the costs depend on the ADT of the road carried by the considered bridges. The
prediction of the ADT may be provided by Equation 20, which considers their increment or
decrement percentage within a time interval of 10 years by means of ρ (Equation 21). The
speed of the traffic flow (Equation 22) depends on the practical capacity of the route and on
the total traffic flow over time (Equation 23). According to Caltrans (2009), the peak period
lasts five hours per day and the percent coefficients K 1 and K 2 of Equation 23 are equal to
8.48% and 3.03%, respectively. Upon these considerations, the traveling speed and the
marginal travel time within the network are accordingly affected depending also on the route
the users are using (normal route or bypass detour).
All the assumed values of the included basic statistical descriptors and their distribution
types are summarized in Table 3. According to Ang (2010), a practical evaluation of epis-
temic uncertainty may usually rely on intuitive/engineering judgments, which lead to a spe-
cific reasonable range of possibilities, with an associated plausible distribution.
Latin Hypercube (McKay et al. 1979) sampling is used with 20,000 trials for the basic
random variables, allowing the probabilistic calculation of the consequences, and for the
simulation of the ground motion.
LIFE-CYCLE RISK
Seismic and traffic vulnerabilities determine the service states of each bridge (S1 to S4).
Although losses are generated when a structural failure occurs (S4), service states S2 and S3
may also cause both direct and indirect consequences.
By applying Equations 24–35, life-cycle risks can be calculated. Given that a bridge
survives, conditional probabilities of being in service states S1, S2, and S3 for traffic hazard
are assumed to be 0.94, 0.05, and 0.01, respectively. Such values are based upon considera-
tions on the annual percentage of time for which a bridge is open or subjected to rehabilitation
actions requiring its closure or partial closure. Figure 5 shows risk referred to the year 2011
induced by seismic (a, b, c, d, and e) and traffic hazards (f, g, h, i, and j) for each considered
bridge of the transportation network in Figure 3. It can be noted that risk of bridges 1 and 2 is
relatively smaller than risk for the other three bridges carrying the northbound of I-15. The
results shown in Figure 5 can provide the sensitivity analysis of risk associated with different
routes within the network. The predicted histograms of life-cycle risks referred to the year
2048 and, based on the simulated earthquakes and consequences, are shown in Figure 6.
According to the proposed framework, it can be noticed that the expected predicted risk
at year 2048 can be three times higher than the risk associated with the year 2011. Moreover,
although ADT is predicted to slightly increase over time and considering the future value of
money (Equation 12), seismic risk significantly increases over time for each considered
bridge (Figures 5 and 6). It was found that seismic risk is very sensitive to the decay of
the fragility parameters; therefore seismic life-cycle analyses must address these aspects.
From a parallel analysis, it is found that, based on a maximum reduction over time of
the fragility parameters of about 25%, time-dependent risk is about 44% higher than risk
obtained with time-independent fragility parameters. Both Figures 5 and 6 help analyze dif-
ferent routes for the small network studied in this paper, but for complex networks, a com-
prehensive method that includes traffic redistribution must be considered in order to account
for the increasing number of alternative routes.
Figure 7a and 7b show the profiles of the mean value of direct and indirect life-cycle
risks, respectively, for the five bridges of the network. As shown, direct risks are consistent
with bridge dimensions and types, indicating that bridge 1 is the smallest among the others,
while indirect risk is found greater for those bridges carrying more traffic (i.e., 3, 4, and 5),
causing more disruption in case of closures or failures. Figure 7c and 7d report on the mean
and dispersion of risk showing the profiles (mean and 90% and 95% percentiles) associated
with direct and indirect risks for bridge 5 of Figure 3. Based on the proposed model, it can be
Table 3. Statistical descriptors and deterministic parameters used for the evaluation of
the consequences. The costs refer to their values at year 2011, which corresponds to year 33 of the
profiles.
Distribution
Random variables Mean COV type References
ADT Varies DNA DNA FHWA (2010)
ADTT/ADT ratio Varies DNA DNA FHWA (2010)
Compensation for truck drivers ðUSD∕hÞ 29.28 0.31 LN AASHTO
(2003)
Detour speed ðkm∕hÞ 30 0.15 LN Assumed
Vehicle occupancies for cars 1.50 0.15 LN AASHTO
(2003)
Vehicle occupancies for trucks 1.05 0.15 LN AASHTO
(2003)
Wage for car drivers ðUSD∕hÞ 23.36 0.28 LN AASHTO
(2003)
Epistemic uncertainty factor for injury costs 1 0.30 LN Assumed
Epistemic uncertainty factor for loss of human 1 0.30 LN Assumed
life costs
Epistemic uncertainty factor for material damage 1 0.20 LN Assumed
costs
Epistemic uncertainty factor for material damage 1 0.20 LN Assumed
costs
Epistemic uncertainty factor for rebuilding costs 1 0.20 LN Assumed
Epistemic uncertainty factor for rehabilitation 1 0.20 LN Assumed
costs
Epistemic uncertainty factor for operating costs 1 0.20 LN Assumed
Epistemic uncertainty factor for time loss costs 1 0.20 LN Assumed
Fixed rehabilitation costs (USD) 50,000 0.20 LN Assumed
Length of detour ðkmÞ Varies DNA DNA FHWA (2010)
Value of a statistical life (USD) 6,200,000 0.45 LN USDOT
(2011)
Operating costs for cars ðUSD∕kmÞ 0.40 0.19 LN AASHTO
(2003)
Operating costs for trucks ðUSD∕kmÞ 0.56 0.19 LN AASHTO
(2003)
Inventory costs ðUSD∕hÞ 3.81 0.20 LN AASHTO
(2003)
Min. Max.
Annual discount rate (%) 2 3 Uniform Assumed
Injury costs (USD) 651,000 6,200,000 Uniform USDOT
(2011)
Rebuilding costs ðUSD∕m2 Þ 993 − 1,655 − Uniform Caltrans
1,324 2,207 (2010)
Note: LN = log-normal distribution; COV = coefficient of variation; DNA = does not apply.
15400 5000
15200 Seismic Hazard Traffic Hazard
15000 4000
Bridge 1 Bridge 1
Samples
1000 3000
800
600 Year 2011 2000 Year 2011
400 20,000 Samples 20,000 Samples
1000
200
0 0
0 1 2 3 4 5 0.0 0.2 0.4 0.6 0.8 1.0
(a) (f)
16200 5000
15800 4000
Bridge 2 Bridge 2
Samples
1000 3000
800
600 Year 2011 2000 Year 2011
1000
200
0 0
0 1 2 3 4 5 0.0 0.2 0.4 0.6 0.8 1.0
(b) (g)
16800 5000
4000
1000 Bridge 3 Bridge 3
Samples
800 3000
600 Year 2011 2000 Year 2011
1000
200
0 0
0 10 20 30 40 0.0 0.5 1.0 1.5 2.0
(c) (h)
15600 5000
15400 Seismic Hazard 4000
Traffic Hazard
Samples
800 3000
600 Year 2011 2000 Year 2011
1000
200
0 0
0 10 20 30 40 0.0 0.5 1.0 1.5 2.0
(d) (i)
15800 5000
4000
Samples
800 3000
600 Year 2011 2000 Year 2011
1000
200
0 0
0 10 20 30 40 0.0 0.5 1.0 1.5 2.0
Total Seismic Risk (USD millions) Total Traffic Risk (USD millions)
(e) (j)
Figure 5. Histograms of the current total risk (direct and indirect) due to (a, b, c, d, e) seismic
hazard and (f, g, h, i, j) traffic-induced hazard for each bridge within the network at the year 2011.
14800 6000
14400
Seismic Hazard 5000 Traffic Hazard
Bridge 1 Bridge 1
Samples
4000
1200
3000
800 Year 2048 Year 2048
2000
20,000 Samples 20,000 Samples
400 1000
0 0
0 2 4 6 8 10 12 14 16 0.0 0.5 1.0 1.5 2.0 2.5 3.0
(a) (f)
15200 6000
14800
Seismic Hazard 5000 Traffic Hazard
Bridge 2 Bridge 2
Samples
4000
1200
3000
800 Year 2048 Year 2048
2000
20,000 Samples 20,000 Samples
400 1000
0 0
0 2 4 6 8 10 12 14 16 0.0 0.5 1.0 1.5 2.0 2.5 3.0
(b) (g)
16400 4000
16000
Seismic Hazard Traffic Hazard
Bridge 3 3000 Bridge 3
Samples
1200
2000
800 Year 2048 Year 2048
20,000 Samples 1000 20,000 Samples
400
0 0
0 20 40 60 80 100 120 140 0 2 4 6 8 10
(c) (h)
15200 4000
Samples
1200
2000
800 Year 2048 Year 2048
400
0 0
0 20 40 60 80 100 120 140 0 2 4 6 8 10
(d) (i)
15200 4000
Samples
1200
2000
800 Year 2048 Year 2048
400
0 0
0 20 40 60 80 100 120 140 0 2 4 6 8 10
Total Seismic Risk (USD millions) Total Traffic Risk (USD millions)
(e) (j)
Figure 6. Histograms of the predicted total risk (direct and indirect) due to (a, b, c, d, e) seismic
hazard and (f, g, h, i, j) traffic-induced hazard for each bridge within the network at the year 2048.
400 1200
Mean Seismic Direct Risk 20,000 Seismic Direct Risk 20,000
350
+ Samples 1000 + Samples
Direct Risk (USD tousands)
Direct Risk (USD tousands)

300 Mean Traffic Direct Risk Traffic Direct Risk
Bridge 4 800
Mean Life-cycle
250 Percentile 95%
Life-cycle
Bridge 3
200 600
Bridge 5 Percentile 90%
150 Mean total
Bridge 2 400 direct risk
100 Bridge 1
200
50 Year 2011 Year 2011
Bridge 5
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (years) Time (years)
(a) (c)
10 45
9 Mean Seismic Indirect Risk 20,000 40 Seismic Indirect Risk 20,000
+ Samples + Samples
8
Indirect Risk (USD millions)
Indirect Risk (USD millions)
Mean Traffic Indirect Risk 35 Traffic Indirect Risk

7
30
Mean Life-cycle
Percentile 95%
6 Bridge 5
Life-cycle
25
5
Bridge 4 20 Percentile 90%
4
Bridge 3 15 Mean total
3 direct risk
Bridge 2
2 10
Bridge 1
1 Year 2011 5 Year 2011
Bridge 5
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (years) Time (years)
(b) (d)
Figure 7. Profiles of the mean values of (a) direct and (b) indirect life-cycle risks of each bridge
within the transportation network due to seismic and traffic hazards. Profiles of the mean values
and 90% and 95% percentiles of (c) direct and (d) indirect life-cycle risks for the selected bridge 5
of Figure 3 due to seismic and traffic hazards.
concluded that total life-cycle risk is very sensitive to the indirect consequences that are
significantly larger than the direct ones. Accordingly, higher dispersion has been found
for indirect risk, which involves a great number of random variables adding further uncer-
tainty. The contribution of the single types of consequences to total risk for the entire network
is illustrated in Figure 8. Mean value profiles of all the types of direct (Figure 8a) and indirect
(Figure 8b) risks are computed. The most significant loss contributions are found in reha-
bilitation costs and time-loss costs. Therefore, the parameters involved in the analysis of the
costs associated with the loss of time require special attention for their estimation. A relevant
outcome to be noticed is that, according to the proposed model, expected rehabilitation costs
0.8
Direct Life-cycle Risks (USD millions)
Seismic Direct Risk Direct Network Risk

0.7
+
Traffic Direct Risk
0.6 Mean risk due to
rehabilitation costs
0.5 20,000 Samples
Mean risk due to
0.4 rebuilding costs
0.3 Mean risk due to
material damage costs
0.2
0.1
0.0
0 10 20 30 40 50 60 70
(a)
20
Indirect Life-cycle Risks (USD millions)
18 Seismic Indirect Risk Indirect Network Risk

+
16 Traffic Indirect Risk
14 Mean risk due to time loss
20,000 Samples
12
10
8 Mean risk due
Mean risk due to
to fatalities
6 operating costs Mean risk due
4 to injuries
2
0
0 10 20 30 40 50 60 70
Time (years)
(b)
Figure 8. Profiles of the mean values of life-cycle network risks associated with (a) direct
consequences: rebuilding costs, rehabilitation costs, and material damage costs; and (b) indirect
consequences: operating costs, time-loss costs, loss of human life costs, and injury costs.
are greater than expected rebuilding costs depending on bridges and network service states
(Figure 8a).
Finally, the profiles of the life-cycle network risk are obtained. In Figure 9a, the risk
contribution of each bridge with respect to the network risk is reported. Figure 9b shows
the profiles of the mean value and 90% and 95% percentiles of the total life-cycle risk asso-
ciated with the whole bridge network. Percentiles give an indication of the dispersion of the
obtained data and can set the basis for the calibration of life-cycle risk-informed management
of bridges within a transportation network.
30
Total Life-Cycle Risk
25
20,000 Samples Bridge 5
Bridge 4
Risk (USD millions)
20
Total Life-Cycle
Bridge 3
Bridge 2
15
Bridge 1
10
Mean total network risk
0
0 10 20 30 40 50 60 70
Time (years)
(a)
150
Total Life-Cycle Network Risk
Network Risk (USD millions)
20,000 Samples
Percentile 95%
100
Total Life-Cycle
Percentile 90%
50
Mean total network risk
0
0 10 20 30 40 50 60 70
Time (years)
(b)
Figure 9. (a) Profile of total life-cycle network risk including the risk contribution of each bridge
within the network. (b) Profiles of the mean value and 90% and 95% percentiles of total life-cycle
network risk.
CONCLUSIONS
Recently, life-cycle management of structures and infrastructures addressed issues related
to risk-informed decision analysis (Ellingwood 2005). In this context, decisions may be
based on the quantification of monetary values of most of the consequences caused by poten-
tial structural failures or malfunctions.
This paper presented some important issues emerging from the life-cycle risk assessment
of spatially distributed aging bridges within a transportation network. A detailed consequence
analysis has been performed in order to evaluate the effects associated with different bridge
service states (serviceability). Two of the most common hazards are considered in this study.
The time-dependent effects of traffic hazard, including aging phenomena and seismic hazard,
for an earthquake-prone region have been investigated.
The following conclusions are obtained:
1. When life-cycle risk is assessed under multiple hazards, the predominant hazard
must be adequately modeled in order to achieve reliable estimates of total risk.
The current case study helps to understand that, for the considered region, it is cru-
cial to model the seismic hazard (earthquake scenarios) with a powerful tool such as
the probabilistic ground motion and associated fragility analysis. In fact, for the
studied network, it was found that the only inclusion of risk induced by traffic
hazard provides an incomplete analysis that may lead to a wrong decision process.
2. As expected, the risk induced by seismic hazard increases over time due to structural
aging consideration. However, a maximum reduction over time of the fragility para-
meters of about 25% leads to a much higher increase in risk (about 44% higher than
risk obtained with time-independent fragility parameters). Additional risk increment
is due to the predicted increase of ADTs over time.
3. The failure of any of the three bridges belonging to the I-15 (i.e., bridges 3, 4, and 5
in Figure 3) generates dramatically high consequences. Within the network, risk for
the bridges belonging to Centre City Parkway (i.e., bridges 1 and 2 in Figure 3) is
almost negligible.
4. It can be noted that the dispersion of seismic risk increases over time depending on
the uncertainty affecting the prediction. Histograms and percentiles give an indica-
tion of such dispersion in the obtained risk profiles. The inclusion of epistemic
uncertainties enhances the probabilistic model but adds further dispersion to the
initial sample.
5. According to the proposed framework, it is found that expected rehabilitation costs
are greater than expected rebuilding costs. Whilst this is true in this case, this obser-
vation may change depending on the extent of damage, on the method of computa-
tion, and on the funding model.
The implementation of risk profiles into a life-cycle framework helps the development of
risk-informed optimal maintenance planning. The provided approach is useful for the
evaluation of risk associated with single bridges within small networks that may be ranked.
Based on this, appropriate maintenance prioritization plans could be adopted (Zhu and
Frangopol 2013). However, for complex networks, methods that include traffic redistribution
must be considered.
ACKNOWLEDGMENTS
The support from the National Science Foundation through grant CMS-0639428
and the U.S. Federal Highway Administration Cooperative Agreement Award
DTFH61-07-H-00040 is gratefully acknowledged. The opinions and conclusions presented
in this paper are those of the authors and do not necessarily reflect the views of the
sponsoring organizations.
REFERENCES
American Association of State Highway and Transportation Officials (AASHTO), 2003. A Manual
of User Benefit Analysis for Highways, 2nd edition, Washington, D.C., 288 pp.
Adey, B., Hajdin, R., and Brühwiler, E., 2003. A risk-based approach to the determination of
optimal interventions for bridges affected by multiple hazards, Engineering Structures 25,
903–912.
Akgül, F., 2002. Lifetime System Reliability Prediction for Multiple Structure Types in a Bridge
Network, Ph.D. Thesis, University of Colorado at Boulder.
Akiyama, M., Frangopol, D. M., and Matsuzaki, H., 2011. Life-cycle reliability of RC bridge
piers under seismic and airborne chloride hazards, Earthquake Engineering & Structural
Dynamics 40 , 1671–1687.
Ang, A. H-S, 2010. Significance of uncertainty in the calculation of risk and reliability: Emphasis
on Decisions for protection against natural hazards, International Symposium on Reliability
Engineering and Risk Management (ISRERM2010), 23–26 September 2010, Shanghai, China.
CD-ROM.
Ang, A. H-S, and Tang, W.H., 2007. Probability Concepts in Engineering: Emphasis
on Applications to Civil and Environmental Engineering, 2nd edition, Wiley and Sons,
New York, 406 pp.
Basöz, N., and Mander, J., 1999. Enhancement of the Highway Transportation Lifeline Module in
HAZUS, National Institute of Building Sciences (NIBS), Washington, D.C.
Bocchini, P., and Frangopol, D. M., 2011. Generalized bridge network performance analysis with
correlation and time-variant reliability, Structural Safety 33, 155–164.
Bocchini, P., and Frangopol, D. M., 2012a. Restoration of bridge networks after an earthquake:
Multi-criteria intervention optimization, Earthquake Spectra 28, 426–455.
Bocchini, P., and Frangopol, D. M., 2012b. Connectivity-based optimal scheduling for mainte-
nance of bridge networks, Journal of Engineering Mechanics, DOI: 10.1061/(ASCE)
EM.1943-7889.0000271.
California Department of Transportation (Caltrans), 2009. California Life-Cycle Benefit/Cost
Analysis Model (Cal-B/C), Volume 3, Sacramento, CA.
California Department of Transportation (Caltrans), 2010. Comparative Bridge Costs, Sacra-
mento, CA.
Campbell, K. W., and Bozorgnia, Y., 2007. Campbell-Bozorgnia NGA Ground Motion Relations
for the Geometric Mean Horizontal Component of Peak and Spectral Ground Motion Para-
meters, PEER Report 2007/02, Pacific Earthquake Engineering Research Center, University of
California, Berkeley, CA.
Cesare, M., Santamarina, J. C., Turkstra, C. J., and Vanmarcke, E., 1993. Risk-based bridge
management, Journal of Transportation Engineering 119, 742–750.
Chang, S. E., Shinozuka, M., and Moore, J. E., 2000. Probabilistic earthquake scenarios:
Extending risk analysis methodologies to spatially distributed systems, Earthquake
Spectra 16, 557–572.
Cornell, C. A., 1968. Engineering seismic risk analysis, Bulletin of Seismological Society of
America 58, 1583–1606.
Cosentino, P., Ficara, V., and Luzio, D., 1977. Truncated exponential frequency-magnitude
relationship in earthquake statistics, Bulletin of Seismological Society of America 67,
1615–1623.
Decò, A., and Frangopol, D. M., 2011. Risk assessment of highway bridges under multiple
hazards, Journal of Risk Research 14, 1057–1089.
Ellingwood, B. R., 2005. Risk-informed condition assessment of civil infrastructure: State of
practice and research issues, Structure and Infrastructure Engineering 1, 7–18.
Federal Highway Administration (FHWA), 2010. National Bridge Inventory (NBI) database.
U.S. Department of Transportation, Washington, D.C., http://www.fhwa.dot.gov/bridge/nbi
.htm.
Frangopol, D. M., 2011. Life-cycle performance, management, and optimization of structural
systems under uncertainty: Accomplishments and challenges, Structure and Infrastructure
Engineering 7, 389–413.
Frangopol, D. M., and Bocchini, P., 2012. Bridge network performance, maintenance and opti-
mization under uncertainty: Accomplishments and challenges, Structure and Infrastructure
Frangopol, D. M., and Liu, M., 2007. Maintenance and management of civil infrastructure based
on condition, safety, optimization, and life-cycle cost, Structure and Infrastructure Engineer-
ing 3, 29–41.
Ghosh, J., and Padgett, J. E., 2010. Aging considerations in the development of time-dependent
seismic fragility curves, Journal of Structural Engineering 136, 1497–1511.
Gibbons, A. M., 1985. Algorithmic Graph Theory, Cambridge University Press, Cambridge, UK,
514 pp.
Golroo, A., Mohaymany, A. S., and Mesbah, M., 2010. Reliability based investment prioritiza-
tion in transportation networks, Proceedings of the 89th annual meeting of the Transportation
Research Board of the National Academies, 10–14 January 2010, Washington, D.C.,
CD-ROM.
Jones, T., Middelmann, M., and Corby, N., 2005. Natural Hazard Risk in Perth, Western
Australia, GeoCat No. 63527, Geoscience Australia, Canberra.
Leemis, L. M., 1995. Reliability, Probabilistic Models and Statistical Methods, Prentice Hall,
Englewood Cliffs, NJ, 288 pp.
Liu, M., and Frangopol, D. M., 2005. Time-dependent bridge network reliability: Novel
approach, Journal of Structural Engineering 131, 329–337.
Liu, M., and Frangopol, D. M., 2006. Probability-based bridge network performance evaluation,
Journal of Bridge Engineering 11, 633–641.
Mander, J. B., 1999. Fragility Curve Development for Assessing the Seismic Vulnerability of
Highway Bridges, Research Progress and Accomplishments: 1997–1999, Multidisciplinary
Center for Earthquake Engineering Research (MCEER), Buffalo, NY.
Martin, W. A., and McGuskin, N. A., 1998. Travel Estimation Techniques for Urban Planning,
National Cooperative Research Program (NCHRP) Report 365, Transportation Research
Board, National Research Council, Washington, D.C.
MathWorks, 2009. Statistics Toolbox™ 7 User’s Guide. The MathWorks, Inc., Natick, MA,
1,749 pp.
Maunsell Ltd., 1999. Serviceable Life of Highway Structures and Their Components, Final
Report, Birmingham (UK), Highways Agency, London, UK.
McKay, M. D., Conover, W. J., and Beckman, R. J., 1979. A comparison of three methods for
selecting values of input variables in the analysis of output from a computer code, Techno-
metrics 21, 239–245.
Nielson, B. G., and DesRoches, R., 2007. Analytical seismic fragility curves for typical bridges in
the central and southeastern United States, Earthquake Spectra 23, 615–633.
Padgett, J. E., and DesRoches, R., 2007. Bridge functionality relationships for improved seismic
risk assessment of transportation networks, Earthquake Spectra 23, 115–130.
Padgett, J. E., DesRoches, R., and Nilsson, E., 2010. Regional seismic risk assessment of bridge
network in Charleston, South Carolina, Journal of Earthquake Engineering 14, 918–933.
Sánchez-Silva, M., and Rackwitz, R., 2004. Socioeconomic implications of life quality index in
design of optimum structures to withstand earthquakes, Journal of Structural Engineering
130 , 969–977.
Spicer, R. S., and Miller, T. R., 2010. Final Report to the National Highway Traffic Safety
Administration: Uncertainty Analysis of Quality Adjusted Life Years Lost, Pacific Institute
for Research and Evaluation, Calverton, MD.
Stein, S. M., Young, G. K., Trent, R. E., and Pearson, D. R., 1999. Prioritizing scour vulnerable
bridges using risk, Journal of Infrastructure Systems 5, 95–101.
Transportation Research Board (TRB), 2000. Highway Capacity Manual, Special Report 209,
National Research Council, Washington, D.C.
U.S. Department of Transportation (USDOT), 2011. Treatment of the Value of Preventing Fatal-
ities and Injuries in Preparing Economic Analyses—2011 Revision, Washington, D.C.
U.S. Geological Survey (USGS), 2011. Earthquake Search Tool. Department of the Interior,
Reston, VA, http://earthquake.usgs.gov/earthquakes/eqarchives/epic/.
van Noortwijk, J. M., and Klatter, H. E., 2004. The use of lifetime distributions in bridge
maintenance and replacement modelling, Computers and Structures 82, 1091–1099.
Wells, D. L., and Coppersmith, K. J., 1994. New empirical relationships among magnitude,
rupture length, rupture width, rupture area, and surface displacement, Bulletin of the Seismo-
logical Society of America 84, 974–1002.
Yang, S-I, Frangopol, D. M., and Neves, L. C., 2004. Service life prediction of structural systems
using lifetime functions with emphasis on bridges, Reliability Engineering & System Safety 86,
39–51.
Zhu, B., and Frangopol, D. M., 2013. Risk-based approach for optimum maintenance of
structures under traffic and earthquake loads, Journal of Structural Engineering 139, DOI:
10.1061/(ASCE)ST.1943-541X.0000671.
(Received 10 June 2011; accepted 15 March 2012)
Virtual Earthquake Engineering Laboratory
Capturing Nonlinear Shear, Localized
Damage and Progressive Buckling of Bar
In Ho Choa) M.EERI
We embarked upon developing a novel parallel simulation platform that is

rooted in microphysical mechanisms. Primarily aiming at large-scale
reinforced-concrete structures exposed to cyclic loading, we sought to settle the
question as to how to capture nonlinear shear, localized damage and progressive
buckling of reinforcing bar. We proposed a tribology-inspired three-dimensional
(3-D) interlocking mechanism in the well-established framework of multidirec-
tional smeared crack models. Strong correlation between random material property
and localized damage has been shown, notably at the global system level. An
automated platform has been suggested to capture progressive buckling
phenomena. Validation and applications straddle a wide range, from small
laboratory tests to large-scale 3-D experiments, successfully offering a clear causal
pathway between underlying physical mechanisms and the unresolved issues
addressed above. [DOI: 10.1193/1.4000095]
INTRODUCTION
In the earthquake engineering community, numerical simulation has played an essential
role in increasing resilience of infrastructures, providing the optimal rehabilitation remedy,
and even generating precise probabilistic fragility curves in the loss estimation process. Still,
we have seen a substantial gap between our simulation capability and actual damage and
failure modes of real structures during recent big earthquakes. Filling the gap by
harmoniously combining physics-based mechanisms and cutting-edge parallel computers
was the original motivation of our novel attempt.
The challenges that we are trying to tackle herein include the nonlinear shear, localized
damage and progressive buckling of reinforcing bar. Indeed, those issues have long been
completely unresolved, and are closely tied to the limitations of state-of-the-art simulation
technology of earthquake engineering fields.
For describing nonlinear behavior of shear-critical reinforced-concrete (RC) structural
members, a large number of numerical and theoretical models have been proposed,
which proved successful in providing important insight into global nonlinear responses.
Some works are based on well-designed panel tests (Mansour and Hsu 2005, Hsu and
Mansour 2005, Vecchio 1999), while the popular fiber section models are often employed
by others (Orakcal and Wallace 2006, Yazgan and Dazio 2011). Even a simple multilinear
a)
Depart. of Mechanical and Civil Engineering, California Institute of Technology, Pasadena, CA 91125
103
104 IN HO CHO
moment-curvature relationship has been effectively utilized for particular cases where
flexural behavior is dominant (Adebar and Ibrahim 2002).
Although all of these attempts proved powerful in understanding salient nonlinear beha-
vior of shear-critical RC structural members, most of them fundamentally rely on particular
types of experimental conditions such as well-controlled force and displacement boundary
conditions or geometric simplicity. Also, the limitation of the popular fiber section approach
is well recognized – the key restrictive assumption that plane section remains plane after
deformation. More importantly, localized damages, which exert strong influence on non-
linear degradation of RC systems, are often dealt with by some averaged quantities, normally
reflected from global force-displacement response. But, in reality, the localized failure modes
appear to emerge in various ways depending on mechanical characteristics of RC members
under earthquake loadings (e.g., Hanson 1996).
The other challenge can be found in steel bar models. To date, a multitude of attempts
have been made to describe nonlinear behavior of reinforcing bars under cyclic loading. For
instance, Kunnath et al. (2009) proposed a generic model by incorporating the initiation of
compressive buckling, low-cycle fatigue fracture, and cyclic strength degradation. Focused
on postyield buckling, Dhakal and Maekawa (2002a) suggested a simple model requiring
only a few parameters, and they derived an analytical method for estimating buckling length
(Dhakal and Maekawa 2002b). Rodriguez et al. (1999) paid considerable attention to the
postbuckling stress-strain behavior of reinforcing bars. Monti and Nuti (1992) combined
plasticity theory and an empirical buckling model so as to cover various hardening models.
Pantazopoulou (1998) pointed out the significant role of the properties of the surrounding
system. Bae et al. (2005) presented intriguing factors such as initial geometric imperfection
and the ratio of the ultimate strength to the yield strength.
Despite such valuable accomplishments, the aforementioned formulations regarding
compressive buckling tend to rely on common assumptions – such as idealized deformation
shape of a short bar, firmly fixed ties, and two ends of bar allowing no horizontal displace-
ment. In reality, however, the initial alignment of a reinforcing system deforms in a very
complicated way due to the rigorous interaction with the surrounding system of cover
and core concrete, horizontal steels such as ties and stirrups, and other longitudinal steels.
Hence the progressive buckling phenomenon essentially calls for a comprehensive and hol-
istic approach to crushing/spalling of brittle material, yielding and lengthened buckling
length of reinforcing bars, and so on.
One might suggest that we can simply perform highly detailed three-dimensional analysis
because powerful parallel computing resources (e.g., OpenSeesSP and OpenSeesMP) have
become available in our field. Indeed, with such advanced parallel computing technologies at
hand, popular continuum-based plasticity models – for example, quasi-brittle model by Simo
and Ju (1987a, b), ductile material under cyclic loading by Steglich et al. (2005), localized
shear band model by Ortiz (1987) and Jirásek (1998) – could be successful candidates to
resolve the challenges of nonlinear shear and localized damage.
However, it should be stressed that those continuum-based models are often handicapped
by the sensitivity to tangent stiffness (thus causing convergence problems), the lack of
VIRTUAL EARTHQUAKE ENGINEERING LABORATORY 105
physical counterparts of salient parameters, and the difficulty in handling evolution of

yielding surface under cyclic loading.
Thus we embarked upon developing a novel parallel simulation platform. Our first set of
goals include: (1) delineating nonlinear shear behavior by means of a reasonable microphy-
sical mechanism, (2) capturing progressive buckling of reinforcing bar, and (3) detecting
localized damage at the global system level.
Toward this end, we shall first revisit the “fixed-type” multidirectional smeared crack
model, which had been adopted to fully retain the physical nature of cracks. Next we
will discuss the tribology-inspired 3-D interlocking model, which is responsible for
generating accurate nonlinear shear resistance and resolving pathological nature of the
“fixed-type” crack model. Then, we shall touch upon a novel automated platform for
modeling progressive buckling of reinforcing bars by comprehensive consideration of the
reinforced system as a whole.
Validations and applications straddle a wide range from small laboratory tests to large-
scale RC structures exposed to severe cyclic loading. Notably, clear detection of localized
damage shall be presented, which appears to be related to the randomness over the entire
domain. It is noteworthy that behind all of the novelty lies the cutting-edge parallel comput-
ing technology (see Cho and Hall 2012).
REVISITING THE “FIXED-TYPE” MULTIDIRECTIONAL

SMEARED CRACK MODEL
The fixed-type smeared crack model is regarded to best retain the physical attribute of
crack, since the primary crack surface remains unchanged by subsequent loadings. The
downside of the fixed crack model, however, is also well known – spurious large stress
development under nonproportional loading. This deleterious attribute, therefore, brought
many supplementary alternatives after Rashid (1968) – for example, de Borst and Nauta
(1985), Weihe et al. (1998), and even rotating crack models (Cope et al. 1980, Rots 1988
and Jirásek and Zimmermann 1998). Since we placed top priority on preserving physical
counterparts, we adopted the fixed-type smeared crack model rather than the rotating
crack model.
The present smeared crack model allows three orthogonal crack surfaces and utilizes total
strain. For comparable approaches, refer to such various works as Vecchio and Collins
(1986), and Selby and Vecchio (1993, 1997); the clear contrast to their works lies in the
scale of the present crack model, which is assumed to be far smaller than that used in
those practical smeared crack models. For detailed flow of the fixed-type smeared crack
model, see the references above or the author’s thesis (Cho 2012).
For updating microscopic stresses on the three crack surfaces, we integrated the
advantageous attributes of the previous accomplishments of tension/compression models
(see Figure 1) – that is, we adopted the Thorenfeldt et al. concrete model (1987), which
is augmented by the linear re/unloading model of Taucer et al. (1991) and Moelands and
Reinhardt’s model (Reinhardt 1984) with c ¼ 0.31 for nonlinear tensile softening regime.
106 IN HO CHO
σ [MPa]
20
εp σ
0
ft
-20
εr 0.6
c = 1.0
0.31
-40
t u
-60 ε
-0.006 -0.003 0
Figure 1. Microscopic stress functions (εr and εp = strains defining un-/reloading path; f t and
εt = tensile strength and associated strain; εu = ultimate strain with tensile resistance; c =
softening parameter; origin of each model is provided in the text).
It is important to note that the mesh objectivity regarding tensile crack can be realized by
adopting the notion of “crack band width” h (e.g. Bažant and Oh 1983) such that
εðu
EQ-TARGET;temp:intralink-;e1;41;433 g dε ¼ GIf ∕h (1)

εt
where g is the microscopic stress function; GIf is Mode-I fracture energy in [J=m2 ]; and h is
crack band width in meters.
Similar to the trend of tribology (the science and engineering of interaction between
relatively moving surfaces), the pioneering work of 2-D interlocking models by Walraven
(1994) can be characterized not only by the stress transfer over opened crack directly depend-
ing on contacting areas between 2-D semicircle and its indentation, but also by a probabilistic
method used for realizing rough crack surface. Inspired by tribology as well as the 2-D
interlocking model, we have derived a 3-D interlocking model for describing the degradation
of tangent shear resistance at the cracked material. Figure 2 shows the base couple of rigid
particle–soft matrix, while the inset compares a typical asperity model consisting of rigid
plate and soft hemisphere widely used in tribology (after Greenwood and Williamson
1996). The single asperity model in tribology assumes that the plastic deformation happens
on the hemisphere part due to severe contact with rigid flat plate (e.g., Jackson and Green
2005). Conversely, in the present work, it is assumed that the permanent plastic deformation
Figure 2. The base couple of rigid particle–soft matrix for 3-D interlocking model being
proposed herein; (inset) typical asperity model in tribology, presented for comparison.
Permanent damage
Cyclic loading
on soft matrix
Figure 3. Permanent plastic damage on the soft matrix by cyclic movement.
Figure 4. Micrographs revealing micro-cracking along the aggregate boundaries of a concrete

specimen that is caused by severe compression (actual cross section of concrete specimen
showing crack patterns emphasized by arrows; from Elkadi (2005) with permission from
Delft University Press).
occurs at the soft matrix part only, and the ideal sphere remains intact during the whole cyclic
loading process (see Figure 3).
The rigid particle–soft matrix assumption is strongly tied to the physical nature of micro-
cracking. As shown in Figure 4, for instance, of concrete under excessive compression, the
micro-cracking tends to develop in the weak matrix first, and to grow along the boundaries of
relatively stiff aggregates. Such irregular, zigzagged configurations of opened crack surfaces
can be realistically described by a large number of the fabrics of rigid sphere-soft indentation.
DERIVATION OF NONLINEAR SHEAR RESISTANCE
Using the physics-rooted mechanism addressed above, we have obtained a calcuation for
nonlinear shear resistance as follows. Let us assume that one ideal particle of diameter Dmax
causes shear resistance. If it is displaced by d relative gap distance between crack surfaces,
we can estimate the horizontal and vertical projections of contact area of the moving sphere in
terms of d (see shaded area in Figure 5).
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
With c ¼ ðDmax ∕2Þ2 − d2 as depicted in Figure 5, we obtain the projected areas of
hemisphere as:
Ah ¼ πc2 ∕2 ;
EQ-TARGET;temp:intralink-;e2;62;162 Av ¼ ðDmax ∕2Þ2 arctan ðc∕dÞ − c × d (2)
By virtue of the persistently retained direction of cracks, notably with fixed-type cracks,
the tangent shear stiffness can be thought to be directly affected by contacting areas:
G ∝ ðAv þ μAh Þ
108 IN HO CHO
Figure 5. Projections of contact area of the ideal hemisphere with diameter Dmax .
This proportionality of shear resistance to actively contacting area is rooted in the central
idea of previous research (Walraven 1994), in which total frictional force in the horizontal
direction is assumed to be solely determined by the contacting area of the 2-D semicircle and
its indentation.
For a concise form, after introducing ε~ ¼ 2d∕Dmax , normalization with ðDmax ∕2Þ− 2 yields
# pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi π $
G ∝ arctan ε~ − 2 − 1 − ε~ 1 − ε~ 2 þ μð1 − ε~ 2 Þ (4)
2
It is natural to expect considerable difference in strength between the rigid particle and the
soft matrix, which can be envisaged as the strength contrast between aggregates and cement
paste for an instance of concrete. Hence appropriate reduction is to be taken into account by
use of a reduction factor, denoted Ccs herein. Although C cs has a physical meaning of
strength difference between key constituents, determining a proper value is not tractable.
Plasticity theory might be used to obtain a highly accurate value for Ccs , as is done in tri-
bology for steel material (Jackson and Green 2005). However, considering strong heteroge-
neity in the materials under consideration, the empirical approach appears more attractive,
and we attained Ccs ¼ 1:66e-4 from the validation against the interlocking test (Figures 9–
10). This value was used in all numerical simulations presented here. Finally, the tangent
shear stiffness on crack surface is proposed as
# pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi π $
G0 2 −
Gð~εÞ ¼ Ccs arctan ε~ − 1 − ε~ 1 − ε~ þ μð1 − ε~ Þ
2 2 2 (5)
ð1 þ μÞ π 2
where G0 = elastic shear modulus; μ = friction coefficient (0.4 is used throughout the
simulations); ε~ ¼ 2d∕Dmax . The choice of the fixed frictional coefficient μ has been
made following the well-proven work by Walraven (1994).
We can see the relationship between ε~ and ε in a one-dimensional case given by
d d Dmax ∕2 D ∕2
ε¼ ¼ ¼ ε~ max (6)
L Dmax ∕2 L L
where L = length of element and ε = normal strain to the crack surfaces.

As seen in Figure 6, the tangent shear stiffness decreases with ε~ (or equivalently with d)
such that Gjε~ ¼0 or d¼0 ¼ Ccs G0 and Gjε~ ¼1 or d¼0.5Dmax ¼ 0. To expand on the necessity of the
1 1
0.8 0.1 0.8 0.1
0.4 0.4
G/CcsG0
G/CcsG0
0.6 0.9 0.6 0.9
0.4 0.4
0.2 0.2
0 0
0 0.5 1 0 0.5 1
2d/Dmax 2d/Dmax
(a) 3D Hemisphere (b) 2D Semicircle
Figure 6. Degradation of tangent shear stiffness with several friction coefficients: (a) when a 3-D
hemisphere is used; (b) when a 2-D semicircle is adopted for interlocking mechanism, exhibiting
relatively stiff nature compared to the case with 3-D hemisphere.
3-D interlocking model in resolving the overstiffness problem of the fixed-type smeared
crack model, the variation of shear stiffness obtained from the 3-D and 2-D interlocking
model is compared in Figure 6. It is apparent that the 2-D semicircle-based model would
reproduce relatively larger shear stiffness than the 3-D hemisphere, which would render
it less favorable for alleviating the problem of overstiffness.
As shown in Figure 7a, when the crack is completely closed, the maximum tangent shear
stiffness is regained from the interlocking mechanism. It is noteworthy that once tensile crack
takes place, we always employ the interlocking mechanism for evaluating nonlinear shear
stiffness, which leads to the “weak” upper limit of tangent shear stiffness in the presence of
Mode-I crack – cf. the “strong” upper limit proposed by Regan (1971) in which full initial
shear stiffness is ideally regained after complete closure of tensile crack. In Figure 7c, we can
see that if the crack fully opens more than half of the ideal particle size, there remains no
resistance to shear.
In the intermediate stage when the crack begins to reopen, as depicted in Figure 7b, the
shear resistance between crack surfaces is governed by the interlocking mechanism. No resis-
tance is assumed during the unloading process, which is consistent with experimental results.
During the reloading process, somewhat early resistance is gained as denoted by recontacting
Slope:
Slope:
Slope
(a) (b) (c)
Figure 7. Variation of tangent shear stiffness depending on crack opening.

110 IN HO CHO
(a) Loading (b) Unloading (c) Reloading (d) Loading
Figure 8. Interlocking states during cyclic loading (all shear strain and shear stress terms signify
those on the crack surfaces).
shear strain γ r in Figure 7b, which appears to be tied to the remaining debris of matrix left by
shearing off and grinding.
However, this early resistance during reloading gradually disappears as the horizontal
displacement increases, and thus an expression for γ r has been suggested such that
γ r → γ max as γ >> 0
# $
γ r ¼ jγ max j × a þ bð1 − exp ð− c × jγ maxjÞÞ
EQ-TARGET;temp:intralink-;e7;41;397 for γ > 0 (7)
where a, b, and c are positive constant with a þ b¼ 1 and γ maxðγ minÞ is the maximum
(minimum) shear strain ever experienced on the crack surfaces. Some recommendable values
(i.e., a ¼ 0.8; b¼ 0.2; c ¼ 1.0, which have been empirically calibrated from the interlocking
experimented presented herein (Figures 9–10)) have shown favorable performance through-
out all simulations presented herein. It is of practical importance to note that γ maxðγ minÞ is not
a fixed value but rather a constantly updated value as time proceeds during nonlinear ana-
lysis. Also, the exponential decay of γ r is rooted in the knowledge from the well-documented
0.1
0.05
-0.05
-0.1
-0.1 -0.05 0 0.05 0.1
(a) (b) (c)
Figure 9. Interlocking experiment: (a) actual experimental specimen; (b) numerical modeling
with initial Mode-I crack, marked by arrows in the middle layer; (c) unstructured random particle
distribution over the middle layer of 2,500 elements. (Note: for visualization purpose, 25 and
2,500 elements are shown in (b) and (c), respectively.)
1.00
0.50
σnt[MPa]
0.00
-0.50
slip [mm]
-1.00
-2 0 2
(a) Experiment (from Briseghella and Gori 1984) (b) Predicted slip-shear stress
Figure 10. Horizontal slip-resultant force responses (force is represented by corresponding shear
stress). The numbers in (a) signify the cycle number recorded during the experiment.
experiments (Eligehausen et al. 1983) in which damage on weak concrete material caused by
cyclic loadings is expressed in exponential form.
RANDOM PARTICLE DISTRIBUTION

When actual RC structures are exposed to seismic loading rather than a discrete crack,
multiple crack surfaces are more likely to emerge at random locations, coalesce, and govern
the nonlinear shear behavior (Hanson 1996). All of these physical facts led us to the attractive
notion of tribology. To describe asperity contact, Jackson and Green (2005) generated
random asperity heights from the uncompromised Gaussian distribution for the entire contact
surface, and performed integration to draw the resultant global force-slip relationship
(cf. similar work by Polycapou and Etsion 1999, Liu et al. 2000). In light of its accuracy,
the present platform has adopted the Gaussian distribution as the base probabilistic distribu-
tion for generating particle sizes for the 3-D interlocking model.
To clarify the distribution process of the ideal particle, let Dh be a homogeneous random
field of ideal particle size for the interlocking mechanism in the physical domain Ù. In the
present study, Ù represents the entire domain of structure on which nonlinear degradation is
defined. Among various methods (e.g., Li and der Kiureghian 1993), the midpoint method
has been adopted to discretize the random field Dh into a vector of random variable
D ¼ fD1 ; D2 ; ...g associated with a probability density function f D . Following the notion
of the midpoint method as done by many (e.g., Carmeliet and de Borst 1995, Gutiérrez
and de Borst 1999), a random variable Di defined on a subdomain Ùi∈ Ù can be represented
by
Di ¼ Dh ðxci Þ
where xci is a vector representing the center of the subdomain Ùi.

Furthermore, we assumed Ùi to be equal to a finite element of the structure under con-
sideration. In this context, each subdomain – that is, a finite element – holds one ideal particle
size that is generated from the Gaussian distribution.
112 IN HO CHO
INTERLOCKING DURING CYCLIC MOVEMENT

Physical interpretation of the interaction between the rigid particle and soft matrix provides
clear insight into realistic behavior of cyclic shear resistance. Figure 8a shows linearly increas-
ing shear stress with the slope induced from the current crack opening, whereas Figure 8b
shows no active contacting for 0 < γ < γ max . Figure 8c describes early development of resis-
tance during reloading. When the current strain lies between the recontacting shear strain γ r and
the previous maximum strain γ max , we reduced shear resistance by α1 Gð~εÞ, where α1 ∈ ð0; 1Þ.
As reloading further proceeds beyond γ max it is assumed that full contacting has been achieved,
and afterward shear stress is regarded to fully obey the interlocking mechanism as depicted in
Figure 8d. Thus at each incremental step, a new tangent shear stiffness Gcr ij should be updated
from the 3-D interlocking model with current crack strains as
#
cr G0 for εcr cr
i ; εj ≤ 0
Gij ≡ (9)
α1 Gð~εÞ otherwise
where εcr
i = normal crack strain in the ith normal direction on the crack surfaces;
ε~ ¼ 2L∕Dmax × max ðεcr cr
i ; εj Þ; α ∈ ð0; 1Þ for jγ r j ≤ jγj ≤ jγ max j.
VALIDATION AGAINST AN INTERLOCKING EXPERIMENT

In order to validate the 3-D interlocking model, a well-documented experiment of rough
crack by Briseghella and Gori (1984) was selected. In the experimental program, a plain
concrete block of dimension 0.2 × 0.2 × 0.2 m is initially separated into two parts, leaving
a discrete rough Mode-I crack with width 1 mm. Afterwards, the lower block is horizontally
loaded in a reverse cyclic manner by displacement control.
For the numerical simulation, the cracked blocks in Figure 9a were modeled by three
layers of finite elements in the vertical direction (each layer consists of 2,500 hexahedral
elements). As shall be addressed in the following section, such a large number of elements
is not necessarily required. Indeed, if the shearing area is fixed, a small number of elements
appears to be enough to obtain the almost identical resultant shear resistance. As shown in
Figure 9b, the middle layer was intentionally designed to hold initial Mode-I cracking as well
as ensuing interlocking behavior, and the arrows stand for the directions normal to the
cracked surface. Initially, ideal particles were generated from the Gaussian distribution
and assigned to all elements on the middle layer as shown in Figure 9c.
Figure 10 compares the horizontal slip and resultant force responses from the experiment
and prediction. For clarity, the unloading and reloading paths predicted with zero slopes are
not shown in Figure 10b, and only ascending and descending paths are shown. As easily
seen, the overall nonlinear response of shear force transfer over the rough crack surfaces
has been successfully reproduced.
Although not significant, however, some discrepancy is noticeable at large slip range, which
appears to be related to the fact that in the simulation, Mode-I crack with an initial opening (1 mm)
was modeled to be smeared within the middle layer of one element height. Unlike the ideal experi-
mental setup with the fixed crack spacing, the cracked solid elements in the middle layer might
deform in a way that the Mode-I crack becomes slightly larger than the initial value or than reality.
Whereby additional loss of contacting area might happen, thus leading to reduction of peak shear
resistance.
1.10
1.05
F/F121
1.00
0.95 Uniform Distribution

Normal Distribution
0.90
0 1000 2000
Number of elements
Figure 11. Normalized maximum horizontal forces with varying element numbers for two ran-
dom particle distributions: normal and uniform distributions.
EFFECT OF NUMBER OF ELEMENTS AND MEAN

DIAMETER OF PARTICLES
With increasing elements through mesh refinement, the diversity of sizes and total number
of particles also increases. Figure 11 presents the parametric study on the element size,
performed with two representative probabilistic distributions having the same mean diameter
0.019 m: normal distribution Nð0.019; 0.006332 Þ and uniform distribution Uð0.0001; 0.038Þ.
The aforementioned rough crack experiment is numerically simulated with a varying number of
elements, 112~502. The vertical axis of Figure 11 stands for the predicted maximum horizontal
forces normalized by that with 112 elements. Regardless of the type of probabilistic distribution
employed, we can confirm the desired mesh objectivity, provided the total cross-sectional
shearing area is fixed.
The mean value used in the probabilistic distribution for random particle size, however,
significantly affects the resultant horizontal forces. Figure 12 gives the positive peak forces
collected from the numerical simulations of the experiment with initial crack width of
1 mm. As expected, the increased mean size of particle distribution pushes upward the
range of forces in a nonlinear manner. It should be noted that with a mean size larger than
a certain value – for example, 25 mm herein – the resulting horizontal forces appear to converge
to the specific upper bound, which can be physically conceived by the almost occupied inden-
tations.
4.E+04
Peak horiz. force [N]
Mean:
3.E+04 Incremented
2.E+04 5mm
1.E+04 2mm
0.E+00
0.4 0.9 1.4 1.9
Horizontal slip [mm]
Figure 12. Effect of mean particle size on resultant horizontal force from 3-D interlocking
model.
114 IN HO CHO
“SMART” REINFORCING BAR MODEL WITH

EVOLVING BUCKLING LENGTH
Departing from valuable knowledge of the previous studies summarized in the introduc-
tion, we aimed to propose a “smart” reinforcing steel model that holds general applicability.
In essence, the “smart” reinforcing steel model is a sort of platform, mostly dealing with
information about microstates of surrounding materials, topological change (i.e., boundary
condition) of the reinforced system, etc., giving rise to a new paradigm of analysis of com-
posite structures. In the near future, further enrichment and sophistication of the platform will
be open to researchers in our fields.
To systematically handle the evolution of buckling length of longitudinal steel bars, we
define the base unit BU i , which consists of the ith steel bar and a set of its surrounding ele-
ments Si ði ∈ ½1; n s &, n s the total number of steel bars). In light of topological information, the
construction of all units and inter-unit information transfer are automatically carried out in a
parallel algorithm (see Cho 2012 for a detailed description and parallelization).
Physically, it is hard to expect the onset of compressive buckling of a longitudinal bar
when some portion of the surrounding elements did not crush (or spall). As briefly illustrated
in Figure 13, a criterion for evolution of buckling length has been made in such a way that the
topological transition can be triggered only when all surrounding elements enter at least par-
tially crushed state. After the transition, we regard bars to lose their surrounding elements
and, in turn, they can coalesce with adjacent bars, constructing new base unit pivoting around
the lengthened bar.
Topological transition of BU m can be correlated to the dissipated energy of surrounding
elements in Sm as
ð
σ dV ≥ Eth
Sm
with a specific energy threshold E th . To evaluate the energy state of the material, two internal
state variables are introduced: one is defined at the integration point level and the other at the
buckling length
All surrounding elements of
crushed (or spalled) Steel(1) BU1
Steel(2) BU2
Steel(3) BU3
extended
Steel(4) BU3
base unit
Steel(5) BU3
Figure 13. Example of topological transition triggered only when all surrounding elements
entered a certain damage level, which in turn lengthened buckling lengths of bars.
ðkÞ
element level. First, let αi be an internal state variable of the kth crack surface, k ∈ f1; 2; 3g
defined on the ith integration point, describing uncrushed state by 1 and crushed state by 2
#
ðkÞ 1 for min εcr k > εth
αi ¼ t (11)
2 otherwise
with εth strain threshold for entering crush phase, which is assumed in this study as the strain
associated with the compressive strength.
Then we can further define an element-level internal variable λj , which can signify intact,
partially crushed, and fully crushed states of an element by use of simple integer values 0, 1,
and 2, respectively. In detail, when all the integration points in an element undergo crushed
phase we assign 2 to λj . When only some of the integration points in an element enter crushed
phase, we regard the element partially crushed denoted by 1, while intact state of an element
is denoted by 0 if none of the integration points experience crushed phase. With the aid of the
simple integer-based state variable we have:
8
>
<0 βj ≤ NINTðjÞ × 3
λj ¼ 1 NINTðjÞ × 3 < βj ≤ NINTðjÞ × 6 (12)
>
:2 βj > NINT × 6
ðjÞ
where NINTðjÞ = number of integration points of the jth element ∈ Sm and
XðjÞ
NINT X
3
ðkÞ
βj ¼
EQ-TARGET;temp:intralink-;e13;62;374 αi : (13)
i¼1 k¼1
While these three categories might not be sufficient for fully describing a wide spectrum of
damage states, they can effectively detect the most critical states of surrounding concrete mate-
rials. Since the internal variable λj is essentially tied to the states definition of advanced material
models, further sophistication of the internal variable shall be a natural future research topic.
For the conservative prediction, if all surrounding elements enter the partially crushed
phase, the topological transition is assumed to take place, which can be ascertained by
nX
ðSm Þ
EQ-TARGET;temp:intralink-;e14;62;234 λj ≥ nðSm Þ (14)
j¼1
where nðSm Þ means the number of entities (i.e., surrounding elements) in Sm . Hence, once the
above condition is violated, the topological transition is triggered and the buckling length of
the corresponding reinforcing bar can evolve by merging with adjacent bars’ buckling
lengths, giving new buckling length:
X ðiÞ
Lb ≡ kL0 (15) EQ-TARGET;temp:intralink-;e15;62;131
ðiÞ
L0 f or ∀e1i ∈ BU m
116 IN HO CHO
ðiÞ
where Lo is initial buckling length of the ith steel bar e1i , and k is effective length factor set as
0.5 for fixed boundary condition. And, in turn, this evolved buckling length feeds the latest
information to the process of new buckling initiation point ε' as
rffiffiffiffiffiffiffiffi
ε' f y Lb ε'
¼ 55 − 2.3 ; ≥ 7 (16)
εy 100 Ds εy
where ε' is the intermediate strain where buckling initiates, following the key notion of Dha-
kal and Maekawa (2002a); εy and f y are the strain and stress ½MPa& at yielding; Lb is the
buckling length updated; and Ds is the diameter of the bar cross section. The suggested
bar model explained so far is schematically illustrated in Figure 14.
Figure 15 compares the analytical prediction with the well-known experiment of single
bar (Monti and Nuti 1992). The analytical response exhibits overall success in capturing
compressive buckling phenomena over a wide range of L∕D ratio. It is noteworthy, however,
that the lack of some advanced aspects – such as flat plateau regime after initial yielding,
smooth transition to buckling, and so on – appears to exert influence on the noticeable dis-
crepancy (e.g., the case of L∕D ¼ 11 in Figure 15). Such sophisticated features are readily
available in literature (e.g., Kunnath et al. 2009), and their inclusion shall be a natural future
extension.
To further understand the evolving buckling length, a simple reinforced concrete system
is simulated by cyclic vertical loading as illustrated in Figure 16a. Relatively weak compres-
sive strength is assigned to one or two bottom layers (Figure 16b) in hopes of confirming the
: Onset of buckling
×: Onset of rupture
Figure 14. Generalized Menegotto-Pinto bar model capturing compressive buckling. The para-
meter bis for stiffness reduction after exceeding yielding point (εy , f y ). (εr , σ r ) is the point where
load reversal happens while (ε0 , σ 0 ) means the intersection point between lines of initial and
reduced stiffness. The ultimate strain εu signifies tensile rupture. Following Rodriguez et al.
(1999), ε'p is introduced for capturing early buckling at positive strain after large tensile loading.
The parameters c1 and c2 are recommended to be 0.2 and 0.02 by Dhakal and Maekawa (2002a).
σ
σ
Figure 15. Comparison between experimental diagram (top row, from Monti and Nuti 1992) and
analytical results from the proposed bar model (bottom row), confirming the overall success in
capturing compressive buckling phenomena over various L∕D ratios.
Cyclic loading One layer crushed Two layers crushed

15
0
σ [MPa]
-15
-30 Concrete
-45
Steel bar at center -0.02 -0.01 0.00
Crushed parts
(a) (b) (c)
Figure 16. Illustration of sample study on different buckling modes, for which weak material
property of concrete has been assigned to different parts of the structure: (a) cyclically loaded test
setup of three layers of concrete with vertical steel bar at center (initial L∕D of each bar element at
one layer is 5); (b) one or two layers is made to quickly enter into crushed phase; (c) example
result of concrete stress-strain history showing the excessively crushed state.
evolving buckling length in accordance with concrete crushing (Figure 16c), and thus clear
contrast in buckling modes.
Figure 17a shows clear contrast in bar buckling modes. When the two concrete layers
enter into the crushed phase, the ratio L∕D also automatically evolves to 10, twice the initial
value of 5. Thus, the loss of two concrete layers causes far earlier bar buckling than the case
with one layer being crushed. Another simulation result presented in Figure 17b assures the
successful prediction of the early initiation of buckling after transversal from excessive
tensile loading, as suggested by Rodriguez et al. (1999). The significant role of evolving
buckling length is obvious in the early emergence of buckling.
118 IN HO CHO
600 600
L/D=5 L/D=5
400 400
L/D=10 L/D=10
200 200
σ [MPa]
σ [MPa]
0 0
-200 -200
-400 -400
-600 -600
-0.20 -0.10 0.00 0.10 -0.20 -0.10 0.00
(a) (b)
Figure 17. Clear contrast in compressive buckling modes of bar induced by automatically evol-
ving buckling length: (a) when two layers of concrete are crushed, the ratio L∕D evolved to 10,
twice the initial value of 5, thus causing the compressive buckling of bar to happen earlier than the
case with one layer being crushed; (b) early buckling after reversal from excessive tensile loading
is largely affected by the evolved buckling length.
APPLICATIONS TO LARGE-SCALE STRUCTURES
H-SHAPED WALL SYSTEM UNDER CYCLIC LOADING

An experiment of real scale RC structure exhibiting sufficient complexity of both
geometry and nonlinear behavior had been selected: Model DP1, the three-dimensional
H-shaped wall system under cyclic loading conducted by Palermo and Vecchio (2002).
In both web and flange, the concrete strength is 21.7 MPa and the strain at the peak is
0.00204. All bars are D6 type with 7-mm diameter, and the yielding stress and the correspond-
ing strain are 605 MPa and 0.00318, respectively. All the reinforcing bars are explicitly mod-
eled by space truss elements with “perfectly bonded” assumption. The finite element model
consists of 7,784 solid elements and 11,212 nodes. Figure 18 shows the dimensional details of
DP1. It is noteworthy that this particular experimental program contains no horizontal
stirrups/hoops.
To successfully utilize the 3-D interlocking model for this particular experiment, ideal
particles were generated from the Gaussian distribution, for which the mean value was
19 mm (ranging from 0.1 mm to 38 mm). Then they were randomly distributed over the
entire domain as depicted in the inset of Figure 18.
According to Figure 19a, without the 3-D interlocking model, the simulation predicts
overstiff responses as expected, which indeed confirms the pathological nature of the
fixed-type smeared crack model at real scale structure level. Contrariwise, Figure 19b asserts
that the positive contribution of the 3-D interlocking mechanism to the smeared crack model
is remarkably significant. The hardening behavior beyond peak point (i.e., displacement
loading greater than ( 8 mm) is markedly alleviated by the interlocking mechanism,
showing successful agreement with experimental response.
Remarkably, material heterogeneity of domain introduced by the 3-D interlocking model
appears to be correlated with the emergence of localized damage. In the absence of the 3-D
interlocking model, damage induced by the smeared cracking appears to be so smoothly
in [mm] 4000
4415
Cyclic loading
640
2020
75 95
3045
2885
95
Figure 18. H-shaped wall system details and random particle distribution for modeling irregular
asperity configuration of crack surfaces (inset).
2000 2000
1000 1000
Force [kN]
Force [kN]
0 0
-1000 -1000
Experiment Experiment
Prediction Prediction
-2000 -2000
-12 -8 -4 0 4 8 12 -12 -8 -4 0 4 8 12
Displacement [mm] Displacement [mm]
(a) Without interlocking model (b) With interlocking model
Figure 19. Effect of 3-D interlocking model on the global force-displacement responses
(simulated up to 12 mm near the peak load-carrying capacity).
scattered over the wall system that it rarely leads to any noticeable formation of the localized
damage (Figure 20a). By introducing 3-D interlocking, however, the simulation allows the
development of noticeable localized damage, notably in the form of out-of-plane bulging
with vertical directivity on the web part as marked by the dashed line in Figure 20b. Indeed,
the localized damage with vertical directivity is one of the unique features of this particular
experimental program (probably due to relatively thin web thickness, coarse reinforcing bars
on the web, excessively wide flanges, etc.), while a typical shear wall system is normally
failed with diagonal directivity. The random particles over the “entire” domain are believed
to be essential to capture such a unique localized damage (cf. comparable discoveries in the
heterogeneous granular medium by Shahinpoor 1980 and Andrade et al. 2007).
120 IN HO CHO
(a) Without interlocking model (b) With interlocking model
Figure 20. Deformed shape (amplified) with/without 3-D interlocking model. (Note: By intro-
ducing the 3-D interlocking mechanism, the localized damage with vertical directivity marked in
dashed arrow can be captured.)
FOUR-STORY T-SHAPED WALL SYSTEM UNDER CYCLIC LOADING

The next application used for validation is more enriched by asymmetric geometry and
sophistication of reinforcing steels – the four-story T-shaped wall system (Thomsen IV and
Wallace 1995, 2004). The wall system was loaded by cyclic displacement up to its ultimate
damage states—namely, fully developed diagonal cracks on web, and crushing and spalling
of concrete accompanied by compressive buckling of longitudinal reinforcing bars.
The experimental program, denoted by TW2 hereafter, is approximately one-quarter
scale of part of the original prototype building, similar to a typical multistory office structure
in a high seismicity zone such as Los Angeles. Along with random particle distribution over
the entire domain, the detailed geometric information is given in Figure 21. TW2 is 3.66 m
tall and 102 mm thick, and the length of both web and flange is 1.22 m. Stiff floor is placed at
914-mm intervals.
The strength of concrete and corresponding strain are 41.7 MPa and 0.002, respectively.
Tensile strength is 5.63 MPa. The yield strength and corresponding strain are commonly
414 MPa and 0.00207 for longitudinal steels (Grade 60 No. 3 and No. 2) and hoops and
ties ðD ¼ 4:75 mmÞ, and more detailed properties of reinforcing steels can be found else-
where (Thomsen IV and Wallace 1995, 2004). In total, 67,410 linear hexahedral elements for
concrete and 20,646 space truss elements for all reinforcing bars are generated (i.e., 84,211
nodes). It should be stressed that no artificial separation between cover and core concrete,
which is normal in popular fiber-section models, has been included.
Figure 22 compares the resultant force-displacement plots from predictions with and
without 3-D interlocking model; the upper and lower limits of the recorded force range
from the experiment are given by the dashed lines for comparison. Overall, the 3-D inter-
locking model appears to successfully coordinate with the fixed-type smeared crack model in
the simulation, enforcing the range of load-carrying capacity to be accurately bounded
(Figure 22a). Contrariwise, without 3-D interlocking mode, the overstiff nature of the
smeared crack model remains clearly visible (Figure 22b). Such overestimated response
seems more apparent in the positive displacement range (i.e., flange under compression)
than in the negative range (i.e., flange under tension). In the case of the flange under com-
pression, with asymmetric placement of the wide flange, a large portion of the bottom web
undergoes tension and shear-governed deformation. This can be restated in more practical
terms such that the so-called neutral axis of the bottom section is located near the wide
Figure 21. Overall geometry of TW2 and (inset) random particle distribution over entire domain
for modeling irregular asperity configuration of crack surfaces.
400 400
> 28% error
300 300
189.78 kN
Lateral Load [kN]
Lateral Load [kN]
200 200
100 100
0 0
-100 -100
Prediction
-200 Prediction -200
-300 -362.94 kN -300 Experiment
Experiment
-400 Range -400 Range
-0.10 -0.05 0.00 0.05 0.10 -0.10 -0.05 0.00 0.05 0.10
Displacement [m] Displacement [m]
(a) With interlocking model (b) Without interlocking model
Figure 22. Effect of 3-D interlocking model on global force-displacement response (numbers
next to dashed line show the maximum lateral load range from the experiment): (a) with
the interlocking model, well-bounded lateral load response has been generated, whereas (b)
without interlocking model, error is escalating (exceeding even 28%) and unboundedly increas-
ing. Both cases commonly underestimate (in negative range) load-carrying capacity due mainly to
the exclusion of confinement effect consideration.
122 IN HO CHO
flange, causing web under tension and flange under compression (Thomsen IV and Wallace
2004). Therefore, the interlocking mechanism, which always manifests itself in the presence
of actively opened cracks, significantly influences the global resistance when the flange is
under compression (i.e., positive displacement regime in Figure 22). On the other hand, when
the flange is under tension (negative displacement range), both predictions with and without
3-D interlocking model appear to commonly underestimate the load-carrying capacity of the
system compared to the experimental record. This common underestimation is most likely
tied to lack of consideration of confinement effect in both simulations. In particular, despite
the full realization of reinforcing system, the compressive stress is still updated from the
microscopic stress function, and relatively large compressive strain can develop in the
web due to the wide width of the flange, together producing weak compressive stress
from the postpeak regime. Thus it would be a natural extension in the future research to
incorporate appropriate physical mechanisms capable of describing lateral confinement effect
into the present framework.
As shown in Figure 23, it is fascinating to capture the irrecoverable damage, notably in a
localized manner. Emphasized by dashed arrows in the figure, the damage is concentrated at
the web part with apparent diagonal directivity. Intrinsic randomness of particles in 3-D inter-
locking model is believed to have a strong influence on such a localized damage of the web.
Furthermore, compressive crushing at the web toe is also well predicted and shown in a
somewhat exaggerated manner.
It is apparent that longitudinal steel placed at the outermost position of web toe is experi-
encing progressive buckling according to the excessively large compressive strain as shown
Figure 23. Deformed shape (amplified) at the last step of the simulation, revealing concentrated
severe damage at web boundary part and the localized damage with diagonal directivity (marked
by dashed arrows) by virtue of the 3-D interlocking model. In the right photo, actual failure
state with severe damage at web and diagonal cracks is provided for comparison (photo from
Thomsen IV and Wallace 1995).
600 600
400 400
200 200
σ [MPa]
σ [MPa]
0 0
-200 -200
-400 -400
-600 -600
-0.10 0.00 -0.002 0.008
Hysteretic response of A Hysteretic response of B
B
A
Figure 24. Apparent contrast in the predicted hysteretic responses of longitudinal steel bars
depending on their locations; that is, we confirm buckling of bar located at the outermost posi-
tions of web toe (denoted A) with excessive compressive strain, whereas no buckling at the bar of
flange (denoted B). Inset at left top is the actual photo with spalling and crushing over a large
number of tie spacings, which bears out the progressive buckling of longitudinal bars (inset from
Thomsen IV and Wallace 2004).
in Figure 24. Contrarily, from the bars at the flange, typical stress-strain response with
smooth transition after yielding is identified, with relatively small strain range ð< 0.01Þ.
The predicted compressive buckling length is well consistent with the realistic failure
mode (inset of Figure 24) of the region near web toe experiencing severe crushing and spal-
ling, notably over a large number of tie spacings.
CONCLUSIONS
We presented a novel parallel simulation platform to settle the unresolved questions as to
nonlinear shear, localized damage and progressive buckling of reinforcing bars. Inspired by
tribology, we adopted the fabric of rigid particle–soft matrix, of which interaction generates
nonlinear shear stress on crack surfaces under cyclic loading. For multiple rough crack sur-
faces, we distribute random particles from the Gaussian distribution over the “entire” domain.
Validation of the 3-D interlocking model against an experiment reveals remarkable accuracy
and possibility of the model.
Departing from well-established previous researches, we presented a platform for a
“smart” steel bar model. With the aid of an automated and parallelized platform, compre-
hensive consideration of nonlinear degradation of the reinforced system of bar and surround-
ing materials offered a smart platform to capture the progressive buckling phenomena.
Simulation of real scale structures with the aid of the parallel platform is itself opening
new possibilities of “top-down” validation, offering a clear causal pathway between under-
lying physical mechanisms and the particular global response. From the large-scale applica-
tions, the well-known pathological nature of the fixed-type smeared crack model has been
successfully resolved by the inclusion of a 3-D interlocking mechanism. Notably, the
124 IN HO CHO
randomness of salient parameters over the entire domain appears to play an essential role to
cause localized damage.
ACKNOWLEDGMENTS
Regarding experimental data for validation, the warm hospitality of Professor Daniel
Palermo, Professor Frank J. Vecchio, and Professor John W. Wallace is appreciated. All
numerical simulations related to the present work were run on GARUDA, a high-performance
computing cluster hosted within the civil engineering department at Caltech. The purchase and
installation of GARUDA was in large part possible thanks to the Ruth Haskell Research Fund,
the Tomiyasu Discovery Fund, and Dell Inc. Professor S. Krishnan’s warm support with the
cluster is deeply appreciated. Special thanks are due to Professor J. F. Hall for his consistent
support and productive discussion.
REFERENCES
Adebar, P., and Ibrahim, A. M. M., 2002. Simple nonlinear flexural stiffness model for concrete
structural walls, Earthquake Spectra 18, 407–426.
Andrade, J. E., Baker, J. W., and Ellison, K. C., 2008. Random porosity fields and their
influence on the stability of granular media, Int. J. Numer. Methods Eng. in Geomechanics
32, 1147–1172.
Bae, S., Mieses, A. M., and Bayrak, O., 2005. Inelastic buckling of reinforcing bars, J. Struct.
Eng. 131, 314–321.
Bažant, Z. P., and Oh, B. H., 1983. Crack band theory for fracture of concrete, Matériaux
et Constructions 16, 155–177.
Briseghella, L., and Gori, R., 1984. Aggregate interlock cyclic response of R.C. critical section,
Proc., 8th World Conf. on Earthquake Engineering, San Francisco, Prentice-Hall Inc., Engle-
wood Cliffs, N.J.
Carmeliet, J., and de Borst, R., 1995. Stochastic approaches for damage evolution in standard and
non-standard continua, Int. J. Solids Structures 32, 1149–1160.
Cho, I., 2012. Virtual Earthquake Engineering Laboratory with Physics-Based Degrading Mate-
rials on Parallel Computers, Ph.D. Thesis, California Institute of Technology, Pasadena, CA.
Cho, I. H., and Hall, J. F., 2012. A parallelized implicit nonlinear FEA program for real scale RC
structures under cyclic loading, Journal of Computing in Civil Engineering 26, 356–365.
Cope, R. J., Rao, P. V., Clark, L. A., and Norris, P., 1980. Modeling of reinforced concrete
behavior for finite element analysis of bridge slabs, Numerical Methods for Nonlinear
Problems I (Taylor, ed.), Pineridge Press, Swansea, 457–470.
de Borst, R., 2001. Some recent issues in computational failure mechanics, Int. J. Numer.
Methods Eng. 52, 63–95.
de Borst, R., and Nauta, P., 1985. Non-orthogonal cracks in a smeared finite element model,
Engineering Computations 2, 35–46.
Dhakal, R., and Maekawa, K., 2002a. Modeling for postyielding buckling of reinforcement.,
J. Struct. Eng. 128, 1139–1147.
Dhakal, R., and Maekawa, K., 2002b. Reinforcement stability and fracture of cover concrete in
reinforced concrete members, J. Struct. Eng. 128, 1253–1262.
Elkadi, A. S. K., 2005. Fracture Scaling of Concrete under Multiaxial Compression. Ph.D. thesis,
Delft University of Technology, the Netherlands.
Eligehausen, R., Popov, E. P., and Bertero, V. V., 1983. Local Bond Stress-Slip
Relationships of Deformed Bars under Generalized Excitations, EERC Report, UCB/
EERC-83/23.
Greenwood, J. A., and Williamson, J. B. P., 1966. Contact of nominally flat surfaces, Proc.
R. Soc. A 295, 300–319.
Gutiérrez, M. A., and de Borst, R., 1999. Numerical analysis of localization using a visco-
plastic regularization: Influence of stochastic material defects, Int. J. Numer. Methods Eng.
44, 1823–1841.
Hanson, R. D., 1996. Evaluation of reinforced concrete members damaged by earthquakes,
Earthquake Spectra 12, 457–478.
Hicher, P. Y., 2008. Constitutive Modeling of Soils and Rocks (Hicher, Pierre-Yves, and Shao,
Jian-Fu, eds.), ISTE Ltd. and John Wiley & Sons, Inc., 143–185.
Hsu, T. T. C., and Mansour, M. Y., 2005. Stiffness, ductility, and energy dissipation of rc
elements under cyclic shear, Earthquake Spectra 21, 1093–1112.
Jackson, R. L., and Green, I., 2006. A statistical model of elasto-plastic asperity contact between
rough surfaces, Tribology International 39, 906–914.
Jirásek, M., 1998. Embedded crack models for concrete fracture. Computational Modelling of
Concrete Structures, Balkema, Rotterdam.
Jirásek, M., and Zimmermann, T., 1998. Rotating crack model with transition to scalar damage,
Journal of Engineering Mechanics 124, 277–284.
Kunnath, S. K., Heo, Y. A., and Mohle, J. F., 2009. Nonlinear uniaxial material model for rein-
forcing steel bars, J. Struct. Eng. 135, 335–343.
Li, C. C., and der Kiureghian, A., 1993. Optimal discretization of random fields, J. Eng. Mech.
119, 1136–1154.
Liu, Z., Neville, A., and Reuben, R. L., 2000. Analytical solution for elastic and elastic-plastic
contact models, Trib. Trans. 43, 627–634.
Mansour, M. Y., and Hsu, T. T. C., 2005. Behavior of reinforced concrete elements under cyclic
shear: Part II – Theoretical model, J. Struct. Eng. 131, 54–65.
Monti, G., and Nuti, C., 1992. Nonlinear cyclic behavior of reinforcing bars including buckling,
J. Struct. Eng. 118, 3268–3284.
Orakcal, K., and Wallace, J. W., 2006. Flexural modeling of reinforced concrete walls – experi-
mental verification. ACI Structural Journal 103, 196–206.
Ortiz, M., 1987. An analytic study of the localized failure modes of concrete, Mechanics of
Materials 6, 159–174.
Palermo, D., and Vecchio, F. J., 2002. Behavior of three-dimensional reinforced concrete shear
walls, ACI Structural Journal 99, 81–89.
Pantazopoulou, S. J., 1998. Detailing for reinforcement stability in RC members, J. Struct. Eng.
124, 623–632.
Peerlings, R. H. J., de Borst, R., Brekelmans, W. A. M., and de Vree, H. P. J., 1996.
Gradient-enhanced damage for quasi-brittle materials, Int. J. Numer. Methods Eng.
39, 3391–3403.
Polycarpou, A. A., and Etsion, I., 1999. Analytical approximations in modeling contacting rough
surfaces, ASME J. Tribol. 121, 234–239.
Rashid, Y. R., 1968. Ultimate strength analysis of prestressed concrete pressure vessels, Nuclear
Engineering and Design 7, 334–344.
126 IN HO CHO
Regan, P., 1971. Shear in Reinforced Concrete: An Experimental Study, Report to the Construc-
tion Industry Research and Information Association, Imperial College, Department of Civil
Engineering, London, England, 93–110.
Reinhardt, H. W., 1984. Fracture mechanics of an elastic softening material like concrete, Heron 29.
Rodriguez, M. E., Botero, J. C., and Villa, J., 1999. Cyclic stress-strain behavior of reinforcing
steel including effect of buckling, J. Struct. Eng. 125, 605–612.
Rots, J. G., 1988. Computational Modeling of Concrete Fracture, Ph.D. thesis, Delft University of
Technology, the Netherlands.
Selby, R. G., and Vecchio, F. J., 1993. Three-dimensional Constitutive Relations for Reinforced
Concrete, Tech. Rep. 93–02, University of Toronto, Department of Civil Engineering,
Toronto, Canada.
Selby, R. G., and Vecchio, F. J., 1997. A constitutive model for analysis of reinforced concrete
solidm Can. J. Civ. Eng. 24, 460–470.
Shahinpoor, M., 1980. Statistical mechanical considerations on the random packing of granular
materials, Powder Technology 25, 163–176.
Simo, J. C., and Ju, J. W., 1987a. Strain- and stress-based continuum damage models – I.
Formulation, Int. J. Solids Structures 23, 821–840.
Simo, J. C., and Ju, J. W., 1987b. Strain- and stress-based continuum damage models – II.
Computational aspects, Int. J. Solids Structures 23, 841–869.
Steglich, D., Pirondi, A., Bonora, N., and Brocks, W., 2005. Micromechanical modelling of
cyclic plasticity incorporating damage, Int. J. Solids Structures 42, 337–351.
Taucer, F., Spacone, E., and Filippou, F. C., 1991. A Fiber Beam-Column Element for Seismic
Response Analysis of Reinforced Concrete Structures, EERC Report, UCB/EERC-91/17.
Thomsen IV, J. H., and Wallace, J. W., 1995. Displacement-Based Design of Reinforced
Concrete Structural Walls: Experimental Studies of Walls with Rectangular and T-Shaped
Cross Sections, Rep. No. CU/CEE-95/06, Department of Civil and Environmental Engineer-
ing, Clarkson University, Potsdam, New York.
Thomsen IV, J. H., and Wallace, J. W., 2004. Displacement-based design of slender reinforced
concrete structural walls-experimental verification, J. Struct. Eng. 130, 618–630.
Thorenfeldt, E., Tomaszewicz, A., and Jensen, J. J., 1987. Mechanical properties of high-strength
concrete and applications in design, Proc. Symp. Utilization of High-Strength Concrete,
Stavanger, Norway.
Vecchio, F. J., 1999. Towards cyclic load modeling of reinforced concrete. ACI Structural Jour-
nal 96, 193–202.
Vecchio, F. J., and Collins, M. P., 1986. The modified compression field theory for reinforced
concrete elements subjected to shear, ACI Journal Proceedings 83, 219–231.
Weihe, S., Kröplin, B., and de Borst, R., 1998. Classification of smeared crack models based on
material and structural properties, Int. J. Solids Structures 35, 1289–1308.
Walraven, J. C., 1994. Rough cracks subjected to earthquake loading, J. Struct. Eng. 120,
1510–1524.
Yazgan, U., and Dazio, A., 2011. Simulating maximum and residual displacements of rc
structures: I. Accuracy, Earthquake Spectra 27, 1187–1202.
(Received 4 October 2011; accepted 19 February 2012)
Seismic Behavior and Shaking Direction
Influence on Adobe Wall Structures
Reinforced with Geogrid
Stefano Bossio,a) Marcial Blondet,b) M.EERI, and Satwant Rihal,c) M.EERI
The aim of this work was to assess and validate the construction technology
proposed in the manual published by the Pontificia Universidad Católica del Perú
(PUCP), Building Hygienic and Earthquake-Resistant Adobe Houses Using
Geomesh Reinforcement. In order to validate the proposed technology, two
full-scale adobe wall house models were built and subjected to dynamic tests
on the PUCP’s unidirectional shaking table. Geogrid was used as reinforcement
for the adobe walls. One of the models was aligned with the direction of shaking,
so that its longitudinal walls were subjected to in-plane forces and its transverse
walls to out-of-plane forces. The other model was turned 45° with respect to the
direction of shaking in order to have all the walls subjected to both in-plane and
out-of-plane forces. The results of both tests were also compared to those
obtained during previous tests of an unreinforced adobe model. [DOI:
10.1193/1.4000096]
INTRODUCTION
It is well known that vernacular earthen construction (adobe, tapial) is extremely vulner-
able to earthquakes. Every time an earthquake occurs in areas where this type of construction
is common, there is widespread destruction, significant economic loss, and regrettable inju-
ries and deaths. Researchers at the Catholic University of Peru (PUCP) have been working
since the 1970s to develop and implement simple and cost-effective reinforcement solutions
to reduce the vulnerability of adobe structures (Vargas et al. 2005). In recent years, several
testing programs at the PUCP’s Structures Laboratory have demonstrated that it is possible to
improve the seismic behavior of adobe structures using geogrid (or geomesh) reinforcement
(Blondet et al. 2005, 2006, 2008b, Torrealva and Acero 2005, Peralta 2009). During cyclic
static and seismic simulation tests, adobe structures reinforced with geogrid maintained their
integrity, providing enhanced resistance and deformation capacity. This technology has also
been used recently to repair and retrofit existing adobe buildings (Zapata and
Torrealva 2010).
The earthquake of 15 August 2007 caused extensive damage in the Peruvian regions of
Lima, Ica, and Huancavelica. Almost 600 people died, and more than 90,000 buildings were
destroyed (INDECI 2009). Few confined masonry or reinforced-concrete buildings suffered
significant damage. Most of the structures that collapsed or suffered extensive damage were
a)
University College London, Gower Street, London, WC1E 6BT
b)
Catholic University of Peru, Avenida Universitaria 1801, Lima 32, Peru
c)
California Polytechnic State University, San Luis Obispo, CA 93407
59
60 S. BOSSIO, M. BLONDET, AND S. RIHAL
made with adobe (Blondet et al. 2008c). In response to this devastating earthquake, PUCP
decided to carry out a training and reconstruction project in the affected areas using this new
adobe house construction technology (Blondet et al. 2008a). As part of the technology trans-
fer implementation plan PUCP distributed the manual Building Hygienic and Earthquake-
Resistant Adobe Houses Using Geomesh Reinforcement (Vargas et al. 2007). The manual is
still being used to build new adobe houses in the areas affected by the Pisco earthquake.
This paper describes the results of a shaking table test program developed to assess and
validate the construction system as published by PUCP and as put into practice after the 2007
Peru earthquake.
PROJECT DESCRIPTION
The aim of this project was to validate the reinforced-adobe construction system
described in the manual published and distributed by PUCP (Vargas et al. 2007) after
the 15 August 2007 earthquake in order to advance earthquake-resistant adobe house con-
struction in seismic areas. For this purpose, two full-scale adobe house models with geogrid
reinforcement were tested at the unidirectional shaking table of the PUCP’s Structures
Laboratory. The first reinforced model, called M1, was identical to an unreinforced
adobe house model (M0) tested at the shaking table during a previous project (Blondet
et al. 2006). Model M0 represented an unreinforced traditional adobe house, similar to
those that have collapsed or suffered significant damage during past earthquakes. According
to the Peruvian adobe code requirements (RNE 2006), although its adobe walls have
adequate strength, this model would have required special seismic reinforcement due to
the large wall slenderness. The geogrid-reinforced models M1 and M2 satisfy recently
approved design requirements in the Peruvian adobe code (Torrealva 2009).
Shake table excitation for both M0 and M1 was in the longitudinal direction of the
models (parallel to the east and west walls). Therefore, the east and west walls were subjected
to in-plane seismic loads and the transverse (north and south) walls were subjected to out-
of-plane loads. The second reinforced model (M2) was tested by placing it at 45° with respect
to the shaking table movement. In this case, all walls were subjected to both in-plane and
out-of-plane seismic loads.
The specific objective of the project was to assess the effectiveness of the geogrid rein-
forcement to provide the adobe walls the capacity to withstand seismic loads. This was done
by comparing the dynamic behavior and performance of the three models tested. Of particular
interest were the evaluation of the increase in strength and stiffness provided by the geogrid
to the adobe walls, and the assessment of the influence of the orientation of the unidirectional
ground shaking on the response of the reinforced walls.
MODEL DESCRIPTION AND CONSTRUCTION

Models M1 and M2 were reinforced with Tensar BX4100 geogrid (Tensar 2009). They
were built with the configuration shown in Figures 1 and 2. Similar adobe housing models
were tested in previous research projects (Torrealva and Acero 2005, Blondet et al. 2006).
Each model consisted of four walls measuring 3.25 m (10.66 ft.) long, 0.25 m (0.82 ft.) wide,
and variable height. Adobe bricks used were 250 mm (10 in.) × 250 mm (10 in.) × 70 mm
(2.8 in.) (L × W × H) in size. They were made from soil, straw, and coarse sand in the
SEISMIC BEHAVIOR AND SHAKING DIRECTION INFLUENCE ON ADOBE WALL STRUCTURES 61
Figure 1. Model configuration of ground plan.
Figure 2. Isometric view of the model.
proportional volume of 5:1:1. Vertical and horizontal joints were 20 mm (0.8 in.) wide and
were filled with mud mortar. The volume proportion of the mud mortar mixture was 3:1:1
(soil:straw:coarse sand). The minimum allowable compressive strength of this type of adobe
masonry is 200 kN/m2 and the minimum allowable shear strength is 25 kN/m2 according to
Peruvian national standards (RNE 2006).
East and west walls were identical and had a central window opening. The door opening
was in the south (front) wall, whereas in the north (back) wall there were no openings. The
roof had a slope of 8 percent in the north-south direction The models were built over a rein-
forced-concrete square ring beam 3.25 m (10.66 ft.) long each side and 0.30 m (1 ft.) by
0.30 m (1 ft.) cross section, which simulated a rigid house foundation. Furthermore, the
ring beam was used to attach the model to the shaking table and as a support during trans-
portation from the building area to the test site. Figure 3 provides elevation views of all
four walls.
Door and window lintels consisted of pieces of cane placed in three layers and tied up
every 0.40 m (1.31 ft.) with #16 wire (Figure 4). Placing these “flexible” lintels was intended
to avoid the large amount of cracks that usually appear through the wall-lintel corners during
shaking, absorbing the tensile stresses that concentrate in these areas. A wooden collar beam
was placed on top of the walls to provide some load transfer and displacement compatibility
at wall intersections. The roof system behaved therefore as a semi-rigid diaphragm.
The roof was placed upon the wooden collar beam. It consisted of 2 × 6-in. wooden
beams placed every 0.55 m (1.80 ft.) in the north-south direction. Strips of wood measuring
1½ × 2 in. were placed every 0.40 m (1.31 ft.) perpendicular to the north-south beams, as
shown in Figure 5.
Figure 3. Elevation of walls: (a) south wall; (b) east wall; (c) west wall; (d) north wall.
Figure 4. Cane lintel.
Figure 5. Roof framing.
The Tensar BX4100 geogrid (Tensar 2009) used in this project is available in Peru at a
cost of U.S. $1.15/m2. The total amount of geogrid per model was 95.14 m2 (1,024 ft.2) with a
cost of U.S. $109.40. The geogrid is conveniently sold in rolls of 3 or 4 m (9.8 or 13.1 ft.) in
width and 75 m (246 ft.) in length, as can be seen in Figure 6. Tensar offers other types of
geogrid, such as the BX1100 and BX1200, which have been used in previous projects
(Torrealva and Acero 2005; Blondet et al. 2006), with a cost of U.S. $1.45/m2 and
U.S. $3.45/m2, respectively. Although the tensile strength of the BX4100 is the lowest
among these geogrids, its use was based on economic criteria. Table 1 shows a comparison
of mechanical properties of the above-mentioned geogrids. Tensile strength tests were carried
out to verify the nominal tensile strengths specified by the supplier. All strength values
measured were higher than the ones provided by the supplier, which were considered reliable
(Bossio 2010).
During the setup of the adobe walls, plastic strips were placed between adobes (Figures 7
and 8) as a means of attaching the geogrid to the walls. These mooring points were composed
Figure 6. Geogrid roll.
of four plastic strips of 0.60 m (2 ft.) long each and placed 0.25 m (10 in.) apart horizontally and
every three adobe layers vertically. Eight pieces of geogrid were cut corresponding to each face
of the model. After each piece was placed upon the face of the wall, the plastic strips were
passed through the geogrid holes and tied, fastening the reinforcement to the model (Figure 9).
A piece of geogrid was wrapped around the foundation ring to simulate the anchorage of
geogrid inside the plinth as specified in the manual (Figure 10). The lower parts of the geo-
grids, which wrap the walls, were tied to the foundation geogrids with plastic thread strips.
On the upper part of the walls, the vertical geogrids of opposite faces were tied with plastic
strips and fastened to the wooden collar beams. All geogrid overlaps were 0.40 m and tied
with plastic strips (Figure 11).
Table 1. Mechanical properties of geogrids (Tensar 2009)
BX1100 BX1200 BX4100
Property Units MD XMD MD XMD MD XMD

Aperture mm 25 33 25 33 33 33
dimensions (in) (1.0) (1.3) (1.0) (1.3) (1.3) (1.3)
Minimum rib mm 0.76 0.76 1.27 1.27 0.76 0.76
thickness (in) (0.03) (0.03) (0.05) (0.05) (0.03) (0.03)
Tensile strength @ kN/m 4.10 6.60 6.00 9.00 4.00 5.50
2% strain (lb/ft) (280) (450) (410) (620) (270) (380)
Tensile strength @ kN/m 8.50 13.40 11.80 8.00 8.00 10.50
5% strain (lb/ft) (580) (920) (810) (550) (550) (720)
Ultimate tensile kN/m 12.40 19.00 19.20 28.80 12.80 13.50
strength (lb/ft) (850) (1,300) (1,310) (1,970) (880) (920)
MD: Main direction; XMD: Cross main direction

Figure 7. Placing plastic strips.
Figure 8. Setting adobe bricks.
Figure 9. Placing and tying the piece of geogrid.

Figure 10 Foundation geogrid.
Figure 11 Tying the geogrid overlaps.
Figure 12. Preparing the mortar.

Figure 13. Placing the first mud plaster layer.
The wall finish was placed in two phases. A first layer of mud plaster 10 mm (0.4 in.) thick
was placed using a mixture with a proportion of 3:2:1 (soil:straw:coarse sand). Figures 12 and 13
show this process. Then a second layer of mud plaster 15 mm (0.6 in.) thick of the same mixture
was placed (Figure 14). Finally, two days after the second layer was finished, the cracks caused
by drying shrinkage of the mud plaster were repaired by applying a fluid mixture of soil and fine
sand in the proportional volume of 1:1 to the wall’s surface, using a paintbrush. The model was
then left to dry for one month before the test (Figure 15).
In contrast to the reinforced models M1 and M2 that were completely plastered, only the
east wall was plastered in model M0. Furthermore, the wooden collar beam that ties all walls
was not placed in model M0. The roof was directly placed upon strips of wood fitted on the
Figure 14. Placing the second plaster layer.

Figure 15. Finished model.
walls in order to simulate the typical construction in unreinforced adobe houses in the rural
areas of Peru.
TEST PROTOCOL AND MODEL INSTRUMENTATION

The shaking table input signal used in these tests was obtained from an accelerogram
recorded in Lima during the 31 May 1970 earthquake that occurred near the central
coast of Peru. The east-west component was subjected to digital filtering and double inte-
gration to obtain a displacement record that was used as the command signal for the shaking
table. Figure 16 shows the 30-second displacement command signal normalized to 1 mm
(0.04 in.) of peak amplitude.
Each test consisted of three phases, defined by the peak command displacement for the
shaking table platform. The first phase had a maximum displacement of 30 mm (1.2 in.), the
second phase a maximum of 80 mm (3.2 in.), and the third phase a maximum of 130 mm
(5.2 in.). The expected peak table accelerations, based on previous similar tests, were
approximately 0.30 g, 0.70 g, and 1.20 g, respectively.
No equivalence has been established between the PUCP shaking table motion and a
specific earthquake ground motion. This is because the unidirectional shaking table can
Figure 16. Command signal with maximum displacement normalized to 1 mm.

reproduce only one component of the ground shaking, thus inducing lower stresses than those
that would have occurred in the field, where the ground shakes in all directions simulta-
neously. For instance, in the Pisco area, where most of the adobe houses collapsed or suffered
significant damage during the 2007 earthquake, the modified Mercalli intensity was between
VII and VIII (USGS 2007). The estimated peak ground acceleration within these intensity
levels is 0.34 g. Most of the adobe houses were also destroyed during the Nazca, Peru, earth-
quake of 1996 in the areas where the MM intensity was between VII and VIII (EERI 1996).
Similarly, in the Atico, Peru, 2001 earthquake, the ground acceleration was lower than 0.30 g
(Fierro 2001). Therefore, although Peruvian traditional unreinforced adobe houses collapse at
ground acceleration levels of about 0.30 g, similar structures subjected to unidirectional shak-
ing table motions collapse at acceleration levels above 0.60 g. To date this effect has not been
studied thoroughly for adobe construction, and therefore unidirectional shaking table tests are
used mainly to compare the relative seismic performance of similar structures with different
reinforcement configurations, such as in this study.
It is expected that the comparison of the seismic response of model M1 (where the long-
itudinal walls are subjected to in-plane forces only and the transverse walls to out-of-plane
forces only) to that of model M2 (where all four walls are subjected simultaneously to both
in-plane and out-of-plane forces) will be a first step towards the study and understanding of
the effects of multidirectional shaking on the dynamic behavior of adobe structures.
Nevertheless, from the observed damage of adobe houses during past earthquakes, and in
numerous shaking table tests carried out with the same command signal, the three test phases
applied in this study may be considered analogous to mild, moderate, and strong earthquakes.
It is expected that a mild earthquake would cause minor cracking on the walls of a typical
unreinforced adobe house; a moderate earthquake would cause significant cracking,
especially in the intersections of the adobe walls; and a strong earthquake would cause
total or partial collapse of the structure.
Before each of the three test phases was carried out and at the end of the test, the model
was subjected to a sequence of rectangular displacement pulses (Figure 17) that induced free
vibration motion. This was done to estimate the vibration period and equivalent viscous
damping of the models.
Figure 17. Free vibration command signal.

Figure 18. Ground plan and elevation distribution of instruments in models M0 and M1.
The instrumentation in models M0 and M1 was based on previous projects carried out at
PUCP (Torrealva and Acero 2005; Blondet et al. 2006). Ten accelerometers were placed on
the model to record absolute accelerations (referred as A# in Figure 18) and eight linear
variable differential transformers (LVDTs) were used to record absolute displacements
(referred as D# in Figure 18). Additionally, one LVDT, one accelerometer, and one
force sensor were located at the platform actuator and recorded test data as well.
It was necessary to modify the instrumentation layout in model M2 to record response
data properly due to the change of orientation of the model. Twelve accelerometers and
Figure 19. Ground plan and elevation distribution of instruments in model M2.
twelve LVDTs (referred as A# and D#, respectively, in Figure 19) were placed on the model.
The same instruments were located at the table platform as in the previous models.
UNREINFORCED MODEL M0 TEST

The unreinforced adobe model was tested in a previous project (Blondet et al. 2006),
where shaking table tests were carried out using the same test parameters. Results from
this test were used as a reference and for comparison with the results from the reinforced
models. Figure 20 shows the state of this model before the test.
During the first phase, there was no damage in the model (Figure 21a). During the second
phase, the model suffered serious damage. Large diagonal cracks occurred in the east and west
walls due to shear forces, and vertical cracks due to out-of-plane forces occurred on the north
and south walls (Figure 21b). The difference between the east wall (plastered) and the west wall
(unplastered) was noticeable. While the first wall was barely damaged, the second wall was
severely affected. At the end of this phase, all LVDTs were removed from the model to prevent
them from being damaged due to the imminent collapse of the structure. Thus, no displace-
ments were recorded in the following test phase. During the third phase, the model suffered a
partial collapse: North and south walls collapsed due to out-of-plane failure. Although east and
west walls remained standing, they suffered significant diagonal cracking and some adobe
bricks broke into pieces. This damage pattern is typical of contemporary vernacular adobe
houses in many countries, which are built to imitate the architectural features of clay brick
masonry houses: They are irregular in plan and elevation, do not have seismic reinforcement,
and their walls have large window and door openings. When an earthquake occurs, the out-
of-plane seismic forces produce large vertical cracks at the corners, and in-plane shear forces
produce diagonal cracks in the walls. The walls are thus broken in large independent pieces,
which fall down, causing in turn the collapse of the roof (Blondet and Villa Garcia 2004). In
this case, due to the unidirectional shaking in the longitudinal (north-south) direction, the east
and west walls were not subjected to out-of-plane forces and remained standing, barely sup-
porting the roof with the total integrity of the model almost lost (Figures 21c and 21d). This test
Figure 20. Unreinforced model—M0.

Figure 21. Damage in model M0.
underscores the importance of a collar beam in preventing or at least delaying the out-of-plane
collapse of the adobe walls due to strong seismic shaking.
REINFORCED MODEL M1 TEST

This model was fully reinforced with Tensar BX4100 geogrid. East and west walls of the
model were parallel to the movement of the shaking table (Figure 22).
During the first phase, tiny cracks appeared in the corner of the window openings and in
the inside corners of the walls (Figure 23a). These cracks were only in the mud plaster. There
was no structural damage in the model. In the second phase these cracks grew significantly
and slight structural damage was observed (Figure 23b). During the third phase, larger
relative displacements between walls were observed. In the areas where maximum shear
and out-of-plane forces occurred, the plaster fell from the wall (Figure 23c). After the
test was finished, the plaster was removed and it was discovered that the geogrid was
torn in the regions near the window corners, due to stress concentrations at these locations.
Most of the geogrid damage occurred near the bottom corner of the west wall, where large
diagonal cracks also appeared (Figure 23d). In conclusion, moderate damage was observed in
the model, and the integrity of the structure was maintained.
Figure 22. Reinforced model—M1.
Figure 23. Damage in M1 model: (a) post-phase 1; (b) post-phase 2; (c) post-phase 3; (d) geogrid
failure.
Figure 24. Reinforced model—M2.
REINFORCED MODEL M2 TEST

This model was fully reinforced with Tensar BX4100 geogrid as well. The model was
rotated 45° with respect to the shaking table movement (Figure 24).
No damages were noticed after the first test phase (Figure 25a). During the second phase,
small cracks in the mud plaster were observed in the corners of the windows and door open-
ings (Figure 25b). No structural damage was observed during this phase. During the third
phase, diagonal cracks occurred in the east and west walls due to shear and out-of-plane
forces (Figure 25c). However, the model slid approximately 20 mm from its base during
this phase (Figure 25d). Only slight structural damage—with cracks less than 20 mm in
width—was observed in the model at the end of the test.
COMPARISON OF RESULTS
Special emphasis was placed on comparing the results of the unreinforced model to the
reinforced models in order to evaluate the contribution of the geogrid reinforcement tech-
nology. In addition, it was of interest to evaluate the influence on the seismic behavior of
changing the orientation of the model relative to the shaking direction.
After comparing the seismic behavior of the models and classifying the damage levels, it
was evident that the geogrid reinforcement technology reduced the damages significantly
from a nearly collapse state in model M0 to a moderate damage state in model M1. In
some corners of the windows and door openings, the geogrid was found to be torn up,
which demonstrates that the reinforcement worked properly, confining the walls and with-
standing tensile stresses. Table 2 shows a comparison of the seismic behavior of the models at
each phase of the test.
Plaster separation and wall cracking that does not go through the width of the wall is
considered slight damage. The presence of larger cracks across the width of the wall, usually
Figure 25. Damage in model M2: (a) Post-phase 1; (b ) post-phase 2; (c) post-phase 3; (d) post-
phase 3.
stepwise within the mortar-block interface or near wall openings and corners but that do not
cause the separation of the wall into large pieces, is considered moderate damage. Wall
cracking that causes separation of the wall in large pieces, usually followed by partial
collapse, is considered serious damage.
In general, model M2 suffered less damage than model M1. Although this difference was
clearly observed during the tests, the imposed displacements of the shaking table could have
Table 2. Seismic behavior of the models
Maximum displacement D0 (mm) M0 M1 M2(*)

30 No damage No damage No damage
80 Serious damage Slight damage No damage
130 Partial collapse Moderate damage Slight damage
p
*In the direction of the walls the displacement of the platform was D0 ∕ 2.
Table 3. Damping factors in free vibration tests
Description Model VL0 VL1 VL2 VL3

M0 8.6% 9.2% 13.8% –
East wall M1 7.9% 11.5% 11.9% 16.0%
M2 8.9% 10.1% 10.4% 11.1%
M0 7.0% 8.1% 8.5% –
West wall M1 8.6% 9.9% 11.4% 17.1%
M2 6.2% 6.6% 10.1% 10.2%
M0 8.4% 9.4% 9.4% –
South wall M1 6.9% 10.7% 11.7% 14.7%
M2 6.0% 8.3% 8.5% 8.8%
M0 11.2% 13.6% 13.9% –
North wall M1 8.7% 8.3% 9.6% 12.7%
M2 7.8% 8.3% 9.1% 9.2%
had an influence on these results. In model M2, all walls were subjected to both shear and out-
of-plane stresses, but in the direction parallel to the walls the displacement of the platform was
70% of the maximum displacement of M1. This difference could not be avoided due to the
limitation in the maximum allowed displacement of the table platform of 150 mm (6 in.).
The increase in damage in the models was also estimated from the increase of the damp-
ing factor measured during the free vibration tests carried out before phase 1 (VL0) and after
each of the three phases of the dynamic tests (VL1, VL2, and VL3, respectively). Table 3
shows the damping factors obtained from the accelerometers located at the top center of each
wall using the method of logarithmic decrement.
An increment of the damping factor values was observed in all four walls after each phase
of the test. Values in models M1 and M2 almost doubled at the end of the test. However, a
lower increment in these values occurred in model M2 as corroborated with the lower
damages observed. On average, the increment of values in model M0 was greater than
the ones in the reinforced models, although this could only be corroborated with the instru-
ments up to phase 2.
Using the data from the sensor at the top center of the north wall that measured transverse
displacements of the wall and the force sensor, lateral force versus relative displacement
graphs for each model were obtained (Figure 26).
For models M0 and M1 this instrument was D4 (Figure 18). For model M2, the sum of
the components of D1 and D2 (Figure 19) in the transverse direction of the wall was used.
The damage in this wall was mostly due to out-of-plane bending, thus this instrument was
taken as representative of this type of damage. The relative displacement (DR) was obtained
using the following equation:
DR# ¼ D# − D0
Figure 26. Lateral force vs. north wall top displacement, phase 2 (80 mm peak table displace-
ment): (a) unreinforced model (M0); (b) Reinforced model (M1); (c) ) reinforced model (M2).
where DR# is the relative displacement of the instrument # (mm), D# is the total displacement
of the instrument # (mm), and D0 is the table platform displacement (mm).
Data from the force sensor represented the amount of force needed to move the platform
according to the command signal. In order to obtain the lateral force applied at the base of
the model, this value had to be reduced by the inertial forces due to the platform and
foundation ring masses. The lateral force V applied to the base of the whole structure
was thus computed as:
V ¼ F a − ðPp þ Pa Þ × A0
where Vis the lateral force applied at the base of the model (kN), F a is the force applied to the
platform (kN), Pp is the platform weight ¼ 176 ; 58 ðkNÞ, Pa is the foundation ring weight
(depends on model) (kN), and A0 is the acceleration at the table platform in terms of g.
The graphs in Figure 26 describe the performance of the north wall during phase 2 of
testing. In model M2, the equation to obtain the lateral, in-plane force (i.e., the base force
parallel to the north-south model axis) was changed due to the orientation of this model with
respect to the direction of table motion. The lateral force applied to the model was obtained
with Equation 3 below:
p
VM2 ¼ V∕ 2 (3)
where VM2 is the lateral force component applied at the base of the model M2 (kN), and V is
the total lateral force applied at the base of the model (kN).
Figure 26 shows that during phase 2 both reinforced models remained elastic, and thus
undamaged, whereas unreinforced model M0 went beyond the elastic range, dissipated a
significant amount of energy, and suffered major damage, almost reaching collapse state
at the end of this phase.
Data from the accelerometers at the top of the north wall in phase 2 were compared in
Figure 27 in order to observe the influence of the geogrid reinforcement. The bottom graph
corresponds to the accelerometer located on the platform of the shaking table.
The accelerations recorded in the reinforced models were significantly lower than
the ones recorded in model M0. Maximum accelerations in models M1 and M2 were
73% and 51% of those in model M0, respectively. Also, it is possible to note that for
model M0 the dominant frequency of vibration at the end of the phase is significantly
lower than at the beginning, which indicates loss of global stiffness due to damage in
the wall.
It was of special interest to compare the graphs of lateral force versus relative displace-
ment envelopes of the adobe walls. The procedure to obtain the envelope of a wall’s behavior
in a cyclic test (e.g., Torrealva 2009) was modified by the authors in order to apply it to a
dynamic test. Data from the LVDTs at top center of the north and east walls were used to
calculate relative displacements. In order to obtain the response envelopes, the total duration
of each phase was divided into 12 intervals of 2.5 seconds each. For each interval, a graph
of lateral force versus relative displacement was obtained. Then, the data points where
Figure 27. Recorded accelerations for phase 2—top north wall.
minimum and maximum displacements and forces occurred were identified and selected.
Therefore, four points for each interval and 48 points for each phase were obtained.
These data points were plotted on the first quadrant by taking absolute value of each ordinate,
and the envelope of all points was generated. Lateral forces were divided by the mass of the
whole structure so as to eliminate the influence of this variable in the graphs. The ordinates
thus represent seismic coefficient or pseudoacceleration (base shear per unit mass). Markers
indicate where change of phase occurred. Figure 28 shows the envelopes for the west wall
(subjected to in-plane forces in models M0 and M1) and Figure 29 for the north wall
(subjected to out-of-plane forces in models M0 and M1).
A great difference is shown in Figure 28 between the envelope of model M1 (reinforced)
and model M0 (unreinforced). Model M1 reached approximately 155% of the base shear force
of model M0. Approximately 10% of this strength increase can be attributed to the extra 25 mm
(1 in.)-thick mud plaster applied on each face of model M1 walls (600 mm total longitudinal
wall thickness), whereas only the east wall was plastered in M0 (550 mm total longitudinal wall
thickness). Therefore, the remaining strength increase was provided by the geogrid. The dif-
ference in elastic stiffness is also noticeable in Figure 28, where model M1 is significantly
stiffer than M0. (A comparison of the average slopes of the lateral force versus relative dis-
placement graphs shows that the lateral stiffness of M1 is twice that of M0.) It seems that the
mud plaster placed on the walls of M1 was able to provide adequate confinement to the rein-
forcement, thus allowing for a good integration between the adobe and geogrid.
These results prove that the reinforcement technology provides a considerable enhance-
ment in the seismic strength capacity of the adobe models. Moreover, large displacements
and damages occurred in M0 during phase 2, whereas in the reinforced models these did not
occur until phase 3. Furthermore, model M0 was nearly incapable of withstanding forces at
the end of phase 2. The damage at this state was due to in-plane shear forces. Significant
diagonal cracks appeared throughout the wall in one or two directions (X shape), reducing the
wall section and ultimately leading to its collapse, as shown in Figure 21.
A large amount of displacement ductility, of about 8, was observed in model M1, which
reached a maximum displacement of 83 mm (3.3 in.) with a reduction of the maximum lateral
force of 50%. A lower displacement ductility of approximately 4.4 was observed in model M2
due to the lower input in the table platform, with a reduction of 10% of the maximum lateral
force. The walls in these two models were still capable of dissipating energy at the end of the test.
Similar behavior was observed for the north wall, as shown in Figure 29. For the
same amount of displacement, the walls in model M2 withstood lower lateral forces
Figure 28. Lateral force vs. relative displacement envelope—west wall.

Figure 29. Lateral force vs. relative displacement envelope—north wall.
(approximately 74%) than those in model M1. This trend was observed up to phase 2.
However, during phase 3 a sliding occurred at the base of model M2 (Figure 25d), thus
not letting the walls reach their real strength capacity. No conclusion could be made for
this phase. Nevertheless, from data of the first two phases of the tests, it was evident that
the action of combined (and in-phase) forces in parallel and perpendicular direction to the
walls reduced the amount of lateral force that the structure could withstand.
Finally, Table 4 presents peak values of relative displacement and base shear force cal-
culated for the east and west walls. Angular distortion, computed as relative displacement
divided by the height of the instrument, and average shear stress are also presented there.
It is interesting to note that, during phase 2, the average peak shear stress computed for
model M0 was 41.6 kN/m2. This value, slightly higher than the elastic shear strength of the
unreinforced adobe walls (Figure 26a), is consistent with the shear strength of 41 kN/m2
obtained during cyclic lateral testing of a full-scale unreinforced adobe wall (Blondet
et al. 2005). The presence of the geogrid and the additional mud plaster increased this
strength by 47%. Since only the east wall was plastered in model M0, the deformation
of the unplastered west wall was almost five times larger than that of the east wall,
where the mud plaster increased its thickness and thus its lateral stiffness and strength.
Although both models M1 and M2 were clearly within their elastic range, due to the reduced
input in its longitudinal direction, the response parameters estimated for model M2 are con-
sistently lower than those of model M1.
During phase 3, the transverse walls of model M0 collapsed. Therefore, measurements
taken during that phase are unreliable and are not shown on Table 4b. Behavior of models M1
and M2 was clearly beyond the elastic range. The negative slope of the force-displacement
envelope indicates progressive stiffness and strength deterioration of both models due to
spalling of the mud plaster, cracking of the adobe walls, and tearing of the geogrid in
model M1. Sliding of model M2 over its concrete base most probably provided some iso-
lation and prevented the development of larger shear forces in its longitudinal walls. As in the
case of phase 2, response parameters estimated for model M2 are consistently lower than
those of model M1.
Table 4. Summary of peak response of longitudinal walls during phases 2 and 3
(a) During phase 2 (D0 ¼ 80 mm)
Both walls
West wall East wall (average values
(peak values) (peak values) at peak force)
Relative Angular Relative Angular Shear Shear

displ. distortion displ. distortion force stress Longitudinal table
Model mm % mm % kN kN/m2 accel. g
M0 71.1 3.36% 15.0 0.71% 77.9 41.6 0.741

M1 5.2 0.24% 4.2 0.20% 130.2 61.3 0.721
M2 2.7 0.13% 1.7 0.08% 78.5 36.9 0.492
(b) During phase 3 (D0 ¼ 130 mm)
Both walls
West wall East wall (average values
(peak values) (peak values) at peak force)
Longi
Relative Angular Relative Angular Shear Shear Relative Angular tudinal
displ. distortion displ. distortion force stress displ. distortion table
Model mm % mm % kN kN/m2 mm % accel. g
M1 83.5 3.95% 64.5 3.05% 147.0 69.2 10.1 3.50% 1.121
M2 29.3 1.39% 33.1 1.57% 97.9 46.1 3.1 1.48% 1.076
CONCLUSIONS AND RECOMMENDATIONS

• The geogrid reinforcement proposed in the manual significantly reduces the seismic
vulnerability of adobe wall structures, reducing the damage level and maintaining the
structure’s integrity even when subjected to large displacements and accelerations in
the shaking table at the PUCP’s Structures Laboratory. Further research that includes
variations in the configuration of the models to simulate typical adobe structures built
in different parts of Peru should be carried out in order to corroborate these results.
• The reinforcement technology demonstrated itself to be compatible with the defor-
mation response of adobe structures and capable of developing ductile behavior
when the structures are subjected to dynamic lateral forces.
• Model M2 was rotated 45° in order to obtain in-plane and out-of-plane responses in
all walls so as to represent real lateral movement of a structure during an earthquake.
Combined in-plane shear and out-of-plane stresses in adobe wall structures are
shown to influence their seismic behavior, reducing the maximum shear force in
the elastic phase that they could withstand when subjected to only one type of stress.
Further experimental and analytical research is required to study and understand the
effect of multidirectional seismic shaking on the behavior of adobe walls.
• The dynamic performance of the proposed construction technology published in the

manual Building Hygienic and Earthquake-Resistant Adobe Houses Using Geo-
mesh Reinforcement, tested in the PUCP’s shaking table, demonstrated that it allows
the adobe wall structure to withstand movements associated with strong earthquakes
while suffering only moderate damage.
ACKNOWLEDGEMENTS
The authors are grateful to Luis Guzman-Barron, PUCP former rector, who provided
funds for this project; to Professor Julio Vargas who co-directed the experimental phase;
to engineer Felipe Montoya, who helped develop the final version of the paper; to the
undergraduate and graduate students who helped during the tests; and to the Structures
Lab personnel.
REFERENCES
Bossio, S., 2010. Evaluación del Comportamiento Sísmico e Influencia de la Dirección del
Movimiento en Módulos de Adobe Reforzado con Geomalla, Tesis para optar el Título
Profesional de Ingeniero Civil (Evaluation of the Seismic Behavior and Influence of Seismic
Movement Direction on Adobe Modules Reinforced with Geogrid, Thesis to obtain the Civil
Engineering Professional Diploma), Pontificia Universidad Católica del Perú, Lima.
Blondet, M., and Villa Garcia, G., 2004. Earthquake resistant earthen buildings?, Proceedings,
13th World Conference on Earthquake Engineering, Vancouver, B.C., Canada.
Blondet, M., Torrealva, D., Villa García, G., and Ginocchio F y Madueño, I., 2005. Using
industrial materials for the construction of safe adobe houses in seismic areas, EarthBuild2005
International Conference, University of Technology, Sydney, Australia.
Blondet, M., Vargas, J., Velasquez, J., and Tarque, N., 2006. Seismic reinforcement of adobe
houses using external polymer mesh, Proceedings, First European Conference on Earthquake
Engineering and Seismology, Geneva, Switzerland.
Blondet, M., Vargas, J., Patron, P., Stanojevich, M., and Rubiños, A., 2008a. Human develop-
ment approach for the construction of safe and healthy adobe houses in seismic areas,
Proceedings, 14th World Conference on Earthquake Engineering, Beijing, China.
Blondet, M., Vargas, J., and Tarque, N., 2008b. Available low-cost technologies to improve the
seismic performance of earthen houses in developing countries, Proceedings, 14th World Con-
ference on Earthquake Engineering, Beijing, China.
Blondet, M., Vargas, J., and Tarque, N., 2008c. Observed behavior of earthen structures during
the Pisco (Peru) earthquake of August 15, 2007, Proceedings, 14th World Conference on
Earthquake Engineering, Beijing, China.
Fierro, E. 2001. Initial Report on 23 June 2001 Atico, Peru Earthquake—Part II, published in
http://www.eeri.org/lfe/pdf/peru_arequipa_initial_reconnaissance_part2.pdf.
Earthquake Engineering Research Institute (EERI), 1997. Learning from earthquakes: The Nazca,
Peru, earthquake of November 12, 1996, Special Earthquake Report from the January 1997
Newsletter.
Instituto Nacional de Defensa Civil (INDECI), 2009. Lecciones Aprendidas del Sur—Sismo de
Pisco 15 Agosto 2007 (Lessons Learned from the South—15 August 2007 Pisco Earthquake),
Instituto Nacional de Defensa Civil, Lima, Peru, 232 pp.
Peralta, G., 2009. Resistencia a Flexión de Muros de Adobe Reforzados con Geomallas—
Influencia del Tipo de Tarrajeo, Tesis para optar el Título Profesional de Ingeniero Civil
(Flexural Strength of Adobe Walls Reinforced with Geogrid—Influence of the type of Plaster,
Thesis to obtain the Civil Engineering Professional Diploma), Pontificia Universidad Católica
del Perú, Lima, Peru.
Reglamento Nacional de Edificaciones (RNE), 2006. Norma Técnica E.080 Adobe (Technical
Code E.080 Adobe), Lima, Peru.
Tensar International Corporation (Tensar), 2009. Product Specification Tensar Biaxial Geogrid,
Tensar, Atlanta, GA, 13 pp.
Torrealva, D., 2009. Diseño Sísmico de Muros de Adobe Reforzados con Geomalla (Seismic
design for reinforced adobe walls with geogrids), Pontificia Universidad Católica del Perú,
Departamento de Ingeniería, Publicación DI-SIC-2009-01, Lima, Peru.
Torrealva, D., and Acero, J., 2005. Reinforcing adobe buildings with exterior compatible mesh:
The final solution against the seismic vulnerability?, SismoAdobe 2005: International Seminar
on Architecture, Construction and Conservation of Earthen Buildings in Seismic Areas, Lima,
Peru.
U.S. Geological Survey (USGS), 2007. ShakeMap 2007gbcv, Published in http://earthquake.usgs
.gov/earthquakes/shakemap/global/shake/2007gbcv.
Vargas, J., Blondet, M., Ginocchio, F., and Villa Garcia, G. 2005. 35 años de investigaciones en
sismo adobe: La tierra armada (35 years of research on SeismicAdobe: Reinforced earth),
SismoAdobe 2005: International Seminar on Architecture, Construction and Conservation
of Earthen Buildings in Seismic Areas, Lima, Peru.
Vargas, J., Torrealva, D., and Blondet, M., 2007. Building Hygienic and Earthquake-Resistant
Adobe Houses using Geomesh Reinforcement for Arid Zones, 1st edition, Pontificia
Universidad Católica del Perú, Lima, Peru, 41 pp.
Zapata, I., and Torrealva, D., 2010. La Restauración de la Casa O´Higgins en Lima (Restoration
of the O’Higgins House in Lima), Congreso Internacional de Rehabilitación del Patrimonio
Arquitectónico y Edificación CICOP 2010, Centro de Extensión Pontificia Universidad
Católica de Chile, Santiago, Chile.
Limit State Strength of Unreinforced
Masonry Structures
A. H. Akhaveissy a)
This paper presents a new formula to estimate the ultimate lateral force of
unreinforced masonry structures. The ratio of the wall’s height to the wall’s
width is used to predict the ultimate lateral load of the wall. The coefficient
is determined by the numerical implementation of an interface model to simulate
the behavior of mortar joints in masonry walls. The numerical predictions are
compared with the FEMA guidelines and the experimental data. The comparisons
show that the loads that are predicted by the proposed formula have a lower error
percentage than the FEMA guidelines. Thus, the proposed formula can be used to
analyze unreinforced masonry structures. [DOI: 10.1193/1.4000097]
INTRODUCTION
Masonry is the oldest building material that is still widely used in building industries. In
the past two decades, there have been important new developments in the materials and appli-
cations of masonry. Masonry is a composite material that consists of units and mortar joints.
Masonry buildings are constructed in many earthquake-prone parts of the world due to their
low costs. The seismic performance of existing unreinforced masonry buildings in North
America was considered in a state-of-the-art paper (Bruneau 1994) that described the various
failure modes of unreinforced masonry buildings that were subjected to earthquake excita-
tion. The damage to the existing buildings was evaluated for different earthquake scenarios.
Thus, knowledge of the seismic behavior of the buildings is necessary to evaluate the seismic
performance of these types of buildings. A pushover analysis is often used to evaluate the
seismic performance of a building and to determine the capacity curve. The analysis traces
the entire evolution of the structural response to an increasing external loading. When eval-
uating the structure’s ultimate load-carrying capacity, the final stage of the plastic response–
plastic collapse is of primary interest. The plastic limit load can be assessed without analyzing
the entire history of the response (Jirasek and Bazant 2002). Thus, this paper considers the
ultimate strength of an unreinforced masonry wall.
In general, both micro- and macro-modeling approaches are used to analyze masonry
buildings. However, the macro-modeling approach is more practice-oriented due to its
lower time and memory requirements and its user-friendly mesh generation. When using
the macro-element method to analyze an unreinforced masonry building, the compressive
strength of a masonry unit is an important parameter. A masonry unit includes the mortar
joints and the masonry bricks. The compressive strengths of masonry units constructed with
a)
Department of Civil Engineering, Faculty Engineering, Razi University, P.O. Box: 67149–67346, Kermanshah,
Iran. Tel.: +98 831 427 4535; fax: +98 831 427 4542; Email address: Ahakhaveissy@razi.ac.ir
1
2 A. H. AKHAVEISSY
different mortars were evaluated (Kaushik et al. 2007a and 2007b, Tena-Colunga et al. 2009,
Chaimoon and Attard 2009, Vasconcelos and Lourenco 2009).
The static pushover curve was studied using the boundary element method for unrein-
forced masonry walls (Alessandri and Brebbia 1987). In the analysis, the masonry buildings
were modeled using a no-tension material with infinite compression strength. The predicted
results correspond well with the experimental data. A macro-element approach to the three-
dimensional seismic analysis of masonry buildings was applied (Brencich et al. 1998). The
base shear force of the full model is approximately 25% higher than the value that was
calculated for the plane structure. A constitutive model was developed based on the homo-
genized anisotropic elasto-plasticity. The effect of the anisotropy was taken into account
using a fictitious isotropic stress and strain space. In this model, the classical theory of plas-
ticity can be used to model nonlinear behavior in the isotropic spaces (Lopez et al. 1999).
The in-plane shear behavior of hollow brick masonry panels was evaluated (Gabor et al.
2006). The masonry’s nonlinear behavior was modeled in the ANSYS 5.4 commercial soft-
ware by assuming that the mortar joint had an elastic-perfectly plastic behavior (according to
the Drucker-Prager model). Thus, the micro-element method was used to analyze the panels.
The experimental results and the results of the numerical analysis correspond well.
A simple equilibrium model was used to estimate the ultimate capacity of masonry shear
walls. The model was based on the strut-and-tie schemes that represent the combination of
the compression and the tension stress fields at the ultimate condition. The model results and
the experimental results for the dry-joint and the mortar-joint masonry correspond well
(Roca 2006).
Milani et al. (2007) performed a three-dimensional homogenized limit analysis to deter-
mine the ultimate lateral load of full masonry structures. Linearized homogenized surfaces
for masonry in six dimensions (Milani et al. 2006, Cecchi and Milani 2008) were obtained
and implemented in a finite element code. From the comparisons of the predicted results from
the three-dimensional homogenized limit analysis and the experimental data, the analysis has
an error of approximately 12%. Milani et al. (2010) also used a three-dimensional homo-
genized limit analysis for full masonry buildings that were reinforced by fiber-reinforced
plastic (FRP). The errors between the predicted results and the experimental data for a
two-story masonry building with and without FRP are 9.4% and 4.6%, respectively. Milani
(2011b) used the three-dimensional homogenized limit analysis to determine the limit load of
a wall under in-plane and out-of-plane loading.
SAP2000 v.10, a software package with a user-friendly interface that is widely used by
practicing engineers, was used to seismically analyze masonry buildings (Pasticier et al.
2008). Two unreinforced stone-masonry walls in the Catania Project were modeled with
SAP2000 v.10. The static pushover curves from the analyses were compared with predicted
results from the SAM code, which was developed by the University of Pavia, the Genoa
research group, and the Basilicata research group (Pasticier et al. 2008). The Basilicata
research group used a no-tensile-strength, macro-element model with crushing and shear
failures, while the Genoa research group used a finite element model with layer failures.
The predicted ultimate base shear forces for wall A are 1,682 kN, 1,339 kN, 1,115 kN,
and 1,395 kN according to the Genoa research group, the SAP2000, the SAM code, and
LIMIT STATE STRENGTH OF UNREINFORCED MASONRY STRUCTURES 3
the Basilicata research group, respectively. The predicted ultimate base shear forces for wall
B are 650 kN, 474 kN, and 508 kN according to the Genoa research group, the SAP2000 and
the SAM code, and the Basilicata research group, respectively. As shown by these results,
there are differences between the different studies. Thus, practicing engineers may be con-
fused as to which codes or research are the most applicable or the most precise.
Unreinforced masonry buildings are analyzed by a two-step approach (Milani et al.
2009). In step 1, the ultimate bending moment–shear force strength domains of the masonry
spandrels were derived by a heterogeneous upper-bound finite element limit analysis, and the
results were stored in a database. In step 2, an equivalent frame model of the masonry wall
was assembled. In the frame model, the spandrels were modeled as elastic Timoshenko
beams. At each analysis step, a check was performed to determine whether the internal forces
of the coupling beams were smaller than the failure loads stored in the database (see step 1).
The shear force and the bending moment capacity of the piers were simply estimated accord-
ing to the Italian Design Code. The proposed analysis approach was capable of deriving the
pushover curve of the unreinforced masonry walls (Milani et al. 2009).
Using various aspect ratios and wall-opening positions, a finite element analysis was
conducted for a single-story, one-room masonry building that was subjected to a seismic
force of varying directions (Shariq et al. 2008). The response spectrum method was employed
for the analysis. The predicted results show that the critical direction of the seismic force to
develop the maximum stresses in the walls of a room occurs when the opening is along the
short wall of the room. The maximum principal tensile stress occurred in the short wall, and
the maximum shear stress occurred in the long wall.
The seismic fragility of an unreinforced masonry low-rise building was studied using a
structural modeling method. The method utilizes a simple, composite nonlinear spring and
divides the wall into distinct areas or segments. Each segment of the unreinforced masonry
wall is then represented by a nonlinear spring, and the springs are assembled in series and in
parallel to match the segment topology of the wall (Park et al. 2009). Rota et al. (2010)
presented a new analytical approach to derive the fragility curves of masonry buildings.
The methodology was based on nonlinear stochastic analyses of building prototypes.
Monte Carlo simulations were used to generate the input variables from the probability den-
sity functions of the mechanical parameters.
The ultimate load-carrying capacities of unreinforced masonry structures, such as piers or
spandrel beams, are often calculated by assuming a perfectly plastic material behavior. The
calculation methods are based on the lower-bound or the upper-bound theorems. The upper-
bound method seeks a feasible collapse mode for the minimal load-carrying capacity,
whereas the lower-bound method seeks a stress distribution in equilibrium for the maximum
load-carrying capacity. Thus, Kawa et al. (2008) developed a procedure to assess the strength
of brick masonry based on the homogenization theory. The approach invokes a lower-bound
analysis, and the critical load was obtained by solving a constrained optimization problem.
The Mohr-Coulomb criterion intercepted by Rankine’s cutoff in the tension domain was
accepted for the brick and the mortar failure conditions. The element level was subjected
to different loadings with various loading directions. However, the approach was not applied
to analyze masonry buildings.
4 A. H. AKHAVEISSY
In this study, the final stage of the pushover curves of unreinforced masonry structures is
evaluated. An explicit formulation is proposed to estimate the ultimate load-carrying capacity
of the unreinforced masonry structures. Masonry is a composite material consisting of units
and mortar joints; therefore, the explicit formulation is evaluated based on the modeling of
the units and the mortar joints in the micro-modeling process (Akhaveissy 2011). The brick
units and the mortar joints are modeled by plane stress elements and interface elements,
respectively, using the appropriate constitutive law for each element. Thus, the unreinforced
masonry structure’s response is more accurate than those of other models; this accuracy,
combined with the use of the appropriate constitutive law, increases the reliability of the
structural responses. Consequently, practicing engineers can accurately determine the
base shear force–carrying capacity of unreinforced masonry buildings under seismic excita-
tion using the concepts described in this paper.
THE INTERFACE MODEL

Masonry is a composite material consisting of units and mortar joints. A masonry wall is
constructed by laying pieces of bricks on top of each other using mortar as a cohesion
mechanism. The mortar joint and the bricks can be modeled by interface elements and iso-
parametric elements, respectively. When a masonry wall is subjected to a lateral force, the
mortar joints may slide. Modeling this phenomenon with standard finite elements leads to
numerical ill-conditioning due to the high aspect ratios. The idea of overcoming this problem
by introducing special elements is not new (Lourenço 1994). However, the joint elements
used in this paper were developed by Giambanco et al. (2001), Shieh-Beygi and Pietruszczak
(2008), and Brasile et al. (2010). Sekiguchi et al. (1990) applied six-noded joint elements in
an elasto-viscoplasticity analysis based on the tangential and normal stiffnesses of joint ele-
ments related to the force-displacement relationship. Akhaveissy (2011) reformulated the six-
noded joint element based on the tangential and the normal elasticity modules related to the
stress-strain relationship, as shown in Figure 1.
The elastic stiffness matrix ðK e Þ of a joint element is defined in the standard finite ele-
ment as ð
K e ¼ BT D B dV:
ve
Figure 1. Schematic representation of the global and the local axes of a joint element.
The B matrix is defined as

1
B¼ N; (2)
w
where w and N are the thickness of the mortar joint and the shape function matrix, respec-
tively. As shown below, K e in Equation 1 can be evaluated using the line integral shown
below.
ð1 "# $2 # $2 %1∕2
∂x ∂y
Ke ¼ t BT D B þ dξ; (3)
∂ξ ∂ξ
−1
where t is the width of the joint element and D is the stress-strain matrix in the local coor-
dinate system. The elastic stress-strain relation is defined as
& ' " %& ' & '
dτ ET 0 dγ 1 rξ 1
dσ ¼ ¼ ¼ D d ε; d ε ¼ ¼ r: (4)
dσ 0 E n dεn w r w
¯ ¯ ¯ η
E T and E n are the tangential and the normal elastic modules, respectively, in the local
coordinates of the joint element, and r is illustrated as
8 9
>
> a1 >
>
> a2 >
> >
>
%> >
& ' &
rξ topξ − bottomξ
' " < > =
a3
r¼ ¼ ¼ −N 1 −N 2 −N 3 N 4 N 5 N 6 ¼ Na:
rη topη − bottomη > a4 >
>
> >
>
>
> a >>
: 5>
> ;
a6
& ' " %" % # $
ui n 0 cosθ sin θ ∂y∕∂ξ
ai ¼ ; Ni ¼ i ; θ ¼ tan −1 (5)
vi 0 ni − sin θ cos θ ∂x∕∂ξ
EQ-TARGET;temp:intralink-;ue5;62;302
The shape function ni that is associated with node i is expressed in terms of the local ξ
coordinate as
# $
ni ¼ 2 ξξi 1 þ ξξi
1
i ¼ 1; 3; 4; 6
: (6) EQ-TARGET;temp:intralink-;e6;62;218
ni ¼ 1 − ξ 2 i ¼ 2; 5
The derived x and y values in terms of the local ξ coordinate in Equations 3 and 5 are
components of the Jacobean matrix. Irreversible discontinuous displacements occur when the
stress state reaches a limit condition. Akhaveissy (2011) defined the elastic domain by three
convex limit surfaces that intersect in a non-smooth fashion: the Coulomb criterion for the
shear stress state, the tension cutoff for the tensile stress state, and the compression cutoff for
the compressive strength. The limit functions, which are reported in the stress space (see
Figure 2), take the following form:
6 A. H. AKHAVEISSY
Figure 2. Yield condition represented in the stress space.
" "
ΦI ¼ "τn " þ σ n tanφ − C ¼ 0
ΦII ¼ σ n − SðεÞ ¼ 0 ; (7)

ΦIII ¼ σ n þ F C ðεÞ ¼ 0
where φ is the internal friction angle of the contact layer; ε is the normal strain; C, S, and F C
are the cohesion, the tensile strength, and the compressive strength of the contact layer,
respectively; and σ n and τn are the normal and shear stresses on the interface surface, respec-
tively. These stresses are calculated in the local coordinates.
A vertical linearized cap model, instead of a spherical cap model, was used for the tension
and compression cutoffs for the tensile stress state and the compressive strength, respectively.
Milani (2011a and 2011b) also used these assumptions. The predicted homogenized failure
surfaces in the stress space by Milani (2011a and 2011b) correspond well with the predicted
failure surface and the von Mises criterion of the elastic-perfectly plastic behavior. Thus, in
this study, the brick behavior is evaluated based on the von Mises criterion using an elastic-
perfectly plastic behavior. The mortar joints have no tensile strength after cracking occurs;
thus, a softening incremental model is required to consider the real behavior of the mortar
joints after cracking under increased loading and thus assess the real capacity curve. This
model type enables the internal stresses due to the cracking of the material to be redistributed.
Therefore, the intact particles of the material tolerate the contribution of the cracked particles.
The limit analysis can be used to calculate the ultimate lateral load. Although the limit ana-
lysis can determine the limit load, the nonlinear incremental method is used for the analyses
in this study. Therefore, a softening incremental model is used to assume a mortar tensile
stress of zero after cracking to evaluate the entire history of the response and failure mode in
the history of loading. To avoid ill-conditioning in the numerical analysis, the tensile stress
linearly decreases from the maximum tensile strength to zero (see Figure 4b). Akhaveissy
(2011) provides more details of the formulation of the interface elements. In the next section,
the micro-modeling of mortar joints and bricks is shown to accurately model the failure of
unreinforced masonry walls. In the finite element analysis, two-dimensional, eight-node and
six-node isoparametric elements are used to simulate the blocks and the contact elements,
respectively, that are used for the mortar joints and units. A formulation to estimate the
ultimate shear strength of the unreinforced masonry walls is then presented based on
the micro-modeling analyses. Thus, practical engineers can accurately and quickly determine
the ultimate base shear force of unreinforced masonry buildings.
NUMERICAL EXAMPLE
THE MASONRY SHEAR WALL EXAMPLE

This example simulated the mechanical response of the masonry wall illustrated in
Figure 3. Experimental tests on a masonry wall were performed by Vermeltfoort et al.
(1993). The wall was made of wire-cut solid clay bricks, which had dimensions of
210 × 52 × 100 mm3 and 10-mm-thick mortar joints. The wall was characterized by a
height-to-width ratio of one, with dimensions of 1000 × 990 mm2. Two stiff steel beams
at the horizontal boundaries of the test setup are used to clamp the top and the bottom
of the wall. The geometry of the model and the adopted boundary conditions are shown
in Figure 3. At the top horizontal side of the wall, the vertical degrees of freedom are con-
strained using a uniform distribution of springs with a stiffness of k that simulate the stiffness
of the test apparatus (Giambanco et al. 2001). Two initial vertical loads p were applied to the
top of the wall. The loads p were 0.3 MPa (30 N/cm2) and 1.21 MPa (121 N/cm2).
Table 1 provides the elastic properties of the bricks and the joint with the interface para-
meters that characterize the strength properties. The parameters were determined for an initial
vertical pressure of 0.3 MPa by Akhaveissy (2010, 2011). The wall with an initial vertical
pressure of 0.3 MPa was also analyzed (Akhaveissy 2011). The wall is analyzed for both of
Figure 3. (a) Solid masonry wall geometry and the initial vertical load of 30 N/cm2 and (b) the
boundary and the loading conditions (Giambanco et al. 2001, Vermeltfoort et al. 1993).
8 A. H. AKHAVEISSY
Table 1. Properties of the bricks and the mortar joints (in terms of N/cm2) (Giambanco et al.
2001)
Brick Mortar joint
Applied pressure (N/cm2) E ν Fc Et E ν C ϕ Fc Ft

30 1,670,000 0.15 175 0 78,200 0.14 35 37° 150 25
121 1,670,000 0.15 175 0 78,200 0.14 22 37° 150 16
the initial vertical pressures in this study. The failure mode was determined to be bed-joint
sliding. Figure 4 shows the softening behavior of the mortar in tension and the behavior of
mortar in compression. The behavior of the bricks is assumed to be elastic-perfect plastic and
is modeled by the von Mises criterion.
In Table 1, E is the elastic modulus; E t is the tangential modulus; ν is the Poisson ratio; F c
is the compressive strength; F t is the tensile strength; C is the shear strength; and ϕ is the
friction angle.
The compressive strengths of the mortar joint and the brick in Table 1 were not used by
Giambanco et al. (2001). The value of the stiffness k has been previously calibrated such that
the final wall collapse occurs due to a slip-type mechanism (Giambanco et al. 2001). The
stiffness values of 9 × 104 N/cm and 45 × 104 N/cm, which were adopted to simulate the
constraint, were provided by the test apparatus for the applied initial pressures of
0.3 MPa and 1.21 MPa, respectively. In the numerical analysis, the bricks and the joints
are represented by the plane stress continuum elements (eight-noded) and the line interface
elements (six-noded), respectively. Each brick is modeled with 2 × 1 elements. The wall is
modeled by 160 eight-noded isoparametric elements, 222 interface elements, and
2,086 degrees of freedom. The error for the displacement convergence criterion is 1e-5.
Figure 5 compares the predicted load-displacement curve and the test data for both of
the applied initial pressures. As shown in the figure, there is a significant relationship
between the predicted results and the test data.
Figure 4. (a) Compressive and (b) tensile behavior for the mortar joint used in the analysis.
Figure 5. Comparison between the test data and the predicted load-displacement curves.
As shown in Figure 5, the predicted ultimate load is 93.25% of the observed ultimate load
for the applied initial pressure of 0.3 MPa. Similarly, the predicted ultimate load is 98% of the
observed ultimate load for the applied initial pressure of 1.21 MPa. After point B in Figure 5,
the analysis did not converge as the incremental load increased. The maximum number of
iterations was approximately 7,500. After the final iteration, the top displacement of the wall
was 16.35308 mm, which is similar to the Euclidean norm of 0.16839 for the force criterion
(i.e., 16,839 times the reference value, while the reference value was 1e-5). The Euclidean
norm shows that the structure should have collapsed for this amount of loading. Therefore,
points A and B in Figure 5 show the ultimate displacement at the top of the wall. The wall has
a decreased ultimate shear force at these displacements. The program for the analysis is based
on the force control method instead of the displacement control method. In the force control
method, the incremental load increases at each step. Thus, the force degradation cannot be
calculated in this method. Consequently, the force degradation is not shown in the pushover
curve. However, the collapse of the wall and the force degradation are recognized when the
analysis cannot converge during the loading process, as described above. Therefore, the
degradation of the external force is also not shown in the pushover curves of the subsequent
analysis, which use the force control method.
The load-displacement curve that was obtained from analysis corresponds well with the
test data. The shear and the tension failures occurred at the bottom of the wall due to a com-
bination of the shear force and the bending moment. At the top of the wall, the tensile and
shear failures occurred in the vertical and the horizontal mortars, respectively. Figure 6 shows
the failure of the wall for the initial pressure of 0.3 MPa.
As shown in Figure 5, the failure displacements of the wall under the different initial pres-
sures indicate that the masonry wall is fragile. Therefore, the behavior of the masonry walls is
governed by the force control instead of the displacement control. As such, it is useful to esti-
mate the resistant lateral load of an unreinforced masonry wall when evaluating the tolerable
shear force of a masonry building. The ultimate lateral loads of masonry walls with different
10 A. H. AKHAVEISSY
Figure 6. Deformed shape for the last step with an applied pressure of 0.3 MPa (point A from
Figure 5).
height-to-width ratios are then predicted using the applied procedure that was shown in the
previous examples. The calibrated parameters in Table 1 are used for the analysis. The thick-
ness and the width of the wall are 100 mm and 500 mm, respectively. The initial pressures at the
top of the wall are the same as those in the previous examples (0.3 MPa and 1.21 MPa). Figure 7
shows the lateral load–lateral displacement curve for the wall with a width of 500 mm and
different height-to-width ratios under an initial pressure of 0.3 MPa. The lateral load–lateral
displacement curves for the different height-to-width ratios did not converge in Figure 7
because the force control method was used. Thus, the displacement of these points is the ulti-
mate lateral displacement, and the force degradation occurs after these points.
The wall was also analyzed for an initial pressure of 1.21 MPa using the same height-to-
width ratios. Thus, Table 2 shows the ultimate lateral loads for both initial pressures.
Figure 7. Lateral load–lateral displacement curves for different height-to-width ratios and an
initial pressure of 0.3 MPa.
Table 2. The ultimate lateral loads (kN) of a 500-mm-wide wall for different height-to-
width ratios (H/L) and initial pressures of 0.3 and 1.21 MPa
Ultimate lateral load (kN) for an Ultimate lateral load (kN) for an
H/L initial pressure of 0.3 MPa initial pressure of 1.21 MPa
4 7.26 10.89
3 9.63 14.36
2 14.30 20.35
1.5 18.50 27.00
1 23.38 37.13
0.5 33.00 48.75
0.22 35.20 49.5
Figure 8 shows the variation of the ultimate load versus the height-to-width ratios that are
expressed in Table 2.
The ultimate lateral loads for the height-to-width ratios of less than 0.22 are assumed to
be the same as the ultimate load for the height-to-width ratio of 0.22. Thus, the predicted
curves in Figure 8 are scaled to the ultimate lateral load for the height-to-width ratio of 0.22.
These scaled curves are referred to as the strength curves of a masonry wall. Figure 9 shows
the scaled curves and the average of the curves.
The curves in Figure 9 were determined based on a wall with a width of 500 mm; the
curves were then drawn in terms of the height-to-width ratio. As shown in Figure 9, the
strength curves for the initial pressures of 0.3 and 1.21 MPa are similar. Therefore, the ulti-
mate lateral load of the unreinforced masonry walls can be estimated based on the average
curve shown in Figure 9. In Figure 9, α is the dimensionless ratio of the height to the width;
thus, α can be used for every width of the wall and a new formulation can be determined
based on this parameter. The different failure modes of an unreinforced masonry wall can be
Figure 8. Curves of the ultimate lateral load versus the height-to-width ratio.
12 A. H. AKHAVEISSY
Figure 9. Scaled curves of the ultimate load of the wall with H/L = 0.22.
predicted by a closed-form solution. The possible failure modes for unreinforced masonry
walls are shown in Figure 10. These modes are evaluated from the observed collapse of the
unreinforced masonry walls based on results of different numerical analyses to determine the
pushover curves shown in Figure 7. The failure modes are related to the applied initial ver-
tical stress on the wall and the height-to-width ratio of the wall.
Figure 10a shows the full diagonal failure of a wall. This failure mode is common in walls
with height-to-width ratios of approximately one. Figure 10b shows the partial failure of a wall
on an oblique line. This failure mode is common in walls with height-to-width ratios of greater
than one. Figure 10c shows the typical failure mode for walls with height-to-widths ratio of less
than one. Figures 10a and 10b imply that the lateral strength of an unreinforced masonry wall
consists of two internal resistant parts relating to the shear strength of the horizontal mortar
joints and the tensile strength of the vertical mortar joints. The extent to which the height of the
vertical mortar joints in the masonry wall contribute to the lateral strength of the wall (see
Figures 10a and b) is dictated by Mohr’s circle. The principal plane in Mohr’s circle
shows the maximum tensile and the compressive normal stresses. The angle of the principal
plane with the horizontal axis is determined by the status of the stress point (see Figure 11b).
Thus, the principal plane is assumed to start from the top corner of the wall and to continue to
the vertical edge of the wall (see Figure 11a). The contribution of the vertical mortar joints to
the tensile strength is determined by the xmin , which is the distance between the top corner of the
wall and the cross-sectional point of the vertical edge and the principal plane (see Figure 11a).
Based on the upper-bound theorem, the contribution of the horizontal mortar joints to the shear
strength is assumed to be the shear strength of the entire length of the horizontal mortar joints or
the width of the wall (see Figure 10c). However, if the behavior is elastic-perfect plastic (see
Figures 5 and 7), then the strength of the mortar joints is assumed to be the limit of the tensile
and shear strengths of the tensile area and the cross-sectional area of the wall, respectively. The
shear strength can be accepted by the Mohr-Coulomb criteria (ATC 1999, Gabor et al. 2006,
Kawa et al. 2008). It is important to note that the mortar joints can tolerate tensile and shear
Figure 10. Different failure modes of an unreinforced masonry wall: (a) full diagonal failure of a
wall, (b) partial failure of a wall on an oblique line, and (c) typical failure mode for walls with
height-to-widths ratio of less than one.
stresses when the tensile stress in the mortar joints is less than the tensile strength. Thus, the
cross-sectional area of the mortar joints can be used to calculate the lateral strength of the wall.
A model of a material with limited tensile strength is used rather than a no-tension material
model. When a no-tension material model is used in the analysis, the equilibrium of the com-
pressive and the tensile internal forces at the cross section of the wall with bending behavior
and a height-to-width ratio of greater than one is not satisfied. Thus, a limited-tension model is
preferred to a no-tension model. Consequently, the resistant lateral force of the wall is included
in the tensile strength of the collapsed vertical surface and the shear strength of the collapsed
horizontal surface. Accordingly, the resistant lateral force is
PU ¼ τU × L × t þ F t × t × xmin ;
where PU is the resistant lateral force; xmin is the effective length of the wall in tension (see
Figure 11); L is the width of the wall; t is the thickness of the wall; τU is the ultimate shear
strength, which can be calculated by the Mohr-Coulomb criterion; and F t is the tensile
strength of the mortar joints. The effective length of the wall in tension can be represented
14 A. H. AKHAVEISSY
Figure 11. (a) Stress element on the failure surface or the principal plane according to Figure 10a
and (b) the status of the stress point.
in terms of L and θ (see Figure 11a). The first and the second terms of the right-hand for-
mulation are the shear and tensile strengths, respectively.
In Figure 11, θ is the angle between the horizontal and the principal planes. The angle is
determined by Mohr’s circle of the stress point in solid mechanics. The stresses that are
applied at a point on the principal plane are included in both the initial vertical pressure
on the top of the wall and the failure shear stress during the collapse of the wall. The initial
pressure is assumed to be constant, and the weight of the masonry wall is neglected. There-
fore, the status of the stress point may be similar to that illustrated in Figure 11b. From solid
mechanics, the angle of the principal plane is determined by
τxy 2 × τu
tanð2θÞ ¼ ¼
σx − σy σ0
; (11)
2
xmin ¼ LtanðθÞ
where σ 0 is the initial vertical stress on the top of the wall. As shown in Figure 11a, when the
principal plane is the cross section of the wall height, the effective length can be represented
in terms of the width of the wall and the angle of the principal plane (see Equation 11). The
numerical expressions show that the principal plane may be in the cross section with the
width of the wall. Consequently, the effective length can be determined from Equation 12
when the principal plane is in the cross section with the width of the wall.
xmin ¼ ð1 − tanðθÞÞ × h (12)
Therefore, the resistant lateral force (shown in Equation 10) can be determined from Equa-
tions 11 and 12. Equation 10 does not account for different height-to-width ratios, the failure
pattern, the staggered disposition of the bricks, or the limited compressive strength of the
masonry unit. Instead, these effects are accounted for by adding α to the equation. As
previously mentioned, α is determined by a series of nonlinear analyses for unreinforced
masonry walls with different height-to-width ratios. The yield surface shown in Figure 2
was used in the nonlinear analyses. The yield surface was defined by three convex limit sur-
faces that intersect in a non-smooth fashion: the Mohr-Coulomb criterion for the shear stress
state, the tension cutoff for the tensile stress state, and the compression cutoff for the compres-
sive strength. FEMA (ATC 1999), Gabor et al. (2006), and Kawa et al. (2008) have used the
Mohr-Coulomb criterion to establish the shear strength of masonry walls. Therefore, the pre-
dicted results of the nonlinear analyses, such as the compressive stress and α, limit the com-
pressive strength of the masonry unit. Additionally, the bricks and the mortar joints of the walls
are geometrically modeled in the analyses by eight-noded isoparametric serendipity elements
and six-noded contact elements, respectively. When geometrically modeling the unreinforced
masonry walls, the staggered brick disposition can be represented by the serendipity and the
contact elements. Consequently, α is included in all of the above effects. Thus, the lateral
strength can be determined using Equations 10–12, the α coefficient shown in Figure 9,
and the shear factor. The assumed shear factor is 0.88 instead of the typical 0.85. Accordingly,
the formulation shown below is used to estimate the ultimate lateral load.
P ¼ 0.88αPU
PU ¼ τU × L × t þ F t × t × xmin
τU ¼ C þ σ#0 tanðφÞ $ %: (13)
L × tanðθÞ 2τ U
xmin ¼ min ; θ ¼ 0.5 × tan−1
ð1 − tanðθÞÞ × h σ0
In Equation 13, α is obtained from Figure 9; θ is the angle of the principal plane; σ 0 is the
initial pressure on the top of the wall; C and φ are the cohesion and the friction angle of the
mortar joints, respectively; L and h are the width and the height of the wall, respectively; and
F t is the tensile strength of the mortar joints. From Equation 13 and Table 1, the ultimate load
is obtained for the masonry wall in example 1 (Vermeltfoort et al. 1993), as shown in Table 3.
To show the ability of Equation 13, different examples using the finite element procedure
and the formulation that is given in the FEMA 273 (ATC 1999) guideline are compared with
the ultimate load that is predicted by the equation.
Table 3. The ultimate lateral load by the proposed method for the masonry wall of example 1
h
L σ 0 (MPa) τu (MPa) θ xmin (mm) A (Fig. 9) P (kN) Eq. 13 Ptest (kN)
1;000
0.30 .5761 37.70 227 .707 39.02 50
990
1;000
1.21 1.130 30.92 401.1 .707 73.60 72.75
990
16 A. H. AKHAVEISSY
Figure 12. Dimensions of the stone wall: (a) the irregular stone wall with bonding mortar, the
boundary conditions, and the initial loading; and (b) the modeling of the interface element.
VALIDATIONS
THE STONE-MASONRY SHEAR WALL EXAMPLE

Irregular stone masonry with bonding mortar is representative of large stone-block con-
struction (e.g., monumental buildings). The data were obtained from the ancient stone-
masonry shear wall test (Senthivel and Lourenço 2009). The average compressive strength,
the tensile strength, and Young’s modulus of the stone were 69.2 N/mm2, 2.8 N/mm2, and
20,200 N/mm2, respectively. The average compressive strength of the mortar was 3.0 N/mm2.
Figure 12 and Table 3 present the experimental stone-masonry shear wall test. The dimen-
sions of the wall were fixed at 1,000 mm (length) × 1,200 mm (height) × 200 mm (width),
and the height-to-length ratio was 1.2. The dimensions of the sawn stone that was used in the
wall were 200 mm (length) × 150 mm (height) × 200 mm (width). The monotonic lateral load
was conducted with a low axial precompression load level of 100 kN ðσ 0 ¼ 0.5 N∕mm2 Þ.
Figure 4 shows the softening behavior of the mortar in tension and the behavior of the
mortar in compression. Senthivel and Lourenco (2009) did not use the compressive strengths
Table 4. Properties of the stone units and the mortar joints (N/mm2) (Senthivel and
Lourenco 2009)
Stone unit Mortar joint
E ν Fc Et E ν C ϕ Fc Ft
20,200 0.2 69.2 0 3.494 0.11 0.1 21.80° 3.0 0.032
of the mortar joint and the brick that are shown in Table 4. The ultimate lateral load that is
obtained from Equation 13 and Table 4 is
τU ¼ C þ σ 0 tanðφÞ ¼ 0.1 þ 0.5 × tanð21.8Þ ¼ 0.3MPa

$ % $ %
2τU 2 × 0.3
θ ¼ 0.5 × tan−1 ¼ 0.5 × tan−1 ¼ 25.097°
σ0 0.5
#
L × tanðθÞ ¼ 1;000 × tanð25.097Þ ¼ 468.37 mm
xmin ¼ min ⇒ xmin ¼ 468.37mm
ð1 − tanðθÞÞ × h ¼ ð1 − tanð25.097ÞÞ × 1;200 ¼ 637.95mm
PU ¼ τU × L × t þ F t × t × xmin ¼ 0.3 × 1;000 × 200 þ 0.032 × 200 × 468.37 ¼ 62;997:57 N
h 1;200
¼ ¼ 1.2 ⇒ α ¼ 0.6384 ∶is obtained f rom Fig : 9
L 1;000
P ¼ 0.88α PU ¼ 0.88 × 0.6384 × 62;997:57 ¼ 35;391:5N ¼ 35.4 kN
To show the ability of Equation 13 for calculating the ultimate lateral load, the example is
also analyzed using the FEMA 307 (ATC 1999) guideline, as shown below.
0.75 0.75
vme ¼ ðC þ f a tanφÞ ¼ ð0.1 þ 0.5 × tanð21.8ÞÞ ¼ 0.15
1.5 1.5
An ¼ 1;000 × 200 ¼ 200;000 mm2
PCE ¼ f a An ¼ 0.5 × 200;000 ¼ 100;000N
L 1;000
¼ ¼ 0.833 ⇒ β ¼ 0.833
h ef f 1;200
V bjs ¼ An × vme ¼ ð200;000Þ × 0.15 × 10−3 ¼ 30kN
$ % $ %
L 1;000
× 10−3 ¼ 37.5kN
V r ¼ 0.9αPCE ¼ 0.9 × 0.5 × ð100;000Þ ×

h ef f 1;200
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
$ % sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
$ %
fa 0.5
V dt ¼ vme An β 1þ ¼ 0.15 × 200;000 × 0.833 × 1þ × 10−3 ¼ 52.02kN
vme 0.15
$ %$ % $ %
L fa 0.5
V tc ¼ αPCE 1− ¼ 0.5 × ð100;000Þ × 0.833 × 1 − × 10−3
h ef f 0.7 × f me0
0.7 × 3
¼ 31.73kN
Referring to FEMA 307 (ATC 1999), if V r or V tc are the lowest value and f a < 100 psi
(0.7 MPa), then the predicted mode is wall-pier rocking with V r as the capacity. When V r ,
V tc , V dt , and V bjs have the lowest predicted capacity, the associated behavior modes are wall-
pier rocking, preemptive toe crushing, preemptive diagonal tension, and bed-joint sliding,
respectively. Thus, the governing behavior mode for the ancient stone-masonry shear
wall that was determined by FEMA 307 is the bed-joint sliding mode with a capacity of
30 kN. The capacity obtained in this study is approximately 35.4 kN. It is important to
note that the proposed method cannot predict the behavior mode. However, the proposed
model can accurately predict the ultimate lateral shear force. The wall is also analyzed
18 A. H. AKHAVEISSY
Figure 13. Comparison of the test data and the predicted load-displacement curves.
using micro-modeling in the finite element process. For the numerical analysis, the stone-
masonry units are represented by plane stress continuum elements (eight-noded), whereas the
joints are represented by line interface elements (six-noded). Each stone-masonry unit is
modeled with 2 × 1 elements. The wall is modeled by 48 eight-noded isoparametric elements,
60 interface elements, and 1,024 degrees of freedom, and an error of 1e-4 is used for the
displacement convergence criterion. Figure 13 compares the predicted load-displacement
curve that was obtained from the micro-modeling analysis (Akhaveissy 2011), the analytical
solution that was obtained in this study, and the test data. As shown in the figure, the pre-
dicted results and the test data are very similar.
The experimental ultimate lateral load from Figure 13 is 38.7 kN. The estimated lateral
loads based on Equation 13 and FEMA 307 are 35.4 kN and 30 kN, respectively. Therefore,
the errors are 8.5% and 22.5% for the proposed formula and FEMA 307, respectively. These
results confirm that Equation 13 is appropriate for estimating the lateral load of the wall.
Using Equation 13, various unreinforced masonry walls are evaluated in Table 5. The
ultimate load that was predicted by Equation 13 is compared with those obtained by ATC-43
and FEMA 273 (ATC 1999). In Table 5, f m is the compressive strength of the masonry prism,
and β is a material parameter to define the tensile strength. The walls in Table 5 are 250 mm
thick. In Table 5, f a is the initial vertical stress on the top of the wall and is the same as σ 0 .
The fourth column in Table 5 shows the ultimate shear strength in terms of the cohesion
strength, the internal friction coefficient, and the initial vertical stress on the top of the
wall. These ultimate shear strengths were accepted by FEMA 273 and ATC-43 (ATC
1999). In Table 5, θ and xmin in columns 5 and 6 are calculated by Equations 11 and 12,
while α is determined by column 1 in Table 5 and the average curve in Figure 9. In column
8, the ultimate lateral strength force of each wall mentioned in column 1 of Table 5 is cal-
culated using Equation 13. The obtained experimental lateral force of each wall is shown in
column 9, and the ultimate lateral forces that were obtained by ATC-43 and FEMA 273 (ATC
1999) are presented in column 11. The errors associated with the ultimate lateral forces that
Table 5. Comparison between this study, FEMA 273, and ATC-43 (ATC 1999)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
h fa xmin P by Eq. Test Error by FEMA 273, ATC-43

(m)
Ft ¼ β × f m
L (MPa) (MPa) τu (MPa) θ (mm) α 13 (kN) (kN) Eq. 10 (%) (kN), error (%)
3
1.5 0.1 × 7.9 1.245 0.813f a 29.20 838 0.408 195.8 185 5.8% 275, 48.6%
2
1.5 0.1 × 7.9 0.67 .206 + 0.813f a 32.97 702 0.593 219.3 227 −3.4% 213, −6.2%
1.35 0.1 × 6.2 0.63 .23 + 0.57f a 30.93 541 .586 117.7 114 3.2% 74, −35%
1.0
2.0
LIMIT STATE STRENGTH OF UNREINFORCED MASONRY STRUCTURES
1.0 0.1 × 6.2 1.11 .23 + 0.57f a 28.62 545 0.408 107.8 108 −0.1% 108, 0.0%
6 ft
11.42 f t 0.1 × 1,740 psi 93 psi 0.0 + f a 31.71 2.3 ft 0.948 146.3 kips 157 kips −6.8% 177 kips, 12.7%
6 ft
9.5 f t 0.1 × 1,740 141 psi 0.0 + f a 31.71 2.3 ft 0.894 164.2 kips 164 kips 0.1% 186 kips, 13.4%
psi
19
20 A. H. AKHAVEISSY
Table 6. Comparison between the ultimate lateral load by the proposed method and those
of previous studies (Magenes and Calvi 1997, Chen et al. 2008)
P by Error by Error by
h t Ft ¼ β % f m f a τu xmin Eq. 13 Test Eq. 13 references
(m)
L (mm) (MPa) (MPa) (MPa) θ (mm) α (kN) (kN) (%) (%)
2 250 0.07 0.6 .21 + 0.81fa 33.34 658 0.408 66.72 72 −7.3 −42.4
1
1.35
250 0.07 0.6 .21 + 0.81fa 33.34 462 0.586 93.9 85 10.5 −26
1
3
380 0.07 1.24 .14 + 0.55fa 26.49 747 0.408 175.7 185 −5 −33.6
1.5
2
380 0.07 0.68 .21 + 0.81fa 32.96 703 0.594 236 227 4 −48.6
1.5
were estimated in this study compared with the experimental data are shown in column 10.
Column 11 shows the error associated with the ultimate lateral forces that were obtained by
ATC-43 and FEMA 273 (ATC 1999) compared with the experimental data. Comparing
columns 10 and 11 shows that this study’s error is less than those of FEMA 273 and
ATC-43 (ATC 1999).
To assess the efficiency of the proposed formula, different experimental results are
shown in Table 6. The columns in Table 6 are calculated in the same manner as those
in Table 5.
DIFFERENT UNREINFORCED MASONRY WALLS

Formica et al. (2002) proposed a strategy for the nonlinear analysis of brick masonry
walls. Thus, the ability of the proposed method to nonlinearly analyze brick masonry
walls is assessed by comparing the results that are obtained by this method with those
obtained by the previous study’s strategy (Formica et al. 2002). The same parameters are
used in this study as those used in Formica et al. (2002). Figure 14 shows the boundary
conditions and the geometries of the walls.
The walls shown in Figure 14 are 120 mm thick. The friction angle, the cohesion, and the
tensile strength used to analyze the walls shown in Figure 14 are 30 degrees, 0.23 MPa, and
0.4 MPa, respectively (Formica et al. 2002). The ultimate lateral loads of the various walls are
calculated using Equation 13 and shown in Table 7.
In Table 7, the ultimate lateral load of the walls shown in Figures 14b and 14c are deter-
mined by the shear strength of both of the piers. Both of the piers in Figures 14b and 14c
tolerate the lateral force. To show the ability of Equation 13, the example is also analyzed
using FEMA 307 (ATC 1999), as shown in Table 8.
By comparing Tables 7 and 8, the predicted results obtained by the proposed formulation
are more accurate than those predicted by FEMA 307. Figure 15 compares the closed-form
solution of this study, the results of FEMA 307, and the results of Formica et al. (2002) for the
Figure 14. Boundary conditions and the geometries of the walls (mm) (Formica et al. 2002).
Table 7. The ultimate lateral loads obtained by the proposed method
P (Formica
Wall h fa τu xmin P by Eq. et al. 2002) Error by
figure L (MPa) (MPa) θ (mm) α 13 (kN) (kN) Eq. 13 (%)
Figure 14a 1;040 0.5 .518 32.13 386.8 0.5869 34.3 36.0 −4.7
770
1;250
Figure 14b 0.15 .317 38.34 261.4 0.3544 2 × 10.02 = 20.04 20.7 −3.2
515
1;320
Figure 14c 770 0.15 .317 38.34 276 0.4839 2 × 18.1 = 36.2 36.2 0.0
different walls shown in Figure 14. The ultimate lateral strengths of the walls shown in Fig-
ures 14a and 14c are similar; thus, to illustrate the strengths more clearly, the results for both
of the walls are shown in Figures 15a and 15b, respectively. The closed-form solution of this
study is more similar to the results of Formica et al. (2002) shown in Figure 15 than the results
predicted by FEMA 307.
22 A. H. AKHAVEISSY
Table 8. The ultimate lateral load obtained by the formulation provided by FEMA 307
Shear force
Wall h fa vme V bjs V r V tc V dt Shear (Formica et al. Error
figure L (MPa) (MPa) (kN) (kN) (kN) (kN) force (kN) 2002) (kN) (%)
1;040 0.5 0.2593 23.96 30.80 28.10 30.40 30.8 36.0 −14.4
Figure 14a
770
1;250
Figure 14b 0.15 0.1583 9.78 3.44 3.62 9.15 2 × 3.44 = 6.88 20.7 −66.8
515
1;320 0.15 0.1583 14.72 7.36 7.75 13.76 2 × 7.36 = 14.72 36.2 −59.3
Figure 14c
770
A SINGLE-STORY UNREINFORCED MASONRY BUILDING

The model was validated using a full-scale single-story unreinforced masonry building
that was tested in the laboratory by Paquette and Bruneau (2003, 2004, 2006). Figure 16
shows the western wall of the tested model. The parapets of the western and the eastern
walls were 254 mm tall (Paquette and Bruneau 2003, 2004, 2006).
The compressive strengths of the brick and the mortar were 109 and 9.24 MPa, respectively,
and the compressive and the tensile strengths of the masonry were 22.2 and 0.18 MPa,
respectively (Paquette and Bruneau 2003). The numerical analysis of this study uses these
strengths for the western wall. The elasticity modulus of the masonry specimen was assumed
to be 850 times the compressive strength of the masonry specimen (Paquette and Bruneau 2003,
2004, 2006, CSA 2004). The thickness of the wall was 190 mm. The gravity load, 2.4 kN/m2,
was applied to the diaphragm, which had dimensions of 4,091 mm × 5,610 mm. Ten wood joists-
were applied to the diaphragm to transmit the gravity load to the western and eastern walls. The
net span of the wood joist was 5,310 mm (Paquette and Bruneau 2003, 2004, 2006). Therefore,
the gravity load on each wall was 6.37 kN/m.
The Drucker-Prager criterion is used to determine the cohesion strength and the friction
angle of the mortar joints. As shown in Figure 6, the failure of the mortar joint typically
occurs in tension. Thus, the tensile strength of the mortar joints (0.18 MPa) is used to cal-
culate the cohesion strength and the friction angle. Therefore, these parameters are deter-
mined by analyzing an infinitesimal element under shear stress by the Drucker-Prager
criterion. If only a shear stress was applied to the element, the principal stress in the principal
stresses space of solid mechanics would be the same as the shear stress. Therefore, element-
level loading is only a cyclically loaded shear stress. Figure 17 shows the shear stress–shear
strain curve of the element under cyclic loading for the cohesion strength and the friction
angle of 0.078 MPa and 31.9 degrees, respectively.
Figure 17 shows the ultimate shear stress for the element level. There is a maximum
principal normal stress of 0.18 MPa for the cohesion strength and the friction angle of
0.078 MPa and 31.9 degrees, respectively. Thus, these parameters are used in Equation 13
to determine the ultimate base shear of the building. First, the ultimate lateral load of the wall
is determined based on the resistance of the piers. Thus, it is assumed that the plastic hinges in
an equivalent frame have only been formed in the piers. Second, the piers and the spandrels
Figure 15. (a) Comparisons of the results for the walls shown in Figures 14a and 14b and (b) the
results for the wall shown in Figure 14c.
are considered using the lateral force calculated in step 1 using the gravity load. Third, the
calculated internal shear force based on the analysis of the equivalent frame in the second step
is compared with the shear strengths of the piers and the spandrels. If the internal shear force
of the spandrels exceeds the shear strength of the spandrels, the assumption applied in the first
step is incorrect, and the ultimate lateral load should be determined using the other plastic
hinge formations (i.e., plastic hinges in the spandrels or plastic hinges in both the piers and
the spandrels). However, if the internal shear force of the spandrels is less than their shear
strength, the assumption applied in the first step is correct and the shear force that was cal-
culated in the first step is the resistance base shear force for the wall.
24 A. H. AKHAVEISSY
Figure 16. Dimensions of the western wall (mm) (Paquette and Bruneau 2003).
Figure 17. Shear stress–shear strain curve by the Drucker-Prager criterion.

First step:
The shear strength for the piers is determined with the resistance piers (see Figure 18).
The shear strengths for all of the piers are calculated in Table 9 for the structure shown in
Figure 18. An initial vertical stress of 0.0335 MPa is applied on the top of the western wall,
which is obtained by dividing the uniform load to the thickness of the wall. The ultimate
shear stress is calculated based on Equation 13 as
τu ¼ C þ σ 0 tanðϕÞ
$ % þ 0.0335 × tanð31.9Þ ¼ 0.0988MPa
¼ 0.078
2 × 0.0988
θ ¼ 0.5tan−1 ¼ 40.189
0.0335
Figure 18. Resistance piers used to determine the ultimate shear force.
Table 9. Shear strength predicted by Equation 13
h
Name of pier Ft (mm) θ xmin (mm) α P by Eq. 13 (kN)
L
953
Window 0.18 40.18 148 0.497 6.96
578
2;470
Central 0.18 40.18 219 0.472 14.10
1;410
2;470
Door 0.18 40.18 219 0.307 6.49
883
Σ 27.55
To compare this study with the FEMA guidelines, Table 10 shows the ultimate lateral
shear strength of the wall that is estimated by FEMA 307. The initial vertical stress, f a , and
the expected masonry friction strength of mortar, vme , for the wall are 0.0335 MPa and
0.0494 MPa, respectively.
As shown in these tables, the shear strengths of the western wall that are predicted by the
proposed formulation and are provided in FEMA 307 are 27.55 kN and 8.43 kN, respectively.
The base shear strength of the western wall was also estimated by Akhaveissy and Desai
(2011) and Akhaveissy (2011) to be 26.57 kN and 25.73 kN, respectively.
Second step:
In this step, the wall is analyzed using the lateral load that was calculated in step 1
(27.55 kN). Figure 19 shows the internal forces on the equivalent frame due to the equivalent
static force of the earthquake. The internal forces are determined by a portal frame assump-
tion analysis. An initial vertical stress of 0.0335 MPa is applied on the top of the western wall
by dividing the uniform load by the thickness of the wall. Therefore, the vertical stress on
each pier is 0.0335 MPa. With the free span of door and the window spandrels (610 mm), a
26 A. H. AKHAVEISSY
Table 10. The ultimate lateral load estimated by the formulation given in FEMA 307
L
Pier V bjs (kN) V r (kN) V tc (kN) V dt (kN) Shear force (kN)
h
578
Window 5.43 2.01 4.71 2.22 2.01
953
1;410 5.11
Central 13.24 4.61 11.49 4.61
2;470
Door 883 8.29 1.81 5.47 2.01 1.81
2;470
Σ 8.43
shear force of 1.94 is applied on the cross section of the spandrels due to the initial vertical
stress. Thus, the shear force due to the initial vertical stress is combined with the shear force
on the cross section of spandrels due to the equivalent static force of the earthquake (see
Figure 19). Accordingly, the total shear force is 12.06 kN for the window and the door
spandrels.
In Figure 19, N and V represent the internal axial and the internal shear forces due to the
lateral force of 27.55 kN. The subscripts of N and V denote that the forces are due to the
earthquake. By comparing the shear force of the piers due to the lateral load shown in Fig-
ure 19 and the shear strength of the piers shown in Table 9, the internal shear force of the piers
is shown to be similar to the shear strength of the piers. Thus, the piers have collapsed, and
plastic hinges have formed in the piers. Next, the collapse of the spandrels is considered.
Figure 19 shows that the axial force of the door spandrel is less than the axial force of window
spandrel. Therefore, the door spandrel is more critical than the window spandrel. The shear
strength of the door spandrel is calculated below according to the internal axial force shown
in Figure 19.
7200
σ0 ¼ ¼ 0.043MPa
882 × 190
τU ¼ C þ σ 0 tanðφÞ ¼ 0.078 þ 0.043 × tanð31.9Þ ¼ 0.1047MPa
$ % $ %
−1 2τU −1 2 × 0.1047
θ ¼ 0.5 × tan ¼ 0.5 × tan ¼ 39.20°
σ0 0.043
#
xmin ¼ min L × tanðθÞ ¼ 882 × tanð39.20Þ ¼ 719.4mm
⇒ xmin ¼ 112.5mm
ð1 − tanðθÞÞ × h ¼ ð1 − tanð39.20ÞÞ × 610 ¼ 112.5mm

PU ¼ τU × L × t þ F t × t × xmin ¼ 0.1047 × 882 × 190 þ 0.18 × 190 × 112.5 ¼ 21;393:1N
h 610
¼ ¼ 0.692 ⇒ α ¼ 0.8637 ∶is obtained f rom Fig: 9
L 882
P ¼ 0.88α PU ¼ 0.88 × 0.8637 × 21393.1 ¼ 16260N ¼ 16.26kN
Figure 19. The internal forces due to the equivalent static force of the earthquake in the equiva-
lent frame of the eastern wall.
The lateral force and the gravity load applied a total shear force of 12.06 kN on both of the
spandrels. Because the total shear force is only 12.06 kN and the shear strength of the span-
drels is 16.26 kN, the spandrels can tolerate the total shear force. Thus, the plastic hinges are
not formed in the spandrels. Therefore, the assumptions applied in step 1 are correct, and the
lateral resistance force of the wall can be calculated from the shear strength of the piers.
Consequently, the resistance base shear force of the wall is 27.55 kN (see Table 9). The
limit loads obtained using FEMA 307 and the closed-form solution obtained from the pro-
posed formulation correspond well with the test data (Paquette and Bruneau 2003, 2004,
2006). The failure pattern between the experimental data and the result predicted by the pro-
posed formulation are compared in Figure 20. It is important to note that there is a constant
Figure 20. Comparison of the ultimate lateral forces obtained from both the test data and the
closed-form solution for the wall shown in Figure 19.
28 A. H. AKHAVEISSY
Figure 21. Comparison of the failure patterns obtained from the closed-form solution and the
experimental data (Paquette and Bruneau 2003, 2004, 2006).
shear force applied to each pier. Thus, shear forces of the same magnitude but opposing
directions are applied to both ends of each pier. Therefore, when the shear force is close
to the shear strength, both the top and the bottom of the pier collapse (see Figure 21b).
The observed failure and the failure that was predicted by the proposed formulation are
shown in Figure 21; as shown in the figure, the failure only occurs in the piers. Thus,
the base shear force and the failure pattern that were predicted by the closed-form solution
are similar to those of the observed data.
The predicted shear strength of the western wall has an error of approximately 4.3%
compared with the test data (Paquette and Bruneau 2003, 2004, 2006). However, a shear
strength of approximately 8.43 kN is predicted using FEMA 307, corresponding to an
error of approximately 68%. Therefore, the proposed model (see Equation 13) can be
used to accurately analyze practical masonry structures similar to those considered in this
study. Akhaveissy and Desai (2011) also analyzed this wall using a new model in the finite
element method. The model, referred to as DSC/HISS-CT, was based on the Disturbed State
Concept (DSC) with a modified hierarchical single yield surface (HISS) plasticity. The model
uses two HISS yield surfaces for the compressive and tensile behaviors. The model para-
meters used to analyze masonry walls should be calibrated by the compressive stress-strain
curve of the masonry unit. The model uses 18 parameters to analyze the masonry wall. There-
fore, it is difficult to analyze masonry walls using this model. Furthermore, the mesh for the
analysis of the western wall contained 720 eight-node isoparametric elements, 2,527 nodes,
and 4,926 degrees of freedom. Therefore, the DSC/HISS-CT analysis has a longer duration
than that of the closed-form solution proposed in this study. Thus, the proposed model may
be preferable when predicting the ultimate lateral force of unreinforced masonry structures.
CONCLUSION
In this study, an interface model for modeling the mechanical response of mortar joints in
masonry walls was numerically implemented. The interface laws were formulated in the
elasto-plasticity framework for nonstandard materials with softening in the mortar joints
due to the applied shear and tensile stresses. The theoretical framework of this study was
entirely based on the plasticity theory. The finite element formulation was based on
eight-noded isoparametric quadrilateral elements and six-noded contact elements. The von
Mises criterion was used to simulate the behavior of the units. Interface laws were formulated
for the contact elements to simulate the softening behavior of the mortar joints under tensile
stress. A normal linear cap model was also used to limit the compressive stress. The cap-
abilities of the interface model and the effectiveness of the computational procedure were
investigated using numerical examples to simulate the response of a masonry wall tested
under shear in the presence of an initial pre-compression load. The experimental results pro-
vided in the literature were compared with the numerical analysis results. The computer pre-
dictions correspond well with the test data. The predicted ultimate load of the masonry wall is
approximately 93.25% of the ultimate load obtained from the test data. The proposed model
was also used to simulate a stone-masonry shear wall. The predicted load-displacement curve
corresponded with the observed data. In addition, a closed-form solution is proposed based
on results of a finite element procedure. The closed-form solution is more accurate than the
ATC and FEMA 307 in predicting the ultimate lateral load of unreinforced masonry walls.
Thus, the proposed closed-form solution can be used to accurately analyze masonry struc-
tures similar to those considered in this study. The proposed model can be used to predict the
base shear force of unreinforced masonry structures under earthquake acceleration in the
nonlinear matrix analysis method. In future research, the presented closed-form solution
can be developed for two-noded linear elements in a finite element framework to nonlinearly
analyze buildings similar to high-rise unreinforced masonry buildings and unreinforced
masonry arch bridges. It is important to note that the analysis of the high-rise masonry build-
ings using plane elements or shell elements in the finite element method requires extensive
computer memory and computing time.
REFERENCES
Akhaveissy, A. H., 2010. Numerical modelling of unreinforced masonry walls using contact ele-
ment, Paper 346, Proceedings of the Tenth International Conference on Computational Struc-
tures Technology, Civil-Comp Press, Stirlingshire, Scotland,
Akhaveissy, A. H., 2011. Lateral strength force of URM structures based on a constitutive model
for interface element, Latin American Journal of Solids and Structures 8, 445–461.
Akhaveissy, A. H., and Desai, C. S., 2011. Unreinforced masonry walls: Nonlinear finite element
analysis with a unified constitutive model, Archives of Computational Methods in Engineering
18, 485–502.
Alessandri, C., and Brebbia, C.A., 1987. Strength of masonry walls under static horizontal loads:
Boundary element analysis and experimental tests, Engineering Analysis 4, 118–134.
Applied Technology Council (ATC), 1999. FEMA 307: Evaluation of Earthquake Damaged
Concrete and Masonry Wall Buildings, Technical Resources, Federal Emergency Manage-
ment Agency, Washington, D.C.
Brasile, S., Casciaro, R., and Formica, G., 2010. Finite element formulation for nonlinear analysis
of masonry walls, Computers and Structures 88, 135–143.
Brencich, A., Gambarotta, L., and Lagomarsino, S., 1998. A macroelement approach to the three-
dimensional seismic analysis of masonry buildings, 11th European Conference on Earthquake
Engineering, Balkema, Rotterdam (ISBN: 9054109823).
Bruneau, M., 1994. State-of-the-art report on seismic performance of unreinforced masonry
buildings, Journal of Structural Engineering 120, 230–251.
30 A. H. AKHAVEISSY
Canadian Standard Association (CSA), 2004. Design of masonry structures, S304.1, Ontario,
Canada.
Cecchi, A., and Milani, G., 2008. A kinematic FE limit analysis model for thick English bond
masonry walls, International Journal of Solids and Structures 45, 1302–1331.
Chaimoon, K., and Attard, M. M., 2009. Experimental and numerical investigation of masonry
under three-point bending (in-plane), Engineering Structures 31, 103–112.
Chen, S. Y., Moon, F. L., and Yi, T., 2008. A macro element for the nonlinear analysis of in-plane
unreinforced masonry piers, Engineering Structures 30, 2242–2252.
Formica, G., Sansalone, V., and Casciaro, R., 2002. A mixed solution strategy for the nonlinear
analysis of brick masonry walls, Comput. Methods Appl. Mech. Engrg. 191, 5847–5876.
Gabor, A., Ferrier, E., and Jacquelin, H. P., 2006. Analysis and modeling of the in-plane shear
behavior of hollow brick masonry panels, Construction and Building Materials 20, 308–321.
Giambanco, G., Rizzo, S., and Spallino, R., 2001. Numerical analysis of masonry structures via
interface models. Comput. Methods Appl. Engrg. 190, 6493–6511.
Jirasek, M., and Bazant, Z. P., 2002. Inelastic Analysis of Structures, John Wiley & Sons, Ltd,
England.
Kaushik, H. B., Rai, D. C., and Jain, S. K., 2007a. Stress-strain characteristics of clay
brick masonry under uniaxial compression, Journal of Materials in Civil Engineering
ASCE 19(9), 728–739.
Kaushik, H. B., Rai, D.C., and Jain, S. K., 2007b. Uniaxial compressive stress-strain model for
clay brick masonry, Current Science 92, 497–501.
Kawa, M., Pietruszczak, S., and Shieh-Beygi, B., 2008. Limit states for brick masonry based on
homogenization approach, International Journal of Solids and Structures 45, 998–1016.
Lopez, J., Oller, S., Onate, E., and Lubliner, J., 1999. A homogeneous constitutive model for
masonry, Int. J. Numer. Meth. Engng. 46, 1651–1671.
Lourenço, P. B., 1994. Analysis of Masonry Structures with Interface Elements: Theory and
Applications, Report 03-21-22-0-01, Delft University of Technology, Delft, Netherlands.
Lourenço, P. B., 1996. Computational Strategies for Masonry Structures, Dissertation, Delft
University.
Magenes, G., and Calvi, G. M., 1997. In-plane seismic response of brick masonry walls,
Milani, G., 2011a. Simple homogenization model for the non-linear analysis of in-plane loaded
masonry walls, Computers and Structures 89, 1586–1601.
Milani, G., 2011b. Simple lower bound limit analysis homogenization model for in-and out-
of-plane loaded masonry walls, Construction and Building Materials 25, 4426–4443.
Milani, G., Beyer, K., and Dazio, A., 2009. Upper bound limit analysis of meso-mechanical
spandrel models for the pushover analysis of 2D masonry frames, Engineering Structures
31, 2696–2710.
Milani, G., Lourenço, P., and Tralli, A., 2006. Homogenised limit analysis of masonry walls, Part
I: Failure surfaces, Computers and Structures 84, 166–180.
Milani, G., Lourenço, P., and Tralli, A., 2007. 3D homogenized limit analysis of masonry
buildings under horizontal loads, Engineering Structures 29, 3134–3148.
Milani, G., Miani, E., and Tralli, A., 2010. Approximate limit analysis of full scale FRP-rein-
forced masonry buildings through a 3D homogenized FE package, Composite Structures 92,
918–935.
Paquette, J., and Bruneau, M., 2003. Pseudo-dynamic testing of unreinforced masonry building
with flexible diaphragm, Journal of Structural Engineering, ASCE, 129, 708–716.
with flexible diaphragm, Paper 2609 1, 3th World Conference on Earthquake Engineering
Vancouver, B.C., Canada.
with flexible diaphragm and comparison with existing procedures, Construction and Building
Materials 20, 220–228.
Park, J., Towashiraporn, P., Craig, J. I., and Goodno, B. J., 2009. Seismic fragility analysis of
low-rise unreinforced masonry structures, Engineering Structures 31, 125–137.
Pasticier, L., Amadio, C., and Fragiacomo, M., 2008. Non-linear seismic analysis and
vulnerability evaluation of a masonry building by means of the SAP2000 V.10 code,
Roca, P., 2006. Assessment of masonry shear-walls by simple equilibrium models, Construction
and Building Materials 20, 229–238.
Rota, M., Penna, A., and Magenes, G., 2010. A methodology for deriving analytical fragility
curves for masonry buildings based on stochastic nonlinear analyses, Engineering Structures
32, 1312–1323.
Sekiguchi, K., Rowe, K. R., Yee Lo, K., and Ogawa, T., 1990. Time step selection for 6-noded
nonlinear joint element in elasto-viscoplasticity analysis, Computer and Geotechnics 10,
33–58.
Senthivel, R., and Lourenço, P. B., 2009. Finite element modelling of deformation characteristics
of historical stone masonry shear walls, Engineering Structures 31, 1930–1943.
Shariq, M., Abbas, H., Irtaza, H., and Qamaruddin, M., 2008. Influence of openings on seismic
performance of masonry building walls, Building and Environment 43, 1232–1240.
Shieh-Beygi, B., and Pietruszczak, S., 2008. Numerical analysis of structural masonry: Mesoscale
approach, Computers and Structures 86, 1958–1973.
Tena-Colunga, A., Juarez-Angeles, A., and Salinas-Vallejo, V. H., 2009. Cyclic behavior of com-
bined and confined masonry walls, Engineering Structures 31, 240–259.
Vasconcelos, G., and Lourenço, P. B., 2009. Experimental characterization of stone masonry in
shear and compression, Construction and Building Materials 23, 3337–3345.
Vermeltfoort, A. T., Raijmakers, T. M. J., and Janssen, H. J. M., 1993. Shear test on masonry
walls, Proc. 6th North Am. Masonry Conf, Drexel University, Philadelphia, 1183–1193.
(Received 8 August 2011; accepted 9 February 2012)
The Use of Stochastic Optimization
in Ground Motion Prediction
Margaret Segoua) M.EERI, and Nicholas Voulgarisb) M.EERI
In recent years stochastic optimization has become popular in modeling highly

complex natural systems as opposed to traditional approaches, such as regression
analysis. This paper tests the efficiency of regression analysis over mathematical
optimization by evaluating the performance of stochastic (genetic algorithms,
simulated annealing) and deterministic algorithms (nonlinear least squares, pattern
search) in solving minimization problems. Qualitative and quantitative comparison
of these solvers reveals that the implementation of genetic algorithms with initial
population development, using Latin Hypercube sampling, outperformed other
algorithms. Using the aforementioned solver, ground motion prediction equations
are derived for the estimation of horizontal peak ground acceleration, velocity, and
5% damped spectral acceleration ordinates at discrete period estimators between
0.05 s and 2 s. Ground motion for stiff and soft soil sites is amplified by a constant
factor with respect to rock site. These strong motion prediction equations are pro-
posed for application in the range M4.5–M6.6 and for distances up to 150 km.
[DOI: 10.1193/1.4000098]
INTRODUCTION
The contribution of optimization to geophysics has been growing over the last decades
due to its efficiency in modeling complex natural systems by determining the best solution
from a set of alternative solutions (Goldberg 1989). However, the main advantage of opti-
mization is its ability to reach the optimal solution for any system even under extreme com-
putational environments, when data sets are limited as well as when vast data sets of high
diversity require the determination of a unique solution describing sufficiently the sam-
ples given.
In the first part of this paper the efficiency of traditional regression analysis to model
strong ground motion is tested against mathematical optimization. Furthermore the perfor-
mance of stochastic (genetic algorithms, simulated annealing) and deterministic solvers
(nonlinear least squares, pattern search) for a specific type of optimization problem,
known as minimization, is evaluated. The second part is related to an application example
aiming to present the adequate performance of the optimum solver, determined previously,
even when a rather modest data set is used to derive ground motion prediction equations. The
problems of strong motion data from the Euro-Mediterranean countries have been discussed
a)
Department of Geophysics-Geothermics, Faculty of Geology and Geoenvironment, Panepistimioupoli Zografou,
Athens, Greece 15784. Current Address: 325 Middlefield Road, Earthquake Science Center, USGS, Menlo Park,
CA, USA, 94025
b)
Department of Geophysics-Geothermics, Faculty of Geology and Geoenvironment, Panepistimioupoli Zografou,
Athens, Greece 15784
283
284 M. SEGOU AND N. VOULGARIS
by several researchers in the past (Akkar and Bommer 2007, Danciu and Tselentis 2007,
Skarlatoudis et al. 2004), identifying it as a case of limited resources.
In cases where strong motion data are scarce, the contribution of optimization in deter-
mining ground motion prediction equations could prove important (Tavakoli and Pezeshk
2005, Graizer and Kalkan 2009). On the other hand, in countries where dense seismological
arrays provide hundreds of strong motion records (Kao et al. 2010) the determination of the
optimal solution for the whole data set represents an important goal for seismologists and
engineers.
DATA SOURCES AND PROCESSING
The strong motion data set presented in this section serves two purposes: (1) to provide a
subset of strong motion data used for testing purposes among different solvers and (2) to
derive ground motion prediction equations for our application example, after the determina-
tion of the optimum solver.
Unprocessed strong motion records have been provided by the web site of the European
Strong Motion Database (Ambraseys et al. 2002), the Geodynamic Institute of the National
Observatory of Athens, and ITSAK institute in Thessaloniki.
The database described in Segou et al. (2008) includes strong motion data from Greece,
Italy, Turkey, and Iran. However, the poor distribution for M > 6.6 lead to limit the original
data set up to this magnitude range (see Table 1). To reduce the epistemic uncertainty of the
equations we adopted redetermined seismic parameters (Moss 2009), provided by the web site
of the International Seismological Centre (http://www.isc.ac.uk/). For the conversion from
body wave to moment magnitude the relationship of Scordilis (2006) was used. Available
strong motion records are limited to free-field and basement-level records of up to two-
story buildings. Strong motion records have been processed by Proschema software
(Segou and Voulgaris 2010) following a subtle processing scheme, described as follows.
PROCESSING STRONG MOTION RECORDS

A common sampling rate of 200 samples per second has been introduced for all record-
ings, and a quality control procedure described in Douglas (2003) has been used. Analog
records have undergone instrument adjustment in order to compensate for the instrument’s
deficiency to record ground motion near its natural frequency, whereas for digital records no
instrument adjustment was needed since they have flat instrument response for a wide period
range (Wang et al. 2003). For digital records the calculated mean was determined from the
pre-event segment and removed, whereas for analog records the entire record length was used
for this purpose. An infinite impulse response (IIR) filter of Butterworth type with design
specifications corresponding to fourth order band-pass filter, phase-preserving implementation
and optimal low cutoff frequencies, depending on single component requirements, was applied
Table 1. Coefficients of the ground motion model determined by regression analysis
a b c d e h
PGA 1.63778 −0.55005 0.15239 1.29326 −0.40947 0.85108

THE USE OF STOCHASTIC OPTIMIZATION IN GROUND MOTION PREDICTION 285
for removing long-period and high-frequency noise. Horizontal components have been com-
bined into their orientation-independent geometrical mean GMRotI50 (Boore et al. 2006).
Spectral ordinates further used for developing ground motion prediction equations fall inside
the usable data bandwidth of each record, calculated as being greater and smaller, by a factor of
1.25, than the high and low cutoff frequencies of the band-pass filter, respectively.
GROUND MOTION PREDICTION MODEL

The functional form of our ground motion prediction equation used for both parts of this
study, the testing phase and the application example case, is:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
log 10 ðYÞ ¼ a þ bM þ cM 2 þ ðd þ eMÞlog 10 Repi 2 þ ð H − hÞ2 þ f 1 RS þ f 2 SS
þ e1 ST s þ e2 SFs þ ålog 10 ðYÞ ð1Þ
Y ¼ PGA; PGV; SAðTÞ

The dependent variables of the mathematical model include peak ground acceleration
(PGA-cm/s2) and velocity (PGV-cm/s), combined in the geometrical mean of the horizontal
components. Pseudo-spectral acceleration is expressed in terms of GMRotI50 (PSA-cm/s2)
(Boore et al. 2006) for 5% of critical damping and discrete period estimators ranging from
0.05 s to 2 s.
Independent variables are considered to be the magnitude M, epicentral distance Repi
(km), and depth H (km). Seismic magnitude M corresponds to moment magnitude,
which lies between 4.5 and 6.6. Equation 1 includes a quadratic-magnitude term to account
for ground motion saturation for large earthquakes ðM ≥ 6.5Þ at short distances, a practice
followed by many researchers in the past (Bolt and Abrahamson 2003, among others). Mag-
nitude-dependent distance slope, expressed through the geometrical spreading term
ðd þ eMÞ, accounts for varying decay rates of ground motion with respect to seismic mag-
nitude (Anderson 2000). Random variables were introduced in Equation 1 for modeling soil
site conditions ðe1 ; e2 Þ and style of faulting ðf 1 ; f 2 Þ. The terms SS and RS and STs and SFs are
dummy variables for strike-slip and reverse style of faulting and stiff and soft soil site class,
respectively. In the case of SS ¼ RS ¼ 0 and STs ¼ SFs ¼ 0 , the estimated prediction cor-
responds to normal style of faulting at a rock site.
DISTANCE METRIC
The distance metric in Equation 1 corresponds to an adjusted hypocentral distance
derived by applying a correction factor h to the focal depth parameter H. In cases where
there is no evidence for constant focal depth, as in the database used, the exclusion of
the depth parameter leads to higher sigma estimation. Scherbaum et al. (2006) suggest
the use of the hypocentral distance for the stochastic simulation of strong ground motion
because it provides the lowest-misfit stochastic models for most empirical equations. In addi-
tion, Akkar and Bommer (2010) state the importance of using the focal depth as a parameter
in ground motion prediction equations since Joyner-Boore distance only accounts for the
horizontal distance measured at the surface. The use of other distance metrics, frequently
considered in seismic hazard assessment, such as the Joyner-Boore distance (Joyner and
Boore 1981) and closest distance to rupture (Idriss 1995), was rejected on the grounds
that kinematic modeling of the seismic rupture is not available for the majority of earthquakes
included in the database, making the determination of these distance metrics difficult.
The alternative method, that of deducing Rjb and Rrup from Repi using an empirical relation
(Scherbaum et al. 2004), increases the standard error of the prediction. The same skepticism
on the use of a distance metric is discussed in Danciu and Tselentis (2007).
LOCAL SITE CONDITIONS CLASSIFICATION SCHEME

Local site conditions classification follows U.S. National Earthquake Hazards Reduction
Program (NEHRP) scheme, which is based on the average shear-wave velocity in the upper
30 m of the site, known as V s30 . Boore et al. (1993) were the first to develop site categories
using the V s criterion, which has been further adopted by ground motion prediction relations
and building codes, such as the 1997 Uniform Building Code and the 2000 International
Building Code. Due to the scarcity of strong motion records in recording sites belonging
to site class A and E, these records have been unified with soil classes B and D, respectively.
The authors note that site characterization remains a critical issue for some Mediterranean
countries, where geophysical surveys at strong motion sites should be undertaken to improve
the quality of the V s30 parameter. Furthermore, an alternative site classification scheme
should be considered since the efficiency of the aforementioned parameter for site charac-
terization is questionable especially in cases of considerable sediment thickness.
METHODS
The purpose of this section is to compare traditional regression analysis and optimization
techniques using stochastic and deterministic algorithms, known as solvers, often used for the
mathematical model optimization (Scherbaum et al. 2006, Tavakoli and Pezeshk 2005). The
aim is to present the advantages and disadvantages of each case and select the optimum solver
for Equation 1 following an information theoretic approach. The selection of the appropriate
solver is directly linked with the representation of the problem (Rothlauf 2006). Here we
focus on the selection of the optimum solver for the function described in Equation 1.
The suggestion of the suitable type of mathematical function for modeling strong ground
motion is not within the scope of this paper.
REGRESSION ANALYSIS
Based on the maximum likelihood criterion, statistical analysis provides a systematic
technique for the empirical determination of a mathematical model, using observed values
and model predictors, known as regression analysis. However, the nonlinear form of Equa-
tion 1 poses limitations concerning its solution. Linearization of Equation 1, through a set of
linear equations, would not adequately represent the problem at hand due to the model’s
nonlinear character (Draper and Smith 1981).
The steepest decent method combined with Levenberg-Marquardt elements was consid-
ered for implementing regression analysis in this study. Regression analysis results following
the mixed effects technique (Lindstrom and Bates 1990) are presented in Table 1, where style
of faulting and soil classification have been considered as random effects. However, two
problems arise: First, it is not possible to constrain efficiently the arithmetic values of the
coefficients so that they remain consistent with their physical meaning (Akkar and Bommer
2007), and second, the inadequate distribution of data (see Figures 1–2) with respect to the
random variables leads to poor determination of the STs , SFs and SS, RS terms
(see Equation 1). The implementation of two-stage regression analysis was not possible
due to the number of single recordings, which in this case can lead to underestimation
of aleatory variability (Spudich et al. 1999).
Figure 1. Magnitude-distance distribution with respect to site conditions.
Figure 2. Magnitude-distance distribution with respect to style of faulting.

OPTIMIZATION
In theory, traditional regression analysis is expected to provide one possible solution, as a
set of coefficients, for any given equation. The challenge that optimization problems address
is the necessity for determining the best solution for data sets representing complex physical
systems. As described previously, the effort lies in determining the optimal solution for the
mathematical model of Equation 1, which requires the implementation of constrained opti-
mization techniques. The mathematical problem corresponds to the minimization of misfit
represented by the sum of squares of the residuals, between the logarithms of observed and
predicted values (Equation 2). Constrained minimization problems are established on some
basic pillars, which are: (1) the existence of a candidate theoretical solution to initiate opti-
mization; (2) the existence of an objective function to evaluate the minimization of the misfit;
(3) a set of linear constraints serving as bounds for the coefficients’ determination; and (4) the
determination of convergence criteria (Rothlauf 2006).
X
OF ¼
EQ-TARGET;temp:intralink-;e2;41;475 ð pred i − obsi Þ2 (2)
i
It should be noted that for consistency reasons the same objective function, linear con-
straints, and convergence criteria (see “Testing Optimization Solvers,” below) have been
used during the implementation of the aforementioned solvers.
Techniques used during optimization to exhaust the search space are classified in three
groups: (1) calculus-based techniques, (2) guided random search techniques, and (3) numera-
tive techniques (Filho et al. 1994). Calculus-based versus guided random search techniques
were tested through comparison of different solvers whereas enumerative algorithms were
discarded since “they cannot compete to the robustness race” when compared with the afore-
mentioned techniques mainly due to the characteristics of their search domains (Said 2005).
OPTIMIZATION USING CALCULUS-BASED TECHNIQUES

Calculus-based techniques are further divided into direct and indirect search methods
(Filho et al. 1994). Indirect search methods, such as nonlinear least squares, are the most
common approach in data fitting problems in earth sciences corresponding to the implemen-
tation of the maximum likelihood criterion (Dennis 1977, Draper and Smith 1987) for deter-
mining the optimum solution that sets the value of the objective function in Equation 2
to zero.
Pattern search, on the other hand, is a direct search method used broadly in its generalized
form in optimization of noncontinuous and nondifferentiable functions (Hookes and Jeeves
1961, Dolan et al. 2003). An initial population of possible solutions provides multiple starting
points, while during optimization the available search space is either increasing or decreasing,
depending on a gradient (Audet and Dennis 2002), in an effort to improve the solutions sug-
gested at a previous step. These solutions are then evaluated for their effectiveness to minimize
the misfit using an objective function (Equation 2). Although criticized for their efficacy to
search sufficiently large solution spaces, direct search methods are considered to be the sim-
plest variation of deterministic algorithms for optimization, (Goldberg 1989).
OPTIMIZATION USING GUIDED RANDOM SEARCH TECHNIQUES

Guided random search techniques are classified in both genetic algorithms and simulated
annealing (Filho et al. 1994). Both algorithms use information in order to guide their search for
the optimal solution of the system. The development of genetic algorithms is based on natural
selection principles whereas simulated annealing is based on thermodynamic processes.
Genetic Algorithms
Fogel et al. (1966) developed genetic algorithms (GA), alternatively known as evolutionary
programming, as “a technique in which candidate solutions to given tasks are represented as
finite-state machines, which are evolved by randomly mutating their state-transition diagrams.”
Holland (1975) focused on how genetic operators observed in nature, such as survival of the
fittest, crossover, and mutation, could be introduced into evolutionary computing. During the
last decades, genetic algorithm applications have been described in the works of De Jong
(1975), Grefenstette (1986), Goldberg (1989), and Davis (1991), and discussed thoroughly
in Mitchell (1996), who points out their strength in determining solutions for complex natural
systems. The robustness of genetic algorithms in geophysics has been only recently described
by Stoffa and Sen (1991) and Tavakoli and Pezeshk (2005). GAs are not limited by restrictive
assumptions concerning the continuity, the existence of derivatives, and the unimodality of the
function in terms of computational geometry. This is an advantage over regression analysis,
which often determines local minima as the solution of a given equation (Goldberg 1989) while
at the same time the suggested solution is highly dependent on a single point of initialization.
In the following paragraph, we focus on the initialization, the process of improvement, and
the destination of constrained minimization by using stochastic solvers such as GAs. By keep-
ing the analogy to biological systems, a number of chromosomes/strings form the genetic pre-
scription for the development and operation of the organism. In our case the chromosomes/
strings are composed of six genes/characters, representing the set of coefficients of Equation 1.
GAs start with a possible solution, or a set of possible solutions, corresponding to a the-
oretic attenuation curve and continue with optimization in order to determine the optimal
solution for the given data set. Both initialization options have been tested, corresponding
either to a single starting point (simple genetic algorithm) or multiple random-generated start-
ing points (genetic algorithm with initial population development), forming an initial popu-
lation of candidate solutions, each one satisfying the given linear constraints for the
determination of the coefficients of Equation 1. In order to ensure well-dispersed and random
initial population development for the adequate representation of the search space, Latin
Hypercube sampling was used (Diaz-Gomez and Hougen 2006). The technique was elabo-
rated by Iman et al. (1981) as stratified sampling without replacement, whereas in recent
years risk analysis software1 employs Latin Hypercube sampling for population development
in preference of Monte Carlo approach.
During optimization every candidate solution/string satisfying the given linear con-
straints is evaluated for its effectiveness in minimizing the misfit, through the objective func-
tion of Equation 2, hence bringing the response value of the system near a desired value.
1
http://www.palisade.com/risk/
Within the same generation (group of solutions) fitness scaling serves the purpose of ranking
each candidate solution to facilitate the selection of the best solutions that should survive in
the next generation. In that way, mimicking nature, the fitter solution survives and can be a
parent to the next generation, whereas worse fit solutions are penalized.
In the present study a number of 200 generations is considered, each one with a population
size of 600 individuals/solutions. Stochastic operators, such as crossover, mutation, and sur-
vival of the fittest guarantee the diversity of the population (Pan 1995), forcing the GA to
search the solution space intensively, thereby reducing the possibility that the algorithm
will return a local minimum (Goldberg 1989). In terms of survival, in this study, two elite
individuals/solutions pass to the next generation. A crossover fraction equal to 0.8 specifies
the percentage of individuals/solutions, other than elite children, that will be produced by
crossover in the next generation. Crossover mimics natural recombination between two parent
chromosomes/solutions during which the offspring chromosome/solution has changed values
of specific genes/numbers with respect to the parent genes/numbers. In this research, two-point
crossover has been implemented where each selected parent string is divided into three seg-
ments and the latter are exchanged between the initial parent strings, providing at the end, new
offspring strings. Mutation, on the other hand, alters a randomly selected character/coefficient
within a string/solution to create a new possible solution. Adaptive mutation, used in this study,
generates new directions in the search space, with respect to the last successful or unsuccessful
generation, bounded every time by the linear constraints set for the coefficients of Equation 1.
By combining the aforementioned stochastic operators, three versions of GAs aim to test
the relation between population diversity and performance:
1. Simple genetic algorithm (SGA)
2. Genetic algorithm with initial population development (GADP)
3. Hybrid genetic algorithm (HGA)
Hybrid genetic algorithm (HGA) introduces a solution, determined a priori by the simple
genetic algorithm (SGA), to a deterministic solver, which requires the existence of derivatives,
to provide local optima once the SGA has determined the neighborhood of the global optima
(Oh et al. 2004). The authors included this enhanced evolutionary algorithm in their compar-
ison to test the efficiency of the interaction between stochastic and deterministic solvers.
Simulated Annealing
Simulated annealing is a meta-heuristic algorithm proposed by Kirkpatrick et al. (1983)
and Cerny (1985) for the determination of global minima. It mimics the physical process by
which metals are slowly cooled so that eventually their crystal structure is frozen using a
minimum energy configuration, corresponding to the determination of the optimal solution
(Bertsimas and Tsitsiklis 1993).
Optimization starts from a randomly generated initial population and a hypothesis for a
parameter, known as temperature, slowly decreasing from 100°C by a factor of 0.0059°C in
the process of determining the optimal solution. During implementation each new possible
solution is evaluated for its effectiveness in minimizing the objective function (Equation 2),
and when a lower misfit value is found the suggested solution is adapted. Simulated anneal-
ing (Bohachevsky et al. 1986) has been an important solver in stochastic minimization
problems reported to successfully determine global minima. However, the authors agree with
the results of Ingber (1993) that this algorithm requires fine-tuning to each specific problem
relative to other solvers.
TESTING OPTIMIZATION SOLVERS

In order to test the efficiency of the optimization solvers in determining the coefficients of
Equation 1 for peak ground acceleration, a subset of the database was used corresponding to
rock site conditions and unspecified style of faulting. The solvers implemented for this test
correspond to: (1) simple genetic algorithm (SGA), (2) genetic algorithm with initial popula-
tion development through Latin Hypercube sampling (GADP), (3) hybrid genetic algorithm
(HGA), (4) simulated annealing algorithm (SA), (5) nonlinear least squares (NLLSQ), (6)
nonlinear least squares with initial population development (NLLSQDP), and (7) pattern
search algorithm (PS). Although the authors provided some basic principles of the solvers
in the previous section (see “Method”), details and theoretical comparison between the sol-
vers can be found in recent literature (Wetter and Wright 2003, Gabere 2007, Alander 1998,
El-Mihoub et al. 2006, Solomatine 1998, among others).
Before evaluating the performance of these solvers it is meaningful to describe their rela-
tive computational efficiencies. The major difference between the two main categories of
stochastic solvers is that genetic algorithms can either automatically produce a starting
point (SGA) or be enhanced by initial population development (GADP), whereas simulated
annealing requires a priori definition of initial conditions (Davis 1987). The latter require-
ment applies for deterministic algorithms (NLLSQ, NLLSQDP, PS) as well. The difference,
however, between simulated annealing and deterministic solvers (NLLSQ, NLLSQDP, PS)
is that the latter depend on the existence of derivatives in order to continue their iterations.
The performance evaluation of the optimization solvers follows a qualitative and quan-
titative approach. Analytically, by qualitative criteria the authors refer to inability of the sol-
ver to determine a possible solution (1) within the given number of generations, (2) satisfying
the linear constraints, and (3) in a timely manner (Aytug and Koehler 2000). When the above
criteria failed, optimization was implemented again with different starting points, which is
assumed to be the main source of error leading to a solver’s failure. The alternative solution—
that of relaxing tolerance criteria in case of a solver’s failure—was not considered since it
would jeopardize the final comparison between different solvers. Failure of a solver to pro-
duce a solution for a second time lead to its exclusion from the quantitative comparison. In
that sense, nonlinear least squares (NLLSQ) solver failing the (2) criterion has been imple-
mented again using multiple starting points (NLLSQDP). After the second failure to deter-
mine a feasible solution, by returning the initial conditions, nonlinear least squares (NLLSQ,
NLLSQDP) has been excluded from the comparison from this point forward. It is noted that
the initial conditions describe only the theoretical ground motion prediction equation, sub-
jectively set by the programmer in the beginning.
Table 2 presents the results of the quantitative comparison of the optimization solvers
together with the coefficients of Equation 1 and numerical details used during optimization,
such as the linear constraints introduced in the form of lower and upper bounds. It is noted
that the convergence criterion, alternatively known as tolerance, was set to 1E-06 for the
purpose of this test. The quantitative comparison has been based on two criteria: (1) the
Table 2. Initial and boundary conditions for the seven solvers used in the constrained
minimization problem, including the coefficients determined by each solver, the corresponding
standard error of the successful models, and their average sample log-likelihood value.
Initial conditions 1.0000 0.1500 0.0030 −0.5000 0.0100 0.0100

Lower bounds 1.0000 0.0100 0.0010 −1.5000 0.0001 0.0010
Upper bounds 3.0000 0.5000 1.0000 −0.0001 1.0000 2.0000
a B c d e h σ LLH
NLLSQ 1.70166 0.35714 0.00100 −1.22962 0.00010 0.00100

NLLSQDP 1.70166 0.35714 0.00100 −1.22962 0.00010 0.00100
GA 1.82357 0.32641 0.00279 −1.27552 0.00755 0.00116 0.3464 1.8184
GADP 2.62122 0.11622 0.01568 −1.47212 0.03988 0.00136 0.3426 1.8044
HGA 1.70170 0.35713 0.00100 −1.22961 0.00010 0.00100
PS 1.50000 0.01000 0.00100 −0.00010 0.00010 1.00100
SA 1.73819 0.32360 0.00608 −1.25514 0.00264 0.55508 0.3484 1.8259
standard error, expressed in logarithm base 10, and (2) the average sample log-likelihood
(LLH) value (Scherbaum et al. 2009) of the resulting ground motion prediction equation
as derived by a specific solver.
The standard error ðσ k Þ has been calculated as the mean of absolute residuals between
observed and predicted by the ground motion model g k (where k denotes the index of the
solver), described in Equation 3.
1
σk ¼ jobsi − g k ðxi Þj (3)
N
Assuming that the set of observations adequately describes nature, the likelihood Lðg k jxÞ
of the model g k given the set of observations x would represent how close the model g k
describes reality. According to Scherbaum et al. (2009) the average sample log-likelihood
(LLH) estimator (Equation 4) has been calculated as the mean of log-likelihood values over N
number of x samples:
N
1X
hlog 2 ðLðg k jxÞÞi ¼ log 2 ðg k ðxi ÞÞ (4)
N i¼1
Delavaud et al. (2009) use the latter estimator as a ranking criterion for testing the applic-
ability of ground motion prediction equations in seismic hazard studies, stating that the major
advantage of this criterion is that it is not sample-size dependent. In this study, the determi-
nation of the optimal algorithm for the constrained minimization problem is based on an
information theory approach based on this estimator. It is noteworthy that this study extends
the applicability of this criterion in areas outside the selection of attenuation relations for
seismic hazard purposes.
Figure 3 is a graphical representation of ground motion prediction equations as a result of
constrained minimization (see Equations 1 and 2) using different solvers for the case of a
magnitude 6.0 event at a rock site for unspecified style of faulting.
Figure 3. Peak ground acceleration versus distance for M6 and unspecified style of faulting, as
determined by simple genetic algorithm (SGA), genetic algorithm with initial population devel-
opment (GADP), and simulated annealing (SA) algorithm. The comparison involves solvers that
have successfully determined the coefficients of Equation 1.
Two main conclusions can be drawn from the results of the test: First, deterministic algo-
rithms, such as nonlinear least squares (NLLSQ, NLLSQDP), fail to determine the whole set
of coefficients since the values of the coefficients c, e, and h (Table 2) remain fixed to their
lower boundary value. The above remark emphasizes the weakness of deterministic algo-
rithms leading to the determination of local minima instead of returning a global solution
for the minimization problem. Second, the effectiveness of genetic algorithms in solving
minimization problems, even in their simpler parameterization (SGA), is supported by
their ranking following the LLH criterion. Thus, the use of genetic algorithms aided by initial
population development (GADP) by Latin Hypercube sampling is suggested for the con-
strained minimization problem set by Equation 2.
RESULTS
The coefficients of Equation 1 have been determined following a model parameter sto-
chastic optimization using genetic algorithms aided by initial population development,
described in the previous paragraph (Table 3). Figures 4 and 5 present the predicted ground
motion within the 1–150 km distance range for magnitude 5 and 6 for different soil site
conditions and normal style of faulting for peak ground and spectral acceleration, respec-
tively. The coefficients determined here also support the findings of Danciu and
294
Table 3. Coefficients of the ground motion prediction equation of this study
a b c d e h f1 f2 e1 e2 σ
PGA 1.92909 0.21829 0.00328 −1.06750 0.01016 0.01005 0.09664 0.08438 0.12297 0.09175 0.35530
PGV 0.34099 0.20796 0.02299 −1.19999 0.02258 0.02000 −0.03281 −0.02242 0.06292 0.14592 0.37480
PSAT(s)
0.05 2.66156 0.06961 0.01093 −1.09640 0.01102 0.48724 0.09653 0.09038 0.13052 0.06987 0.36470
0.08 3.50000 −0.14358 0.03575 −1.20000 0.01000 0.02000 0.00936 −0.00319 0.00544 −0.09032 0.36940
0.10 3.46395 −0.15202 0.03044 −1.09855 0.01001 0.01026 0.16362 0.13974 0.14210 0.02642 0.37690
0.15 3.00000 −0.10000 0.03800 −1.01000 0.01000 0.01030 −0.06279 −0.01569 −0.01095 −0.11561 0.37060
0.20 2.97121 −0.10620 0.03596 −1.01754 0.01003 0.01038 0.11329 0.14147 0.13557 0.07669 0.38230
0.25 3.00000 −0.15000 0.03999 −1.00001 0.02000 0.01999 0.01706 0.04995 0.07150 0.05547 0.39150
0.30 1.80981 0.16064 0.02101 −0.98343 0.01019 0.01146 0.10489 0.06650 0.14998 0.16916 0.38460
0.35 3.38155 −0.19764 0.02209 −0.49144 0.00416 1.99696 0.05220 0.04669 0.00793 0.13319 0.43480
0.40 1.90000 0.16000 0.02131 −0.99765 0.01000 0.01200 0.04822 −0.10739 −0.00781 0.05642 0.40320
0.45 2.94919 −0.13791 0.01918 −0.40942 0.00031 1.99986 0.04598 0.04097 0.02112 0.16015 0.42760
0.50 0.16184 0.54075 0.00163 −1.07036 0.01754 0.01646 0.07518 0.02552 0.18241 0.27060 0.37820
0.55 1.25865 0.20002 0.01109 −0.45422 0.00011 1.99997 0.05510 0.05492 0.08827 0.25781 0.45430
0.60 0.25266 0.30127 0.01008 −0.23860 0.00034 0.13595 0.03641 0.03832 0.12404 0.26115 0.43130
0.66 1.28057 0.21751 0.00117 −0.39226 0.00014 1.99994 0.03265 0.02476 0.06171 0.20877 0.40380
0.70 0.85749 0.30001 0.00109 −0.44564 0.00054 2.49910 0.03389 0.02735 0.07343 0.22261 0.39470
0.76 1.65758 0.13293 0.00329 −0.47992 0.01117 2.49999 0.02121 0.01575 0.03888 0.17570 0.37780
0.80 0.50000 0.30000 0.03000 −1.10000 0.02580 0.02000 −0.11698 −0.16848 0.04742 0.15695 0.38760
0.86 1.62528 0.10145 0.00451 −0.37808 0.00741 2.49498 0.11442 0.11698 0.05826 0.21481 0.41030
0.90 1.50302 0.12837 0.00324 −0.40799 0.01022 2.49997 0.11215 0.11109 0.06231 0.21796 0.40830
0.96 1.27202 0.18049 0.00191 −0.44827 0.00987 2.99999 0.10062 0.09488 0.06248 0.21062 0.40180
1.00 0.46026 0.26484 0.02273 −1.12771 0.05347 0.01861 −0.03319 −0.07071 0.16793 0.28830 0.39720
1.05 1.76823 0.09425 0.00111 −0.79840 0.06880 2.99585 0.09255 0.08648 0.04745 0.20373 0.41150
1.10 1.24260 0.16870 0.00211 −0.50353 0.02203 2.99947 0.08923 0.08824 0.05160 0.21070 0.41830
1.15 0.97917 0.22428 0.00183 −0.44920 0.00570 2.00429 0.07672 0.07805 0.04654 0.20661 0.41300
M. SEGOU AND N. VOULGARIS
1.20 1.47429 0.02689 0.02194 −0.45340 0.00022 0.01249 0.07868 0.06421 0.04537 0.20285 0.40490
1.25 1.16815 0.11955 0.01538 −0.47769 0.00029 0.01112 0.07678 0.06247 0.04982 0.20837 0.40450
1.30 1.23549 0.08814 0.01502 −0.47051 0.01068 0.00152 0.06975 0.06255 0.04633 0.21132 0.41930
1.35 1.23549 0.08814 0.01502 −0.47051 0.01068 0.00152 0.08221 0.08290 0.04706 0.22415 0.43360
1.40 1.82383 0.01378 0.01070 −0.80432 0.06667 0.01262 0.08281 0.09169 0.02857 0.21103 0.44210
1.45 1.54842 0.02244 0.01605 −0.45561 0.01092 0.01282 0.08558 0.09289 0.02293 0.20755 0.44700
1.50 0.39548 0.20522 0.02300 −1.19997 0.02001 0.01999 −0.05657 −0.04702 0.05007 0.14052 0.36800
1.55 1.22956 0.14492 0.00300 −0.63697 0.04322 0.03480 0.07220 0.07911 0.01964 0.19133 0.43260
1.60 1.34373 0.06662 0.01296 −0.56841 0.02959 0.00478 0.03451 0.04733 −0.00980 0.15430 0.42460
1.65 1.26049 0.08710 0.01300 −0.61238 0.03071 0.00467 0.04281 0.05912 0.00855 0.16525 0.41350
1.70 1.21309 0.11551 0.00989 −0.65834 0.03472 0.02641 0.02968 0.04183 0.00016 0.14996 0.40520
1.75 1.12806 0.10887 0.01411 −0.58629 0.01733 0.01027 0.02629 0.02966 0.00265 0.14668 0.39930
1.80 0.90498 0.18153 0.00825 −0.63404 0.02229 0.02505 0.02435 0.01544 0.00560 0.14259 0.39410
1.85 0.63004 0.24228 0.00510 −0.56258 0.01013 0.01009 0.02403 −0.00295 0.00631 0.13455 0.39150
1.90 0.83639 0.15966 0.01333 −0.57944 0.01011 0.01053 0.05052 0.02341 0.03073 0.15964 0.39100
1.95 0.58206 0.24806 0.00517 −0.58143 0.01035 0.01001 0.02496 −0.01930 −0.00097 0.11413 0.38720
2.00 0.68400 −0.16646 0.07296 −0.89049 0.02068 0.01891 0.09381 0.02055 0.27907 0.37352 0.41350
THE USE OF STOCHASTIC OPTIMIZATION IN GROUND MOTION PREDICTION
295
Figure 4. Predicted ground motion in terms of peak ground acceleration for seismic magnitude
equal to M5 and M6, at rock, stiff, and soft soil site conditions. Style of faulting corresponds to
normal.
Figure 5. Predicted ground motion in terms of peak ground acceleration for (a) pseudo-spectral
acceleration at 0.1 s and (b) pseudo-spectral acceleration at 2.0 s, for M5 and M6, at rock, stiff,
and soft soil site conditions. Style of faulting is normal.
Tselentis (2007) and Skarlatoudis et al. (2003) that the ground motion of strike-slip and thrust
events does not differ significantly for Greek earthquakes (Figure 4). In Figures 5a and 5b
soft soils have lower predicted amplitudes than stiffer soils, possible evidence of nonlinear
deamplification. Boore and Atkinson (2008) report the same observation, noting that this
effect is more pronounced at short periods, such as 0.1 s (Figure 5a) than at long periods,
such as 2.0 s (Figure 5b), where soft soil sites display amplification over stiff soil sites. Segou
et al. (2010), using the same database, have considered nonlinearity in soil behavior for the
same data set using the amplification factors of Boore and Atkinson (2008). However, there
has been no hard evidence that this approach provides better prediction for the stiff and soft
soil sites of this database.
Figures 6 and 7 display residuals, calculated as the difference between the logarithm of the
predicted value from the logarithm of observation, versus magnitude for peak ground accel-
eration (Figure 6) and spectral acceleration (Figures 7a and 7b). In Figure 6 the standard devia-
tions per soil class for sites with measured or estimated V s30 parameter are presented. For rock
and stiff soil sites the standard deviation of sites with measured V s30 is almost 10% lower than
the corresponding for estimated V s30 sites (Figure 6). Moreover, the standard deviation from
rock sites, both measured and estimated, is 20% lower than the corresponding one for stiff and
soft sites. This probably reflects the fact that for rock sites the geological description, which
becomes essential when the V s30 value or other site classification scheme is missing, was more
accurate. After the above remarks it is clear that providing precise measurements of the average
shear wave velocity at the sites (Moss 2008) could further reduce the epistemic uncertainty of
the particular model. Furthermore, an overestimation is observed in lower magnitudes
(Figures 6 and 7), up to magnitude 5, which the authors believe is related to the poor metadata
quality associated with weak events since the majority of observations in that range corre-
sponds to sites with estimated V s30 values (Figure 6). At this point the authors do not
favor magnitude-dependent sigma since there is no conclusive evidence that this is an artifact
Figure 6. Residuals between predicted and observed ground motion for peak ground accelera-
tion versus magnitude.
Figure 7. Residuals between predicted and observed ground motion for (a) pseudo-spectral
acceleration at 0.1 s and (b) pseudo-spectral acceleration at 2.0 s versus magnitude.
due to inadequate site class information. Furthermore, no correlation can be found between the
residuals and distance for peak ground acceleration (Figure 8).
Comparison of the proposed prediction model for peak ground acceleration with those
previously published for the area of Greece (Makropoulos and Burton 1985, Theodulidis and
Papazachos 1992, Skarlatoudis et al. 2003, Danciu and Tselentis 2007) is shown in Figure 9
for M6 at rock site conditions and distance ranges 1–150 km. The aforementioned models are
abbreviated as MK85, TP92, SK03, and DT07. Since these models, especially SK03 and
DT07, have used a very similar data set with the one used in this study and functional
Figure 8. Residuals between predicted and observed ground motion for peak ground accelera-
tion versus distance.
Figure 9. Comparison of the predicted peak ground acceleration values for distances up to
150 km for M6 and rock site conditions between the proposed model and Makropoulos and
Burton (1985) (MB85), Theodulidis and Papazachos (1992) (TP92), Skarlatoudis et al.
(2003) (SK03), and Danciu and Tselentis (2007) (DT07) models.
forms, it is the method of coefficient determination that influences more the predicted values.
It should be noted that MK85 and SK03 use hypocentral whereas DT07 and TH92 use epi-
central distance. For the calculations of this study a uniform depth = 10 km has been used. In
order to convert from M s and mb for the TP92 and MB85 models, the empirical equation of
Scordilis (2006) was implemented. Moreover, the TP92 and SK03 models both use horizon-
tal components and DT07 uses their arithmetic mean whereas MK85 does not provide any
documentation about the combination of horizontal components. Following Douglas (2003)
and Sharma et al. (2006), the use of either both horizontal components or their combination
into arithmetic or geometric mean produces similar results. Predicted values based on the
geometric mean and GMRotI50—used in the present study—are in good agreement
(Watson-Lamprey and Boore 2007). Based on the above no adjustment was necessary
for the combination of horizontal components. In Figure 9, the higher values predicted
by TP92 are probably due to (1) the multiple-step least square method, strongly correlating
the zero-order coefficient with the predicted values, and (2) using at close distances and large
magnitudes ðM s > 7.0 Þ recordings from Japan and Alaska. DT07 and SK03 models have
used similar data sets and both lack a second-order magnitude term in their functional
form. In the SK03 model the developers note that during regression they failed to determine
one coefficient based on their available data, which was afterwards adopted from another
study. The remaining coefficients of the SK03 model have been derived using least squares
optimization in one step whereas DT07 uses a nonlinear mixed-effects regression technique.
For the SK03 and DT07 models the implementation of deterministic algorithms, such as least
squares, and regression techniques presents similar results. We note that the coefficients of
the proposed model have been determined simultaneously and independent from each other.
Although the authors included the MB85 model in this comparison, since it has been exten-
sively used in the early era of seismic hazard in Greece, no useful conclusions can be drawn
regarding the influence of the regression technique. The reason behind this is that this model
has been “found by trial and error” as “an average equation for the peak ground acceleration”
representing a number of eight formulae (Makropoulos and Burton 1985). Additionally, the
global models that were averaged do not correspond to rock site conditions, making it dif-
ficult to determine the site class of the final predicted value following the MB85 model. As
shown in Figure 9 the proposed model successfully addresses the problem of scarce data in
the near field from large earthquakes, which resulted in low predicted values at close dis-
tances for the SK04 and DT07 models, due to the stochastic optimization procedure.
CONCLUSIONS
In this paper the coefficients of the ground motion model have been determined using the
optimum solver derived by our analysis corresponding to genetic algorithms with initial
population development using Latin Hypercube topology. The role of stochastic operators
(survival of the fittest, mutation, crossover) guarantees the diversity in future populations, but
the percentage of new solutions introduced to the next generation as a result of these opera-
tors should be calibrated on the basis of initial population diversity and available data sets.
Moreover, the connection between population diversity and the performance of genetic algo-
rithms performance was investigated using a single starting point and multiple randomly
generated initial populations. The results support the significance of initial population devel-
opment in enhancing the efficiency of the stochastic solvers. However, the selection of the
appropriate solver for optimization is directly linked with the mathematical equation describ-
ing the ground motion prediction equation. In that sense, it is necessary to evaluate the
performance of the candidate solvers for their efficiency to determine global minima
using data-driven explicit criteria, such as the average sample log-likelihood.
Using a strong motion data set from Mediterranean countries and the optimum solver,
previously defined, ground motion prediction equations for peak ground acceleration, velo-
city, and spectral acceleration for periods ranging from 0.05 s to 2 s for 5% damping were
developed. These strong motion prediction equations are proposed for application in the
range M4.5–M6.6 and for distances up to 150 km.
ACKNOWLEDGMENTS
The first author expresses sincere thanks to Dr. Boore for instructive comments during deri-
vation of strong ground motion prediction equations using regression analysis. Many thanks
also to Assistant Professor Tzanis for exchanging ideas regarding ground motion prediction
equations applicability. This manuscript has been greatly improved after the recommendations
of three anonymous reviewers. This research has been partly funded by the Greek Scholarship
Institute (IKY). The authors would like to thank the responsible editor and three anonymous
reviewers for their invaluable comments, leading to the final edition of this manuscript.
APPENDIX
Table 1. Earthquake list used in this study.
No. records M Depth (km) Eartquake Name D M Y H m s
1 5.8 13 Ionian 4 11 1973 15 52 12

1 5.1 15 Ionian 4 11 1973 16 11 36
1 4.8 13 Patra 29 1 1974 15 12 43
12 6.5 6 Friuli 6 5 1976 20 0 13
1 5.1 10 Amfissa 29 12 1977 16 52 59
1 4.5 26 Achaia 18 5 1978 0 18 49
3 6.2 3 Central 20 6 1978 20 3 22
1 5.3 6 Volvi 4 7 1978 22 23 28
6 5.8 4 Valnerina 19 9 1979 21 35 37
1 5.1 12 Almiros 16 7 1980 0 6 58
1 5.5 5 Almiros 11 8 1980 9 15 59
1 5.0 5 Almiros 26 9 1980 4 19 21
2 6.2 18 Alkyonides 24 2 1981 20 53 37
1 5.9 8 Alkyonides 25 2 1981 2 35 51
2 5.6 10 Preveza 10 3 1981 15 16 20
1 4.8 10 Paliambela 10 4 1981 8 33 32
1 4.8 15 Panagoula 25 5 1981 23 4 0
1 5.2 15 Lefkada 27 5 1981 15 4 2
2 6.3 9 Kefallonia 17 1 1983 12 41 31
1 5.4 15 Kefallonia 17 1 1983 15 53 56
1 5.4 5 Kefallonia 31 1 1983 15 27 2
1 5.2 25 Etolia 16 3 1983 21 19 41
1 5.9 30 Heraklio 19 3 1983 21 41 42
1 5.2 25 Kefallonia 23 3 1983 19 4 6
3 5.8 13 Kefallonia 23 3 1983 23 51 5
3 6.1 2 Agion Oros 6 8 1983 15 43 53
3 5.2 12 Ierissos 26 8 1983 12 52 9
1 4.9 30 Heraklio 1 3 1984 9 8 30
4 5.6 7 Umbria 29 4 1984 5 2 59
12 5.9 8 Lazio-Abruzzo 7 5 1984 17 49 43
1 4.9 24 Corinth Gulf 17 8 1984 21 22 58
1 5.3 6 Zakynthos 4 10 1984 10 15 12
1 5.2 11 Kremidia 25 10 1984 9 49 15
1 5.3 30 Elis 13 8 1985 13 49 14
3 5.0 15 Preveza 31 8 1985 6 3 47
2 5.7 18 Drama 9 11 1985 23 30 43
2 6.0 1 Kalamata 13 9 1986 17 24 34
3 5.3 10 Kalamata 15 9 1986 11 41 28
1 4.9 20 Tsipiana 1 2 1987 5 35 36
(continued )
1 5.0 9 Kounina 14 5 1987 6 29 11

2 5.4 6 Dodecanese 5 10 1987 9 27 2
1 5.2 5 Corinth Gulf 3 4 1988 3 35 7
1 4.5 1 Ionian 24 4 1988 10 10 33
2 5.5 21 Etolia 18 5 1988 5 17 42
1 5.4 15 Etolia 22 5 1988 3 44 15
2 5.5 17 Corinth Gulf 5 7 1988 20 34 52
1 5.5 12 Kyllini 22 9 1988 12 5 39
1 4.9 1 Kyllini 30 9 1988 11 3 54
1 5.2 5 Kyllini 30 9 1988 13 2 54
3 5.6 25 Kyllini 16 10 1988 12 34 5
1 4.3 10 W. Peloponnese 19 10 1988 0 27 999
2 4.5 18 Kyllini 31 10 1988 2 59 51
1 5.2 5 Ionian 13 12 1988 11 0 35
2 4.9 10 Patra 22 12 1988 9 56 50
1 5.4 10 W. Peloponnese 7 6 1989 19 46 0
1 4.8 23 Patra 31 8 1989 21 29 31
3 5.8 9 Potenza 5 5 1990 7 21 29
1 5.0 15 Aigio 17 5 1990 8 44 6
2 5.7 29 Ionian 16 6 1990 2 16 0
2 5.2 10 Plati 8 8 1990 0 35 7
1 4.9 5 Sarti 3 9 1990 5 35 49
1 5.1 8 Sarti 9 9 1990 19 0 38
1 5.9 1 Griva 21 12 1990 6 57 43
1 4.9 28 Messini Gulf 1 6 1991 5 42 2
1 5.4 3 Kefallonia 23 1 1992 4 24 17
1 4.5 9 Kefallonia 1 2 1992 9 5 18
2 6.6 10 Erzican 13 3 1992 17 18 40
2 5.2 12 Mataranga 30 5 1992 18 55 40
2 6.0 12 Tithorea 18 11 1992 21 10 41
1 5.3 5 Gulf 4 2 1993 2 22 59
1 4.6 4 Sarti 24 2 1993 11 13 0
1 5.5 1 Near 5 3 1993 6 55 6
1 5.9 25 Kallithea 18 3 1993 15 47 0
2 5.1 3 Pyrgos 26 3 1993 11 45 16
2 5.1 10 Pyrgos 26 3 1993 11 56 14
2 5.5 10 Pyrgos 26 3 1993 11 58 15
1 5.0 10 Pyrgos 26 3 1993 12 49 13
2 5.6 20 Ionian 13 6 1993 23 27 0
5 5.5 13 Patra 14 7 1993 12 31 50
1 4.5 10 Patra 14 7 1993 12 39 13
1 4.1 10 Patra 14 7 1993 12 54 7
(continued )
2 5.4 10 Gulf 4 11 1993 5 18 37

4 5.5 5 Komillion 25 2 1994 2 30 50
1 4.5 10 Ionian 27 2 1994 22 34 52
1 5.1 18 Ionian 10 4 1994 19 46 21
3 6.0 30 South 23 5 1994 6 46 16
1 5.8 12 Florina 1 9 1994 16 12 41
1 4.8 7 W. Peloponnese 27 10 1994 7 2 30
1 4.6 13 Aigio 12 11 1994 1 6 0
1 5.2 21 W. Peloponnese 29 11 1994 14 30 999
1 4.8 17 W. Peloponnese 21 12 1994 10 22 0
1 4.2 5 Ionian 1 1 1995 17 22 0
1 4.3 8 Ionian 16 1 1995 7 26 0
1 4.8 10 W. Peloponnese 16 3 1995 6 25 999
1 5.0 23 Arnaia 3 5 1995 21 43 27
3 5.3 14 Arnaia 4 5 1995 0 34 11
1 4.8 10 Ionian 10 5 1995 22 46 0
1 6.1 14 Kozani 13 5 1995 8 47 15
1 4.8 18 Kozani 14 5 1995 14 46 57
2 5.4 25 Kozani 15 5 1995 4 13 57
1 5.4 20 Kozani 17 5 1995 4 14 25
1 5.4 10 Kozani 19 5 1995 6 48 49
1 4.9 26 Kozani 20 5 1995 21 6 25
1 4.9 18 Aigio 28 5 1995 19 57 0
3 5.1 27 Kozani 11 6 1995 18 51 48
1 4.1 13 Ierissos 12 6 1995 1 31 19
2 5.5 5 Aigio 15 6 1995 0 31 0
1 5.1 25 Aigio 15 6 1995 4 51 0
1 4.7 9 Aigio 15 6 1995 7 1 0
1 4.3 10 Aigio 18 6 1995 12 9 0
1 4.5 10 Aigio 19 6 1995 6 19 0
1 4.5 11 Aigio 25 6 1995 1 6 0
1 4.9 12 Aigio 5 7 1995 18 25 0
1 4.3 10 Ionian 17 7 1995 19 13 0
1 5.1 10 W. Peloponnese 13 8 1995 5 17 0
1 5.0 15 Aigio 28 9 1995 6 17 0
4 6.4 10 Dinar 1 10 1995 15 57 13
1 5.0 26 Ionian 5 10 1995 8 21 0
1 4.4 23 Ionian 6 10 1995 2 10 0
1 4.3 5 Ionian 11 10 1995 9 48 0
1 4.8 10 Ionian 17 12 1995 8 22 0
1 4.2 15 Ionian 18 5 1996 12 33 0
1 4.6 15 Ionian 26 5 1996 21 44 0
(continued )
1 4.5 10 Ionian 1 6 1996 9 24 0

1 5.0 10 W. Peloponnese 6 6 1996 16 26 0
1 4.9 6 Ionian 4 7 1996 21 57 0
1 4.3 6 Ionian 4 7 1996 22 25 0
2 4.9 15 Central Greece 11 8 1996 11 43 45
1 4.1 5 Ionian 13 8 1996 21 51 0
14 5.7 7 Umbria 26 9 1997 0 33 16
18 6.0 6 Umbria 26 9 1997 9 40 30
4 6.3 1 Kalamata 13 10 1997 13 39 40
3 6.0 10 W. Greece 18 11 1997 13 8 0
1 5.5 25 Strofades 18 11 1997 13 14 0
1 4.4 18 Ionian 23 4 1998 12 3 32
1 5.0 25 Ionian 8 10 1998 3 50 0
1 4.5 10 W. Greece 2 2 1999 14 27 51
10 5.8 9 Athens 7 9 1999 11 56 51
1 4.7 20 W. Greece 21 10 1999 8 45 48
2 5.0 13 W. Greece 27 4 2000 19 3 46
1 5.5 8 Ionian 26 5 2000 1 28 22
1 5.0 21 W. Greece 25 7 2000 19 33 59
4 6.0 19 W. Greece 26 7 2001 0 21 39
1 4.8 25 Ionian 2 9 2001 6 59 25
1 5.5 10 W. Greece 16 9 2001 2 0 48
1 5.1 6 Central Greece 7 12 2001 19 44 53
1 4.9 6 W. Greece 5 4 2002 7 55 51
2 5.6 15 Aegean 21 5 2002 20 53 31
2 4.9 15 Attiki 5 9 2002 22 19 52
1 5.5 19 Ionian 2 12 2002 4 58 57
1 5.1 15 Kalamata 29 4 2003 1 51 20
19 5.8 12 Lefkada 14 8 2003 5 14 55
1 5.1 26 Ionian 14 8 2003 12 18 15
3 5.4 10 Ionian 14 8 2003 16 18 4
2 5.1 8 Ionian 16 11 2003 7 22 50
1 5.5 24 W. Greece 31 1 2005 1 5 29
1 5.0 26 Ionian 19 9 2005 22 17 17
2 5.6 19 W. Greece 18 10 2005 15 26 0
1 5.3 10 W. Greece 18 10 2005 15 36 24
3 5.4 10 W. Greece 4 4 2006 22 5 3
1 5.5 10 W. Greece 11 4 2006 0 2 42
3 5.5 10 W. Greece 11 4 2006 17 29 28
3 5.6 10 W. Greece 12 4 2006 16 52 1
1 5.3 10 W. Greece 19 4 2006 15 16 25
REFERENCES
Akkar, S., and Bommer, J. J., 2007. Prediction of elastic displacement response spectra in Europe
and the Middle East, Earthquake Engineering and Structural Dynamics 36, 1275–1301.
Alander, J. T., 1998. An Indexed Bibliography of Genetic Algorithms and Simulated
Annealing: Hybrids and Comparisons, Report for the Department of Electrical Engineering
and Automation, University of Vaasa, Finland.
Ambraseys, N., Smit, P., Sigbjornsson, R., Suhadolc, P., and Margaris, B., 2002. Web page for
European Strong-Motion Data, European Commission, Research-Directorate General, Envir-
onment and Climate Programme, http://www.isesd.cv.ic.ac.uk/, accessed 7 March 2010.
Anderson, J. G., 2000. Expected shape of regressions for ground-motion parameters on rock,
Bulletin of Seismological Society of America 90, S43–S52.
Audet, C., and Dennis, J. E., 2002. Analysis of generalized pattern searches, SIAM Journal on
Optimization 13, 889–903.
Aytug, H., and Koehler, G. J., 2000. New stopping criterion for genetic algorithms, European
Journal of Operational Research 126, 662–674.
Bertsimas, D., and Tsitsiklis, J., 1993. Simulated annealing, Statistical Review 8, 10–15.
Bohachevsky, I. O., Johnson, M. E., and Stein, M. L., 1986. Generalized simulated annealing for
function optimization, Technometrics 28, 209–217.
Bolt, B. A., and Abrahamson, N. A., 2003. Estimation of strong seismic ground motions,
in International Handbook of Earthquake and Engineering Seismology (V. W. Lee,
H. Kanamori, P. C. Jennings and C. Kisslinger, eds.), Academic Press, Inc., vol. B,
983–1001.
Boore, D. M., Joyner, W. B., and Fumal, T. E., 1993. Estimation of response spectra and peak
accelerations from western North American earthquakes: An interim report, Open File Report
94–127.
Boore, D. M., Watson-Lamprey, J., and Abrahamson, N. A., 2006. Orientation-independent mea-
sures of ground motion, Bulletin of Seismological Society of America 96, 1502–1511.
Boore, D. M., and Atkinson, G. M., 2008. Ground-motion prediction equations for the average
horizontal component of PGA, PGV, and 5%-damped PSA at spectral periods between 0.01 s
and 10.0 s, Earthquake Spectra 24, 99–138.
Cerny, V., 1985. A thermodynamical approach to the travelling salesman problem: An efficient
simulation algorithm, Journal of Optimization Theory and Applications 45, 41–51.
Danciu, L., and Tselentis, G.-A., 2007. Engineering ground-motion parameters and attenuation
relationships for Greece, Bulletin of Seismological Society of America 97, 162–183.
Davis, L., 1987. Genetic Algorithms and Simulated Annealing, Hyperion Books, London, 224 pp.
Davis, L., 1991. Handbook of Genetic Algorithms, Van Nostrand Reinhold, New York, 384 pp.
De Jong, K. A., 1975. An analysis of the behavior of a class of genetic adaptive systems, Dis-
sertation Abstracts International 36, 5140B.
Delavaud, E., Scherbaum, F., Kuehn, N., and Riggelsen, C., 2009. Information-theoretic selection
of ground-motion prediction equations for seismic hazard analysis: An applicability study
using Californian data, Bulletin of Seismological Society of America 99, 3248–3263.
Dennis, J. E., Jr., 1977. Nonlinear Least-Squares: State of the Art in Numerical Analysis, ed.
D. Jacobs, Academic Press, 269–312.
Diaz-Gomez, P. A., and Hougen, D. F. 2006. Genetic algorithms for hunting snakes in hyper-
cubes: Fitness function and open questions, in Proceedings of the Seventh ACIS International
Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distrib-

uted Computing, Las Vegas, 389–394.
Douglas, J., 2003. Earthquake ground motion estimation using strong-motion records: A review
of equations for the estimation of peak ground acceleration and response spectral ordinates,
Earth-Science Reviews 61, 43–104.
Dolan, E. D., Lewis, R. M., and Torczon, V., 2003. On the local convergence of pattern search,
SIAM Journal on Optimization 14, 567–583.
El-Mihoub, T. A., Hopgood, A. A., Nolle, L., and Battersby, A., 2006. Hybrid genetic algorithms:
A review, Engineering Letters 13, 124–137.
Filho, J. L., Treleaven, P. C., and Alippi, C., 1994, Genetic-algorithm programming envirin-
ments, Computer 27, 28–43.
Fogel, L. J., Owens, A. J., and Walsh, M. J., 1966. Artificial intelligence through simulated evo-
lution, John Wiley, New York.
Gabere, M. N., 2007. Simulated Annealing Driven Pattern Search. Algorithms for Global Opti-
mization, MSc Thesis, School of Computational and Applied Mathematics, University of the
Witwatersrand.
Goldberg, D. E., 1989. Genetic Algorithms in Search, Optimization and Machine Learning,
Addison Wesley Longman, 412 pp.
Graizer, V., and Kalkan, E., 2009. Prediction of response spectral acceleration ordinates based on
PGA attenuation, Earthquake Spectra 25, 36–69.
Grefenstette, J. J., 1986. Optimization of control parameters for genetic algorithms, IEEE Trans-
actions SMC 16, 122–128.
Holland, J. H., 1975. Adaptation in Natural and Artificial Systems: An Introductory Analysis with
Applications to Biology, Control, and Artificial Intelligence, University of Michigan Press,
Ann Arbor (second edition: MIT Press, 1992).
Hookes, R., and Jeeves, T. A., 1961. “Direct search” solution of numerical and statistical pro-
blems, Journal of the Association for Computing Machinery 8, 212–229.
Idriss, I. M., 1995. An overview of earthquake ground motions pertinent to seismic zonation, in
Proceedings of the 5th International Conference on Seismic Zonation, Nice, 17–19 October
1995, vol. 3, 2111–2126.
Iman, R. L., Helton, J. C., and Campbell, J. E., 1981. An approach to sensitivity analysis of
computer models, Part 1. Introduction, input variable selection and preliminary variable
assessment, Journal of Quality Technology 13, 174–183.
Ingber, L., 1993. Simulated annealing: Practice versus theory, Journal of Mathematical Compu-
tational Modelling 18, 29–57.
Joyner, W. B., and Boore, D. M., 1981. Peak horizontal acceleration and velocity from strong-
motion records including records from the 1979 Imperial Valley, California, earthquake, Bul-
letin of Seismological Society of America 71, 2011–2038.
Kao, C. Y., Chung, J. K., and Yeh, Y. T., 2010. A comparative study of the least squares method
and the genetic algorithm in deducing peak ground acceleration attenuation relationships, Ter-
restrial, Atmospherical and Oceanic Sciences 21, 869–878.
Kirkpatrick, S., Gelatt, C. D., Jr., and Vecchi, M. P., 1983. Optimization by simulated annealing,
Science 220, 671–680.
Makropoulos, K. C., and Burton, P. W., 1985. Seismic hazard in Greece II. Ground Acceleration.
Tectonophysics 117, 259–294.
Mitchell, M., 1996. An Introduction to Genetic Algorithms, MIT Press, Cambridge, 221 pp.
Moss, R., 2008. Quantifying measurement uncertainty of thirty-meter shear-wave velocity, Bul-
Moss, R., 2009. Reduced Uncertainty of Ground Motion Prediction Equations through Bayesian
Variance Analysis, PEER Report 2009/105.
Lindstrom, M. J., and Bates, D. M., 1990. Nonlinear mixed effects models for repeated measures
data, Biometrics 46, 673–687.
Oh, I-S., Lee, J-S., and Moon, B-R., 2004. Hybrid genetic algorithms for feature selection, IEEE
Transactions on Pattern Analysis and Machine Intelligence 26, 1424–1437.
Pan, Z., Chen, Y., Kang, L., and Zhang, Y., 1995. Parameter estimation by genetic algorithms for
nonlinear regression, in Proceedings of International Conference on Optimization Technique
and Applications, G. Z. Liu (ed.), World Scientific, Hackensack, NJ, 946–953.
Rothlauf, F., 2006. Representations for Genetic and Evolutionary Algorithms, Springer Verlag,
Berlin, 289 pp.
Said, Y. H. 2005. Genetic algorithms and their applications, in Handbook of Statistics 24, C. R.
Rao, E. J. Wegman, and J. L. Solka (eds.), 359–390.
Scherbaum, F., 2004. On the conversion of source-to-site distance measures for extended earth-
quake source models, Bulletin of Seismological Society of America 94, 1053–1059.
Scherbaum, F., Cotton, F., and Staedke, H., 2006. The estimation of minimum-misfit stochastic
models from empirical ground-motion prediction equations, Bulletin of Seismological Society
of America 96, 427–445.
Scherbaum, F., Delavaud, E., and Riggelsen, C., 2009. Model selection in seismic hazard ana-
lysis: An information-theoretic perspective, Bulletin of Seismological Society of America 99,
3234–3247.
Scordilis, E. M., 2006. Empirical global relations converting Ms and mb to moment magnitude,
Journal of Seismology 10, 225–236.
Segou, M., Voulgaris, N., Makropoulos, K. C., and Stavrakakis, G. N., 2008. A review of the
Greek strong motion database: Needs, improvements and future development, in Proceedings
of the 31st General Assembly of the European Seismological Commission, Hersonissos, Crete,
Greece, 7–12 September 2008, 422–427.
Segou, M., and Voulgaris, N., 2010. Proschema: A Matlab application for processing strong
motion records and estimating earthquake engineering parameters, Computers & Geosciences
36, 977–986.
Segou, M., Voulgaris, N., and Makropoulos, K., 2010. On the sensitivity of ground motion pre-
diction equations in Greece, Bulletin of Geological Society of Greece (Special Issue: Proceed-
ings of the 12 th International Congress of the Geological Society of Greece, Planet Earth:
Geological Processes & Sustainable Development) XLII, Paper 2163.
Sharma, M. L., Bhattacharya, A., Bungum, H., and Kumar, A., 2006. Definition of horizontal
component of strong ground motion using Indian region data set, Indo Norwegian Workshop
on Seismic Hazard and Risk Assessment, IHC, New Delhi, India, 17–18 March 2006.
Skarlatoudis, A. A., Papazachos, C. B., Margaris, B. M., Theodulidis, N. P., Papaioannou, C.,
Kalogeras, I., Scordilis, E. M., and Karakostas, V. G., 2003. Empirical peak ground-motion
predictive relations for shallow earthquakes in Greece, Bulletin of Seismological Society of
America 93, 2591–2603.
Solomatine, D. P., 1998. Genetic and other global optimization algorithms: Comparison and use
in calibration problems, in Proceedings of the 3rd International Conference on Hydroinfor-
matics, Balkema Publishers, Copenhagen, Denmark, 1021–1028.
Stoffa, P. L., and Sen, M. K., 1991. Nonlinear multiparameter optimization using genetic algo-
rithms: Inversion of plane wave seismograms, Geophysics 56, 1794–1810.
Spudich, P., Joyner, W. B., Lindh, A. G., Boore, D. M., Margaris, B. M., and Fletcher, J. B.,
1999. SEA99: A revised ground motion prediction relation for use in extensional tectonic
regimes, Bulletin of the Seismological Society of America 89, 1156–1170.
Tavakoli, B., and Pezeshk, S., 2005. Empirical-stochastic ground-motion prediction for eastern
North America, Bulletin of Seismological Society of America 95, 2283–2296.
Theodulidis, N. P., and Papazachos, B. C., 1992. Dependence of strong ground motion on mag-
nitude-distance, site geology and macroseismic intensity for shallow earthquakes in Greece: I,
Peak horizontal acceleration, velocity and displacement, Soil Dynamics and Earthquake Engi-
neering 11, 387–402.
Wang, G. Q., Boore, D. M., Igel, H., and Zhou, X. Y., 2003. Some observations on colocated and
closely-spaced strong ground motion records of the 1999 Chi-Chi, Taiwan Earthquake, Bul-
Watson-Lamprey, J. A., and Boore, D. M., 2007. Beyond SaGMRotI: Conversion to SaArb,
SaSN, and SaMaxRot, Bulletin of the Seismological Society of America 97, 1511–1524.
Wetter, M., and Wright, J., 2003. Comparison of generalized pattern search and a genetic algo-
rithm optimization method, in Proceedings of the Eighth International Conference on Build-
ing Simulation (IBPSA), Eindhoven, Netherlands, 1401–1408.
A FEMA P695 Study for the Proposed
Seismic Performance Factors for
Cold-Formed Steel Special
Bolted Moment Frames
Atsushi Sato a) and Chia-Ming Uangb)
The objective of this study was to verify the adequacy of the proposed Seismic
Performance Factors for the newly developed Cold-Formed Steel Special Bolted
Moment Frames in the AISI S110 Seismic Standard. The FEMA P695
methodology, Qualification of Building Seismic Performance Factors, was used
for this purpose. A total of 13 archetype designs, representing two seismicity
and two gravity load levels, were designed and analyzed. The computed results
from all individual archetypes and four performance groups showed that the
proposed seismic performance factors met the acceptance criteria and could
provide a sufficient margin against collapse under the maximum considered
earthquake. [DOI: 10.1193/1.4000099]
INTRODUCTION
The American Iron and Steel Institute (AISI) recently published S110, Standard for Seis-
mic Design of Cold-Formed Steel Structural Systems–Special Bolted Moment Frames (AISI
2007). Special bolted moment frame (SBMF) is intended for use as freestanding light storage
mezzanines, elevated office support platforms, equipment support platforms, and small build-
ings in all seismic areas. It is composed of cold-formed hollow structural section (HSS) col-
umns and double C-section beams connected by snug-tight high-strength bolts (see Figure 1).
If needed, bearing plates can be welded to the connection region to increase the bearing
strength of the bolted connection. The S110 limits the SBMF to one story with a height
up to 10.7 m.
The design provisions for the SBMF in S110 are based on the results of full-scale
testing of nine beam-column subassemblies (Uang et al. 2004) and the associated analytical
studies. Testing showed that the bolted moment connection can provide a reliable source of
energy dissipation through bolt friction and bearing. Figure 2a shows a typical hysteresis
behavior of the bolted connection. A yield-like behavior was produced by bolt slippage,
which was then followed by significant hardening due to bolt bearing. Based on the concept
of instantaneous center of rotation of an eccentrically loaded bolt group (Crawford and
Kulak 1971), a numerical model that can properly simulate the cyclic response of the
a)
Nagoya Institute of Technology, Showa-Ku, Gokiso-Cho, Bldg. 24-305, Nagoya, Japan 4668555
b)
University of California, San Diego, La Jolla, CA 92093-0085
259
260 A. SATO AND C.-M. UANG
For Double C-section

Beam only
HSS Column
C
L
C-section Beam
VIEW A-A
B Bolt Bearing
A c A
Plate (Optional)
Bolt Bearing
Plate a
(Optional)
CL b C
L
a
h C-section
C-section Beam
Beam
HSS Column HSS Column
CL B C
L
ELEVATION VIEW B-B
(a) SBMF (b) Typical bolted moment connection detail
Figure 1. Typical SBMF.
moment connection was developed (Uang et al. 2010). Testing also showed that the beam-
column subassemblage was able to dissipate a significant amount of energy as long as
the beam and column were designed to remain essentially elastic by the capacity design
principles.
The current design procedure permits the designer to reduce the elastic seismic design
forces corresponding to the design basis earthquake by an R factor. The designer then only
needs to perform linear elastic analysis. A structure thus designed actually is expected
to deform inelastically to dissipate seismic energy, and those structural components
(e.g., beams and columns in SMBF) that are expected to remain elastic would experience
much higher seismic forces. Instead of requiring the designer to perform nonlinear analysis to
evaluate this “amplified” seismic effect, ASCE 7 (ASCE 2005, 2010) provides an empirical
system overstrength factor, Ω0 , for this purpose. With the connection model for SBMF
mentioned above, however, the nonlinear response of the bolt group can be explicitly
considered in design, and S110 provides a design procedure to facilitate the calculation
of maximum moment that can be developed at the design story drift level (Sato and
Uang 2009). Therefore, specifying an empirical Ω0 for SBMF is unnecessary when the
procedure in S110 is used.
The above capacity design procedure requires the design story shear to be calculated
first. Previously, the design practice has been to treat SBMF as an ordinary moment frame
with a response modification coefficient, R, value of 3.5. This value is deemed appro-
priate for SBMF because cyclic testing (see Figure 2b) showed a high ductility capacity
of the frame. Based on the moment connection model and a series of nonlinear time his-
tory analysis, a statistical evaluation showed that, for a given ductility factor, the seismic
force reduction for an SBMF can be higher than that of an elasto-perfectly plastic single
degree-of-freedom system (Sato and Uang 2010). It is the significant hardening effect in
the moment connection that contributes to this higher seismic force reduction. The
A FEMA P695 STUDY FOR THE PROPOSED SEISMIC PERFORMANCE FACTORS FOR COLD-FORMED STEEL SBMF 261
Story Drift Ratio (%)

-10 -5 0 5 10
150 150 60
100 100 40
Moment (kN-m)
Moment (kN-m)
Load (kN)
50 50 20
0 0 0
-50 -50 -20
-100 -100 -40
-150 -150 -60
-0.06 -0.04 -0.02 0 0.02 0.04 0.06 -200 -100 0 100 200
Slip-Bearing Rotation (rad) Story Drift (mm)
(a) Bolted moment connection behavior (b) Typical global response of SBMF
Figure 2. Typical hysteresis behaviors.
following deflection amplification factor, Cd , was also derived (Sato and Uang 2010):
R
Cd ¼ (1)
1.2
which corresponded to a C d value of about 3.0.

The objective of this study was to evaluate if the proposed seismic performance factors
(R and C d ) are acceptable and can provide a sufficient margin against collapse under max-
imum considered earthquake ground motions. The evaluation procedure developed by
FEMA P695 (FEMA 2009), Qualification of Building Seismic Performance Factors, was
used for this purpose.
STRUCTURAL SYSTEM INFORMATION
IDENTIFICATION OF SBMF ARCHETYPE CONFIGURATION

Figure 3 shows the two-dimensional frame geometry that was used to model the one-
story SBMF structures; both one-span and two-span frames were considered. With this con-
figuration, a set of archetype designs was developed to represent the archetype design space.
This set of designs was organized into four performance groups per P695 to represent the
range of design ground motion intensity and gravity loading:
• Maximum seismic (Dmax ) and low gravity loading
• Maximum seismic (Dmax ) and high gravity loading
• Minimum seismic (Dmin ) and low gravity loading
• Minimum seismic (Dmin ) and high gravity loading
ARCHETYPE STRUCTURAL DESIGN

Table 1 shows the design data of each archetype design. The high and low seismic demands
in terms of code-specified spectral intensities were represented by the maximum and minimum
demands possible in seismic design category (SDC) D (ASCE 2005). The archetypes were
designed for a Site Class D condition (soft soil) with SDS ¼ 1.0g as well as SD1 ¼ 0.6g
for SDC Dmax and SDS ¼ 0.5g and SD1 ¼ 0.2g for SDC Dmin . The corresponding maximum
Tributary area
Bay
Beam
Bolted Connection Bolted Connection
Height
Bay
Column
Span Span Span Span
(a) Plan (b) Elevation
Figure 3. Archetype analysis model for SBMF.
Table 1. Archetype structural design parameters
Key archetype design parameters
Seismic design criteria

Archetype
design ID Number Span Bay Height T
number of span (m) (m) (m) SDC R (sec) SMT ðTÞ ðgÞ
Maximum seismic ðDmax Þ and low gravity designs
1 1 6.10 4.88 4.88 Dmax 3.5 0.87 1.04
2 2 6.52 6.10 3.12 Dmax 3.5 0.51 1.50
3 2 7.12 6.10 4.88 Dmax 3.5 0.77 1.16
4 2 6.52 3.66 9.14 Dmax 3.5 1.36 0.66
Maximum seismic ðDmax Þ and high gravity designs
5 2 6.52 4.88 4.88 Dmax 3.5 0.82 1.10
6 1 6.10 4.88 4.88 Dmax 3.5 0.73 1.24
7 2 6.52 4.27 7.32 Dmax 3.5 1.08 0.84
8 1 6.79 4.88 3.35 Dmax 3.5 0.57 1.50
Minimum seismic ðDmin Þ and low gravity designs
9 1 6.71 6.10 2.59 Dmin 3.5 0.43 0.71
10 2 6.10 5.18 3.66 Dmin 3.5 0.95 0.32
11 2 7.32 4.27 4.88 Dmin 3.5 1.10 0.27
Minimum seismic ðDmin Þ and high gravity designs
12 2 7.38 3.66 5.49 Dmin 3.5 0.90 0.33
13 1 6.71 4.27 2.90 Dmin 3.5 0.46 0.66
considered earthquake (MCE) hazard spectral values were equal to 1.5 times these design spec-
tral values.
Following the P695 methodology, the archetype models were designed based on the
following seismic load combinations:
ð1.2 þ 0.2SDS ÞD þ ρQE þ L
ð0.9 − 0.2SDS ÞD þ ρQE

where D is the structural self weight and superimposed dead loads, and L is the design live load
(see Table 2.) QE is the effect of horizontal seismic forces resulting from the base shear, V.
Following P695, the redundancy factor, ρ, was conservatively assumed to be 1.0 in all cases.
The structural design was in accordance with S110. The beams were specified to be
ASTM A653 galvanized cold-formed steel C-section members with a nominal yield stress
ðF y Þ and a ultimate strength ðF u Þ of 345 and 483 N/mm2, respectively. ASTM A500 Grade B
HSS members with F y and F u values of 317 and 483 N/mm2, respectively, were used for the
columns. Tables 2 and 3 summarize the member and connection designs based on the seismic
load combinations in Equations 2 and 3.
DEVELOPMENT OF NONLINEAR STRUCTURAL ARCHETYPE MODEL

The nonlinear analyses were performed by using the software OpenSees (Open System
for Earthquake Engineering Simulation) (PEER 2006). The model shown in Figure 3
included beam and column elements as well as a zero-length rotational spring to simulate
the nonlinear behavior of the bolted beam-to-column moment connections. Figure 4 shows
sample correlation of the bolted moment connection response based on the proposed hyster-
esis rule (Uang et al. 2010). Fiber element was used to model the yielding of the beams and
Table 2. Archetype structure model sizes
Archetype Live load, Dead load,

design ID Beama (mm) Column (mm) L (N/m2) D (N/m2)
1 508CS89 × 3.4 HSS254 × 254 × 6.4 2394 2633
2 406CS89 × 3.4 HSS203 × 203 × 6.4 2394 575
3 406CS89 × 3.4 HSS254 × 254 × 6.4 2394 575
4 508CS89 × 3.4 HSS305 × 305 × 6.4 5985 575
5 508CS89 × 3.4 HSS254 × 254 × 6.4 5985 575
6 406CS89 × 3.4 HSS254 × 254 × 6.4 5985 575
7 508CS89 × 3.4 HSS305 × 305 × 6.4 5985 575
8 508CS89 × 3.4 HSS203 × 203 × 6.4 5985 575
9 305CS89 × 2.7 HSS203 × 203 × 6.4 2394 575
10 406CS89 × 3.4 HSS203 × 203 × 6.4 2394 2633
11 406CS89 × 3.4 HSS254 × 254 × 6.4 2394 2633
12 508CS89 × 3.4 HSS254 × 254 × 6.4 5985 575
13 406CS89 × 3.4 HSS203 × 203 × 6.4 5985 575
a
Double C-sections.
Table 3. Bolt configurations
Design ID a (mm) b (mm) c (mm)

1 76 254 159
2 76 152 108
3 76 152 159
4 76 254 210
5 76 254 159
6 76 152 159
7 76 254 210
8 76 254 108
9 64 76 108
10 76 152 108
11 76 152 159
12 76 254 159
13 76 152 108 25.4-mm-diameter snug-tight high-strength bolts.
Test Simulated
Moment (kN-m)
Moment (KN-m)
150 150
100 100
50 50
0 0
-50 -50
-100 -100
-150 -150
-0.06 -0.04 -0.02 0 0.02 0.04 0.06 -0.06 -0.04 -0.02 0 0.02 0.04 0.06
Slip-Bearing Rotation (rad) Slip-Bearing Rotation (rad)
(a) Specimen 2: Double C-section Beam: 305CS×89×2.7, Column: HSS203×203×6.4, Bolt bearing
plate: 3.4mm
Moment (kN-m)
Moment (KN-m)
150 150
100 100
50 50
0 0
-50 -50
-100 -100
-150 -150
-0.06 -0.04 -0.02 0 0.02 0.04 0.06 -0.06 -0.04 -0.02 0 0.02 0.04 0.06
(b) Specimen 3: Double C-section Beam: 406CS×89×2.7, Column: HSS203×203×6.4, Bolt bearing
plate: N/A
Moment (kN-m)
Moment (KN-m)
200 200
100 100
0 0
-100 -100
-200 -200
-0.06 -0.04 -0.02 0 0.02 0.04 0.06 -0.06 -0.04 -0.02 0 0.02 0.04 0.06
(c) Specimen 9: Double C-section Beam: 508CS×89×3.4, Column: HSS254×254×6.4, Bolt bearing
plate: N/A
Figure 4. Sample correlation of bolted moment connection response.

columns; no buckling was considered in the analysis. The justification for not considering
local buckling in the analysis is presented later.
The expected yield stress ð¼ Ry F y Þ was used to compute the strength for the beam and
column, where Ry is equal to 1.1 and 1.4 for the beam and column, respectively (AISI 2007).
To model the hardening region of the bolted moment connection response, the bearing
strength at bolt holes is a function of the expected tensile strength ð¼ Rt F u Þ. The following
Rt values were used: 1.1 for the beam and 1.3 for the column (AISI 2007).
For either nonlinear static or dynamic analysis, P695 requires that the following gravity
loads be imposed to the model structure:
1.05D þ 0.25L
A 5% damping was assumed for the models. The P-delta effect was also included in the
analysis by including a pin-connected leaning column to the frame.
NONLINEAR STATIC ANALYSIS
DEFINITION OF SYSTEM OVERSTRENGTH AND DUCTILITY CAPACITY

For capacity design, empirical system overstrength factors have been provided in the build-
ing code (ASCE 2005) to allow the designer to calculate the ultimate strength of the building
frame while avoiding the need to perform nonlinear analysis. In this study, nonlinear static (or
pushover) analyses were conducted in accordance with P695 to help validate the behavior of
models and to provide statistical data on the system overstrength, Ω0 , and the ductility capacity,
μc . The system overstrength factor for a given index archetype model, Ω, is defined as follow:
V
Ω ¼ max (5)
V
and the period-based ductility capacity for a given index archetype model, μT , is defined as
follows:
δu
nT ¼ (6)
δy; ef f
See Figure 5 for the definition of δu , δy; ef f , and V max , where V is the design base shear. Note
that P695 assumes the maximum strength of the system is reached before it degrades to 80%
of the peak strength. Nevertheless, the response of a properly designed SBMF is very dif-
ferent from that depicted in Figure 5. Figure 6 shows a typical pushover curve of an SBMF.
When the beams and columns are designed to remain essentially elastic by the capacity
design principles per S110, an SBMF at large story drift will continue to harden due to
the bearing action of the bolt group. Since S110 sets the maximum story drift limit to
0.03 h, the corresponding ductility capacity, μSDL c , can be defined as follows:
0.03h
μSDL
c ¼ (7)
Δy
where h is the story height.

Base Shear
Vmax
0.8Vmax
o δy,eff δu Roof Disp.

×µ T
Figure 5. Idealized nonlinear pushover curve (FEMA 2009).
Base Shear
K
VDBE
×1/R
V0.03h
V
Vy
o ∆y ∆d ∆0.03h Story Drift
Figure 6. General response of SBMF.
It is also difficult to implement the archetype overstrength definition (Equation 5) to

SBMF because, within the story drift range of practical interest, a properly designed
SBMF usually does not reach its peak strength and then degrade. Instead, the lateral strength
of the frame continues to increase with the story drift. Note that S110 provides a design
procedure, which is based on the concept of instantaneous center of rotation of an eccen-
trically loaded bolt group, to calculate the maximum base shear for a given design story
drift. Since this procedure explicitly calculates the seismic load effect with overstrength,
E mh , the empirical system overstrength factor is not needed. Nonetheless, the archetype over-
strength factor, Ω, as defined below is calculated, which corresponds to the maximum story
drift limit.
V 0.03h
Ω¼ (8)
V
where V 0.03h is the base shear at a story drift of 0.03 h.
SYSTEM OVERSTRENGTH AND DUCTILITY CAPACITY

Figure 7 shows some sample pushover analysis results. The pseudo-yielding base shear
V y and the corresponding yield displacement Δy (¼ V y ∕K, where K is the lateral stiffness)
are summarized in Table 4.
From pushover curves, key parameters that are used to calculate the Ω and μSDL
c values are
listed in Table 4. The average value of μSDL
c ranges from 4.42 to 4.99, with an average value
without P-Delta effect with P-Delta effect
150 80
Base Shear (kN)
Base Shear (kN)

VDBE = 95.6 kN 60 VDBE = 51.5 kN
100
40
50
V = 27.3 kN 20 V = 14.7 kN
0 0
0 2 4 6 0 2 4 6
Story Drift Ratio (%) Story Drift Ratio (%)
(a) Archetype 2 (b) Archetype 6
120 120
Base Shear (kN)
Base Shear (kN)

90 90
60 60
VDBE = 44.9 kN
30 30 VDBE = 26.4 kN
V = 12.8 kN
V = 7.52 kN
0 0
0 2 4 6 0 2 4 6
(c) Archetype 10 (d) Archetype 13
Figure 7. Sample nonlinear static analysis results.
of 4.61. The average value of Ω varies from 1.90 to 2.09 for the maximum seismic perfor-
mance groups, and is much higher for the minimum seismic performance groups.
NONLINEAR DYNAMIC ANALYSIS
INCREMENTAL DYNAMIC ANALYSIS

Nonlinear time-history response for each of the archetypes was evaluated for a set of
predefined earthquake ground motions. The P695 procedure for conducting nonlinear
dynamic analyses is based on the concept of incremental dynamic analysis (Vamvatsikos
and Cornell 2002), in which each ground motion is scaled to increase intensities until
the structure reaches a collapse point. The objective of the analyses is to establish the median
collapse capacity and collapse margin ratio for each of the archetypes.
GROUND MOTION AND GLOBAL RESPONSE AT DESIGN BASIS EARTH-

QUAKE
Twenty-two earthquake ground motion record pairs from sites located 10 km or more
from the fault rupture, referred to as the “Far-Field” record set, were used for the collapse
assessment (FEMA 2009). Table 5 summarizes the key information of these ground motion
records.
Table 4. Archetype structural properties
ID h K Vy V V 0.03h Δy Δ0.03h
(m) (kN/m) (kN) (kN) (kN) (mm) (mm) Ω μSDL
c

1 4.88 538 17.7 19.8 38.4 32.7 146 1.94 4.47
2 3.12 1462 28.6 27.3 47.6 19.6 93.7 1.74 4.79
3 4.88 704 20.3 23.2 36.0 29.0 146 1.55 5.05
4 9.14 217 15.1 12.6 30.0 69.9 274 2.38 3.82
mean 1.90 4.53
5 4.88 807 26.5 28.1 57.4 32.8 146 2.04 4.47
6 4.88 475 13.6 14.7 24.4 28.4 146 1.66 5.14
7 7.32 406 18.9 18.6 43.1 46.7 219 2.32 4.70
8 3.35 865 24.2 19.9 46.3 27.9 101 2.33 3.60
mean 2.09 4.48
9 2.59 1277 17.4 6.63 21.6 13.7 77.7 3.26 5.67
10 3.66 954 24.5 12.8 37.1 25.7 110 2.90 4.28
11 4.88 699 20.3 10.9 35.7 29.2 146 3.27 5.01
mean 3.14 4.99
12 5.49 567 23.6 7.21 46.7 41.7 165 6.48 3.95
13 2.90 1165 20.6 7.52 35.2 17.8 86.9 4.69 4.89
mean 5.59 4.42
For each archetype, the maximum considered earthquake (MCE) spectral acceleration,
SMT , corresponding to the fundamental period of the model, T, is summarized in Tables 1 and
6. These MCE values are equal to 1.5 times those of the design basis earthquake (DBE) in
accordance with ASCE 7-05.
Figure 8 shows sample responses of a few archetypes produced by the second component
of Ground Motion 1 (NORTH/MUL279 recorded from the 1994 Northridge earthquake).
Note the slight negative slope of the response of Archetypes 2 and 6 in the plateau (i.e.,
bolt slippage) region due to P-delta effect. But this undesirable effect was overcome by
the significant hardening of the bolted moment connection response at higher drift levels.
INCREMENTAL DYNAMIC ANALYSIS

Sample results from the incremental dynamic analysis are shown in Figure 9. The ground
motion intensity, ST , is defined as the spectral acceleration at the fundamental period of the
structure. Every point along each curve was produced from a single nonlinear dynamic ana-
lysis of one archetype model subjected to one ground motion record scaled to one intensity
level. Repeating the same process, each curve was produced by scaling the same ground
motion record to the full range of intensity values. For a given ground motion intensity,
the scatter of the story drift ratio represents the differences in the response of each archetype
Table 5. Earthquake event and PEER-NGA data for the far-field record set (FEMA 2009)
Earthquake Peak motion*
GM PGA PGV Normalized

ID M Year Name Component 1 Component 2 (g) (cm/sec) factor
1 6.7 1994 Northridge NORTH/ NORTH/ 0.52 63 0.65
MUL009 MUL279
2 6.7 1994 Northridge NORTH/ NORTH/ 0.48 45 0.83
LOS000 LOS270
3 7.1 1999 Düzce, DUZCE/ DUZCE/ 0.82 62 0.63
Turkey BOL000 BOL090
4 7.1 1999 Hector HECTOR/ HECTOR/ 0.34 42 1.09
Mine HEC000 HEC090
5 6.5 1979 Imperial IMPVAL/ IMPVAL/ 0.35 33 1.31
Valley H-DLT262 H-DLT352
6 6.5 1979 Imperial IMPVAL/ IMPVAL/ 0.38 42 1.01
Valley H-E11140 H-E11230
7 6.9 1995 Kobe, KOBE/ KOBE/ 0.51 37 1.03
Japan NIS000 NIS090
8 6.9 1995 Kobe, KOBE/ KOBE/ 0.24 38 1.10
Japan SHI000 SHI090
9 7.5 1999 Kocaeli, KOCAELI/ KOCAELI/ 0.36 59 0.69
Turkey DZC180 DZC270
10 7.5 1999 Kocaeli, KOCAELI/ KOCAELI/ 0.22 40 1.36
Turkey ARC000 ARC090
11 7.3 1992 Landers LANDERS/ LANDERS/ 0.24 52 0.99
YER279 YER360
12 7.3 1992 Landers LANDERS/ LANDERS/ 0.42 42 1.15
CLW-LN CLW-TR
13 6.9 1989 Loma LOMA/ LOMA/ 0.53 35 1.09
Prieta CAP000 CAP090
14 6.9 1989 Loma LOMA/ LOMA/ 0.56 45 0.88
Prieta G03000 G03090
15 7.4 1990 Manjil, MANJIL/ MANJIL/ 0.51 54 0.79
Iran ABBAR—L ABBAR—T
16 6.5 1987 Superstition SUPERST/ SUPERST/ 0.36 46 0.87
Hills B-ICC000 B-ICC090
17 6.5 1987 Superstition SUPERST/ SUPERST/ 0.45 36 1.17
Hills B-POE270 B- POE360
18 7.0 1992 Cape CAPEMEND/ CAPEMEND/ 0.55 44 0.82
Mendocino RIO270 RIO270
19 7.6 1999 Chi-Chi, CHICHI/ CHICHI/ 0.44 115 0.41
Taiwan CHY101-E CHY101-N
(continued )
Earthquake Peak motion*
GM PGA PGV Normalized

ID M Year Name Component 1 Component 2 (g) (cm/sec) factor
20 7.6 1999 Chi-Chi, CHICHI/ CHICHI/ 0.51 39 0.96
Taiwan TCU045-E TCU045-N
21 6.6 1971 San SFERN/ SFERN/ 0.21 19 2.10
Fernando PEL090 PEL180
22 6.5 1976 Friuli, FRIULIA/ FRIULIA/ 0.35 31 1.44
Italy A-TMZ000 A-TMZ270
*The larger value of two components is reported.
model when subjected to different ground motions with different frequency characteristics.
According to P695, such record-to-record variability is fairly constant for different structural
models and ground motion record sets. As such, only the median collapse intensity needs to
be computed, as explained in the following.
EVALUATION OF COLLAPSE MARGIN AND ACCEPTANCE CRITERIA
MEDIAN COLLAPSE INTENSITY AND COLLAPSE MARGIN RATIO

For each archetype model, the spectral acceleration at collapse, SCT , was computed for
each of the 44 ground motions, from which the median collapse intensity, S^ CT , was com-
puted. S^ CT was obtained by scaling all the records in the record set to the MCE intensity, SMT ,
and then by increasing the intensity until one-half of the scaled ground motion records caused
collapse. The ratio between the median collapse intensity, S^ CT , and the MCE intensity, SMT ,
is the collapse margin ratio (CMR).
S^
CMR ¼ CT (9)
SMT
The value of S^ CT depends on how the collapse is defined. Using a very large story drift
ratio to define S^ CT gives an upper bound (i.e., nonconservative) estimate of CMR. On the
other hand, using the code-specified story drift limits (ASCE 2005) to define S^ CT is too con-
servative because the actual story drift capacity is generally higher. To establish a realistic
story drift level to define collapse, the story drift capacity of nine cyclically tested, full-scale
beam-column subassemblies were evaluated (see Figure 10). The test matrix included beam
and column sizes with varying width-to-thickness (w/t) ratios (Uang et al. 2010). All except
specimen 3 experienced local buckling at higher drift levels. Specimens 4, 8, and 9 had higher
w∕t ratios, which resulted in lower story drift capacities due to severe local buckling. Based
on the test results of these three specimens, limits on width-to-thickness ratios were also
established and incorporated in S110. The average story drift capacity of all nine specimens
was about 0.06 radians. Since the beam and column of all the archetypes satisfied the S110
seismic compactness requirements, it was judged that a story drift capacity of 0.06 radians
was a reasonable value to define the lower bound collapse limit state.
150 6

100 4
Base Shear (kN)
50 2
0 0
-50 -2
-100 -4
-150 -6
-6 -4 -2 0 2 4 6 0 5 10 15 20 25 30 35
Story Drift Ratio (%) Time (sec)
(a) Archetype 2
90 6

60 4
Base Shear (kN)
30 2
0 0
-30 -2
-60 -4
-90 -6
-6 -4 -2 0 2 4 6 0 5 10 15 20 25 30 35
(b) Archetype 6
120 6
80 4
Base Shear (kN)
40 2
0 0
-40 -2
-80 -4
-120 -6
-6 -4 -2 0 2 4 6 0 5 10 15 20 25 30 35
(c) Archetype 10
120 6
80 4
Base Shear (kN)
40 2
0 0
-40 -2
-80 -4
-120 -6
-6 -4 -2 0 2 4 6 0 5 10 15 20 25 30 35
(d) Archetype 13
Figure 8. Sample nonlinear dynamic analysis results (GM1, Component 2, DBE level).
The ultimate story drift capacity is also limited by the bearing deformation in the energy-
dissipating bolt group. Based on the instantaneous center of rotation concept, S110 provides a
procedure to calculate this story drift capacity. The calculated capacity ranged from 0.08 to
8 6
6
4
ST (g)
ST (g)
4
2
2
SMT=1.50g SMT=1.24g
0 0
0 2 4 6 0 2 4 6
Maximum Story Drift Ratio (%) Maximum Story Drift Ratio (%)
3
8
2 6
ST (g)
ST (g)
4
1
2
SMT=0.32g SMT=0.66g
0 0
0 2 4 6 0 2 4 6
Maximum Story Drift Ratio (%) Maximum Story Drift Ratio (%)
Figure 9. Sample incremental dynamic analysis results (44 ground motion records).
No strength degradation
Specimen with Specimen with

Beam Buckling Column Buckling
w/t =109 w/t =147 w/t = 31 w/t = 40
0.12
0.10
Drift Capacity (rad.)
0.08
0.06
0.04
0.02
0.00
3 1 2 4 5 6 7 8 9
Specimen No.
Figure 10. Story drift capacity from full-scale testing.

0.13 radians for the archetypes, larger than that (= 0.06 radians) assumed to define collapse in
this study.
Figure 11 shows the sample IDA analysis results for calculating the lower bound S^ CT and
CMR values. These lower bound values for all the archetypes are summarized in Table 6.
ADJUSTED COLLAPSE MARGIN RATIO

A characteristic of ground motions that can affect collapse capacity is the spectral shape
(Backer and Cornell 2006). For rare ground motions in California, such as MCE ground
motions, the spectral shape is much different than the shape of a design spectrum in
ASCE 7. That is, the shape of the spectrum of rare ground motions is peaked at the period
of interest, and drops off more rapidly at periods that are longer or shorter than the period of
interest. As the structure’s period elongates in the inelastic response, this characteristic makes
the rare ground motions less damaging from the collapse capacity point of view. To remove
the conservatism from using the ground motion record set in Table 5, P695 uses a spectral
shape factor, SSF, to compute the modified collapse margin ratio, ACMR:
ACMR ¼ SSF × CMR
SSF is a function of the fundamental period, T, the ductility capacity, μc , and the seismic
criteria used for the design of the index archetype as defined by seismic design category.
Values of μc listed in Table 6 are consistent with the definition of using a story drift of
0.06 radians to define the lower bound collapse limit state. Values of SSF and ACMR
for each archetype are also listed in Table 6. The mean ACMR value for each performance
group is also provided.

Calculated values of ACMR need to be compared with acceptable values. P695 defines
acceptable ACMR values in terms of the total system uncertainty and the established limits on
collapse probability due to MCE ground motions. For the latter, P695 suggests that, on aver-
age, the probability of collapse for each performance group be limited to 10%:
ACMRi ≥ ACMR10%
But P695 also recognizes that potential “outliers” with a performance group could have
collapse probabilities that exceed this value. Therefore, a limit of twice that value, or 20%, is
suggested as an evaluation criterion for each archetype:
ACMRi ≥ ACMR20%
In Equations 11 and 12, ACMRi and ACMRi represent the individual and average ACMR
values. ACMR10% and ACMR20% are the acceptable values with a collapse probability of 10%
and 20%, respectively. Table 7 shows sample values of ACMR10% and ACMR20% . Note that
these values are a function of the total system collapse uncertainty, βTOT . This parameter
reflects the fact that systems that have more robust design requirements, more comprehensive
test data, and more detailed nonlinear analysis models have less collapse uncertainty. Accord-
ing to P695, quality ratings (“Superior,” “Good,” “Fair,” and “Poor”) for design requirements,
1
8 SMT=
CMR=SCT/SMT=1.59
Collapse Probability
0.8 1.50g
6
0.6
ST (g)
4
0.4
SCT=2.38g SCT = 2.38g
2 0.2
SMT=1.50g
CMR=SCT/SMT=1.59
0 0
0 2 4 6 0 2 4 6 8
Maximum Story Drift Ratio (%) Collapse Spectral Acceleration, SCT (g)
(a) Archetype 2
6 1
SMT=
CMR=SCT/SMT=1.57
0.8 1.24g
4 0.6
ST (g)
0.4
2 SCT=1.95g SCT = 1.95g
SMT=1.24g 0.2
CMR=SCT/SMT=1.57
0 0
0 2 4 6 0 2 4 6
(b) Archetype 6
1
3 SMT=
CMR=SCT/SMT=2.63
0.8 0.32g
2 0.6
ST (g)
0.4
1 SCT = 0.84g
SCT=0.84g
0.2
SMT=0.32g CMR=SCT/SMT=2.63
0 0
0 2 4 6 0 1 2 3
(c) Archetype 10
1
SMT=
8 CMR=SCT/SMT=4.03
0.8 0.66g
6
0.6
ST (g)
4 0.4
SCT=2.66g SCT = 2.66g
2 0.2
CMR=SCT/SMT=4.03
SMT=0.66g
0 0
0 2 4 6 0 2 4 6 8
(d) Archetype 13
Figure 11. Sample IDA results and collapse probability (lower bound).
test data, and nonlinear models are then translated into quantitative values of uncertainties as
follow: β ¼ 0.10 for “Superior,” β ¼ 0.20 for “Good,” β ¼ 0.35 for “Fair,” and β ¼ 0.50 for
“Poor.” In this study, the design requirement (i.e., S110) was rated as “Good” and βDR ¼ 0.20.
Since cyclic testing of nine full-scale beam-column subassemblies showed very consistent
slip-bearing behavior in the bolt group, the quality of test data was rated as “Superior”
with βTD ¼ 0.10. Applying the concept of instantaneous center of rotation has provided a
very good correlation to simulate the slip-bearing action in the bolt group (e.g., see Figure 4).
Thus, the modeling uncertainty was rated as “Good” with βTD ¼ 0.20. Together with a value of
0.4 fixed in P695 for βRTR to account for the record-to-record uncertainty, the total system
collapse uncertainty is
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
βTOT ¼ β2RTR þ β2DR þ β2TD þ β2MDL ¼ 0.50
With the above information, P695 assigns a value of 0.50 for βTOT . From Table 7, the
values of ACMR10% and ACMR20% are 1.90 and 1.52, respectively. A comparison of the
calculated and acceptable ACMR values in Table 6 shows that both Equations 11 and 12
Table 6. Summary of collapse results for archetype designs, final collapse margins, and
acceptance criteria
IDA analysis Computed collapse Acceptance

results margin check
Archetype
design ID SMT ðTÞ; S^CT (T), Accept. Pass/
number (g) (g) CMR μc SSF ACMR ACMR fail
1 1.04 1.35 1.30 8.9 1.44 1.82 1.52 Pass
2 1.50 2.38 1.59 9.6 1.35 2.15 1.52 Pass
3 1.16 1.67 1.44 10.1 1.41 2.03 1.52 Pass
4 0.66 1.17 1.77 7.9 1.58 2.80 1.52 Pass
mean 2.20 1.90 Pass
5 1.10 1.54 1.40 8.9 1.43 2.00 1.52 Pass
6 1.24 1.95 1.57 10.3 1.40 2.20 1.52 Pass
7 0.84 1.31 1.56 9.4 1.49 2.32 1.52 Pass
8 1.50 2.35 1.57 7.2 1.34 2.10 1.52 Pass
mean 2.16 1.90 Pass
9 0.71 2.29 3.23 11.4 1.14 3.68 1.52 Pass
10 0.32 0.84 2.63 8.5 1.24 3.26 1.52 Pass
11 0.27 0.85 3.15 10.1 1.28 4.03 1.52 Pass
mean 3.66 1.90 Pass
12 0.33 1.43 4.33 7.9 1.23 5.33 1.52 Pass
13 0.66 2.66 4.03 9.8 1.15 4.63 1.52 Pass
mean 4.98 1.90 Pass
Table 7. Acceptable value of adjusted collapse margin ratio
Total system collapse uncertainty βTOT ACMR10% ACMR20%

0.45 1.78 1.46
0.50 1.90 1.52
0.55 2.02 1.59
are satisfied. Therefore, the proposed seismic performance factors can provide a sufficient
margin against collapse due to maximum considered earthquake ground motions.
EFFECT OF BOLT TENSION VARIATION
BACKGROUND
According to S110, high-strength bolts used in an SBMF are to be in a snug-tight con-
dition. Based on the cyclic test results of nine full-scale beam-column subassemblies, it was
shown that using a 75.6 kN bolt tension for the 25.4-mm-diameter high-strength bolts with a
slip coefficient k ¼ 0.19 would provide a satisfactory correlation with the test results in the
slip range (Uang et al. 2010). To evaluate the effect of bolt overtightening or undertightening
on the seismic response of SBMF, including story drift ratio and maximum moment in the
bolted connection, a sensitivity study was conducted. The effect of the variation of bolt ten-
sion on the collapse margin was also evaluated.
STRUCTURAL MODEL
Four archetypes (ID numbers 2, 4, 6, and 7 from Table 1) were considered in the sensi-
tivity study. For each archetype, three models were considered. The standard (ST) model
assumes that the bolts were properly tightened to the snug-tight condition such that the
bolt tension is the same as that assumed ðT ¼ 75.6 kNÞ. The overtightened (OT) model
assumes that all the bolts were overtightened such that the bolt tension is 50% larger
(i.e., T ¼ 113 kN). The undertightened (UT) model assumes that the bolt tension is 70%
of that assumed in the standard model.
NONLINEAR STATIC AND DYNAMIC ANALYSES

A comparison of the static pushover analysis results can be made in Figure 12. The var-
iation in the bolt tension mainly affects the response in the slip range, but not too much in the
bearing range. A comparison of the sample global responses of archetype 6 for three bolted
tension models is shown in Figure 13. Figure 14 summarizes the maximum story drift ratio
produced by all 44 ground motions. Figure 15 shows a summary of the base shear. The
average value and the standard deviation for both the story drift ratio and base shear are
listed in Table 8.
Even with the assumption that the bolt tension would be increased by 50% due to over-
tightening, it is observed from Table 8 that both the story drift ratio and base shear (or
150 80
Base Shear (kN)
Base Shear (kN)

60
100 OT OT
ST 40 ST
UT UT
50
20
0 0
0 2 4 6 0 2 4 6
120
80
Base Shear (kN)
Base Shear (kN)

60 90
OT OT
40 ST 60 ST
UT UT
20 30
0 0
0 2 4 6 0 2 4 6
Figure 12. Sensitivity of bolt tension force on pushover analysis.
90 90
60 60
Base Shear (kN)
Base Shear (kN)
30 30
0 0
-30 -30
-60 -60
-90 -90
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
(a) UT model (b) ST model
90
60
Base Shear (kN)
30
0
-30
-60
-90
-6 -4 -2 0 2 4 6
(c) OT model
Figure 13. Global response of archetype 6 (GM5, Component 1, DBE level).

Maximum Story Drift Ratio (%)

8 Comp. 1 8 Comp. 1
Comp. 2 Comp. 2
6 6
4 4
2 2
0 0
10
11
12
13
14
15
16
17
18
19
20
21
22
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
1
2
3
4
5
6
7
8
9
Ground Motion No. Ground Motion No.

8 Comp. 1
Comp. 2
6
0
10
11
12
13
14
15
16
17
18
19
20
21
22
1
2
3
4
5
6
7
8
9
Ground Motion No.

(c) OT model
Figure 14. Maximum story drift at DBE level (Archetype 6).
70 Comp. 1 70 Comp. 1
60 Comp. 2 60 Comp. 2
Base Shear (kN)
Base Shear (kN)
50 50
40 40
30 30
20 20
10 10
0 0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Ground Motion No. Ground Motion No.

70 Comp. 1
60 Comp. 2
Base Shear (kN)
50
40
30
20
10
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Ground Motion No.

(c) OT model
Figure 15. Maximum base shear at DBE level (Archetype 6).
moment in the bolted connection) are not very sensitive to the variation of bolt tension. In this
case, overtightening reduces the story drift ratio by 2.7% to 3.4%, although the maximum
moment in the bolted connection is increased by 4.4% to 18%. If the bolt tension was reduced
by 30% due to undertightening, it is also observed from Table 8 that both the story drift ratio
Table 8. Bolt tension effect on the maximum base shear and story drift
Maximum Maximum
Height Bolt tension*, base shear story drift
Model (m) T (kN) (kN) (%)
UT 48.9 [27.1] 3.2 [1.2]
2 3.12 ST 48.9 [23.1] 2.9 [1.2]
OT 54.7 [18.2] 2.7 [1.2]
UT 23.6 [11.1] 2.8 [0.97]
4 9.14 ST 24.5 [9.3] 2.7 [0.92]
OT 28.9 [10.2] 2.7 [1.0]

UT 23.1 [13.3] 3.2 [1.1]
6 4.88 ST 24.5 [12.9] 3.0 [1.2]
OT 28.0 [10.7] 2.9 [1.1]
UT 37.8 [20.0] 3.0 [1.1]

ST 39.1 [19.1] 2.9 [1.2]
7 7.32
OT 40.9 [16.9] 2.8 [1.1]
*UT: Undertightened T ¼ 52.9 kN; ST: Standard T ¼ 75.6 kN; OT: Overtightened T ¼ 113 kN, []: standard deviation
1 1
SMT= SMT=
0.8 1.50g
0.8 0.66g
0.6 0.6
0.4 0.4
UT UT
0.2 ST 0.2 ST
SCT SCT
OT OT
0 0
0 2 4 6 8 0 1 2 3 4 5
Collapse Spectral Acceleration, SCT (g) Collapse Spectral Acceleration, SCT (g)
1 1
SMT= SMT=
1.24g 0.8 0.84g

0.8
0.6 0.6
0.4 0.4
UT UT
0.2 SCT ST 0.2 ST
SCT
OT OT
0 0
0 2 4 6 0 1 2 3 4 5
Collapse Spectral Acceleration, SCT (g) Collapse Spectral Acceleration, SCT (g)
Figure 16. Sensitivity of bolt tension to collapse probability.

Table 9. Summary of collapse results for archetype designs, final collapse margins, and
acceptance criteria (sensitivity of bolt tension)
IDA analysis results Computed collapse margin Acceptance check

Archetype
design ID SMT ðTÞ; S^CT (T), Accept. Pass/
number ðgÞ (g) CMR μc SSF ACMR ACMR fail
UT 2.17 1.45 13.7 1.35 1.96 1.52 Pass
2 ST 1.50 2.38 1.59 9.6 1.35 2.15 1.52 Pass
OT 2.64 1.76 6.4 1.31 2.31 1.52 Pass
UT 1.12 1.70 11.2 1.58 2.69 1.52 Pass
4 ST 0.66 1.17 1.77 7.9 1.58 2.80 1.52 Pass
OT 1.22 1.85 5.2 1.46 2.70 1.52 Pass
UT 1.77 1.43 14.7 1.40 2.00 1.52 Pass
6 ST 1.24 1.95 1.57 10.3 1.40 2.20 1.52 Pass
OT 2.05 1.65 6.9 1.37 2.26 1.52 Pass
UT 1.29 1.54 13.4 1.49 2.29 1.52 Pass
7 0.84
ST 1.31 1.56 9.4 1.49 2.32 1.52 Pass
OT 1.38 1.64 6.3 1.41 2.31 1.52 Pass
UT: Undertightened T ¼ 52.9 kN; ST: Standard T ¼ 75.6 kN; OT: Overtightened T ¼ 113 kN
and base shear are not very sensitive to the variation of bolt tension. In this case, undertigh-
tening increases the story drift by 3.3% to 10%, and the maximum moment in the bolted
connection is reduced by up to 5.7%. Therefore, this sensitivity study concludes that the
overtightening or undertightening is not a major concern.

Figure 16 shows the collapse probability obtained from the IDA. A summary of the lower
bound collapse analysis results of four archetype designs is shown in Table 9. Again, the
acceptance criteria (Equation 12) were met.
CONCLUSIONS
Following the FEMA P695 procedure, a total of 13 special bolted moment frame arche-
type designs representing two seismicity and two gravity load levels were designed and ana-
lyzed. Expecting inelastic action to occur in the bolted moment connections only, a nonlinear
spring element was developed and used in the nonlinear static and dynamic analyses. Beam
and column modeling allowed for yielding, but buckling was not modeled in the analyses.
Therefore, a lower bound approach was used to compute the collapse margins, where col-
lapse was defined as that when the story drift ratio reached 6% of the story height.
The lower bound approach showed that all 13 individual archetype designs and four
performance groups met the P695 acceptance criteria. It was concluded that the seismic per-
formance factors (R ¼ 3.5 and Cd ¼ 3.0) proposed were acceptable and would provide a
satisfactory margin against collapse under the maximum considered earthquake ground
motions.
Additionally, a sensitivity study of the bolt tension force on story drift ratio, maximum
moment in the bolted connection, and collapse margins were studied. Overtightening reduced
the story drift ratio by up to 3.4%, although the maximum moment in the bolted connection
was increased by 4.4% to 18%. On the other hand, undertightening increased the story drift
ratio by 3.3% to 10%, and the maximum moment in the bolted connection was reduced by up
to 5.7%. Evaluating the collapse margin, the effect of bolt tension was not significant and
again the acceptance criteria were met. Therefore, overtightening or undertightening of the
bolts was not a major concern.
In the process of code adoption, the proposed R value ð¼ 3.5Þ was incorporated into
Table 12.2-1 of ASCE 7-10 (ASCE 2010). Although this P695 study demonstrated that
a value of C d ð¼ 3.0Þ lower than the R value was acceptable due to the significant bearing
hardening effect in the bolt groups, the code committee opted to set C d equal to R, resulting in
a C d value of 3.5 in the table. Furthermore, since S110 provides a design procedure to cal-
culate the ultimate strength of the structural system and, therefore, the use of Ω0 would be
unnecessary, for consistency with the format for other systems, the code committee decided
to assign a value of 3.0 to Ω0 . But a footnote is provided such that the designer still can use
S110 to explicitly compute the amplified seismic effects for capacity design instead of using
the empirical Ω0 factor.
ACKNOWLEDGMENTS
Funding for this project was provided by the American Iron and Steel Institute. The
authors would like to acknowledge Mr. Ken Wood for designing the archetypes.
REFERENCES
American Iron and Steel Institute (AISI), 2007. Standard for Seismic Design of Cold-Formed
Steel Structural Systems–Special Bolted Moment Frames, AISI S110, Washington, D.C.
American Society of Civil Engineering (ASCE), 2005 and 2010. Minimum Design Loads for
Buildings and Other Structures, ASCE/SEI 7-05 and ASCE/SEI 7-10, Reston, VA.
Backer, J. W., Cornell, C. A., 2005. A vector-valued ground motion intensity measure consisting
of spectral acceleration and epsilon, Earthquake Engineering and Structural Dynamics 34,
1193–1217.
Crawford, S. F., and Kulak, G. L., 1971. Eccentrically loaded bolted connections, Journal of the
Structural Division 97, 765–783.
Federal Emergency Management Agency (FEMA), 2009. Qualification of building seismic per-
formance factors, FEMA P695, Washington, D.C.
Pacific Earthquake Engineering Center (PEER), 2006. Open System for Earthquake Engineering
Simulation (Opensees), University of California, Berkeley.
Sato, A., and Uang, C. M., 2009. Seismic design procedure development for cold-formed steel–
special bolted moment frames, Journal of Constructional Steel Research 65, 860–868.
Sato, A., and Uang, C. M., 2010. Seismic performance factors for cold-formed steel special bolted
moment frames, Journal of Structural Engineering 136, 961–967.
Uang, C. M., Sato, A., Hong, J. K., and Wood, K., 2010. Cyclic testing and modeling of cold-
formed steel special bolted moment frame connections. Journal of Structural Engineering
136, 953–960.
Vamvatsikos, D., and Cornell, C. A., 2002. Incremental dynamic analysis, Earthquake Engineer-
ing and Structural Dynamics 31, 491–514.
(Received 18 November 2010; accepted 19 February 2012)
Simulation of Reinforced Concrete Frames
with Nonductile Beam-Column Joints
Sangjoon Parka) M.EERI, and Khalid M. Mosalama) M.EERI
The accurate prediction of shear strength and flexibility of beam-column

joints without transverse reinforcement is essential to assess the seismic perfor-
mance of nonductile reinforced concrete (RC) buildings characterized by having
such unreinforced beam-column joints. In this study, a multilinear backbone
curve to represent the moment-rotation relationship of an unreinforced corner
beam-column joint is proposed. The modeling parameters of the backbone
curve are estimated based on experimental results of four corner joint specimens
recently tested by the authors. Furthermore, the proposed backbone curve is mod-
ified to be applicable to interior and roof beam-column joints. These backbone
curves are validated by accurate reproduction of the force-drift responses of the
four corner joint specimens and eight other exterior and interior joint specimens
from literature. Using these backbone curves, nonlinear dynamic analyses are
performed on three hypothetical building frames. The analytical results demon-
strate the importance of joint flexibility for seismic assessment of nonductile RC
buildings. [DOI: 10.1193/1.4000100]
INTRODUCTION
Many reinforced concrete (RC) buildings designed prior to the 1970s still exist in the
western United States and in other seismically active regions worldwide. These existing
older RC buildings were usually designed for gravity load only or with little consideration
of seismic forces and detailing, which led to insufficient shear reinforcement in columns and
beam-column joint regions. Accordingly, these existing older RC buildings are considered
vulnerable to earthquake excitations mainly due to shear failure of columns and beam-
column joints. By equilibrium, shear demands on columns and beam-column joints are cor-
related and thus the beam-column joint shear failure can precede the column shear failure
depending on the structural design details of RC buildings. In past earthquakes, failure of
beam-column joints without transverse reinforcement in the joint region, referred to hereafter
as unreinforced joints, was demonstrated (Uang et al. 1999, Sezen et al. 2000, Li et al. 2008,
Günay and Mosalam 2010). Besides the joint shear failure due to limited strength, large non-
linear deformation of such joints can significantly change the lateral response of RC framed
structures. For example, joint deformation results in increase of interstory drift, which may
lead to column shear and axial failures (Elwood and Moehle 2003).
In spite of the important role of unreinforced joints for the seismic evaluation of non-
ductile existing RC buildings, few accurate shear strength and moment-rotation relationships
(referred to hereafter as backbone curve) are available. The recommendations of ASCE/SEI
a)
Dept. of Civil and Environmental Engineering, University of California, Berkeley 94720
233
234 S. PARK AND K. M. MOSALAM
41-06 (2007), in short ASCE 41, are widely adopted to estimate the shear strength and defor-
mation capacity for unreinforced joints. These recommendations have not yet been experi-
mentally evaluated. Park (2010) recently tested four full-scale unreinforced corner joint
specimens with different ratios of beam-to-column cross-sectional heights and beam long-
itudinal reinforcement ratios. It is worth mentioning that the experimental program in Park
(2010) focused on the seismic performance of unreinforced corner joints. Therefore, the
beams and columns were designed with proper detailing instead of accounting for the
whole spectrum of poor seismic details found in older RC buildings, such as widely spaced
column hoops and lap splices. The measured joint responses from these tests provided bench-
mark data to develop a backbone curve of unreinforced corner joints as described in this
paper. For strength prediction in the developed backbone curve, a simplified version of
an analytical model proposed by Park and Mosalam (2012a) is adopted in this study.
A backbone curve is proposed in this study to simulate the behavior of unreinforced
corner (and also exterior) joints under earthquake-type loading. For validation of the pro-
posed backbone curve, the four corner joints tested by the authors and briefly discussed
in this paper and eight other tests from the literature (four exterior joints from Wong
(2005) and four interior joints from Alire (2002)) are simulated using OpenSees (2010).
In addition, nonlinear dynamic analyses are performed for three hypothetical RC building
frames using the developed backbone curves to preliminarily investigate the contribution of
joint flexibility to the lateral response of nonductile RC buildings.
The scope of this study is limited to the behavior of unreinforced joints without addressing
failures of columns and beams adjacent to these joints. For this purpose, this study considers
strong column/weak beam configuration designed with deformed reinforcing bars and anchorage
details of the beam longitudinal top and bottom bars bent into the joint region with 90° standard
hooks. It is worth mentioning that strong column/weak beam configuration is found in some RC
buildings built prior to the 1970s (Moiser 2000). The focus of this paper is mainly on the devel-
opment of a more accurate backbone curve that represents a monotonic envelope of joint shear
force and deformation response than that proposed in the ASCE 41. In building frame simulations,
the developed backbone curve can be augmented with available cyclic strength degradation mod-
els (Lowes and Altoontash 2003, Mitra and Lowes 2007). Developing rules for strength degrada-
tion due to cyclic loading and formulating accurate and practical damage model for the studied
joints are out of the scope of this study, and more research is needed in that regard.
JOINT SHEAR STRENGTH PREDICTION

Most current standards and codes—for example, ACI 318-11 and ACI/ASCE 352-02—have
recommendations for shear strength of seismically detailed reinforced joints, while recommen-
dations for shear strengths of unreinforced joints are proposed in ASCE 41 for seismic rehabi-
litation of existing buildings based on the prestandards of FEMA 273 and FEMA 356 (BSSC
1997 and 2000). According to ASCE 41, the nominal joint shear strength is defined as follows,
pffiffiffiffi
V n ¼ γ n f c0 bj h c
pffiffiffiffi
where γ n is the joint shear strength as a multiple of f c0 , f c0 is the concrete compressive
strength, h c is the column cross-sectional height, and bj is the effective joint width,
bj ¼ ðbb þ bc Þ∕2. Note that bb and bc are the respective widths of the beam and column
SIMULATION OF REINFORCED CONCRETE FRAMES WITH NONDUCTILE BEAM-COLUMN JOINTS 235
cross sections. For unreinforced exterior joints without lateral beam, the joint shear strength is
γ n ¼ 6 psi0.5 ½0.5 MPa0.5 &, which is also applicable to corner joints. Priestley (1997) pro-
posed a joint shear strength model basedpffiffiffiffion an experimentally
pffiffiffiffi determined principal tensile
stress limit of the joint panel, σ 1 ¼ 5 f c0 psi ½0.42 f c0 MPa&. The joint shear strength V n
corresponding to the principal tensile stress limit is given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
" #
P
V n ¼ Ag σ 1 þ σ 1
2
(2)
Ag
where P is the column axial load (compression is positive), and Ag is the gross area of the
column cross section. Park and Mosalam (2012b) indicated that the shear strength of the
unreinforced exterior and corner joints is affected by two main parameters: (1) joint aspect
ratio, which is defined as the ratio of beam-to-column cross-sectional heights, h b∕h c , and (2)
beam longitudinal tension reinforcement ratio, which is directly related to the joint shear
demand and bond resistance around the beam bars. The effect of the joint aspect ratio
on the joint shear strength can be explained by the strut and tie approach where a steeper
diagonal strut is developed in the high aspect ratio of a joint region and this steeper strut is
less effective to equilibrate the joint horizontal shear force. Two shear strength models have
been developed reflecting the effect of the two main parameters using semi-empirical and
analytical approaches, and these two models are simplified for practicality into a third model
(Park and Mosalam 2012a) and expressed as follows:
$ %
pffiffiffiffi cosθ
V n ¼ k 12 f c bj h c
0 ; θ ¼ tan−1 ðh b∕h c Þ; psi units (3)
cosðπ∕4Þ
where k is a strength factor to account for the effect of the beam longitudinal reinforcement ratio
as defined in Figure 1a. The factor k is bounded between 0.4 and 1.0 such that the joint shear
strength is, for example, 4 < γ n ≤ 12 for a joint aspect ratio h b∕h c ¼ 1.0, that is, θ ¼ π∕4. The
lower bound of 4 represents the typical tensile strength of concrete and the upper bound of 12
represents the shear strength of reinforced joints as recommended in ACI-ASCE 352-02. This
model assumes that the shear strength of the unreinforced joints cannot exceed that of similar
reinforced joints. Similar bounds of the joint shear strength are observed in the evaluation of
many previous tests as documented in Park and Mosalam (2012b). A joint shear index, SI j , in
the abscissa of Figure 1a represents the joint shear demand at the onset of yielding of the beam
longitudinal tension bars and is expressed as follows:
" #" #
As f y hb
SI j ¼ pffiffiffiffi 1 − 0.85 (4)
bj h c f c0 H
where As and f y are, respectively, the area and yield stress of the beam tensile longitudinal
reinforcement, and H is the height between the upper and lower column inflection points.
As f y is the horizontal tensile force induced by the beam longitudinal tension bars at yield, and
0.85 h b∕H is an approximate ratio of the column shear force to As f y and is obtained by global
equilibrium and assuming a constant internal moment arm of the beam cross section. The
model showed accurate predictions of the shear strength of previous tests from literature as
shown in Figure 1b. The shear strength of eight specimens described in the subsequent sec-
tion is predicted by three different models and their predictions are compared to the test
300
300
V
V n,test (kip)
[kip]
jh,test
k
250
250
1.0 200
200
150
150
100
100
0 .4 Mean
Mean ==0.97
0.97
COV = 0.13
COV = 0.13
50
50
SIj
cos θ cos θ 00
Vn,model
V (kip)
[kip]
jh,model
4 12
cos π 4 cos π 4 00 50
50 100
100 150
150 200
200 250
250 300
300
(a) Consideration of beam reinforcement ratio (b) Comparison of measured to calculated strength
Figure 1. Joint shear strength model proposed by Park and Mosalam (2012a): (a) consideration
of beam reinforcement ratio; (b) comparison of measured to calculated strength.
Table 1. Comparison of test results with model predictions
Method of shear 3-D Corner Joint Specimens 2-D Exterior Joint Specimens
strength prediction (Park 2010) (Wong 2005)
[kips]
SP1 SP2 SP3 SP4 BS-L-300 JA-NN03 BS-L-450 BS-L-600
Test result 155.7 228.7 131.3 167.6 113.5 68.4 70.9 63.8
109.9 109.1 110.2 115.5 54.9 63.0 56.8 65.7
ASCE 41
(1.42) (2.10) (1.05) (1.45) (2.07) (1.09) (1.25) (0.97)
142.4 184.3 158.5 171.7 86.5 68.5 81.0 90.4
Priestley 1997
(1.09) (1.24) (0.83) (0.98) (1.31) (1.00) (0.88) (0.71)
Park and Mosalam 159.4 218.2 142.6 168.1 102.0 69.5 82.0 71.8
2012a (0.98) (1.05) (0.92) (1.00) (1.11) (0.98) (0.86) (0.89)
Note: Number in parenthesis is the ratio between the test result and the model prediction.
results in Table 1. The comparison shows that the simplified model by Park and Mosalam
(2012a) is sufficiently accurate for unreinforced exterior and corner joints and thus is adopted
in the simulations of this paper.
EXPERIMENTAL INVESTIGATION
DESCRIPTION OF EXPERIMENTS
Four full-scale three-dimensional (3-D) unreinforced corner joint tests have been con-
ducted. The test matrix addressed two parameters: (1) joint aspect ratio, and (2) beam long-
itudinal reinforcement ratio. The configuration and design details of the specimens are
illustrated in Figure 2.
Beam Reinforcement Ratio

low high
SP1 SP2
4-#6 4-#8
beam
beam
stirrup stirrup
18’’
18" 18’’
18"
#3@3’’ #3@3’’
4-#6 16" 4-#7
16’’
16" 16’’
Corner joint low
Aspect Ratio (hb /hc)
column
column
8-#8 8-#8
18’’
18" hoop 18’’
18" hoop
#3@3’’ #3@3’’
18"
18’’ 18"
18’’
SP3 SP4
4-#6 4-#8
H =145 “
L=96”
beam
beam
stirrup stirrup
30’’
30"
#3@3’’ 30’’
30"
#3@3’’
EW NS 4-#6 4-#7
16’’
16" 16’’
16"
high
column
column
8-#10 8-#10
18’’
18" hoop 18’’
18" hoop
#3@3’’ #3@3’’
18’’
18" 18’’
18"
Test setup
Figure 2. Details of the full-scale 3-D unreinforced beam-column corner joint specimens.
The lateral load was applied in a quasi-static (0.02 in/sec) manner through vertical dis-
placement control at the end of the beam, which is assumed to be an inflection point of the
beam. This lateral load alternated between the two beams, that is, one beam remained at a
stationary level during the loading of the other orthogonal beam. Variable axial loads were
applied to the column to simulate the overturning effect on the column under earthquake
shaking. The column axial loading equation was derived from the pushover analysis of a
hypothetical prototype structure obtained by modifying the Van Nuys Holiday Inn building
model (Krawinkler 2005). Based on the pushover analysis results, the relationship between
the column axial force and the beam shear was approximated by a linear equation, Equation 5.
The derived equation was applicable to specimens SP3 and SP4, which have the same dimen-
sions as the first-story beams and columns in the considered prototype frame. The coefficient
α in Equation 5 was adjusted for SP1 and SP2 such that similar column axial loads were
applied at the onset of beam yielding for each pair of specimens having the same amount
of beam top reinforcement, that is, (SP1 and SP3) and (SP2 and SP4). During the tests, the
beam shear forces for the east-west and north-south directions, V b;EW and V b;NS , respectively,
at each loading step were recorded in real time and these forces directly determined the
applied column axial load according to Equation 5,
Papplied¼ −95 þ αðV b;EW þ V b;NS Þ

where α ¼ 4 for specimens SP1 and SP2 and α ¼ 2 for specimens SP3 and SP4. The column
axial loads for the four specimens varied between 0.07f c0 Ag in tension and 0.25f c0 Ag in com-
pression. P-delta effect was not considered in the test setup for simplicity.
JOINT SHEAR STRESS AND SHEAR STRAIN RESPONSES

Joint shear stresses, vjh , are evaluated using a constant moment arm between the tension
and compression resultants in the beam L-shaped cross section including the effective
slab width, estimated by ASCE 41. The constant moment arm is defined as a fraction
of the effective depth, d b, of the beam cross section. This fraction is selected as
0.875 d b until beam reinforcement yields based on jd b ¼ d b − βd b∕3 where βd b is the neu-
tral axis depth in compression and β is estimated as 0.4 (Hakuto et al. 1999). After yielding,
the internal moment arm is assumed to be 0.9d b as commonly adopted in estimating the
horizontal joint shear forces (Wong 2005). The effective depth of the L-shaped beam
cross section is estimated using the measured concrete cover depths and the slab top rein-
forcement within the effective slab width. Joint shear strain, γ xy , and rotation at the beam-
joint interface, θs , due to bar slip and/or crack opening were separately measured in the tests,
as shown in Figure 3.
The joint shear stress–shear strain relationships in the EW direction of the four specimens
are plotted in Figure
pffiffiffiffi 4. Note that the joint
pffiffiffiffishear stresses on the vertical axes of these plots are
normalized by f c0 , that is, γ ¼ vjh ∕ f c0 . The results show that the joint shear strength
increased with the increase of the beam reinforcement ratio, that is, from SP1 to SP2
and from SP3 to SP4, but decreased with the increase of the joint aspect ratio, that is,
from SP1 to SP3 and from SP2 to SP4. These results are consistent with the trends observed
in many previous tests documented in Park and Mosalam (2012b). The average joint shear
strains of the values measured at the downward and upward peak loads are 0.006 rad for SP1,
0.007 rad for SP2 and SP3, and 0.008 rad for SP4. The evaluated joint shear stress–shear
strain responses are compared to two backbone curves proposed by ASCE 41 and Priestley
(1997) in Figure 4.
In addition to the large discrepancy in the strength prediction presented in Table 1,
the two models—ASCE 41 and Priestley (1997) in Figure 4—are conservative for pre-
dicting the post-peak drift capacity of unreinforced corner joints. In the ASCE 41 back-
bone curve, the joint shear strength is γ n ¼ 6 psi0.5 ½0.5 MPa0.5 & starting from zero shear
Figure 3. Instrumentation for measuring the joint deformations: (a) joint shear strain; (b) rotation
at the beam-joint interface.
[ psi 0.5 ] [psi 0.5 ]

12 12
10 10
8 (+) (+) 8 (+) (+)
6 6
4 4
2 2
0 0
-2 -2
-4 -4
-6 -6
-8 ASCE 41 -8 ASCE 41
-10 Priestley (1997) -10 Priestley (1997)
-12 -12
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Joint shear strain ( 10-2 rad) Joint shear strain ( 10-2 rad)
(a) SP1-EW ( b) SP 2 - E W
[ psi 0.5 ] [ psi 0.5 ]
12 12
10 10
(+) (+) (+) (+)
8 8
6 6
4 4
2 2
0 0
-2 -2
-4 -4
-6 -6
-8 ASCE 41 -8 ASCE 41
-10 Priestley (1997) -10 Priestley (1997)
-12 -12
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Joint shear strain ( 10-2 rad) Joint shear strain ( 10-2 rad)
(c) SP3-EW ( d) SP 4 - E W
Figure 4. Joint shear stress versus shear strain responses: (a) SP1-EW, (b) SP2-EW, (c) SP3-EW,
(d) SP4-EW.
strain with distinct drops at two critical shear strain values of 0.005 rad and 0.010 rad
regardless of the joint aspect ratio and the beam reinforcement ratio. In the backbone
curve by Priestley (1997), the joint shear strength is maintained with rigid-plastic beha-
vior up to shear strain of 0.007 rad followed by linear degradation up to shear strain of
0.02 rad and then another linear degradation up to complete loss of shear capacity at
shear strain of 0.04 rad.
The envelopes relating bar stress to deformation measured at the beam-joint interface of
the four specimens are plotted in Figure 5a as illustrated by the insert in the figure where the
shown deformation is the relative deformation between the column and the beam. The bar
stress did not reduce after the peak loads (marked by arrows for the four specimens in
Tensile stress (ksi) SP3SP1 (psi0.5)

80 12
10
bottom bars
60 SP2 8
40 SP4 6 (+) SP1
4 SP2
20 SP3
(+) 2
s SP4
0 0
s (−)
-20 -2
-4
top bars
-40 SP1
SP4 SP2 -6
-60 SP2 SP3 -8
SP4 -10
-80 -12 s(+)
-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 -3 -2 -1 0 1 2 3
SP1 SP3at beam-joint interface (in.)
Deformation
Deformation at beam-joint interface (in.) Rotation at beam-joint interface ( 10-2 rad)
(a) Bar stress versus deformation (b) Joint shear stress versus rotation
Figure 5. Measured responses at the beam-joint interface: (a) bar stress versus deformation;
(b) joint shear stress versus rotation.
Figure 5a), which implies that the capacity of the anchorage was maintained in the loading
subsequent to the peak loads. It is observed that the deformation at the beam-joint interface
is not significantly dependent on the bar stress, which is related to the beam longitudinal
reinforcement ratio. This observation is based on measuring similar deformations at the
peak load for each pair of specimens having a different beam reinforcement ratio (and accord-
ingly different stress in the beam longitudinal bars) but the same joint aspect ratio, that is,
(SP1 and SP2) and (SP3 and SP4). It is noted that the average deformation at the peak loads is
smaller in the high joint aspect ratio specimens, that is, 0.13 in. for SP3 and SP4, than in the
low joint aspect ratio specimens, that is, 0.21 in. for SP1 and SP2. This smaller deformation
leads to the smaller rotation at the beam-joint interface of SP3 and SP4 as shown in Figure 5b.
Accordingly, it is postulated that the rotation at the beam-joint interface is more influenced by
the joint aspect ratio than by the beam reinforcement ratio.
BACKBONE CURVE FOR JOINT MACRO-MODEL

In conventional simulations of RC buildings, joints are modeled as a node where one-
dimensional (1-D) beam and column elements intersect, that is, the orthogonality between
beams and supporting columns is maintained during the analysis. However, this orthogon-
ality is not sustained in unreinforced joints because of shear failure and shear deformation. To
account for joint shear failure and flexibility in building simulations, a nonlinear joint macro-
model using a single rotational spring is pursued in this study. Joint macro-models in Bidda
and Ghobarah (1999) and Lowes and Altoontash (2003) use concrete softening models from
Hsu (1988) and Vecchio and Collins (1986), respectively, to define the constitutive relation-
ship of the joint panel, but these softening models are inappropriate for lightly reinforced
joints and especially for unreinforced joints LaFave and Shin (2005). In this study, a back-
bone curve is developed empirically based on the joint responses measured and visual obser-
vations from the four corner joint specimens discussed in the previous section. The responses
in terms of the joint shear stresses normalized by the overall maximum, that is, joint shear
strength, versus the joint shear strain and rotation at the beam-joint interface are shown in
Figure 6.
Rotations at the beam-joint interface can be either explicitly modeled as additional rota-
tional springs or implicitly included in the joint rotational spring. The implicit modeling is
adopted in this study because it is a practical option for simulating a building with a large
number of degrees of freedom. The rotation of the joint spring, θj , is defined as the sum of the
joint shear strain and the rotation at the beam-joint interface, that is, θj ¼ γ xy þ θs . It is worth
mentioning that rotation at the column-joint interface is assumed to be negligible because of
the considered strong column/weak beam configuration, as mentioned before. As a
monotonic envelope to represent joint shear stress and strain relationship, a multilinear back-
bone curve (Figure 7) is developed based on the peak values of the first cycle to a specified
displacement in the joint responses presented in Figure 6 and the observed joint damage
progression. Modeling parameters are defined at the following physical states: (1) λ1 and
θa represent initial joint cracking, (2) λ2 and θb represent either beam reinforcement yielding
or significant opening of existing joint crack, (3) θc corresponding to the normalized peak
loading represents either further propagation of existing joint crack or additional joint crack
opening, and (4) λ3 and θd represent the normalized residual joint shear stress and rotation
when the joint damage is severe. It is noted that the joint shear strength, V n and corresponding
γ n , are predicted using Equation 3 in the simulations. The modeling parameters are evaluated
n n
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 SP1 -0.4 SP1
SP2 SP2
-0.6 SP3 -0.6 SP3
-0.8 SP4 -0.8 SP4
-1 -1
-3 -2 -1 0 1 2 3 -2 -1 0 1 2
Joint shear strain ( 10-2 rad) Rotation at beam-joint interface ( 10-2 rad)
(a) Joint shear strain (b) Rotation at beam-joint interface
Figure 6. Normalized joint shear stress versus joint deformation: (a) joint shear strain;
(b) rotation at beam-joint interface.
Figure 7. Proposed backbone curve for unreinforced exterior and corner joint tests.
from the tested four corner joint specimens, and the selected mean values for the backbone
curve are presented in Figure 7.
The parameters λ1 , λ2 , θa, and θb are found to be insensitive to the joint aspect ratio
and the beam longitudinal reinforcement ratio because the joint shear strains and the
beam-joint interface rotations of the four specimens were similar before yielding of beam
longitudinal reinforcement as shown in Figure 6. Close to the peak loads, the rotations at
the beam-joint interface were affected by the joint aspect ratio. To address this effect of
the joint aspect ratio on the beam-joint interface rotation, the parameter θc is empirically
proposed as follows:
θc ¼ 0.0325 − 0.0125 h b∕h c

Equation 6 gives 0.020 and 0.012 for the joint aspect ratios, h b∕h c , of 1/1 (SP1 and SP2)
and 5/3 (SP3 and SP4), respectively. It is worth mentioning that Equation 6 shows good
agreement with the results of the previous specimens from Clyde et al. (2000) and Pantelides
et al. (2002). The parameters λ3 and θd given in Figure 7 are selected by best fitting the
negative slope of the global load-displacement responses because the test measurements
were not reliable after severe joint damage.
VERIFICATION OF THE PROPOSED BACKBONE RELATIONSIHP
CORNER JOINT SPECIMENS

Simulations are performed on the tested four corner joint specimens using the proposed
backbone curve for verification. Since the lateral loads were applied in an alternating pattern
between the two orthogonal beams, two-dimensional (2-D) analysis is assumed to be reason-
able to simulate the responses of the EW and NS directions separately. To address the effect
of the slab, the beam cross section is modeled as an L-shape with effective slab width accord-
ing to ASCE 41. The strength and ultimate strain of core concrete confined by hoops for
Figure 8. Modeling of beam-column subassemblies of the tested corner specimens: (a) conven-
tional rigid joint model; (b) model with joint rotational spring; (c) constitutive model for joint
rotational spring.
column and closed stirrups for beam are calculated using the models by Mander et al. (1988),
Qi and Moehle (1991), and Saatçioğlu and Razvi (1992).
The beam-column subassemblies are modeled with two types of idealizations using
OpenSees (2010). The first considers the 1-D beam and column elements intersecting at
the joint (rigid joint), Figure 8a, while the second is similar to the first but the joint region
is modeled with a rotational spring and joint offsets, Figure 8b. The beam and column ele-
ments are modeled using nonlinearBeamColumn elements based on the force method for-
mulation where there is no need to define a plastic hinge length. Five integration points are
assigned to each element. The constitutive model for the joint rotational spring is defined as
the relationship between moment and rotation at the center of the joint panel, which is cal-
culated from the proposed backbone curve shown in Figure 7. Knowing the dimensions of
the RC frame and its beam-column joints, the moment at the center of the joint, M j , is
obtained from the joint shear stress as follows,
pffiffiffiffi
γ f c0 bj h c L 1
Mj ¼ ; η¼ − (7)
η ðL þ h c ∕2Þjd b H
where L is the length from the beam inflection point to the column face. The M j − θj relation-
ship is implemented using a multilinear hysteresis material (Pinching4) available in Open-
Sees (2010) as shown in Figure 8c. The envelope of each positive and negative direction
consists of a quad-linear curve that connects four key points, which are defined by the para-
meters of the proposed backbone curve.
In Figure 8c, nM n and pM n represent the respective negative and positive moments
obtained from the joint shear strength, γ n . θmin and Mðθmin Þ (θmax and Mðθmax Þ) are the mini-
mum (maximum) rotation and the corresponding moment in the hysteretic response until the
current excursion. The adopted hysteresis rules follow a trilinear unloading-reloading path as
follows: (1) linear unloading with the initial stiffness until the moment reaches a fraction
(defined by the parameter uForce) of nM n or pM n ; (2) linear pinching until reaching the
reloading points specified as fractions (defined by the parameters rDisp and rForce) of
θmin , Mðθmin Þ or θmax , Mðθmax Þ (refer to Figure 8c); and (3) upon reaching the reloading
points, linear loading to θmin , Mðθmin Þ or θmax , Mðθmax Þ. The parameters for unloading
and reloading in pinching4 material are selected as follows: uForce = 0.05, rForce =
0.25, and rDisp = 0.5 to match the experimental load-drift responses, which show that
the beam loads at zero displacement were positive (i.e., upward) because the slab pulled
down the beam as the transverse beam orthogonal to the loaded beam remained at a negative
position. It is to be noted that the strength and stiffness degradation parameters in pinching4
material (OpenSees 2010) are ignored for simplicity and because these material damage para-
meters are insignificant in modeling structural collapse (Haselton et al. 2008). The joint panel
finite size represented by rigid links in Figure 8b is modeled by the joint offset command in
OpenSees (2010).
The simulated load-displacement responses are compared to the test results in Figure 9
where the post-peak strength degradation of the tested corner joint specimens was gradual
although most of the damage was concentrated in the joint region. This moderate strength
degradation is attributed to the presence of the transverse beam and floor slab rather than to
the well-confined beams and column. In other words, the damage of the core concrete in the
corner joints was reduced by the monolithically cast orthogonal transverse beam and floor
slab. Further investigation is needed to support this conclusion. In the simulations with the
proposed backbone curve, the strengths of the four specimens are calculated using Equation 3
as presented in Table 1. Furthermore, additional simulations are performed using the ASCE
41 backbone curve with explicit addition of a bilinear bond-slip model (Lehman and Moehle
2000) at the beam-joint interface, and the envelopes of the simulated responses are included
as thick solid lines in Figure 9.
Based on the comparison of the simulated and test results, the implicit joint macro-
modeling with the proposed backbone curve accurately predicts the load-displacement
responses of the tested four specimens with slight discrepancies at the peak loads due
to the shear strength prediction imperfection of the tested four specimens using Equation 3
as listed in Table 1. On the other hand, the rigid joint modeling clearly shows poor pre-
diction of the responses. Moreover, the ASCE 41 backbone curve with the added bond-slip
model (Lehman and Moehle 2000) does not successfully reproduce the load-displacement
responses because of its underestimation of the shear strength and post-peak drift capacity
as shown in the joint responses in Figure 4. It is also shown that the contribution of the bar
50 50
40 40
30 30
Applied Beam Load (kip)

20 20
10 10
0 0
-10 -10
-20 Test -20 Test
Rigid joint Rigid joint
-30 -30
ASCE41 ASCE41
-40 Proposed model -40 Proposed model
-50 -50
-8 -6 -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 8
Beam Tip Displacement (in.) Beam Tip Displacement (in.)
(a ) Sp e c i m e n S P 1 ( b ) S p e c i me n S P 2
80 80
60 60
40 40
20 20
0 0
-20 -20
-40 Test -40 Test
Rigid joint Rigid joint
-60 ASCE41 -60 ASCE41
Proposed model Proposed model
-80 -80
-8 -6 -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 8
Beam Tip Displacement (in.) Beam Tip Displacement (in.)
(c ) Sp e c i m e n S P 3 ( d ) S p ec i me n S P 4
Figure 9. Comparison of simulated and experimental results of tested corner joints: (a) specimen
SP1; (b) specimen SP2; (c) specimen SP3; (d) specimen SP4.
slip model is negligible, except for specimen SP3, because the beam longitudinal bar
stresses remained elastic at the maximum joint shear stresses predicted by the ASCE 41
strength recommendation.
PLANAR EXTERIOR JOINT SPECIMENS (WONG 2005)

Four unreinforced exterior joint specimens without lateral beams and floor slab, referred
to as planar (2-D) exterior joints, tested by Wong (2005), are selected for verifying the pro-
posed backbone curve for joint aspect ratios ranging from 1.0 to 2.0. Test setup and details of
the selected four specimens are illustrated in Figure 10. The joint aspect ratios of specimens
BS-L-300, JA-NN03, BS-L-450, and BS-L-600 are 1.00, 1.33, 1.50, and 2.00, respectively.
Specimens JA-NN03 and BS-L-450 were tested using load control until the beam yielded and
thereafter testing continued by displacement control, while specimens BS-L-300 and BS-L-
600 were tested under displacement control only. The column axial loads were kept constant
at 0.03f c0 Ag for JA-NN03 and 0.15f c0 Ag for the other three specimens.
Figure 10. Test setup and tested specimens from Wong (2005).
The force-displacement responses of the four specimens are simulated by modeling the
joints with rotational springs based on the proposed backbone curve depicted in Figure 7. For
these simulations, the displacement histories are reproduced from the load-displacement plots
illustrated in Wong (2005). The shear strengths of the selected specimens are predicted using
Equation 3 as presented in Table 1. Figure 11 compares the simulated responses as dashed
lines to the test result as solid lines. It is noted that the dashed horizontal lines in Figure 11 for
the specimens with low joint aspect ratios, namely BS-L-30 and JA-NN03, indicate the
experimentally determined yield levels of the beam longitudinal reinforcement.
For each specimen, the simulated responses are symmetric for both loading directions
because beam longitudinal bars are identically placed at the top and bottom without an over-
hanging slab (unlike the corner joint simulations discussed in the previous subsection). On
the other hand, the experimental responses were not symmetric because the joint strength was
not maintained for both loading directions if the joint was severely damaged by the previous
rightward or leftward loading. The simulations using the proposed model are in good agree-
ment with the experimental responses in terms of the displacement at the peak force and post-
peak behavior, although the predicted peak forces have some mismatch that arises from
imperfection of the model predictions. Consequently, the simulations of the four specimens
tested by Wong (2005) reasonably confirm the adequacy of the proposed backbone curve for
simulating the seismic response of unreinforced planar exterior joints.
Figure 11. Simulations of four test specimens (Wong 2005) using the proposed backbone curve:
(a) BS-L-300 ðh b∕h c ¼ 1.00Þ; (b) JA-NN03 ðh b∕h c ¼ 1.33Þ; (c) BS-L-450 ðh b∕h c ¼ 1.50Þ;
(d) BS-L-600 ðh b∕h c ¼ 2.00Þ.
MODELING OF INTERIOR AND ROOF JOINTS

For analysis of a whole building frame, strength predictions and backbone curves for
interior and roof joints are also needed in addition to those for exterior (including corner)
joints that are previously presented. A principle of joint shear strength recommendations in
ACI 352-02 and ASCE 41 provisions is adopted for prediction of shear strength for unrein-
forced interior and roof joints, as illustrated in Figure 12. This principle is that joint shear
strength is affected by joint confinement conditions provided by the surrounding beams and
columns and that there exists certain shear strength ratios among the different types of
joints. Table 2 presents the ratio of shear strengths for three joint types with respect to
the exterior joint shear strength. It is noted that the strength ratio factor for interior joints
is selected as the ratio obtained from the recommendations of ASCE 41 because the ratio per
ACI 352-02 significantly underestimates the maximum shear strength of interior joints
ðγ n ¼ 15 psi0.5 Þ compared to the results of interior joint specimens tested by Alire
(2002) where the measured maximum shear strength was γ n ¼ 25 psi0.5 . Using the strength
Figure 12. Joint shear strengths code recommendations: (a) ACI 352-02; (b) ASCE 41.
ratio factor, Γ, Equation 3 is modified for predicting the joint shear strengths for interior
and roof joints as follows:
$ pffiffiffiffi %
cosθ
V n ¼ k Γ × 12 f c bj h c
0 ; psi units (8a)
cosðπ∕4Þ
" #
SI j − X 1 cosθ cosθ
k ¼ 0.4 þ 0.6 ≤ 1.0; X 1 ¼ Γ × 4 ; X ¼ Γ × 12 ; psi units (8b)
X2 − X1 cosðπ∕4Þ 2 cosðπ∕4Þ
Table 2. Ratios of shear strengths of different joint types
ACI 352-02 ASCE 41

Selected
Joint type *1
γn Γ *1 γn Γ *1 Γ
Exterior 12 1.00 6 1.00 1.00

Roof exterior 8 0.66 4 0.66 0.66
Interior 15 1.25 10 1.67 1.67
Roof interior 12 1.00 6*2 1.00 1.00
*1
Γ ¼ γ∕γ exterior
*2
This value is not explicitly specified, but it can be postulated that this joint type has similar strength to that of the exterior
joint because these two joints are geometrically identical with a 90° turn.
where Γ is listed in the last column of Table 2. It is noted that the joint shear index, Equa-
tion 4, is also applicable to other joint types because it is derived based on global equilibrium.
For interior joints, the total cross-sectional area of beam longitudinal reinforcement in tension
becomes the sum of the cross-sectional area of top bars in one side of the joint and bottom
bars in the other side of the joint. For roof joints, the height, H, between upper and lower
column inflection points used in Equation 4 is taken as half of the top story height.
The modeling parameters of the backbone curve for an interior joint may be different
from those for an exterior joint because two longitudinal beams are framing into the interior
joints instead of one beam in the exterior joints. To determine the modeling parameters of the
backbone curve and to evaluate the above modification of the shear strength prediction for
interior joints, four unreinforced interior joint specimens tested by Alire (2002) are simulated
with varying parameters of the backbone curve to best fit the test results. The simulations
with the selected modeling parameters as dashed lines are compared to the test results shown
in Figure 13. The strength predictions are comparatively acceptable even though the predic-
tions are somewhat conservative, especially for positive column shear in Figure 13d.
The selected joint rotation parameters of the backbone curve, θ’s in Figure 7, for all
unreinforced beam-column joint types are presented in Table 3. Because of the lack of
Figure 13. Simulations of tested four interior joint specimens (Alire 2002): (a) PEER-0850; (b)
PEER-0995; (c) PEER-1595; (d) PEER-4150.
Table 3. Selected joint rotation parameters of the backbone curves for different types of
joints
Joint type θa θb θc θd
Exterior
0.0025 0.005
Roof exterior 0.0325 − 0.0125 ðh b∕h c Þ θc þ 0.03
Interior
0.0050 0.010
Roof interior
test data on unreinforced roof joints, it is assumed that the proposed strength prediction and
backbone curves for exterior and interior joints (located in lower and middle stories) can be
used for exterior and interior joints in the top story, respectively. It should be noted that other
parameters for the backbone curve in terms of normalized joint shear force, λ’s in Figure 7,
are kept the same for all joint types as those defined in Figure 7.
BUILDING FRAME SIMULATIONS WITH JOINT FLEXIBILITY

BUILDING DESCRIPTION
Three hypothetical building frames are designed with guidelines from the design details
of the Van Nuys Holiday Inn building model (Krawinkler 2005) used as the prototype of the
tests summarized in Figure 2. The transverse frame of the structure, which has three bays, is
considered for the presented simulations. The lateral load-resisting system of the considered
frames is an RC moment frame. The building was designed per the 1964 City of Los Angeles
Building Code and was built in 1966. Therefore, the design details of this structure can be
considered to represent nonductile older-type RC buildings. The frame dimensions, beam and
column cross sections, and design details are presented in Figure 14.
A reference building frame is designed to be identical to the original details of the Van
Nuys Holiday Inn structure, except for the shear reinforcement in the columns and beams as
RSB6 RSB5 RSB4

Roof
FSB5 FSB4 8ft.-6.5in.
FSB6
7F
FSB6 FSB5 FSB4 8ft.-6in.
6F
30" 22.5" 22"
5F
4F 14" 14" 14"
8ft.-6in. 2FSB- FSB- RFSB-
FSB6 FSB5 FSB4
3F
2FSB6 2FSB5 2FSB4 8ft.-6in.
2F 20" 20"
C9 C18 C27 C36 13ft.-6in.

14" 14"
20ft.-1in. 20ft.-10in. 20ft.-1in. N
1C9&1C36 1C18&1C27
(a) Elevation (b) Cross-sections of beams and columns
Figure 14. Prototype building frame and cross sections: (a) elevation and (b) cross sections of
beams and columns.
Table 4. Details of the reference (Type I) building
Dimension f y; column Concrete

Floor C-9 C-18 C-27 C-36
bc × h c [in.] [ksi] f c0 [psi]
5,000 1st 8-#9 12-#9 12-#9 8-#9
60
4,000 2nd 6-#7 8-#9 8-#9 6-#7
Column 14 × 20 3rd 6-#7 8-#9 8-#9 6-#7

4th 6-#7 6-#9 6-#9 6-#7
60 3,000
5th–7th 6-#7 6-#7 6-#7 6-#7
bb × h b f y; beam
f c0 [psi] Floor Location B4 B5 B6
[in.] [ksi]
Top 3-#9 3-#9 2-#9
14 × 30 40 4,000 2nd (2FS-)
Bottom 2-#9 2-#8 2-#9
Top 3-#10 3-#10 3-#10
3rd–4th (FS-)
Bottom 2-#8 2-#8 2-#8
Top 3-#9 3-#9 3-#9
Beam 14 × 22.5 40 3,000 5th–6th (FS-)
Bottom 2-#8 2-#8 2-#8
Top 2-#9 3-#8 2-#9
7th (FS-)
Bottom 2-#8 2-#8 2-#8
Top 2-#8 3-#7 2-#8
14 × 22 40 3,000 Roof (RS-)
Bottom 2-#7 2-#7 2-#7
discussed below. This building frame is referred to hereafter as “Type I” and the design details
are presented in Table 4. From the shear strength prediction of the beam-column joints in Type
I frame using Equation 3, the shear strengths of the beam-column joints are close to the lower
bound at which the failure of the beam-column joint occurs at the ultimate flexural strength of
the beams. To investigate the change of the lateral response for different levels of joint shear
strength, two additional building frames are considered by simply increasing the yield strength
of the beams, f y;beam , and columns, f y;column , from those used in a Type I building frame. With
the increase of f y;beam and f y;column by 25 percent, the shear strengths of the beam-column joints
in the second building frame (referred to as “Type II”) are located between the upper and lower
bounds so that the failure of the beam-column joints is accompanied by yielding of the beams’
longitudinal reinforcement but less than their ultimate strengths. The beam-column joints in the
third building frame (referred to as “Type III”) are specified with the increase of f y;beam and
f y;column by 62.5 percent so that most of the beam-column joints are subjected to shear failure
prior to yielding of the beams longitudinal reinforcement.
Modeling all the seismic deficiencies found in older RC-framed structures, especially
shear-critical columns, will change the output of these simulations because of premature col-
umn shear failures, anchorage loss, or P-delta effects. Prior to conducting complete collapse
analysis (Talaat and Mosalam 2009) of nonductile RC buildings, including all nonseismic
details, the dynamic analyses in the next subsection are only focused on the investigation
of the influence of shear failure and flexibility of the unreinforced beam-column joints on
the lateral responses. For this purpose, beams and columns in the three building types (I,
II, and III) are assumed to have enough shear reinforcement to prevent their shear failures.
NONLINEAR DYNAMIC ANALYSES

The beam-column joints in the hypothetical building frames are modeled in two different
ways, namely rigid joint modeling and joint rotational spring modeling as shown in Figures 8a
and b, respectively. The fundamental periods of the three building frames are presented in
Table 5. It is obvious that the fundamental periods of the three building frames are greater
in the model with the proposed joint springs than in the model with rigid joints because
the rotational springs make the frames more flexible. Note that the fundamental periods of
the three building frames modeled with rigid joints are the same because the initial stiffness
values of these frames are identical. The change of the fundamental period according to
whether joint flexibility is considered or ignored in the computational model produces different
dynamic responses. Hence, nonlinear time-history analyses are performed for the three building
frames under multiple ground motions scaled to produce increasing spectral accelerations.
The ground motion acceleration recorded during the 1994 Northridge earthquake
(Tarzana station), shown in Figure 15, is used for these dynamic analyses. This ground
Table 5. Fundamental periods of the analyzed three building frames, T1 [sec]
Type I Type II Type III
Rigid Proposed joint Rigid Proposed joint Rigid Proposed joint

joints springs joints springs joints springs
1.29 1.75 1.29 1.67 1.29 1.60
Spectral Response Acceleration (g)
1.5 3 0.1g at T 1
0.5g at T 1
Ground acceleration (g)
2.5 1.0g at T 1
1
2
0.5
1.5
0 1.0g
1
-0.5 0.5g
0.5
0.1g
-1 0
0 5 10 15 20 25 0 0.5 1 1.5 2
periodTT(sec)
Period (sec.)
Time (sec)
Figure 15. Selected ground motion (Northridge earthquake at Tarzana) and scaling procedure.
Figure 16. Maximum interstory drift results of the analyzed three building frames: (a) Type I
building frame; (b) Type II building frame; (c) Type III building frame; (d) increase of interstory
drift due to joint rotation.
acceleration history is scaled such that the scaled ground motions would produce a spectral
acceleration at the natural period of the frame ðT 1 Þ of 0.1 g to 1.0 g with increments of 0.1 g,
based on the ASCE-7 (2006) design response spectrum. It is noted that the average of the
fundamental periods listed in Table 5 (referred to as T 1 in Figure 15 and is taken as 1.48 sec)
is considered for the scaling of the ground motion.
Figures 16a–16c illustrate the simulation results of maximum interstory drift, θmax , versus
spectral acceleration at the fundamental period of the three building frames, SaðT 1 Þ. As
expected, the proposed joint rotational spring model introduces additional interstory drifts
to those of the rigid joint models. The increase of the maximum interstory drift due to joint
rotation is more significant as the spectral acceleration increases, but this increase reduces for
larger spectral accelerations as the beams and columns are subjected to large strain hardening
in the rigid joint model. This behavior is reflected in Figure 16d, where the increase of the
interstory drift due to the joint rotation falls sharply at higher spectral acceleration (> 0.70 g)
for the more flexible building frames like Type I, where at SaðT 1 Þ > 0.9 g, maximum inters-
tory drift is greater in the model with rigid joints because of yielding of first-story columns.
The results of the three building frames are assembled in Figure 17a. Between the two
different types of joint idealizations, a significant difference of the spectral accelerations is
1.0
Probability of Exceedance
0.8
Proposed joint model

0.6
rigid joint model

0.4
0.2
0.0
0
0.0 0.2 0.4 0.6 0.8 1.0
0
Sa (T1) (g)
(a) Comparison of interstory
i drifft for the mod
del (b) Shhift of the fraagility curve ffor significantt
with
h rigid joints and
a proposed joint springs structural ddamage (schem matic)
Figure 17. Effect of joint flexibility on seismic assessment of nonductile RC buildings: (a) com-
parison of interstory drift for the model with rigid joints and proposed joint springs; (b) shift of the
fragility curve for significant structural damage (schematic).
observed around 2 percent interstory drift at which a performance level is defined as sig-
nificant structural damage (Celik and Ellingwood 2010). Consequently, this difference pro-
duces a significant shift of the seismic vulnerability functions for nonductile RC buildings
designed with unreinforced joints such as the fragility curve for a certain structural damage
limit state as schematically shown in Figure 17b. Therefore, the importance of joint flexibility
should be carefully recognized in the earthquake simulation and seismic assessment of non-
ductile RC buildings having unreinforced beam-column joints.
CONCLUDING REMARKS
Based on the measured joint shear stress rotation and visual observations of the tested
corner joint tests, a multilinear backbone curve for unreinforced corner beam-column joints is
developed. The accuracy of the proposed backbone curve is demonstrated by accurate repro-
duction of the load-displacement responses for the tested four corner joint specimens and four
planar exterior joint specimens from literature. The ASCE 41 provisions are shown to be
conservative for predicting the shear strength and deformability of unreinforced corner joints.
Furthermore, the proposed backbone curve and adopted shear strength model are modified
for interior and roof joints. Available test data on four interior joint specimens are success-
fully simulated using the modified backbone curve.
Three hypothetical building frames are designed, and nonlinear dynamic analyses are
performed to investigate the change of the structural responses when joint flexibility is con-
sidered or ignored in the computational model. The results of the nonlinear dynamic analyses
emphasize the importance of joint flexibility for the assessment of the seismic performance of
nonductile reinforced concrete buildings containing unreinforced joints. Therefore, backbone
models, such as the proposed ones, reflecting realistic behavior of unreinforced beam-column
joints are recommended in structural modeling of such buildings.
In a future study, further refinement of the proposed backbone model of unreinforced

beam-column joints will be conducted. This refinement will take into account the strength
degradation due to cyclic loading, and different types of reinforcement (e.g., smooth bar) and
anchorage details benefiting from available literature.
ACKNOWLEDGMENTS
The research presented in this paper was supported primarily by the National Science
Foundation (NSF) award #0618804 (J. P. Moehle, PI) through the Pacific Earthquake Engi-
neering Research Center (PEER). Any opinions, findings, and conclusions or recommenda-
tions expressed in this paper are those of the authors and do not necessarily reflect those of the
NSF or PEER.
REFERENCES
Alire, D. A, 2002. Seismic evaluation of existing unconfined reinforced concrete beam-column
joints, MSCE Thesis, University of Washington, 291 pp.
American Concrete Institute (ACI Committee 318), 2011. Building Code Requirements for
Structural Concrete (ACI 318-11) and Commentary (318R-11), American Concrete Institute,
Farmington Hills, MI.
American Concrete Institute (ACI-ASCE Committee 352), 2002. Recommendations for Design
of Beam-Column Connections in Monolithic Reinforced Concrete Structures (ACI 352R-02),
American Concrete Institute, Farmington Hills, MI.
American Society of Civil Engineers (ASCE/SEI), 2006. Minimum Design Loads for Buildings
and Other Structures (ASCE/SEI 7-05), American Society of Civil Engineers, Reston, VA.
American Society of Civil Engineers (ASCE/SEI), 2007. Seismic Rehabilitation of Existing
Buildings (ASCE/SEI 41-06), American Society of Civil Engineers, Reston, VA.
Biddah, A., and Ghobarah, A., 1999. Modelling of shear deformation and bond slip in reinforced
concrete joints, Struct. Eng. Mech. 7, 413–432.
Building Seismic Safety Council (BSSC), 1997. NEHRP Guidelines for the Seismic Rehabilita-
tion of Buildings, FEMA-273, Federal Emergency Management Agency, Washington, D.C.
Building Seismic Safety Council (BSSC), 2000. NEHRP Guidelines for the Seismic Rehabilita-
tion of Buildings, FEMA-356, Federal Emergency Management Agency, Washington, D.C.
Celik, O. Z., and Ellingwood, B. R., 2010. Seismic fragilities for non-ductile reinforced concrete
frames: Role of aleatoric and epistemic uncertainties, Struct. Safety 32, 1–12.
Elwood, K. J., and Moehle, J. P., 2003. Shake Table Tests and Analytical Studies on the Gravity
Load Collapse of Reinforced Concrete Frames, PEER Report 2003/01, University of Cali-
fornia, Berkeley, 346 pp.
Günay, S. M., and Mosalam, K. M., 2010. Structural Engineering Reconnaissance of the April 6,
2009, Abruzzo, Italy, Earthquake and Lessons Learned, PEER Report 2010/105, University of
California, Berkeley, 46 pp.
Hakuto, S., Park, R., and Tanaka, H., 1999. Effect of deterioration of bond of beam bars
passing through interior beam-column joints on flexural strength and ductility, ACI Struct.
J. 96, 858–864.
Haselton, C. B., Liel, A. B., Lange, S. T., and Deierlein, G. G., 2008. Beam-Column Element
Model Calibrated for Predicting Flexural Response Leading to Global Collapse of RC Frame
Buildings, PEER Report 2007/03, University of California, Berkeley, 134 pp.
Hsu, T. T. C., 1988. Softened truss model theory for shear and torsion, ACI Struct. J. 85,
624–634.
Krawinkler, H. (ed.), 2005. Van Nuys Hotel Building Testbed Report: Exercising Seismic Per-
formance Assessment, PEER Report 2005/11, University of California, Berkeley, 250 pp.
LaFave, J. M., and Shin, M., 2005. Discussion of “Modeling Reinforced Concrete Beam-Column
Joints Subjected to Cyclic Loading” by Laura N. Lowes and Arash Altoontash, J. Struct. Eng.
131, 992–993.
Li, B., Wang, Z., Mosalam, K. M., and Xie, H., 2008. Wenchuan earthquake field reconnaissance
on reinforced concrete framed buildings with and without masonry infill walls, The 14th World
Conference on Earthquake Engineering, Beijing, China.
Lehman, D. E., and Moehle, J. P., 2000. Seismic Performance of Well-Confined Concrete Bridge
Columns, PEER Report 1998/01, University of California, Berkeley, 295 pp.
Lowes, L. N., and Altoontash, A., 2003. Modeling reinforced-concrete beam-column joints sub-
jected to cyclic loading, J. Struct. Eng. 129, 1686–1697.
Mander, J. B., Priestley, M. J. N., and Park, R., 1988. Theoretical stress-strain model for confined
concrete, J. Struct. Eng. 114, 1804–1826.
Mitra, N., and Lowes, L. N., 2007. Evaluation, calibration, and verification of a reinforced con-
crete beam-column joint model, J. Struct. Eng. 133, 105–120.
Moiser, W. G, 2000. Seismic assessment of reinforced concrete beam-column joints, MSCE the-
sis. University of Washington, 218 pp.
OpenSees, 2010. Open System for Earthquake Simulation, PEER, University of California, Ber-
keley, http://opensees.berkeley.edu.
Park, S., 2010. Experimental and analytical studies on old reinforced concrete buildings with
seismically vulnerable beam-column joints, Ph.D. Dissertation, University of California, Ber-
keley, 245 pp.
Park, S., and Mosalam, K. M., 2012a. Analytical model for predicting the shear strength of
unreinforced exterior beam-column joints, ACI Struct. J. 109, 149–159.
Park, S., and Mosalam, K. M., 2012b. Parameters for shear strength prediction of exterior beam-
column joints without transverse reinforcement, Eng. Struct. 36, 198–209.
Priestley, M. J. N., 1997. Displacement-based seismic assessment of reinforced concrete build-
ings, J. Earthquake Eng. 1, 157–192.
Qi, X., and Moehle, J. P., 1991. Displacement Design Approach for Reinforced Concrete Struc-
tures Subjected to Earthquakes, UCB/EERC-91/02, Earthquake Engineering Research Center,
University of California, Berkeley, 186 pp.
Saatçioğlu, M., and Razvi, S. R., 1992. Strength and ductility of confined concrete, J. Struct. Eng.
118, 1590–1607.
Sezen, H., Elwood, K. J., Whittaker, A. S., Mosalam, K. M., Wallace, J. W., and Stanton, J. F.,
2000. Structural Engineering Reconnaissance of the August 17, 1999 Earthquake: Kocaeli
(Izmit), Turkey, PEER Report 2000/09, University of California, Berkeley, 154 pp.
Talaat, M., and Mosalam, K. M., 2009. Modeling progressive collapse in reinforced concrete
buildings using direct element removal, Earthquake Eng. Struct. Dyn. 38, 609–634.
Uang, C. M., Elgamal, A., Li, W. S., and Chou, C. C., 1999. Ji-Ji Taiwan Earthquake of Sep-
tember 21, 1999: A Brief Reconnaissance Report, University of California, San Diego, http://
www.structures.ucsd.edu/Taiwaneq.
Vecchio, F. J., and Collins, M. P., 1986. The modified compression field theory for reinforced
concrete elements subjected to shear, ACI Struct. J. 83, 219–231.
Vollum, R. L., 1998. Design and analysis of exterior beam column connections, Ph.D. Disserta-
tion, Imperial College of Science Technology and Medicine, University of London, 603 pp.
Wong, H. F., 2005. Shear strength and seismic performance of non-seismically designed rein-
forced concrete beam-column joints, Ph.D. Thesis, Hong Kong University of Science and
Technology, 341 pp.
(Received 7 March 2011; accepted 29 February 2012)
Engineering Characteristics of Ground
Motion Records from the 2010 Mw 7.2 El
Mayor–Cucapah Earthquake in Mexico
Yun Liaoa) M.EERI, and Jorge Menesesb) M.EERI
This paper presents an engineering analysis of ground motion records from

the 2010 El Mayor–Cucapah earthquake. Spectral accelerations from the 1979
Imperial Valley earthquake are also compared with NGA Boore and Atkinson
2008 (BA08) and SEA99 ground-motion prediction equations (GMPEs). Both
GMPEs predict slower distance attenuation than demonstrated by the 2010
earthquake but capture distance attenuation shown by the 1979 earthquake.
Intra-event residuals of both GMPEs relative to the 2010 earthquake data
exhibit larger scattering than the 1979 earthquake data. H/V ratios indicate that
some soft-soil sites show predominant site periods of about 0.4 to 0.7 seconds,
but the remaining sites show lengthened predominant periods of about 1.0 to
2.0 seconds. Rock sites show no pronounced predominant periods despite the
presence of a weak H/V ratio peak at 0.2 seconds. [DOI: 10.1193/1.4000101]
INTRODUCTION
The main shock of the Mw 7.2 El Mayor–Cucapah earthquake occurred at 3:40 PM
(PDT) on 4 April 2010 in Baja California, Mexico, with the hypocenter located at
32.259ºN, 115.287ºW, and a depth of approximately 10 km (USGS 2011). The epicenter
was approximately 47 km SSE of Mexicali, Baja California, 51 km SSE of Calexico,
California, and 180 km SE of San Diego (EERI 2010). The earthquake was felt in a
large area encompassing northern Baja California in Mexico and Southern California,
Arizona, and Nevada, with a Mercalli intensity of as high as IX in areas near the fault rupture
and VIII in Mexicali, which is the nearest densely populated area. Figure 1 shows the location
of the epicenter as well as the location of the rupture, distribution of shaking intensities, and
contours of peak ground acceleration (PGA) values. Using the methods developed by Wald
et al. (1999), the United States Geological Survey (USGS) generated the modified Mercalli
intensity map and the peak ground acceleration contours.
Two people were killed and hundreds were injured in Baja California. The earthquake
caused damage in a wide region from the Sea of Cortez in the south to the Salton Sea in the
north. Extensive damage occurred to irrigation systems, water treatment facilities, buildings,
bridges, earth dams, and roadways. Extensive liquefaction and ground failure occurred in
Mexicali Valley, which left wheat and hay fields submerged under water, and destroyed
many canals that irrigate fields in the predominantly agricultural region. A comprehensive
a)
Kleinfelder, 1330 Broadway Blvd., Suite 1200, Oakland, CA 94612
b)
Kleinfelder, 5015 Shoreham Place, San Diego, CA 92122
177
178 Y. LIAO AND J. MENESES
CISN ShakeMap for Sierra El Mayor Earthquake

Sun Apr 4, 2010 03:40:42 PM PDT M 7.2 N32.26 W115.29 Depth: 10.0km ID:14607652
33.5˚N
7
7
Salton City
7
Ocotillo Wells
33˚N Brawley 7
7
21 El Centro
7
14
Yuma
Mexicali
21
28
Tecate
14
32.5˚N 42
35
San Luis R o Colorado
42
Guadalupe Victoria
35
32˚N 28
Ensenada
21
0 50 km
7
31.5˚N
-117˚
-117˚W -116˚W
-116˚ -115˚
-115˚W
PERCEIVED
SHAKING Not felt Weak Light Moderate Strong Very strong Severe Violent Extreme
POTENTIAL none none none Very light Light Moderate Moderate/Heavy Heavy Very Heavy
DAMAGE
PEAK ACC.(%g) <.17 .17-1.4 1.4-3.9 3.9-9.2 9.2-18 18-34 34-65 65-124 >124
PEAK VEL.(cm/s) <0.1 0.1-1.1 1.1-3.4 3.4-8.1 8.1-16 16-31 31-60 60-116 >116
INSTRUMENTAL
INTENSITY I II-III IV V VI VII VIII IX X+
Figure 1. Locations of the epicenter (start) and fault rupture (solid line), distribution of shaking
intensities, and contours of peak ground acceleration values (g) in percent of gravity (Liao and
Meneses 2012).
description of observed damage is presented in the Earthquake Engineering Research

Institute (EERI) reconnaissance report (EERI 2010) and Geotechnical Extreme Events
Reconnaissance (GEER) preliminary report (GEER 2010).
This earthquake is the largest one in northern Baja California since the Laguna Salada
earthquake in 1892. The earthquake generated a number of strong ground motion recordings
over a large area encompassing Baja California, Mexico, and Southern California. The main
shock of the earthquake was recorded by 497 strong motion stations with the largest PGA of
about 0.54 g recorded in the El Centro Array #11, California (closest distance to fault rupture
plane ðRrupÞ ¼ 21.2 km) and about 0.53 g recorded in the Michoacan de Ocampo (MDO)
station, Baja California ðRrup ¼ 17.5 kmÞ. The recorded PGA was about 0.02 g to 0.05 g in
the San Diego area and about 0.01 g in the Los Angeles area.
The amplification of ground motions due to the local site condition plays an important
role in seismic damage (e.g., Idriss 1991, Singh 1998). Idriss (1991) found that the PGA
ENGINEERING CHARACTERISTICS OF GM RECORDS FROM THE EL MAYOR–CUCAPAH EARTHQUAKE 179
values recorded at soft-soil sites during the 1989 Loma Prieta earthquake were significantly
greater than those recorded at other sites at distances of about 45 to 100 km. The use of the
site soil stiffness represented by the shear wave velocity in the upper 30 m of soil deposits
ðV S30 Þ was then initiated in the early 1990s (Borchert 1994) by assigning a range of V S30 to
different site categories (e.g., NEHRP categories). More recently, V S30 has been included as
an independent parameter in the ground-motion prediction equations (GMPEs) (Boore et al.
1997, Campbell and Bozorgnia 2008 (CB08), Boore and Atkinson 2008 (BA08), Chiou and
Youngs 2008 (CY08), and Abrahamson and Silva 2008 (AS08)). However, prior to this
earthquake, there has not been one single large earthquake that recorded so many (more
than 30) strong ground motions at soft-soil sites (Chiou and Youngs 2008). Hence the
study on effects of soft soil on site response has been limited by the scarcity of recordings
at soft-soil sites. Such a limitation makes the strong motion records at soft-soil sites from this
earthquake very valuable for studying the effect of soft soil on site responses.
This paper presents a study on characterizing strong ground motions recorded during the
main shock of the 2010 El Mayor–Cucapah earthquake with a focus on the effect of soft-soil
deposits on site response characteristics. Ground motion parameters evaluated include spec-
tral accelerations, frequency content, Arias intensity, and ground motion duration. Whenever
possible, the recorded ground motion parameters are compared with predictions by available
GMPEs or attenuation relationships. Liao and Meneses (2012) compared four Next
Generation Attenuation (NGA) GMPEs (i.e., AS08, BA08, CB08, and CY08) with observed
spectral accelerations from the 2010 El Mayor–Cucapah earthquake in terms of intra-event
residuals with respect to V S30 and the Joyner-Boore distance (Rjb). Due to the many
similarities between the 2010 El Mayor–Cucapah earthquake and the 1979 Imperial Valley
earthquake main shock, spectral accelerations from the 1979 Imperial Valley earthquake are
also compared with the NGA BA08 and the SEA99 GMPEs. SEA99 (Spudich et al. 1999)
was developed for predicting earthquake ground motions in extensional tectonic regimes
such as the Imperial Valley and the Mexicali Valley.
In addition, the horizontal-to-vertical (H/V) ratio technique proposed by Nakamura
(1989) was used to estimate predominant site periods of soft-soil sites with V S30 between
180 and 250 m/s and rock sites with V S30 ¼ 748 m=s, which is an important parameter
for assessing building damage and in evaluating local site amplification.
SEISMOTECTONIC SETTING
The seismotectonic setting of the region is dominated by the northwestern movement of
the Pacific plate with respect to the North American plate. Since 1892, this region has been
historically struck by several earthquakes with magnitudes as large as 7. The 1892 earthquake
occurred along the Laguna Salada fault system, but surface offsets associated with that event
lie farther northwest than the epicenter of the 4 April 2010 main shock. The 1940 Imperial
Valley earthquake occurred farther to the north and on the Imperial fault, and had a
magnitude of 6.9. An event of M 7.0 or 7.1 occurred in this region in 1915, and then in
1934 an M 7.0 to 7.2 event ruptured the Cerro Prieto fault with up to several meters of surface
slip (Mueller and Rockwell 1995).
The April 4 main shock occurred along a strike-slip segment of the North American–
Pacific transform plate boundary along a southeastern extension of the Elsinore and Laguna
Salada fault system. Rupture was initially slow and involved normal faulting. After a pause
lasting about six seconds, the main slip release was asymmetric yet bilateral, with rupture
propagating rapidly to the northwest along the Pescadores and Borrego faults and aftershocks
crossing 10 km north of the U.S.-Mexico border. Accompanying this was a less rapid south-
eastward propagation along the newly named Indiviso fault, with aftershocks occurring down
to the northern tip of the Gulf of California (GEER 2010).
STRONG GROUND MOTION STATIONS

The main shock of 4 April 2010’s El Mayor–Cucapah event was recorded by 497 digital
strong motion instruments, 168 of which are within 200 km, and 27 of which are located in
the Imperial Valley and Mexicali Valley. Most of the stations in the Imperial Valley and the
Mexicali Valley are located on young soft sediments with a V S30 of less than 250 m/s. The
closest instruments are the El Centro Array #12, with an Rrup of about 15.8 km in California
and the Riito station with an Rrup of about 13.6 km in Baja California.
Ground motion records and stations along with local site conditions, Rrup, and V S30
values were obtained from the Center for Engineering Strong Motion Data website
(CESMD 2010). It is noted that V S30 values were measured at 23 of the 168 stations, a major-
ity of which were less than 250 m/s. These V S30 values were either measured by downhole
method or spectral analysis of surface waves (SASW) method and are available in the NGA
database flatfile of the Pacific Earthquake Engineering Research Center (PEER). V S30 values
of the remaining station sites are inferred from the local geological conditions (Wills and
Clahan 2006). Idriss (2008) showed that no apparent systematic bias in prediction of ground
motion parameters results from using inferred values of V S30 based on local geological con-
ditions. Hence, all stations with measured and inferred V S30 values are used in this study.
Note that only ground motions recorded at free-field sites are used in this study because
ground motions recorded within a structure or facility were very likely to have been impacted
by soil–structure interaction. Figure 2 shows approximate locations of stations within 200 km
of the fault rupture plane ðRrup ≤ 200 kmÞ. As seen in Figure 2, soft-soil sites with V S30 less
than 250 m/s are all located in the Imperial and Mexicali Valleys. Table 1 presents the station
locations, characteristics, and ground motion parameters from these stations. It is noted in this
study that the Rrup distance is the same as the Joyner-Boore distance ðRjbÞ based on the fault
geometry (WGCEP 2008).
Both Imperial and Mexicali Valleys are underlain by deep sediments. Figure 3 shows
contours of depth-to-shear wave velocity 1 km/s ðZ 1.0 Þ and depth-to-shear wave velocity
2.5 km/s ðZ 2.5 Þ in the Imperial Valley (Caltrans 2010). These contour maps were generated
using data from the Community Velocity Model (CVM) version 4.0 (Shantz and Merriam
2009). As seen in Figure 3, Z 1.0 is between 400 and 700 m, and Z 2.5 is between 3,000 and
5,000 m. Based on these deep soil conditions, amplification of ground motions at intermedi-
ate to long periods (greater than 0.5 seconds) is expected in these regions.
STRONG GROUND MOTION RECORDS

The California strong motion instruments are owned and maintained by the California
Strong Motion Instrumentation Program (CSMIP) and the United States Geological Survey
(USGS), although several smaller networks also operate in the region. In addition, the 4 April
Figure 2. Locations of strong ground motion stations, the epicenter (star), and fault rupture (solid
line) of the 2010 El Mayor–Cucapah earthquake (Liao and Meneses 2012).
2010 main shock was recorded by a total of 13 strong motion instruments from the Red de
Acelerógrafos del Noroeste de México(RANM) in Mexico. These instruments are owned and
maintained by the Centro de Investigación Científica y de Educación Superior de Ensenada
(CICESE).
Strong motion records obtained in California were downloaded from the CESMD
website (CESMD 2010). These records were filtered with a 0.07–40 Hz bandpass filter
and baseline corrected by CESMD. Records obtained in Baja California were downloaded
from the CICESE website (CICESE 2010). These records were baseline corrected and
filtered with corner frequencies of between 0.1 and 25 Hz. Note that records obtained in
both California and Baja California had not been processed following the protocols of
the NGA project.
Figure 4a plots the acceleration, velocity, and displacement time histories of the main
shock recorded at Station 931 and Station MDO, respectively. Figure 4b presents a plot
of 5% damped pseudo-acceleration response spectra for fault normal (FN) and fault parallel
(FP) directions at Station 931 and Station MDO, respectively. The data are rotated into the
182
Table 1. Strong ground motion parameters of the 2010 El Mayor–Cucapah earthquake
Station Latitude Longitude V S30 2 Rrup3 PGA4 ðD5−95 Þ5 Ia6 T m7

1
Station name code Agency (°) (°) (m/s) (km) (g) (s) (cm/s) (s)
Riito RII RANM 32.166 −114.961 207.5 13.6 0.39 26.6 434.7 0.40
El Centro Array #12 931 USGS 32.7177 −115.6379 196.9 15.8 0.36 30.6 297.0 0.68
Geotermica GEO RANM 32.4 −115.24 207.5 15.8 0.29 43.0 322.5 0.83
Michoacan De Ocampo MDO RANM 32.464 −115.316 207.5 17.5 0.53 33.0 615.6 0.56
Calexico Fire Station 5053 USGS 32.6695 −115.4925 231.2 23.1 0.27 39.0 200.0 0.64
Chihuahua CHI RANM 32.488 −115.242 207.5 24.3 0.22 52.0 219.6 0.60
Saltillo SAL RANM 32.422 −115.13 207.5 24.6 0.15 54.4 158.0 0.80
El Centro—Imperial & Ross 1711 CGS 32.78 −115.567 209 25.4 0.38 24.3 349.5 0.52
El Centro Differential Array TAM RANM 32.55 −115.236 207.5 29.6 0.22 45.8 200.1 0.65
Tamaulipas 1794 CGS 32.773 −115.447 186.2 33.4 0.21 40.4 101.0 0.53
El Centro—Hwy.8/Meloland Overpass 464 USGS 32.8 −115.4746 207.5 33.5 0.19 44.8 103.0 0.59
El Centro—Meloland Geotechnical Array 5028 USGS 32.8292 −115.5055 210.5 33.9 0.25 31.2 147.5 0.40
Bonds Corner 5054 USGS 32.693 −115.3382 223 35.7 0.24 31.2 150.5 0.42
Meloland, E. Holton Rd. and Meloland Rd. DRE SCSN 32.805 −115.447 209 35.7 0.23 56.2 193.5 0.50
Holtville Post Office 5055 USGS 32.8114 −115.3786 202.9 41.2 0.19 41.9 91.0 0.58
Rancho San Luis RSL RANM 32.116 −115.841 748 43.6 0.06 33.8 9.2 0.27
Westmorland—Fire Station 11369 CGS 33.037 −115.623 193.7 46.1 0.16 62.0 60.0 0.63
Brawley Airport 5060 USGS 32.991 −115.5128 208.7 46.9 0.18 34.3 62.5 0.42
Y. LIAO AND J. MENESES
Anza Borrego S.P.—Tierra Blanca 2659 CGS 32.942 −116.276 748 53.7 0.04 41.7 4.9 0.77
Calipatria Fire Station 5061 USGS 33.1266 −115.5159 205.8 59.4 0.07 104.9 35.5 1.02
Rock Hill RXH SCSN 33.183 −115.623 209 61.3 0.04 78.2 6.0 0.96
Ocotillo Wells—Veh. Rec. Area 5375 USGS 33.1557 −116.1679 370.8 64.6 0.08 21.7 12.0 0.55
Santa Isabel Viejo SIV RANM 31.871 −115.816 748 70.0 0.03 32.9 2.7 0.37
Niland—Fire Station 11023 CGS 33.239 −115.513 207.5 70.7 0.09 68.6 24.5 0.68
Salton City 11628 CGS 33.28 −115.984 324.5 71.2 0.12 17.7 24.0 0.40
Borrego Springs—Scripps Clinic 5220 USGS 33.1937 −116.3289 271.4 76.7 0.05 33.5 4.6 0.80
Bombay Beach BOM SCSN 33.365 −115.73 354 79.2 0.05 95.8 10.4 0.99
Winterhaven—Sheriff Substation 22 CGS 32.739 −114.636 280 81 0.06 41.8 16.0 0.56
Rancho Agua Caliente RAC RANM 32.02 −116.301 515 81 0.07 33.1 16.8 0.18
Bombay Beach—Bertram 11627 CGS 33.397 −115.776 390 82.4 0.05 50.8 5.5 0.80
Frink 11684 CGS 33.398 −115.657 345.4 83.8 0.04 59.0 4.5 1.03
North Shore—Durmid 11591 CGS 33.421 −115.831 338.5 85 0.06 50.7 12.0 0.97
Borrego Springs BOR SCSN 33.268 −116.417 748 88.3 0.03 28.6 1.0 0.27
Heroes de la Independencia HDI RANM 31.615 −115.882 748 89.6 0.05 31.4 6.6 0.24
Alpine—Fire Station 2467 CGS 32.838 −116.775 748 92.2 0.03 35.6 1.9 0.22
Salton Beach SLB SCSN 33.485 −115.866 354 92.2 0.01 73.0 1.2 0.74
North Shore—Hwy. 111 & Parkside Dr. 11217 CGS 33.511 −115.92 209 95.5 0.05 56.7 10.0 1.06
Valle de la Trinidad VTR RANM 31.399 −115.714 455 97.1 0.03 35.0 2.4 0.47
El Monte County Park EML SCSN 32.891 −116.846 748 100 0.03 31.5 1.9 0.21
North Shore Salton Sea 2 NSS2 SCSN 33.556 −115.946 354 100.7 0.02 65.1 2.7 0.72
TresHermanos TRH RANM 31.69 −116.19 515 101 0.05 32.8 8.1 0.25
Mecca—CVWD Yard 11625 CGS 33.564 −115.988 345.4 102.2 0.03 77.3 5.4 0.86
Glamis, Black Mountain Rd. GLA SCSN 33.051 −114.827 748 102.4 0.04 31.6 3.0 0.47
ENGINEERING CHARACTERISTICS OF GM RECORDS FROM THE EL MAYOR–CUCAPAH EARTHQUAKE
Mecca Fire Station 5270 USGS 33.572 −116.0772 207.5 104.7 0.06 52.8 11.0 0.76
Otay Lakes Park OLP SCSN 32.608 −116.93 455 104.7 0.04 29.9 6.1 0.24
San Diego Road Dept. SDR SCSN 32.736 −116.942 748 106 0.03 30.7 2.0 0.26
(continued )
183
184
1
Toro Canyon TOR SCSN 33.575 −116.226 354 109 0.02 52.2 0.5 0.55
El Cajon—Marshall & Vernon 2140 CGS 32.813 −116.975 349 110.1 0.02 29.6 1.2 0.24
Santee—Mission Gorge & Cottonwood 2143 CGS 32.84 −116.974 349 110.5 0.04 39.7 3.9 0.23
Spring Valley—Jamacha& Gillespie 3152 CGS 32.71 −117.007 387 111.9 0.04 35.2 4.7 0.43
Cactus City, Airway beacon CTC SCSN 33.655 −115.99 387 112.2 0.02 53.6 2.0 0.71
Dos Picos County Park DPP SCSN 32.999 −116.942 748 112.2 0.03 31.3 1.8 0.22
Chihuahua Valley—Private Residence 5221 USGS 33.38 −116.68 748 114 0.02 32.2 1.1 0.32
Big Chuckawalla Mtns. BC3 SCSN 33.655 −115.454 748 116 0.01 80.7 0.3 0.74
Hinds Plumbing Plant 817 MWD 33.7084 −115.6281 684.9 118.2 0.01 43.7 0.2 0.68
CICESE CIC RANM 31.868 −116.664 748 118.4 0.02 37.6 2.7 0.48
Coachella—6th & Palm 12076 CGS 33.678 −116.178 207.5 118.4 0.04 66.9 6.4 0.81
San Diego—54th & College Grove 3148 CGS 32.735 −117.08 387 118.9 0.05 29.1 5.0 0.59
Poway—Community Rd.& Civic Center 3260 CGS 32.955 −117.042 349 119.6 0.04 33.8 3.0 0.18
San Diego—45th & Orange 3154 CGS 32.754 −117.097 387 120.6 0.03 39.9 3.5 0.74
Anza—Pinyon Flat 5044 USGS 33.6076 −116.4541 724.9 121.3 0.01 31.7 0.3 0.37
La Quinta—Bermudas & Durango 12951 CGS 33.671 −116.301 345.4 121.5 0.08 25.5 9.5 0.42
San Diego—Hwy. 15 & Ocean View 3147 CGS 32.702 −117.12 387 122.4 0.05 46.2 5.0 1.04
Indio—Monroe &Carreon 12904 CGS 33.706 −116.237 207.5 123.1 0.05 51.6 5.3 0.77
Indio—Riverside Co. Fairgrounds 12543 CGS 33.715 −116.221 207.5 123.5 0.04 57.2 4.6 0.78
Palomar PLM SCSN 33.354 −116.863 748 124.9 0.02 44.5 0.9 0.67
Palomar Mountain—Palomar Observatory 12330 CGS 33.354 −116.863 748 124.9 0.03 36 1 0.64
Mountain Center—Pine Mtn. Rnch 5223 USGS 33.5778 −116.5899 338.5 125.3 0.04 16.1 1.6 0.47
San Diego—8th & J St. 3145 CGS 32.71 −117.158 387 126 0.04 47.9 4.5 0.95
Anza, Mitchell Rd. DNR SCSN 33.567 −116.631 349 126.6 0.06 20.7 6.0 0.59
Indio—Jackson Rd. 5294 USGS 33.7462 −116.2163 207.5 126.7 0.03 41.0 4.0 0.74
San Diego—UCSD Hospital Grounds 3746 CGS 32.754 −117.164 387 126.9 0.04 33.5 3.7 0.68
San Diego—I5 & Laurel 3146 CGS 32.73 −117.169 387 127.2 0.03 53.6 4.2 1.01
San Diego—Mira Mesa & Camino Ruiz 3156 CGS 32.915 −117.142 387 127.6 0.02 32.1 1.3 0.43
Indian Wells—Hwy 111 & El Dorado 12966 CGS 33.723 −116.338 207.5 128.1 0.05 31.5 4.2 0.46
Anza Fire Station 5160 USGS 33.5559 −116.6745 338.5 128.1 0.02 35.3 0.6 0.68
San Diego—Balboa & Mt. Abernathy 3151 CGS 32.822 −117.175 387 128.7 0.03 27.1 2.4 0.57
Idyllwild—Kenworthy Fire Station 5372 USGS 33.6159 −116.6212 338.5 130.5 0.05 15.6 3.0 0.32
Cahuilla Valley—Fire Station 5241 USGS 33.512 −116.798 349 132.1 0.05 27.1 3.7 0.29
Palm Desert—Country Club & Portola 12952 CGS 33.758 −116.374 207.5 133 0.06 34.5 7.4 0.70
San Diego—Ocean Beach Fire Station 3121 CGS 32.749 −117.242 387 134.1 0.02 40.3 1.5 0.76
La Jolla—UCSD Hospital Grounds 3234 CGS 32.877 −117.227 515 134.5 0.05 34.5 6.5 0.44
San Diego—VA Medical Center 5440 USGS 32.875 −117.23 387 134.7 0.06 29.2 8.0 0.35
Pacific Beach—Mission & Grand 3150 CGS 32.795 −117.255 387 135.8 0.03 34.4 4.0 0.90
Anza—Tripp Flats Training 5222 USGS 33.6022 −116.7557 684.9 136.9 0.03 22.5 1.3 0.32
Rancho Mirage—Gerald Ford & Bob Hope 12953 CGS 33.787 −116.41 207.5 137.3 0.04 35.5 3.7 0.60
San Marcos—Mission & Twin Oaks 13142 CGS 33.144 −117.165 387 137.4 0.02 35.1 0.9 0.24
Valley
Radec—Sage &Cottonwood School Rds. 12092 CGS 33.482 −116.911 338.5 137.4 0.04 41.0 5.6 0.48
Mirage MGE SCSN 33.818 −116.369 209 139 0.03 75.2 2.5 0.73
Thousand Palms—Vly Preserve HQ 5443 USGS 33.845 −116.311 354 139.9 0.04 18.8 1.5 0.36
Thousand Palms—Post Office 5068 USGS 33.82 −116.4 209 140.3 0.02 50.9 2.3 0.69
Encinitas—Village Park &Mtn. Vista 13141 CGS 33.054 −117.245 515 140.9 0.02 33.3 1.3 0.70
Cathedral City—Ramon & Desert Vista 12108 CGS 33.815 −116.461 209 142.1 0.04 37.0 3.6 0.55
Idyllwild—Keenwild Fire Sta. 5232 USGS 33.7077 −116.7174 845.4 143.9 0.01 21.0 0.1 0.57
Palm Springs—Airport 12025 CGS 33.829 −116.512 207.5 145.5 0.03 39.8 2.6 0.61
Palm Springs Desert Museum PSD SCSN 33.824 −116.55 354 146.6 0.01 47.7 0.2 0.39
Idyllwild—Hwy. 243 & Pine Crest 12116 CGS 33.747 −116.715 684.9 147.3 0.04 7.5 2.5 0.27
Blythe—Fire Station 10021 CGS 33.61 −114.715 354 147.6 0.03 70.9 4.3 0.96
Sage—Fire Station A 12188 CGS 33.605 −116.938 748 148.6 0.01 36.9 0.1 0.52
(continued )
185
186
1
Sky Valley—Fire Station #56 5435 USGS 33.9168 −116.3898 345.4 149.9 0.02 30.2 0.7 0.49
Fun Valley 5069 USGS 33.9252 −116.3903 345.4 150.7 0.02 31.4 0.9 0.42
West Wide Canyon WWC SCSN 33.941 −116.409 748 153 0.01 35.2 0.5 0.38
Oceanside B—Fire Station 13333 CGS 33.201 −117.331 515 154.1 0.02 39.2 1.4 0.85
Temecula—6th & Mercedes 13172 CGS 33.497 −117.151 370.8 155.8 0.03 52.8 3.5 1.05
Murietta Hot Springs—Skinner Dam 720 USGS 33.583 −117.072 748 156 0.02 29.2 0.5 0.37
North Palm Springs Fire Station #36 5295 USGS 33.9247 −116.5479 345.4 156.5 0.03 25.0 2.3 0.48
Domenigoni Reservoir DGR SCSN 33.65 −117.009 748 156.8 0.01 38.9 0.3 0.48
Oceanside—Fire Station No. 1 5430 USGS 33.198 −117.378 387 158.1 0.01 35.4 0.5 1.18
Desert Hot Springs—Fire Station 12149 CGS 33.962 −116.509 345.4 158.7 0.02 31.6 1.3 0.54
Hemet—Acacia & Stanford 12923 CGS 33.745 −116.932 338.5 159.6 0.04 47.5 3.3 0.70
Banning, Hwy. 243, San Jacinto Mtns. SLR SCSN 33.834 −116.797 748 159.6 0.03 30.1 1.0 0.34
Hemet—Stetson Ave Fire Station 12331 CGS 33.729 −116.98 338.5 161.2 0.04 38.8 3.5 0.52
San Jacinto—Valley Cemetery 12202 CGS 33.76 −116.961 338.5 162.6 0.03 42.7 2.0 0.56
Silent Valley—Poppet Flat 12206 CGS 33.851 −116.853 684.9 164.2 0.01 25.8 0.2 0.52
San Jacinto—CDF Fire Station 25 12102 CGS 33.787 −116.959 271.4 164.7 0.05 53.0 6.5 0.90
San Jacinto—CDF Fire Station 12673 CGS 33.787 −116.959 271.4 164.7 0.05 56.3 6.5 0.93
Hemet—Cawston& Devonshire 13093 CGS 33.751 −117.015 338.5 165.2 0.05 29.2 4.8 0.61
Black Rock Canyon Campground 2 BLA2 SCSN 34.069 −116.39 387 165.7 0.01 101.5 0.1 0.93
Jodan Farms Nursery, Murrieta MUR SCSN 33.6 −117.195 748 166 0.03 37.2 1.6 0.33
Cabazon 5073 USGS 33.9176 −116.7833 345.4 166.6 0.02 33.1 1.0 0.70
Mt. San Jacinto Campus MSJ SCSN 33.808 −116.968 280 167 0.06 56.9 8.5 0.82
San Jacinto—MWD West Portal 5289 USGS 33.8218 −116.9684 338.5 168.2 0.01 48.7 0.8 0.73
Murrieta—Clinton Keith & Bear Creek 13081 CGS 33.565 −117.264 748 168.7 0.01 32.8 0.3 0.51
Morongo Valley 5071 USGS 34.0486 −116.5782 345.4 170.1 0.04 19.9 2.2 0.63
Menifee Valley—Murrieta & Scott 13929 CGS 33.649 −117.205 503 170.2 0.01 35.4 0.3 0.32
Joshua Tree—Fire Station 22170 CGS 34.131 −116.315 379.3 170.2 0.01 35.9 0.6 0.72
Blythe BLY SCSN 33.75 −114.524 748 171.1 0.01 79.2 0.2 0.80
Yucca Valley—Hwys. 62 & 247 22960 CGS 34.122 −116.416 354 172.1 0.01 29.5 0.3 0.97
Homeland—Hwy. 74 & Sultanas 13924 CGS 33.748 −117.127 338.5 172.2 0.03 27.5 1.1 0.26
Wildomar—Palomar &Gruwell 13097 CGS 33.604 −117.279 349 172.4 0.03 46.1 2.5 0.66
Sun City—I215 & McCall Blvd. 13930 CGS 33.715 −117.191 338.5 174 0.02 27.0 1.1 0.45
Beaumont, I-10 at 79 BBS SCSN 33.921 −116.981 387 177.5 0.02 56.0 0.9 0.87
Beaumont—6th & Maple 12919 CGS 33.93 −116.973 370.8 177.8 0.02 32.1 1.9 0.81
Canyon Lake—Vacation Dr. & San Joaq 13096 CGS 33.699 −117.266 424.8 178.1 0.01 34.0 0.3 0.31
Nuevo—11th & Lakeview 13095 CGS 33.818 −117.133 684.9 178.1 0.03 20.7 1.9 0.58
Pioneertown—Bronco & Wyandot 22078 CGS 34.155 −116.504 387 178.3 0.01 24.9 0.3 0.51
Snow Peak SNO SCSN 34.035 −116.808 748 178.8 0.01 79.4 0.3 0.85
Lake Elsinore—Graham & Poe 13922 CGS 33.669 −117.332 622.9 180.7 0.01 33.7 0.2 0.92
Perris—San Jacinto & C St. 13928 CGS 33.787 −117.23 518 182.1 0.01 28.1 0.1 0.47
Moreno Valley—Alessandro & Moreno 13927 CGS 33.921 −117.173 338.5 189 0.02 30.0 1.4 0.38
Beach
Yucaipa Valley—Calimesa & Co. Line Rd. 23920 CGS 34.004 −117.059 370.8 189.2 0.03 33.4 0.6 0.78
Mill Creek Ranger Station 5076 USGS 34.0795 −117.0448 370.8 195.1 0.02 21.0 1.2 0.60
Mentone Fire Station #9 5162 USGS 34.0702 −117.1211 271.4 198.5 0.01 34.4 0.4 0.85
Reche Canyon—Olive Dell Ranch 5037 USGS 34.0047 −117.2239 488 199 0.01 24.4 0.2 0.64
Notes: 1USGS = United States Geological Surveys; CGS = California Geological Survey; RANM = the Red de AcelerografsdelNoroeste de Mexico (RANM) network; SCSN =
Southern California Seismic Network.
2
Average shear wave velocity in the upper 30 meters. Underlined values are measured and others are estimated using the statistical relationships between V S30 and geological units
proposed by Wills and Clahan (2006).
3
Closest distance to the faul trupture plane.
4
Peak ground acceleration. Geometric mean of two as-recorded components.

5
Significant duration. The time interval between 5% and 95% of the Arias intensity.
6
Arias intensity. For a definition see Arias (1970).
7
Mean period. For its definition see Rathje et al. (2004).
187
(a)
(b)
Figure 3. (a) Contours of depth (in meters) to shear wave velocity of 1,000 m/s, and (b) contours
of depth to shear wave velocity of 2,500 m/s in the Imperial Valley (adapted from Caltrans 2010).
FN and FP directions based on a fault strike of N130°E (Wei et al. 2011). It is noted that
velocity and displacement waveforms are rich in significant long-period energy content at
periods of about 6 to 7 seconds. This long-period energy was observed throughout the
Imperial Valley but not in the Mexicali Valley. Figure 4b indicates no polarization of ground
motion in the FN direction. Data from other stations that could have developed rupture
forward directivity were also examined, and no clear polarization of ground motion in
the FN direction was observed. This finding is in agreement with the GEER report
(GEER 2010). Figure 4b also clearly illustrates that long-period energy was observed at
Station 931 but not at Station MDO.
STRONG GROUND MOTION PARAMETERS

Ground motion parameters of the records obtained during the main shock from the
4 April 2010 El Mayor–Cucapah were evaluated in this study, including spectra acceleration,
frequency content, Arias intensity, and ground motion duration. The results of some
important ground motion parameters are summarized in Table 1. Whenever applicable,
these parameters are compared with the predictions by available GMPEs or attenuation
relationships. Because the 1979 Imperial Valley main shock and the 2010 El Mayor–
Cucapah earthquake bear many similarities, spectral accelerations from the 1979 Imperial
Valley main shock are also compared with the BA08 and the SEA99 GMPEs.
During the comparison of observed spectral accelerations with the BA08 and the SEA99
GMPEs, intra-event residuals for each ground motion station considering the appropriate
site-source distance and local site conditions were calculated to more accurately evaluate
the predictability of the GMPE as follows (e.g., Margaris et al. 2010):
ðRÞi ¼ lnðIM i Þrec − lnðIM i ÞGMPE

1 PN
η¼ ðRÞi (2)
N i¼1
ðRÞintra
i ¼ ðRÞi − η (3)
where ðRÞi is the value of the total residual from recording i; ðIM i Þrec is the value of ground-
motion intensity measure from recording i; ðIM i ÞGMPE is the median value of that same IM
from the appropriate GMPE; N is the number of ground-motion intensity measures; η is the
value of the inter-event residual for the earthquake; ðRÞintra
i is the value of the intra-event
residual from recording i. The ground-motion intensity measures used here are PGA and
spectral accelerations at 0.3-, 1.0-, and 2.0-second periods.
COMPARISON TO BOORE AND ATKINSON 2008 GMPE

In the past decade, under a PEER’s NGA project, five teams have developed and pre-
sented their respective GMPEs for shallow crustal earthquakes in western North America,
namely, AS08, BA08, CB08, CY08, and Idriss 2008 (I08). With the exception of I08, these
GMPEs require V S30 as an independent input to consider the effect of the local site condition.
Station MDO Station 931
0.5
FN FN
Acc. (g)
-0.5
0.5
Acc. (g)
FP FP
0
-0.5
70
Vel. (cm/s)
FN FN
0
-70
70
Vel. (cm/s)
FP FP
0
-70
60
Disp. (cm)
FN FN
0
-60
60
Disp. (cm)
FP FP
0
-60
0 50 100 150 0 50 100 150
Time (s) (a) Time (s)
10 10
1 1
Sa (g)
Sa (g)
0.1 0.1
MDO - FP 931 - FP
MDO - FN 931 - FN
0.01 0.01
0.01 0.1 1 10 0.01 0.1 1 10
Period (s) Period (s)
(b)
Figure 4. (a) Time history plots of Station MDO (IBC Site Class D, Rrup ¼ 17.8 km), and
Station 931 (El Centro Array #12, IBC site class D, Rrup ¼ 15.8 km), (b) plots of fault normal
(FN) and fault parallel (FP) response spectra for Stations MDO and 931.
Figure 5 shows plots of 5% damped pseudo-spectral accelerations at periods of PGA

(0.01), 0.3, 1.0, and 2.0 seconds versus Rjb distances. For comparison purposes, the median
and the 16th- and 84th-percentile predictions of the BA08GMPE are also plotted in Figure 5
for V S30 ¼ 250 m=s, for V S30 ¼ 425 m=s, and for V S30 ¼ 600 m=s. Comparisons in Figure 5
cannot reflect the actual local site conditions at ground motion stations due to a single V S30
value being used for each comparison.
Intra-event residuals of the aforementioned NGA GMPEs (except the I08 GMPE) with
respect to the 2010 El Mayor–Cucapah earthquake were investigated by Liao and Meneses
(2012) in terms of intra-event residuals with respect to V S30 and Rjb. They found that spectral
accelerations are in general underpredicted by the four NGA GMPEs at small distances from
10
PGA (g)
0.1
0.001
10
Sa (T = 0.3 s) (g)
0.1
0.001
10
Sa (T = 1.0 s) (g)
0.1
0.001
10
Sa (T = 2.0 s) (g)
0.1
0.001
10 100 10 100 10 100
Rjb (km) Rjb (km) Rjb (km)
(a) (b) (c)
Figure 5. Recorded spectral accelerations (5% damping) from 2010 El Mayor–Cucapah earth-
quake plotted as a function of Rjb. For reference, the median and the 16th and 84th percentile
predictions of the BA08GMPE are also shown: (a) V S30 ¼ 250 m=s; (b) V S30 ¼ 425 m=s;
(c) V S30 ¼ 600 m=s.
about 10 to 30 km and are overpredicted at large distances from about 80 to 200 km; they also
found that intra-event residual of the NGA GMPEs exhibit a negative trend with distance,
which is suggestive of faster attenuation in the 2010 earthquake data than in the NGA
GMPEs; the intra-event residuals were found to show no noticeable trend with spectral per-
iods less than 4.0 seconds.
COMPARISON TO SEA99 GMPE

Spudich et al. (1999) developed the SEA99 GMPE, which is appropriate for predicting
earthquake ground motions in extensional tectonic regimes. Ground motions in extensional
tectonic regimes appear to be lower than other tectonic regimes. The dataset used for
developing the SEA99 GMPE includes more than 30 ground motions recorded during the
1979 Imperial Valley main shock, which occurred on the Imperial fault. Figure 6 shows the
location of the epicenter of the 1979 Imperial Valley main shock as well as the location of
the rupture and distribution of shaking intensities. The 1979 Imperial Valley main shock
Figure 6. Locations of the epicenter (start) and fault rupture (solid line), distribution of shaking
intensities of 1979 Imperial Valley main shock (after http://www.cisn.org/shakemap/sc/shake/
Imperial_Valley/intensity.html).
and the 2010 El Mayor–Cucapah earthquake bear many similarities. The 1979 Imperial Valley
main shock occurred on the Imperial fault with a strike of N143°E (Archuleta 1984), very similar
in orientation to the fault strike of N130°E for the 2010 El Mayor–Cucapah earthquake. Both
earthquakes occurred on the primarily strike-slip faults. As such, observed spectral accelera-
tions during the 1979 Imperial Valley main shock are compared to the predictions by the SEA99
GMPE as well as the BA08 GMPE. The same ground motions with Rjb distances less than
100 km as used in Spudich et al. (1999) are downloaded from 2005 PEER Ground Motion
Database (PEER 2005), and have been processed following the protocols of the NGA project.
Figure 7 shows plots of 5% damped pseudo-spectral acceleration at periods of PGA, 0.3,
1.0, and 2.0 seconds versus the Rjb distance. For comparison purposes, the median and the
16th- and 84th-percentile predictions of the SEA99 GMPE are also plotted in Figure 7 for soil
and rock sites. The gray solid and dashed lines represent the predictions by the SEA99 GMPE
for an Mw 6.5 earthquake, whereas the black solid and dashed lines represent the predictions
by the SEA99 GMPE for an Mw 7.2 earthquake. As seen in Figure 7, the SEA99 GMPE
predicts pseudo-spectral accelerations without significant bias with distances for the 1979
Imperial Valley earthquake. However, the SEA99 GMPE underpredicts pseudo-spectral
accelerations with distances from about 10 to 40 km for the 2010 El Mayor–Cucapah
earthquake. This similar trend is observed for the predictions by the BA08 GMPE relative
to the 2010 El Mayor–Cucapah earthquake.
Figure 8 presents the intra-event residuals of the SEA99 GMPE as well as the BA08
GMPE. As it can be seen, residuals of the SEA99 and the BA08 GMPEs do not show a
noticeable trend with distance for the 1979 Imperial Valley earthquake. However, residuals
of the SEA99 GMPE exhibit a negative trend with distances for the 2010 El Mayor–Cucapah
earthquake, which suggests faster attenuation in the 2010 earthquake data than in the SEA99
GMPE. Intra-event residuals of both the SEA99 and the BA08 GMPEs relative to the 2010 El
Mayor–Cucapah earthquake exhibit larger scattering than the 1979 Imperial Valley
earthquake. In general, one earthquake can have higher variability in ground motions
than another earthquake due to differences in the rupture pattern (both temporal and spatial)
or in the local wave propagation. The larger scattering observed in the 2010 data but not in the
1979 data can be mainly attributed to the local wave propagation effects (e.g., basin effects)
on the ground motions in the area of the Los Angeles Basin and its surroundings.
The median intra-event residuals along with the 95% confidence intervals with respect to
the spectral period for the SEA99 GMPE are presented in Figure 9. As it can be seen, the
SEA99 GMPE is not biased relative to the data with spectral periods less than 4.0 seconds.
Together with the study by Liao and Meneses (2012), both the SEA99 and the BA08 GMPEs
provide unbiased estimates of pseudo-spectral accelerations with spectral periods of less than
4.0 seconds.
COMPARISON TO FREQUENCY CONTENT PREDICTION EQUATION

The dynamic response of geotechnical and structural systems is significantly affected by
the frequency content of an earthquake ground motion. The resonance occurs when the fre-
quency content of an earthquake ground motion closely matches the natural period of a geo-
technical or a structural system. As a consequence, larger seismic forces are exerted on the
system and more severe damage may occur (e.g., Kramer 1996). The frequency content of
10
PGA (g)
0.1
2010 Earthquake 2010 Earthquake
0.001
10
Sa (T = 0.3 s) (g)
0.1
0.001
10
Sa (T = 1.0 s) (g)
0.1
0.001
10
Sa (T = 2.0 s) (g)
0.1

1979 Earthquake
1979 Earthquake
0.001
0.1 1 10 100 0.1 1 10 100
Rjb (km) Rjb (km)
(a) Soil (b) Rock
Figure 7. Recorded spectra acceleration (5% damping) from the 2010 El Mayor–Cucapah earth-
quake and 1979 Imperial Valley earthquake plotted as a function of Rjb. For reference, the median
(gray solid line: 1979 Imperial Valley earthquake; black solid line: 2010 El Mayor–Cucapah
earthquake) and the 16th and 84th percentile (gray dashed lines: 1979 Imperial Valley earth-
quake; black dashed lines: 2010 El Mayor–Cucapah earthquake) predictions of the SEA99
GMPE are also shown: (a) Soil; (b) Rock.
ground motions is most completely characterized by Fourier amplitude spectra (e.g., Brune
1970, 1971) or acceleration response spectra (e.g., Newmark and Hall 1982). However, it is
often useful to characterize the frequency content of an earthquake ground motion with a
single scalar parameter that can be easily compared with the natural period of a system
to evaluate the possibility of resonance conditions. Such scalar frequency content parameters
have been incorporated into seismic design procedures for landslides and earthfills
(e.g., Blake et al. 2002, Stewart et al. 2003).
2
PGA PGA
Intra-event Residual
-1 2010 Earthquake 2010 Earthquake

-2
2
T = 0.3 s T = 0.3 s

-2
2
T = 1.0 s T = 1.0 s
1

-2
2
T = 2.0 s T = 2.0 s

-2
0.1 1 10 100 200 0.1 1 10 100 200
Rjb (km) Rjb (km)
(a) (b)
Figure 8. Plots of intra-event residuals for (a) the SEA99 and (b) the BA08 GMPEs showing their
distribution with respect to Rjb. The dashed lines in each plot are plus/minus one standard deviation
with respect to the SEA99 GMPE and the BA08 GMPE, respectively (2010 earthquake: 2010 El
Mayor–Cucapah earthquake; 1979 earthquake: 1979 Imperial Valley earthquake).
Various scalar parameters have been proposed to characterize the frequency content of
strong ground motions (e.g., Schnabel 1973, Vanmarcke 1976, Boore 1983, Rathje et al.
1998, 2004). Rathje et al. (1998, 2004) discussed the relative advantages and disadvantages
of several single valued parameters—namely, the predominant period (T p), the mean period
(T m ), the smoothed spectral predominant period (T o), and the average spectral period
(T avg )—and developed empirical predictive equations for them. They concluded that T m
can be better predicted because it best distinguishes the frequency content of strong ground
Median (2010 Earthquake)

95% C.I. (2010 Earthquake)
0.6
Median (1979 Earthquake)
95% C.I. (1979 Earthquake)
0.3
0.0
-0.3
-0.6
0.01 0.1 1 2
Period (s)
Figure 9. Median intra-event residuals and 95% confidence interval (C.I.) with respect to
SEA99 GMPE.
motions. Note that the simplified geotechnical site (SCS) classification system developed by
Rodriguez-Marek et al. (2001) was adopted for developing these equations. The SCS system
differentiates between rock (soil depth < 6 m), shallow stiff soil (soil depth < 60 m), and deep
stiff soil (soil depth > 60 m).
Figure 10 shows the calculated T m values of the recorded ground motions compared to
the predictive relationship proposed by Rathje et al. (2004). The comparisons suggest that the
predictive equation of T m provided a good match to the data trends at SCS site class D, but
overestimated about half of the observed values at SCS siteclass C by approximately one
standard deviation. Such an overestimation may be an indication that these recordings
are particularly rich in high-frequency content.
COMPARISON TO ARIAS INTENSITY PREDICTION EQUATION

Arias intensity (I a ) has been recognized to correlate well with several commonly used
demand measures of structural performance, liquefaction, and seismic slope stability. Several
attenuation relationships have been developed in the past decades (Wilson and Keefer 1985,
Kayen and Mitchell 1997, Paciello et al. 2000, Travasarou et al. 2003).
Values of I a calculated from the ground motions recorded from the main shock of this
earthquake are compared to the Travasarou et al. (2003) relationship in Figure 11. This rela-
tionship adopted the SCS classification system as well. It appears that the Travasarou et al.
relationship predicts a good match to the data trends at distances from approximately 50 to
200 km for SCS site classes B and C. For SCS site class D, the Travasarou et al. relationship
provides a good match only at intermediate and long distances, from approximately 50 to
200 km. For stations with distances from approximately 10 to 50 km, the I a values of
recorded ground motions are at least one standard deviation above the median of the I a
prediction. The misfit increases with decreasing distance. These stations at short, intermedi-
ate, and long distances have very similar V S30 values with an average of approximately
1
Tm (s)
SCS Site Class D SCS Site Class C

0.1
10 100 200 10 100 200
Rrup (km) Rrup (km)
1
Tm (s)
SCS Site Class B

0.1
10 100 200
Rrup (km)
Figure 10. Recorded T m values of the 2010 El Mayor–Cucapah earthquake plotted as a function
of Rrup. For reference, the median and plus/minus one standard deviation predictions of the Rathje
et al. (2004) relationships are also shown.
200 m/s at short distances and 230 m/s at intermediate and large distances. This underpredic-
tion of the I a values for site class D at small distances is likely to be a result of higher soil
amplification at higher shaking level and at short distances.
COMPARISON TO SIGNIFICANT DURATION PREDICTION EQUATION

Ground motion duration has been incorporated into seismic performance design pro-
cedures for seismic slope displacements (e.g., Bray and Rathje 1998) and for liquefaction-
induced lateral spread (Rauch and Martin 2000). Ground motion duration is often
quantitatively represented by the significant duration ðD5−95 Þ, which measures the time interval
between accumulation of 5% and 95% of the Arias intensity.
The values of ðD5−95 Þ of the recorded ground motions are compared to the Kempton and
Stewart (2006) significant-duration prediction equation (Figure 12). This equation was devel-
oped to predict the significant duration of an earthquake ground motion as a function of
magnitude, closest site-source distance, near-surface shear wave velocity, basin depth,
and near-fault parameters. As shown in Figure 12, the comparison shows that the Kempton
and Stewart equation provided good predictions to the observed significant duration values
for sites with V S30 greater than 250 m/s but underpredicted for sites with V S30 less than
250 m/s.
1000
Ia (cm/s)
10
SCS Site Class D SCS Site Class C

0.1
10 100 200 10 100 200
Rrup (km) Rrup (km)
1000
Ia (cm/s)
10
SCS Site Class B

0.1
10 100 200
Rrup (km)
Figure 11. Recorded Arias intensity values of the 2010 El Mayor–Cucapah earthquake plotted as
a function of Rrup. For reference, the median and plus/minus one standard deviation predictions of
the Travasarou et al. (2003) relationships are also shown.
SITE RESPONSE CHARACTERISTICS

Nakamura (1989) proposed a methodology of using the H/V ratio of Fourier amplitude
spectra to estimate local site response characteristics. Although the absolute level of the H/V
ratio depends on the wave field and source mechanism, extensive studies (e.g., Field 1996,
Theodulidis et al. 1996, Yamazaki and Ansary 1997, Bonilla et al. 1997, Satoh et al. 2001,
Rodríguez and Midorikawa 2003, Rodriguez-Marek et al. 2010, Meslem et al. 2010, Celebi
et al. 2010) have demonstrated that the H/V ratio shape shows good stability and is
insensitive to source location and mechanism. If a clear impedance contrast exists at the
site, the H/V ratio has been shown to provide an adequate estimate of the predominant period
of a site, which is an important parameter for assessing building damage (Fellahiet al. 2003,
Gosar 2007) and in evaluating local site amplification. In this study, the H/V ratio technique
was used to investigate the site response characteristics.
The main shock of the 4 April 2010 El Mayor–Cucapah earthquake generated a total of
36 ground motion records from stations located on soft-soil sites (V S30 ranging from about
180 to 250 m/s) with Rrup ranging from about 14 to 150 km. On the other hand, a total of 24
ground motions were recorded on rock sites ðV S30 ¼ 748 m=sÞ. These rock sites are between
about 44 and 200 km from the fault rupture plane. The locations of these soft-soil and rock
sites are presented in Figure 2. The H/V ratio is calculated as the Fourier amplitude spectral
120
Vs30 < 250 m/s Vs30 = 250 - 600 m/s
80
D5-95 (s)
40
0
10 100 200 10 100 200
Rrup (km) Rrup (km)
120
Vs30 > 600 m/s
80
D5-95 (s)
40
0
10 100 200
Rrup (km)
Figure 12. Recorded significant durations of the 2010 El Mayor–Cucapah earthquake plotted as
a function of Rrup. For reference, the median and plus/minus one standard deviation predictions of
the Kempton and Stewart (2006) prediction equation are also shown for different V S30 ranges.
ratio between the geometric mean of the two horizontal components and the vertical
component given by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
H FAh1 ðTÞ ⋅ FAh2 ðTÞ
¼ (4)
V FAv ðTÞ
where FAh1 ðTÞ, FAh2 ðTÞ, and FAv ðTÞ are the smoothed Fourier amplitude spectra for the two
horizontal and vertical components of a ground motion record, respectively. T is the spectral
period. These Fourier amplitude spectra were smoothed by a Parzen window of 0.4 Hz
bandwidth.
Figure 13a shows plots of H/V ratios of Fourier amplitude spectra of 36 ground motions
on the soft-soil sites along with the geometric mean and mean plus/minus one standard
deviation lines. Overall, these soft-soil sites appear to show broadband predominant
site periods ranging from around 0.4 to 2.0 seconds corresponding to the peak of the geo-
metric mean H/V ratio curve. Nonetheless, a closer examination of the H/V ratio curves
suggests that there appear to be double peaks, with the first peak occurring from about 0.4
to 0.7 seconds followed by a second peak from about 1.0 to 2.0 seconds. In general, a
typical predominant site period of about 0.6 to 0.8 seconds should be anticipated for a
similar soft-soil site with V S30 of between 180 and 250 m/s (Zhao et al. 2006). The observed
lengthening in the predominant site period for some of the soft-soil sites is likely caused by
the nonlinear behavior of soil deposits under a relatively high level of excitation. The
10 10
H/V Ratio
H/V Ratio
1 1
0.1 0.1
0.01 0.1 1 4 0.01 0.1 1 4
Period (s) Period (s)
(a) (b)
10
Rock Sites
Soil Sites with Vs30 = 180 - 250 m/s
H/V Ratio
0.1
0.01 0.1 1 4
Period (s)
(c)
Figure 13. H/V ratios of Fourier amplitude spectra of the 2010 El Mayor–Cucapah earthquake:
(a) all soft sites with V S30 ¼ 180 – 250 m/s; (b) rock sites with V S30 ¼ 748 m/s; thick lines
represent the geometric mean H/V ratios and dashed lines represent the plus/minus one standard
deviations of the H/V ratios; and, (c) geometric mean H/V ratios for the soft-soil sites and for the
rock sites.
predominant site period has been observed to be lengthened when the soil deposit behaves
more nonlinearly during a strong shaking (e.g., Wen et al. 1994, Zeghal and Elgamal 1994,
Aguirre and Irikura 1997). Figure 13b plots H/V ratios of Fourier amplitude spectra of 24
ground motions on rock sites along with the geometric mean and mean plus/minus one
standard deviation lines. In general, these rock sites showed no pronounced predominant
site period despite the presence of a weak H/V ratio peak at 0.2 seconds, which is typical for
a rock site. For ease of comparison, the geometric mean H/V ratios for soft-soil sites and for
rock sites are plotted in Figure 13c. It is noted that the predominant site periods, amplitudes,
and shapes of the H/V ratios are remarkably different between the two site classes, which
revalidates the capability of the H/V ratio to estimate the predominant period of soil
deposits.
SUMMARY AND CONCLUSIONS

The main shock of the Mw 7.2, 4 April 2010 El Mayor–Cucapah earthquake generated a
number of ground motion recordings at soft-soil sites, most of them located in the Imperial
Valley and Mexicali Valley characterized by deep, soft young sediments with V S30 of less
than 250 m/s. These recordings are very valuable for studying the effect of soft soil on site
response characteristics. Spectral accelerations evaluated from the recorded ground motions
were compared with the BA08 GMPE. In general, the BA08 GMPE underpredicts observed
spectral accelerations at soil sites with V S30 less than 250 m/s over the distance range from
about 10 to 30 km but overpredicts at the site with V S30 greater than 250 m/s over distances
from about 80 to 200 km. Liao and Meneses (2012) observed that intra-event residuals of the
NGA GMPEs with respect to the 2010 El Mayor–Cucapah earthquake show a negative trend
with distance for sites with V S30 less than 250 m/s but no noticeable trend over the V S30 range
and spectral periods less than 4.0 seconds.
The recorded spectral accelerations from the 2010 El Mayor–Cucapah earthquake as well
as the 1979 Imperial Valley main shock are also compared with the SEA99 GMPE. In
general, both the SEA99 and the BA08 GMPEs predict spectral accelerations with distance
with no significant bias for the 1979 Imperial Valley earthquake. Intra-event residuals of both
the SEA99 and the BA08 GMPEs do not show a noticeable trend with distance for the 1979
Imperial Valley earthquake. However, residuals of both the SEA99 and BA08 GMPEs
exhibit a negative trend with distance for the 2010 El Mayor–Cucapah earthquake. Data
from the 2010 El Mayor–Cucapah earthquake exhibit larger scattering than data from the
1979 Imperial Valley earthquake, which can be mainly attributed to the local wave propaga-
tion effects (e.g., basin effects) on the ground motions in the Los Angeles Basin and its
surroundings.
Comparing the calculated T m values to the Rathje et al. (2004) relationship suggests that
this relationship generally provides a good match to the data trends at SCS site class D but
likely provides overestimation at SCS site classes B and C. The Travasarou et al. (2003)
relationship predicts a good match to the data trends at distances from approximately 50
to 200 km for SCS site classes B and C. For SCS site class D, the Travasarou et al. prediction
provides a good match only at intermediate and long distances from approximately 50 to
200 km. For stations within about 50 km, the I a (Arias intensity) values of recorded ground
motions are at least one standard deviation above the median of the I a prediction. The misfit
increases with decreasing distance. The comparison of the computed ðD5−95 Þ shows that the
Kempton and Stewart (2006) prediction equation predicts good estimates of significant dura-
tion of ground motions for sites with V S30 greater than 250 m/s but may underpredict for sites
with V S30 less than 250 m/s.
Results of H/V ratios reveal that some soft-soil sites with V S30 between 180 and 250 m/s
show a predominant site period of about 0.4 to 0.7 seconds but others show lengthened pre-
dominant site periods of about 1.0 to 2.0 seconds, which likely results from the nonlinear
behavior of soil deposits under a high level of excitation. Rock sites show no pronounced
predominant site period despite the presence of a weak H/V ratio peak at 0.2 seconds. The
remarkably different predominant site periods, amplitudes, and shapes of the H/V spectral
ratios between the two sites revalidate the capability of the H/V spectral ratio method for
estimating predominant site periods, which is an important parameter for assessing building
damage.
ACKNOWLEDGMENTS
The authors gratefully appreciate insightful comments and suggestions from Professor
Adrian Rodriguez-Marek of Virginia Tech University and three anonymous reviewers.
Their comments and suggestions led to significant improvement of the manuscript. Any opi-
nions, findings, and conclusions expressed in this material are those of the authors.
REFERENCES
Abrahamson, N.A., and Silva, W.J., 2008. Summary of the Abrahamson and Silva NGA ground-
motion relations, Earthquake Spectra 24, 67–97.
Aguirre, J., and Irikura, K., 1997. Nonlinearity, liquefaction, and velocity variation of soft
soil layers in Port Island, Kobe, during the HyogokenNanbu earthquake, Bulletin of the
Seismological Society of America 87, 1244–1258.
Archuleta, R.J., 1984. A faulting model for the 1979 Imperial Valley earthquake, Journal of
Geophysical Research 89, 4559–4585.
Blake, T. F., Hollingsworth, R. A., and Stewart, J. P., 2002. Recommended Procedures for
Implementation of DMG Special Publication 117: Guidelines for Analyzing and Mitigating
Landslide Hazards in California, Southern California Earthquake Center, Los Angeles, CA.
Bonilla, L. F., Steidl, J. H., Lindley, G. T., Tumarkin, A. G., and Archuleta, R. J., 1997. Site
amplification in the San Fernando Valley, California: Variability of site-effect estimation
using the S-wave, coda, and H/V methods, Bulletin of the Seismological Society of America
87, 710–730.
Borcherdt, R. D., 1994. Estimates of site-dependent response spectra for design (methodology
and justification), Earthquake Spectra 10, 617–653.
Boore, D. M., 1983. Stochastic simulation of high-frequency ground motions based on
seismological models of the radiated spectra, Bulletin of the Seismological Society of America
73, 1865–1894.
Boore, D.M., and Atkinson, G.M., 2008. Ground-motion prediction equations for the average
horizontal component of PGA, PGV, and 5%-damped PSA at spectral periods between
0.01 s and 10.0 s, Earthquake Spectra 24, 99–138.
Boore, D. M., Joyner, W., and Fumal, T., 1997. Equations for estimating horizontal response
spectral and peak acceleration from western North American earthquakes: A summary of
recent work, Seismological Research Letters 68, 128–153.
Bray, J.D., and Rathje, A.M., 1998. Earthquake-induced displacements of solid-waste landfills,
ASCE Journal of Geotechnical Engineering 124, 242–253.
Brune, J. N., 1970. Tectonic stress and the spectra of seismic shear waves from earthquakes,
Journal of Geophysics Research 75, 611–614.
Brune, J. N., 1971. Correction, Journal of Geophysics Research 76, 5002.
California Department of Transportation (Caltrans), 2010. Caltrans Seismic Design Criteria, ver-
sion 1.6, Sacramento, CA.
California Integrated Seismic Network (CISN), 2011. TriNet ShakeMap: Estimated instrumental
intensity Web page, http://www.cisn.org/shakemap/sc/shake/Imperial_Valley/intensity.html,
accessed 2 December 2011.
Campbell, K. W., and Bozorgnia, Y., 2008. NGA ground motion model for the geometric mean
horizontal component of PGA, PGV, PGD and 5% damped linear elastic response spectra for
periods ranging from 0.01 to 10 s, Earthquake Spectra 24, 139–171.
Celebi, M., Bazzuro, P., Chiaraluce, L., Clemente, P., Decanini, L., Desortis, A., Ellsworth, W.,
Gorini, A., Kalkan, E., Marcucci, S., Milana, G., Mollaioli, F., Olivieri, M., Paolucci, R.,
Rinaldis, D., Rovelli, A., Sabetta, F., and Stephens, C., 2010. Recorded motions of the 6
April 2009 Mw 6.3 L’Aquilla, Italy earthquake and implications for building structural
damage: Overview, Earthquake Spectra 26, 651–684.
Center for Engineering Strong Motion Data (CESMD), 2010. Website, http://www
.strongmotioncenter.org, accessed 14 August 2010.
Centro de InvestigaciónCientífica y de Educación Superior de Ensenada (CICESE), 2010.
Website, http://www.cicese.edu.mx, accessed 1 December 2010.
Chiou, B. S, and Youngs, R. R., 2008. An NGA model for the average horizontal component of
peak ground motion and response spectra, Earthquake Spectra 24, 173–215.
Earthquake Engineering Research Institute (EERI), 2010. Reconnaissance Report on the 2010,
Mw 7.2 El Mayor–Cucapah Earthquake (J. Meneses et al., eds.), Report No. 2010–02.
Fellahi, A., Alaghebandian, R., and Miyajima, M., 2003. Microtremor measurements and
building damage during the Changureh-Avaj, Iran, earthquake of June 2002, Journal of
Natural Disaster Science 25, 37–46.
Field, E. H., 1996. Spectral amplification in a sediment-filled valley exhibiting clear basin-edge
induced waves, Bulletin of Seismological Society of America 86, 991–1005.
Geotechnical Extreme Event Reconnaissance (GEER), 2010. Preliminary Report on Seismolo-
gical and Geotechnical Engineering Aspects of the April 4, 2010 Mw 7.2 El Mayor–Cucapah
(Mexico) Earthquake (J. Stewart et al., eds.), Report No.GEER–023.
Gosar, A., 2007. Microtremor HVSR study for assessing site effects in the Bovec Basin (NW
Slovenia) related to 1998 Mw 5.6 Earthquakes, Engineering Geology 91, 178–193.
Idriss, I. M., 1991. Earthquake ground motions at soft soil sites, Proceedings of the 2nd
International Conference on Recent Advances in Geotechnical Earthquake engineering
and Soil Dynamics, St. Louis, MO, 2265–2272.
Idriss, I. M., 2008. An NGA empirical model for estimating the horizontal spectral values gen-
erated by shallow crustal earthquakes, Earthquake Spectra 24, 217–242.
Kayen, R. E., and Mitchell, J. K., 1997. Assessment of liquefaction potential during earthquake
by Arias intensity, ASCE Journal of Geotechnical and Geoenvironmental Engineering 123,
1162–1174.
Kempton, J. J., and Stewart, J. P., 2006. Prediction equations for significant duration of
earthquake ground motions considering site and near-source effects, Earthquake Spectra
22, 985–1013.
Kramer, S. L., 1996. Geotechnical Earthquake Engineering, Prentice Hall, Upper Saddle River,
NJ, 653 pp.
Liao, Y., and Meneses, J., 2012. Comparison of ground motions from the 2010 Mw 7.2 El
Mayor–Cucapah earthquake with the Next Generation Attenuation ground motion prediction
equations, Bulletin of Earthquake Engineering, DOI: 10.1007/s10518-012-9358-7.
Margaris, B., Athanasopoulos, G., Mylonakis, G., Papaioannou, C., Klimis, N., Theodulidis, N.,
Savvaidis, A., Efthymiadou, V., and Stewart, J. P., 2010. The 8 June 2008 Mw6.5 Achaia-Elia,
Greece earthquake: Source characteristics, ground motions, and ground failure, Earthquake
Spectra 26, 399–424.
Meslem, A., Yamazaki, F., Maruyama, Y., Benouar, D., Laouami, N., and Benkaci, N., 2010. Site
response characteristics evaluated from strong motion records of the 2003 Boumerdes,
Algeria, earthquake, Earthquake Spectra 26, 803–823.
Mueller, K. J., and Rockwell, T. K., 1995. Late quaternary activity of the Laguna Salada fault in
northern Baja California, Mexico, Geological Society of America Bulletin 107, 8–18.
Nakamura, T., 1989. A method for dynamic characteristics estimation of subsurface using
microtremor on the ground surface, Quarterly Report of Railway Technical Research Institute
30, 25–33.
Newmark, N. M., and Hall, W. J., 1982. Earthquake Spectra and Design, Engineering
Monograph on Earthquake Criteria, Structural Design, and Strong Motion Records, vol. 3,
Earthquake Engineering Research Institute, Oakland, CA, 103 pp.
Paciello, A., Rinaldis, D., and Romeo, R., 2000. Incorporating ground motion parameters related
to earthquake damage into seismic hazard analysis, Proceedings of the Sixth International
Conference on Seismic Zonation: Managing Earthquake Risk in the 21st Century, Earthquake
Engineering Research Institute, Oakland, CA.
Pacific Earthquake Engineering Center (PEER), 2005. PEER NGA Database, http://peer.berkeley
.edu/nga/, accessed 22 November 22.
Rathje, E. M., Abrahamson, N. A., and Bray, J. D., 1998. Simplified frequency content estimates
of earthquake ground motions, Journal of Geotechnical and Geoenvironmental Engineering,
ASCE, 124, 150–159.
Rathje, E. M., Faraj, F., Russell, S., and Bray, J. D., 2004. Empirical relationships for frequency
content parameters of earthquake ground motions, Earthquake Spectra 20, 653–675.
Rathje, E. M., Stokoe, II, K. H., and Rosenblad, B., 2003. Strong motion station characterization
and site effects during the 1999 earthquakes in Turkey, Earthquake Spectra 19, 217–242.
Rauch, A. F., and Martin, J. R., 2000. EPOLLS model for predicting average displacements
on lateral spreads, ASCE Journal of Geotechnical and Geoenvironmental Engineering
126, 360–371.
Rodriguez-Marek, A., Bay, J. A., Park, K., Montalva, G. A., Cortez-Flores, A., Wartman, J., and
Boroschek, R., 2010. Engineering analysis of ground motion records from the 2001 Mw 8.4
southern Peru earthquake, Earthquake Spectra 26, 499–524.
Rodriguez-Marek, A., Bray, J. D., and Abrahamson, N. A., 2001. An empirical geotechnical
seismic site response procedure, Earthquake Spectra 17, 65–87.
Rodríguez, V. H. S., and Midorikawa, S., 2003. Comparison of spectra ratio techniques for
estimation of site effects using microtremor data and earthquake motions recorded at the sur-
face and in boreholes, Earthquake Engineering and Structural Dynamics 32, 1691–1714.
Satoh, T., Kawase, H., and Matsushima, S., 2001. Differences between site characteristics
obtained from microtremors, S-wave, and codas, Bulletin of the Seismological Society of
America 91, 313–334.
Schnabel, P. B., 1973. Effects on Local Geology and Distance from Source on Earthquake
Ground Motions, Ph.D. Thesis, University of California, Berkeley.
Shantz, T., and Merriam, M., 2009. Development of the Caltrans Deterministic PGA Map and
Caltrans ARS Online, Report No. 071409, Caltrans, Sacramento, CA.
Singh, J. P., 1998. Site specific ground motions for high-tech seismic design, ATC/JSCA 8th
U.S.-Japan Workshop on the Improvement of Structural Design And Construction Practices,
Honolulu, HI.
Spudich, P., Joyner, W. B., Lindh, A. G., Boore, D. M., Margaris, B. M., and Fletcher, J. B.,
1999. SEA99: A revised ground motion prediction relation for use in extensional tectonic
regimes, Bulletin of the Seismological Society of America 89, 1156–1170.
Stewart, J. P., Blake, T. M., and Hollingsworth, R. A., 2003. A screen analysis procedure for
seismic slope stability, Earthquake Spectra 19, 697–712.
Theodulidis, N., Bard, P. Y., Archuleta, R., and Bouchon, M., 1996. Horizontal-to-vertical
spectral ratio and geological conditions: The case of Garner Valley downhole array in
Southern California, Bulletin of the Seismological Society of America 86, 306–319.
Travasarou, T., Bray, J. D., and Abrahamson, N. A., 2003. Empirical attenuation relationship for
Arias intensity, Earthquake Engineering and Structural Dynamics 32, 1133–1155.
Vanmarcke, E. H., 1976. Structural response to earthquake, in Seismic Response and Engineering
Decisions (C. Lomnitz and E. Rosenblueth, eds.), Elsevier, Amsterdam, 287–338.
U.S. Geological Survey (USGS), 2011. Earthquake data Web page for the 4 April 2010 magni-
tude 7.2 event in Baja California, Mexico, http://earthquake.usgs.gov/earthquakes/
eqinthenews/2010/ci14607652/, accessed 16 February 2011.
Wald, D. J., Quitoriano, V., Dengler, L., and Dewey, J. W., 1999. Utilization of the internet for
rapid community intensity maps, Seismological Research Letters 70, 680–697.
Wei, S. J., Fielding, E., Leprince, S., Sladen, A., Avouac, J-P, Helmberger, D., Hauksson, E.,
Chu, R. S., Simons, M., Hudnut, K., Herring, T., and Briggs, R., 2011. Superficial simplicity
of the 2010 El Mayor–Cucapah earthquake of Baja California in Mexico, Nature Geoscience
4, 615–618.
Wen, K., Beresnev, I., and Yeh, Y. T., 1994. Nonlinear soil amplification inferred from downhole
strong seismic motion data, Geophysical Research Letters 21, 2625–2628.
Wilson, R. C., and Keefer, D. K., 1985. Predicting areal limits of earthquake-induced landsliding,
Evaluating Earthquake Hazards in the Los Angeles Region: An Earth-Science Perspective
(J. I. Ziony, ed.), USGS Professional Paper 1360.
Wills, C. J., and Clahan, K. B., 2006. Developing a map of geologically defined site-condition
categories for California, Bulletin of the Seismological Society of America 96, 1483–1501.
Working Group on California Earthquake Probabilities (WGCEP), 2008. The Uniform California
Earthquake Rupture Forecast, version 2 (UCERF 2), USGS Open Report 2007–1437.
Yamazaki, F., and Ansary, M. A., 1997. Stability of H/V spectrum ratio of earthquake ground
motion, Transactions of the 14th International Conference on Structural Mechanics in Reactor
Technology(SMiRT 14), Lyon, France, August 17–22.
Zeghal, M., and Elgamal, E. W., 1994. Analysis of site liquefaction using earthquake records,
Journal of Geotechnical Engineering, ASCE, 120, 996–1017.
Zhao, J. X., Irikura, K., Zhang, J., Fukushima, Y., Somerville, P. G., Asano, A., Ohno, Y.,
Oouchi, T., Takahashi, T., and Ogawa, H., 2006. An empirical site-classification method
for strong-motion stations in Japan using H/V response spectral ratio, Bulletin of the
(Received 15 March 2011; accepted 1 March 2012)
Seismic Performance Assessment
of Unreinforced Masonry Buildings
with a Hybrid Modeling Approach
Alper Aldemir,a) M. Altuğ Erberik,b) I. Ozan Demirel,a) and Halûk
Sucuoğlu,c) M.EERI
This study proposes a hybrid modeling approach for the seismic perfor-
mance assessment of unreinforced masonry buildings. The method combines
finite-element and equivalent-frame approaches such that more powerful features
of each approach are utilized. The finite-element approach is used to model the
masonry components of different geometrical and material characteristics with a
high level of accuracy. Then this numerically simulated database is used in the
analytical modeling of masonry buildings with equivalent beams and columns
instead of spandrels and piers. Thus it becomes possible to model a masonry
building as a frame structure that can simply be analyzed in order to capture
the global behavior. The method has been verified by comparing the analytical
results with the previous experimental findings. The last part of the study is
devoted to the implementation of the method to an existing masonry building
that was damaged during a severe earthquake. [DOI: 10.1193/1.4000102]
INTRODUCTION
Performance-based seismic assessment procedures are becoming increasingly popular in
the field of earthquake engineering. Such procedures enable estimation of the seismic
performance of building structures in terms of damage as well as direct and indirect losses.
Numerous research studies have been conducted on the seismic performance of steel and
reinforced concrete (RC) frame buildings in the past. However, this is not valid for unrein-
forced masonry (URM) buildings, mainly for two reasons. First, most of the residential URM
buildings all over the world are nonengineered structures that have been constructed with
traditional practices. Therefore it is very difficult to use standard analysis methods for
this highly nonstandard type of construction. Second, estimating the nonlinear behavior
of URM buildings under seismic action is a challenging task. According to Abrams
(2001), despite being the oldest construction material, masonry is still the least understood
in terms of strength and deformation characteristics.
Apart from the above discussions, there exists a considerable URM building stock in
earthquake-prone regions of the world that has to be evaluated in terms of seismic safety
considerations. In addition, there are also a vast number of historical masonry structures
a)
Research Assistant, Middle East Technical University, Civil Engineering Department, 06800 Ankara, Turkey
b)
Associate Professor, Middle East Technical University, Civil Engineering Department, 06800 Ankara, Turkey
c)
Professor, Middle East Technical University, Civil Engineering Department, 06800 Ankara, Turkey
33
34 A. ALDEMIR ET AL.
that require seismic performance assessment. Therefore, no matter how difficult the modeling
of URM structures is, this task should be accomplished in order to mitigate financial and
physical losses and to prevent the loss of cultural heritage during future earthquakes.
There have been several efforts to assess and rank the seismic performance of existing
URM building stocks, most of them residential (D’Ayala et al. 1997, D’Ayala 2005,
Kappos et al. 2006, Park et al. 2009, Erberik 2010). This paper proposes a novel method
to assess the seismic performance of URM structures that makes use of both continuum-
based finite-element modeling and one-dimensional line-element-based structural modeling.
Hence it may be referred to as a “hybrid” method, in which finite-element (FE) technique is
employed in order to obtain the response of masonry wall components in the local sense and
equivalent frame modeling is employed to obtain the seismic response of masonry structures
in the global sense. Finally, the proposed method is validated by using both experimental and
field data.
MODELING STRATEGIES FOR UNREINFORCED MASONRY BUILDINGS

Due to the diversity of material properties and geometrical layout, coupled with the high
level of complexity in modeling the anisotropic and nonhomogeneous nature of masonry wall
components, estimating the nonlinear behavior of unreinforced masonry structures under
vertical and horizontal loading is not simple. In the past, several modeling strategies
with different levels of complexity have been utilized for this purpose. These approaches
range from sophisticated FE micro-modeling to approximate limit analysis. The proposed
method uses two of these modeling strategies, which are explained in detail in the following
section.
FINITE-ELEMENT-BASED MODELS VERSUS EQUIVALENT-FRAME MODELS

FE-based modeling is the most sophisticated tool that can be used for predicting the
behavior of masonry components or structures from the linear elastic stage to the complete
loss of strength. Lourenço (1996) summarized FE-based modeling strategies with regards to
the level of sophistication used for the analysis of URM structures as follows:
Detailed micro-modeling: This requires discrete modeling of mortar, brick units with conti-
nuum elements, and unit-mortar interface with discontinuous elements. This technique is
applicable to small structural parts or components of masonry, such as a pier or a spandrel
beam, for which the heterogeneous nature of stress and strain states are to be modeled in
detail.
Simplified micro-modeling: In this approach, brick units are modeled with continuum elements
whereas the behavior of mortar joints and unit-mortar interface is lumped in discontinuous
elements.
Macro-modeling: Units, mortar, and the interface between them are smeared into a macro con-
tinuum model with few degrees of freedom. This technique is applicable to large structural
components, or even the structure itself, in order to capture the global behavior.
Several studies exist in which the FE method is employed with different levels of com-
plexity in order to simulate the behavior of masonry components and structures (Lofti and
Shing 1994, Gambarotta and Lagomarsino 1997, Gabor et al. 2006, Attard et al. 2007, Casolo
and Pena 2007, Lourenço et al. 2007).
SEISMIC PERFORMANCE ASSESSMENT OF URM BUILDINGS WITH A HYBRID MODELING APPROACH 35
Figure 1. Schematic representation of EF approach: (a) perforated masonry wall, (b) representa-
tion of masonry piers and spandrels by EF members (thick solid lines are rigid-end zones),
(c) details of an EF member with plastic hinges lumped at certain locations.
The equivalent-frame (EF) approach is a simple and efficient way for predicting the
seismic response of URM structures. The structure is modeled with one-dimensional line
elements that represent the piers and spandrels of the actual structure (Figures 1a and
1b). Plastic hinges can be lumped at certain locations (end and/or mid- sections) of the
EF member in order to describe the local nonlinear behavior of the wall component
under consideration through moment-rotation and shear force–shear displacement relation-
ships (Figure 1c). Hence it becomes possible to analyze a masonry structure as a frame struc-
ture by using conventional analysis software. Due to its simplicity and efficiency, many
researchers use this approach in the analysis of URM buildings under gravity and earthquake
loads (Magenes and Fontana 1998, Salonikios et al. 2003, Roca et al. 2005, Penelis 2006,
Pasticier et al. 2008, Belmouden and Lestuzzi 2009).
Some researchers compared the FE approach with the EF approach regarding the lateral
load response of URM structures. Karatoni and Fardis (1992) considered three different
methods for the analysis of URM structures: the finite element, the equivalent frame,
and the shear-beam model. They observed that only the FE method could capture the
key features of in-plane and out-of-plane masonry response with an accurate prediction
of damage distribution. However, they also added that the selection of the FE approach
for seismic response analysis may not be economically justified except for the analysis
of historical masonry structures. Kappos et al. (2002) stated that the EF modeling yields
reasonably accurate results for practical purposes regarding the nonlinear analysis of masonry
buildings.
A HYBRID MODELING APPROACH FOR SEISMIC PERFORMANCE

ASSESSMENT OF UNREINFORCED MASONRY BUILDINGS
The FE and EF approaches have their pros and cons, as discussed above. The FE
approach is robust and can estimate actual behavior with good accuracy, but it is compu-
tationally intense, especially in the case of dynamic response prediction of large masonry
structures. On the other hand, the EF approach is quite practical since continuum elements
are converted to line elements and the analysis can be simply achieved by using any con-
ventional frame analysis software. The drawback is that it is not possible to capture the key
details of masonry response. Considering the basic advantages of these two modeling
approaches and balancing their pros and cons, this study proposes a hybrid approach for the
seismic performance assessment of unreinforced masonry buildings that employs the stron-
gest part of each method: The FE approach is used to model the masonry wall components
with a high level of accuracy and for capturing all behavioral modes. Then this information is
used in the analytical modeling of the building with equivalent beams and columns instead of
spandrels and piers. In the global sense, a masonry building is modeled as a frame structure
that can be analyzed practically. Moreover, nonlinear spring elements attached to the ends of
linear frame elements are able to simulate the local nonlinear behavior of masonry walls.
Inelastic response in masonry piers and spandrels develops in reality by cracking and crush-
ing of masonry, which results in various nonlinear deformation mechanisms including slid-
ing, rocking, diagonal tension cracking, and toe crushing. Such nonlinear behavior can be
simulated most accurately by a nonlinear finite element model, which is done at the member
level in this study.
The equivalent frame model is employed at the structural level for computational effi-
ciency, where the nonlinear lateral load–lateral deformation response obtained from the
finite-element model of the members is simplified into bilinear force-deformation relation-
ships, and these relationships are assigned to nonlinear springs in the equivalent frame mem-
bers. Although nonlinear lumped springs are utilized for the idealization of plastic behavior
of masonry piers and spandrels, these elements do not represent flexural yielding as in the
case of reinforced concrete or steel frame members. While nonlinear springs indicate plas-
tification of the assigned section when utilized for reinforced concrete or steel, the main
objective in their employment is to simulate nonlinear force-deformation response, which
is obtained by sophisticated FE analysis for a masonry element. Local masonry behavior
for each nonlinear spring is derived with high accuracy from a huge database of masonry
components with varying dimensions, material properties, and vertical stress, which was
obtained by elaborate FE analysis.
The details of the hybrid method, which is composed of five steps, are shown in Figure 2.
In the first step, FE analyses are carried out in order to construct the comprehensive database
of masonry wall response by using an FE program (ANSYS is employed in this step). In the
second step, the response curves obtained from FE analysis are idealized in terms of basic
structural parameters. A bilinear idealization is used herein with four parameters: yield
strength, yield displacement, ultimate strength, and ultimate displacement. In the third
step, nonlinear regression analysis is employed in order to obtain an empirical equation
in terms of the four structural parameters for representing the idealized response with an
adequate level of accuracy. Once this equation is obtained, it can be assigned to the nonlinear
plastic hinges of the EF members to simulate the shear force-displacement relationship of a
masonry wall component in a building structure. After constructing the EF model by EF
members with their corresponding nonlinear hinges and rigid end offsets, the analytical
model can be easily analyzed by any structural analysis program with nonlinear dynamic
analysis capability. In this study, SAP2000 is used. The first three steps of the method
are enclosed by a dashed- line box. This means that these three steps will be performed
only once. After obtaining the empirical equations at the end of the third step, this informa-
tion can be used repeatedly to assess the seismic performance of any number of URM
buildings.
Figure 2. Detailed outline of the hybrid method.
FINITE-ELEMENT MODELING OF MASONRY WALL COMPONENTS

The first task is to develop a database that includes the local behavior of masonry wall
components with different characteristics so that it becomes possible to determine the
response characteristics of any wall segment with different properties within the database.
For this purpose, the FE program ANSYS (2007) is employed. Element type “Solid65” is
selected to model masonry wall components. It is a three-dimensional brick element with
eight nodes at each corner. The Solid65 element was chosen for its capability of exhibiting
cracking and crushing whenever its tensile or compressive capacity is reached. This property
gives the opportunity to observe the crack pattern of the analyzed component model and also
to identify its failure mode.
The material properties of masonry are simulated by assigning a combined material
model to the Solid65 element. A combined material model is used because the FE
model program is not able to model masonry explicitly; hence by using the specific properties
of two different material models, it is possible to develop a more realistic model for masonry
in terms of cracking, crack closing, crushing, stress relaxation, and shear transfer. These two
models are the five-parameter concrete model proposed by Willam and Warnke (1975) and
the multilinear isotropic plasticity model in ANSYS (Figure 3a). Two models are combined
in such a way that the inner plasticity model always governs the response (Figure 3b). Further
details on the FE model can be found elsewhere (Aldemir 2010).
Verification of the FE model is achieved by comparing the analytical results with the
experimental results obtained from different series of unreinforced masonry wall tests.
Figure 3. (a) Multilinear isotropic plasticity model used in modeling masonry wall components,
(b) simple sketch showing the combined envelope of two different plasticity models.
The first series of tests were conducted at ETH Zurich (Ganz and Thürlimann 1984). The wall
specimen considered for verification is a masonry panel (3600 ! 2000 ! 150 mm3) with two
flanges (150 ! 2000 ! 600 mm3) that is composed of hollow clay brick units. Mechanical
properties of the wall specimen are given in Table 1. Initially, a uniformly distributed pres-
sure (0.61 N/mm2) is applied on top of the wall. Then the wall specimen is subjected to
monotonically increasing displacement until failure. As reported by Ganz and Thürlimann
(1984), diagonal shear cracks appear at the ultimate stage and some of the masonry blocks in
the middle part of the wall sally from the wall panel.
The FE model of the wall specimen is presented in Figure 4. As seen in the figure, a
concrete slab and foundation are part of the tested specimen and are also modeled by con-
sidering the material properties given in the technical report by Ganz and Thürlimann. The
optimum mesh density is achieved by a process of trial and error, the details of which are
given in Aldemir (2010).
The comparisons of analytical and experimental results are shown in Figures 5 and 6.
From the comparison of load-displacement curves for two cases (experimental data is shown
with dots and analytical data is shown with a solid line) in Figure 5, it can be inferred that the
analytical model simulates the behavior of the wall specimen in acceptable limits. In addition,
the crack pattern obtained from the analytical model at the end of FE analysis (shown in
Figure 6a with a well-defined diagonal crack) agrees reasonably well with the experimental
results (shown in Figure 6b).
Table 1. Mechanical properties of the wall specimen tested in ETH Zurich
Mechanical parameter Value Unit
Modulus of elasticity ðEm Þ 2460 MPa

Shear modulus ðG m Þ 1130 MPa
Compressive strength ðf m Þ 7.61 MPa
Tensile strength ðf mt Þ 0.28 MPa
Poisson's ratio ðνÞ 0.18 _
Figure 4. The FE model in ANSYS of the wall specimen tested in ETH Zurich.
Figure 5. The comparison of capacity curves of the wall specimen from the FE analysis and the
experiment.
The other two specimens considered for verification represent two different classes of
masonry wall panels with respect to geometry (i.e., squat and slender piers). The first
one has been extracted from a series of tests conducted again at ETH Zurich (ElGawady
et al. 2005). The specimen has a height-to-length ratio equal to 0.45. The second one
has a height-to-length ratio around 1.8, which was tested in the structural laboratory of
the University of Illinois at Urbana-Champaign (Abrams et al. 2007). The common point
in both of the specimens is that they were subjected to cyclic loading patterns. Hence
the experimental monotonic load-displacement relationships were obtained through the
envelopes of the hysteresis curves. Both specimens were modeled in ANSYS and the capa-
city curves were obtained. Figures 7a and 7b show the comparison of the numerical results
with the experimental data. In both cases, the force and displacement capacity seems to be
simulated within acceptable limits. Hence it can be concluded that the FE model satisfactorily
simulates the local force-displacement relationship of masonry wall components.
DEVELOPMENT OF A DATABASE FOR MASONRY WALL COMPONENTS

After verification, the next step is to develop the database for in-plane capacity curves of
masonry wall components with different geometry, material properties, and loading
Figure 6. (a) The crack pattern from the FE analysis, (b) the damage observed at the end of the
test in ETH Zurich.
Figure 7. The comparison of capacity curves of (a) squat and (b) slender wall specimens from the
experiment with the ones obtained through FE analysis.
conditions. Accordingly, masonry wall components are classified with respect to three major
structural parameters. The first classification is according to aspect ratio. By definition, it is
the ratio of the wall height to wall length (H/L) in the horizontal direction. In this category,
masonry piers with eight different aspect ratios ranging from H/L = 0.25 to 2.0 are consid-
ered. The reason for this broad range of classification is that the mode of failure highly
depends on the aspect ratio and that many different aspect ratios can be encountered in prac-
tice. It is impossible to define an analytical model by using a predefined aspect ratio; that is,
length or height along with the thickness should also be known. Thus, a reference length of
1 m and a reference thickness of 0.2 m are assumed for the FE model.
The second classification is according to the compressive strength of masonry ðf m Þ. In
this study, only clay brick masonry units are considered and the values are determined with
respect to the local masonry construction practice and the related documentation. Hence three
subclasses are proposed in terms of the mean compressive strength, for which the values are
2 MPa (low strength), 5 MPa (moderate strength), and 8 MPa (high strength). The modulus of
elasticity of masonry ðE m Þ is taken as 2,000 MPa for all cases in accordance with the material
test results of clay bricks obtained from previous studies (Bayulke 1992).
The last classification for masonry components is made according to vertical stress (level
of axial load). Similar to concrete members, unreinforced masonry panels tested under higher
axial loads exhibit larger lateral load capacity with a less ductile displacement response than
their counterparts that are subjected to lower vertical stresses. Furthermore, the level of axial
load can also change the failure mode. Six levels of axial load are selected for classification,
where level of axial load is defined as a percentage of the compressive strength of masonry
pier under consideration. These classes can be listed as p ¼ 0.05 f m , 0.10 f m , 0.20 f m , 0.30 f m ,
0.40 f m , and 0.50 f m . The combination of these three subclasses makes up 144 different
masonry wall components to be modeled analytically. All of the analytical models are ana-
lyzed by using the built-in nonlinear solution techniques and convergence algorithms
of ANSYS.
IDEALIZATION OF CAPACITY CURVES OBTAINED BY THE FE ANALYSES

After the analyses are completed, the analytical database is expressed in the form of the
capacity curves of masonry wall elements with different geometrical and material properties.
In the second step of the hybrid approach, the analytical capacity curves are idealized accord-
ing to the FEMA 356 method (ASCE 2000) in the form of a bilinear envelope with four
parameters: yield force ðF y Þ, yield displacement ðδy Þ, ultimate force ðF u Þ, and ultimate dis-
placement ðδu Þ (Figure 8). These parameters also represent the local limit states of masonry
wall components. The former stands for the serviceability limit state at which significant
cracking takes place with a significant reduction in stiffness of the wall, whereas the latter
represents the ultimate limit state at which either stability is lost or there exists a significant
reduction in the wall strength.
The analytical database is limited in such a way that all the capacity curves are obtained
for the reference length of L ¼ 1 m and reference thickness of t ¼ 0.2 m. However, these
geometrical properties differ in practice. Hence in order to examine the effect of length and
thickness of masonry wall on the aforementioned capacity curve parameters, parametric ana-
lyses are conducted by varying length and thickness and comparing the results with the ones
obtained from the wall models with reference values of these parameters. The additional FE
analysis results reveal that the effects of change in both length and thickness of the wall
components are reflected almost linearly in most of the cases on the capacity curve para-
meters. Hence the effect of length and thickness can be incorporated by simply multiplying
Figure 8. Idealization of the analytically obtained capacity curve in bilinear fashion and the
major response parameters used in the idealization process for the masonry wall component
with L ¼ 1 m, H ¼ 1 m, t ¼ 0.2 m, p ¼ 0.1 f m , and f m ¼ 5 MPa.
the reference values with the corresponding ratios. Further details on the parametric analysis
can be found elsewhere (Aldemir 2010).
EMPIRICAL RELATIONS FOR CAPACITY CURVE PARAMETERS THROUGH

NONLINEAR REGRESSION
Nonlinear regression analysis is used in order to obtain empirical relationships for capa-
city curve parameters [F y (in kN), δy (in mm), F u (in kN), and δu (in mm)] in terms of the
aforementioned structural parameters [f m (in MPa), H ∕L, p (in MPa), L (in m), and t (in m)].
Many different mathematical models have been examined and the ones with the highest R2
values are selected as the most appropriate. As a result, the following two empirical relation-
ships are proposed for estimating the displacement-based and force-based capacity curve
parameters, respectively:
δ̈ ¼ C 1 ! pC2 ! eC3 f m ! ðH ∕LÞC4 ! L ! t
EQ-TARGET;temp:intralink-;e1a;41;287 (1a)
F̈ ¼ C 1 ! pC2 ! f Cm3 ! eC4 ðH ∕LÞ ! L ! t

EQ-TARGET;temp:intralink-;e1b;41;252 (1b)
In the above equations, δ̈ and F̈ stand for the estimated values of displacement-based para-
meters (δy and δu ) and force-based parameters (F y and F u ) of the capacity curve. C i (i = 1– 4)
denotes the regression coefficients that have been obtained for the best fit equations as given in
Table 2. The effect of length and thickness are reflected in a proportional manner for the
capacity curve parameter of interest, for the reasons that were previously discussed.
Next, a parametric study is conducted by considering the idealized force-displacement
relationships of ten masonry wall components with different characteristics as given in
Table 3. The results of the parametric study are presented in Figures 9a to 9d. For the
sake of comparison, the capacity curves obtained from the FE analysis in ANSYS (thin
black lines) are shown together with idealized curves (thick gray lines) for the selected
wall components. In most of the cases, there is a satisfactory match between the capacity
curves obtained by Equations 1a and 1b and the ones obtained from ANSYS.
Table 2. Regression coefficients obtained for the estimation of δy , F y , δu , and F u
Regression coefficients
Capacity curve parameters C1 C2 C3 C4
δy (mm) 0.587 0.543 0.095 1.426

F y (kN) 70.629 0.604 0.414 −0.931
δu (mm) 0.477 −0.540 0.319 1.414
F u (kN) 70.431 0.498 0.501 −0.856
Table 3. Properties of the masonry wall components that are used in the parametric
analysis
Wall No L (m) H (m) H/L t (m) f m (MPa) p (MPa)

1 1.0 1.0 1.0 0.2 5.0 0.5 (0.1 f m )
2 1.0 0.5 0.5 0.2 5.0 0.5 (0.1 f m )
3 1.0 2.0 2.0 0.2 5.0 0.5 (0.1 f m )
4 1.0 1.0 1.0 0.1 5.0 0.5 (0.1 f m )
5 1.0 1.0 1.0 0.3 5.0 0.5 (0.1 f m )
6 1.0 1.0 1.0 0.2 5.0 0.25 (0.05 f m )
7 1.0 1.0 1.0 0.2 5.0 1.0 (0.2 f m )
8 1.0 1.0 1.0 0.2 5.0 1.5 (0.3 f m )
9 1.0 1.0 1.0 0.2 2.0 0.2 (0.1 f m )
10 1.0 1.0 1.0 0.2 8.0 0.8 (0.1 f m )
In all of the figures, there is a reference wall component (Wall 1) with the following
characteristics: L ¼ 1 m, H ¼ 1 m, t ¼ 0.2 m, p ¼ 0.1 f m , and f m ¼ 5 MPa. First the
influence of aspect ratio on the idealized capacity curves is examined (Figure 9a). Hence
three different H values are considered. In the case of H ¼ 0.5 m (Wall 2), a squat wall
with H ∕L ¼ 0.5 is considered, whereas in the case of H ¼ 2 m (Wall 3), a slender wall
with H ∕L ¼ 2.0 is taken into consideration. As expected, the squat wall possesses higher
strength but lower displacement capacity when compared to the slender wall. In Figure 9b,
the influence of thickness is examined. As stated before, when the thickness of the wall is
increased, strength is also increased proportionally. An increase in the ultimate displacement
capacity is also monitored for this case. Figure 9c presents the influence of vertical stress on
the capacity of masonry walls. From the figure it is revealed that as the vertical stress
increases, the strength of the wall increases but the ultimate displacement capacity decreases.
Finally, in Figure 9d, the influence of compressive strength of masonry is considered. It is
observed that as the masonry strength increases, both the force capacity and the displacement
capacity of the wall increase significantly. Considering these observations, it can be con-
cluded that the idealized capacity curves for masonry walls exhibit physically consistent
trends for all the cases.
The match between the results obtained from the empirical equations and the FE analyses
can be examined in more detail by the introduction of an error function, which is defined as
Figure 9. The effect of (a) aspect ratio, (b) thickness, (c) vertical stress, and (d) masonry strength
on the idealized capacity curves of masonry wall components. Thin black lines show the results of
the FE analysis whereas thick gray lines represent the results from Equations 1a and 1b.
the ratio of the value obtained by using the empirical equations to the value obtained from the
bilinearization process of the benchmark FE curve, for the capacity curve parameter under
consideration. The results are shown in Figure 10 in terms of the capacity curve parameters
(Fy, δy , F u , and δu ) for all the masonry wall components used in the parametric study. The
error measure generally varies in the range 0.8–1.2 with few exceptions for the ratios in terms
of ultimate displacement. This is not surprising, given that ultimate displacement capacity is
the most difficult performance parameter to estimate since it is not possible to attain a stan-
dard definition for the collapse or near-collapse limit states. But overall, the error measure
reveals that the basic formulations given in Equations 1a and 1b can be used with confidence
to estimate the capacity parameters of a wall masonry component.
CONSTRUCTION OF THE EQUIVALENT-FRAME MODEL

Masonry wall components are represented with frame members in the EF model. In order
to simulate the nonlinear behavior of masonry walls as well as their linear elastic behavior,
shear hinges are assigned at the bottom ends of equivalent frame idealizations of masonry
piers (Figure 11). The properties of the shear hinge—such as yield force ðF y Þ, ultimate force
ðF u Þ, and ultimate displacement ðδu Þ are determined from Equations 1a and 1b depending on
the aspect ratio, compressive strength, axial load, length, and thickness of the pier. The reason
of assigning a shear hinge instead of a moment hinge is that contrary to typical reinforced
concrete or steel frame members, whose nonlinear behavior is governed by flexure, masonry
piers and spandrels are generally squat members where shear deformations are effective in
the elastic and inelastic ranges. As stated by Teran-Gilmore et al. (2009), shear deformations
dominate over flexural deformations as level of damage increases. However, utilization of
Figure 10. Error measure for the capacity curve parameters by considering the masonry walls
given in Table 2.
Figure 11. Illustration of the EF member and the shear hinge that has been used in order to
simulate the elastic and inelastic behavior of masonry wall components.
shear hinges does not imply that nonlinearity is restricted to squat members failing in shear-
dominated modes such as diagonal tension and sliding. Since the force deformation relation
of these hinges is taken from the simplified FE analysis results in which the failure mode is
included, these hinges are applicable to slender members as well as squat members.
Axial stiffness of frame elements remains constant throughout the analysis. Lateral stiff-
ness of frame elements is composed of flexural stiffness and shear stiffness (see Equation 2)
where H is the height, EI is the flexural stiffness, AG is the shear stiffness, β is the boundary
condition parameter (i.e., 3 for cantilever boundary conditions, 12 for both ends fixed bound-
ary conditions), and κ is the shear coefficient (1.2 for rectangular cross-section geometry).
! "−1
H3 κH
Ke ¼ þ (2)
βEI AG
In EF modeling, it is important to determine the effective heights of masonry components

of a perforated masonry wall, which is directly related to lateral elastic stiffness of the wall.
Figure 12. EF model (a) with no rigid-end offsets, (b) with Dolce offsets, and (c) with full rigid-
end offsets.
For this purpose, generally rigid-end offsets are employed. There are different approaches in
the literature to assign fixity at the end of EF members. One method, proposed by Dolce
(1989), is to take only a portion of equivalent beam-column interaction as rigid; another
approach is to take this interaction as fully rigid. In order to decide which approach to
use, a comparative study has been performed on a two-story, one-bay masonry building,
which has been investigated before by Salonikios et al. (2003). The structure is first analyzed
by using FE analysis and then modeled as an EF with no rigid-end offsets, Dolce offsets, and
full rigid-end offsets (Figure 12). After the EF models are generated, lateral stiffnesses of
equivalent frame members are calculated by the computer software, which takes into account
geometrical and material properties, boundary conditions, and rigid-end zones.
When the results of FE analysis and frame analyses are compared, it is observed that the
frame with Dolce offsets give the closest match to the FE results in terms of base shear and
lateral roof displacement. Hence throughout the proposed approach, Dolce offsets are
employed. The details of this comparative study can be obtained elsewhere (Demirel 2010).
VERIFICATION OF THE PROPOSED METHOD

After constructing the EF by using line elements whose rigid-end offsets are defined
according to the approach proposed by Dolce (1989), dead load analysis is conducted
with the model in order to determine the axial loads acting on piers. Utilizing the axial
load level, compressive strength, and the geometric properties, nonlinear shear hinges are
placed at the bottom of each column element representing a masonry pier. Then the analytical
model is ready to be analyzed by any structural analysis software for nonlinear static or
dynamic analyses. Data preparation, run time, and post-processing with the EF model
are significantly less time-consuming and more practical compared to the FE analysis.
For the verification of the proposed method, the masonry specimens tested at the ISMES
laboratory at Bergamo are considered. Finally, the proposed method is applied to an existing
brick unreinforced masonry building in Afyon, Turkey, that was damaged during the 1995
Dinar earthquake.
VERIFICATION USING EXPERIMENTAL DATA

A half-scale, two-story brick unreinforced masonry building that was tested on a shake
table at ISMES laboratory in Bergamo is selected for verification of the proposed nonlinear
EF model. The specimen was subjected to base excitations in two orthogonal directions and
the vertical direction simultaneously by using the signals recorded at Calitri during the 1980
Irpinia, Italy, earthquake ðM w ¼ 6.5Þ. Five different levels of seismic excitation (P7–P11)
were applied to the specimen by scaling the actual signals. It was reported that the excitation
P7 caused a fairly linear response on the specimen, whereas during the excitation P11 the
behavior was highly nonlinear. The excitation P9 was identified as the event marking the
yield point of the masonry specimen with a base shear coefficient of 0.19W in one of
the orthogonal directions, where W is the total weight of the building specimen. Ultimate
force capacity and ultimate roof drift were measured as 0.22W and 0.003 in the same ortho-
gonal direction, respectively. In terms of material properties, the compressive masonry
strength, modulus of elasticity, and Poisson’s ratio were determined as 2.2 MPa,
920 MPa, and 0.2, respectively. Further details about the test specimen are reported by Ben-
edetti et al. (1996).
For verification, the test specimen is modeled in the SAP2000 platform as an EF struc-
ture. By using the proposed hybrid method, masonry walls of the test specimen are replaced
with EF members (Figure 13a). The effective heights of members and the corresponding rigid
offset lengths are calculated using the Dolce method. After obtaining the geometrical proper-
ties of each masonry wall component, linear gravity load analysis is conducted and compres-
sive stress on each wall component is determined. After effective height and axial load level
of EF members are obtained, shear hinges are assigned according to the capacity parameters
calculated by Equations 1a and 1b (Figure 13b). The analysis does not consider out-of-plane
stiffnesses of the EFs.
Nonlinear time-history analyses are conducted by using the seismic excitations P7–P11.
Cyclic behavior is simulated by the multilinear degrading hysteresis model developed by
Takeda et al. (1970), for which the idealized capacity curves obtained from empirical for-
mulation constitute the backbone curve. The reason for using the Takeda model is that,
among the available hysteresis models in SAP2000, it seems to be the most suitable one.
The backbone curve parameters of the model are supplied from FE analysis, where the geo-
metrical and mechanical properties of masonry walls are taken into consideration. This means
that the inherent behavior of walls in terms of displacement and force capacities are also
reflected to some degree. Under this backbone curve, the cyclic behavior is assumed to
be governed by Takeda rules. Rayleigh damping is used in time-history analyses since
the first two translational vibration modes were experimentally measured.
Comparison of nonlinear time-history results with the experimental data is presented in
Figures 14a–14f. This comparison is given in terms of floor accelerations of the first and
second stories since this quantity was directly measured during the tests. Among five levels
of seismic excitation, P7 (initial loading with results in linear elastic behavior), P9 (inter-
mediate loading in which nonlinearity starts with significant cracking), and P11 (final loading
in which a high level of nonlinearity is observed with significant damage) are presented in the
figure.
Figure 13. (a) EF model of the test specimen (dimensions refer to the half-scale model), (b)
idealized nonlinear capacity curves assigned to shear hinges of the corresponding frame members
in both directions.
The comparison of acceleration time histories reveals that general trends in the variation of
recorded accelerations during the tests are successfully captured by the proposed model, espe-
cially for the excitations with low intensity (P7). Maximum differences between the recorded
and the calculated values of floor acceleration occur in the case of the excitation with the high-
est intensity (P11) and in the second story. Experimentally, significant yielding was reported to
take place at the end of excitation P9, whereas analytically it is observed at the end of excitation
P8. Furthermore, the ultimate lateral resistance of the test specimen was reported as 0.22W
experimentally, whereas it is calculated as 0.17W by using the proposed approach.
Although there are some differences between the experimental and analytical findings,
the values are in reasonable limits and this validates the use of the proposed method for the
seismic performance assessment of unreinforced masonry buildings. The differences can be
attributed to the assumptions and limitations of the proposed approach and also to the fol-
lowing:
• Although the test specimen is modeled as three-dimensional, out-of-plane resistance
of the walls is neglected. Thus, interaction of out-of-plane walls with the in-plane
resisting walls (i.e., flange effect) is not taken into account.
Figure 14. Comparison of the time histories of floor accelerations obtained through nonlinear
time-history analyses and experimental shake table tests.
• Base accelerations in three orthogonal directions are applied simultaneously in the

experiment, whereas excitations in each direction are applied individually during
analysis.
• In the experiment, the wooden slab of the test specimen seems to behave like a
flexible diaphragm rather than a rigid diaphragm, which is one of the major assump-
tions in the modeling and analysis stages. Therefore the actual lateral force distribu-
tion can be different from the distribution that has been obtained during the analysis.
• It is assumed in the proposed method that nonlinear behavior is observed in masonry
piers only. Spandrels are modeled as elastic members. However, although ultimate
failure is due to piers, significant spandrel damage was observed during the experiment.
• The Takeda model, which is used to simulate the cyclic behavior of unreinforced
masonry components, is not the best model to represent the actual masonry beha-
vior, which is strongly dependent on the failure mechanism of the pier. Energy dis-
sipation capacity of this model is eventually larger than the actual energy dissipation
capacity of the test specimen. This issue induces a limitation for the proposed model:
During simulation of dynamic behavior using the EF model, the Takeda model, or
analogous models, the simple models are successful in capturing the behavior just in
the early nonlinear range; however, the results should be used with caution when the
behavior is highly nonlinear since such simple models are unable to represent dif-
ferent behavior modes of masonry wall panels like flexure, sliding, or diago-
nal shear.
VERIFICATION USING FIELD DATA

In the last part of the study, seismic performance assessment of an existing brick unrein-
forced masonry building that was damaged during the 1995 Dinar earthquake ðM w ¼ 6.4Þ is
carried out. The photo and plan layout of the ground-story walls of the considered building
are presented in Figures 15a and 15b. The damage state of this building after the earthquake
Figure 15. (a) Photo, (b) first-story plan layout of the existing URM building in Dinar, together
with observed damage states of masonry walls in that story.
was examined by the technical teams from the Middle East Technical University, who used
the standard damage evaluation form for rural masonry structures developed by the Turkish
Ministry of Public Works and Settlement, and it was reported as “moderate damage“
(METU-EERC 1996). The observed damage states of the first-story walls as reported by
the technical team are also listed in Figure 15b.
It is a two-story building with almost regular plan geometry. The existence of reinforced
concrete floor slabs enables the formation of rigid diaphragm action. The masonry walls were
constructed by using solid brick units, which had been produced at local furnaces nearby. For
this type of masonry unit, the compressive strength, modulus of elasticity, and Poisson’s ratio
can be taken as 5 MPa, 2,000 MPa and 0.2, respectively, based on previous research (Bayulke
1992, METU-EERC 1996). Unit weight of the concrete slab is taken as 25 kN/m3 and unit
weight of the brick masonry walls is taken as 18 kN/m3 for the weight calculations.
The EF model of the considered building is constructed with 20 wall components in the
ground story and 22 wall components in the second story (Figure 16). Walls are assumed to
resist in-plane forces acting on their strong direction only. Effective heights of the perforated
walls are calculated according to method proposed by Dolce.
Modal analysis is conducted in order to identify the dynamic characteristics of the
building. The periods of vibration for the first three modes are obtained as 0.12 s,
0.10 s, and 0.07 s, respectively. The first two are predominantly translational modes in
Y- and X-directions, respectively, whereas the third one is a torsion-dominant mode.
After determining the capacity curve parameters for each wall component in the model
and estimating the force-displacement relationships by using Equations 1a and 1b, the
pushover analysis of the case study building is conducted in two orthogonal directions.
Due to the presence of a long structural wall (W20 in Figure 15), the shear capacity of
the building in the X-direction is approximately 1.5 times larger than the corresponding capa-
city in the Y-direction. The capacity curve in the Y-direction is shown in Figure 17, on which
the local limit states of structural walls in the first story are mapped. These limit states are
taken from the idealized capacity curves of masonry walls, for which the deformation limit
states δ ¼ δy , δ ¼ 0.75 δu , and δ ¼ δu are denoted by the letters A, B, and C, respectively, in
the figure. The numbers in front of the letters in Figure 17 (i.e., 4A, 1C, 3B, etc.) stand for the
number of walls that attained the specified limit states at the corresponding load step. It is
observed that all masonry walls in the considered direction have reached their local yield
limit state up to a drift ratio of 0.1%, which can be considered as the “immediate occupancy”
(IO) limit state. This drift ratio was also accepted by some researchers as a reasonable value
corresponding to the onset of structural damage (Calvi 1999, Erbay 2007). In later stages, one
of the walls reaches 75% of its displacement capacity when the total drift is equal to 0.24%
and the same wall fails at a drift ratio of 0.3%. Hence it can be concluded that the next
performance limit for the building, which can be considered as “life safety” (LS) limit
state, occurs between 0.25% and 0.3%. This value is also close to the values recommended
by Calvi (1999) and Tomazevic (2007). At the drift ratio of 0.46%, 6 out of 13 walls in the
first story either reach 75% of their shear capacities or fail at their ultimate capacities. This is
identified as the “collapse prevention” (CP) limit state, after which the building is heavily
damaged or collapses. The proposed values for this ultimate performance level are in good
agreement with the previous research conducted by Calvi (1999), who proposed a drift ratio
Figure 16. Computer model of the case study URM building in Dinar.
Figure 17. Capacity curve of the case study building in the Y-direction on which the local limit
states of structural walls in the first story are mapped.
of 0.5%. Some other researchers proposed a drift ratio of 1.0% for CP limit state. However,
this value seems high for the considered unreinforced masonry buildings since the dominant
failure mode is generally brittle and caused by shear. The proposed performance limits for the
case study building are also shown in Figure 17 by vertical dashed lines.
The performance point of the case study building is calculated by utilizing the assessment
procedure in the Turkish Earthquake Code (2007). Accordingly, the elastic design spectrum
provided by the code is used, with proper selection of seismic zone coefficient and site class
for the case study building. The capacity curve is idealized by using the bilinear approxima-
tion in FEMA 356 (ASCE 2000). Then both curves are converted to ADRS format by using
the formulations in ATC 40 (ATC 1996). Demand and capacity curves are compared in
ADRS format and the performance point is obtained as 8.8 mm (0.16% drift ratio) after
Figure 18. (a) EW component of the ground motion record in the 1995 Dinar earthquake,
(b) comparison of the nonlinear static (PO) and dynamic (TH) analysis results together with
the previously obtained performance point.
several iterations. The performance point is shown on the capacity curve in Figure 17 by the
star symbol. According to the assessment procedure in the Turkish Earthquake Code (2007),
the case study building is expected to be in the life safety performance level. It is observed
that some of the critical walls according to the analysis results (W1, W2, W3, W12) are also
the ones that were damaged during the earthquake. Walls 11 and 13, which seem to exceed
the elastic limit according to pushover analysis, have been reported to experience light-to-
moderate damage with some diagonal cracks. Hence the behavior simulated by using the
idealized curves of the wall components seems to reasonably agree with the field observa-
tions after the earthquake.
Nonlinear time-history analysis is also carried out for the case study building by using the
actual ground motion records of the 1995 Dinar earthquake. Figure 18a presents the EW
component of the ground motion record for the Dinar earthquake (applied to the Y-direction
of the building model), with a peak ground acceleration of 0.25 g. Cyclic behavior of EF
members is defined according to the hysteresis model proposed by Takeda et al. (1970).
Figure 18b illustrates the base shear coefficient versus roof displacement relationship of
the case study building in the Y-direction together with the capacity curve obtained from
pushover analysis. The previously obtained performance point is also shown on the same
curve as a solid dot. It is observed that a significant amount of energy is dissipated due
to the cyclic nature of motion, indicating that some of the members have responded in
the nonlinear range. However, the results obtained from the time-history analysis underes-
timate the expected performance obtained by the code procedure. The main reason for this
discrepancy is the selected hysteresis model for nonlinear behavior. As discussed before, the
Takeda model is a simple stiffness-degrading model, which is more suitable for ductile and
well-detailed reinforced concrete members. Hence the actual behavior of brittle masonry wall
components with significant strength degradation and different failure modes cannot be simu-
lated by using this hysteresis model. But it should also be noted that it is difficult to find in the
literature simple hysteresis models with few parameters that satisfactorily simulate the beha-
vior of masonry wall components, and that development of such a hysteresis model is out of
the scope of this study. If a hysteresis model, which is more suitable for the dynamic behavior
of the masonry components, can be utilized, the EF approach has the ability to yield better
results in time-history analysis.
CONCLUSIONS
The work presented in this study mainly focuses on the nonlinear modeling and structural
assessment of unreinforced masonry buildings through finite-element and equivalent-frame
methods. A hybrid approach is proposed such that it utilizes both methods in an efficient way,
considering the powerful features of each method: Local behavior of masonry wall compo-
nents are represented in detail by employing the FE analysis, and this information is used in a
simplified manner by employing the EF method in developing the analytical model of the
unreinforced masonry building under consideration. Hence it becomes possible to analyze
unreinforced masonry buildings in a practical way by using any standard structural analysis
platform. The proposed approach has been verified both in the component level and at the
structural level by comparing the analytical results with the previous experimental findings.
In the last part of the study, the proposed approach is utilized for an existing unreinforced
masonry building that was damaged in a previous earthquake. Both nonlinear static and
dynamic analyses were applied to the generated EF model of the case study building.
The results were satisfactory for the purpose of simulating the global behavior. Based on
the limitations and assumptions of the proposed hybrid approach, the conclusions can be
stated as follows:
• It is possible to model local behavior of masonry wall components in detail by using
the finite-element method. Since the available material database of ANSYS does not
contain suitable models to simulate masonry behavior specifically, different failure
criteria have been used together to develop a more realistic model for masonry in
terms of cracking, crack closing, crushing, stress relaxation, and shear transfer. The
obtained results have a good match with the experimental data from literature.
• The proposed approach enables quick and simple estimation of lateral capacity
curves of masonry wall components in terms of basic structural parameters like com-
pressive strength, axial load ratio, and aspect ratio. Once the element database is
constructed, the idealized capacity curves of masonry components can be obtained
easily without performing extra FE analysis and by only using the simple empirical
relationships. The empirical formulations have proven to be satisfactory after com-
parisons with actual FE analysis of masonry components.
• The hybrid approach is tested by using both experimental and field data. The results
are satisfactory by considering the limitations and assumptions involved. Perfor-
mance limit states can be attained through pushover analysis of the EF model of
the building. Hence it becomes possible to estimate the force and displacement capa-
city of unreinforced masonry buildings in a simple and realistic way through frame
analysis. Nonlinear time-history analysis also yields encouraging results, keeping in
mind the limitations induced by the use of simple hysteresis models. To obtain a
closer match to the actual behavior, a more complicated hysteresis model that
captures the cyclic behavior and the corresponding failure modes of masonry
wall components can be used by sacrificing the simplicity that the proposed
approach offers.
• Overall, the proposed approach seems to be a useful tool since the main focus of this
study is to propose a simple EF model that can be used by structural engineers for
capturing the global behavior of unreinforced masonry structures with reasonable
accuracy. Based on the assumptions and limitations of the approach, it should be
used with caution, especially in estimating the dynamic response of unreinforced

masonry structures by using simple hysteresis models.
REFERENCES
Abrams, D. P., 2001. Performance-based engineering concepts for unreinforced masonry
building structures, Progress in Structural Engineering and Materials 3, 48–56.
Abrams, D., Smith, T., Lynch, J., and Franklin, S., 2007. Effectiveness of rehabilitation on seis-
mic behavior of masonry piers, Journal of Structural Engineering, ASCE, 133, 32–43.
Aldemir, A., 2010. A Simple Seismic Performance Assessment Technique for Unreinforced
Brick Masonry Structures, Master’s Thesis, Middle East Technical University Civil Engineer-
ing Department, Ankara, Turkey.
American Society of Civil Engineers (ASCE), 2000. Prestandard and Commentary for the
Seismic Rehabilitation of Buildings, prepared for the SAC Joint Venture, published by the
Federal Emergency Management Agency, Report No. FEMA-356, Washington, D.C.
ANSYS Inc., 2007. Basic Analysis Guide for ANSYS 11, SAS IP Inc.
Applied Technology Council (ATC), 1996. Seismic Evaluation and Retrofit of Concrete Build-
ings, volumes 1 and 2, Report No. ATC-40, Redwood City, CA.
Attard, M. M., Nappi, A., and Tin-Loi, F., 2007. Modeling fracture in masonry, Journal of Struc-
tural Engineering, ASCE, 133, 1385–1392.
Bayulke, N., 1992. Masonry Structures, Ministry of Public Works and Settlement, Disaster
Affairs General Directorate, Ankara, Turkey, 110 pp. (in Turkish).
Belmouden, Y., and Lestuzzi, P., 2009. An equivalent frame model for seismic analysis of
masonry and reinforced concrete buildings, Construction and Building Materials 23, 40–53.
Benedetti, D., and Pezzoli, P., 1996. Shaking Table Tests on Masonry Buildings: Results and
Comments, ISMES Report, Politecnico di Milano, Italy.
Calvi, G.M., 1999. A displacement-based approach for vulnerability evaluation of classes of
buildings, Journal of Earthquake Engineering 3, 411–438.
Casolo, S., and Pena, F., 2007. Rigid element model for in-plane dynamics of masonry walls
considering hysteretic behavior and damage, Earthquake Engineering and Structural
Dynamics 36, 1029–1048.
D’Ayala, D. F., 2005. Force and displacement-based vulnerability assessment for traditional
buildings, Bulletin of Earthquake Engineering 3, 235–265.
D’Ayala, D., Spence, R., Oliveira, C., and Pomonis, A., 1997. Earthquake loss estimation for
Europe’s historic town centers, Earthquake Spectra 13, 773–793.
Demirel, I. O., 2010. A Nonlinear Equivalent Frame Model for Displacement Based Analysis of
Unreinforced Brick Masonry Buildings, Master’s Thesis, Middle East Technical University
Civil Engineering Department, Ankara, Turkey.
Dolce, M., 1989. Models for In-Plane Loading of Masonry Walls, Corso sul consolidamento degli
edifici in muratura in zona sismica, Ordine degli Ingegneri, Potenza.
ElGawady, M., Lestuzzi, P., and Badoux, M., 2005. In-plane seismic response of URM walls
upgraded with FRP, Journal of Composites for Construction, ASCE, 9, 524–535.
Erbay, O. O., 2007. A Methodology to Assess Seismic Risk for Populations of Unreinforced
Masonry Buildings, Report No. 07-10, Mid-America Earthquake Center, University of Illinois
at Urbana-Champaign.
Erberik, M. A., 2010. Seismic risk assessment of masonry buildings in Istanbul for effective risk
mitigation, Earthquake Spectra 26, 967–982.
Gabor, A., Bennani, A., Jacquelin, E., and Lebon, F., 2006. Modeling approaches of the in-plane
shear behavior of unreinforced and FRP strengthened masonry panels, Composite Structures
74, 277–288.
Gambarotta, L., and Lagomarsino, S., 1997. Damage models for the seismic response of brick
masonry shear walls. Part I: The mortar joint model and its applications, Earthquake Engi-
neering and Structural Dynamics 26, 423–439.
Ganz, H. R., and Thürlimann, B., 1984. Tests on Masonry Walls Under Normal and Shear
Loading, Report No: 7502-4, Institute of Structural Engineering, ETH, Zurich, Switzerland
(in German).
Kappos, A.J., Panagopoulos, G., Panagiotopoulos, C., and Penelis, G., 2006. A hybrid method for
the vulnerability assessment of R/C and URM buildings, Bulletin of Earthquake Engineering
4, 391–413.
Kappos, A.J., Penelis, G. G., and Drakopoulos, C. G., 2002. Evaluation of simplified models for
lateral load analysis of unreinforced masonry buildings, Journal of Structural Engineering,
ASCE, 128, 890–897.
Karatoni, F. V., and Fardis, M. N., 1992. Computed versus observed seismic response and
damage of masonry buildings, Journal of Structural Engineering, ASCE, 118, 1804–1821.
Lofti, H. R., and Shing, P. B., 1994. Interface model applied to fracture of masonry structures,
Journal of Structural Engineering, ASCE, 120, 63–80.
Lourenço, P. B., 1996. Computational Strategies for Masonry Structures, Ph.D. Dissertation,
Delft University of Technology, the Netherlands.
Lourenço, P. B., Milani, G., Tralli, A., and Zucchini, A., 2007. Analysis of masonry structures:
Review of and recent trends in homogenization techniques, Canadian Journal of Civil
Magenes, G., and Fontana, A. D., 1998. Simplified non-linear seismic analysis of masonry build-
ings, in Proceedings of the Fifth International Masonry Conference, vol. 8, London, England,
190–195.
METU-EERC, 1996. Repair and Rehabilitation Project for Moderately Damaged Masonry
Structures after 1995 Dinar Earthquake, Technical Report, Earthquake Engineering Research
Center, Civil Engineering Department, METU Ankara.
Park, J., Towashiraporn, P., Craig, J. I., and Goodno, B. J., 2009. Seismic fragility analysis of
low-rise unreinforced masonry structures, Engineering Structures 31, 125–137.
Pasticier, L., Amadio, C., and Fragiacomo, M., 2008. Non-linear seismic analysis and
vulnerability evaluation of a masonry building by means of the SAP2000 V.10 code,
Penelis, G. G., 2006. An efficient approach for pushover analysis of unreinforced masonry
(URM) structures, Journal of Earthquake Engineering 10, 359–379.
Roca, P., Molins, C., and Marí, A.R., 2005. Strength capacity of masonry wall structures by the
equivalent frame method, Journal of Structural Engineering, ASCE, 131, 1601–1610.
Salonikios, T., Karakostas, C., Lekidis, V., and Anthoine, A., 2003. Comparative inelastic
pushover analysis of masonry frames, Engineering Structures 25, 1515–1523.
Takeda, T., Sozen, M. A., and Nielsen, N. N., 1970. Reinforced concrete response to simulated
earthquakes, Journal of Structural Engineering, ASCE, 96, No. ST12, 2257–2573.
Teran-Gilmore, A., Cuevas, O. Z., and Garcia, J. R., 2009. Displacement-based seismic assess-
ment of low-height confined masonry buildings, Earthquake Spectra 25, 439–464.
Tomazevic, M., 2007. Damage as a measure for earthquake resistant design of masonry
structures: Slovenian experience, Canadian Journal of Civil Engineering 34, 1403–1412.
Turkish Earthquake Code, 2007. Specifications for the Buildings to Be Constructed in Disaster
Areas, Ministry of Public Works and Settlement, Ankara, Turkey.
Willam, K. J., and Warnke, E. D. 1975, Constitutive model for the triaxial behavior of concrete, in
Proceedings of the International Association for Bridge and Structural Engineering, vol. 19,
1–30.
(Received 14 June 2011; accepted 1 March 2012)
OPINION PAPER
The Impact of Earthquakes

on the Domestic Stock Market
Bert Scholtensa) and Yvonne Voorhorsta)
How do financial markets respond to the impact of earthquakes? We inves-

tigate this for more than 100 earthquakes with fatalities in 21 countries from five
continents in the period 1973–2011. Using an event study methodology we con-
clude that there are significant negative effects on stock market value. We find
that the stock market’s response to earthquakes is more pronounced in recent
years than in the 1970s and 1980s. There is no difference in the responses to
the most and least severe earthquakes or to those in high-income and low-income
countries. There is hardly a significant difference in the response to earthquakes
in German law–based countries and English or French law–based countries. This
suggests that the stock market is not very sensitive regarding key characteristics
of earthquakes. [DOI: 10.1193/1.4000103]
INTRODUCTION
Cochrane (2004), Smith and Petley (2009), and DesRoches et al. (2001) mention a wide
range of issues that have to be accounted for when assessing the impact of earthquakes—for
example, losses regarding property and infrastructure, injuries, iconic losses, environmental
impact, interindustry effects, survivor and unemployment benefits, rebuilding assistance, and
social and systemic losses. Most studies that try to assess the impact of earthquakes find that
they have a significant and negative impact, but there is no agreement about the size of the
impact or about the most appropriate methodology.
Economists have tried to assess the economic impact of earthquakes as well. Albala–
Bertrand (1993) and Hallegatte et al. (2007) investigate the impact of earthquakes on the
economic growth rate. Ramcharan (2007) finds that the exchange rate regime does impact
on the economy’s adjustment after real shocks like earthquakes and hurricanes. Wein and
Rose (2011) investigate the cost of business interruption due to natural disasters. Others look
into the consequences of earthquakes for particular regions, like Nakagawa et al. (2007) for
Tokyo and Tantala et al. (2008) for New York. In addition, some studies look into the impact
of earthquakes on financial markets, like Shelor et al. (1991) and Naoi et al. (2009), who
investigate the housing prices in California and Japan, respectively, and Lamb (1998) and
Kleidt et al. (2009), who look into response of the (re)insurance industry. Worthington and
Valadkhani (2004) and Bernhard and Leblang (2007) use information from the capital
a)
University of Groningen, Department of Economics, Econometrics and Finance, PO Box 800, 9700 AV
Groningen, The Netherlands.
325
326 B. SCHOLTENS AND Y. VOORHORST
markets to investigate the value impact of natural disasters. Here, it is interesting to point out
that several scholars regard stock markets as seismographs of the economy (e.g., Kenett et al.
2012). Earthquakes have inspired Maillet and Michel (2003) to develop measures of stock
market volatility in analogy with the Richter scale.
We use stock market information to arrive at an estimate of the economic value impact of
earthquakes. The stock market participants’ response to shocks provides us with a number
that is available relatively fast, which is a huge advantage over the macro-economic approach
where data is available only with a very considerable time lag. Stock market information
might complement other methods that are employed to get a quick overview about the impact
of earthquakes on society (such as the U.S. Geological Survey’s PAGER system, http://
earthquake.usgs.gov/earthquakes/pager/). This can be relevant for regional and national
authorities and for international aid organizations. The more timely and accurate information
they have, the more effective and efficient their efforts can be. Therefore, the stock market
response to an earthquake may be helpful to authorities that have to manage the disaster, as it
can complement other types of information. Furthermore, it can be useful in entrepreneurial
risk management, for example regarding capital requirements for (re)insurers such as the
European Union’s Solvency II. In addition, investors in the stock market may find this useful
in order to know what to expect and how to hedge or benefit from earthquakes.
We will use the event study methodology to relate the impact of earthquakes to economic
value. Several articles use this approach to investigate the impact of natural disasters on
capital markets. For example, Lamb (1998) analyzes how Hurricane Andrew affects the
stock value of U.S. insurers. Yamori and Kobayashi (2002) investigate if Japanese insurers
benefited from the 1995 Hanshin-Awaji earthquake. Worthington and Valadkhani (2004)
estimate how storms, floods, cyclones, earthquakes, and wildfires impact the Australian
stock market. We contribute to this literature, but our approach is somewhat different.
We use a homogeneous event, namely earthquakes, which make any findings more
straightforward from a policy perspective. Furthermore, we use a large sample of events
(more than 100) instead of focusing on the impact of one single event. This improves
the robustness of the analysis and offers more scope for generalization. Of course, we
are well aware of the fact that each earthquake is unique. Therefore, we will look into several
characteristics of earthquakes and their location in order to find out if these characteristics
indeed matter. More specifically, we look into earthquake severity, number of fatalities,
country income, legal system, and old versus recent earthquakes.
First we will explain our data and the estimation method. Then we present our results,
followed by our conclusion and policy recommendations.
DATA AND METHOD

The earthquake information we use is derived from the National Geophysical Data Center
(http://www.ngdc.noaa.gov). Other economic researchers use this database too (e.g., Brounen
and Derwall 2010). We combine it with stock return data from Datastream, which started in
1973. Since 1973, the National Geophysical Data Center has reported 643 earthquakes with at
least one fatality. The median number of fatalities was five. In our analysis, we include earth-
quakes with at least five fatalities, which reduced the number to 353. Because of substantial
overlap among the earthquakes in the estimation and event window, 101 earthquakes remained.
THE IMPACT OF EARTHQUAKES ON THE DOMESTIC STOCK MARKET 327
Most earthquakes occurred in Indonesia (18), China (13), Japan (12), and Turkey (8)
(see Online Appendix A for a list of the earthquakes used in the analysis).
We use the event study methodology to investigate whether earthquakes significantly
impact the national stock markets of the countries in which the earthquake occurs. The
event study is a well-established approach in finance and economics (MacKinlay 1997)
that relies on the assumption that stock market participants are forward-looking. It is
based on a model that holds that prices will only move when new information—that is, infor-
mation that is unexpected for the market participants—becomes available. Event studies are
ubiquitous in capital market research. Kothari and Warner (2005) report that during the per-
iod 1974–2005 the top five finance journals published 565 articles that used event study
results. Lamb (1998) establishes that stock markets are informationally efficient, even in
times of natural disasters. Mitchell and Netter (1994) summarize the essentials of the method:
An event study is a statistical technique that estimates the stock price impact of occurrences such as
mergers, earnings announcements, and so forth. The basic notion is to disentangle the effects of
two types of information on stock prices—information that is specific to the firm under question
(e.g., dividend announcement) and information that is likely to affect stock prices marketwide
(e.g., change in interest rates).
The event study provides a measure about the economic value effect of an event. (See
Online Appendix B for a brief outline of the general structure of an event study.) This type of
study is very useful as, given rationality in the financial markets, the effect of the event will
be reflected immediately in asset prices (Corrado 2011). The economic impact of an event
can be measured by using asset prices that are observed over a short time period. This con-
trasts with more direct measures, which may require many months or even years of observa-
tion (Campbell et al. 1997). Furthermore, it is not always clear whether damage and
casualties result from the earthquake itself or from a chain of related events.
A key issue in event studies is the assumption that there are normal returns and, hence,
abnormal returns. In this respect, basically two approaches are available to calculate the
abnormal returns, namely statistical and economic models. Models in the first category
are based on the statistical assumptions concerning the behavior of asset returns and they
do not depend on any theoretical arguments (Campbell et al. 1997). Economic models
are mainly based on assumptions about investor behavior (MacKinlay 1997, Corrado
2011). The convention with the statistical models is to assume that asset returns are jointly
multivariate normal and independently and identically distributed over time (Campbell et al.
1997). This assumption allows for the development of exact finite-sample distributional
results for the estimators and statistics; inferences using the normal return models are robust
to deviations from the assumption (Campbell et al. 1997). The constant mean return model
assumes that the mean return of a given security is constant through time. The market model
is a statistical model that relates the return of any given security to the return of the stock
market portfolio (Brown and Warner 1980). It relies on a linear specification, assuming the
asset weights in the market portfolio are constant, and follows from the assumed joint nor-
mality of asset returns. This market model assumes that there is a stable linear relation
between the market return and the security return. It removes the portion of the return
that is related to variation in the market’s return and, as such, reduces the variance of
the abnormal return. This may lead to increased ability to detect event effects (Campbell
et al. 1997). Economic models restrict the parameters of statistical models to provide more
constrained normal return models. The capital asset pricing model (CAPM) is commonly
used in event studies (Corrado 2011). This model imposes constraints on the constant
term relative to the market model. Furthermore, it uses a constant risk-free rate. The
CAPM investigates what returns to expect for a particular stock. It is based on variables
that are easily observed in the market, namely the risk-free interest rate (i.e., domestic
government bond return), the market-risk or equity-risk premium, the variance of the market,
and the covariance of the stock with the market. As such, the CAPM is basically a regression
of stock returns on market returns.
We use CAPM as the basis for our event study as it is validated by many studies and can
be regarded as the standard methodology (Brown and Warner 1980, Brown and Warner
1985, Campbell et al. 1997). This model accounts for the price (return) development on
the stock market that is involved in the unexpected news (the earthquake) in relation to
a global benchmark (the global stock market index). Relating to an international benchmark
takes out part of the systematic and interrelated price movements in the different international
markets (Park 2004, Campbell et al. 2010). Instead of using firm-specific stock market infor-
mation, sectoral or country indices can be used as the methodology is robust in this perspec-
tive (Brown and Warner 1985, Bos 1994, Park 2004, Chukwuogor 2008). The CAPM also
accounts for the country-specific risk of the stock market. Then, expected returns are com-
pared with actual stock market returns surrounding the earthquake. Thus we explicitly
assume that the earthquake is an unexpected event from the perspective of the investment
community. This event may affect the value of corporations that get hit. We test whether
earthquakes result in differences between the normal (i.e., expected) and the actual returns
in the stock market.
To arrive at abnormal returns, we first need to calculate returns. The return of the country
i’s stock on day tis defined as the log of the stock market index on day tminus the log of the
index on day t− 1. The market and risk-adjusted returns model calculates the beta of the
index, which reflects the volatility of the stocks’ returns vis-à-vis a benchmark, in our
case an international stock market index called the Datastream Total World Market
Index. This beta is calculated by relating the covariance of the stock return and the market
return to the variance of the market return. The specific risk (the alpha) of the firm is defined
as the stock’s return minus the beta times this return. To arrive at the abnormal return of
country ion day t, we subtract the alpha and beta times the world market return from
the country’s return.
We first calculate the return on the domestic stock market index1 in the estimation win-
dow, which is a period of time that predates the earthquake. These returns are the normal or
expected returns. In the event window, which is the period of time surrounding the earth-
quake, we take the difference between the actual returns and the expected returns. This is the
abnormal return. We average these returns for all earthquakes, and this is the average
abnormal return (AAR) for that particular day in the event window. We test whether
1
We use the total market return indices provided by Datastream for Australia, Chile, China, Colombia, France,
Greece, India, Indonesia, Italy, Japan, Mexico, New Zealand, Pakistan, Peru, the Philippines, Russia, Turkey, the
United States, and South Africa.
these abnormal returns are significantly different from zero. If so, we conclude that the earth-
quake can be associated with a significant stock market response.
We define the event day (day 0) as the day on which the news about the earthquake
becomes available to the market. When the event occurs on a nontrading day (in some
cases, earthquakes can result in several days’ closure of the stock market), the first trading
day is defined as the event day (Worthington and Vladkhani 2004). We use an estimation
window of 100 trading days to estimate the normal returns: −100; −1. This is in line with the
approach used by Shelor et al. (1991) and Bernhard and Leblang (2007). In the event window
we investigate the returns from day −1 to day +5 (Brown and Warner 1985). This window is
much shorter than that of most other studies (like Shelor et al. 1991, Bernhard and Leblang
2007, Kleidt et al. 2009). Our window is based on the immediate response that shows up in
previous studies. For each event, we use the data from the respective national and global
market indices during the −100; +5 time period. Apart from investigating average abnormal
returns of the earthquakes, we also investigate the cumulative average abnormal return
(CAAR). This CAAR calculates the response of the stock market for a particular multiday
period. The CAAR of period t− x, tis the sum of the average abnormal returns from day t− x
up to and including day t.
The descriptive statistics of returns of the country indices in the estimation window of the
101 earthquakes are in Table 1. They show that the returns are skewed and leptokurtic. As
such, the returns series are not normally distributed and we will engage in nonparametric
testing. For this nonparametric test, we use the Wilcoxon signed-rank test (Newbold et
al. 2007), which accounts for the sign of the difference as well as for the size of the differ-
ences between pairs. For the pairs, the Wilcoxon test investigates the difference in the
medians.
Several studies argue that there is a huge negative economic impact from an earthquake
(Albala-Bertrand 1993). For example, Cavallo et al. (2010) suggest that the 2010 Haiti earth-
quake had a lower-bound economic impact of $8.1 billion. However, Nakamura (2011)
argues that the impact of the March 2011 earthquake in Japan is not straightforward. On
the one hand, it reduced the supply capacity of the economy. At the same time, however,
if weak demand also impacts on output, the overall impact on prices and the economy is not
straightforward. We will test the hypothesis that an earthquake does significantly impact on
Table 1. Descriptive statistics of the market- and risk-adjusted returns

in the estimation window ðN ¼ 101Þ
Mean 0.0000
Median −0.0003
Standard deviation 0.0159
Minimum −0.1320
Maximum 0.1780
Skewness 0.3895
Kurtosis 11.5800
the national stock market of the country it hits. The null hypothesis is that there is no
significant impact.
To substantiate our findings, we engage in a number of sensitivity checks. We inves-
tigate whether the market response did change over time, if the market response to the
least severe earthquakes (in terms of fatalities) is weaker than to the most severe earth-
quakes, and if markets react more strongly to earthquakes in high-income countries than
to those in low-income countries. We also investigate whether the legal system is a mod-
erating factor. In addition, we want to find out whether the criterion of five fatalities is a
decisive factor. The motivation for the analysis and details will be discussed in the results
section.
RESULTS
The main results are displayed in Table 2. They show that there is a negative and sta-
tistically significant abnormal return on the event day and on day 3 in the event window. On
the event day, this average abnormal return is −0.53% (which amounts to about −12% on an
annualized basis). Apart from day 2, all abnormal returns are negative. Because some of the
existing research suggests there may be a somewhat lagged response, we look into several
time intervals in the event window in Table 3. This shows that this indeed may be the case as
the cumulative adjusted returns are statistically significant and negative in all subperiods.
During the first two days, they are −0.63% (−9% annualized) and during the complete
event period they are on average −1.01% (−7% on an annual basis), which suggests
there is some initial overreaction. We did a robustness test with the mean returns model
Table 2. Estimation results of the market- and risk-adjusted returns model for average
abnormal returns in the event window ðdf¼ 100Þ
Day AAR p-value (Wilcoxon test)

0 −0.0053 0.02
1 −0.0010 0.33
2 0.0003 0.37
3 −0.0021 0.01
4 −0.0011 0.24
5 −0.0009 0.33
Table 3. Estimation results of the market- and risk-adjusted returns model for cumulative
average abnormal returns in the event window ðdf¼ 100Þ
Period CAAR p-value (Wilcoxon test)

[0; 1] −0.006 0.018
[0; 3] −0.008 0.010
[0; 5] −0.010 0.011
and the market-adjusted returns model (not reported here but available upon request with the
contacting author). The former does show negative (cumulative) average abnormal returns,
but not significantly so at the 5% level. The latter gives results that are very similar to those of
the market and risk adjusted returns regarding the order of magnitude of both the coefficients
and the p-values of the significance tests.
Our findings are in line with the impact of earthquakes on the real estate market (as
studied by Shelor et al. 1991), with that of natural disasters on the Australian stock market
by Worthington and Valadkhani (2004), with that of disasters on international markets by
Bernhard and Leblang (2007), and with the impact of catastrophes on insurance compa-
nies by Kleidt et al. (2009). However, the (cumulative) average abnormal returns of the
earthquakes (Tables 2 and 3) show a less pronounced response and the significance is
slightly less compared to the studies that focus on heterogeneous and extreme disasters.
Our findings suggest that domestic stock markets respond in a straightforward and sig-
nificant manner to earthquakes and that there is a slight overreaction. For our 101 earth-
quakes, the response within one business week shows an average decline of more than 6%
of the market value of the firms that are listed on the domestic stock exchange. This is
about half the size of the response on the day the news about the earthquake breaks to the
market.
Miller and Keipi (2005) and Cannock et al. (2011) point out that it is crucial for autho-
rities to understand the potential impact of earthquakes and prepare strategies for managing
these disasters. Given that stock markets are efficient, even in times of stress (Lamb 1998),
our findings might be helpful for these authorities as they complement more conventional
instruments that are used to assess the impact of disasters (see Cochrane 2004, Smith and
Petley 2009). Furthermore, it helps entrepreneurs to assess the expected impact of an earth-
quake, which is important in their hedging and planning decisions. For investors, the find-
ings are relevant as they might arrive at speculative profits in the first days after the
earthquake due to the overreaction, and they can include the impact of the shocks in
the positioning of their investment portfolios. However, it is important to find out whether
stock markets are sensitive to the specifics of an earthquake. To this extent, we engage in a
number of sensitivity checks to find out whether there are significant differences as to the
stock market response to earthquakes along time, severity, per capita income, and legal
system.
First, we investigate whether the stock market response did change over time. Given the
changes in the relative number of people living in cities, continued population growth, and
increasing welfare (see Maddison 2007, Grimm et al. 2008), there might be reason to expect
that human and material value destroyed in earthquakes has risen with the course of time
(Kunreuther and Heal 2012). In addition, the fact that (re)insurance companies have increas-
ingly ended the insurance coverage for earthquakes in many countries might have had an
impact (Swiss Re 2012). In fact, we hypothesize there will be a stronger response to
more recent earthquakes. To test this, we split the sample in three equal parts and compare
the oldest earthquakes with the youngest in our sample. This means that we compare the
stock market response between the first and the last 30 earthquakes in the sample and
leave out the ones in between. We do this because we are not sure whether there actually
is a break point in the series and because there is no clear cutoff point that separates “old” and
“new” earthquakes. We calculate (cumulative) average abnormal returns of the two

subgroups and perform a t-test on the differences. We report the main findings in Online
Appendix C. The differential within the event window is positive (i.e., the average abnormal
returns are higher for the more recent earthquakes), but it is not statistically significant. When
we compare the cumulative average abnormal returns of the two subgroups, the positive
differential of almost 2% in the event period is statistically significant. We conclude that
more recent earthquakes result in a more pronounced response and that this differential
is statistically significant. It suggests that increased welfare and urbanization and reduced
coverage could have an impact on the amount of economic value at risk, and that the
stock markets do account for these changes.
Second, we compare the market response to the least and most severe earthquakes as
measured by the number of fatalities. Here, we expect a more pronounced reaction to the latter.
We hypothesize that there is more damage with the heavier earthquakes that destroy more
economic value. Samardjieva and Badal (2002) find a positive association between the number
of casualties and the strength of earthquakes. Page et al. (1975) and Zhai and Xie (2007) find
the same for damage to buildings and damage in general, respectively. We therefore hypothe-
size that the response to the most severe earthquakes is significantly stronger than that to the
least severe earthquakes in the sample. To test this, we compare the stock market response to
the 33% most severe earthquakes with that to the 33% least severe earthquakes. Thus, we leave
out one-third of the earthquakes because we do not have an a priori about what is severe and it
is difficult to arrive at an objective criterion to distinguish more and less severe earthquakes. We
calculate (cumulative) average abnormal returns of the two subgroups and perform a t-test on
the differences. The results are in Online Appendix D. We do not detect significant (i.e., 95%
confidence) differences in the stock market response to the most and the least severe
earthquakes. In fact, an earthquake is sufficient for a response and the stock market does
not seem to account for severity of the earthquake in terms of fatalities. This could be
due to the fact that short-term effects are difficult to grasp for stock market participants.
This finding also shows that the stock market is not very sensitive as to the value impact
of earthquakes.
Third, we want to find out if stock markets in high-income countries react more strongly
than those in low-income countries. This is based on the study of Anbarci et al. (2005), who
find that the number of fatalities resulting from an earthquake is inversely related to a
country’s per capita level of income. They point out that it is society’s capability to arrive
at what all parties perceive to be an agreeable distribution of the burden of collective action
that causes the relatively wealthy to self-insure against the disaster while leaving the poor to
its mercy. Keefer et al. (2010) suggest that especially in low-income countries, there will not
be as much focus on the costs and casualties of an earthquake. Also, in poor economies there
will be a lack of funds to mitigate the impact of earthquakes. This would suggest that
the markets in low-income countries respond in a less pronounced way to earthquakes
than those in high-income countries. To test this, we compare the stock market response
in countries that are classified by the World Bank as low-income or medium-low income
countries with those in high-income countries (http://www.data.worldbank.org). It appears
that there are 44 earthquakes in the former group and 42 in the latter. We leave out the 15
earthquakes that occurred in upper-medium-income countries. The results are in Online
Appendix E. We find that there is no statistically significant difference in the responses

between the two groups. Thus, in contrast to Anbarci et al. (2005) and Keefer et al.
(2010), stock market investors always seem to respond negatively to an earthquake, irrespec-
tive of the country that is hit.
Furthermore, we compare whether the legal system is a moderating factor in the
stock market response to earthquakes. Anbarci et al. (2005) hint at the role of regulation
in the severity of earthquakes. Keefer et al. (2010) expect that in common law countries,
institutions are regarded as more reliable, which would result in a more modest response
to earthquakes than in civil law countries where there would be less confidence in earth-
quake-mitigation policies. Djankov et al. (2002) and La Porta et al. (2008) conclude that
strict regulation with respect to market entry is to be associated with corruption. La Porta
et al. (2008) find that countries with a civil law origin (German or French) have more
market entry requirements that those with a common law origin. They also suggest that
corruption will be more likely to occur in civil law–based countries. In case there is more
corruption, we expect that any regulations with regards to, for example, earthquake-
proof building is less likely to be enforced and, therefore, that more value is lost
from earthquakes in civil law countries than in common law countries. We investigate
the earthquakes in 19 common law–based countries and 82 civil law countries (56
French-based law and 26 German-based law). The results are in Online Appendix F.
We do find that there is a negative and statistically significant response to earthquakes
in civil law countries (in particular German law–based countries) and not so in common
law countries. We performed a Kruskal-Wallis test on the intergroup differences (see
Newbold et al. 2007) and did not find significant results (not reported). There is
only a marginally significant (i.e., at the 10% level) difference between the responses
to earthquakes in German and English law–based countries. This finding does not really
support the hypothesis that the response to earthquakes is more pronounced in civil law
than in common law countries. Therefore, we conclude there are no statistical significant
differences in the stock market response to earthquakes among common law and civil
law countries. It suggests that the legal system, like the income level, is not decisive for
the stock market response.
Finally, we want to find out if the criterion of five deadly fatalities is a crucial one with
respect to our analysis. To this extent, we do the basic analysis and all the sensitivity tests on
the basis of two alternative criteria. One is where we use four deadly fatalities to select earth-
quakes for our sample (111 earthquakes), the other is where we only include the 80 earth-
quakes with 10 or more fatalities. The basic results only marginally differ. As such, the five
deadly fatalities criterion does seem to be robust. In case of old versus recent earthquakes,
we find that with four fatalities, the significance of the differential drops, as is the case with
the most and least severe earthquakes. We also find that with ten deadly fatalities, there is a
significant difference in the response between earthquakes in high-income versus those in
low- and low-middle income countries. Here, it shows that the value effect of earthquakes in
high-income countries is significantly larger than that in low- and lower-middle-income
countries. When we compare earthquakes among legal systems, the number of deadly
fatalities (four, five, or ten) does not seem to matter.
CONCLUSION
Earthquakes have an economic impact. We try to investigate how the stock market grasps
this impact and if it does account for various properties of the earthquakes. For 101 earthquakes
in 21 countries in the period 1974–2011, we find evidence that earthquakes do have a signif-
icant impact on the domestic stock market. In particular, the average abnormal returns on the
event day and on day 3 after the event are highly significant. On an annualized basis, we find
that the losses on average are within the range of 6–12% of the total market value of the firms
that are listed on the domestic stock exchange. Our results are generally robust to the way in
which we estimate normal returns. We also investigate whether the stock market is sensitive to
key attributes of the earthquakes. In this respect, we find that the response to recent earthquakes
(2003–2011) is more pronounced than to those in the period 1974–1995. However, the stock
market response to the most severe earthquakes does not significantly differ from that to the
least severe in our sample. Furthermore, there is no significant difference in the response in
low- and lower-middle-income countries compared to high-income countries or in common
law and civil law countries. This suggests that stock market investors in general tend to respond
in an identical manner to earthquakes, namely that they estimate a loss of 6–12% of total market
capitalization. However, at the same time, it also shows that the stock market is not very
sensitive to the specifics of earthquakes.
These results support the findings from the literature that earthquakes have a negative
impact on the economy. But our study finds less pronounced results and the significance is
slightly lower than in most other studies (Shelor et al. 1991, Kleidt et al. 2009, Naoi et al.
2009). This might result from the fact that we include many more earthquakes than previous
studies do (see Lamb 1995, Kleidt et al. 2009), we have a more homogeneous sample than
most other studies (cf. Worthington and Valadkani 2004, Bernhard and Leblang 2007), and
we do not have a bias regarding the most severe earthquakes (cf. Shelor et al. 1995, Naoi
et al. 2009).
As to our policy recommendations, we come up with the following. Although stock markets
are often regarded as the seismographs of the economy and the business community, they are
not very precise when it comes to assessing the impact of earthquakes. This is because they
hardly differentiate along the severity of the earthquake, the income levels of those who are hit,
and the perceived management quality of the authorities. As such, regional and national autho-
rities and (international) aid agencies need not specifically look at the stock market to arrive at a
rapid assessment of the impact of an earthquake. Instead, they better rely on seismographic,
geological, geographical, and social-economic information (Smith and Petley 2009). For entre-
preneurs, however, stock markets can help to assess the expected impact of an earthquake,
which is important in their hedging and planning decisions. For investors, our findings are
relevant as they might arrive at speculative profits in the first days after the earthquake due
to the initial overreaction. Furthermore, investors can include the impact of the shocks in
the positioning of their investment portfolios.
We conclude that the stock market response to earthquakes provides a very crude signal
about the economic value lost. This information is only very partially complementary to
other impact indicators and should in no case be used to replace geological and seismic
analysis.
REFERENCES
Albala-Bertrand, J. M., 1993. Natural disaster situations and growth: A macroeconomic model for
sudden disaster impacts, World Development 21, 1417–1434.
Anbarci, N., Escaleras, M., and Register, C. A., 2005. Earthquake fatalities: The interaction of
nature and political economy, Journal of Public Economics 89, 1907–1933.
Bernhard, W., and Leblang, D., 2007. Standing Tall When the Wind Shifts: Financial Market
Responses to Elections, Disasters and Terrorist Attacks, Centro de Investigacion y Docencia
Economicas, Toluca (Mexico), No. 163.
Bos, J. W. D., 1994. Stock market efficiency: The evidence from FTA indices of eleven major
stock markets, De Economist 142, 455–473.
Brounen, D., and Derwal, J., 2010. The impact of terrorist attacks on international stock markets,
European Financial Management 16, 585–598.
Brown, S. J., and Warner, J. B., 1980. Measuring security price performance, Journal of
Financial Economics 8, 205–258.
Brown, S. J., and Warner, J. B., 1985. Using daily stock returns: The case of event studies,
Journal of Financial Economics 14, 3–31.
Campbell, C. J., Cowan, A. R., and Salotti, V., 2010. Multi-country event study methods, Journal
of Banking and Finance 34, 3078–3090.
Campbell, J. Y., Lo, A. W., and MacKinlay, A. C., 1997. The Econometrics of Financial Markets,
Princeton University Press, Princeton, NJ.
Cannock, G., Silva Yon, J., Jara Trujillo, C. S., O’Connor La Rosa, R., and Bonifaz, F. S., 2011.
Economic Impact of the 2007 Earthquake in Peru in the Water and Sanitation Sector: What
Did Unpreparedness Cost the Country?, World Bank Water and Sanitation Program Technical
Paper 63627.
Cavallo, E. A., Powell, A., and Becerra, O., 2010. Estimating the direct economic impact of the
earthquake in Haiti, Development 120 , 298–312.
Chukwuogor, C., 2008. Stock markets returns and volatilities: A global comparison, Interna-
tional Research Journal of Finance and Economics 15, 7–22.
Cochrane, H., 2004. Economic loss: Myth and measurement, Disaster Prevention and Manage-
ment 13, 290–296.
Corrado, C. J., 2011. Event studies: A methodology review, Accounting & Finance 51, 207–234.
DesRoches, R., Comerio, M., Eberhard, M., Mooney, W., and Rix, G. J., 2011. Overview of the
2010 Haiti earthquake, Earthquake Spectra 27, S1–S21.
Djankov, S., La Porta, R., Lopez-de-Silanes, F., and Shleifer, A., 2002. The regulation of entry,
Quarterly Journal of Economics 117, 1–37.
Grimm, N. B., Faeth, S. H., Golubiewski, N. E., Redman, C. L., and Briggs, J. M., 2008. Global
change and the ecology of cities, Science 319, 756–760.
Hallegatte, S., Hourcade, J., and Dumas, P., 2007. Why economic dynamics matter in assessing
climate change damages: Illustration on extreme events, Ecological Economics 62, 330–340.
Keefer, P., Neumayer, E., and Plümper, T., 2010. Earthquake Propensity and the Politics of
Mortality Prevention, World Bank Policy Research Working Paper 5182.
Kenett, D. Y., Raddant, M., Lux, T., and Ben-Jacob, E., 2012. Evolvement of uniformity and
volatility in the stressed global financial village. PLoS ONE 7, e31144, doi:10.1371/journal.
pone.0031144.
Kleidt, B., Schiereck, D., and Sigl-Gueb, C., 2009. Rationality at the eve of destruction: Insurance
stocks and huge catastrophic events, Journal of Business Valuation and Economic Loss
Analysis 4.
Kothari, S. P., and Warner, J. B., 2005. Econometrics of event studies, in Handbook of Corporate
Finance: Empirical Corporate Finance (Espen Eckbo, B., ed.), Elsevier/North-Holland,
Amsterdam, the Netherlands.
Kunreuther, H., and Heal, G., 2012. Managing catastrophe risk, in Encyclopedia of Energy, Nat-
ural Resources and Environmental Economics (Shogren, J., ed.), forthcoming, http://ssrn
.com/abstract=2049456.
La Porta, R., Lopez-de-Silanes, F., and Shleifer, A., 2008. The economic consequences of legal
origins, Journal of Economic Literature 46, 285–332.
Lamb, R. P., 1998. An examination of market efficiency around hurricanes, Financial Review 33,
163–172.
MacKinlay, A. C., 1997. Event studies in economics and finance, Journal of Economic Literature
35, 13–39.
Maddison, A. 2007. Contours of the World Economy, 1–2030 AD: Essays in Macro-Economic
History, Oxford University Press, Oxford and New York.
Maillet, B., and Michel, T., 2003. An index of market shocks based on multiscale analysis, Quan-
titative Finance 3, 88–97.
Miller, S., and Keipi, K., 2005. Strategies and Financial Instruments for Disaster Risk Manage-
ment in Latin America and the Caribbean, Sustainable Development Department Technical
Paper Series ENV-145, Inter-American Development Bank, Washington, D.C.
Mitchell, M. L., and Netter, J. M., 1994. The role of financial economics in securities fraud cases:
Applications at the Securities and Exchange Commission, Business Lawyer 49, 545–590.
Nakagawa, M., Saito, M., and Yamaga, H., 2007. Earthquake risk and housing rents: Evidence
from the Tokyo Metropolitan Area, Regional Science and Urban Economics 37, 87–99.
Nakamura, K., 2011. The impact of the earthquake on the output gap and prices, Bank of Japan
Review E-4, 1–7.
Naoi, M., Seko, M., and Sumita, K., 2009. Earthquake risk and housing prices in Japan: Evidence
before and after massive earthquakes, Regional Science and Urban Economics 39, 658–669.
Newbold, P., Carlson, W. L., and Thorne, B., 2007. Statistics for Business and Economics.
Pearson, Upper Saddle River, NJ.
Page, R. A., Joyner, W. B., and Blume, J. A., 1975. Earthquake shaking and damage to buildings,
Science 22, 601–608.
Park, N. K., 2004. A guide to using event study methods in multi-country settings, Strategic
Management Journal 25, 655–668.
Ramcharan, R., 2007. Does the exchange rate regime matter for real shocks? Evidence from
windstorms and earthquakes, Journal of International Economics 73, 31–47.
Samardjieva, E., and Badal, J., 2002. Estimation of the expected number of casualties caused by
strong earthquakes, Bulletin of the Seismological Society of America 92, 2310–2322.
Shelor, R. M., Anderson, D. W., and Cross, M. L., 1991. The impact of the California earthquake
on real estate firms’ stock value, Journal of Real Estate Research 5, 335–340.
Smith, K., and Petley, D. N., 2009, Environmental Hazards: Assessing Risk and Reducing
Disaster, 5th edition, Routledge, Abingdon, United Kingdom.
Swiss Re, 2012. Lessons from Recent Major Earthquakes, Swiss Re, Zürich, Switzerland.
Tantala, M. W., Nordenson, G. J. P., Deodatis, G., and Jacob, K., 2008. Earthquake loss estima-
tion for the New York City metropolitan region, Soil Dynamics and Earthquake Engineering
28, 812–835.
Wein, A., and Rose, A., 2011. Economic resilience for the shakeout earthquake scenario, Earth-
quake Spectra 27, 559–573.
Worthington, A., and Valadkhani, A., 2004. Measuring the impact of natural disasters on
capital markets: An empirical application using intervention analysis, Applied Economics
36, 2177–2189.
Yamori, N., and Kobayashi, T., 2002. Do Japanese insurers benefit from a catastrophic event?
Reactions to the 1995 Hanshin-Awaji earthquake, Journal of the Japanese and International
Economies 16, 92–108.
Zhai, C., and Xie, L., 2007. A new approach of selecting real input ground motions for
seismic design: The most unfavourable real seismic design ground motions, Earthquake
Engineering and Structural Dynamics 36, 1009–1027.
(Received 17 October 2011; accepted 21 September 2012)
TECHNICAL NOTE
Estimating Economic Losses from

Earthquakes Using an Empirical Approach
Kishor Jaiswal,a) M.EERI, and David J. Wald,b) M.EERI
We extended the U.S. Geological Survey’s Prompt Assessment of Global

Earthquakes for Response (PAGER) empirical fatality estimation methodology
proposed by Jaiswal et al. (2009) to rapidly estimate economic losses after sig-
nificant earthquakes worldwide. The requisite model inputs are shaking intensity
estimates made by the ShakeMap system, the spatial distribution of population
available from the LandScan database, modern and historic country or sub-country
population and Gross Domestic Product (GDP) data, and economic loss data from
Munich Re’s historical earthquakes catalog. We developed a strategy to approxi-
mately scale GDP-based economic exposure for historical and recent earthquakes
in order to estimate economic losses. The process consists of using a country-
specific multiplicative factor to accommodate the disparity between economic
exposure and the annual per capita GDP, and it has proven successful in hindcast-
ing past losses. Although loss, population, shaking estimates, and economic data
used in the calibration process are uncertain, approximate ranges of losses can be
estimated for the primary purpose of gauging the overall scope of the disaster and
coordinating response. The proposed methodology is both indirect and approxi-
mate and is thus best suited as a rapid loss estimation model for applications like
the PAGER system. [DOI: 10.1193/1.4000104]
INTRODUCTION
Estimating realistic economic impact due to a large-scale earthquake disaster can be a
daunting task. Due to the inherent complexity of estimating infrastructural and institutional
impacts (in terms of direct and indirect losses), the true value of economic impact from such
disasters may only be approximately known. Months or years following large earthquakes,
the estimates on economic losses can still remain in flux; for example, losses for the 2010
Haiti earthquake and the 2011 Japan earthquake are only partially tallied after several months
following the earthquake. Tallies of the economic impact for the 1994 Northridge, California,
earthquake continued to grow for roughly a decade. Seligson and Eguchi (2005) revised
the initial estimate of $44 billion in 1997 to $57 billion in 2004 mainly by adding business-
interruption and casualty-related losses and by revising the insured losses. Even then, in both
estimates, the assumed cost of $20 billion related to deductible, insured, and uninsured losses
were highly approximate values.
a)
U.S. Geological Survey, Golden CO (contracted through Synergetics Incorporated) kjaiswal@usgs.gov
b)
U.S. Geological Survey, Golden CO
309
310 K. JAISWAL AND D. J. WALD
In the realm of post-earthquake response planning, it is necessary that the stakeholders be

alerted about the potential economic disaster that may be unfolding after a large earthquake.
A recent example pertains to the M5.8 23 August 2011 Virginia earthquake; the lack of
obvious catastrophic losses early on was taken as an indication that losses would be rather
minimal, though actual tallies a month later well exceeded $100 million (Bausch 2011, writ-
ten communication), and continued to grow thereafter. Is it possible to project such losses in
the immediate aftermath of such events?
The U.S. Geological Survey’s Prompt Assessment of Global Earthquakes for Response
(PAGER) program aims to inform humanitarian aid decisions in the hours or days after the
earthquake before firsthand, ground-truth observations are available to assess the true scope
of humanitarian needs. In order to institutionalize a globally consistent operational earth-
quake response procedure, impact assessments for earthquake fatalities and economic losses
for both developing and developed parts of the world are needed (Wald et al. 2011).
This article describes a procedure for rapidly estimating the economic consequences in
the aftermath of a large earthquake disaster. First, an estimate is made of the country- or
region-specific economic loss ratio (defined here as total direct economic loss normalized
by the total economic exposure), which varies as a function of shaking intensity and is cali-
brated against historical earthquakes losses. The economic value of all the physical assets
exposed at different locations in a given area is generally not known and can be extremely
difficult to compile at a regional scale, let alone at a global scale. In the absence of such a
dataset, we estimate the total gross domestic product (GDP) exposed at each shaking intensity
by multiplying the per-capita GDP of the country by the total population exposed at that
shaking intensity level. We then scale the total GDP estimated at each intensity by an expo-
sure adjustment multiplier to account for the disparity between wealth and/or economic assets
and the annual GDP.
The economic exposure obtained using this procedure is thus a proxy for the economic
value of the actual inventory (which includes buildings, infrastructure and other assets) that is
exposed to the earthquake. Similarly, the economic loss estimated using this approach repre-
sents a direct loss to building and other infrastructure, and does not include economic impacts
due to indirect losses. In essence, this approach bypasses the requirement of detailed building
inventories, which may or may not be available for certain parts of the world.
This article illustrates the development of a country- or region-specific economic loss
ratio model using economic loss data available for global earthquakes from 1980 to
2007. Further, it illustrates the economic loss estimation for a selected earthquake and
demonstrates suitability of this approach within the PAGER system.
MACROECONOMIC STUDIES OF PAST EARTHQUAKES

Earthquake-induced damage and losses are an outcome of several complex and interde-
pendent phenomena that usually happen within several seconds of strong ground shaking
during an earthquake. In order to understand the causes of damage (to buildings, infrastruc-
ture, and other facilities) and the resulting losses (casualties, economic impacts), we first need
to establish the physical damageability relations and mortality vulnerability at various shak-
ing levels. Detailed damage and loss analyses thus require a number of ingredients, for
ESTIMATING ECONOMIC LOSSES FROM EARTHQUAKES USING AN EMPIRICAL APPROACH 311
example, an inventory of building and infrastructure exposure; relative vulnerability of built

environment to ground shaking; an assessment of socioeconomic wealth associated with the
exposed assets; and an understanding of the region’s productivity, economic growth, and
resiliency (Brookeshire et al. 1997, Kircher et al. 2006). Compilation of such factors at a
global scale is a mammoth task in itself, and despite the availability of data for certain
parts of the world, the knowledge and tools are often unavailable to perform quantitative
risk and loss analyses. In addition, the availability of required data and models does not
ensure useful loss estimates unless the estimates are calibrated against historical earthquake
losses.
Several researchers have attempted to simplify the problem of economic loss assessment
using a macroeconomic-based empirical approach (Chan et al. 1998, Dunbar et al. 2002,
Yong et al. 2001). In principle, these approaches consisted of developing an empirical rela-
tionship between loss/GDP and the seismic intensity, using dozens of significant historical
earthquakes, for the cases of low-, mid-, and high-income countries. For example, Chan et al.
(1998) divided nations according to their GDPs into categories of high income (GDP >
$8,300), medium-income ($675–$8,300), and low-income (< $675) economies. The loss-
GDP relations derived using these broad categories thus provided limited insight when
estimating economic vulnerability of an individual country. The macro-economic based
approach requires historical data on Loss/GDP ratios at each intensity unit in order to con-
strain these empirical relationships. Such data are rarely available for most historical earth-
quakes worldwide. Also, it is quite challenging to estimate Loss/GDP ratios at a local scale,
due to (a) poor reporting in the literature on extent of the affected zone/damage footprint of
individual earthquakes, (b) limited knowledge about local GDP of an affected region of a
country, and (c) only aggregate losses available for individual earthquakes. In order to cir-
cumvent this, researchers often associate the reported losses with some “arbitrary zone” of
maximum earthquake loss, which is an area within 30 km of the epicenter of an earthquake
(Dunbar et al. 2002, Yong et al. 2001). This overly simplified hazard characterization may
be feasible for small- to moderate-sized earthquakes with concentrated exposure limited to
the epicentral area. However, for most global earthquakes the terms of earthquake losses are
dictated by underlying complexities of earthquake fault type, dimension and type of earth-
quake rupture, site conditions (bed rock vs. soft alluvium), and the spatial distribution of
exposed assets as well as their relative vulnerability to ground shaking.
A recent macroeconomic-based approach includes a global risk map for economic losses
for six major natural hazards, including earthquakes, by Columbia University under the
umbrella of the ProVention Consortium of the World Bank (Dilley et al. 2005). The meth-
odology and underlying dataset were further improved in the Global Assessment Report on
Disaster Risk Reduction (ISDR 2009). In order to estimate economic exposure, the study
relied on raster distribution of the gross domestic product (GDP) defined in terms of a pur-
chasing power parity (or PPP value).
Total earthquake loss from any single earthquake is not limited by GDP and it can exceed
the GDP of the region or country. For example, the 6 November 1988 Burma-China earth-
quake caused losses on the order of 4.6 times the nominal GDP of the region (i.e., a total of
$269 million of economic loss for the regional GDP of $58 million at the time of earthquake,
see Table I of Chan et al. 1998); the 19 August 1992 Kyrgyzstan earthquake resulted in
log(loss)-to-log(GDP) ratios of the order of 5.694 (Dunbar et al. 2002); and the 2010 Haiti
earthquake losses exceeded the total nominal GDP of the country (http://www.
haitispecialenvoy.org/relief-and-recovery/key-statistics/). Therefore, in order to characterize
actual earthquake exposure in any given earthquake, we need to (a) find a way to apportion
the total GDP of the country to the part of the country that is affected by the earthquake
and (b) account for disparity between national GDP and economic value of assets that
are exposed in a given area. In the following section, we describe the methodology to esti-
mate country-specific economic loss functions and an approach to characterize economic
exposure in lieu of the requirement to compile building inventory and economic exposure
data worldwide.
GROSS DOMESTIC PRODUCT (GDP)

Since GDP estimates are primarily used to reflect the economic exposure in a given
region, it is important to describe this term in more detail. GDP measures the total output
of goods and services for final use produced by residents and non-residents, regardless of the
allocation to domestic and foreign claims (World Bank 1994). The use of official exchange
rates to convert national currency figures to U.S. dollars does not reflect the relative domestic
purchasing powers of currencies. In order to describe the local purchasing power, the U. N.
International Program (ICP) has developed measures of real GDP on an internationally com-
parable scale, using purchasing power parity of currencies (PPPs) instead of exchange rates
as conversion factors. Thus, the PPP value is better suited for cross-border comparisons of
living standards, specifically relevant to non-traded goods and services, but that is not the
primary goal of the present investigation. Hence, the more general description in terms of
nominal value of GDP (instead of its PPP value) is relevant and sufficient for exposure ana-
lyses needs. The GDP dataset is generally compiled by several different agencies including
the World Bank, the United Nations, and the International Monetary Fund (IMF).
The Economic Statistics Branch of the United Nations Statistics Division (UNSD) main-
tains and annually updates the National Accounts Main Aggregates database. These data are
compiled from the official data reported to UNSD through the annual National Accounts
Questionnaire, supplemented by data estimates for any years and countries with incomplete
or inconsistent information (http://unstats.un.org/unsd/default.htm). In the present investiga-
tion, we used the per capita nominal GDP estimates (which are simply referred as GDP)
compiled by UNSD for all countries and regions since 1970. Note that the per capita
GDP estimates are available at current prices (which are prices of the current reporting period,
also known as nominal GDP), and thus they already account for exchange rate changes and
inflation adjustments for different countries of the world (http://unstats.un.org/unsd/snaama/
Introduction.asp). Historical accounts of GDP estimates were important for correlating
reported economic losses from past earthquakes directly with the associated economic expo-
sure (as measured in terms of the GDP) at the time of these events.
METHODOLOGY
We define the economic loss ratio, r, as the total direct economic loss (which includes
structural, non-structural and content losses) normalized by the total economic exposure as:
Direct Economic Loss

Loss Ratio r ¼ (1)
Total Economic Exposure
Note that the term Damage Ratio (DR) used in earthquake loss estimation studies (e.g.,
ATC 13 1985) is used to characterize earthquake damage to a particular structure. DR is
defined as repair cost divided by replacement (or total) cost. At any given population center,
we may observe single or several types of structures such as wood frame, concrete, steel
frame and masonry buildings. In order to compute shaking-induced economic losses to
these building types, we need to establish the ground motion-damage relationships for
each structure type. Such analyses are feasible only at places where both the building
stock inventory and vulnerability data for each structure type exist. In order to circumvent
this, we attempt to characterize the total direct economic loss (which can be a combined
assessment of losses from different structure types) using a single term r, and more impor-
tantly, not be dependent upon the arduous requirement of building stock inventory and vul-
nerability data.
Similar to DR, the loss ratio r varies with Modified Mercalli (MM) shaking intensity s.
By definition, as the shaking intensity increases, the susceptibility to earthquake damage and
losses increases. Similarly, the loss ratio r (defined here as ratio of total direct economic loss
to total economic exposure) can assume any value between 0 and 1 at a given intensity s.
These criteria, in addition to widespread historic precedence (based on earthquake damage
and loss analysis in the past), we choose a two-parameter lognormal cumulative distribution
function of shaking intensity s to define the loss ratio function:
! " #$
1 s
rðsÞ ¼ ϕ ln (2)
β θ
where ϕ is the standard normal cumulative distribution function. The parameter θ represents
the mean of the natural logarithm of shaking intensity s and the parameter β represents the
standard deviation of lnðsÞ. In the present application, the shaking intensity s ranges from
5.0 to 9.0, where total exposure at MMI IX and above is aggregated and assigned to IX.
Unfortunately, for the last thirty years, per capita GDP estimates are unavailable at a
resolution higher than the country level. Using per capita GDP at the year of an earthquake
and total population subjected to a specific shaking intensity level, one can compute the total
GDP exposed to the shaking intensity level at the time of earthquake. We use country-level
estimates to apportion the total GDP of the country into the region’s GDP under the assump-
tion that per capita GDP estimates are uniform within a country:
Total GDPðregion;intensity ¼sÞ ¼ Per Capita GDPcountry × Populationðregion;intensity ¼sÞ

The total economic exposure at a given intensity can be computed using total GDP
exposed to that intensity multiplied by the country- or region-specific exposure correction
factor α as shown below
E co:Exposureðintensity ¼sÞ ¼ αregion × Total GDPðregion;intensity Þ
Note that the exposure correction factor is simply the ratio of per capita wealth to per
capita GDP estimate of the country. The data on per capita wealth are directly taken from the
World Bank (2006) study. This study considered several important factors including esti-
mates of capital produced, physical capital stock, consumption, investment, natural resources
and other factors in order to compute the per capita wealth in the year 2000 for 119 countries.
The wealth estimate per capita is an important and useful indicator of economic well being
of a region or country. Since the per capita wealth data are only available for the year 2000
(and not for other years in the past), they are assumed constant here. However, if per capita
wealth data become available at higher resolution and at different time period, one could
compute the exposure correction factor αregion specific to each region, and for each year
for hindcasting historical losses. Similarly, it is also plausible to infer per capita economic
exposure using other indicators, for example, household wealth data (Davies et al. 2007).
The total expected economic loss following an earthquake could be estimated by sum-
ming the product of total economic exposure and loss ratio at each shaking intensity level
using the following equation
X
EðLÞ ¼ rðsÞ × Eco:Exposureðintensity ¼sÞ (5)
The loss ratio function rðsÞ depends on the two free parameters, θ and β for each country
or geographic region k. If we suppose that Oi is the recorded economic loss (in million U.S.
Dollars, USD, taken from the Munich Re catalog) for an earthquake i, and there are N such
earthquakes for that country, then we can determine the parameters of the loss ratio function
in such a way that the total error ε between expected loss EðLÞ (also termed as model esti-
mated loss abbreviated as Ei ) and recorded losses Oi is minimized using the following norm
proposed by Jaiswal et al. (2009):
2vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N u N ! " #$
6ut 1X 7 u 1X Ei 2
ε ¼ ln 4 ½E i − Oi % 5 þ t2 ln (6)
N i¼1 N i¼1 Oi
The choice of the above norm is based on our limited analysis. For a given country, the
catalog has many earthquakes with a lower value of economic losses and few earthquakes
that have sizable value of economic losses. The goal is to search for the parameters that best
hindcast the losses in bulk of these historical earthquakes, both at high and low ends of eco-
nomic loss. It is plausible to use another, possibly statistically robust and efficient norm that
could achieve the same purpose discussed above.
A standard iterative search technique available in MATLAB (R2007a) is used to
minimize the objective function shown in Equation 6. The parameters θ and β obtained
using this procedure are country- or region-specific, as described in Jaiswal and Wald (2011).
From the PAGER operational system point of view, since we do not have all the informa-
tion that may be necessary to systematically quantify the total variability associated with
actual economic impacts from any given earthquake, we can only use the model’s predict-
ability to hindcast past earthquake losses as a measure to infer the uncertainty that may be
associated with future losses. In order to depict such uncertainty in a forward sense, we
estimate the standard deviation ζ of the natural logarithm of actual or catalog-recorded
loss, that is, lnðOi Þ given the natural logarithm of model-estimated loss lnðEi Þ as:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N h i2
u 1 X
ζ¼t lnðOi Þ − μðln Oi jln Ei Þ (7)
N − 2 i¼1
The uncertainty measure ζ is assumed to be constant here between high and low losses.
For simplicity, the expected value of lnðOi Þ given lnðE i Þ is assumed to be the same as model-
estimated loss given by Equation 5. The sampling uncertainties and the uncertainties asso-
ciated due to heteroschedasticity or due to the factors that are not fully reflected in historical
earthquakes catalogs are not accounted in the present calculations. However, given the objec-
tive of the PAGER system to provide quick assessment of likely impact, we chose to
disseminate earthquake losses both in terms of color-coded alert levels (selected based
on median loss estimate) and in terms of associated uncertainty about the model-estimated
losses obtained using Equation 7.
EMPIRICAL PARAMETERS USING JAPANESE EARTHQUAKES

In order to combine earthquake shaking, population exposure and historical earthquake loss
data, we use two key datasets, namely the EXPO-CAT catalog (Allen et al. 2009; http://
earthquake.usgs.gov/research/data/pager/expocat.php) and the Munich Reinsurance’s NatCat
Service (NatCatSERVICE 2009) database (which was made available to the PAGER team
through the GEM Foundation, Italy). The EXPO-CAT dataset provides an estimate of popula-
tion exposure at each shaking intensity level for historical earthquakes that occurred between
1973 and 2007 using an Atlas of ShakeMaps (Allen et al. 2008) and the LandScan population
database (http://www.ornl.gov/sci/landscan/). Munich Re’s database covers earthquakes
between 1980 and 2007 and the economic damage data represents total direct economic losses
reported for the year of an earthquake. The term ‘direct economic loss’ here represents the total
direct losses incurred due to structural and nonstructural damage to buildings (of different occu-
pancy types), costs associated with the structural content damage, and potential income and
other losses such as temporary relocation and costs due to loss of business functionality. In
contrast, the ‘indirect economic loss’ refers to any damage or loss that is not caused by factors
that are associated with direct losses, e.g., an earthquake’s impact on the regional economy,
disruptions caused in earthquake zones indirectly affecting the overall business-supply chain
well beyond the earthquake affected area (Brookeshire et al. 1997).
Significant efforts were necessary to post-process these two datasets, especially associa-
tion of earthquakes between the two catalogs (by using the magnitude, location and time of
earthquake occurrence), and removing events for which damage and losses incurred primar-
ily due to non-shaking hazards.
To illustrate the model development for an example country, we used 79 earthquakes in
Japan that caused losses in excess of US$100,000 (since this is the lowest threshold of eco-
nomic loss reported in the catalog) from Munich Re’s catalog and mapped them with popula-
tion exposure information from EXPO-CAT. Using the above methodology we obtain the
parameters θ ¼ 10.29 and β ¼ 0.1, with results shown in Figure 1. The gray line in Figure 1a
Modeling historical earthquake losses for Japan

10 6
Estimated Loss (mil. $)
10 4
10 2
10 0
10 0 10 2 10 4 10 6
Recorded Loss (mil. $)
10 0
10 −2
10 −4
Loss Ratio
10 −6
10 −8
10 −10
10 −12
V VI VII VIII IX
MM Intensity
Figure 1. Diagram (a) showing estimated vs. catalog recorded economic losses for historical
earthquakes in Japan obtained using (b) economic loss ratio as a function of shaking intensity
ðθ ¼ 10.29; β ¼ 0.1Þ.
shows a one-to-one match of recorded loss (from the Munich Re’s catalog) and the estimated
earthquake loss from the model. The two dashed lines represent one order of magnitude in
estimated versus recorded loss. Figure 1b shows the loss ratio function plotted using the two
parameters and equation 2. The high standard deviation ζ ¼ 2.05 as shown in Figure 1a,
indicates a large spread on estimated losses from the model, especially due to earthquakes
of moderate size that were reported to cause little damage. The loss ratio function defined in
terms of two parameters is a cumulative lognormal distribution function of shaking intensity,
and is plotted between intensity V to IX. The estimated economic loss ratio at intensity IX
reaches almost 10 percent indicating significant vulnerability in terms of damage and loss for
economic exposure (with α ¼ 13.40) at such shaking levels in Japan.
GLOBAL COVERAGE AND PAGER IMPLEMENTATION

As with the case for Japan, there are several dozen countries in the world where many
damaging earthquakes have occurred during the past three decades, and where it was possible
to derive a country-specific loss ratio function using historical events alone. However, in
order to derive such functions for other places, we relied on additional data and inferences
that would help us associate countries with similar vulnerability traits. The regionalization
scheme that was originally proposed for PAGER’s global fatality model (Jaiswal et al. 2009)
was a good start because the scheme, besides other factors, relied heavily on building stock
characterization and annual GDP data to associate countries. We considered each region (that
may constitute two or more countries) where many damaging earthquakes have occurred in
the past, to develop a regional loss ratio function. Even when grouping earthquakes from
different countries in order to derive a regional model, the key was to hindcast the economic
losses strictly based on each individual earthquake. This is achieved by considering each
earthquake-specific population exposure, the GDP data, and the exposure correction factor
of the country where earthquake occurred, and searching for best parameters that hindcast the
reported losses in that earthquake. Again the numerical search algorithm allows solving for
all the earthquakes of the region (despite that they occurred in different countries of the same
region) simultaneously via minimizing the Equation 6. In this scheme, an individual country
with sufficient earthquakes can be used to contribute to the regional model but still retains its
own country-based loss ratio parameters. Table 1 provides economic loss model parameters
for selected countries as a quick reference to the readers. Jaiswal and Wald (2011) details the
process implemented in deriving regional loss ratio functions for many countries. In addition
we provide an electronic database of the most recent version of the PAGER loss model
parameters for all the countries of the globe that are implemented within the operational
PAGER system.
Table 1. Empirical model parameters for selected countries. Interested readers are referred to
Jaiswal and Wald (2011) electronic database that provides the most recent version of the model
parameters used by the PAGER system.
Per capita
GDP (2009 Number of
Country θ β ζ estimates) Earthquakes
Albania 9.61 0.10 1.31 $4,174 20
Australia 8.88 0.10 2.15 $48,253 18
Chile 9.73 0.10 1.14 $10,091 17
Italy 9.03 0.10 2.50 $38,640 20
Japan 10.29 0.10 2.05 $38,578 79
Nigeria 8.64 0.10 2.15 $1450 5
Trinidad and Tobago 9.65 0.11 1.73 $18,153 10
Turkey 9.46 0.10 1.74 $10,031 62
United States (without California) 11.51 0.15 1.54 $45,230 12
California 9.60 0.10 2.50 $45,230* 38
*per capita GDP for California is assumed to be the same as for the entire United States.
In the forward calculation within the PAGER system, we use the most recent GDP dataset
in addition to the LandScan1 population data to estimate earthquake losses. In our current
implementation, the USGS ShakeMap system automatically triggers the PAGER system that
ultimately produces earthquake impact alerts using an earthquake impact scale (EIS, Wald et
al. 2011). Soon after an earthquake, there could be large uncertainties associated with many
of the earthquake parameters such as location, depth, focal mechanism, source characteriza-
tion or choice of ground motion prediction or the ground motion-intensity conversion equa-
tions. The PAGER loss estimate can thus vary through time between different versions as
new data arrive to constrain the ShakeMap of a given earthquake. For example, within the
first hour of the M9.0 11 March 2011 Tohoku, Japan earthquake, PAGER estimated a ‘red
alert’ for economic losses using the model described in the earlier section of this article. The
estimate indicated a high likelihood that the shaking-related economic impact could amount
to billions of dollars. However, for the first version of PAGER, that was produced 23 minutes
following this earthquake, we estimated a “yellow” alert for economic losses (indicating tens
of millions of dollars of economic losses) mainly due to lower estimate (M7.9) of magnitude
at that time. The earthquake magnitude was revised to M8.8 within several minutes, which
resulted in the loss estimates increasing to the level of a red alert. Further, as the new data
(ground motion amplitudes, felt intensities, and fault dimensions) were acquired during the
days following this earthquake, newer versions of ShakeMap were produced and thus the
process helped revise the PAGER loss estimates as shown in Figure 2.
Clearly, the tsunami-related losses dominated the overall damage and casualties; none-
theless, the shaking-related impact was substantial. On 24 June 2011, the Cabinet Office of
Government of Japan estimated a total economic loss of US$209.8 billion2 from tsunami and
shaking-related causes, not including nuclear meltdown related issues. The World Bank
study indicated that the total losses could be as much as US$235 billion3. Early reports sug-
gested that the shaking-related causes could have amounted to at least US$77 billion.
Figure 3 shows the earthquake impact alert for the M5.8 Virginia earthquake produced by
the operational PAGER system indicating an orange alert for economic losses (median, i.e.,
50th percentile economic loss value exceeding US$100 million). The PAGER alert indicated
that there was a low likelihood of shaking related fatalities but the economic losses could be
widespread, and such an earthquake could require regional level response based on historical
experience.
Table 2 provides a list of earthquakes with estimated and reported economic losses for
some of the significant earthquakes of 2010–2011. There are large variations in reported
economic losses for many earthquakes in the list. We used widely cited loss estimates
and compared them with the PAGER estimated median economic losses obtained from
the most recent version of the ShakeMap. In general, the loss estimates varied from 25%
to 250% of the reported losses and are clearly within the order of magnitude loss-ranges
originally desired for the PAGER rapid loss estimation system.
1
http://www.ornl.gov/sci/landscan/
2
http://www.reuters.com/article/2011/06/24/japan-economy-estimate-idUSL3E7HN3CM20110624
3
http://www.latimes.com/business/la-fgw-japan-quake-world-bank-20110322,0,3799976.story
M 9.0, NEAR THE EAST COAST OF HONSHU, JAPAN PAGER

Origin Time: Fri 2011-03-11 05:46:24 UTC (14:46:24 local)
Location: 38.30°N 142.37°E Depth: 29 km Version 15
FOR TSUNAMI INFORMATION, SEE: tsunami.noaa.gov Created: 22 weeks, 6 days after earthquake
Estimated Fatalities Red alert for shaking-related fatalities and Estimated Economic Losses
economic losses. High casualties and
extensive damage are probable and the
disaster is likely widespread. Past red alerts
have required a national or international
response.
Estimated economic losses are 0-1% GDP

of Japan.
Estimated Population Exposed to Earthquake Shaking

ESTIMATED POPULATION - -* 13,803k* 21,142k* 8,416k* 9,464k* 34,740k* 5,816k* 257k 0
EXPOSURE (k = x1000)
ESTIMATED MODIFIED
MERCALLI INTENSITY
PERCEIVED SHAKING Not felt Weak Light Moderate Strong Very Strong Severe Violent Extreme
Resistant
Structures none none none V. Light Light Moderate Moderate/Heavy Heavy V. Heavy
POTENTIAL
DAMAGE Vulnerable
Structures none none none Light Moderate Moderate/Heavy Heavy V. Heavy V. Heavy
*Estimated exposure only includes population within the map area.
Population Exposure population per ~1 sq. km from Landscan Structures:

Overall, the population in this region resides
in structures that are resistant to earthquake
shaking, though some vulnerable structures
exist. The predominant vulnerable building
types are non-ductile reinforced concrete
frame and heavy wood frame construction.
Historical Earthquakes (with MMI levels):

Date Dist. Mag. Max Shaking
(UTC) (km) MMI(#) Deaths
1998-06-14 363 5.7 VII(428k) 0
1994-12-28 263 7.7 VII(132k) 3
1983-05-26 369 7.7 VII(174k) 104
Recent earthquakes in this area have caused
secondary hazards such as tsunamis,
landslides, and fires that might have
contributed to losses.
Selected City Exposure

from GeoNames.org
MMI City Population

IX Iwanuma 42k
IX Furukawa 76k
IX Hitachi 186k
IX Takahagi 34k
VIII Karasuyama 18k
VIII Shiogama 60k
VII Yokohama 3,574k
VII Tokyo 8,337k
IV Nagoya 2,191k
IV Kobe 1,528k
III Osaka 2,592k
PAGER content is automatically generated, and only considers losses due to structural damage. bold cities appear on map (k = x1000)
Limitations of input data, shaking estimates, and loss models may add uncertainty.
http://earthquake.usgs.gov/pager Event ID: usc0001xgp
Figure 2. An automated PAGER impact estimate showing “Red Alert” based on both fatalities,
as well as the economic losses using PAGER’s operational models at the time of earthquake. Note
that the likelihood estimates for fatalities (shown in the bar chart above) indicated that actual
fatalities could very well also lie in an “Orange Alert” level thresholds.
M 5.8, VIRGINIA PAGER

Origin Time: Tue 2011-08-23 17:51:04 UTC (13:51:04 local)
Location: 37.94º N 77.93º W Depth: 6 km Version 3
Created: 2 hours, 5 minutes after earthquake
Estimated Fatalities Orange alert level for economic losses.

Significant damage is likely and the disaster is
Estimated Economic Losses
potentially widespread. Estimated economic
losses are less than 1% of GDP of the United
States. Past events with this alert level have
required a regional or national level response.
Green alert level for shaking-related fatalities.

There is a low likelihood of casualties.
Estimated Population Exposed to Earthquake Shaking

ESTIMATED POPULATION - -* 19k* 9,627k* 2,285k 76k 23k 10k 0 0
EXPOSURE (k = x1000)
ESTIMATED MODIFIED
MERCALLI INTENSITY
PERCEIVED SHAKING Not felt Weak Light Moderate Strong Very Strong Severe Violent Extreme
Resistant none none none V. Light Light Moderate Moderate/Heavy Heavy V. Heavy
POTENTIAL Structures
DAMAGE Vulnerable
Structures none none none Light Moderate Moderate/Heavy Heavy V. Heavy V. Heavy
*Estimated exposure only includes population within the map area.
Population Exposure population per ~1 sq. km from Landscan Structures:

Overall, the population in this region resides
in structures that are resistant to earthquake
shaking, though some vulnerable structures
exist.
Historical Earthquakes (with MMI levels):
There were no earthquakes with significant

population exposure to shaking within a 400
km radius of this event.
Selected City Exposure

from GeoNames.org
MMI City Population

VII Louisa 2k
VI Gordonsville 2k
VI Newington 21k
VI Orange 4k
VI Weber City 1k
VI Lake Monticello 10k
V Virginia Beach 425k
V Washington 552k
IV Richmond 191k
IV Baltimore 611k
IV Annapolis 36k
PAGER content is automatically generated, and only considers losses due to structural damage. bold cities appear on map (k = x1000)
Limitations of input data, shaking estimates, and loss models may add uncertainty.
http://earthquake.usgs.gov/pager Event ID: us082311a
Figure 3. An automated PAGER impact estimate showing an “Orange Alert” based on the
operational economic loss model at the time of earthquake. It took several weeks to quantify
the actual economic impact after this earthquake, and the most recent estimate indicates that
the actual loss exceeded $100 million as of 20 September 2011.
Table 2. Loss estimates produced by the PAGER system for recent worldwide earthquakes.
Estimated Direct
Shaking-related Loss Fraction
Reported Losses Losses (median (Reported/
Earthquake (in USD) estimate* in USD) Estimated)
M7.0 January 12, 2010 7.8 billion† 3.0 billion 1.43
Haiti earthquake (4.3 billion direct)
M8.8 September 24, 2010 30 billion‡ 15 billion 1.0 to 2.0
Chile earthquake
M7.0 September 3, 2010 2.2 to 2.9 billion§ 2 billion 1.10 to 1.45
Canterbury, New Zealand
earthquake
M6.1 February 22, 2011 16.5 to 25 billion** 37 billion 0.45 to 0.68
Christchurch, New Zealand
earthquake
M9.0 March 11, 2011 At least 77 billion†† 31 billion 0.76 to 1.52
Tohoku, Japan earthquake (direct losses due
to shaking only)
M6.0 June 13, 2011 4.83 billion‡‡ 4.0 billion 1.21
Christchurch, New Zealand
earthquake
M7.1 October 23, 2011 500 million to 2.0 billion 0.25 to 0.50
Eastern Turkey earthquake 1.0 billion§§
M5.8 August 23, 2011 200 to 300 million*** 236 million 0.85 to 1.27
Virginia, USA earthquake
*Median value of shaking-related direct economic losses from most recent version of ShakeMap system
†
http://www.haitispecialenvoy.org/relief-and-recovery/key-statistics/
‡
http://www.aon.com/attachments/reinsurance/201012_if_annual_global_climate_cat_report.pdf
§
http://www.nzherald.co.nz/business/news/article.cfm?c_id=3&objectid=10709579
**http://www.stuff.co.nz/business/rebuilding-christchurch/4984173/Quake-rebuild-will-eat-into-GDP
††
http://www.itas.fzk.de/tatup/113/khua11a.htm
‡‡
http://en.wikipedia.org/wiki/June_2011_Christchurch_earthquake
§§
http://www.erra.pk/TurkeyEQ2011/FactFile.asp
***http://www.washingtonpost.com/local/region-tallies-earthquake-damage-mostly-uninsured/2011/08/24/
gIQAFdxScJ_story.html
The EIS scheme allows us to not only provide useful and actionable information as
quickly as possible, but also makes the user aware of inherent uncertainties associated
with rapid estimation procedures like PAGER that work on a global scale.
LIMITATIONS
Regional economic exposure estimated at each intensity level using the per capita GDP,
exposure correction factor and the total population exposed at that level has certain draw-
backs. It is plausible that such approach may not be sufficient to represent the true physical
exposure for certain areas where the actual population density and the scale of economic
activities are poorly correlated with each other. Also, the assumption of uniform per capita
GDP in a given year throughout the country, as well as a stationary exposure correction factor
α (due to limited available data) can add to these deficiencies. For example, we do not correct
the economic exposure separately for areas of high industrial activities versus those of low
economic activity. Similarly, we do not account for the high disparity of population or eco-
nomic growth between urban and rural areas (compared to the natural averages that are
already accounted here) in certain countries. In the future, it may be possible to improve
the empirically estimated economic exposure through the inclusion of finer resolution
(at ~1 sq km grid) per capita GDP dataset that were produced within the Natural Hotspot
study (Dilley et al. 2005).
In addition, PAGER loss estimation is also subject to additional uncertainties or inac-
curacies within the input datasets, such as the LandScan population database, shaking hazard
estimates obtained from the ShakeMap system, population exposure estimates obtained from
EXPO-CAT database, and estimates of per capita GDP derived for certain countries. Similar
to the Jaiswal et al. (2009) fatality estimation model, the present approach does have certain
deficiencies, specifically (1) it is country- or region-specific, a resolution too coarse for some
countries, (2) the approach does not quantify the errors associated with reported economic
losses that are available through Munich Re’s historical earthquake catalog, and (3) there are
uncertainties in estimating the exact population count exposed per intensity level. Any sys-
tematic uncertainty due to these factors may be included in the estimation of the uncertainty
parameter ζ obtained using Equation 7; however, any unaccounted-for uncertainties within
the input dataset or inaccuracies associated with the earthquake source parameters can poten-
tially result in significant under or over-estimation of losses.
DISCUSSION AND CONCLUSIONS

The empirical economic loss-modeling approach presented in this paper is comparable to
the earthquake fatality estimation model described by Jaiswal et al. (2009). However, in this
study, the shaking-intensity dependent loss ratio is defined as the total direct economic loss
normalized by the total economic exposure in a given area at the time of the earthquake. The
earthquake specific economic exposure at each intensity level is calculated for each earth-
quake by multiplying the total population at a given shaking intensity by per capita GDP for
the year in which the earthquake occurred and an exposure correction factor of that country.
The loss ratio as a function of shaking intensity is described by using a country-specific,
two-parameter, lognormal cumulative distribution function. The loss ratio function when
multiplied by the economic exposure associated with each shaking-intensity level provides
an estimate of the total expected economic loss.
Earthquake loss estimation is a challenging problem. In the realm of rapid economic loss
estimation within the PAGER system, the purpose is to produce both actionable and accep-
table economic loss estimates that can help its users determine appropriate levels of response
in the initial hours of an earthquake disaster. While detailed inventory-based earthquake loss
estimation models exist, at least domestically (e.g., FEMA’s HAZUS program or commercial
insurance industry models), PAGER’s goal is to serve to the global earthquake disaster needs.
Empirically derived loss estimation models, though enticing, are subject to some drawbacks.
These models can be developed only at places where sufficient historical earthquake damage
and loss data exist, or where sound inferences are possible by virtue of supplementary
information related to seismic vulnerability. Thus, this model is not intended to challenge or
replace detailed inventory-based loss estimation models with simpler empirically derived
ones. Fatality and economic loss based alerts produced in such a way by the PAGER system
help identify the relevant and significant shocks, from the thousands of earthquakes that are
recorded every year by the USGS National Earthquake Information Center.
ACKNOWLEDGMENTS
Munich Reinsurance’s NatCat Service provided the economic loss data for earthquakes
that occurred between 1980 and 2008. Other online data sources include the United Nations
Statistical Service (UNSD), and the World Bank. David Perkins, Margaret Hopper, Nico
Luco, and Jill McCarthy of USGS are thanked for their reviews. The GEMVEM project
funded by the GEM Foundation provided partial support for this research work. Mention
of trade names or commercial products does not constitute endorsement or recommendation
for use by the U.S. Government.
REFERENCES
Allen, T. I., Wald, D. J., Hotovec, A. J., Lin, K., Earle, P. S., and Marano, K. D., 2008. An Atlas of
ShakeMaps for Selected Global Earthquakes, U.S. Geological Survey Open-File Report 2008-
1236, 47 pp.
Allen, T. I., Wald, D. J., Earle, P. S., Marano, K. D., Hotovec, A. J., Lin, K., and Hearne, M.,
2009. An atlas of ShakeMaps and population exposure catalog for earthquake loss modeling,
Bull. Earthq. Eng. 7, DOI: 10.1007/s10518-009-9120-y.
Applied Technology Council (ATC), 1985. Earthquake Damage Evaluation Data for California,
Applied Technology Council, Redwood City, California.
Bausch, D., 2011. Written communication.
Brookeshire, D. S., Chag, S. E., Cochrane, H., Olson, R. A., Rose, A., and Steenson, J., 1997. Direct
and indirect economic losses from earthquake damage, Earthquake Spectra 13, 683–701.
Chan, L. S., Chen, Y., Chen, Q. F., Liu, J., Dong, W., and Shah, H., 1998. Assessment of global
loss based on macroeconomic indicators, Natural Hazards 17, 269–283.
Chen, Y., Chen, Q. F., and Chen, L., 2001. Vulnerability analysis in earthquake loss estimate,
Natural Hazards 23, 349–364.
Davies, J. B., Sandstrom, S., Shorrocks, A., and Wolff, E. N., 2007. The World Distribution
of Household Wealth, UNU-WIDER Research Paper, World Institute for Development
Economic Research (UNU-WIDER).
Dilley, M., Chen, R. S., Deichmann, U., Lerner-Lam, A. L., Arnold, M., Agwe, J., Buys, P.,
Kjekstad, O., Bradfield, L., and Yetman, G., 2005. Natural Disaster Hotspots: A Global
Risk Analysis, Synthesis Report, http://sedac.ciesin.columbia.edu/hazards/hotspots/
synthesisreport.pdf (accessed 3 May 2010).
Dunbar, P., Bilham, R., and Laituri, M., 2002. Earthquake loss estimation for India based on
macroeconomic indicators, in Risk Science and Sustainability: Science for Reduction of
Risk and Sustainable Development of Society (T. Beer, and A. Ismail-Zadeh, Eds.), Kluwer
Academic Publishers, Dordrecht, 163–180.
Federal Emergency Management Agency (FEMA), 2006. HAZUS-MH MR2 Technical Manual.
Washington, D.C., http://www.fema.gov/plan/prevent/hazus/hz_manuals.shtm (accessed 22
August 2008).
International Strategy for Disaster Reduction (ISDR), 2009. Global Assessment Report on Dis-
aster Risk Reduction, United Nations, Geneva, Switzerland.
Jaiswal, K., Wald, D. J., and Hearne, M., 2009. Estimating Casualties for Large Earthquakes
Worldwide Using an Empirical Approach, USGS Open File Report, OF 2009-1136, 83 pp.
(http://pubs.usgs.gov/of/2009/1136/).
Jaiswal, K., and Wald, D. J., 2011. Rapid estimation of the economic consequences of global
earthquakes, USGS Open File Report, OF 2011-1116, pp 47. (http://pubs.usgs.gov/of/
2011/1116/).
Kircher, C., Seligson, H., Bouabid, J., and Morrow, G., 2006. When the Big One Strikes Again-
Estimated Losses due to a repeat of the 1906 San Francisco earthquake. Earthquake Spectra
22(S2), S297–S339.
NatCatSERVICE, 2008. Natural Catastrophes Worldwide 1980–2008, Münchener Rückversi-
cherungs-Gesellschaft, Geo Risks Research (data made available through GEM Foundation,
Pavia Italy).
Seligson, H. A., and Eguchi, R. T., 2005. The true cost of earthquake disasters: an updated tabu-
lation of losses for the 1994 Northridge earthquake, Proceedings, International Symposium on
Earthquake Engineering Commemorating the Tenth Anniversary of the 1995 Kobe Earth-
quake Int. Symp. Earthq. Eng. 2005, Kobe, Japan, 13–16 January, 6 pp.
Wald, D. J., Jaiswal, K. S., Marano, K., and Bausch, D., 2011. An earthquake impact scale,
Natural Hazards Review 12, 125–139.
World Bank, 1994. World Development Report 1994: Infrastructure for Development, Oxford
University Press, Inc., New York, 264 pp.
World Bank, 2006. Where Is the Wealth of Nations?: Measuring Capital for the 21st Century,
World Bank, Washington, D.C., 188 pp.
(Received 7 November 2011; accepted 2 May 2012)
Global Earthquake Fatalities
and Population
Thomas L. Holzera) M.EERI, and James C. Savagea)
Modern global earthquake fatalities can be separated into two components:

(1) fatalities from an approximately constant annual background rate that is inde-
pendent of world population growth and (2) fatalities caused by earthquakes with
large human death tolls, the frequency of which is dependent on world popula-
tion. Earthquakes with death tolls greater than 100,000 (and 50,000) have
increased with world population and obey a nonstationary Poisson distribution
with rate proportional to population. We predict that the number of earthquakes
with death tolls greater than 100,000 (50,000) will increase in the 21st century to
8.7!3.3 (20.5!4.3) from 4 (7) observed in the 20th century if world population
reaches 10.1 billion in 2100. Combining fatalities caused by the background rate
with fatalities caused by catastrophic earthquakes (>100,000 fatalities) indicates
global fatalities in the 21st century will be 2.57!0.64 million if the average post-
1900 death toll for catastrophic earthquakes (193,000) is assumed. [DOI:
10.1193/1.4000106]
INTRODUCTION
From the perspective of global earthquake fatalities, the 21st century began ominously. In
2001, the M7.6 Bhuj, India, earthquake killed 20,085 people. In 2003, the M6.6 Bam, Iran,
earthquake killed 31,000 people. In 2004, the M9.3 Sumatra-Andaman, Southeast Asia,
earthquake killed 228,000 people, mostly by a tsunami that swept across the Indian
Ocean. In 2005, the M7.6 Kashmir, Pakistan, earthquake killed 87,351 people. In 2008,
the M7.9 Wenchuan, China, earthquake killed 88,287 people. And in 2010, the M7.0
Haiti earthquake killed 222,570 people. All told, approximately 699,000 people died in earth-
quakes during the century’s first decade. While large losses of life in earthquakes have been
observed before, the occurrence of so many very deadly earthquakes within a single decade is
unprecedented. These very deadly earthquakes—along with the 2011 Tohoku, Japan, M9.0
earthquake, with its 22,626 death toll and huge financial impact—have prompted speculation
that the planet is experiencing the consequences of the large urban population growth in the
20th century (e.g., Bilham 2010). (Earthquake fatalities for earthquakes reported here are
described at the USGS’s world earthquake deaths Web site, http://earthquake.usgs.gov/
earthquakes/world/world_deaths.php, as of 26 April 2011. Uncertainty in the Haitian esti-
mated death toll is discussed later.)
The growing world population suggests that earthquakes should become increasingly
deadly because the expected number of earthquake deaths is proportional to the exposed
population and their vulnerability to earthquakes. This article explores the relation between
a)
U.S. Geological Survey, MS 977, 345 Middlefield Rd., Menlo Park, CA 94025
155
156 T. L. HOLZER AND J. C. SAVAGE
global earthquake fatalities and world population. In addition to examining the relation
between global fatalities and world population, we postulate that earthquakes with large
death tolls can be represented by a nonstationary Poisson process in which their annual prob-
ability of occurrence is proportional to world population. The purpose of our investigation is
to use historical trends to predict both global fatalities and the number of catastrophic
(>100,000 fatalities) earthquakes in the 21st century.
While a correlation of global fatalities with world population is a reasonable expecta-
tion, it assumes that (1) the population of seismically hazardous areas as a proportion of
world population has remained approximately constant and that (2) the seismic vulner-
ability of buildings has remained approximately constant. We tested the first assumption
by compiling the aggregate population of the 31 countries that occupy the most seismi-
cally hazardous parts of the planet. The population as a percentage of world population
of these 31 countries remained remarkably constant from 1500 to 2010, approximately
62%. The countries were selected from the Global Seismic Hazard Map, a probabilistic
seismic hazard map, which estimates peak ground accelerations (PGA) with a probability
of exceedance of 10% in 50 years (GSHAP 2011). The 31 countries included those
where PGA exceeded 0.2 g in part of the country. The 62% compares favorably to
Bilham’s (2009) estimate that 65% of the world’s large cities today may be subject
to seismic shaking. The second assumption that building vulnerability has not changed
significantly at the global scale is more difficult to evaluate. Some countries, for exam-
ple, the United States and Japan, have decreased their risk by implementing seismic
resistant design while others may have increased theirs. An example of the latter phe-
nomenon was reported by Wyss (2005). He observed a disproportional increase of earth-
quake fatalities with population growth in an investigation of earthquake fatalities in the
front of the Himalaya. He concluded that seismic risk in the region was increasing in part
because buildings that were more vulnerable to lethal collapse during shaking were repla-
cing the traditional more seismically resistant wood-frame and bamboo construction. In a
more global investigation, Wyss and Trendafiloski (2011) used trends in the ratio of
earthquake injuries to fatalities, or the casualty ratio, to infer if newer buildings are
less seismically vulnerable. They concluded that the global casualty ratio in the last cen-
tury had increased as a result of improved building practice, which they interpreted to
indicate that newer buildings were less vulnerable or more seismically resistant. The
decrease of vulnerability, however, was not very statistically robust, and the investigation
was limited to the 20th century. For our investigation, we assume that the seismic vul-
nerability of new buildings did not, and will not during the 21st century, significantly
improve at the global scale.
Only a few investigations have explored the predictive potential of historical trends to
forecast future global earthquake fatalities. By excluding earthquakes with large numbers
of fatalities, Bilham (2009) fit the underlying residual fatality trends from 1500 to 1999
with curves obeying a power law. His residual curves generally underestimated later
fatality rates, which limited their predictive value. In an earlier investigation, Nishenko
and Barton (1996), compared the cumulative number of earthquakes to fatalities per
earthquake. As pointed out by Bilham (2009), their analysis indicated that one earth-
quake with a million fatalities may be expected per century if world population reaches
10 billion. Based on statistical regression, Nichols and Beavers (2008) proposed that
GLOBAL EARTHQUAKE FATALITIES AND POPULATION 157
the mean annual number of fatal earthquakes for the time period from 1870 to 1999 could
be related to world population by simply multiplying world population in billions by a
factor of four.
EARTHQUAKE FATALITY CATALOGS

The completeness and reliability of earthquake fatality catalogs presents a significant
challenge to investigators of historical trends of earthquake fatalities (e.g., Guidoboni
and Ebel 2009, Hough and Bilham 2006). Records of early earthquakes (pre-1900) are
incomplete, and the historical fatality count is systemically underestimated (Bilham
2009). Reasons for the incompleteness of historical catalogs include a lack of systematic
reporting and the loss of records. Even official estimates of earthquake fatalities in very
recent events may be inexact because of imprecise and incomplete record keeping. This
uncertainty for modern events is illustrated by reported death tolls for the 2010 Haiti earth-
quake, which have ranged from 45,000 to 316,000 (Muggah and Kolbe 2011). We use here
the 222,570 fatality estimate that was posted on the USGS Web site (cited in the Introduction
here) when this manuscript was submitted. This uncertainty can cause fatality estimates for
individual earthquakes to vary significantly between catalogs. In addition, seismological
aspects—including the location, source parameters, shaking levels, and even the dates—
of older earthquakes often are poorly documented. Interpretations are dependent on felt
and often incomplete reports of shaking to infer that an earthquake was responsible for fatal-
ities. The poor seismological documentation of some older events is a result of the absence of
widespread and systematic instrumental earthquake monitoring. Such monitoring did not
begin until the end of the 19th century. In fact, the ability to instrumentally locate earthquakes
with magnitudes greater than M6.5 dates back only to 1904 (Lee and Brillinger 1979).
To partially compensate for the inherent uncertainty in available earthquake fatality cat-
alogs, the present investigation made use of multiple catalogs. For our analyses of statistical
trends, we relied on three multicentury catalogs: Utsu (2002), the National Oceanic and
Atmospheric Administration (NOAA) online catalog, and the U.S. Geological Survey
(USGS) catalog of the world’s most destructive earthquakes. We also used two modern
short-term catalogs to highlight some of the characteristics of global earthquake fatalities.
We used these latter catalogs for illustrative purposes simply because we suspect that
they are more complete. Earthquake fatalities in all of these catalogs are typically total fatal-
ities for an earthquake, that is, they include fatalities from secondary effects, such as tsunamis
and landslides.
The two modern short-term catalogs that we used were compiled by the USGS. The
shortest and most recent catalog was compiled by Kristin Marano and colleagues for
their investigation of earthquake fatalities caused by secondary effects, that is, nonshaking
(Marano et al. 2010). It covers the period from 1968 to 2008 and includes earthquakes with
one or more reported fatalities. It is presumably the most complete of all of the catalogs.
A second valuable USGS catalog (USGS 2011a), available at http://earthquake.usgs.gov/
earthquakes/world/world_deaths.php, includes earthquakes since 1900 that have killed
more than 1,000 people. Both catalogs were updated to 2011 with fatalities by year
(USGS 2011b; http://earthquake.usgs.gov/earthquakes/eqarchives/year/).
The multicentury catalogs were compiled by Utsu (2002), the NOAA Geophysical Data
Center, and the USGS. The Utsu (2002) catalog is his DEQ50 in which earthquakes with 50
or more fatalities that occurred between 1500 and 2000 are listed. It is available digitally in
Lee et al. (2002). The NOAA catalog, which is referred to here as the NOAA2010 catalog to
distinguish from earlier printed versions (e.g., Dunbar et al. 1992), is the National Geophy-
sical Data Center’s Significant Earthquake Database (2011), available online at http://www
.ngdc.noaa.gov/nndc/struts/form?t=101650&s=1&d=1. It includes earthquakes from 2150
BCE to 2011 CE. The USGS catalog (USGS 2011c) lists the most destructive earthquakes
from 800 CE to 2011 CE. This catalog includes only earthquakes with death tolls of
more than 50,000 and is maintained at http://earthquake.usgs.gov/earthquakes/world/most_
destructive.php. Earthquakes with more than 50,000 fatalities that are included in at least one
of the three catalogs are compared in Table 1. For completeness, earthquakes that Utsu (2002)
labeled as great are also listed; he did not specify fatalities for these earthquakes, but we
assumed “great” implied approximately 50,000 or more fatalities.
Although we were reluctant to modify these catalogs other than to update them to 2011,
we reviewed individual earthquakes with exceptionally large reported death tolls in the three
multicentury catalogs. Descriptions of older earthquakes often are sparse and subject to alter-
native interpretation. In particular, counts of fatalities may be poorly documented. An exam-
ple of a major reinterpretation is the natural disaster in Calcutta, India, in 1737. Many
earthquake fatality catalogs include (or have included) this earthquake, which is estimated
to have killed 300,000 people. Bilham (1994), however, concluded that it was probably a
cyclone or large weather event. Our review prompted us to delete only one earthquake in the
longest USGS catalog: the 1789 Palu, Turkey, earthquake, which is reported to have killed
51,000 people. Although it was originally described by Ambraseys (1989), it was omitted
from his updated compilation of historical earthquakes in the Mediterranean and Middle East
(Ambraseys 2009).
The Utsu (2002) catalog contains many descriptive entries in the fatalities column.
Fatality estimates for 257 of the 1,025 fatal earthquakes listed in the catalog are described
by descriptive terms. Accordingly, we substituted 50, 300, 3,000, 30,000, and 50,000 fatal-
ities, respectively, for “many” (227 entries); “100s” (9 entries); “1,000s” (13 entries);
“10,000s” (3 entries); and “great” (5 entries) in his catalog. With these substitutions,
descriptive designations account for approximately 0.393 million of the 5.272 million
total fatalities in the catalog. Although observations and conclusions presented here reflect
these substitutions, we repeated the analyses without the substitutions to test if our con-
clusions remained valid, which they did. We will refer to the catalog with these substitu-
tions and the post-2000 earthquakes included as the updated Utsu (2002) catalog. Reference
to the Utsu (2002) catalog without the phrase “updated” indicates that the unmodified
catalog was used.
World population estimates were obtained from two sources. For pre-1900
population estimates, we relied on Biraben (1980). For post-1900 populations and
projections, we relied on compilations by the United Nations Department of
Economics and Social Affairs, Population Division, available at http://www.un.org/
esa/population/ (UNDESA 2011). Information on the population histories of cities
was from Chandler (1987).
Table 1. Comparison of earthquakes with more than 50,000 fatalities in three earthquake fatal-
ity catalogs used in this investigation. Earthquake names shown in bold, normal, and italics type,
respectively, are USGS, NOAA2010, and Utsu (2002) catalog designations.
Time Fatalities
(calendar
year) Earthquake Country USGS NOAA2010 Utsu (2002)
856 Damghan Iran 200,000 200,000 –
893 Ardabil Iran 150,000 180,000 –
1033 Ramala Israel – 70,000 –
1042 Tabriz Iran – 50,000 –
1101 Khorasan Iran – 60,000 –
1138 Aleppo Syria 230,000 230,000 –
1169 Halab Syria – 80,000 –
1202 Southwestern Syria – 1,100,000 –
1268 Asia Minor Silicia 60,000 50,000 –
1290 Chihli China 100,000 100,000 –
1505 Kabul Afghanistan no estimate great
1556 Shaanxi China 830,000 830,000 830,000
1667 Shemacha Azerbaijan 80,000 80,000 80,000
1693 Sicily Italy 60,000 93,000 54,000
1695 Shaanxi China – 20,000 52,600
1703 Jeddo Japan – – great
1718 Gansu China – 73,000 75,000
1727 Tabriz Iran 77,000 77,000 77,000
1739 Ningxia China 50,000 50,000
1755 Lisbon Portugal 70,000 60,000 62,000
1779 Tabriz Iran 200,000 50,000
1783 Calabria Italy 50,000 30,000 35,000
1832 Hindu Kush Afghanistan – no estimate great
1844 Miyaneh Iran – – great
1861 Padang Indonesia – no estimate great
1868 Quayaquil Ecuador – 70,000 40,000
1908 Messina Italy 72,000 82,000 82,000
1920 Ningxia China 200,000 200,000 235,502
1923 Kanto Japan 142,800 99,331 142,807
1935 Quetta Pakistan – 60,000 60,000
1948 Ashgabat Turkmenistan 110,000 19,800 19,800
1970 Chimbote Peru 70,000 66,794 66,794
1976 Tangshan China 255,000 242,000 242,800
1990 Manjil-Rudbar Iran 40,000–50,000 40,000 35,000
2004 Sumatra-Andaman Indian Ocean 227,898 175,827 –
2005 Kashmir Pakistan 86,000 86,000 –
2008 Wenchuan China 87,587 87,652 –
2010 – Haiti 222,570 222,570 –
FATALITY TRENDS
Three aspects of earthquake fatalities trends are relevant to the present investigation.
These aspects are illustrated in Figures 1 to 3. They include: (1) the significant contributions
of earthquakes with large death tolls to global totals, (2) the approximately constant back-
ground fatality rate when these earthquakes are removed, and (3) the increase in the number
of earthquakes with large death tolls in modern time.
The significant contribution from earthquakes with large death tolls is demonstrated for
the 20th century in Figure 1. Cumulative fatalities for two time periods, 1900–2011 and
1968–2011, are plotted. The two time histories are based on the post-1900 USGS and Marano
et al. (2010) catalogs. The salient feature of both time histories is that earthquakes with large
death tolls are superimposed on a smooth background trend. The impact of earthquakes with
large death tolls on the cumulative total is particularly clear in Figure 1. In fact, more than half
of the global earthquake fatalities in the 20th century were caused by only seven earthquakes.
Figure 2, based on the updated Utsu (2002) catalog, illustrates the dominant role of earth-
quakes with large death tolls for a longer time interval, 1500–2011 (see Table 1 for dates of
earthquakes with fatalities greater than 50,000). When all earthquakes in the updated catalog
are considered, the time history of global fatalities is dominated by steps that reflect the
occurrence of earthquakes with large death tolls. Removal of earthquakes with death
tolls greater than 100,000, however, produces an approximately linear background residual
curve, particularly after 1668. By excluding earthquakes with decreasing numbers of fatal-
ities (i.e., 100,000, 50,000, and 10,000), curvature of the resulting residual curves increases
slightly with decreasing thresholds for exclusion (Figure 2a). Of course, the magnitude of the
steps in the curves decreases when earthquakes with successively smaller death tolls are
removed. This results in smoother curves.
Figure 1. Cumulative global earthquake fatalities from 1900 to 2011 for earthquakes with 1000
or more deaths and from 1968–2011 for earthquakes with one or more deaths.
(a)
(b)
Figure 2. Earthquake fatality data from updated Utsu (2002) catalog. (a) Cumulative global
earthquake fatalities, 1500–2011. Different curves show cumulative fatalities for earthquakes
for which only indicated death tolls were included. (b) Cumulative global earthquake fatalities
versus world population, 1500–2011.
The large relative contribution from earthquakes that exceed different death tolls to global
earthquake fatalities is indicated by the separation of the curves in Figure 2a. The comparison
indicates that earthquakes with more than 10,000, 50,000, and 100,000 fatalities, respec-
tively, were responsible for 82%, 52%, and 25% of global fatalities. These estimates of rela-
tive contribution are for the time period 1600–2011 in order to remove the large impact of the
1556 Shaanxi, China, earthquake, which is reported to have killed 830,000 people.
Figure 3. Individual earthquakes from 800 to 2011 with more than 50,000 fatalities from USGS
catalog. The 1990 M7.4 western Iran earthquake in which reported fatalities ranged from 40,000
to 50,000 is included.
Estimated world population as a function of time is also shown in Figure 2a. By combin-
ing the population and fatality data in Figure 2a, we can show cumulative earthquake fatal-
ities versus population after 1600 (Figure 2b). Cumulative global earthquake fatalities
obviously are not a linear function of world population. In fact, the ratio of global earthquake
fatalities to world population has been steadily decreasing. This conclusion applies to the
background fatality trends as well (Figure 2b).
A modern increase in the number of earthquakes with large death tolls is suggested
in Figure 3. It shows the time history from 800–2011 of earthquakes in the world’s most
destructive earthquake (>50,000 fatalities) catalog published by the USGS. This catalog
includes only earthquakes with death tolls greater than 50,000. The time history indi-
cates that the number of these earthquakes began to increase in the late 17th century
relative to earlier centuries. No earthquakes with more than 50,000 fatalities were
reported in the 19th century, but this is a period for which the updated Utsu (2002)
catalog documents only 669,151 earthquake fatalities. Both the number of earthquakes
with death tolls of more than 50,000 and the total earthquake fatalities, 1,564,241,
increased in the 20th century to their highest level. In all, 16 of the 22 earthquakes
in the catalog occurred after 1666. The time history in Figure 3 also suggests that
the frequency of earthquakes with more than 100,000 fatalities has increased, particu-
larly since 1900.
We assume these trends are real and not an artifact of catalog incompleteness. Although
we were reluctant to use pre-1600 portions of catalogs for statistical analysis because of
concern about catalog completeness (Hough and Bilham 2006), we assumed that earthquakes
with extremely large death tolls (greater than 50,000 fatalities) were more likely to be
reported and their occurrence preserved in the historical record than are earthquakes with
smaller death tolls.
PREDICTED CATASTROPHIC EARTHQUAKES

AND GLOBAL FATALITIES IN THE 21ST CENTURY
The basic objective of this investigation was to establish the relation between global
earthquake fatalities and world population and to use it to predict future global earthquake
fatalities. Figures 1 and 2 indicate that global fatalities are not directly proportional to world
population, but in fact are dominated by the intermittent but increasing numbers of earth-
quakes with large death tolls that are superimposed on an approximately constant background
rate that appears largely independent of population growth. These observations prompted us
to estimate fatalities in the 21st century by separately predicting fatalities from a background
rate and fatalities caused by earthquakes with more than 100,000 fatalities, here referred to as
catastrophic earthquakes, and then combining them. We describe our analysis here.
The background annual rate of fatalities was determined from the updated Utsu DEQ50
catalog by excluding earthquakes with more than 100,000 fatalities (Figure 4a). A linear fit
(solid line in Figure 4a) to the cumulative background fatalities after 1825 closely approx-
imates the post-1825 observations. The fatality rate inferred from the linear fit is 8,895!30
fatalities per year (uncertainty quoted is one standard deviation). Orthogonal polynomials of
degrees 2, 3, and 4 furnish very good fits to the entire time period, 1600–2011 (Figure 4a).
The higher-degree polynomials, however, are more sensitive to short-wavelength trends,
which are of less interest here than the lower-degree polynomials that govern the long-
term trend. Extrapolation of the polynomial fits to the year 2100 suggests that the linear
fit to the post-1825 time period furnishes a minimum estimate of background fatality in
the 21st century (Figure 4A).
Fatality estimates for the 21st century based on the background fatality rate range from
0.890!0.003 million fatalities for a linear extrapolation to 0.98 million and 1.21 million
fatalities, respectively, for extrapolations based on second and third degree orthogonal poly-
nomial fits. The fourth degree polynomial yields 1.70 million, but the extrapolation is domi-
nated by the perturbation of the curve during the first decade of the 21st century, and we will
not consider it here. The curvature in the polynomial fits, however, is determined largely by
the 1600–1825 data, which, as explained below, may not be reliable. Thus, we adopt the
linear fit to the post-1825 data for extrapolation of the background fatalities.
We note that both the background fatality rate and the number of fatal earthquakes in the
updated Utsu (2002) catalog changed rather abruptly in about 1825 (Figure 4a, 4b). Figure 4a
shows that the background annual fatality rate increased from 4,150 to 8,895 fatalities per
year, respectively, for the time periods 1600–1825 and 1825–2011. Figure 4B shows that the
increase of fatality rate coincides with an increase in the reported number of fatal earthquakes
(>50 fatalities) in the updated Utsu (2002) catalog. The number of fatal earthquakes per year
increased from 1.3 from 1600–1825 to 3.8 from 1825–2011. Both increases presumably
reflect improved reporting because the annual number of events with less than 10,000 fatal-
ities before 1825 in the updated Utsu (2002) catalog is less than half of the annual number
after 1825 (Figure 4C). This suggests a bias in reporting, in which smaller (<10,000 fatalities)
5
Cumulative fatalities, millions

Updated Utsu DEQ50 Catalog 4
Fatalities<100k only 3
4 21
Polynomial Degree
3
2100
2011
2
1825
0
1600 1700 1800 1900 2000 2100
Time, calendar year
(a)
1200
Cumulative Number of Earthq.
Updated Utsu DEQ50 Catalog

1000 Fatalities <100k only
./a
800
eq
1825
8
600
3.
400
2011
a
200 eq./
1.3
0
1600 1700 1800 1900 2000 2100
Time, calendar year
(b)
1600-1825 1825-2011
1.5
Number earthquakes/a
Updated Utsu
DEQ50 Catalog
1.0
0.5
50 -100 100-1k 1k-10k 10k-100k

Fatality Range
(c)
1600-1825 1825-2011
6000
Updated Utsu
5000 DEQ50 Catalog
Deaths/a
4000
3000
2000
1000
50-100 100-1k 1k-10k 10k-100k

Fatality Range
(d)
Figure 4. Earthquakes with <100,000 fatalities from the updated Utsu DEQ50 Catalog. (a) Cumu-
lative fatalities (solid gray dots) plotted as a function of time. Linear fit to the 1825–2011 data (solid
line) and orthogonal polynomial fits to the 1600–2011 data (dashed lines with degrees 2, 3, and
4 polynomials labeled in figure). (b) Cumulative number of earthquakes (solid gray dots) plotted
as a function of time. Dashed lines are linear fits to the 1600–1825 and 1825–2011 data. (c) Histo-
gram of the average number of earthquakes/a for the 1600–1825 and 1825–2011 intervals as a func-
tion of fatality/event categories. (d) Histogram of the average number of deaths per year for the
1600–1825 and 1825–2011 intervals as a function of fatality/event categories.
events have been underreported in the past. (We note that the cumulative fatality plot
(Figure 4a) is less sensitive to this bias because most of the fatalities in the updated
Utsu (2002) catalog are associated with earthquakes with more than 10,000 fatalities
(Figures 2b and 4d), which presumably are more reliably reported.)
The increase in the frequency of catastrophic earthquakes with that of world population
prompted us to model their occurrence as a nonstationary Poisson process (see the online
Appendix). In a nonstationary Poisson process, the probability that an event will occur in the
time interval ðt; t þ dtÞ is λðtÞdt. If λðtÞ is a constant, the Poisson process is stationary. Here
we assume that the annual probability of the occurrence of a catastrophic earthquake is pro-
portional to population λðtÞ ¼ αNðtÞ where α is a constant and NðtÞ is the world population at
ð t. Then the cumulative number of events expected in the time interval ðS; tÞ is
time
t
α NðτÞdt. Given a sequence of n events at times t 1 ; t2 ; : : : tn within the time interval
s
n
ðS; TÞ, the maximum likelihood estimate of α is ð T (see the online Appendix).
NðτÞdt
S
Figure 5 shows the nonstationary Poisson process fits to the cumulative number of earth-
quakes with more than 100,000 fatalities in the updated Utsu (2002) DEQ50, NOAA2010,
and USGS catalogs (Table 1). The 100,000 death toll was selected as the threshold, rather
than 50,000, because of the ambiguity in the Utsu (2002) catalog with the descriptor “great.”
It is uncertain whether death tolls in “great” earthquakes exceed 50,000. Post-1500 was
selected because it coincides with the beginning of the time period covered by the Utsu
(2002) DEQ50 catalog. We then determined α for the catastrophic earthquakes (fatalities
greater than 100,000) in each of the three catalogs (updated Utsu 2002, NOAA2010, and
USGS) and compared the cumulative number of earthquakes observed with the expected
number predicted by the nonstationary Poisson distribution (Figure 5). Solid curves in
these plots represent the cumulative number of earthquakes expected from the maximum
likelihood fits to the data for a nonstationary Poisson process in which the probability of
an event in the time interval ðt; t þ dtÞ is αNðtÞ. We evaluated the nonstationary fit for
each catalog by comparing the observed number of catastrophic earthquakes in each of
the subintervals 1600–1836.36, 1836.36–1964.55, and 1964.55–2011, intervals chosen so
that the expected number of events in each interval is the same. We used Pearson’s chi-square
test to calculate the percentage P of independent trials from the nonstationary Poisson dis-
tribution that would be expected to give a better agreement with theory (online Appendix).
Any value of P in the interval (5% to 95%) is deemed acceptable. The values of α and P are
given in the legends of each of the plots in Figure 5. The values of α in the three plots do not
differ significantly, and the values of P are all within the acceptable range. However, some
caution is advised in accepting the significance of the fits in Figure 5 because of the small
number of catastrophic earthquakes in the 1500–2011 interval.
The dashed curves in Figure 5 represent the maximum likelihood fit to a stationary Poisson
process in which the probability of an event in the time interval ðt; t þ dtÞ is constant. The
nonstationary process furnished the better fit to the data in each of the three plots in Figure 5.
If world population were to remain constant at 6.12 billion and α ¼ 0.0098!0.0037 per
billion person years based on the USGS catalog, the nonstationary Poisson process implies
7
Cumulative Number
6
Earthquakes with fatalities >100k only
5 α=0.0084±0.0034
4 P=68.3%
3
2
1
0
1500 1600 1700 1800 1900 2000
(a)
7
NOAA2010 Catalog
6
Cumulative Number

5 α=0.0084±0.0034
P=68.3%
4
3
2
1
0
1500 1600 1700 1800 1900 2000
(b)
8
USGS Catalog
Cumulative Number

6
α=0.0098±0.0037
P=71.5%
4
0
1500 1600 1700 1800 1900 2000
Time, calendar year
(c)
Figure 5. Cumulative number of catastrophic earthquakes (>100,000 fatalities) plotted as a func-

tion of time (a) updated Utsu (2002) DEQ50 Catalog, (b) NOAA2010 catalog, and (c) USGS
catalog (see Table 1). The values of α (in units of per billion people year) and P are quoted
in the legend for each plot. The solid and dashed curves, respectively, show the expected values
for the nonstationary Poisson process and the stationary Poisson process representations.
that the world would experience approximately 6.0!2.3 catastrophic earthquakes in the 21st
century (uncertainty quoted is one standard deviation). If we use the United Nations
(UNDESA 2011) medium-fertility projections of world population for the 21st century,
which will grow from 6.12 to 10.1 billion, 8.7!3.3 catastrophic earthquakes are expected
in the century. This forecast is approximately a doubling of the four catastrophic earthquakes
observed in the 20th century when world population grew from 1.61 billion to 6.12 billion.
Total global earthquake fatalities for the 21st century can be estimated by combining
fatalities caused by catastrophic earthquakes with those caused by the annual background
rate. Estimates of global fatalities are compiled in Tables 2 and 3 for earthquakes with
Table 2. Predicted minimum global earthquake fatalities in 21st century, based on United
Nations (UNDESA 2011) medium-fertility population projections
Catastrophic earthquakes
Fatalities Background fatalities Total fatalities

Catalog Number (millions) (millions) (millions)
Earthquakes with fatalities > 100,000 (Fatalities per earthquake = 100,000)
Utsu (2002) updated 7.5!3.0 0.75!0.30 0.89!0.003 1.64!0.30
NOAA2010 7.5!3.0 0.75!0.30 0.89!0.003 1.64!0.30
USGS 8.7!3.3 0.87!0.33 0.89!0.003 1.76!0.33
Earthquakes with fatalities > 50,000 (Fatalities per earthquake = 50,000)

Utsu (2002) updated 31.1!6.2 1.56!0.31 0.68!0.002 2.24!0.31
NOAA2010 27.1!5.1 1.36!0.26 0.68!0.002 2.04!0.26
USGS 20.5!4.3 1.03!0.22 0.68!0.002 1.71!0.22
Table 3. Predicted global earthquake fatalities in 21st century based on United Nations
(UNDESA 2011) medium-fertility population projections and 1900–2010 average death tolls
of catastrophic earthquakes

Earthquakes with fatalities > 100,000 (Average fatalities per earthquake = 193,000)
Utsu (2002) updated 7.5!3.0 1.45!0.58 0.89!0.003 2.34!0.58
NOAA2010 7.5!3.0 1.45!0.58 0.89!0.003 2.34!0.58
USGS 8.7!3.3 1.68!0.64 0.89!0.003 2.57!0.64
Utsu (2002) updated 31.1!6.2 4.31!0.86 0.68!0.002 4.99!0.86
NOAA2010 27.1!5.1 3.75!0.71 0.68!0.002 4.44!0.71
USGS 20.5!4.3 2.84!0.60 0.68!0.002 3.52!0.60
death tolls greater than 100,000 and 50,000 (entries for the latter will be discussed later). The
tables show fatality estimates using the number of catastrophic earthquakes predicted with
each of the three long-term catalogs. Table 2 compiles minimum fatality estimates; Table 3
compiles “average” fatality estimates. The number of catastrophic earthquakes predicted
with each catalog (see column 2) is based on the United Nations medium-
fertility projection of world population in the 21st century, which is predicted to grow
from 6.12 billion to 10.1 billion. Because fatality estimates based on each catalog agree
at one standard deviation, only estimates based on the USGS catalog are discussed here.
The USGS catalog indicates that a minimum of 1.76!0.33 million fatalities will occur in
the 21st century (Table 2). The minimum estimate was derived by assuming a death toll of
Table 4. Predicted global earthquake fatalities in 21st century based on United Nations
(UNDESA 2011) high-fertility population projections and 1900–2010 average death toll of cat-
astrophic earthquakes

Utsu (2002) updated 9.0!3.6 1.74!0.69 0.89!0.003 2.63!0.69
NOAA2010 9.0!3.6 1.74!0.69 0.89!0.003 2.63!0.69
USGS 10.5!3.9 2.03!0.75 0.89!0.003 2.92!0.75
Utsu (2002) updated 37.5!7.5 5.19!1.04 0.68!0.002 5.88!1.04
NOAA2010 32.7!6.1 4.53!0.84 0.68!0.002 5.21!0.84
USGS 24.7!5.2 3.42!0.72 0.68!0.002 4.10!0.72
only 100,000 for each of the predicted 8.7 catastrophic earthquakes. If the average post-1900
death toll for catastrophic earthquakes (193,000) is assumed, estimated global fatalities
increase to 2.57!0.64 million (Table 3). These estimates compare to 1.564 million fatalities
in the 20th century reported in the updated Utsu (2002) catalog.
Table 4 shows predicted global earthquake fatalities if world population reaches the
upper range projected by the United Nations, 15.8 billion, which is based on a high-fertility
rate. Global fatalities increase slightly to 2.92!0.75 million from 2.56!0.64 million. The
relatively small increase is the result of only a modest increase in the number of predicted
catastrophic earthquakes from 8.7!3.3 to 10.5!3.9.
DISCUSSION
This section addresses the reliability of the catalogs and the sensitivity of our analysis to
alternative assumptions. We conclude with discussion and speculation about the causes of the
observed trends.
As discussed previously, the completeness of earthquake catalogs, particularly for older
events, is not known. In our analysis, we used the catalogs in two very different ways: (1) to
determine and then extrapolate background fatality rates and (2) to count the number of cat-
astrophic earthquakes. The latter use requires a less complete catalog than does the former
use. For the latter use, it is required that the reporting of only catastrophic earthquakes is
approximately complete. For the former use, it is required that most of the earthquake fatal-
ities have been reported.
We believe that the inferred change of the background fatality rate in Figure 4a in about
1825 is a consequence of the incompleteness of the updated Utsu (2002) catalog. As shown in
Figure 4c the annual number of earthquakes with fatalities less than 10,000 in the 1600–1825
interval is less than half that in the 1825–2011 interval. The simple explanation for this dis-
crepancy is that the reporting of events with fewer than 10,000 fatalities was less complete in
the earlier interval than in the later interval. The discrepancy between the two intervals is
broadly consistent with changes in the NOAA catalog inferred by Nichols and Beavers
(2008). They considered fatalities per earthquake in both the full and 20th century portion
of the NOAA catalog as of 2000 and reported that the death toll per earthquake had decreased
from 8,026 to 4,477 fatalities. They hypothesized that the decrease was either caused by
under-reporting of smaller events in the older portion of the catalog or an increase in the
number of fatal earthquakes. Based on our analysis, we prefer their former interpretation.
Although we cannot explain the apparent linearity in the cumulative plots of background
earthquake fatalities versus time (Figure 4a), we think it is real and not a result of deficiencies
in the underlying catalogs. There are biases (e.g., under-reporting of fatalities in the earliest
earthquakes) that should affect those plots, but one would not expect such extraneous effects
to produce a linear plot. Linear dependence is a very special case out of all the possible
relationships, and it is unlikely that an extraneous effect would transform a nonlinear relation
into that special case. Note that Bilham (2009, Figure 12) has plotted curves similar to
Figure 2a in which background earthquakes are taken as those with death tolls less than either
5,000 or 30,000, rather than the death tolls used here. In those cases, he concluded that the
cumulative fatalities associated with the background earthquakes was better explained by a
power law fit ða þ bt 10 Þ than a linear fit. We suspect that the modest increase in the back-
ground rate with time is due to the improved reporting with time of earthquakes with smaller
death tolls.
We used linear extrapolation of the background rate for predictions despite the increase of
the fatality rate in the last ten years, 2001–2011. We do not believe a ten-year record is very
reliable for defining new trends. Nevertheless, as noted, a linear fit provides a minimum
estimate. On this basis, we used the constant rate to predict the background fatalities.
Testing the reliability of reports of catastrophic earthquakes in the earthquake fatality
catalogs is challenging because much of the uncertainty derives from pre-1900 earthquakes
for which documentation may be sparse and interpretations subjective. Without discovery of
new historical information for the oldest earthquakes, we are limited to the available infor-
mation. The existing ambiguity is illustrated by a comparison of the USGS, Utsu (2002), and
NOAA2010 earthquake catalogs (Table 1). The criterion for listing an earthquake in Table 1
was that fatalities exceeded 50,000 in at least one of the catalogs. The NOAA catalog
includes ten catastrophic earthquakes from 800–1500 while the USGS catalog lists only
five, which is a significant difference. It is also informative to note that death tolls of indi-
vidual earthquakes even in the modern era can vary significantly between catalogs. The 1933
Quetta, Pakistan, earthquake is assigned 60,000 fatalities in the Utsu (2002) and NOAA cat-
alogs, and only 30,000 fatalities in the USGS post-1900 catalog of earthquakes with 1,000 or
more fatalities. By contrast, the 1948 Ashgabat, Turkmenistan, earthquake is listed as cat-
astrophic in the USGS catalog (110,000 fatalities), but not in either of the other catalogs
(19,800 fatalities).
Because we could not rigorously evaluate the reliability of the reporting of catastrophic
earthquakes in the three multicentury catalogs, we tested the assumption of a nonstationary
Poisson process with each of the catalogs (Figure 5). This is not a completely satisfying test
because the catalogs do not represent entirely independent observations and presumably
inform each other to some degree. Nevertheless, the similarity of the fits is reassuring
and indicates the nonstationary Poisson process predicts the occurrence of catastrophic earth-
quakes far better than a stationary Poisson process (dashed lines in Figure 5).
We also tested how well the nonstationary Poisson process fit earthquakes in the three
catalogs using earthquakes with 50,000 or more fatalities because the definition of the cat-
astrophic earthquake as an event with more than 100,000 fatalities is arbitrary. The fits were
made for longer time periods than in Figure 5 in order to take advantage of the full catalogs.
The time period considered for the USGS catalog was 800–2011, for NOAA2010 catalog
800–2011, and for the updated Utsu catalog 1500–2011. Results are shown in Figure 6.
25
Cumulative number
20 Earthquakes with
fatalities>50k only
15 α=0.0351±0.0070
P=94.1%
10
0
1500 1600 1700 1800 1900 2000
Time, calendar year
(a)
30
NOAA2010 Catalog
Cumulative number
25
Earthquakes with
20 fatalities>50k only
α=0.0306±0.0057
15
P=63.8%
10
5
0
1000 1200 1400 1600 1800 2000
Time, calendar year
(b)
USGS Catalog
20
Earthquakes with
Cumulative number
fatalities>50k only
15
α=0.0231±0.0049
P=36.5%
10
0
1000 1200 1400 1600 1800 2000
Time, calendar year
(c)
Figure 6. Cumulative number of earthquakes with >50,000 fatalities plotted as a function of

time. (a) Updated Utsu (2002) DEQ50 catalog, (b) NOAA2010 catalog, and (c) USGS catalog
(see Table 1). The values of α (in units of per billion people year) and P are quoted in the legend
for each plot. The solid and dashed curves, respectively, show the expected values for the non-
stationary Poisson process and the stationary Poisson process representations.
In evaluating the significance of the fits, the overall time interval was divided into four sub-
intervals such that the expected number of events in each time interval was the same. The
values of α and P for each fit are shown in the legend for each plot in Figure 6. Values of α
found for the three catalogs do not differ significantly. Values of P found for the three
catalogs are all within the acceptable range, although that for the updated Utsu (2002) catalog
is marginal. There are a sufficient number of events in each of the plots in Figure 6 that we are
confident in the Pearson’s chi-square evaluation of the significance of the fits.
For α ¼ 0.0231!0.0049 per billion person years determined from the USGS catalog
(Figure 6c) and the medium-fertility population projections by the United Nations, one pre-
dicts 20.5!4.3 earthquakes with more than 50,000 fatalities in the 21st century. Values of α
determined from the updated Utsu (2002) and NOAA2010 catalogs do not differ significantly
from the α determined from the USGS catalog, but the fits to the cumulative number of
earthquakes were less satisfactory. On the basis of the fit to the USGS catalog of earthquakes
with 50,000 or more fatalities, we estimate that a minimum of 1.71!0.22 million fatalities
will occur in the 21st century based on the medium-fertility United Nations population
projection (Table 2). If a post-1900 average death toll (138,500 fatalities) for earthquakes
with death tolls greater than 50,000 is assumed, estimated global fatalities increase to 3.52!
0.60 million (Table 3). This estimate is not significantly different from estimated global
fatalities of 2.57!0.64 based on the 100,000 fatality criterion that we used to define a
catastrophic earthquake (Table 3).
We are inclined to place more confidence in the completeness of the USGS catalog with
regard to the reporting of earthquakes with 50,000 or more fatalities because it just focuses on
these earthquakes. In particular, Bruce Presgrave (2011, written communication) indicated
that the catalog, although originally based on the NOAA catalog, is periodically reviewed and
updated in light of new information.
Although we will argue that it is an oversimplification, it is tempting to conclude that the
increase in the frequency of catastrophic earthquakes was caused by the growth in size and
number of cities, which began late in the 17th century with the onset of the industrial revolu-
tion and improved municipal hygiene. Such growth, as has been pointed out by Bilham
(2009), increases the likelihood of catastrophic earthquakes by concentrating population.
This places more people at risk if an earthquake occurs nearby, and increases the potential
for massive loss of life. Without such population concentrations, catastrophic earthquakes are
less likely. The increase in the size and number of cities is documented in Figure 7, which
shows the history of the number of cities of different population sizes. It is based on the
compilation by Chandler (1987). City as used by Chandler (1987) includes suburbs lying
outside the municipal area. The history illustrates how large (>1,000,000 people) cities
are a modern phenomenon. The first city in the modern era to reach a million inhabitants
was Beijing, China, which grew to this size in about 1815 (Chandler 1987). By 1900, only 16
cities had populations greater than one million. Today, 476 cities have populations of a mil-
lion or more (Brinkhoff 2010). Even cities with more than 500,000 inhabitants are histori-
cally recent. In 1800, only six cities had populations in excess of half a million.
To test the hypothesis that the growth of cities is the primary factor controlling the increase
of catastrophic earthquakes, we used the urban population of the world instead of the world
population for NðtÞ in the nonstationary Poisson process. This population as a function of time
Figure 7. History of total number of cities with different populations (data from Chandler 1987).
There were 476 cities in the world in 2010 with populations of more than one million.
was estimated from the number

ð of cities exceeding a given population as shown in Figure 7.
t
However, the number α NðτÞdt predicted at any time t by a nonstationary Poisson process
s
with NðtÞ equal to the urban population increased too abruptly in the 20th century to fit the
observed cumulative number of catastrophic earthquakes for any of the three catalogs in
Table 1. This suggests that part of the world population living outside of cities is also signifi-
cantly contributing to the increase of catastrophic earthquakes. Thus, some catastrophic earth-
quakes result from the devastation of multiple, but closely spaced, villages (e.g., the 2005
Kashmir, Pakistan, earthquake), or in the case of tsunamis, multiple coastal inundations.
Such a conclusion could have been anticipated from Table 1 because not all of the catastrophic
earthquakes listed there are associated with a specific city. This is not meant to imply that the
growth of cities is inconsequential. Clearly, the million-fatality earthquake striking a megacity
proposed by Bilham (2009) is more likely as the number of large cities and their size grows.
Finally, we note that for our predictions to be fulfilled, the percentage of the world popu-
lation living in earthquake-resistant buildings must not proportionally increase during the
21st century. Wyss and Trendafiloski (2011) concluded that seismic design and construction
practices have improved in the last 100 years and that these improvements have protected
populations from building collapse in some countries. The challenge that remains is whether
these improved practices will be adopted in other seismically hazardous areas. As we pre-
viously noted, we estimate that approximately 62% of the world population lives in countries
with a significant seismic hazard. Based on the UNDESA (2011) 2100 medium-fertility
population projections for these countries, we estimate this percentage will decline only mod-
estly to 47% in 2100. Thus, without a significant increase in seismic-resistant buildings at a
global scale, the number of catastrophic earthquakes and earthquake fatalities will continue to
increase, and our predictions are likely to be fulfilled.
CONCLUSIONS
Global earthquake fatalities appear to be controlled by two processes: (1) a steady
background fatality rate, which has remained largely constant and independent of popu-
lation growth since 1825; and (2) large catastrophic (>100,000 fatalities) earthquakes,
which have been increasing in number over time. The annual background fatality rate
is 8; 895!30 deaths per year, which indicates that 0.8895 million fatalities will occur
in the 21st century even if no catastrophic earthquakes occur. Modeling of the three
long-term earthquake fatality catalogs indicates catastrophic earthquakes obey a non-
stationary Poisson process in which their rate is proportional to world population. Accord-
ingly, we predict that there will be 8.7!3.3 catastrophic earthquakes in the 21st century
assuming medium-fertility population projections by the United Nations. This estimate
compares to the four observed in the 20th century. This prediction implies that a mini-
mum of 0.87!0.33 million people will die in catastrophic earthquakes. The combina-
tion of these two estimates indicates minimum global earthquake fatalities in the 21st
century will be 1.76!0.33 million. Based on the average post-1900 death toll of cata-
strophic earthquakes (193,000), global earthquake fatalities in 21st century will be 2.57!
0.64 million.
Earthquakes with more than 50,000 fatalities also appear to obey a nonstationary process.
This indicates that 20.5!4.3 earthquakes with death tolls of more than 50,000 will occur in
the 21st century if world population follows medium-fertility projections by the United
Nations. Minimum global fatality estimates are 1.71!0.22 million, which compares favor-
ably at one standard deviation to the minimum estimate of 1.76!0.33 million fatalities based
on catastrophic earthquakes (>100,000 fatalities). If an average post-1900 death toll
(138,500) for earthquakes with more than 50,000 fatalities is assumed, global earthquake
fatalities in the 21st century will be 3.52!0.60 million, which does not differ significantly
from the estimate of 2.57!0.64 million fatalities based on catastrophic earthquakes
(>100,000 fatalities). Thus, estimates of global earthquake fatalities in the 21st century
are statistically independent of the death toll criterion—50,000 and 100,000 fatalities—
used to define a catastrophic earthquake.
ACKNOWLEDGMENTS
The inspiration for this investigation was a talk by Brian Tucker in which he specu-
lated about the potential impact of population growth on earthquake fatalities. Subsequent
discussions with Brian motivated further investigation. We are grateful to Kristin Marano
for providing a digital copy of the Marano et al. (2010) 1968–2008 earthquake fatality
catalog. Thomas E. Noce performed the compilation of population in seismically hazar-
dous areas. Conversations with Willie Lee about the shortcomings of fatality catalogs
provided us with additional insight. We particularly appreciate discussions with collea-
gues, Thomas C. Hanks and Walter D. Mooney, who read and commented honestly on
early drafts, and the official U.S. Geological Survey reviews by James W. Dewey and
Susan E. Hough. We also are grateful for discussions with and an additional review by
Michael W. Hamburger.
REFERENCES
Ambraseys, N., 2009. Earthquakes in the Mediterranean and the Middle East: A multidisciplin-
ary study of seismicity up to 1900, Cambridge University Press, New York, NY, 947 pp.
Ambraseys, N. N., 1989. Temporary seismic quiescence: SE Turkey, Geophysical Journal 96,
311–331.
Bilham, R., 1994. The 1737 Calcutta earthquake and cyclone evaluated, Seismological Society of
America Bulletin 84, 1650–1657.
Bilham, R., 2009. The seismic future of cities, Bulletin of Earthquake Engineering 7, 839–887.
Bilham, R., 2010. Lessons from the Haiti earthquake, Nature 463, 878–879.
Biraben, J.-N., 1980. An essay concerning mankind’s evolution (in French), Population, 34,
13–25.
Brinkhoff, T., 2010. The Principal Agglomerations of the World - Statistics and Charts as Map,
Diagram, and Table, http://www.citypopulation.de/.
Chandler, T., 1987. Four Thousand Years of Urban Growth: A Historical Census, St. David’s
University Press (Edwin Mellen Press), Lewiston, NY, 656 pp.
Dunbar, P. K., Lockridge, P. A., and Whiteside, L. S., 1992. Catalog of Significant Earthquakes
2150 B.C.–1991 A.D., including Quantitative Casualties and Damages, Volume SE-49,
National Geophysical Data Center, World Data Center A for Solid Earth Geophysics, U.
S. Department of Commerce, National Oceanic and Atmospheric Administration, Boulder,
CO, 320 pp.
Global Seismic Hazard Assessment Program (GSHAP), 2011. http://www.seismo.ethz.ch/static/
GSHAP/.
Guidoboni, E., and Ebel, J. E., 2009. Earthquakes and Tsunamis in the Past—A Guide to Tech-
niques in Historical Seismology, Cambridge University Press, Cambridge, 590 pp.
Hough, S.E., and Bilham, R.G., 2006. After the Earthquakes: Elastic Rebound on an Urban
Planet, Oxford University Press, New York, 321 pp.
Lee, W. H. K., and Brillinger, D. R., 1979. On Chinese earthquake history - An attempt to model
an incomplete data set by point process analysis, Pure Applied Geophysics 117, 1229–1257.
Lee, W. H. K., Kanamori, H., Jennings, P. C., and Kisslinger, C. C. (editors), 2002. International
Handbook of Earthquake and Engineering Seismology, vol. 81A, Academic Press, Amster-
dam, 933 pp.
Marano, K., Wald, D. J., and Allen, T. I., 2010. Global earthquake casualties due to secondary
effects: A quantitative analysis for improving rapid loss analyses, Natural Hazards 52,
319–328.
Muggah, R., and Kolbe, A., 2011. Haiti: Why an accurate count of civilian deaths matters, Los
Angeles Times (Home Edition), Op-Ed, 12 July 2011, Main News, A-11.
National Geophysical Data Center, (NGDC), 2011. Significant earthquake database, available at
http://www.ngdc.noaa.gov/nndc/struts/form?t=101650&s=1&d=1
Nichols, J. M., and Beavers, J. E., 2008. World earthquake fatalities from the past: Implications
for the present and future, Natural Hazards Review 9, 179–189.
Nishenko, S. P., and Barton, C. C., 1996. Scaling laws for natural disasters, in Proceedings,
Reduction and Predictability of Natural Disasters, Rundle, J. B., Turcotte, D. L., and
Klein, W. (editors), Santa Fe Institute Studies in the sciences of complexity, Proceeding
vol. 25, Addison-Wesley Publishing Company, Santa Fe, NM, 3–17.
United Nations Department of Economics and Social Affairs (UNDESA), 2011. World Popula-
tion Prospects: The 2011 Revision, Highlights and Advanced Tables, Working Paper No.
ESA/WP.220, Department of Economic and Social Affairs, Population Division, United
Nations, New York, NY, 142 pp.
U.S. Geological Survey (USGS), 2011a. Earthquakes with 1,000 or more deaths since 1900,
available at http://earthquake.usgs.gov/earthquakes/world/world_deaths.php,
U.S. Geological Survey (USGS), 2011b. Earthquake information by year, available at http://
earthquake.usgs.gov/earthquakes/eqarchives/year/.
U.S. Geological Survey (USGS), 2011c. Earthquakes with 50,000 or more deaths, available at
http://earthquake.usgs.gov/earthquakes/world/most_destructive.php.
Utsu, T., 2002. A list of deadly earthquakes in the world: 1500–2000, in International Handbook
of Earthquake and Engineering Seismology, Lee, W. H. K., Kanamori, H., Jennings, P. C., and
Kisslinger, C. C. (editors), vol. 81A, Academic Press, Amsterdam, 691–717.
Wyss, M., 2005. Human losses expected in Himalayan earthquakes, Natural Hazards 34,
305–314.
Wyss, M., and Trendafiloski, G., 2011. Trends in the casualty ratio of injured to fatalties in earth-
quakes, in Human casualties in earthquakess: Progress in modeling and mitigation, Spence,
R., So, E., and Scawthorn, C. (editors), Springer, Dordrecht, 267–274.
(Received 29 April 2011; accepted 15 April 2012)
A High- and Low-Noise Model
for High-Quality Strong-Motion
Accelerometer Stations
Carlo Cauzzia) and John Clintona)
We present reference noise models for high-quality strong-motion acceler-

ometer installations. We use continuous accelerometer data to derive very
broadband (50 Hz–100 s) high- and low-noise models. The proposed noise mod-
els are compared (1) to the broadband seismometer Peterson (1993) noise models;
(2) the datalogger self-noise and background noise levels at existing Swiss and
Southern California strong-motion stations; and (3) typical earthquake signals
recorded in Switzerland and worldwide. The accelerometer low-noise model
(ALNM) is dominated by instrument noise from the sensor and datalogger.
The accelerometer high-noise model (AHNM) reflects (1) at high frequencies
the acceptable site noise in urban areas, (2) at mid-periods the microseismal
peaks and (3) at long periods the maximum noise observed from well-insulated
sensor/datalogger systems placed in vault quality sites. This study also provides
confirmation of the remarkable capability of modern strong-motion acceler-
ometers to record low-amplitude ground motions with seismic observation
quality over a broad frequency range. [DOI: 10.1193/1.4000107]
INTRODUCTION
Noise models for seismic stations provide a crucial guide to the quality of the seismic
installation and the selected site. Models based on microseismic noise have been derived for
many decades (e.g., Brune and Oliver 1959). The current reference model for stations
equipped with very broadband velocity sensors are the Peterson (1993) high- and low-noise
models, developed by the U.S. Geological Survey (USGS) and the Albuquerque Seismolo-
gical Laboratory (ASL). These upper and lower bounds constrain the range of acceptable
noise amplitudes across a broad period range, derived from high-quality instrumentation
and vaults in the Global Seismic Network (GSN). These models continue to be important
guidelines for evaluating and comparing station site characteristics; for defining technical
specifications for instruments; and for predicting the response of sensor systems to quiet
and noisy background conditions (Peterson 1993, Berger et al. 2004).
In spite of the widespread use of noise models in the seismological community, no com-
parable effort to develop reference models for accelerometer station installation exists in the
engineering community. This may be partly explained by the relatively recent deployment of
very high-quality acceleration sensors recording on 24bit dataloggers with continuous data
observation. Prior to this, the typical strong-motion installation traditionally used up to 19bit
a)
Swiss Seismological Service (SED-ETHZ), Sonneggstrasse 5, 8051 Zürich, Switzerland, carlo.cauzzi@sed.ethz
.ch, j.clinton@sed.ethz.ch
85
86 C. CAUZZI AND J. CLINTON
dataloggers operating in triggered mode recording low dynamic range accelerometers

optimized to record the strongest ground motions (e1 g clip level). Indeed, only two decades
ago, the vast majority of strong-motion systems were analogue, with very limited frequency
response and low dynamic range typically not exceeding 60 dB (see e.g., Trifunac and
Todorovska 2001, Douglas 2003).
Consequently, 1) the background noise was rarely measured as only records reaching pre-
selected minimum acceleration levels well above the background noise were recorded and 2)
even if a continuous recording were made, the datalogger/sensor noise would be so high that
background noise would be observed only at very noisy conditions. Current trends to incor-
porate high-quality strong-motion sensors and dataloggers into broadband seismic networks
(e.g., Clinton et al. 2011) now allow high-fidelity recording of small earthquakes at close
distances, at amplitudes that are far below the traditional trigger levels of the order of
1-to-10 mg for strong-motion stations (see Douglas 2003 and Zambonelli et al. 2011 amongst
others). There is now a need to maximize the amplitude range of the signal by ensuring
candidate strong-motion sites have minimum site noise (e.g., COSMOS 2001).
As research over the last decades has emphasised displacement-based (capacity design)
concepts (Priestley et al. 2007) in earthquake engineering, the ability to recover displace-
ments at longer periods, and hence the waveform quality in general, has become a key para-
meter in the selection of accelerometric traces (Douglas 2003). This is important across a
wide range of engineering applications, from nonlinear time-history analyses of structural
response to basic seismic hazard assessment. At the same time, the progressive introduction
of broadband accelerometric sensors recording on high dynamic range dataloggers within
seismic networks has blurred the traditional boundary between weak and strong ground
motion (Boore 2005a). Networks now routinely record digital ground motion acceleration
records of seismic observation quality (i.e., quality of these data is comparable to that
expected from a seismological broadband sensor installation, from the point of view of
low noise across a broad frequency range), and what are traditionally engineering data (accel-
eration waveforms) are now available continuously in real-time, including aseismic noise.
Boore (2005a) described this new reality as the basis for “a new era in engineering seismol-
ogy.” Further, strong-motion sensors are now routinely being placed at both sites of seismic
vault standard (often co-located with broadband sensors to ensure a given site can record the
full spectrum of earthquake motion) as well as soft soil sites in urban areas. An improved
understanding of the optimal possible performance of typical state-of-the-art accelerometer
stations is a basic prerequisite to appropriate installation and evaluation of their potential
usage in seismology and earthquake engineering studies.
This study presents new reference noise models for high-quality strong-motion acceler-
ometer installations. Continuous accelerometer data from (1) the Swiss Seismological
Service at the ETH Zürich (Clinton et al. 2011) and (2) the Southern California Seismic
Network (SCSN 2011) are used to derive very broadband (50 Hz–100 s) high- and low-
noise models. There is particular emphasis on the potential use of the new models for
strong-motion network operators and researchers.
A note on terminology: to be consistent with naming conventions typical in the seismol-
ogy community, in this manuscript the term broadband (BB) refers to high-gain wide-
passband seismological instrumentation, that is, velocity sensors with flat response to
A HIGH- AND LOW-NOISE MODEL FOR HIGH-QUALITY STRONG-MOTION ACCELEROMETER STATIONS 87
velocity over a broad frequency range (typically spanning between 360 s and 120 s to
between 20 Hz and 100 Hz). The term strong-motion (SM) in the context of this paper,
describes modern low-gain accelerometer sensors, that are also broadband and are charac-
terised by a flat response to acceleration from e200 Hz to DC.
ACCELEROMETRIC DATA USED IN THIS STUDY

The data used in this study are collected by the seismic networks in Switzerland and
Southern California. The Swiss Seismological Service (SED) at the ETH Zürich (ETHZ)
is the Swiss federal agency responsible for monitoring the seismicity of Switzerland and
surrounding areas, providing rapid notification to the local authorities and the public, and
archiving and providing reliable data for seismological and earthquake engineering research
studies. The SED manages both a high-gain broadband/short-period seismometer network
and a low-gain strong-motion network. The former is intended to monitor earthquake activity
at magnitudes well below the human perception threshold (Deichmann et al. 2008), whereas
the latter is principally aimed at engineering purposes.
The strong-motion network in Switzerland (SSMNet) has been progressively modernized
and densified since 2000, when the first high-quality stations were installed co-located with
some broadband stations. High-quality, stand-alone strong-motion stations at sites of engineer-
ing interest began to be installed in 2005. The strong-motion network is currently being rapidly
expanded: by 2018, over 130 strong-motion stations should be operational in Switzerland.
Modern strong-motion stations are integrated into the network in the same manner as
broadband stations, that is, the high dynamic range instruments are continuously monitored
in real-time at high sampling rates. Currently, strong-motion sensors are co-located with broad-
band sensors at 12 sites, and similar instrumentation is installed at 30 (this number is rapidly
increasing) free-field, mainly urban locations.
The geographical distribution of the main broadband and accelerometric stations continu-
ously monitored by the SED is presented in Figure 1. Note the densification of strong-motion
stations in the Basel Area, the Valais and the Graubünden, where the Swiss earthquake
hazard is highest (Wiemer et al. 2009). The Swiss strong-motion sites used in this study
are described in Table 1, along with the recording starting date, sampling rate, sensor
and datalogger type at each co-located/stand-alone station.
The strong-motion sensors operated by the SED (see Table 1) are uniformly Kinemetrics
EpiSensors (Kinemetrics 2011), with 2 g clip level, 155 dB dynamic range and flat frequency
response from 200 Hz (above the Nyquist frequency at the operational sampling rates) to DC.
The majority of dataloggers are Nanometrics Taurus (Nanometrics 2011). Data from strong-
motion co-located with broadband are sampled at 120 sps, and at 250 sps at the stand-alone
strong-motion sites. The high sample rate adopted at stand-alone strong-motion stations was
chosen to allow computation of ground motion parameters of engineering interest, such as
acceleration response spectral ordinates, at up to 100 Hz (see also Douglas and Boore 2011).
Commercial ADSL communication is used for the stand-alone stations. The average data
latency across the network is 2 s. The latency is the time taken for the recorded seismic
data to be made available to processing machines at the SED, and so includes data packing
at source, communications delay, and data unpacking at the SED. The typical Swiss strong-
motion housing is a vault (a concrete cylinder with a metallic pot) placed in free-field
6° E 7° E 8° E 9° E 10° E
Streckeisen STS-2
SDSNet and SSMNet
Guralp CMG-3T
Realtime Stations OTTER SLE
Lennartz LE-3D/5s
SBAF OTER1 STEIN
SBAP SRHB TRULL Lennartz LE-3D/1s
SBAT SMZW FLACH
SAUR ACB WEIN Kinemetrics EpiSensor
SBIS2
SKAF SULZ Guralp CMG-5T
SFRA
ZUR WILA Nanometrics Trillium 40s
BOURR
BALST LIENZ
SBUB
SLUW 47° N
47° N PLONS
MUO
BRANT
SCOU LLS SCUC

BNALP DAVOX
STAF
HASLI
TORNY SZER
WIMIS FUORN STSP

GIMEL SCEL
FIESA VDL
LKBD LAUCH BERNI
FUSIO
AIGLE SENIN LKBD2
SMUK SVIO SVIL
SIOV SIOO SVIT SIMPL
SMUR EMBD
GRYON SIOM
SALAN VANNI SGRA
SNIB
EMV MMK 46° N
DIX
46° N
MUGIO
SED
Schweizerischer Erdbebendienst
JUN2011 Swiss Seismological Service
6° E 7° E 8° E 9° E 10° E
0 20 40 80 km
Figure 1. Geographical distribution and sensor type of the Swiss realtime stations of the SSMNet
(Swiss Strong-Motion Network) and SDSNet (Swiss Digital Seismic Network). The networks are
densified in the regions where the seismic hazard is highest. Twelve sites have co-located strong-
motion and broadband sensors, these installations are optimised for broadband seismological
observation. In general, the stand-alone strong-motion sites are in urban areas.
conditions, as described in Clinton et al. 2011. This housing concept draws from experiences
in other broadband and strong-motion communities, including those in Japan (Aoi et al.
2011), the United States (see SCSN 2011), and Italy (Gorini et al. 2010), and was designed
to minimize anthropogenic noise sources, including electrical noise.
The e120 strong-motion channels acquired continuously in real-time at the SED are pro-
cessed and archived identically to the broadband and short-period sensors also monitored by
the seismic network (Clinton et al. 2011). This allows use of the strong-motion data for routine
automatic network operations, manual locations and seamless archival of continuous data and
extraction of data into event files. Metadata maintenance, and health monitoring—both of
waveform completeness and waveform signal quality—are also kept to the same standards
as the rest of the network. This ensures a high-quality in strong-motion network performance,
and nearly 100% recovery of event data.
The SCSN (2011) is a cooperative project of Caltech and the U.S. Geological Survey
(USGS). All waveform data are recorded at Caltech, processed, archived (1932–present) and
distributed to the research and general public by the Southern California Earthquake Data Center
(SCEDC). With more than 350 seismic stations in Southern California, SCSN is one of the
Table 1. Characteristics of the Swiss strong-motion stations used in this study. All dataloggers are Nanometrics.
Site conditions (EC8 ground type

Recording Sample and V S30 where available, otherwise
Station start date rate, sps Lat,° Lon,° general description, SED 2009) Sensor Datalogger
OTTER 08/26/06 250 47.578 7.604 EpiSensor ES-T Trident
SAUR 11/01/05 250 47.534 7.723 alluvium, urban site EpiSensor ES-T Taurus
SBAF 02/03/05 250 47.584 7.592 EpiSensor ES-T Taurus
B, e420 ms-1(1), urban site
SBAP 01/20/05 250 47.572 7.564 EpiSensor ES-T Taurus

SBAT 03/30/06 250 47.559 7.582 EpiSensor ES-T Taurus
SBIS2 12/02/08 250 47.541 7.584 EpiSensor ES-T Taurus

SBUB 09/02/10 250 47.146 9.434 rock, quiet site in the mountains EpiSensor ES-T Taurus
SCEL 08/19/09 250 46.507 9.855 alluvium, urban site EpiSensor ES-T Taurus
SCOU 04/16/08 250 46.854 7.104 urban site EpiSensor ES-T Taurus
SCUC 05/16/05 250 46.798 10.304 alluvium, urban site EpiSensor ES-T Taurus
SFRA 06/20/07 250 47.504 7.709 alluvium, urban site EpiSensor ES-T Taurus
SGRA 12/15/10 250 46.194 7.836 B, rock slope, urban site EpiSensor ES-T Taurus
SIOO 09/14/05 250 46.233 7.383 alluvium, urban site EpiSensor ES-T Taurus
SIOV 04/19/06 250 46.235 7.364 EpiSensor ES-T Taurus
SKAF 03/22/06 250 47.538 7.720 limestone, urban site EpiSensor ES-T Taurus
SLUW 12/15/10 250 47.046 8.318 alluvium, urban site EpiSensor ES-T Taurus
A, e1475 ms-1(1), urban site
SMUK 10/12/05 250 46.286 6.941 EpiSensor ES-T Taurus

SMUR 10/12/05 250 46.277 6.921 alluvial fan, urban site EpiSensor ES-T Taurus
SMZW 09/01/05 250 47.548 7.647 EpiSensor ES-T Taurus
C, e280 ms-1(1), urban site
SNIB 12/16/10 250 46.177 7.802 B, alluvium, urban site EpiSensor ES-T Taurus
SRHB 09/22/05 250 47.571 7.624 EpiSensor ES-T Taurus
STAF 04/16/08 250 46.805 7.216 alluvial fan, urban site EpiSensor ES-T Taurus
A HIGH- AND LOW-NOISE MODEL FOR HIGH-QUALITY STRONG-MOTION ACCELEROMETER STATIONS
STSP 05/29/06 250 46.627 10.333 alluvial/colluvial deposits, urban site EpiSensor ES-T Taurus
SVIL 07/20/10 250 46.292 7.887 alluvium, urban site EpiSensor ES-T Taurus
SVIO 07/21/10 250 46.291 7.880 alluvium, urban site EpiSensor ES-T Taurus
89
(continued )
90
Site conditions (EC8 ground type
Recording Sample and V S30 where available, otherwise
Station start date rate, sps Lat,° Lon,° general description, SED 2009) Sensor Datalogger
SVIT 06/05/10 250 46.290 7.885 alluvial fan, urban site EpiSensor ES-T Taurus
SZER 05/29/06 250 46.697 10.098 alluvium, urban site EpiSensor ES-T Taurus
SIOM 01/05/11 250 46.229 7.362 alluvium, urban site EpiSensor ES-T Taurus
BNALP* 08/12/04 120 46.870 8.425 EpiSensor ES-T þ STS-2 HRD-24
BOURR* 04/26/00 120 47.394 7.230 EpiSensor ES-T þ STS-2 Trident
DIX* 08/04/99 120 46.080 7.408 Rock(2) EpiSensor ES-T STS-2 HRD-24
A, e1655 ms−1(2)(3)
þ
EMV* 08/12/04 120 46.063 6.899 Rock(2) EpiSensor ES-T STS-2 HRD-24
B, e695 ms−1(1)(2)
þ
FUSIO* 12/01/05 120 46.455 8.663 Rock(2) EpiSensor ES-T þ STS-2 Trident
LIENZ* 10/28/04 120 47.295 9.493 Rock(2) EpiSensor ES-T þ STS-2 Trident
LLS* 03/25/04 120 46.847 9.008 EpiSensor ES-T þ STS-2 HRD-24
MMK* 11/11/00 120 46.051 7.964 Rock(2) EpiSensor ES-T þ STS-2 HRD-24
SENIN* 02/13/02 120 46.363 7.299 Rock(2) EpiSensor ES-T STS-2 HRD-24
A, e3,010 ms−1(2)(3)
þ
SULZ* 05/19/04 120 47.527 8.112 EpiSensor ES-T þ STS-2 Trident
VDL* 10/20/03 120 46.483 9.450 Rock(2) EpiSensor ES-T þ STS-2 Trident
ZUR* 04/14/04 120 47.369 8.581 EpiSensor ES-T STS-2 Trident
A, e1,030 ms−1(1)(2)
(1) (2) (3)

*Co-located broadband and strong-motion sensor; from array measurements; rock site for seismological observations; from MASW investigations.
B, e635 ms−1(1)(2)
C. CAUZZI AND J. CLINTON
densest seismic networks in the world. Careful site selection procedures, sub-networking and
continuous recording enable SCSN to regularly detect events as small as M L < 0.4 . At all broad-
band sites a strong-motion sensor is co-located. Similar to the Swiss national networks, the
majority of sensors deployed in the SCSN are broadband STS-2 sensor, and the EpiSensor
strong-motion sensor. The subset of the SCSN used in the present study comprises e190
three-component strong-motion stations, typically equipped with EpiSensor and Quanterra
dataloggers (typically Q330). No borehole data have been used in this study.
In both networks, the continuous archival of accelerometric waveforms along with the
velocity streams means state-of-the-art seismological community data quality processing
tools can be applied to the data (as described in the following section), allowing direct ana-
lysis, comparison and evaluation of the noise recorded at the strong-motion stations.
ACCELERATION DATA PROCESSING

The continuous strong-motion stations are routinely processed via the health monitoring
software tool PQLX (McNamara and Buland 2004; see also McNamara 2011). This software,
a standard tool for the broadband community, calculates power spectral densities (PSD) from
intervals of continuous data and overall station performance from a given period is illustrated
as a probability density function (PDF). The software allows evaluation and temporal track-
ing of background station noise across the frequency spectrum. The software processes con-
tinuous seismic data (typically miniSEED) with associated station metadata (sensor and
datalogger response description—typically datalessSEED; IRIS 2011). For time periods
on the order of a few hours, restitution of ground motion into acceleration units (ms−2)
is performed through full deconvolution of station response. Hence, the PSD computation
is performed using a similar algorithm as used to develop the Albuquerque Seismological
Laboratory (ASL) new low-noise model (NLNM) and new high-noise model (NHNM)
(Peterson 1993, Bendat and Piersol 1971). The PSD estimate is converted into decibels
(dB) with respect to acceleration (ms−2)2Hz−1, for direct comparison to the NLNM. Finally,
seismic noise PDFs from each of the PSDs are computed over full octave averages taken in
1/8 octave intervals. These PDFs result in a plot of the probability of occurrence of a given
power at a particular period with direct comparison to the high- and low-noise reference.
Examples of these plots for all Swiss stations can be found at http://www.seismo.ethz.ch/
research/groups/alrt/pqlx/index (SED 2011). Other agencies also routinely provide this infor-
mation (e.g., Orfeus 2011). The software stores the PDFs and PSDs values in a local rela-
tional database, and statistics such as minimum, mode, median and maximum for each station
component for a given period of time can be extracted for further processing.
For the present study, using the separate PQLX databases maintained at the SED and
Caltech, we extracted the PDFs computed for the all strong-motion stations for the duration
of their operation. For the Swiss stations, the 5-, 50-, and 95-percentile values of the PSD
distribution are presented in Figure 2 and compared with the Peterson (1993) high- and low-
noise model. From Figures 1 and 2, it is clear that high frequency noise levels at urban sites
only rarely exceed the NHNM of Peterson (1993). Modal PSD values close to the NHNM are
typically reached at stations installed in the Basel area (Switzerland’s second-largest conur-
bation with e830,000 inhabitants), and are exceeded at one station, SVIL, placed on loose
alluvium basin sediments in a small and heavily industrialized town (Visp) in the Canton of
OTTER−Trident−08/26/2006 SAUR−Taurus−11/01/2005 SBAF−Taurus−02/03/2005 SBAP−Taurus−01/20/2005 SBAT−Taurus−03/30/2006 SBIS2−Taurus−12/02/2008 SBUB−Taurus−09/02/2010

−50
−100
−150
−200
SCEL−Taurus−08/19/2009 SCOU−Taurus−04/16/2008 SCUC−Taurus−05/16/2005 SFRA−Taurus−06/20/2007 SGRA−Taurus−12/15/2010 SIOO−Taurus−09/14/2005 SIOV−Taurus−04/19/2006

−50
−100
−150
−200
SKAF−Taurus−03/22/2006 SLUW−Taurus−12/15/2010 SMUK−Taurus−10/12/2005 SMUR−Taurus−10/12/2005 SMZW−Taurus−09/01/2005 SNIB−Taurus−12/16/2010 SRHB−Taurus−09/22/2005

−50
20log10(ms−2Hz−1/2), dB
−100
−150
−200
STAF−Taurus−04/16/2008 STSP−Taurus−05/29/2006 SVIL−Taurus−07/20/2010 SVIO−Taurus−07/21/2010 SVIT−Taurus−06/05/2010 SZER−Taurus−05/29/2006 SIOM−Taurus−01/05/2011

−50
−100
−150
−200
BNALP*−HRD−24−08/12/2004 BOURR*−Trident−04/26/2000 DIX*−HRD−24−08/04/2099 EMV*−HRD−24−08/12/2004 FUSIO*−Trident−12/01/2005 LIENZ*−Trident−10/28/2004 LLS*−HRD−24−03/25/2004

−50
−100
−150
−200
0.1 1 10 100 0.1 1 10 100
MMK*−HRD−24−11/11/2000 SENIN*−HRD−24−02/13/2002 SULZ*−Trident−05/19/2004 VDL*−Trident−10/20/2003 ZUR*−Trident−04/14/2004
−50
−100
−150
−200
0.1 1 10 100 0.1 1 10 100 0.1 1 10 100 0.1 1 10 100 0.1 1 10 100
Period T, s
Figure 2. Noise characteristics for Swiss stand-alone and co-located strong-motion stations. For
each station, using PQLX statistics, the modal PSD values (thick curves) are represented along
with the 5- and the 95-percentile of the PSD distribution (thin gray curves). These station-specific
noise characteristics are compared with the NLNM and NHNM of Peterson (1993), shown by
thick dashed curves. In the title of each subplot, the station name, datalogger type and starting date
of acquisition are also given; all stations are operational today. Consistent with the adopted pro-
cessing tool, the units of y-axis in each subplot are 20log10(ms−2Hz−1/2), dB. On the x-axis of each
subplot the period T (in s) is given. Similar to Table 1, an asterisk attached to the station name
means co-located station, indicating high-quality vault located at generally hard sites.
Valais (SW Switzerland). The modal PSD values at co-located stations are, as expected,
significantly lower (up to two to three orders of magnitude for periods T < 1 s) than
those observed at stand-alone strong-motion sites. These co-located installations are indeed
optimized for broadband seismological observation, earthquake detection and location, and
are thus located at hard-rock sites (with measured V S30 values typically exceeding 1000 ms−1
and reaching e3000 ms−1 at LLS), with minimum anthropogenic disturbances. The main aim
of co-locating strong-motion instruments with broadband at these sites is to allow recording
of peak parameter of ground motions in case of strong shaking, which when in excess of
−1
e13 mms , would saturate (clip) the STS-2 broadband sensor.
PSEUDO SELF-NOISE TESTS FOR DATALOGGERS

The individual station noise levels represent the overall noise recorded by the whole
system, including not only the site, but also both the accelerometer and the datalogger.
In order to better understand and clarify the datalogger contribution to the observed
noise levels, we performed terminated noise tests for dataloggers available at our laboratory
in Switzerland following the method of Evans et al. (2010) and Ringler and Hutt (2010). We
were then able to estimate the self-noise spectra for the dataloggers used at the SED (and also
in many other seismic networks worldwide). Terminated noise tests estimate the datalogger
self-noise by shortening its input by a resistor that simulates the output of low-impedance
sensors such as accelerometers and many modern broadband seismometers. The noise levels
computed for a specific acquisition unit can be coupled with the idealized sensor response by
converting the datalogger input in Volts to ground motion units (e.g., ms−2) for a given sensor.
This of course does not take into account possible noise introduced by a real seismic sensor.
Datalogger self-noise tests were carried out at the SED seismological laboratory of Zürich
at Degenried on Nanometrics dataloggers HRD-24, Taurus and Trident and Quanterra
dataloggers Q330, Q330HR (Quanterra 2011). All these acquisition units are 24bit datalog-
gers, except for the 26bit Q330HR (High-Resolution). 24bit systems can typically provide
20log10(223) e140 dB dynamic range, while 26bit data acquisition system can deliver over
150 dB dynamic range.
Following Evans et al. (2010) experiments were carried out at 200 sps with duration over
several days. Datalogger input were shortened by a 100 Ω resistor to simulate the effect of
strong-motion and broadband sensors as operated by SED. Data were processed via PQLX
for sake of consistency with the processing adopted for permanent stations (see previous
Section). Two dataloggers were tested for each model (with the exception of Q330HR)
and the minimum modal PSD value taken as representative of the datalogger performance
at a given period T. Since PQLX computes PSD values on restituted ground motion, we state
these experiments produce pseudo self-noise levels, because the sensitivity of the (missing)
sensor has been taken into consideration in processing the noise waveforms. We simulated
the EpiSensor with full-scale range at 2 g, þ/−20V differential, sensitivity equal to 1.0197
V/(ms−2), and the STS-2 sensor with bandpass sensitivity equal to 1,500 V/(ms−1).
Results are depicted in Figure 3, along with 5-percentile PSD values as obtained from
co-located (green curves) and stand-alone (red curves) Swiss strong-motion stations. The
Peterson (1993) high- and low-noise models are also shown in Figure 3 for comparison.
The results of the experiments with simulated EpiSensor are in the upper part of Figure 3.
The lowest noise levels from the Swiss network are similar to the Taurus datalogger per-
formance over a broad period range, from 50 Hz to 5 s. At longer periods differences
are observed in the performance of the two tested Taurus, explaining the discrepancy between
the pseudo noise and observed field noise data. The performance of the Trident datalogger
appears to be sensitive to the chosen sample rate: in the tests at 200 sps the long period noise
is e10 dB higher than observed under normal operational conditions at co-located sites,
where the sample rate is 120 sps, as seen in the lower bound of the green curves in Figure 3.
Further, the unusual long period high noise levels observed at a subset of co-located stations
(see Figure 2 also) is due to the use of HRD-24 dataloggers.
While a systematic comparison of typical data acquisition unit is beyond the scope of the
present study, these results show that datalogger performance is the key limiting factor for
long-period low noise at seismological vault-quality strong-motion stations. At periods larger
than 10 s, substituting a HRD-24 (or equivalent) with a Taurus (or equivalent) datalogger
results in at least 10 dB increased resolution.
−80
− [1]
−90 − [2]
HNM
−100
−110
20log10(ms−2Hz−1/2), dB
−120
−130
LNM
−140
−150
−160
−170 HRD−24
−180 Q330
Q330HR
−190 Taurus
Trident
−200
0.01 0.1 1 10 100
Period T, s
Figure 3. Results of the datalogger self-noise tests carried out at the SED. In the upper portion,
the datalogger test results, convolved with the EpiSensor response, are compared with the
Peterson (1993) high- (HNM) and low-noise (LNM) model, as well as with data from Swiss
stand-alone [1] (typically urban, soft soils) and co-located (typically low noise, hard rock) [2]
strong-motion stations. The results of the experiments with simulated STS-2 are located in
the lower portion of the picture.
The results of the experiments with simulated STS-2 are located in the bottom part of
Figure 3. Here, the effect of sensor response on the seismographic system (i.e., both sensor
and datalogger) performance is clearly illustrated. The same data acquisition units coupled to
ideal broadband velocity sensors, produce a system noise performance well below the NLNM
for periods T > 1 s. Results from STS-2 coupled with a Q330HR datalogger are consistent
with Ringler and Hutt (2010).
PROPOSED ACCELEROMETRIC NOISE MODELS

Based on the PSD curves presented in the previous Sections, the proposed accelerometric
high- (AHNM) and low- (ALNM) noise models for high-quality accelerometers are pre-
sented in Figure 4. The amplitude values of the proposed AHNM and ALNM are listed
in Table 2. In Figure 4, the AHNM and ALNM are also compared with Swiss and Southern
California 5-percentile PSD obtained through PQLX from continuous acceleration record-
ings. Since it is not standard practice to couple an EpiSensor with a Q330HR datalogger, we
define the ALNM by the minimum measured resolution observed at Swiss and Southern
California (SCSN) stations, that is, the lower boundary of 5-percentile PSD amplitudes
observed at a large number of hard-rock strong-motion sites. The proposed ALNM is
consistent with noise performance of typical high-quality digitizers. If compared with the
resolution of the strong-motion equipment presently adopted in Japan by K-Net (2011)
and KiK-Net (2011) networks, the proposed ALNM is at least 20 dB below the K-Net02
instrument noise (Aoi et al. 2011) for short periods, T < 10 s.
−80
− [1] ALNM
− [2] AHNM
− [3] HNM
−90
20log10(ms−2Hz−1/2), dB
−100
−110
−120
−130
−140
0.1 1 10 50 100
Period T, s
Figure 4. Proposed accelerometric high- (AHNM) and low- (ALNM) noise models for high-qual-
ity accelerometers. The models are compared with 5-percentile PSD computed through PQLX for
Swiss continuous stand-alone [1] and co-located [2] accelerometric stations, Southern California
continuous strong-motion stations [3] and the Peterson (1993) high-noise model (HNM).
Table 2. Amplitude values of the proposed AHNM and ALNM. Intermediate values can be
computed using linear interpolation over logarithmic period axis (see Figure 4).
Period T, s AHNM, dB ALNM, dB

0.01 −91.50 −135.00
0.10 −91.50 (−)
0.22 −97.41 (−)
0.32 −110.50 (−)
0.80 −120.00 (−)
1.00 (−) −135.00
3.80 −98.00 (−)
4.60 −96.50 (−)
6.30 −101.00 (−)
7.10 −105.00 (−)
10.00 (−) −130.00
150.00 −91.25 −118.25
(–) explicit definition not needed.
The AHNM is dominated by electromagnetic noise for very short periods T < 0.1 s. In
this period range, the most important source of noise is the 50 Hz electromechanical vibration
associated with strong-motion installations within transformer houses. The proposed AHNM
can be conservative in this frequency range since, as recently demonstrated by Cauzzi and
Clinton (2011) within the framework of the ongoing upgrade project of Swiss strong-motion
network (see also SED 2010), relocating stations e12 m away from transform houses is in
most cases sufficient to reduce the amplitude of the 50 Hz peak by at least 2 orders of
magnitude. At frequencies higher than the microseism peaks (0.05 s–3 s) the AHNM reflects
the acceptable site noise in urban/industrialized areas and the contribution to noise amplitude
amplification due to site effects. In this period range, the apparent high noise levels at many
Southern California stations (gray curves in Figure 4) reflect typical cultural and site noise
conditions in the sedimentary Los Angeles basins (e 40 stations in the present dataset). At
mid-periods the peak microseismal energy that can be expected at strong-motion stations
installed near coastlines dominates the AHNM. The Peterson (1993) high-noise
model remains appropriate over this broad period range. At long and very long periods
(7 s–100 s) the proposed AHNM is constrained by the noise of the sensor/datalogger system
for a well-insulated station. The proposed AHNM and ALNM models span 2 orders of
magnitude over a broad frequency range.
It is worth noting here that the proposed noise models for typical state-of-the-art accel-
erometer stations are conceptually different from the Earth’s ambient noise models. The
Peterson (1993) NLNM does not depend on the measuring instrument, while the proposed
ALNM is the result of a particular combination of 140þ dB accelerometric sensors with a
given gain and response (e.g., EpiSensor 2 g) with 24bit dataloggers (e.g., Taurus). However,
the comparison in Figure 4 with a large set of data with different geographical origin and
vault characteristics confirms the validity of the proposed ALNM, further strengthened by the
practical observation that the class of sensors and dataloggers used in this study is also widely
adopted in many modern strong-motion networks. Note that the lower bound of SCSN PSD
curves is characterized by EpiSensors with 2g∕4 g clip level and the latest generation of
Q330-S dataloggers, that is, a digitizer not included in the pseudo noise tests described
in the previous Section. The new AHNM is primarily intended as a minimum reference
level for high-quality strong-motion installations, which are capable of recording acceleration
waveforms over a large magnitude and frequency domain. Achieving this noise level thus
justifies the considerable expense of the high-quality equipment.
DISCUSSION AND CONCLUSIONS

In Figure 5, the proposed acceleration noise models are compared in terms of period and
amplitude with average earthquake amplitudes representing events typically of interest to
both engineers and seismologists. Following Clinton and Heaton (2002), signals are repre-
sented as octave wide bandpassed acceleration, in ms-2. This direct measure of acceleration is
chosen rather than the standard power spectral density to allow direct comparison of the
proposed noise models with expected ground motion amplitudes from transient earthquake
signals, which cannot be uniquely represented by power spectra as the duration must be
arbitrarily selected (Aki and Richards 1980, Clinton and Heaton 2002). The EpiSensor
2g clip level and the theoretical resolution (least significant bit) of 12bit, 16bit, 19bit and
24bit instrumentation are also shown in Figure 5 as gray dashed lines (see also Trifunac
and Todorovska 2001). Note that instrument limits in Figure 5 are scaled down (typically
50%) to account for the bandpassing of the event data (Clinton and Heaton 2002). Hence, the
upper bound of the y-axis in Figure 5 corresponds to a 2 g clip level of the accelerometer.
In addition to the near field strong-motion data presented in the original Clinton and Heaton
(2002) dataset, we augment the teleseismic dataset by including data recorded in Switzerland
at 10,000 and 12,000 km distance during the Chile, 2010, M8.8 and the Tohoku, 2011,
EpiSensor 2g full scale

9.81
M7.5max
Octave wide bandpassed acceleration, ms−2
1
M5.5
Near source ~ 10 km
0.1 M3.5
M7.5 Regional ~ 100 km
Teleseismic ~ 3000 km
0.01 M5.5
M1.5 M3.5
0.001 12bit instrument resolution
Japan, 2011
AHNM
0.0001
16bit M8
1e−05 M1.5 Chile, 2010
19bit
M6 M7
1e−06 ALNM
24bit instrument resolution
1e−07
0.03 0.1 1 10 50 100
Period T, s
Figure 5. Comparison between the proposed acceleration noise models and earthquake data as
processed following Clinton and Heaton (2002). Black curves are computed from local earth-
quakes with distance ~10 km. Red curves represent regional events recorded at ~100 km distance.
Green curves are teleseismic events recorded at more than 3,000 km distance. The M7.5max
curve in the figure was derived using some of the largest near source data from Chi-Chi (Taiwan)
and Kocaeli (Turkey) earthquakes. The black lines with symbols represent the maximum
observed PSD amplitudes at Swiss strong- motion station SULZ during the Chile 2010 and
Tohoku (Japan) 2011 earthquakes, at about 10,000 km.
M9.0 earthquakes. The spectral content of the recorded waveforms clearly exceeds the
ALNM for T > 1 s and the AHNM for T > 10 s. The amplitude levels of the proposed
noise models are converted from dB into acceleration as follows:
dB −1
ms−2 ¼ 10 20 T 2 α;
where T is the period in s and α is a scaling factor that accounts for the use of average-to-peak
amplitudes instead of peak-to-peak amplitudes (Clinton 2004, Clinton and Heaton 2002).
When high-quality strong-motion sensors are installed at site of seismological quality, as
in many locations in Switzerland, stations are capable of operating near the ALNM at high
frequencies. While modern accelerometers can record very small earthquakes (M1.5 at
regional e100 km distance) at quiet sites, cultural noise dominates the signal at high frequen-
cies in noisy urban/industrialized environments, with amplitudes e2 orders of magnitude
larger than the sensor/datalogger noise. This suggests that strong-motion data from high-
quality strong-motion station can be used in noise tomography investigations. Further, a
strong-motion station operating near the proposed AHNM can reliably record acceleration
waveforms from local earthquakes at e10 km distance with M1.5. Regional events are also
recorded down to low magnitudes (eM3) over a broad period range. For ground motions of
engineering significance, for example, for M > 4, station noise typically dominates wave-
forms only at periods larger than e10 s. Also apparent from Figure 5 is the limited capability
of low-resolution (especially 12bit and 16bit) instruments to record long period ground
motions from local, regional and teleseismic events, as well as short period ground motions
from local low-energy events. Further, comparing the proposed ALNM with the resolution of
a 24bit instrument suggests that the theoretical expected performance of the accelerographic
system can be significantly overestimated at periods < 1 s.
The comparison of the proposed noise models with earthquake data in Figure 5 contri-
butes to the ongoing discussion within the engineering seismology community assessing
optimal correction techniques (including filtering) and usable period ranges for computing
ground motion parameters of engineering interest from strong-motion data. Following
Boore (2001), Boore (2005a and b) and Boore and Bommer (2005), Paolucci et al.
(2007) used a global digital strong-motion accelerometric database (albeit dominated by
Japanese data), with magnitude spanning M W 5–7.2 and hypocentral distances
R< 150 km, to show that simple removal of a baseline offset computed on pre-event
time window of the acceleration trace is often sufficient processing if the aim is to calculate
reliable elastic displacement response spectra up to at least 10 s. In other words, when using
high-quality strong-motion data, station noise proves to be significant in data only at periods
> 10 s. These findings, implemented by Cauzzi and Faccioli (2008) in their development of a
global broadband (0–20 s) ground motion prediction equation (GMPE), have been subse-
quently confirmed in similar studies by Akkar and Boore (2009).
Even though sensor and datalogger noise and vault conditions are not the only contri-
butors to long-period drift in strong-motion waveforms—co-seismic tectonic and local tilting
(Clinton 2004, Grazier 2010) are also important—the newly developed noise models provide
a reliable reference framework for waveform selection for modern accelerometric data for
earthquake engineering and engineering seismology.
At the SED, the introduction of continuous monitoring and the automated quality pro-
cessing via PQLX (McNamara et al. 2009) led to significant improvements in the quality of
data collected. As an example, improved insulation of the sensor provides dramatic improve-
ments in the long period performance of the sensors, an improvement that was subsequently
applied to all stations. In terms of high frequency performance, some sensors were found
to record significant high frequency noise from electrical sources that prompted station relo-
cation. The good performance of the Swiss stations is a direct consequence of the migration
to constant observation and easy comparison of the quality of each site.
A remarkable example of the performance of a strong-motion station is presented in
Figure 6, which shows waveforms from the Tohoku, Japan, M9 earthquake of March
acc., cms−2
0.05 acceleration recorded by SM sensor

0
−0.05
0.5 displacement derived from SM recording

disp., cm
0
−0.5
vel., cms−1
0.1 velocity recorded by BB sensor

0
−0.1
0.5
disp., cm
displacement derived from BB recording

0
−0.5
0 1000 2000 3000 4000 5000 6000 7000 8000

Time, s
Figure 6. Acceleration, velocity and 120 s high-pass filtered displacement waveforms from the
Tohoku (Japan) 2011 M9 earthquake recorded at 10,000 km at Swiss station ZUR. Data proces-
sing is described in the text. Label HGN denotes North-South channel of the accelerometer sen-
sor, while HHN is the North-South channel of the broadband velocity sensor.
2011 recorded at e10,000 km at Zürich (station ZUR), where an accelerometer is co-located

with a broadband STS-2 (data are open to the public and available at http://arclink.ethz.ch, last
accessed 18 November 2011). Figure 6 displays the unfiltered acceleration waveform as
recorded by the accelerometer at ZUR. The only processing is removal of the bandpass
gain and a linear trend. Also shown are the displacement traces obtained (1) by double
integration of the acceleration signal (after high-pass filtering with corner frequency
f c ¼ 1∕120 Hz to simulate the response of the STS-2) and (2) by integration of the broadband
velocity waveform (with bandpass gain removed, but without deconvolution). The similarity
of the displacement waveforms demonstrates the quality of the strong-motion record, that is,
the often overlooked capability of an accelerometric station to reliably record relatively weak
motions at very long periods. Also shown in the figure is the velocity waveform as recorded
by the broadband instrument, note that peak ground velocity values are roughly within 10% of
the full scale of the STS-2 seismometer at 13 mms−1, though observed at periods near 100 s.
For network operators installing new strong-motion stations, the proposed noise models
can serve as useful guidelines, though the noise at a site is often only one of many important
considerations. At the SED, these preliminary models now provide the acceptable limits for
candidate sites within the strong-motion network modernisation project (Clinton et al. 2011).
Nevertheless, as the key goal of the network upgrade and densification is to improve the
understanding of the seismic risk in Switzerland, certain urban towns must be instrumented
(e.g., Visp) which are located at sites with basin amplifications so large that any station in the
vicinity will have noise above the AHNM, and we are forced to accept that at some sites the
AHNM will be exceeded at some frequencies.
These noise models, in particular the ALNM, are functions of the sensor/datalogger and
vault system. We have shown that for the currently proposed ALNM model based on the
Swiss and Southern California dataset, the datalogger performance is the limiting factor in
determining the achievable noise floor. It is important to note that in future, as we migrate to
in some cases better and possibly different instrumentation (for example datalogger with
higher dynamic range or strong-motion sensors recording velocity rather than acceleration),
the ALNM would need to be revised. We note that some new strong motion networks prior-
itize station density over sensor quality, often resulting in deployment of cheaper instrumen-
tation (typically MEMS-based) with reduced frequency and dynamic range operating in a
triggered mode. In these networks, station quality tends to be a minor consideration and
the AHNM is not expected to be reached.
ACKNOWLEDGEMENTS
This work has been partially developed within the framework of the project of renewal
and modernization of the Swiss Strong-Motion Network (SSMNet 2011), approved by the
Swiss Federal Council in February 2009, monitored and supervised by a steering committee
headed by the Swiss Federal Office for the Environment (FOEN). The installation of a subset
of Swiss strong-motion stations used in this study was funded by the COGEAR project
(CCES 2011). We thank Egill Hauksson and Rae Yip (Caltech) for providing PQLX statistics
for SCSN (Southern California Seismic Network). The Electronics Laboratory at the SED
provided valuable technical support, both by achieving the high-quality network data for
Switzerland, and in performing the datalogger tests described in the paper. The Engineering
Seismology group at the SED maintains the site characterization database, which is the
source of the geophysical station information. We thank Sabine Wöhlbier for providing
the Seismic Network maps. Marco Olivieri contributed with many helpful discussions.
Joe Steim also provided valuable insight. We would like to thank John Douglas, Marco
Massa and an anonymous reviewer for critically revising the manuscript and providing con-
structive comments and suggestions.
REFERENCES
Aki, K., and Richards, P. G., 1980. Quantitative Seismology: Theory and Methods, W.H. Free-
man and Co., San Francisco, 704 pp.
Akkar, S., and Boore, D. M., 2009. On baseline corrections and uncertainty in response spectra
for baseline variations commonly encountered in digital accelerograph records, Bull. Seism.
Soc. Am. 99, 1671–1690.
Aoi, S., Kunugi, T., Nakamura, H., and Fujiwara, H., 2011. Deployment of New Strong-motion
Seismographs of K-NET and KiK-net, in Earthquake Data in Engineering Seismology.
Geotechnical, Geological, and Earthquake Engineering 14, 167–186.
Bendat, J. S., and Piersol, A. G., 1971. Random Data: Analysis and Measurement Procedures,
John Wiley and Sons, New York, 407 pp.
Berger, J., Davis, P., and Ekstrom, G., 2004. Ambient earth noise: A survey of the Global Seismographic
Network, Journal of Geophysical Research 109, B11–B11307, DOI: 10.1029/2004JB003408.
Boore, D. M., 2001. Effect of baseline corrections on displacements and response spectra for several
recordings of the 1999 Chi-Chi, Taiwan, earthquake, Bull. Seism. Soc. Am. 91, 1199–1211.
Boore, D. M., 2005a. Long-period ground motions from digital acceleration recordings: a new
era in engineering seismology, in Directions in Strong-Motion Instrumentation (P. Gülkan,
and J. G. Anderson, Editors), Springer, Dordrecht, The Netherlands, 322 pp.
Boore, D. M., 2005b. On pads and filters: processing strong-motion data, Bulletin of the
Boore, D. M., and Bommer, J. J., 2005. Processing of strong-motion accelerograms: needs,
options and consequences, Soil Dynamics and Earthquake Engineering 25, 93–115.
Brune, J. N., and Oliver, J., 1959. The seismic noise of the earth’s surface, Bulletin of the
Cauzzi, C., and Faccioli, E., 2008. Broadband (0.05 to 20 s) prediction of displacement response
spectra based on worldwide digital records, Journal of Seismology 12, 453–475.
Cauzzi, C., and Clinton, J., 2011. Improving waveform quality and data accessibility for the
Swiss Strong-motion Network: challenges and issues, Geophysical Research Abstracts 13,
EGU2011-6576.
Clinton, J., Cauzzi, C., Fäh, D., Michel, C., Zweifel, P., Olivieri, M., Cua, G., Haslinger, F., and
Giardini, D., 2011. The Current State of Strong-motion Monitoring in Switzerland,
Earthquake Data in Engineering Seismology, Geotechnical, Geological, and Earthquake
Clinton, J. F., 2004. Modern Digital Seismology - Instrumentation, and Small Amplitude Studies
for the Engineering World, Ph.D. Thesis, California Institute of Technology.
Clinton, J. F., and Heaton, T. H., 2002. The potential advantages of a strong-motion velocity
meter over a strong-motion accelerometer, Seismological Research Letters 73, 332–342.
Competence Center Environment and Sustainability (CCES), 2011. COGEAR: Coupled Seismo-
genic Geohazards in Alpine Regions, http://www.cces.ethz.ch/projects/hazri/COGEAR, last
accessed 25 January 2013.
Consortium of Organizations for Strong-Motion Observation Systems (COSMOS), 2001.
Guidelines for Installation of Advanced National Seismic System Strong-Motion Reference
Stations, available at http://www.cosmos-eq.org/publications/Guidelines_ANSS.pdf (last
accessed 25 January 2013).
Deichmann, N., Baer, M., Clinton, J., Husen, S., Fäh, D., Giardini, D., Kaestli, P., Kradolfer, U.,
and Wiemer, S., 2008. Earthquakes in Switzerland and surrounding regions during 2007,
Swiss Journal of Geosciences 101, 659–667.
Douglas, J., 2003. What is a poor quality strong-motion record?, Bulletin of Earthquake
Douglas, J., and Boore, D. M., 2011. High-frequency filtering of strong-motion records, Bulletin
of Earthquake Engineering 9, 395–409.
Evans, J. R., Followill, F., Hutt, C. R., Kromer, R. P., Nigbor, R. L., Ringler, A. T., Steim, J. M.,
and Wielandt, E., 2010. Method for calculating self-noise spectra and operating ranges
for seismographic inertial sensors and recorders, Seismological Research Letters 81,
640–646.
Gorini, A., Nicoletti, M., Marsan, P., Bianconi, R., de Nardis, R., Filippi, L., Marcucci, S., Palma,
F., and Zambonelli, E., 2010. The Italian strong motion network, Bulletin of Earthquake
Graizer, V., 2010. Strong motion recordings and residual displacements: What are we actually
recording in strong motion seismology?, Seismological Research Letters 81, 635–639.
Incorporated Research Institutions for Seismology, 2011. SEED Reference Manual, available at
http://www.fdsn.org/seed_manual/SEEDManual_V2.4.pdf, last accessed 25 January 2013.
Kiban-Kyoshin Network (KiK-Net), 2011. KiK-Net, http://www.kik.bosai.go.jp, last accessed
25 January 2013.
Kinemetrics, 2011. Kinemetrics EpiSensors, http://www.kinemetrics.com, last accessed

18 November 2011.
Kyoshin Network (K-Net), 2011. K-Net, http://www.k-net.bosai.go.jp, last accessed
25 January 2013.
McNamara, D. E., 2011. PQLX: A Software Tool to Evaluate Seismic Station Performance, avail-
able at http://geohazards.cr.usgs.gov/staffweb/mcnamara/Software/PQLX.html, last accessed
25 January 2013.
McNamara, D. E., and Buland, R. P., 2004. Ambient noise levels in the continental United States,
Bulletin of the Seismological Society of America 94, 1517–1527.
McNamara, D. E., Hutt, C. R., Gee, L. S., Benz, H. M., and Buland, R. P., 2009. A method to
establish seismic noise baselines for automated station assessment, Seismological Research
Letters 80, 628–637.
Nanometrics, 2011. Nanometrics Taurus, http://www.nanometrics.ca, last accessed 25 January 2013.
Observatories and Research Facilities for European Seismology (Orfeus), 2011. ORFEUS PQLX
Interface VEBSN, available at http://www.orfeus-eu.org/cgi-bin/pqlx.cgi, last accessed
25 January 2013.
Paolucci, R., Rovelli, A., Faccioli, E., Cauzzi, C., Finazzi, E., Vanini, M., Di Alessandro, C., and
Calderoni, G., 2008. On the reliability of long period spectral ordinates from digital acceler-
ograms, Earthquake Engineering and Structural Dynamics 37, 697–710.
Peterson, J., 1993. Observations and Modeling of Background Seismic Noise, Open-file report
93-322, U. S. Geological Survey, Albuquerque, NM.
Priestley, M. J. N., Calvi, G. M., and Kowalsky, M. J., 2007. Displacement based seismic design
of structures. IUSS, Pavia, 721 pp.
Quanterra, 2011. Q330S Manual, available at http://www.q330.com, last accessed 18 November 2011.
Ringler, A. T., and Hutt, C. R., 2010. Self-noise models of seismic instruments. Seismological
Research Letters 81, 972–983.
Schweizerischer Erdbebendienst ETH Zürich (SED), 2009. Determination of Site Information for
Seismic Stations in Switzerland, Report SED/PRP/R/004/20090831.
Schweizerischer Erdbebendienst ETH Zürich (SED), 2010. Erneuerung des Starkbebennetz-
werkes in der Schweiz — Jahresbericht 2010, Report SED/SSMNet_R/R/001/20101029.
Schweizerischer Erdbebendienst ETH Zürich (SED), 2011. Station Noise, http://www.seismo
.ethz.ch/research/groups/alrt/pqlx/index, last accessed 25 January 2013.
Southern California Seismic Network (SCSN), 2011. The Southern California Seismic Network
Home Page, http://www.scsn.org, last accessed 25 January 2013.
Swiss Strong-Motion Network (SSMNet), 2011. Swiss Strong-Motion Network Renewal, avail-
able at http://www.seismo.ethz.ch/research/groups/risk/projects/SSMNrenewal/index, last
accessed 25 January 2013.
Trifunac, M. D., and Todorovska, M. I., 2001. A note on the useable dynamic range of accel-
erographs recording translation. Soil Dynamics and Earthquake Engineering 21, 275–286.
Wiemer, S., Giardini, D., Fäh, D., Deichmann, N., and Sellami, S., 2009. Probabilistic seismic hazard
assessment for Switzerland: best estimates and uncertainties, Journal of Seismology 13, 449–478.
Zambonelli, E., de Nardis, R., Filippi, L., Nicoletti, M., and Dolce, M., 2011. Performance of the
Italian strong-motion network during the 2009, L’Aquila seismic sequence (central Italy),
Bulletin of Earthquake Engineering 9, 39–65.
(Received 3 August 2011; accepted 19 December 2011)
APPENDIX
Because the worldwide probability of a large earthquake appears to be constant in time
(Michael 2011), the probability of the occurrence of a catastrophic earthquake in any short
interval dt is thought to be α NðtÞdt, where NðtÞ is some measure of the population at risk at
time t and α is a constant. The occurrence of catastrophic earthquakes is then a nonstationary
Poisson process (see Ogata 1983 for a brief discussion of nonstationary Poisson events and
the maximum likelihood method of analyzing them). Given a sequence of n catastrophic
earthquakes at times t1 ; t2 ; : : : ; tn within the interval ðS; TÞ, the likelihood of that parti-
cular sequence is the product of the likelihoods αNðti Þ of an earthquake at each time t i and the
t
likelihoods Exp½−∫ tiþ1i
αNðtÞdt% of no events in time intervals S to t 1 , t i to t iþ1 , and t n to T.
Thus, the likelihood of the particular sequence is
!Y n " ! ð "
T
L¼ αNðt i Þ Exp − αNðtÞdt
EQ-TARGET;temp:intralink-;e1;62;468 (A1)
i¼1 S
The natural logarithm of the likelihood is

!X
n " ðT
Log e ðLÞ ¼ nLog e ðαÞ þ Log e ðNðt i ÞÞ − α NðtÞdt
i¼1 S
One can then can solve the equation ∂ðLog e ðLÞÞ∕∂α ¼ 0 for the value
n
α ¼ ðT
NðtÞdt
S
that maximizes the likelihood of the particular sequence t 1 ; t 2 ; : : : ; t n . The standard devia-
tion σ α of α is
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u α
σα ¼ u tT
ð (A4)EQ-TARGET;temp:intralink-;e4;62;271
NðtÞdt
S
One can compare as a function of time the observed cumulative number of catastrophic
earthquakes in the time interval ðS; tÞ with the expected number α∫ tSNðtÞdt predicted at
any time t by the nonstationary Poisson process. However, one should realize that the
observed accumulation curve may wander from predicted curve because the abscissa
of the observed cumulative number of events is composed of the sum of independent
random intervals. Because the time of an event then depends upon the time of the pre-
ceding event, the events are not independent. For example, if one of the early intervals
between events is unusually long, the accumulation curve will fall below the predicted
curve, and there is no reason for the subsequent events to compensate for that shortfall
because the intervals are independent. The accumulation curve may then wander from the
predicted curve in a random walk.
ES‐1
The Pearson’s chi-square test (Evans 1955, p. 775) provides a quantitative measure of
how well the data fit the proposed nonstationary Poisson distribution. Divide the time interval
ðS; TÞ into m subintervals such that a significant number ð>5Þ of events is expected in each
subinterval. Then the chi-square statistic ðχ 2 Þ is
m
X ðOi − E i Þ2
χ2 ¼ (A5)
Ei
i¼1
t
where Oi is the number of observed events lying in the ith subinterval, and E i ¼ α∫ tiþ1
i
NðtÞdt
is the expected number of events in that subinterval. The probability P that a smaller value
of χ 2 would be observed if the data were drawn from the proposed nonstationary Poisson
process is given by the chi-square distribution with m-2 degrees of freedom and argument χ 2 .
A probability between 5% and 95% indicates that the data are consistent with proposed
nonstationary Poisson distribution.
REFERENCES
Evans, R. D., 1955. The Atomic Nucleus, Tata McGraw-Hill Publishing Company, Bombay,
India, 1002 pp., available at http://www.archive.org/details/atomicnucleus032805mbp.
Michael, A. J., 2011. Random variability explains apparent global clustering of large earthquakes,
Geophysical Research Letters 38, L21301.
Ogata, Y., 1983. Estimation of the parameters in the modified Omori formula for the aftershock
frequencies by the maximum likelihood procedure, Journal of the Physics of the Earth 31,
115–124.
ES‐2

Earthquake Spectra Volume 29

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Earthquake Spectra Volume 29

Uploaded by

Copyright:

Available Formats

Application of Supplemental Damping

Characteristics to Response Spectrum

For practicing engineers, knowledge of the characteristics of supplemental

responses of elastic single-story asymmetric-plan buildings have also been discussed

CHARACTERIZATION OF THE SUPPLEMENTAL DAMPING

THE EFFECTIVE ONE-STORY BUILDING

Therefore, the damping matrix of the EOSB can be expressed as:

It is well known that the damping matrix of an arbitrary single-story asymmetric-plan

or in terms of Equation 15, the equation of motion of the EOSB becomes

0.9589 0 12.5206 −5.2786 466.2 −69.7

Table 2. The properties of the EOSB (units: kN, m, rad, sec).

M! C! K! ξsd ēsd ρ̄sd

½cz! (sd ¼ ½c!z (sd ¼ 103 ×

0.9287 0 0.6685 0.4455 41.552 14.750

Building M! C! K! ξsd ēsd ρ̄sd

0.9611 0 0.4224 −0.0672 2.7810 −0.2991

M! C! K! ξsd ēsd ρ̄sd

RESPONSE SPECTRUM ANALYSES TO NONPROPORTIONALLY

THE RELATIONSHIPS BETWEEN THE SEISMIC RESPONSES OF THE EOSB

Table 7. Earthquake data.

Building T z (sec.) ξ ē Ω ηz ηθ (1/m)

Building TS Building TSS Building TF

SUMMARY AND CONCLUSIONS

This paper presents a rational framework for the quantitative assessment of

BRIDGE VULNERABILITY UNDER MULTIPLE HAZARDS

where M 0 is the threshold magnitude value, M p is the maximum regional finite

• Definition of the characteristics of each generated seismic event. Information

Bridge fragility analysis

where P½DSjPSA' is the probability of exceedance of a specific damage state DS given a

AjDS ðtÞ ¼ AjDS ½0' ⋅ ð1 − t ⋅ γÞ

Consequence type Eq. no.

Rebuilding CjRebðtÞ ¼ E RebcjRebW j Lj ð1 þ rÞt (13)

þðcS;truck þ cS;cargo Þ 100 24000Sj ðtÞ

þcOp ;truck 100

þðcTC OTruck þ cInv Þ 100

OCar ð1 − 100 Þþ Aj ðtÞLj

þOTruck 100 24000Sj ðtÞ

þOTruck 100 24000Sj ðtÞ

*Based on Stein et al. (1999)

LIFE-CYCLE RISK ASSESSMENT

Hazard type Service state Probability Eq. no.

S1 PjE;S1 ðtÞ ¼ P½NDj 'ðtÞ

S3 PjE;S3 ðtÞ ¼ P½MODj 'ðtÞ þ P½EXT j 'ðtÞ

S4 PjE;S4 ðtÞ ¼ P½COM j 'ðtÞ

S1 PjT;S1 ðtÞ ¼ P½S1j jSurvj 'ðtÞ ⋅ ½1 − PjT;f ðtÞ'

Generate a seismic event associating a

Probabilistic Ground Motion

Define the random location, depth, rupture

Evaluation of the shortest distance between

Evaluation of traffic conditions and detour

Evaluation of direct and indirect

Evaluation of the probability of a bridge in

Evaluation of the bridge risk induced by

NO Is the total number of

CASE STUDY: GROUP OF EXISTING BRIDGES

Figure 3. Schematic layout of the transportation network.

EVALUATION OF THE CONSEQUENCES

Direct Risk (USD tousands)

250 Percentile 95%

Indirect Risk (USD millions)

Mean Traffic Indirect Risk 35 Traffic Indirect Risk

Seismic Direct Risk Direct Network Risk

18 Seismic Indirect Risk Indirect Network Risk

Mean total network risk