Soil Dynamics and Earthquake Engineering 77 (2015) 453–468

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: www.elsevier.com/locate/soildyn

Preliminary probabilistic seismic hazard assessment for the Kingdom

of Saudi Arabia based on combined areal source model: Monte Carlo
approach and sensitivity analyses
Hani Mahmoud Zahran a, Vladimir Sokolov a,n, Salah El-Hadidiy Youssef a,b,
Wael Wassel Alraddadi a
a
National Center for Earthquakes and Volcanoes, Saudi Geological Survey, P.O. Box 54141, Jeddah 21514, Saudi Arabia
b
National Research Institute for Astronomy and Geophysics, Helwan, Cairo, Egypt

art ic l e i nf o a b s t r a c t

Article history: We create and test a framework for probabilistic seismic hazard assessment for the Kingdom of Saudi
Received 8 December 2014 Arabia using Monte Carlo simulation, recently developed models of seismic source zones and modern
Received in revised form ground-motion prediction equations (GMPE). A generalized seismic source model containing 43 zones
18 May 2015
has been compiled and seven GMPEs were selected. The assessment was performed on the basis of 100
Accepted 21 June 2015
synthetic seismic sub-catalogs with duration 10,000 years each. The hazard curves were calculated for
the nodes of 0.251
0.251 grid and the hazard maps were created in terms of PGA, PGV and seismic
Keywords: intensity for rock sites. Preliminary sensitivity analysis was performed to determine the importance of
Seismic hazard the input parameters and the level of uncertainty introduced by the parameters. The developed
Saudi Arabia
framework and the results of PSHA provide a benchmark for the comprehensive seismic hazard and
Deaggregation
seismic risk analysis and up-to-date seismic hazard maps for the Kingdom of Saudi Arabia.
Sensitivity analysis
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction much controversy [3,23,29,47–51,83] (see also discussion in EOS,

Saudi Arabia has experienced considerable earthquakes in the
past [12–15,61]. However the number of documented large earth-
quakes in the Arabian Peninsula is low due to sparsity of popula-
tion. The seismotectonic setting around Saudi Arabia suggests that
large earthquakes can occur along the Red Sea rift zones, Arabian
Gulf and Zagros belt [4,6,8,39,40,45,60,64,66,78,79,82]. Design of
buildings and structures in earthquake prone regions, seismic risk
estimation and management, and insurance business require
information related to expected seismic effect; the expected effect
should be expressed in terms of earthquake ground-motion
parameters, such as seismic intensity, peak amplitudes of ground
motion, pseudo-spectral acceleration (PSA) and ground-motion
time histories. The specification of engineering (or design) ground-
motion parameters is the goal of seismic hazard analysis (SHA). It
involves the quantitative estimation of ground shaking hazard at a
particular site taking into account characteristics of potentially
dangerous earthquakes around the site. The relation between
deterministic and probabilistic approaches for SHA is a subject of
n
Corresponding author. Tel.: þ 966 12 619 5000x2672.
E-mail address: sokolov.v@sgs.org.sa (V. Sokolov).
2003, 2004, 2005). The choice of approach depends on (a) the final
goal: how and where to expect to use the result, and (b) the
parameters of seismicity or likelihood of the worst-case event.
Seismic hazard mapping, development of design codes, retrofit
design, and financial planning of earthquake losses require mostly
probabilistic seismic hazard assessment (PSHA) especially in the
case of low-seismicity areas.
Several studies have been carried out to evaluate the seismic
hazard for different regions of the Arabian Peninsula. The earliest
study for the Kingdom of Saudi Arabia (KSA) was performed by
Thenhaus et al. [80], in which peak ground acceleration (PGA) and
velocity (PGV) were estimated for stiff soil and for 10% probability
of being exceeded in 100 years using several ground motion
prediction equations (GMPE). Al Haddad et al. [8,9] used these
ground-motion models to estimate PGA for 10% of being exceeded
in 50 years and suggested seismic zonation map for KSA to be used
as a basis for seismic design. Seismic hazard for particular areas in
KSA and the neighboring countries was analyzed by Deif et al. [36]
(Sinai Peninsula and Gulf of Aqaba area, PGA and spectral accel-
eration for rock sites, single GMPE); Deif et al. [37] (area of
construction of Makkah–Madinah high-speed rail, PGA and spec-
tral acceleration for rock sites, consideration of several GMPEs in
logic tree scheme); Osman [56] (Makkah area, rock sites, single

http://dx.doi.org/10.1016/j.soildyn.2015.06.011
0267-7261/& 2015 Elsevier Ltd. All rights reserved.
454 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
GMPE); Al-Malki and Al-Amri [11] (southwestern part of KSA, PGA
and spectral acceleration were estimated for bedrock and sedi-
mentary rock sites on the basis of semi-theoretical ground-motion
models obtained by stochastic simulation); Al-Arifi et al. [7]
(northwestern part of KSA, PGA and spectral acceleration for rock
sites, single GMPE).
Seismic hazard for the territory of United Arab Emirates (UAE)
was analyzed by Abdalla and Al-Homoud [1,2] in terms of PGA for
rock sites. Khan et al. [46] considered also spectral acceleration at
different natural periods and used several ground-motion predic-
tion equations in a logic tree scheme showing results of deag-
gregation for major cities. Sicbjornsson and Elnashai [67]
performed PSHA for stiff soil for Dubai in terms of PGA and
uniform hazard response spectra. Mohindra et al. [53] analyzed
seismic hazard (PGA and macroseismic intensity) for Yemen and
considered local soil conditions through amplification coefficients.
Probabilistic seismic hazard assessment for Arabian Peninsula
was performed by Pascucci et al. [59], in which PGA and spectral
acceleration at rock sites were estimated for particular cities.
Uncertainties in seismic sources characteristics (earthquake recur-
rence, maximum magnitude) and ground motion models were
incorporated using a logic-tree framework. Aldama-Bustos et al.
[10] estimated seismic hazard in terms of uniform hazard spectra
for rock sites in the cities of Abu Dhabi, Dubai and Ra's Al Khaymah
in the UAE and showed results of deaggregation.
The conventional Cornell-McGuire approach [32,52] has been
applied in almost all aforementioned studies, and the results were
not suitable for seismic loss analysis for extended portfolios (cities)
and spatially distributed structures (lifelines), because the con-
ventional PSHA does not allow consideration of ground-motion
correlation [24,52]. In this study we present a framework and
results of tentative “current-state” PSHA analysis performed for
KSA using the Monte Carlo approach. The Monte Carlo technique is
a flexible and rigorous tool to characterize various uncertainties in
the assessment and, at the same, it is extendable to earthquake
engineering applications such as probabilistic seismic loss analysis
(for examples, see Refs. [17,19]). The Monte Carlo technique is the
only possible way to analyze multiple-location hazard, i.e. ground
motion occurring simultaneously in several points during a
particular earthquake. The analysis is very important for assess-
ment of damage, loss and seismic risk for spatially distributed
structures (lifelines, critical elements of network, important
objects, urban areas, etc.) [24,35,42,43,58,71–73]. We considered
recently developed models of seismic source zones for the Arabian
Peninsula and surrounding regions, and modern ground-motion
prediction equations. We performed preliminary sensitivity ana-
lysis to determine the importance of input parameters (seismic
zonation, maximum and minimum considered magnitudes, local
site conditions) and the level of uncertainty introduced to the
results by the parameters. Previous PSHA studies for the region did
not contain such analysis.
2. Probabilistic seismic hazard assessment – the method
In probabilistic seismic hazard analysis, seismic hazard is defined as
the annual frequency of exceedance γ of different levels of ground
motion. In other words, the annual frequency γ , with which ground
motion parameter A exceeds a specific value A0 at a specific site, i.e.
γ ½A 4 A0 , is estimated. The return period (or more precisely “the mean
return period”) T γ is defined as the reciprocal of the annual frequency
of exceedance, i.e. T γ ¼ 1=γ . The term “return period” is more fre-
quently used in seismic hazard assessments than the term “annual
frequency of exceedance” due to convenience and usability. For the
Poisson occurrence of earthquakes in time, the probability of observing
at least one exceedance of the given ground motion level A0 in time
interval t, i.e. Pt ½A 4 A0  , is related to annual frequency γ as:
P t ½A 4 A0 j t ¼ 1  expð  γ ½A 4 A0 
tÞ ð1Þ
For t ¼1 year and small enough value γ ½A 4 A0 , the annual
probability of at least one exceedance is numerically almost equal
to the annual frequency of exceedance. Note, however, that these
quantities have different dimensions. Thus, a PSHA result is
represented by the frequency of exceedance, the probability of
exceedance, and the return period. A plot showing the calculated
annual frequencies of exceedance or the annual probabilities of
exceedance for different levels of ground motion parameter is
referred to as “hazard curve”.
The hazard is estimated at a particular site taking into account
characteristics (location, size and occurrence frequencies) of all
potentially dangerous earthquakes around the site. The most
important applications of PSHA results include seismic hazard
mapping, development of design codes, retrofit design, and
financial planning of earthquake losses [51].
The design seismic action is associated with a reference
probability of exceedance P te ½A 4 A0  during finite time period
te (the exposure period). Assuming a Poisson process for the
ground motion occurrence, the reference probability is related to
the annual frequency of exceedance γ and the exposure time as:
P te ½A 4 A0  ¼ 1  expð  γ ½A 4A0 
teÞ ð2Þ
The correspondent reference return period is defined as:
TðA 4 A0 Þ ¼  te=lnð1  P te ½A 4 A0 Þ ð3Þ
Several design levels may be defined in seismic codes. For
example, if the ordinary structure (OS) is designed and con-
structed to withstand the design seismic action without local or
global collapse, the recommended values for reference probability
POS are 10% (0.1) in a 50-year exposure period of engineering
interest, the correspondent return period is T OS ¼ 475 years, and
the annual frequency of exceedance is hence γ OS ¼ 0:002105. Note
that for essential or hazardous facilities (EHF) the collapse pre-
vention requirement may correspond to P EHF ¼ 2% (0.02) in a 50-
year exposure period of engineering interest, the return period
T EHF ¼ 2475 years, and the annual frequency of exceedance is
γ OS ¼ 0:000404. The reference probability 2% in a 50-year exposure
period is also applied for ordinary structures in low-seismicity
regions. For safely critical facilities (SCF, e.g. nuclear and defense-
related facilities) P NCR_SFC ¼ 0:5% (0.005) in a 50-year exposure
period (or 1% in 100-year) and return period T NCR_SFC ¼ 9475 years.
The adopted probability level is related to regional seismicity and
historical progress of seismic design requirements.
2.1. Monte-Carlo approach for probabilistic seismic hazard
assessment
In recent decades, Monte Carlo simulation is frequently used in
PSHA as an alternative to conventional numerical integration (so-
called Cornell-McGuire PSHA) [17,19,38,54,55,71,84]. The calcula-
tion steps are straightforward and intuitive: simulating earthquake
location, generating earthquake magnitude and fault dimensions,
and then estimating ground motion parameters for given earth-
quake characteristics. Simple statistical data analysis is carried out
next to develop seismic hazard curves. Through multiple repeti-
tions of this process, additional uncertainties regarding different
source models, uncertain maximum magnitude and parameters of
the Gutenberg–Richter relationships and choice of GMPEs can be
easily incorporated by treatment of epistemic and aleatory uncer-
tainties as being equivalent. In numerical integration, by contrast,
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
inclusion of various uncertainties and probabilistic models via
logic tree branches can be problematic, because the number of
combinations in a logic tree expands exponentially and discretiza-
tion of non-Gaussian probability distributions can be troublesome
[16]. Also, the Monte Carlo technique is the only possible way to
analyze so-called “multiple-location hazard”, i.e. ground motion
occurring simultaneously during a particular earthquake at several
points along an extended territory. The analysis requires consid-
eration of ground-motion correlation, i.e. similarity of ground
motion variability for different earthquakes (between-earthquake
correlation), and at different locations (within-earthquake correla-
tion). Ground-motion correlation is a key parameter for assess-
ment of damage, loss and seismic risk for spatially distributed
structures (lifelines, critical elements of network, important
objects, urban areas, etc.) [24,35,42,43,58,71–73].
In general, the simulation-based approach to seismic hazard
evaluation consists in two steps: (1) generation of a long-duration
stochastic catalog of earthquakes (or a number of stochastic sub-
catalogs that is more convenient computationally) for given
seismic source parameters (geometry, maximum magnitude,
earthquake recurrence, hypocentral depth); (2) calculation of
ground motion at selected sites of the studied region using all
earthquakes from the stochastic catalogs.
Fig. 1. The models of seismic source zones used by different authors. (a) Pascucci et al. [59]; Deif et al. [37]; Mohindra et al. [53]; Al-Arifi et al. [7].
455
A stochastic earthquake catalog represents seismic process
described by the used input parameters and considers uncertainty
of the parameters. It is compatible with knowledge about regional
seismicity and relations between ground motion parameters and
characteristics of particular earthquakes. On one hand, the dura-
tion of synthetic catalog (or sub-catalogs) should be long enough
to include several events of maximum possible magnitude, as
defined by the correspondent magnitude recurrence. On the other
hand, the number of the sub-catalogs and, correspondingly, the
total number of simulated years T tot , should ensure statistically
reliable determination of annual frequency of exceedance for high
amplitudes of ground motion (i.e., for very low probability of
exceedance) [35,54,55].
The seismic source model describes the spatial and temporal
distribution of earthquakes in a region. Each epicenter within the
source zone is determined randomly assuming that any location
within the source zone has an equal probability of being the
epicenter of the next earthquake. The depth of the given earth-
quake source is also generated randomly considering the possible
depth distribution. The number of earthquakes with particular
magnitude is determined against the magnitude–frequency dis-
tribution for that source zone. For each seismic event of stochastic
catalogs, the distribution of ground motion is calculated using the
456 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
specified ground-motion prediction equation (GMPE) and corre-
sponding random error considering the between-earthquake
variability for the event, the within-earthquake variability at each
site, and, if necessary, ground-motion correlation.
The hazard rates, i.e., the annual frequency of exceedance
γ t ¼ 1 ½A 4 A0  at the individual points, are computed from the statistics
of the simulated earthquakes. The rates were calculated by counting the
number NðA 4 A0 Þ of ground-motion values A exceeding particular
values A0 in relation to the specified time interval T tot (total duration
of stochastic catalogs). If T sub is the duration of every sub-catalog and
the number of sub-catalogs is N cat , then the total duration is
T tot ¼ T sub
N cat . Thus, the annual frequency γ t ¼ 1 ½A 4 A0  is deter-
mined as γ t ¼ 1 ½A 4 A0  ¼ NðA 4 A0 Þ=T tot .
In-house software that is similar, in general, with the software
described by Assatourians and Atkinson [17] has been used. The
software was developed by V. Sokolov and it has been applied in
research related to seismic hazard and loss estimations including
analysis of methodological problems in multiple-site hazard and
loss assessment [71–73].
3. Input data
3.1. Seismic source zones
Our goal is to create a PSHA model that can be used as
benchmark for up-to-date comprehensive seismic hazard and
seismic risk analysis for Saudi Arabia. Therefore, in this study we
do not make an attempt to create a new seismic source model; we
use the available models of seismic source zones reported in the
recent studies. We select only those models, for which all
necessary characteristics, i.e. maximum magnitudes and rate of
occurrence of earthquakes, are reported. At the same time, we
avoid models that provide superfluous detailing for the zones [36].
The composed model covering almost the whole territory of KSA
and neighboring regions has been compiled on the basis of the
models described in the following studies, namely: Pascucci et al.
[59], Deif et al. [37], Mohindra et al. [53], and Al-Arifi et al. [7].
These basic seismic source (BSS) models are presented in Fig. 1. It
is necessary to bear in mind, however, that the analysis of
characteristics of seismicity in the basic models has been per-
formed in different time periods in the past and therefore it did
not consider seismic events occurred during last 5–10 years.
The zones in the considered BSS models may partly or com-
pletely overlap each other. The composed seismic source (CSS)
model has been created as follows. First, we selected so-called
preferred model that provides the most detailed description of
source zones for particular area. The basic model suggested by
Pascucci et al. was used as the preferred model for the north-
eastern and eastern regions; the basic model of Mohindra et al.
was used for the southern region; the basic model of Deif et al. was
used for the south-western and western regions; and the basic
model of Arifi et al. was used for the north-western region.
Second, the zones in preferred BSS models were included, as a
whole, into the composed model. When a zone from preferred BSS
model overlaps with a zone from non-preferred model, the
configuration of the latter was modified correspondingly. Finally
43 seismic source zones are selected for the CSS model (Fig. 2).
Maximum magnitude for the zone of CSS model is assigned as that
in the preferred BSS model zone.
Location and boundaries of every zone in the composed model
were determined by comparison of corresponding characteristics
of the zones in the BSS models. The uncertainty associated with
the boundaries is rather high; therefore we assume that all four
BSS models are equally reliable.
Fig. 2. The composed model of seismic source zones used in this study. Numbers of
zones correspond to the numbers in Table E1 (electronic attachment).
Seismicity of seismogenic zone is described in terms of the
recurrence relationship [44]:
log Nm Z M ¼ a þ bM ; ð4Þ
where N is the number of earthquakes of magnitude M or greater
per unit time. The a value defines the intercept of the relationship
at M equals zero. The parameter b defines the relative proportion
of small and large earthquakes. However, the following truncated
exponential recurrence relationship is commonly used in practice:



exp  β ðM M min Þ  exp  β ðM max  M min Þ
Nm Z M ¼ αðM min Þ
 ;
1 exp  βðM max  M min Þ
ð5Þ
where α ¼ NðMmin Þ; Mmin is an arbitrary reference magnitude,
Mmax is an upper-bound magnitude; β ¼ b
lnð10Þ. This equation
results in the earthquake frequency approaching zero for the
upper-bound magnitude.
The rates of earthquake occurrence (number of earthquakes of
given magnitude) for particular zones in the CSS model are
evaluated using the following procedure. The same number of
synthetic catalogs (100 sub-catalogs with duration 10,000 years) is
generated for every basic seismic source models. The recurrence
relationship, parameters of which are provided by the selected
basic models, is used for determination of number of earthquakes
of given narrow range of magnitude M 70.125 applying truncated
exponential recurrence model. It is assumed that any location
within the source zone has an equal probability of being the
epicenter of the next earthquake.
For every seismic source zone in the composed model, we
calculated from the BSS catalogs the number of earthquake
epicenters, which lay in the zone, using a weighted summation
scheme. The scheme consists of the following steps. First, we
check whether the epicenter of every particular earthquake with
given magnitude range M70.125 from given BSS catalog lies
within the considered CSS zone. If not, zero weight is assigned
to the event; otherwise the weight equal to 1.0 is assigned and we
check whether this epicenter lies within any zone of next BSS
catalog. If the answer is “yes” for the BSS catalog, the weight equal
to 0.5 is assigned to the event, etc. Thus, the weight wi;j that is
assigned to every earthquake i from every BSS model j is
determined as wi;j ¼ 1=ni , where ni denotes the number of basic
seismic source models, for which the earthquake may be found
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468 457

Fig. 3. The weighting scheme used for counting number of events in the Composed
Seismic Source model. Symbols denote particular earthquakes in the BSS sub-
catalogs.

within any zone (Fig. 3). Finally, the total number of earthquakes of
a given magnitude range in the considered zone of the CSS model
P P
Nj NBSS
is determined as NM 7 0:125 ¼ ðwi;j Þ, where NBSS is the num-
i¼1j¼1
ber of basic models, and Nj is the number of earthquakes in the
particular basic model. The number of earthquakes normalized for
1 year is calculated as N 1; M 7 0:125 ¼ N M 7 0:125 =T tot , where T tot is
total duration of all synthetic catalogs in every basic model, i.e.
1,000,000 years (100 sub-catalogs
10,000 years). The N1; M 7 0:125
values are used for estimation of parameters of the Gutenberg–
Richter relationship (coefficients a and b) and of the truncated
exponential recurrence relationship (coefficients α and β). Char-
acteristics of all seismic source zones are listed in Table E1
(Electronic supplement).
In this study we do not consider a seismic source zone in the
central area of the Arabian Peninsula. The Arabian plate is a stable
landmass that does not exhibit any discernable trace of interior
deformation during the late Tertiary, and the interior of the Arabian
plate is also not known to have experienced any significant seismic
events over the past 2000 years [82]. The area was considered as
aseismic in most recent studies (e.g., Refs. [8,9,59,80]).
The synthetic sub-catalogs in the Composed Seismic Source Fig. 4. Examples of synthetic catalogs (earthquake epicenters) generated using the
model (100 sub-catalogs with duration 10,000 years) are gener- composed seismic source (CSS) model, selected duration 1000 years; (a) earthquakes
ated using the obtained recurrence relationships and assuming with magnitudes M 5.0–5.5 (b) intermediate and large earthquakes.
that any location within the source zone has an equal probability
of being the epicenter of the next earthquake. The depth of given to the recently developed GMPEs and to the GMPEs based on large
earthquake source is generated randomly considering uniform amount strong-motion data, i.e. so-called Next Generation of
depth distribution between 5 km and 25 km. Examples of dis- Ground Motion Attenuation Models, NGA [62].
tribution of earthquake epicenters in the synthetic sub-catalogs For peak ground acceleration, the models of Akkar et al. [5],
for the CSS model are shown in Fig. 4. Boore and Atkinson [26] (NGA), Campbell and Bozorgnia [28]
(NGA), and Zhao et al. [85] are used for active shallow crustal
3.2. Ground motion prediction equations sources (i.e. Iran, Zagros, Gulf Region, Dead Sea and Jordan), and
equal weights (0.25) were assigned to these models. The model of
The seismic hazard analysis typically uses multiple GMPEs Akkar et al. [5] supersedes previous GMPEs derived for Europe and
deemed applicable to the region or the site of interest. The use the Middle East, and address shortcomings identified in those
of alternative GMPEs is a popular approach to consider epistemic models. The model of Zhao et al. was derived predominantly from
uncertainties in the correct median motions for given event Japanese data; however it has been identified [25] as a proper
scenarios. In most of the studies related to seismic hazard assess- candidate for selection within PSHA for shallow crustal seismicity.
ment for Arabian Peninsula, alternative ground-motion models The Atkinson and Boore’ [18] model is used for stable continental
were selected for use within the various considered tectonic region of Saudi Arabia in conjunction with the mentioned above
provinces: the active shallow crustal sources, the stable regions equations for the active shallow crustal sources. The largest weight
and extensional zones. In this work we select ground motion (0.60) is assigned to the Atkinson and Boore’ model for stable
prediction equations according to the tectonic regime associated regions and equal weights (0.10) are assigned for crustal source
with the earthquakes in each source zone. The preference is given equations. The model of Pankow and Pechmann [57] that
458 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
supersedes previous study of strong ground motions in exten-
sional tectonic regimes by Spudich et al. [77] is used for exten-
sional zones in the Red Sea and the Indian Ocean. The model of
Zhao et al. [85] does not consider peak ground velocity; therefore
for estimation of hazard in terms of PGV, we use the model
recently developed by Cauzzi et al. [30]. All the GMPEs allow
considering local site effects through average shear-wave velocity
of the upper 30-m column (Vs30).
Hazard in terms of seismic intensity (Modified Mercalli Inten-
sity, MMI, scale) is estimated using PGA–MMI conversion equa-
tions. First, PGA is calculated at a site from every earthquake using
the specified ground-motion prediction equation (GMPE) and
random error that reflects the scatter of peak amplitudes; second,
corresponding seismic intensity is evaluated. We use the conver-
sion equations proposed recently for Turkey by Bilal and Askan
[22], the equation applied by Mohindra el al. [53] for Yemen, and
the worldwide relationship developed by Chernov [31]. The
equations are as follows (PGA in cm/s2):
Bilal and Askan [22]
MMI BA1 ¼ 0:132 þ 3:884log 10 PGA
MMI BA2 ¼  1:692 þ 0:793log 10 PGA þ0:653M W 2:746log 10 Rep
where MW is the moment magnitude and Rep is the epicentral
distance
Mohindra et al. [53]:
MMI MOH ¼  0:715 þ 3:92log 10 PGA
Chernov [31]:
MMI CH ¼ 0:84 þ 3:8log 10 PGA
The resulting intensity to be considered in hazard estimations
is calculated as:
MMI ¼ ðMMI BA1 þ MMI BA2 þ MMI MOH þ MMI CH Þ=4
i.e., equal weights are assigned to different conversion relationships.
The selected ground motion models use different distance
metrics, i.e. the minimum distance between the rupture and the
site (RRUP) and the minimum distance between the surface
projection of the source plane and the site (Joyner–Boore distance,
RJB). For small to intermediate-size earthquakes (M o5.5–6.0), the
hypocentral distance may be used instead of RRUP, and the
epicentral distance – instead of RJB. Determination of the distance
metrics for area located in the vicinity of large earthquakes
requires information about dimensions and orientation of the
Fig. 5. Examples of hazard curves for particular cities.
source plane. The Monte Carlo technique that is based on event-
by-event calculation allows considering variations in strike and
dip angles of the rupture, as well as in the source dimensions. We
applied the following scheme for the source orientation and
dimensions of the M46.0 earthquakes within every source zone.
The dip angles are fixed as 851 (almost vertical fault), and the
strike angles are determined using uniform distribution between
01 and 3601, i.e. all strike angles are equally possible. The source
dimensions are estimated using sampling from normal distribu-
tion considering magnitude–dimension relations provided by
Vakov [81], which define mean values of source area and length,
and the error terms.
4. Results of PSHA and sensitivity analysis
Probabilistic seismic hazard assessment was performed on the
basis of long synthetic catalog (100 sub-catalogs with duration
10,000 years each). The catalog has been generated using the
ð6aÞ composed seismic source model that consists of 43 seismic source
zones (Fig. 2 and Table E1). The hazard curves were calculated for the
nodes of grid 0.251
0.251. Only earthquakes of MW Z4.5 were
considered. The examples of hazard curves, i.e. annual frequencies
ð6bÞ
of exceedance for different values of ground motion parameter, are
shown in Fig. 5. Fig. 6 shows the contour hazard maps in terms of
peak ground acceleration (cm/s2) for a rock condition (average shear
wave velocity of the top 30 m of soil column, Vs30¼800 cm/s)
with 10% and 2% probabilities of exceedance in 50 years, which
ð7Þ
correspond to return periods 475 and 2475 years. The hazard maps in
terms of peak ground velocity (cm/s) and seismic intensity (MMI
scale) are given in Electronic supplement (Figs. E1, E2). Tables 1 and
ð8Þ
2 present the hazard values (PGA, PGV and seismic intensity)
estimated for particular locations (cities).
ð9Þ 4.1. Sensitivity analysis
Sensitivity study is necessary in SHA to determine the impor-
tance of input parameters and the level of uncertainty introduced
to the results by the parameters [19,20,33,34,41,63,65,76]. The
ground-motion models and area-source model were mentioned as
the parameters with the highest sensitivity. Some authors also
pointed to the maximum magnitude (M max ) values and the
magnitude–frequency distribution [33], as well as to the choice
of minimum magnitude (M min ) values [20]. The impact of M min and
M max values depends on ground-motion frequency and return
period (probability of exceedance). Obviously, the degree of
influence depends also on the location of the site, for which the
hazard is assessed.
4.1.1. The area source model
The assumption that future earthquakes will occur randomly
within seismically homogeneous source zones is frequently made
in seismic hazard analysis. The definition of a seismic source zone
is based on the interpretation of the geological, geophysical and
seismological data. In practice, seismic source zones are difficult to
define, because the evaluation and interpretation of the various
types of information available depend strongly on individual
judgment or opinion and different experts may define source
zones differently. The spatio-temporal characterization of seismic
activities and seismotectonic features, which controls the rates of
seismicity in and around the site, may be particularly uncertain in
low-to-moderate seismic regions. On one hand, the boundaries
should mark differences in the statistic of earthquake occurrence;
on the other hand they should reflect an appropriate degree of
conservatism [16].
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468 459

Fig. 6. Schemes of ground motion distribution (PGA, cm/s2, rock sites) (a) 10% probability of being exceeded in 50 years (return period 475 years); (b) 2% probability of being
exceeded in 50 years (return period 2475 years).

At the same time, assumption of sharp changes in seismicity at
the presumed boundaries is not realistic. An abrupt change in
earthquake rates at a boundary may mean that significantly
different hazard levels may be predicted at sites a few kilometers
apart near such a boundary. To avoid discontinuities in seismicity
along the boundaries, Bender [21] suggested to model earthquakes
at each point within the zone as having a normal distribution
about that point. This model is equivalent to one that regards the
possible locations of the source zone as being normally distrib-
uted. The selected standard deviation σ of the distribution reflects
460 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468

Table 1
Results of PSHA for several locations (cities) in KSA, PGA, cm/s2, rock condition.

Location of point Return period (years) City

E, Long N, Lat 60a 475 1000b 2500 10,000

34.83 29.30 160 435 550 700 850 Haql

35.67 27.33 30 80 110 155 255 Duba
36.42 26.25 25 75 105 150 250 Al Wajh
37.25 25.00 30 85 120 175 320 Umm Lajj
39.67 24.50 15 50 70 120 260 Al Madinah
38.08 24.17 30 90 110 160 255 Yanbu’ al Bahr
38.80 23.80 25 75 110 190 390 Badr Hunayn
39.00 22.83 20 75 110 155 265 Rabigh
39.13 21.50 25 70 100 145 230 Jeddah
39.83 21.42 15 30 45 65 115 Makkah
40.33 20.17 20 75 105 155 255 Al Lith
41.09 19.09 25 75 105 150 250 Al Qunfundhan
42.50 18.25 15 40 55 95 200 Abha
42.00 17.66 25 75 110 155 260 Ash Shuqayq
44.17 17.50 15 65 100 175 390 Najran
42.58 16.83 25 80 110 160 260 Jizan
48.50 28.38 10 30 40 60 110 Al Khafji
49.67 27.00 10 30 35 50 85 Al Jubail
50.08 26.45 10 25 30 40 65 Ad Dammam

a
50% of being exceeded in 50 years.
b
5% of being exceeded in 50 years.

Table 2
Results of PSHA for several locations (cities) in KSA, PGV (cm/s) and MM intensity,
rock condition.

Location of point Return period (years) City

a
E, Long N, Lat PGV MM Intensity

475 2475 475 2475

34.83 29.30 26.0 55.0 9.0 10.0 Haql

35.67 27.33 6.0 11.5 6.3 7.4 Duba
36.42 26.25 4.0 8.5 6.1 7.3 Al Wajh
37.25 25.00 4.5 11.0 6.3 7.6 Umm Lajj
39.67 24.50 2.0 4.5 5.3 6.7 Al Madinah
38.08 24.17 4.5 11.0 6.3 7.4 Yanbu’ al Bahr
38.80 23.80 3.5 8.0 6.1 7.5 Badr Hunayn
39.00 22.83 4.0 9.50 6.2 7.4 Rabigh
39.13 21.50 3.5 8.5 6.0 7.2 Jeddah
39.83 21.42 2.0 3.5 4.7 5.9 Makkah
40.33 20.17 4.0 9.5 6.1 7.4 Al Lith
41.09 19.09 4.0 9.5 6.1 7.3 Al Qunfundhan
42.50 18.25 2.0 4.5 5.1 6.5 Abha
42.00 17.66 4.5 10.0 6.2 7.4 Ash Shuqayq
44.17 17.50 2.5 7.0 5.9 7.5 Najran Fig. 7. Distribution of epicenters (crosses) of earthquakes with magnitudes MW
42.58 16.83 4.5 10.0 6.2 7.4 Jizan 5.26–5.75 in the synthetic sub-catalog (selected period 2500 years). (a) “Sharp-
48.50 28.38 3.5 6.5 4.7 5.8 Al Khafji boundary” model of seismic zones; (b) “soft-boundary” model, threshold distance
49.67 27.00 3.5 6.0 4.6 5.6 Al Jubail 0.61.
50.08 26.45 3.0 5.5 4.4 5.3 Ad Dammam

a
MM intensity V corresponds to the range of calculated intensities between
4.6 and 5.5; MMI VI – to 5.6–6.5; MMI VII – to 6.6–7.5; MMI VIII – to 7.6–8.5, etc.
the uncertainty in the source zone location; σ ¼0 (no variability)
reduces to the fixed boundary case. Typically, σ is 25% or less of the
width of the source zone.
In our study we apply a different scheme to consider “soft
boundaries” of area sources. The scheme allows certain number of
earthquake epicenters, which are initially located near the border of a
zone, to be moved symmetrically in the neighboring zone [71]. The
relative percentage of displaced epicenters changes from 0% at a
threshold distance of several tens of kilometers from the border to
50% at the border. The procedure is applied only for earthquakes with
magnitudes less than a threshold magnitude (e.g. 6.0). Examples of
stochastic catalogs calculated considering the “sharp-boundary” and the
“soft-boundary” models of the seismic source zones are shown in Fig. 7.
To analyze the influence of “softness” of the boundaries on results
of seismic hazard assessment, for the case of PGA and PGV estima-
tions we calculate ratio RAT sof t;D between the hazard estimations
obtained for selected locations (cities) as RAT sof t;D ¼ HAZ sof t;D
=HAZ sharp , where D is the threshold distance; HAZ sof t and HAZ sharp
are the PGA or PGV estimations considering the “soft-boundary” and
the “sharp-boundary” models, correspondingly. For the case of
seismic intensity estimations, the increment DIF sof t;D is calculated as
DIF sof t;D ¼ HAZ sof t;D  HAZ sharp . Fig. 8 shows the ratios and intensity
increments for different threshold distances and return periods.
The influence of “soft” boundaries varies depending on mutual
location of the point of calculation and the boundaries of seismic zone,
and on the accepted threshold distance. It seems that the highest
difference between the “soft-boundary” and “sharp-boundary” models
(e.g., cities of Duba, Umm Lajj, Al Madinah, and Badr Hunayn)
corresponds to the points located near but outside the zones with
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
Fig. 8. Influence of the source zone boundary models on the hazard assessment.
Difference between the hazard estimations obtained for selected locations (cities,
see Table 1) using the “sharp-boundary” and “soft-boundary” models of seismic
source zones; influence of threshold distances D and selected return period. The
threshold distance is 0.61 for PGV and MMI.
Fig. 9. Influence of minimum magnitude selection on hazard assessment. Differ-
ence (ratio RAT Mmin ) between the PGA estimations obtained for selected locations
(cities, see Table 1) using M min ¼ 4:0 and M min ¼ 4:5.
461
relatively high-level seismicity as compared with neighboring zones.
The contour hazard maps that were compiled using the “soft-bound-
ary” model (threshold distance of 0.61) are given in Electronic supple-
ment (Figs. E3–E5). Tables E2 and E3 in Electronic supplement present
the correspondent PGA, PGV and seismic intensity values for
particular sites.
What threshold distance, or the Bender's standard deviation in
normal distribution of earthquake location, should be applied for
particular seismic source zonation is a judgment call. When the
locations of boundaries and zonal seismicity have been specified,
corresponding sensitivity analysis should be carried out to demon-
strate balance between conservatism and uncertainty.
4.1.2. The minimum and maximum magnitudes
In probabilistic SHA the choice of minimum magnitude M min ,
on one hand, relates to a rather arbitrary decision: (a) what
earthquakes can be ignored for engineering purposes and
(b) could one disregard contribution to the hazard associated with
high-frequency and high-amplitude vibration from small magni-
tude earthquakes. The level of high-frequency ground motion,
even if the motion is characterized by short duration, may be
important for some fragile electric equipment and for PGA based
emergency shutdown systems that use information from a single
station. On other hand, the minimum magnitude value could be
restricted by the lower magnitude limit of the used strong-motion
prediction model.
There are only a few studies related to the effect of choice of
minimum magnitude (see for example a short review in Ref. [20]).
In this work we use M min ¼ 4:5 for the basic calculations, and check
the influence of selection of a smaller value M min ¼ 4:0 for selected
locations. The influence of the smaller values of minimum magni-
tude in terms of ratio RAT Mmin ¼ PGAMmin ¼ 4:0 =PGAMmin ¼ 4:5 for the
“sharp-boundary” model of seismic sources is shown in Fig. 9. In
general, the lower value of minimum magnitude selected in PSHA
increases the PGA estimations; the difference may reach 35–40% in
considered cases. As expected [20,76], the impact of the minimum
magnitude on the hazard results is decreasing with increasing of
return period, and the influence of M min depends on relative
location of the site and seismic zones, as well as on the character-
istics of the zone. The considered variances in minimum magni-
tude values practically do not affect the PGV and seismic intensity
estimations.
Estimation of maximum possible magnitude that can occur in a
source zone is a difficult task. On one hand, the maximum possible
magnitude may be chosen as the maximum magnitude of histor-
ical catalog. However uncertainties on the historical magnitudes
are, as a rule, not less than 0.5 units of magnitude. Analysis of
seismotectonic and paleoseismic data may lead to rather larger
uncertainty. The values of maximum magnitude assigned to source
zones in the basic seismic source models, which overlap each
other, may differ considerably. We apply a rather simple analysis of
the impact of variation of maximum magnitude on the results of
PSHA by increasing M max values by 0.5 unit of magnitude for all
zones in the composed seismic source model.
The influence of the higher values of maximum magnitudes in
terms of ratio RAT Mmax ¼ HAZ Mmax þ 0:5 =HAZ Mmax for the “sharp-
boundary” model of seismic sources is demonstrated in Fig. 10. In
general, the impact of the maximum magnitude on the PGA
hazard results is not considerable for considered return periods:
in most cases the ratios do not exceed a value of 1.2. It has been
noted [20,76] that the variation of maximum magnitudes is not so
crucial in probabilistic seismic hazard estimation (at least for PGA)
for regions with weak and moderate seismicity. However, for PGV
estimations the influence of variation of maximum magnitudes
may be considerably high, especially for site located in the north-
western part of the studied area.
462 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
Fig. 10. Influence of maximum magnitude selection on hazard assessment. Differ-
ence between the PGA and PGV estimations obtained for selected locations (cities,
see Tables 1 and 2) using the accepted values of maximum magnitudes and the
values increased by 0.5 units. The dashed line marks upper level of the ratios (1.2)
for PGA estimations.
Estimation of hazard in terms of seismic intensity in our study is
based on peak ground acceleration (Eqs. 6–9); therefore it reveals a
similar behavior as that in the case of PGA hazard. The impact of
changes in maximum magnitude values on the intensity hazard results
is apparently insignificant for the considered cases. The immediate
relations between seismic intensity and earthquake characteristics, as
well as relations based on peak velocities or spectral parameters of
ground motion [69], should be used for comprehensive analysis.
4.1.3. Site effect
It is well understood that near-surface geological conditions may
strongly affect earthquake ground motion at a particular site due to
amplification of the shaking amplitude and change of frequency
content of the motion. In certain cases (e.g., building code provisions),
a few generalized site classes have been selected to describe the variety
of local soil conditions and to characterize their effect on ground
motion. A widely used site classification system is based on the
properties of the top 30 m of the soil column (average shear wave
velocity Vs30) disregarding the characteristics of the deeper geology.
Modern ground motion prediction equations include rough character-
ization of site amplification and estimation of possible nonlinear effects
during high-amplitude shaking based on the Vs30 data. To reduce the
uncertainty associated with the description of site effect in GMPE, some
additional measures can be employed, for example the fundamental
frequency of a site [27]. The schemes for consideration of site effect in
PSHA using models for Fourier amplitude spectra and frequency-
dependent site amplification functions are described in Refs.
[68,70,74–76]. However, such estimations require comprehensive
information about the local site conditions. In this work we estimate
the effect of local site amplification comparing calculation of ground
motion for “rock” and “firm soil” conditions applying for “firm soil” the
Fig. 11. Influence of local site conditions on hazard assessment. Difference between
the hazard estimations obtained for selected locations (cities, see Table 1) using the
“firm soil” (Vs30 ¼300 m/s) and the “rock” (Vs30 ¼ 800 m/s) conditions.
value Vs30¼300 m/s in the ground-motion models. The influence of
local site conditions in terms of ratio RAT soil ¼ HAZ soil =HAZ rock (PGA
and PGV) and intensity increment DIF soil ¼ HAZ soil  HAZ rock for the
“sharp-boundary” model of seismic sources is demonstrated in Fig. 11.
In general, the impact is significant both for the PGV and the
PGA estimations: the ratios for PGV are higher than 1.6 and the
ratios for PGA are about 1.4–1.5. However, the site effect in terms
of relative amplification may vary between the sites reflecting the
relationship between dominant frequency range of ground motion
and fundamental frequency of a site (as it specified by applied
GMPEs), as well as the relative numbers of small, moderate and
large earthquakes causing different levels of ground motion.
In our study, calculation of hazard in terms of seismic intensity is
based on peak ground acceleration. Because the PGA–MMI relation
does not consider influence of ground motion duration, the effect of
local site conditions on intensity values seems to be underestimated
for large earthquakes and overestimated for small and intermediate-
size events occurring closely to the sites of interest.
The joint influence of the “soft boundaries” and local site effect
in terms of ratio RAT boundary; site ¼ HAZ boundariy;soil =HAZ sharp;rock and
intensity increment DIF boundary; site ¼ HAZ boundariy;soil  HAZ sharp;rock is
shown in Fig. 12. The ratios may reach values of 2.3–2.4 for PGA
and 2.8–3.0 for PGV; the increment of MM intensity may reach a
value of 1.4. The degree of influence depends on the relative
location between a site and neighboring source zones, as well as
on the characteristics of seismicity of the zones.
4.2. Deaggregation
Probabilistic seismic hazard assessment considers influence
from all earthquakes of all possible magnitudes at all significant
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
Fig. 12. Joint influence of the source zone boundary models and local site
conditions on hazard assessment. Difference between the PGA, PGV, and seismic
intensity hazard estimations obtained for selected locations (cities, see Table 1)
using the “firm soil” (Vs30¼ 300 m/s) together with the “soft-boundary” model
(threshold distance 0.61) and the “rock” (Vs30 ¼ 800 m/s) together with the “sharp-
boundary” model.
distances from the site of interest. Thus PSHA combines contribu-
tions from earthquakes with entirely different damaging potential
into resulting hazard curves. Deaggregation procedure is used to
identify the earthquake scenarios (e.g. magnitude–distance, or M–
R, pairs), which may produce ground motion exceeding particular
level of shaking intensity for the site or region of interest. In other
words, deaggregation determines relative frequency of each M–R
pair being the causative events for selected return period (or
probability of being exceeded).
A significant advantage of using Monte Carlo simulation rather
than numerical integration in hazard analysis is that it facilitates
the production of seismic deaggregation results at any probability
level (i.e. to show hazard contributions by magnitude and dis-
tance), because all information that is needed for seismic deag-
gregation is kept throughout the analysis [17,19]. Examples of
PGA-hazard deaggregation for several locations are shown in
Fig. 13. In general, relatively small-magnitude earthquakes that
may occur near the sites bring the main contribution to PGA
exceedances for short return periods (or relatively high probability
of exceedance). The stronger earthquakes that may occur at
distances up to 100–150 km are also important. For longer return
periods (or lower probability of exceedance), only the nearby
events of various possible magnitudes produce the main contribu-
tion to the PGA hazard.
Fig. 14 compares the results of deaggregation for PGA and PGV
hazard estimated for city of Duba. Similar comparisons for other
selected locations are presented in Electronic supplement (Figs
E6–E9). Characteristics of dominant earthquakes for PGV hazard
463
are characterized by the larger magnitudes and the broader range
of distances than that for PGA hazard, especially for short return
periods, showing magnitude- and distance-dependent influence of
spectral content of ground motion. Indeed, closely occurring
small-magnitude earthquake may frequently cause relatively
high-amplitude acceleration peaks; however amplitudes of ground
velocity would not be high enough, as compared with those from
stronger and more distant events.
5. Discussion
The results obtained in this study using the Monte Carlo
technique show that in general the features of ground motion
distribution are similar to those obtained in the previous studies
using the conventional Cornell–McGuire approach [8,9,80]. The
hazardous areas are located along the shoreline of Red Sea and the
Arabian Gulf. The level of seismic hazard for KSA is highest in the
northwestern part of the country near Gulf of Aqaba. However,
when comparing results of PSHA obtained in this study and in
previous works, it is necessary to bear in mind that the results may
correspond to different geological conditions considered in the
hazard estimations. Different descriptions of a “zone boundary
effect” may have also been used. Therefore, together with PGA
values obtained in the previous studies, we show several estima-
tions that correspond to various input models used in our study
(Table 3).
As can be seen from the comparison for some locations, there is
a discrepancy between the PGA values estimated in our study and
obtained by other researchers. The difference may be explained by
utilization of different models of seismic source zones and
attenuation equations. Unfortunately, sensitivity analysis was not
performed in the previous works; therefore it is not possible to
analyze the source of discrepancy.
One of the main goals of deaggregation is providing informa-
tion about combinations of earthquake magnitude–epicentral
distance that would bring the most contribution to the hazard
for given site and selected return period. These so-called “domi-
nant” earthquakes are used to select a single or several “design” or
“scenario” earthquakes to be considered in advanced engineering
analysis (see, for example, Refs. [52,54]). The relatively broad range
of magnitudes for dominant (i.e. bringing the highest contribution
to hazard) earthquakes that were determined for several sites (i.e.
Duba, Jeddah, Jizan) may be explained as follows. On one hand, the
small earthquakes will occur much more frequently than the
intermediate size events. On the other hand, the scatter of possible
values of ground motion parameter at a site is modeled using
lognormal distribution. Standard deviation of the distribution is
usually about 0.3 of logarithmic units, which corresponds to about
2 times the difference. Thus, if the small earthquakes frequently
occur near the site, they may cause high values of peak accelera-
tion (due to the positive variances in lognormal distribution)
during relatively large number of events, which is comparable
with the number of exceedances from the stronger and less
frequent earthquakes. However, it is necessary to bear in mind
that the high peak accelerations from small earthquakes are
related to sharp single peaks during short and high-frequency
vibration. Therefore, for the purpose of selection of “design”
earthquakes, the results of PGA deaggregation should be analyzed
together with deaggregation of peak velocity and spectral
acceleration PSHA.
Figs. 13–14 show results of deaggregation of PGA hazard
performed using the “sharp-boundary” model of seismic zones.
However, results of PSHA are sensitive to the location of the
boundaries (see Fig. 8). Deaggregation of PGA hazard performed
using the “soft-boundary” models for locations that reveal the
464 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468

Fig. 13. Results of deaggregation for particular cities. Contribution of different magnitude (4.5–7.5) – distance bins to the PGA hazard, “sharp-boundary” model, rock site; left
diagrams – return period 475 years; right diagrams – return period 2475 years.
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468 465

Fig. 14. Location: Duba. Comparison of results of deaggregation for PGA and PGV hazard, “sharp-boundary” model, rock site.

Table 3 model. Thus, the variability of results introduced by the config-

Comparison between results of PSHA (PGA, cm/s2) from different studies. uration of the seismic source zones, as well as by the activity rates
assigned to the zones, should be carefully analyzed.
Location and Refs. Reported This studya
value
SharpB þ R SoftB þR SharpB þ S SoftBþ S
6. Conclusion
Deif et al., [37]
Makkah 68 32 35 42 46 The main goal of this work is to create and test a framework for
Al Madinah 90–95b 45 70 55 100
probabilistic seismic hazard assessment (PSHA) for Kingdom of
Jeddah 90–95b 70 95 100 140
Al-Arifi et al. [7] Saudi Arabia using Monte Carlo technique. A generalized compo-
Haqil 4300b 430 400 500 460 site seismic source model for the region is created on the basis of
Duba 200b 80 125 120 180 seismic zonation models that are available in the literature for
Pascucci et al.
particular territories. The composite model is used for generation
[59]
Jeddah 70 70 95 100 140
of 100 sub-catalogs with duration 10,000 years each. The pre-
Osman [56] liminary PSHA is focused on analysis of sensitivity of the results to
Makkah 80 32 35 42 46 uncertainties in input data, namely: seismic zonation, maximum
a
and minimum considered magnitudes, and possible influence of
SharpB þ R – “sharp-boundary” model together with rock site, SharpB þ S –
local site conditions. The analysis allows understanding the
“sharp-boundary” model together with firm soil site, SoftBþ R – “soft-boundary”
model together with rock site, SoftBþ S – “soft-boundary” model together with influence of uncertainties and selecting most important character-
firm soil site. istics of seismicity and seismic effect for intensive study. The
b
The values were estimated from contour maps. compiled input data, the developed framework and the results of
seismic hazard assessment may provide a bench-mark for the up-
to-date seismic hazard maps for KSA that meet modern standards
highest influence of the boundary “softening” is shown in Fig. 15 of rigor, quality, and engineering practice, and which may be used
and in Electronic supplement (Figs. E10–E12). The “softening” in developing building codes and other structural engineering
allows some events to be located outside a zone of relatively high guidelines, as well as for emergency planning, land-use planning,
level of seismicity. These earthquakes may occur near to a site of and insurance applications.
interest and, therefore, the parameters of dominant earthquakes In this study, as the basic input data for PSHA, we used the
are shifted to smaller epicentral distances and sometimes to larger published models of seismic source zones recently developed for
magnitudes, as compared with those for the “sharp-boundary” the Arabian Peninsula and surrounding regions, as well as the
466 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
Fig. 15. Location – Umm Lajj. Comparison of results of deaggregation for the “sharp-boundary” and for “soft-boundary” (threshold distance 0.61) models of seismic source
zones, return period 475 years.
modern ground-motion prediction equations. The generalized
seismic source model containing 43 zones has been compiled
and several GMPEs, which include the Next Generation of Ground
Motion Attenuation models and the up-to-date model developed
for Europe and the Middle East, were selected to be used in a
weighted combination. The input data were used for probabilistic
seismic hazard assessment in terms of peak ground acceleration,
peak ground velocity and macroseismic intensity.
The results obtained in this study using the Monte Carlo
approach show that the features of ground motion distribution,
in general, are similar to those obtained in the previous studies by
conventional Cornell–McGuire PSHA. The hazardous areas are
located along the shoreline of Red Sea and the Arabian Gulf. The
level of seismic hazard for KSA is the highest in the northwestern
part of the country near Gulf of Aqaba, where hazard estimates for
return period 475 years and rock sites exceed 400 cm/s  2 for PGA,
25 cm/s for PGV, and may reach MMI IX. Preliminary sensitivity
analysis has been performed to determine the importance of input
parameters and the level of uncertainty introduced to the results
by the parameters. The analysis shows that for the given region,
which is characterized by weak and moderate seismicity, the area-
source model (i.e. location of the boundaries of seismic source
zones) is the parameter with the highest impact on the results.
Consideration of influence of local site conditions is also of great
importance. At the same time, it has been found that the variation
of maximum magnitudes (0.5 units of magnitude) assigned to the
seismic source zones is not crucial for PGA hazard, at least for
return period 475 and 2475 years. It has been shown also that
deaggregation analysis is an efficient tool for visualization of
results of sensitivity study.
Future studies should be concentrated in comprehensive ana-
lysis of the up-to-date regional earthquake catalog and more
accurate definition of configuration and seismicity parameters of
areal zones, including maximum magnitude; development of
zoneless models – e.g., kernel estimation of the activity rate
density inferred from the catalog; delineation and quantitative
estimations of the seismic potential of active faults in the region.
Development of regional ground-motion model based on deter-
mination of region-dependent geometrical spreading functions
and anelastic path attenuation models using the records from
small earthquakes is also an important and necessary task. The up-
to-date PSHA should include improvement of ground motion
models through more comprehensive treatment of local site
effects. For example, site-specific (based on Vs30 values) and
PGA-dependent amplification factors (see for example, Ref. [53])
may be used together with rock-site hazard estimations. On the
other hand, to reduce the uncertainty associated with the descrip-
tion of the site effect, some additional measures can be employed,
such as the fundamental frequency of a site and frequency-
dependent site amplification. When estimating seismic hazard in
terms of seismic intensity, the immediate relations between
seismic intensity and earthquake characteristics, as well as rela-
tions based on peak velocities or spectral parameters of ground
motion (e.g. Sokolov [69]), should be used for comprehensive
analysis.
Seismic hazard assessment and sensitivity analysis should be
performed also in terms of spectral acceleration (response spectra)
at different vibration periods. The controlling factors for PSHA
results depend significantly on the site location and probability
level of interest. Thus it is important to conduct a full range of
sensitivity analyses as a part of up-to-date PSHA.
Acknowledgments
The work was carried out at the National Center for Earth-
quakes and Volcanoes, Saudi Geological Survey, the Kingdom of
Saudi Arabia. The constructive comments of anonymous reviewers
are gratefully acknowledged.
Appendix A. Supporting information
Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.soildyn.2015.06.
011.
References
[1] Abdalla JA, Al-Homoud AS. Seismic hazard assessment of United Arab Emirates
and its surroundings. J Earthq Eng 2004;8(6):817–37. http://dx.doi.org/
10.1080/13632460409350510.
[2] Abdalla, JA, and Al-Homoud AS. Earthquake hazard zonation of Eastern Arabia.
In: Proceedings of the 13th world conference on earthquake engineering,
Vancouver, B.C: Canada; 2004 p. 1008.
[3] Abrahamson, NA.. State of the practice of seismic hazard evaluation. In:
Proceedings of the GeoEng2000. Invited papers, Melbourne, Australia: Vol.
1; 2000. p.659–685.
[4] Adams RD, Barazangi M. Seismotectonics and seismology in the Arab region; a
brief summary and future plans. Bull Seismol Soc Am 1984;74:1011–30.
[5] Akkar S, Sandikkaya MA, Bommer JJ. Empirical ground-motion models for
point- and extended-source crustal earthquake scenarios in Europe and the
Middle East. Bull Earthq Eng 2014;12(1):359–87. http://dx.doi.org/10.1007/
s10518-013-9461-4.
[6] Al-Amri AM. Seismotectonics and seismogenic source zones of the Arabian
platform. In: Al Hosani K, et al., editors. Lithosphere dynamics and sedimen-
tary basins: the Arabian plate and analogues. Heidelberg, New York, Dor-
drecht, London: Springer Frontiers in Earth Sciences; 2013.
p. 295–316. http://dx.doi.org/10.1007/978-3-642-30609-9-15.
 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
[7] Al-Arifi NS, Fat-Helbary RE, Khalil AR, Lashin AA. A new evaluation of seismic
hazard for the northwestern part of Saudi Arabia. Nat Hazards
2013;69:1435–57. http://dx.doi.org/10.1007/s11069-013-0756-1.
[8] Al-Haddad, M, Siddiqi, GH, Al-Zaid, R, Arafah, A, Necioglu, A, and Turkelli, N.
Seismic hazard and design criteria for Saudi Arabia. In: Proceedings of the 10th
world conference on earthquake engineering, Balkema: Rotterdam; 1992.
p.449–54.
[9] Al-Haddad M, Siddiqi GH, Al-Zaid R, Arafah A, Necioglu A, Turkelli N. A basis
for evaluation of seismic hazard and design criteria for Saudi Arabia. Earthq
Spectra 1994;10:231–58.
[10] Aldama-Bustos G, Bommer JJ, Fenton CH, Stafford PJ. Probabilistic seismic
hazard analysis for rock sites in the cities of Abu Dhabi, Dubai and Ra's Al
Khaymah, United Arab Emirates. Georisk 2009;3(1):1–29.
[11] Al-Malki MA, Al-Amri AM. Seismic zones regionalization and hazard assessment of SW
Arabian Shield and Southern Red Sea region. In: Al Hosani K, et al., editors. Lithosphere
dynamics and sedimentary basins: the Arabian plate and analogues. Springer Frontiers
in Earth Sciences; 2013. p. 317–31. http://dx.doi.org/10.1007/978-3-642-30609-9-16.
[12] Ambraseys NN, Melville CP. The seismicity of Saudi Arabia and adjacent areas:
part A. London: Department of Civil Engineering, Imperial College of Science
and Technology; 1988 ICST/KACST Project, ESEE Report no 88/11.
[13] Ambraseys NN, Adams RD. The seismicity of Saudi Arabia and adjacent areas:
part B. London: Department of Civil Engineering, Imperial College of Science
and Technology; 1988 ICST/KACST Project, ESEE Report no 88/11.
[14] Ambraseis NN, Melville CP. Evidence for intraplate earthquakes in North-
western Arabia. Bull Seismol Soc Am 1989;79:1279–81.
[15] Ambraseys NN, Melville CP, Adams RD. The seismicity of Egypt, Arabia and the
Red Sea: a historical review. Great Britain: Cambridge University Press; 1994.
p. 181.
[16] Aspinall WP. Scientific uncertainties: a perspective from probabilistic seismic
hazard assessments for low-seismicity areas. In: Rougier J, Sparks S, Hill. L,
editors. Risk and uncertainty assessment for natural hazards. New York,
United States of America: Cambridge University Press; 2013. p. 234–74.
[17] Assatourians K, Atkinson GM. EqHaz: an open-source probabilistic seismic-
hazard code based on the Monte Carlo simulation approach. Seismol Res Lett
2013;84:516–24. http://dx.doi.org/10.1785/0220120102.
[18] Atkinson GM, Boore DM. Earthquake ground-motion prediction equations for
Eastern North America. Bull Seismol Soc Am 2006;96:2181–205. http://dx.doi.
org/10.1785/0120050245.
[19] Atkinson GM, Goda K. Probabilistic seismic hazard analysis of civil infrastruc-
ture. In: Tesfamariam S, Goda K, editors. Handbook of seismic risk analysis and
management of civil infrastructure systems. Cambridge, UK: Woodhead
Publishing Ltd.; 2013. p. 3–28. http://dx.doi.org/10.1533/9780857098986.1.3.
[20] Beauval C, Scotti O. Quantifying sensitivities of PSHA for France to earthquake catalog
uncertainties, truncation of ground motion variability, and magnitude limits. Bull
Seismol Soc Am 2004;94:1579–94. http://dx.doi.org/10.1785/012003246.
[21] Bender B. Modelling source zone boundary uncertainty in seismic hazard
analysis. Bull Seismol Soc Am 1986;76:329–41.
[22] Bilal M, Askan A. Relationships between felt intensity and recorded ground-
motion parameters for Turkey. Bull Seismol Soc Am 2014;104:484–96. http:
//dx.doi.org/10.1785/0120130093.
[23] Bommer JJ. Deterministic vs. probabilistic seismic hazard assessment: an
exaggerated and obstructive dichotomy. J Earthq Eng 2002;6:43–73. http:
//dx.doi.org/10.1080/13632460209350432.
[24] Bommer JJ, Crowley H. The influence of ground motion variability in earth-
quake loss modelling. Bull Earthq Eng 2006;4:231–48. http://dx.doi.org/
10.1007/s10518-006-9008-z.
[25] Bommer JJ, Douglas J, Scherbaum F, Cotton F, Bungum H, Fäh D. On the
selection of ground-motion prediction equations for seismic hazard analysis.
Seismol Res Lett 2010;81:783–93. http://dx.doi.org/10.1785/gssrl.81.5.783.
[26] Boore DM, Atkinson GM. 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. Earthq Spectra 2008;24(1):99–138. http://dx.doi.
org/10.1193/1.2830434.
[27] Cadet H, Bard PY, Duval AM, Bertrand E. Site effect assessment using KiK-net
data: part 2 – site amplification prediction equation based on f0 and Vsz. Bull
Earthq Eng 2011;10:451–89. http://dx.doi.org/10.1007/s10518-011-9298-7.
[28] Campbell KW, Bozorgnia Y. NGA ground motion model for the geometric
mean horizontal component of PGA, PGV, PGD and 5%-damped linear elastic
response spectra at periods ranging from 0.1 s to 10.0 s. Earthq Spectra
2008;24(1):139–71. http://dx.doi.org/10.1193/1.2857546.
[29] Castanos H, Lomnitz C. PSHA: is it a science? Eng Geol 2002;66:315–7.
[30] Cauzzi C, Faccioli E, Vanini M, Bianchini A. Updated predictive equations for
broadband (0.01–10 s) horizontal response spectra and peak ground motions,
based on a global dataset of digital acceleration records. Bull Earthq Eng
2015;13(6):1587–612. http://dx.doi.org/10.1007/s10518-014-9685-y.
[31] Chernov YK. Strong ground motion and quantitative assessment of seismic
hazard. Tashkent, USSR: Fan Publishing House; 1989 [in Russian].
[32] Cornell CA. Engineering seismic risk analysis. Bull Seismol Soc Am
1968;18:1583–606.
[33] Cramer ChH, Petersen MD, Reidle MS. A Monte Carlo approach in estimating
uncertainty for a seismic hazard assessment of Los Angeles, Ventura, and
Orange Counties, California. Bull Seismol Soc Am 1996;86:1681–91.
[34] Cramer ChH, Wheeler RL, Mueller CS. Uncertainty analysis for seismic hazard
in the Southern Illinois basin. Seismol Res Lett 2002;73:792–805. http://dx.
doi.org/10.1785/gssrl.73.5.792.
467
[35] Crowley H, Bommer JJ. Modelling seismic hazard in earthquake loss models
with spatially distributed exposure. Bull Earthq Eng 2006;4:249–73. http://dx.
doi.org/10.1007/s10518-006-9011-4.
[36] Deif A, Abou-Elenean K, El-Hadidy M, Tealeb A, Mohamed A. Probabilistic
seismic hazard maps for Sinai Peninsula, Egypt. J Geophys Eng 2009;6:288–97.
http://dx.doi.org/10.1088/1742-2132/6/3/008.
[37] Deif, A, Zahran, HM, El-Hadidy, MS, Bawajeeh, AO, El-Hadidy, SY, and
Mansoub, TA. Sesimic hazard assessment along Haramein high speed rail
project (Makkah-Madinah). SGS NCEV report; 2009b. 150 pp.
[38] Ebel JE, Kafka AL. A Monte Carlo approach to seismic hazard analysis. Bull
Seismol Soc Am 1999;89:854–66.
[39] El-Isa ZH, Al Shanti A. Seismicity and tectonics of the Red Sea and western
Arabia. Geophys J 1989;97:449–57.
[40] Ghebreab W. Tectonics of the Red Sea region reassessed. Earth-Sci Rev
1998;45:1–44.
[41] Giner JJ, Molina S, Jauregui P. Advantages of using sensitivity analysis in
seismic hazard assessment: a case study of sites in Southern and Eastern
Spain. Bull Seismol Soc Am 2002;92:543–54.
[42] Goda K, Hong HP. Estimation of seismic loss for spatially distributed buildings.
Earthq Spectra 2008;24:889–910. http://dx.doi.org/10.1193/1.2983654.
[43] Goda K, Hong HP. Deaggregation of seismic loss of spatially distributed
buildings. Bull Earthq Eng 2009;7:255–72. http://dx.doi.org/10.1007/
s10518-008-9093-2.
[44] Gutenberg B, Richter CF. Seismicity of the earth and associated phenomena.
Princeton: Princeton University Press; 1944. p. 310.
[45] Khair K, Karakaisis GF, Papadimitriou EE. Seismic zonation of the Dead Sea
Transform fault area. Ann Geofis 2000;43(1):61–79.
[46] Khan Z, El-Emam M, Irfan M, Abdalla J. Probabilistic seismic hazard analysis
and spectral acceleration for United Arab Emirates. Nat Hazards
2013;67:569–89. http://dx.doi.org/10.1007/s11069-013-0586-1.
[47] Klügel JU. Problems in the application of the SSHAC probability method for
assessing earthquake hazards at Swiss nuclear power plants. Eng Geol
2005;78(3–4):285–307.
[48] Klügel JU. Error inflation in probabilistic seismic hazard analysis. Eng Geol
2007;90(3–4):186–92.
[49] Klügel JU. Seismic hazard analysis — Quo vadis? Earth-Sci Rev 2008;88(1–
2):1–32.
[50] Krinitzsky EL. How to combine deterministic and probabilistic methods for
assessing earthquake hazards. Eng Geol 2003;70:157–63.
[51] McGuire RK. Deterministic vs. probabilistic earthquake hazard and risks. Soil
Dyn Earthq Eng 2001;21:377–84.
[52] McGuire RK. Seismic hazard and risk analysis. Oakland, CA: Earthquake
Engineering Research Institute; 2004. p. 240.
[53] Mohindra R, Nair AKS, Gupta S, Sur U, Sokolov V. Probabilistic seismic hazard
analysis for Yemen. Article ID 304235. International Journal of Geophysics
2012;vol. 2012:14. http://dx.doi.org/10.1155/2012/304235.
[54] Musson RMW. Determination of design earthquakes in seismic hazard
analysis through Monte Carlo simulation. J Earthq Eng 1999;3(4):463–74.
[55] Musson RMW. PSHA validated by quazi observational means. Seismol Res Lett
2012;83:130–4.
[56] Osman, A. Seismic hazard analysis and development of ground motion
parameters for Makkah Region in Saudi Arabia. In: Proceedings of the 14
world conference earthq. eng. Lisboa: Portugal; 2012.
[57] Pankow KL, Pechmann JC. The SEA99 ground-motion predictive relations for
extensional tectonic regimes: revisions and a new peak ground velocity
relation. Bull Seismol Soc Am 2004;94:341–8.
[58] Park J, Bazzurro P, Baker JW. Modeling spatial correlation of ground motion
intensity measures for regional seismic hazard and portfolio loss estimations.
In: Kanda, Takada, Furuta, editors. Applications of statistics and probability in
civil engineering. London: Taylor & Francis Group; 2007. p. 1–8.
[59] Pascucci, V, Free, MW, Lubkowski, ZA. Seismic hazard and design require-
ments for the Arabian peninsula region. In: Proceedings of the 14th world
conference on earthquake engineering, Beijing, China; October 12–17 2008.
[60] Pedone R, Lombardo P, Diamantidis D. Seismotectonic regionalization of the Red
Sea area and its application to seismic risk analysis. Nat Hazards 1992;5:233–47.
[61] Poirier JP, Taher MA. Historical seismicity in the near and middle east, North
Africa and Spain from Arabic documents (VIIth–Xviith century). Bull Seism Soc
Am 1980;70(6):2185–201.
[62] Power M, Chiou B, Abrahamson N, Bozorgnia Y, Shantz T, Roblee C. An
overview of the NGA project. Earthq Spectra 2008;24(1):3–21.
[63] Rebez A, Slejko D. Sensitivity analysis on the input parameters in probabilistic
seismic hazard assessment. Soil Dyn Earthq Eng 2000;20:341–51.
[64] Roobol, MJ, and ICF Stewart. Cenozoic faults and recent seismicity in the
northwest Saudi Arabia and the Gulf of Aqaba region. Saudi Geological Report
SGA-TR-2008-7; 2009. 35 p.
[65] Sabetta F, Lucantoni A, Bungum H, Bommer JJ. Sensitivity of PSHA results to
ground-motion prediction relations and logic tree-weights. Soil Dyn Earthq
Eng 2005;25:317–27.
[66] Shoja-Taheri J, Niazi M. Seismicity of the Iranian plateau and bordering
regions. Bull Seism Soc Am 1981;71(2):477–89.
[67] Sicbjornsson R, Elnashai AS. Hazard assessment of Dubai, United Arab
Emirates, for close and distant earthquakes. J Earthq Eng 2006;10(5):749–73.
http://dx.doi.org/10.1080/13632460609350617.
[68] Sokolov V. Hazard-consistent ground motions: generation on the basis of
Uniform Hazard Fourier Spectra. Bull Seismol Soc Am 2000;90(4):1010–27.
468 H.M. Zahran et al. / Soil Dynamics and Earthquake Engineering 77 (2015) 453–468
[69] Sokolov VYu. Seismic intensity and Fourier acceleration spectra: revised
relationship. Earthq Spectra 2002;18(1):161–87.
[70] Sokolov V, Chernov YK. Probabilistic microzonation of the urban territories: a
case of Tashkent city. Pure Appl Geophys 2001;158(12):2295–312.
[71] Sokolov V, Wenzel F. Influence of ground-motion correlation on probabilistic
assessments of seismic hazard and loss: sensitivity analysis. Bull Earthq Eng
2011;9(5):1339–60. http://dx.doi.org/10.1007/s10518-011-9264-4.
[72] Sokolov V, Wenzel F. Spatial correlation of ground-motions in estimating
seismic hazard to civil infrastructure. In: Tesfamariam S, Goda K, editors.
Handbook of seismic risk analysis and management of civil infrastructure
systems, 2013. Cambridge, UK: Woodhead Publishing Ltd.; 2013.
p. 57–78. http://dx.doi.org/10.1533/9780857098986.1.57.
[73] Sokolov V, Wenzel F. On the relation between point-wise and multiple-
location probabilistic seismic hazard assessments. Bull Earthq Eng 2015;13
(5):1281–301. http://dx.doi.org/10.1007/s10518-014-9661-6.
[74] Sokolov V, Loh CH, Wen KL. Site-dependent input ground motion estimations
for the Taipei area: a probabilistic approach. Probab Eng Mech 2001;16
(2):177–91.
[75] Sokolov V, Bonjer KP, Wenzel F. Accounting for site effect in probabilistic
assessment of seismic hazard for Romania and Bucharest: a case of deep
seismicity in Vrancea zone. Soil Dyn Earthq Eng 2004;24:927–47.
[76] Sokolov V, Wenzel F, Mohindra R. Probabilistic seismic hazard assessment for
Romania and sensitivity analysis: a case of joint consideraton of intermediate-
depth (Vrancea) and shallow (crustal) seismicity. Soil Dyn Earthq Eng
2009;29:364–81. http://dx.doi.org/10.1016/j.soildyn.2008.04.004.
[77] Spudich P, Joyner WB, Lindh AG, Boore DM, Margaris BM, Fletcher JB. SEA99: a
revised ground motion prediction relation for use in extensional tectonic
regimes. Bull Seismol Soc Am 1999;89:1156–70.
[78] Talebian M, Jackson JA. A reappraisal of earthquake focal mechanisms and
active shortening in the Zagros mountains of Iran. Geophys J Int 2004;156
(3):506–26. http://dx.doi.org/10.1111/j.1365-246X.2004.02092.x.
[79] Tavakoli F, Ghafory-Ashtiany M. Seismic hazard assessment of Iran. Ann Geofis
1999;42(6):1013–21.
[80] Thenhaus PC, Algermissen ST, Perkins DM, Hanson SL, Diment WH. Probabil-
istic estimates of the seismic ground-motion hazard in Western Saudi Arabia.
U.S. Geol Surv Bull 1989;1868:42.
[81] Vakov AV. Relationships between earthquake magnitude, source geometry
and slip mechanism. Tectonophysics 1996;261:97–113.
[82] Vita-Finzi C. Neotectonics at the Arabian plate margins. J Struct Geol
2001;23:521–30.
[83] Wang Z. Seismic hazard assessments: issues and alternatives. Pure Appl
Geophys 2011;168:11–25. http://dx.doi.org/10.1007/s00024-010-0148-3.
[84] Weatherill G, Burton PW. An alternative approach to probabilistic seismic
hazard analysis in the Aegean region using Monte Carlo simulation. Tectono-
physics 2010;492:253–78. http://dx.doi.org/10.1016/j.tecto.2010.06.022.
[85] Zhao JX, Zhang J, Asano A, Ohno Y, Oouchi T, Takahashi T, et al. Attenuation
relations of strong ground motion in Japan using site classifications based on
predominant period. Bull Seismol Soc Am 2006;96:898–913. http://dx.doi.org/
10.1785/0120050122.