10.1007/978 3 319 58154 5

Geotechnical, Geological and Earthquake Engineering
AlainPecker
EzioFaccioli
AybarsGurpinar
ChristopheMartin
PhilippeRenault
An Overview
of the SIGMA
Research
Project
A European Approach to Seismic Hazard
Analysis
Geotechnical, Geological and Earthquake
Engineering
Volume 42
Series Editor
Atilla Ansal, School of Engineering, zyein University, Istanbul, Turkey
Editorial Advisory Board

Julian Bommer, Imperial College London, U.K.
Jonathan D. Bray, University of California, Berkeley, U.S.A.
Kyriazis Pitilakis, Aristotle University of Thessaloniki, Greece
Susumu Yasuda, Tokyo Denki University, Japan
More information about this series at http://www.springer.com/series/6011
AlainPeckerEzioFaccioli
AybarsGurpinarChristopheMartin
PhilippeRenault
An Overview of the SIGMA

Research Project
A European Approach to Seismic Hazard
Analysis
AlainPecker EzioFaccioli
AP Consultant Studio Geotecnico Italiano
Sceaux, France Milan, Italy
AybarsGurpinar ChristopheMartin
Izmir, Turkey Geoter-Fugro
Auriol, France
PhilippeRenault
Swissnuclear
Olten, Switzerland
ISSN 1573-6059 ISSN 1872-4671(electronic)

Geotechnical, Geological and Earthquake Engineering
ISBN 978-3-319-58153-8ISBN 978-3-319-58154-5(eBook)
DOI 10.1007/978-3-319-58154-5
Library of Congress Control Number: 2017940276
Springer International Publishing AG 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors
or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims
in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Le doute est un tat mental dsagrable mais
la certitude est ridicule.
Doubt is an unpleasant mental state but
certainty is ridiculous.
Voltaire (Franois-Marie Arouet,
16941778)
v
Contents
1 Introduction................................................................................................ 1
1.1 Overview oftheProject Organisation................................................. 1
1.2 Object oftheDocument...................................................................... 4
References.................................................................................................... 4
2 General Concepts andPSHA Background.............................................. 5
2.1 Development ofaSeismotectonic Framework forPSHA.................. 5
2.2 Development ofSeismic Sources andLogic Trees forSource
Definition............................................................................................ 6
2.3 Site Specific vs. Regional Study......................................................... 6
2.4 PSHA AFramework forSeismic Source & Ground
Motion & Site Response Characterization.......................................... 8
2.5 Logic Tree Approach andTreatment ofUncertainties........................ 12
2.5.1 Epistemic Uncertainty vs. Aleatory Variability....................... 12
2.5.2 Logic Tree Methodology........................................................ 13
2.5.3 Site Response.......................................................................... 14
2.5.4 Use ofExperts......................................................................... 16
2.6 Interface Issues Between Work Packages........................................... 18
2.7 Common Required Outputs forSeismic Hazard Results.................... 18
2.7.1 Basic Definitions andRequirements....................................... 19
2.7.2 Common Hazard Results........................................................ 20
2.7.3 Additional Parameters............................................................. 22
References.................................................................................................... 23
3 Seismic Source Characterization.............................................................. 25
3.1 Pre-requisites toDevelop theSSC Models......................................... 26
3.2 Database, Earthquake Catalogue, Magnitude Conversions,
Uncertainties onMetadata.................................................................. 28
vii
viii Contents
3.3 Seismic Source Models....................................................................... 32

3.3.1 Diffuse Seismicity VersusIdentified
Seismogenic Structures........................................................... 32
3.3.2 Seismic Source Characterization Framework......................... 33
3.3.3 Area Source, Fault Sources, Gridded Seismicity.................... 34
3.3.4 Lessons Learned Related toSeismic Source Models............. 40
3.4 Occurrence Processes.......................................................................... 41
3.4.1 Poisson Model......................................................................... 41
3.4.2 Characteristic Model............................................................... 42
3.4.3 Time-Dependent Seismicity Models....................................... 42
3.5 Maximum Magnitude andRecurrence Parameters............................. 42
3.5.1 Maximum Magnitude............................................................. 42
3.5.2 Recurrence Parameters............................................................ 44
3.5.3 Lessons Learned..................................................................... 45
3.6 Logic-Tree Implications...................................................................... 47
3.6.1 Logic Tree Approaches........................................................... 47
3.6.2 Efficient Tools fortheLogic Tree Conception
andWeights Assignment......................................................... 49
3.6.3 Verification andQuality Assurance (QA)............................... 53
References.................................................................................................... 53
4 Rock Motion Characterization................................................................. 57
4.1 Empirical Models andPoint Source Stochastic Models..................... 57
4.1.1 Empirical Ground Motion Attenuation Models...................... 57
4.1.2 Point Source Stochastic Models............................................. 62
4.2 Model Selection andCriteria.............................................................. 63
4.2.1 Modelling Criteria................................................................... 63
4.2.2 Tectonic Consistency.............................................................. 64
4.2.3 Site-Conditions Consistency................................................... 66
4.3 Corrections or Modifications ofPublished Models............................ 66
4.3.1 -VS30 (Simulation-Based) Correction..................................... 67
4.3.2 Data-Based Predictions forHard Rock................................... 71
4.4 Standard Deviation ofModel Predictions; Truncation....................... 73
4.4.1 Sigma Truncation.................................................................... 76
4.5 Approaches fortheVertical Ground Motion Component................... 78
4.6 Logic Tree Implications...................................................................... 79
4.7 Lessons Learned fromtheSIGMA Project......................................... 80
References.................................................................................................... 81
5 Site Response Characterization................................................................ 85
5.1 Soil Characterization........................................................................... 85
5.1.1 Determination oftheProfile Natural Frequency f0................. 86
5.1.2 Determination oftheShear-Wave Velocity Profile
andSite Class.......................................................................... 86
5.1.3 Seismic Instrumentation......................................................... 91
5.1.4 Characterization ofNonlinear Soil Properties........................ 92
Contents ix
5.2 Hazard Assessment at theGround Surface......................................... 93

5.2.1 Direct Evaluation fromGround Motion Prediction
Equations (FpG)...................................................................... 96
5.2.2 Generic Site Specific Approaches (HyG)............................... 97
5.3 Completely Site Specific Approaches (HyS)...................................... 101
5.3.1 Linear Numerical Analyses..................................................... 103
5.3.2 Equivalent Linear Numerical Analyses................................... 104
5.3.3 Nonlinear Numerical Analyses............................................... 107
5.4 Treatment ofUncertainties.................................................................. 111
5.4.1 Fully Probabilistic Generic Site Approach (FpG)................... 111
5.4.2 Hybrid Site Specific Approach (HyS)..................................... 112
5.5 Lessons Learned fromtheSIGMA Project......................................... 114
5.6 Additional Topics inGround Surface Hazard Assessment................. 115
5.6.1 Vertical Ground Motion.......................................................... 115
5.6.2 Maximum Ground Motion: Truncation.................................. 116
References.................................................................................................... 117
6 Seismic Hazard Computation................................................................... 119
6.1 Basic Requirements............................................................................ 119
6.2 Interfaces andBoundary Conditions................................................... 120
6.3 Software Packages.............................................................................. 120
6.3.1 PSHA Software....................................................................... 120
6.3.2 Site Response Analysis Codes................................................ 122
6.4 Sensitivity Analysis............................................................................. 123
6.5 Hazard Disaggregation........................................................................ 126
6.6 Additional Engineering Output Parameters........................................ 127
6.7 Selection ofTime Histories................................................................. 128
6.7.1 Selection Based onUHS......................................................... 128
6.7.2 Selection Based onConditional Spectra................................. 129
References.................................................................................................... 130
7 Interfaces Between Subprojects................................................................ 133
7.1 SSC andGMC Interfaces.................................................................... 133
7.2 GMC andSRC Interfaces................................................................... 135
7.3 Single-Station Sigma.......................................................................... 137
7.4 V/H Models forRock andSoil........................................................... 138
References.................................................................................................... 139
8 Probabilistic Seismic Testing andUpdating ofSeismic
Hazard Results........................................................................................... 141
8.1 PSHA Testing Using Acceleration andMacroseismic
Intensity Data...................................................................................... 142
8.2 Bayesian Update ofPSHA.................................................................. 145
References.................................................................................................... 145
x Contents
9 Summary andWay Forward.................................................................... 147

9.1 Seismic Source Characterization........................................................ 147
9.2 Ground Motion Characterization........................................................ 148
9.3 Site Response Characterization........................................................... 148
9.4 Hazard Calculation.............................................................................. 149
9.5 Risk Assessment................................................................................. 149
Annexes............................................................................................................. 151
Bibliography..................................................................................................... 165
Index.................................................................................................................. 169
Acronyms
ASCR Active shallow crustal region

BPT Brownian passage time
CAV Cumulative absolute velocity
COV Coefficient of variation
EMS European Macroseismic Scale
EPRI Electric Power Research Institute
FFS Finite fault simulation
IAEA International Atomic Energy Agency
IDP Intensity data point
GIS Geographic information system
GMC Ground motion characterization
GMPE Ground motion prediction equation
GR parameters Gutenberg-Richter parameters
NRC Nuclear Regulatory Commission (USA)
PDF Probability density function
PGA Peak ground acceleration
PGV Peak ground velocity
PSA Probabilistic safety assessment
PSHA Probabilistic seismic hazard assessment
QA Quality assurance
RP Return period
RVT Random vibration theory
SCR Stable continental region
SHA Seismic hazard assessment
SSHAC Senior Seismic Hazard Analysis Committee
SOF Style of faulting
SRC Site response characterization
SSC Seismic source characterization
US-DOE Department of Energy (USA)
xi
Chapter 1
Introduction
In recent years, attempts have been made to identify and quantify uncertainties in
seismic hazard estimations for regions with moderate seismicity. These studies have
highlighted the lack of representative data, thereby resulting in predictions of seis-
mic ground motion with large uncertainties. These uncertainties, for which no esti-
mation standards exist, create major difficulties and can lead to different
interpretations and divergent opinions among experts. There is a wide consensus
among the scientific and technical community for the need to improve knowledge
so as to better characterize and, ideally, reduce the uncertainties entering in the cal-
culation of seismic ground motion hazard.
To address this situation, in January 2011, an industrial consortium composed of
the French electric power utility (EDF), the French company AREVA, the Italian
electricity company ENEL (Ente Nazionale per lEnergia eLettrica), and the French
Atomic Energy Commission (CEA) launched an international research and devel-
opment program. This program, named SIGMA (SeIsmic Ground Motion
Assessment, http://www.projet-sigma.com), lasted for 5 years and involved a large
number of international institutions.
1.1 Overview oftheProject Organisation
The main objective of the research programme was to establish a framework to be

used in the future to produce stable and robust hazard estimates. Better characteriza-
tion and more stable uncertainty estimation could provide input for the updating of
regulations. It was expected that total uncertainties will be reduced by significantly
lowering epistemic uncertainty, and subsequently, this research programme would
significantly contribute to the following efforts:
Validate, homogenize and stabilize input databases for seismic hazard
calculations;
Springer International Publishing AG 2017 1

A. Pecker et al., An Overview of the SIGMA Research Project, Geotechnical,
Geological and Earthquake Engineering 42, DOI10.1007/978-3-319-58154-5_1
2 1Introduction
Produce accepted and validated methods and calculation tools;

Reduce total uncertainties;
Improve confidence in seismic hazard assessments; and
Foster technical and scientific exchanges among French and other European
organizations.
The programme was organised around five Work Packages, as follows:
WP 1: Improve knowledge of seismic sources
The main goal was to produce a French catalogue of earthquakes that covers both
the historical and instrumental periods.
WP 2: Improve seismic ground motion prediction
The goal was to develop methodologies and analysis tools for predicting seismic
ground motion that are adapted to the French context and contiguous countries,
and which adopt a realistic representation of aleatory and epistemic
uncertainties.
WP 3: Improve local site conditions representation (Site response)
The goal was to develop methods to determine which sites are potentially subject
to local site effects, and to develop appropriate tools to be used in seismic hazard
calculations.
WP 4: Improve seismic hazard models
The intention was to better identify and quantify uncertainties with the goal to
reduce them, particularly the epistemic uncertainties. It was intended to validate
existing methods, and to explore new directions, for testing probabilistic hazard
curves against observations.
WP 5: Improve on characterization and utilization of seismic ground motion
The studies in this work package were aimed to ensure that results of the overall
project fulfil the engineers and designers needs for the design and operation of
various types of facilities. Its goal was to produce methods and tools for the
development of the needed engineering parameter(s) for the earthquake ground
motion.
Figure 1.1 summarizes the general framework of study in the five main Work
Packages.
To help achieve these ambitious objectives, the project management was orga-
nized around four entities (Fig.1.2):
A Steering Committee (COSS) composed of the industrial financial sponsors,
which is charged with identifying strategic orientations and approval of the tech-
nical and scientific choices;
A management committee (COPIL) composed of the Work Package leaders and
the Project Manager;
A international Scientific Committee (CS) to guarantee high quality scientific
research and development; and
An external committee (SHARP) composed of internationally recognized experts
to give the COSS a highly credible scientific assessment.
1.1 Overview oftheProject Organisation 3
Fig. 1.1 Illustration of relationship between the five technical Work Packages
Fig. 1.2 Management flowchart
As mentioned previously, a large number (30) of worldwide academic, research

and professional institutions contributed to the project and their contributions
resulted in the publication of 75 technical reports reviewed by the Scientific
Committee, 40 publications in peer-reviewed journals and numerous communica-
tions in international conferences, workshops and symposia.
A list of all institutions and members of the different committees can be found in
Annex 1. The most important technical reports (deliverables) mentioned in the pres-
ent document are listed in Annex 2 with the published papers.
The total cost of the programme amounted to 7.5 million Euro over a period of
5years.
4 1Introduction
1.2 Object oftheDocument
The main objective of this document is to present, based on the outcomes of the
SIGMA project, lessons learned from conducting a Probabilistic Seismic Hazard
Assessment (PSHA), including site response, for selected areas in France and Italy.
After a general overview of the elements of a PSHA, the document is organized in
chapters closely related to the work packages: Chap. 3 presents the seismic source
characterization (WP1), Chap. 4 the rock motion characterization (WP2), Chap. 5
the site effects (WP3) and Chap. 6 the seismic hazard computations (WP4). Two
important chapters have been added related to interface issues to be faced in PSHA
between the work packages (Chap. 7) and to the testing of PSHA results (Chap. 8).
The final chapter attempts to summarize the lessons learned and to identify the areas
where additional research is needed.
It must be stressed that not all the topics related to PSHA were covered in the
SIGMA project; nevertheless, they will be mentioned in the document for the sake
of completeness.
It is assumed that the reader is familiar with PSHA and, therefore, the basic con-
cepts are not covered in detail in the present document. The interested reader is
referred to general documents for further details, e.g.: IAEA Safety Standard SSG-9
(2010), USNRC Regulatory Guide RG 1.208 (2007) and the EERI monograph by
McGuire (2004).
References
International Atomic Energy Agency (2010) Seismic hazards in site evaluation for nuclear instal-
lations, Specific Safety Guide SSG-9. International Atomic Energy Agency, Vienna
McGuire RK (2004) Seismic hazard and risk analysis, EERI monograph. Earthquake Engineering
Research Institute, Oakland
NRC (2007) Regulatory guide 1.208, a performance-based approach to define the site-specific
earthquake ground motion. U.S.Nuclear Regulatory Commission, Office of Nuclear Regulatory
Research, Washington, DC
Chapter 2
General Concepts andPSHA Background
2.1 Development ofaSeismotectonic Framework forPSHA
The first step in building the PSHA model is the collection of geological, geophysi-
cal, geotechnical and seismological data from published and unpublished docu-
ments, theses, and field investigations. These data are integrated to develop a
coherent interpretation of a seismotectonic framework for the study region. Its size
can vary depending on the purpose. The international practice for a site-specific
study is to distinguish between the investigations at a regional, near regional and site
vicinity level (e.g. 300 km, 25km and 5km radius in IAEA SSG-9, IAEA (2010)).
In order to include all features and areas with significant potential contribution to
the hazard, it may also be necessary to include information in a radius up to 500km
(e.g. for subduction zones). This framework provides the guiding philosophy for the
identification of seismic sources. Furthermore, the framework should address the
important issues that each expert expects to influence the identification and charac-
terisation of seismic sources in the region. The main topics to be addressed in the
seismotectonic framework include:
Use of pre-existing geological structures to provide a basis for defining the pres-
ent and future seismicity.
Tectonic models that are applicable to contemporary processes, the observed
seismicity, and are compatible with seismic sources.
Spatial distribution of the seismicity in three dimensions, and associated focal
mechanisms and their relation to potential seismic sources.
Implications of contemporary stresses and strains (e.g. earthquake focal mecha-
nisms, geodetics, other kinematic constraints) for defining sources.
Use of historical and instrumental seismicity and seismic source delineation to
provide a basis for defining the locations of future earthquake activity.
The following categories of seismotectonic configurations can be distinguished:
Stable continental region (SCR);

6 2 General Concepts andPSHA Background
Active shallow crustal region (ASCR); and

Subduction region.
In Europe two regimes are usually considered and discussed: Active shallow
crustal region (ASCR, southern part) and stable continental region (SCR, northern
part). Within the SIGMA project the study regions were the South-East of France
and Northern Italy. The former region and regime can be considered as being part of
the extended crust (SCR) and weighted accordingly, while the seismically active
zones of Northern Italy can be predominantly classified as ASCR (see Fig. 4.3). The
subduction regime should also be mentioned in this context, but is applicable only
for some special regions in Europe, as e.g. Southern Italy, Greece, Turkey and
Cyprus. Also, the Vrancea seismic zone in Romania can be treated better with a
subduction-related database, although its tectonics are not clear.
The definition of the appropriate seismotectonic model is very important and has
implications on many aspects of the PSHA.Especially the source parameters such
as the maximum magnitude are closely related with the seismotectonic environment
and depend mostly on tectonic metrics (strain rate, etc.). Models for the magnitude
scaling and associated uncertainties are also dependent on the defined environment
and, thus, have an impact on the hazard.
2.2 D
evelopment ofSeismic Sources andLogic Trees
forSource Definition
Using the seismotectonic framework as a basis, the expert team in charge of seismic
source characterization develops its interpretation for the study region (see Sect.
2.5.4). Alternative interpretations of seismic sources (e.g. large regional sources
with spatial smoothing of seismicity versus localised source zones) and alternative
source zone geometries are usually incorporated in the seismic source models as
weighted alternatives using the logic tree methodology. The logic tree framework
allows, especially for the seismic source characterization, to capture the epistemic
uncertainty lying within the various interpretations. The seismic source zone maps
and the supporting calculations of spatial density functions of seismicity, using ker-
nel density estimation, are a part of the seismic source characterization
assessment.
2.3 Site Specific vs. Regional Study
PSHA for critical infrastructures (such as dams, power supply structures, e.g.
nuclear power plants) is usually done on a site-specific basis and cannot directly be
compared to regional studies (such as national seismic hazard maps as used in
design codes). The goal of regional studies is to provide seismic hazard results at a
2.3Site Specific vs. Regional Study 7
regional or national scale based on a uniform approach. Such a result can of course
only be achieved if a common seismological rock layer is defined and simplified
models are defined in order to keep the computation effort manageable. Usually, the
site response cannot be accurately captured in a regional study and, due to the lack
of appropriate soil data, cannot be measured in an adequate and accurate way. The
seismic source characterization models for regional or site-specific studies can be
compared, as the underlying historical and measured seismic data should theoreti-
cally be the same. Nevertheless, seismic sources are not always defined through
seismicity data. In a site-specific study, the detail of investigation increases as we
approach the site, i.e. regional, near regional and site vicinity scales as defined in
IAEA SSG-9. Therefore, the sources can also be different from a regional study in
which only regional scale tectonic data are considered. On the other hand, the
ground motion characterization can also be quite different, since usually no site-
specific (or even regional) attenuation model exists. Therefore, the choices for ade-
quate models to be used for the PSHA can be different depending on the targets of
the study and the resources allocated to deriving adequate models. For example, in
modern PSHA published ground motion prediction models are adjusted to make
them more site-specific. Furthermore, recent site-specific studies make use of the
single-station sigma concept, which requires some local data and very good knowl-
edge of the investigated site. This is usually not the case for a large scale regional
study.
A site-specific study should not primarily rely on the scarce regional data but
should undertake the effort to collect adequate near regional, site vicinity and site
data at appropriate scales. Such data collection is required by nuclear safety stan-
dards (IAEA SSG-9). They are also cost effective and can scale over time depending
on the available resources. Without more knowledge and data, the penalty to pay for
a site-specific study is the acceptance of large uncertainties. Only site specific data
collection to constrain the model space can lead to a reduction of uncertainties.
There is usually a difference in the approach and possibilities for existing versus
new sites. At a new site the collection of data for the ground can usually be carried
out easily, while at an existing site there are constraints to respect. At a regional
level the available data for an existing site might be richer, as equipment has been
deployed and measurements have been carried out since then. At a new site in a
remote location there might be, in the extreme case, no data at all available, as no
infrastructure is nearby. Of course, it depends on the scope of the study and the
available resources, but the approach should be chosen according to specific safety
objectives and implemented in the context of a long-term perspective. Detailed and
extensive data collection can appear costly at the beginning, but will be valuable for
reduction of uncertainties and updates at a later stage.
2.4 P
SHA AFramework forSeismic Source & Ground
Motion & Site Response Characterization
State-of-the-art seismic hazard studies calculate ground motion exceedance proba-

bilities using earth science hypotheses about the origin and characteristics of earth-
quakes in the considered study region. Scientific uncertainty about the causes and
effects of earthquakes in the study region and about the physical characteristics of
potentially active tectonic features lead to uncertainties in the inputs to the seismic
hazard calculations.
These uncertainties have to be propagated through the entire analysis. The result
is a suite of alternative results (in the form of hazard curves), where each hazard
curve is associated with one set of hypotheses and is assigned a weight that repre-
sents the relative merit or credibility of that set of hypotheses. These curves quantify
the seismic hazard for the study area and its uncertainty at the site, and can be used
as basis for decision making. In addition, this suite of hazard curves implicitly con-
tains information about the sensitivity of the hazard results to the various assump-
tions or parameters and about the contributions of these assumptions and parameters
to the total uncertainty in seismic hazard.
Commonly, the method used to calculate seismic hazard at a site is based on the
approach proposed by Cornell (1968, 1971). The approach is well established in the
literature (Der Kiureghian and Ang 1975; McGuire 1976, 1978) and many studies
are based on this mathematical framework. Calculation of the hazard requires speci-
fication of the following three main inputs:
1. Source geometry: the three dimensional geographic description of the seismic
sources. A seismic source is a portion of the earths crust associated with a fault
with a concentration of historic seismicity or with a volumetric defined region of
the earths crust having similar geological and geophysical characteristics that
may be capable of producing earthquakes, thus describing a zone with homoge-
neous seismic potential (without the ability to identify a precise fault in the
source). The geometry of a seismic source relative to the site and a relationship
between rupture size and magnitude, m, determine the conditional probability
distribution of the distance, r, from the earthquake rupture on the i-th source to
the site for a given magnitude: fR(i)|M(i) (r; m). When the hypocentral or epicentral
distance, rather than the rupture or Joyner-Boore distance, are used as the dis-
tance metric, the distance density distribution is no longer conditioned by mag-
nitude. Most modern ground motion prediction equations (GMPEs) make use of
the rupture or Joyner-Boore distance and thus, the distribution of distance is
conditioned by magnitude, which becomes especially important for M > 5.5.
2. Recurrence: the mean annual cumulative rate of occurrence, i, and the magni-
tude density distribution, fM(i)(m), of earthquakes occurring in each source i
(which can be a specific fault or an area describing a region of diffuse seismic-
ity). This characterisation also includes the maximum magnitude that a source
2.4PSHA AFramework forSeismic Source & Ground Motion & Site Response 9
can produce. Usually, magnitude is characterised in the moment magnitude

scale.
3. GMPEs (previously called attenuation relation, function or equation): an algo-
rithm that allows the estimation of ground-motion amplitude (e.g. peak ground
acceleration, spectral acceleration or any other seismic parameter) at the site as
a function of earthquake magnitude and distance. This characterisation consists
of the following three elements: (1) an algorithm for estimating the median
amplitude, (2) an algorithm for estimating the standard deviation, , that
describes the site-to-site and event-to-event scatter in the log[amplitude] of the
observations for the same magnitude and distance, and (3) an algorithm for the
maximum ground motion that can occur (i.e. an amplitude that has zero
probability of being exceeded, given that magnitude and distance). This last
aspect has not been considered in SIGMA.
These inputs are illustrated in Fig.2.1, parts (a) through (c). Figure2.1a shows
the geometry of a seismic source and the distance distribution for a given value of
magnitude. The cumulative probability distribution function (PDF) is assigned the
form of an exponential distribution, derived on a physical basis from the Gutenberg-
Richter relationship; it follows that the density distribution of magnitude, fM(i)(m),
for an area source is typically defined by the doubly truncated exponential distribu-
tion (Fig.2.1b). Seismicity for a source with the exponential magnitude distribution
is completely specified by the minimum magnitude, m0, maximum magnitude mmax,
and recurrence parameters a and b. Parameter a is a measure of seismic activity and
parameter b is a measure of relative frequency of large versus small events. The log
of the cumulative annual rate of events with magnitude m, log[i fM(i)(m)], is propor-
tional to bm for m mmax.
The ground motion is modelled by a ground motion function, as illustrated in
Fig.2.1c. The GMPE is usually expressed in the form log[A] = log[Amedian(M, R)] +
, where A is ground motion amplitude, M is magnitude, R is distance, and is a
normally distributed random variable with mean zero and standard deviation (M,
R), that represents variability in log[A] for a given magnitude and distance. The
maximum ground motion, log[Amax(M, R)], truncates the upper tail of the distribu-
tion of . For the purpose of the calculations, it is useful to express the attenuation
function as the probability GA|M,R (a*;m, r) = P[A a* |m, r]; namely the probability
that the ground-motion amplitude A is larger than a*, for a given m and r.
These three elements (i.e. source geometry, recurrence and GMPE) can be used
to calculate the annual probability of exceeding amplitude a* at the site, which is
expressed as the following summation using the total probability theorem:
Haz ( a * ) = vi GA|m , r ( a*;m;r ) f M ( i ) ( m ) f R ( i )| M ( i ) ( r;m ) dm dr (2.1)

i rm
in which Haz(a*), often written as (a*), is the mean frequency of exceedance of
ground motions, A, at a site and thus, the annual rate of earthquakes that produce
amplitudes A a* at the site. The summation is performed over all seismic sources
Fig. 2.1 Basic inputs for the calculation of seismic hazard: (a) geometry of seismic source and
distribution of distance; (b) magnitude recurrence model; (c) GMPE
i and faults, respectively. The integration on magnitude in the equation considers

only earthquakes with magnitudes greater than a minimum magnitude m0, typically
taken as moment magnitude 4 or 5. Smaller earthquakes are assumed to produce no
damage to engineered structures, regardless of the ground motion amplitudes they
generate (see also Sect. 2.7.3.3).
Both i and f M ( i ) ( m ) are typically defined in terms of magnitudes greater than
m0, although lower magnitudes are normally considered in the determination of the
rate and magnitude distribution. The hazard equation is formulated using the
assumption that earthquakes (most particularly successive earthquakes) are inde-
pendent in size and location. In most seismic hazard applications, primary interest
is focused on computing probabilities for high (rare) ground motions for which the
probability of two or more exceedances of a* in 1 year is negligible. Thus, the quan-
2.4PSHA AFramework forSeismic Source & Ground Motion & Site Response 11
tity on the right side of the equation, which strictly speaking is the annual rate of
ground motions with amplitude A a*, is a very good approximation to the proba-
bility of exceeding amplitude a* in 1 year. It is commonly assumed that earthquake
occurrences in time are represented by a Poisson random process (Parzen 1962). In
fact, this assumption is not necessary, provided the probability of two or more
exceedances of a* in 1 year is negligible.
The calculation in the hazard equation is performed for multiple values of
exceedance amplitudes a*. The result is a hazard curve, which gives the annual
probability of exceedance as a function of a*. This calculation can be performed for
multiple measures of ground-motion amplitude (e.g. peak ground acceleration, peak
ground velocity, and response spectral acceleration at multiple frequencies). As
most GMPEs are formulated for peak ground and spectral accelerations, which are
also useful parameters for engineering purposes, these are the usual measures that
are used for current PSHAs.
It is useful to understand hazard equation using a deterministic perspective as the
starting point. Suppose we want to determine the ground-motion amplitude for an
earthquake of known magnitude occurring within a certain seismic source and at a
certain distance to the site. It is known that we cannot determine this amplitude
exactly, even for fixed magnitude and distance, because the earthquake source can-
not be fully described by a single parameter (magnitude) and wave propagation
through the earths crust cannot be fully described by a single parameter (distance).
To represent the resulting variability in ground motion, a probability distribution in
the form of GA|M,R (a*; m, r) = P[A a* |m, r] is used, i.e. the attenuation equation
is written as a complementary cumulative probability distribution. This is simply a
method for identifying which of the earthquakes lead to ground motions above the
target value a*.
Suppose now that all potentially damaging earthquakes in a certain seismic
source need to be considered. The integral over magnitude and distance in the equa-
tion is just a mathematical approach for sampling all possible earthquakes that may
occur in the given source, while weighting each earthquake by how frequently it
occurs, given the regional seismicity and geology (this weight is expressed by the
joint probability fM(i)(m) fR(i)|M(i)(r;m) dm dr). Multiplication of this integral by the
rate of occurrence i transforms this probability into units of occurrence per year, as
required for design decisions and for comparison with other natural and man-made
hazards. Note that the most notable distinction between the probabilistic and deter-
ministic character of a seismic hazard assessment is the introduction of the rate of
occurrence in the PSHA, rather than just considering and assuming one single sce-
nario. Finally, the summation samples the earthquakes from all of the seismic
sources in the region.
Another useful result is obtained if separate bins are used to accumulate the
rates from earthquakes in different magnitude ranges (e.g. using one bin for magni-
tudes 5.05.4, another bin for 5.55.9 and so forth), and then to divide these accu-
mulated weights by the total hazard. The result, which is called the magnitude
disaggregation of seismic hazard (McGuire 1995; Bazzurro and Cornell 1999)
indicates which magnitude ranges contribute significantly to seismic hazard. Similar
disaggregation results can be obtained for distance and for the number of standard
deviations (). Furthermore, joint disaggregation results can be obtained, where
separate bins for different combinations of magnitude, distance, and are used (see
Sect. 6.5). The consideration and information about becomes especially relevant
when dealing with lower probability levels, as for example when performing seis-
mic PSA.Furthermore, in terms of hazard computation gives an indication about
the sensitivity of the result to a potential truncation of the aleatory variability
(sigma). This step of disaggregation is also very useful to identify the most contrib-
uting seismic scenarios in terms of magnitude-distance couples: this identification
is helpful for example to select the appropriate time series for databases of real
earthquakes consistent with the UHS (see Sect. 2.7.2.4).
2.5 Logic Tree Approach andTreatment ofUncertainties
2.5.1 Epistemic Uncertainty vs. Aleatory Variability
Modern PSHA studies distinguish between two types of uncertainty, namely epis-
temic uncertainty and aleatory variability. Aleatory variability (sometimes called
randomness) is variability that results from natural physical processes. For example,
the size, depth, and time of the next earthquake on a fault and the resulting ground
motion can be considered to be aleatory. In current practice, these elements cannot
be predicted with sufficient confidence, even with collection of additional data.
Thus, the aleatory variability is irreducible without the inclusion of additional pre-
dictive parameters. However, an estimate of the sigma, not the sigma itself, is always
calculated. On the other hand, epistemic uncertainty (often simply called uncer-
tainty) results from imperfect mathematical models, knowledge about faults and
physical processes that produce earthquakes and associate ground motions. In prin-
ciple, epistemic uncertainty can be reduced with advances in knowledge and the
collection of additional data.
Aleatory variability and epistemic uncertainty are treated differently in PSHA
studies. Integration is carried out over aleatory variabilities to obtain a single hazard
curve, whereas epistemic uncertainties result in a suite of hazard curves based on
multiple assumptions, hypotheses, models or parameter values (see Sect. 2.5.2).
Results are presented as curves showing statistical summaries (e.g. mean, median
and fractiles or percentiles) of the exceedance probability for each ground motion
amplitude. The mean and median hazard curves convey the central tendency of the
calculated exceedance probabilities. The separation among fractile curves conveys
the net effect of epistemic uncertainty in the source characteristics and GMPE on
the calculated exceedance probability.
There are epistemic uncertainties associated with each of the three inputs to the
seismic hazard evaluation, as follows:
2.5Logic Tree Approach andTreatment ofUncertainties 13
Uncertainty about the location of causative faults or seismic areas, about the
seismogenic potential of faults or seismic areas and other geological features, as
a result of (1) uncertainty about the tectonic regime operating in the region and
(2) incomplete knowledge of these geological features. There is also uncertainty
about the geometry of these geological features (e.g. faults dip, borders of areal
sources, the exact location of a fault, the thickness of the seismogenic layer or
alternative interpretations of these geometries).
Uncertainty in recurrence is generally divided into uncertainty in maximum
magnitude, uncertainty in the seismic activity rate i, and uncertainty in param-
eter b.
Uncertainty in the GMPEs arises from uncertainty about the dynamic character-
istics of the earthquake source and wave propagation in the vicinity of the site.
This uncertainty is usually large in regions where few strong motion recordings
are available.
Further discussion on the philosophical and practical issues regarding the dis-
tinction between epistemic uncertainty and aleatory variability in PSHA is provided
by NRC (1997).
Often expert judgment needs to be considered, as the available data are scarce,
especially for very rare events with large magnitude. Nevertheless, the importance
of data collection needs to be pointed out here, as expert judgment cannot replace
real measured data or at least should be guided by it. Conceptually, epistemic uncer-
tainty can be reduced through new data and knowledge which should better con-
strain the space of alternatives. This should already be motivation enough for each
sponsor of a study, e.g. the owners of a critical infrastructure. Furthermore, the
collection of local data, which is mandatory for nuclear facilities (see IAEA SSG-9
and USNRC RG 1.208), can help to better understand site-specific phenomena. In
the long term this will also enable making use of non-ergodic PSHA models (see
e.g. Walling (2009), Walling and Abrahamson (2012)) and, thus, allow for an even
more realistic and site-specific hazard assessment.
2.5.2 Logic Tree Methodology
The epistemic uncertainty about the various inputs that affect seismic hazard is
organised and displayed by means of logic trees (Kulkarni etal. 1984; NRC 1997).
This technique is used for seismic source, ground motion and site response charac-
terisation. For example, each node of a logic tree represents a key seismic source
characteristic affecting seismic hazard. This characteristic may be a discrete state of
nature (e.g. are identified faults seismically active?) or a numerical parameter (e.g.
maximum magnitude on a specific seismic source). In the latter case, the continuous
range of values is approximated by a discrete set of values. Each branch emanating
from a node represents one alternative interpretation of the source characteristic
represented by that node. The collection of all branches emanating from a node is
assumed to be a mutually exclusive and collectively exhaustive set of alternative

interpretations. The weight assigned to each branch indicates the expert or expert
teams assessment of the likelihood that this branch represents the true state of
nature, given existing knowledge and data. These weights are conditional on the
values of preceding (i.e. lower) branches in the logic tree.
Each end branch of the logic tree represents a complete description of the inputs
to the PSHA model presented in the section above, for all seismic sources affecting
the site. Associated with each branch tip, there is a weight, calculated as the product
of the weights of all branches followed, and a hazard curve, calculated using the
hazard equation. These hazard curves, together with the associated probabilities, are
used to calculate statistical summaries of the seismic hazard (e.g. mean, median,
and fractile hazard curves), and to estimate the sensitivity of the results to the vari-
ous inputs.
2.5.3 Site Response
The material (soil) beneath a site affects the amplitude, frequency content, and dura-
tion of earthquake ground motion at the surface or in embedded layers. From a first-
order, engineering perspective, the three most important physical phenomena that
affect the amplitude of ground motions at the site are: (1) impedance contrasts
between the reference rock used for the rock calculations and the soil medium, (2)
resonance effects from energy that is trapped between the surface and the bedrock,
and (3) increased damping. In addition, two- and three-dimensional effects are
sometimes considered. At high amplitudes of motion, non-linearity may have a sig-
nificant effect on the elastic properties and damping of the soil.
2.5.3.1 Methods forImplementing Site Response Effects
The most common approach in practice is to perform the PSHA only for rock condi-
tions and then modify the rock amplitudes to introduce the effects of site response.
The key disadvantage of this approach is that it does not directly propagate the
effects of the aleatory variability and epistemic uncertainty in the amplification
factors.
Conceptually, the most straightforward approach for incorporating aleatory vari-
ability and epistemic uncertainty in the site response into a PSHA is to start with a
site-specific (soil) ground motion equation, which may be obtained empirically or
via modelling. Then the hazard equation can be implemented directly for site-
specific amplitudes, using these site-specific GMPEs. Alternatively, the rock ground
motion model and site amplification model can be treated separately. The advan-
tages of the latter approach are that the required expertise and project workload are
decoupled and more combinations of rock motion and site response models are
allowed. The disadvantage is that some of the source information available to the
rock ground motion model is decoupled from the site response and thus, not avail-
able to the site response model (e.g. source location and depth). Various determinis-
tic and probabilistic approaches are used today and SIGMA has investigated and
compared some of them (see Chap. 5).
In the past, Bazzurro (1998), Bazzurro etal. (1999) and McGuire etal. (2002)
have investigated the accuracy of a number of approximate approaches for the intro-
duction of site response effects in hazard results. NUREG CR-6728 (McGuire etal.
2002) compares several approximate approaches to the direct approach and recom-
mended one (denoted as approach 3) that explicitly includes epistemic and aleatory
uncertainty in site amplification, as well as the dependence of site amplification on
the rock input motion and on the dominant earthquake magnitude.
This approach integrates over all rock amplitudes, calculating the probability of
exceedance of specific soil amplitudes, using means and (log) standard deviations
that are functions of magnitude. The resulting equation is:
a
P AS > a = P AF > m,r ,a f A| M , R ( a;m;r ) f M , R ( m,r )dm dr da (2.2)
a
where P[AS a*] is the probability that soil amplitude AS exceeds a*, m is earth-
quake magnitude, r is distance, a is the rock ground motion amplitude, fM,R(m,r) is
the probability that an earthquake equals m and r, P[AF a*/a|m,r,a] is the proba-
bility that soil amplification factor AF exceeds a*/a given m, r and a, and fA|M,R(a;m,r)
is the probability distribution of a given m and r. The formulation recognizes AF as
being dependent on m, r, and a and integrates over all m and r to calculate P[AS
a*]. In effect it is doing the PSHA on a rock-modified-to-soil attenuation equation.
Bazzurro (1998) found this method to be an accurate way to calculate soil hazard.
If, as in McGuire etal. (2002), one recognizes that soil response is governed primar-
ily by the level of rock motion a and magnitude m of the event, the dependence on
distance can be neglected, and the previous equation simplified accordingly. A
detailed practical oriented discussion of the approaches implemented in SIGMA is
documented in Chap. 5.
To evaluate the soil equation (Eq.2.2), the median site amplification factor (SAF)
and the standard deviation of log(SAF) are required. The function fA(a) is obtained
as the negative derivative of the rock hazard curve, and the dominant earthquake
magnitude is calculated by disaggregating the seismic hazard.
Epistemic uncertainty in soil amplification is treated by including multiple soil
amplification models with weights. For calculation of soil hazard, all possible soil
models (P[AS a*|m,a] in the equation above) are combined with all possible rock
hazard models to calculate a family of soil hazard curves, each curve with its own
weight. Statistics on the soil hazard (e.g. mean and fractiles) are determined from
this family of soil hazard curves. There are several advantages to using the soil
equation over the other alternatives. First, the rock hazard curves can be calculated
using region-wide ground-motion equations, rather than developing a set of
equations for each site. Second, site-specific amplification models can be derived
independently of the seismic hazard study, in the context of soil properties and input
motions only. This is how such models are generally applied. Third, this approach
allows explicit evaluation of the impact of epistemic uncertainty in soil amplifica-
tion, which may point to the need for additional data or modelling, rather than com-
bining epistemic uncertainties in soil response with epistemic uncertainties in
ground motion attenuation and dependence on earthquake magnitude. Finally, if
site-specific amplification models are updated at a later date, for example with addi-
tional site data, the soil hazard can be derived (through the equation above) without
repeating the entire seismic hazard calculation. Nevertheless, the implementation of
this approach should be carried out with great care with regard to consistency among
the rock and soil interface parameters. In particular, double counting of uncertain-
ties has been an issue in the past studies (e.g. including the aleatory variability of the
ground motion again in the determination of the soil amplification). Recently, the so
called single-station sigma concept has been introduced, which attempts to remove
the epistemic part of the site response from the evaluation of the aleatory part of the
rock ground motion (Rodriguez-Marek etal. 2013).
2.5.4 Use ofExperts
In order to gather, evaluate and use data in SHA, experts are necessary. Furthermore,
to cover the diversity of scientific interpretations, one approach is to involve a team
of qualified experts. As a SHA requires a multidisciplinary approach the study
makes use of various specialists with different backgrounds and from various fields
(such as geology, seismology, geophysics, geotechnical engineering, statistics, risk
analysis and computer sciences). The use of experts becomes especially relevant in
the context of the quantification of the uncertainties. In this chapter the criteria for
being considered an expert, a very brief example for the expert selection process,
and the general process to be followed in eliciting the evaluations of experts are
described. Experience has shown that, to be credible and useful, technical analyses
such as those performed for the seismic characterisation, ground motion attenuation
and site response must: (1) be based on sound technical information and interpreta-
tions, (2) follow a structured process that considers all available data, and (3) incor-
porate uncertainties (see SSHAC, NRC 1997). A key mechanism for quantifying
uncertainties is the use of formal expert elicitation. Nevertheless, the term elicita-
tion should be used in a broad sense to include all of the processes involved in
obtaining the technical evaluations of multiple experts. These processes include
reviewing available data, debating technical views with colleagues, evaluating the
credibility of alternative views, expressing interpretations and uncertainties in elici-
tation interviews, and documenting interpretations. In this sense, the evaluation pro-
cess begins with the first project meeting and ends with the finalisation of the
evaluation summaries. The elicitation in the context of SHA should not be con-
fused with the classical definition of elicitation used in the social sciences, which is
a strictly speaking a poll. Within a study, experts can have sometimes multiple roles
(e.g. according to the SSHAC guidelines): they may be acting as proponents and
resource experts, as well as evaluators.
2.5.4.1 Guidance Regarding Expert Elicitations
Comprehensive guidance on processes to be followed for expert elicitations has

been set forth in the document Recommendations for Probabilistic Seismic Hazard
Analysis: Guidance on Uncertainty and Use of Experts by the Senior Seismic
Hazard Analysis Committee (SSHAC) (NRC 1997). The guidance was developed
under sponsorship of the US Nuclear Regulatory Commission (US-NRC), the
Electric Power Research Institute (EPRI), and the US Department of Energy
(US-DOE). The study was conducted with the purpose of drawing on the experience
gained from expert elicitation projects, particularly those conducted for nuclear
power plants in the central and eastern United States, and developing a consensus
position regarding acceptable methodologies. In reviewing the differences in PSHA
estimates conducted by different groups for individual sites, the SSHAC study con-
cluded that the differences were largely due to procedural differences in the manner
in which the PSHA was conducted. Hence, it is concluded that the procedural steps
are as important as the technical analyses that comprise a PSHA.
A basic principle defined by SSHAC (NRC 1997, p.21) is that: The underlying
basis for the inputs [to a PSHA] must be the composite distribution of views rep-
resented in the appropriate scientific community. Expert judgement is used to repre-
sent the informed scientific communitys state of knowledge.
As noted in NUREG/CR-6372 (NRC 1997, p.21), the goal of any formal expert
elicitation process is: To represent the centre, the body, and the range of technical
defensible interpretations that the larger informed technical community would have
if they were to conduct the study. In this context, informed means, hypotheti-
cally, that all, in the community, have a full understanding of the site-specific tech-
nical details. Thus, the experts become informed after having been exposed to the
site-specific data and models and having evaluated them. This happens through a
number of structured workshops in which data are presented, alternative models and
interpretations are debated, and feedback is given. Interaction among the experts
should be encouraged and dedicated elicitations should be conducted in form of
interviews. Since the publication of the SSHAC guidelines in 1997 a number of
PSHA studies have been carried out using this framework and the community has
developed some more practical guidance based on the experience gained from those
projects (NRC 2012).
Nevertheless, it should be mentioned that there are other approaches to collect
expert opinions and integrate them to a PSHA model. The SIGMA project was not
carried out under the SSHAC framework but made use of elements of it for building
site-specific hazard estimates.
2.6 Interface Issues Between Work Packages
A PSHA is fundamentally a multidisciplinary study that involves various experts

from different fields. As the work is often distributed in different work packages that
are usually done by different experts or expert groups, the issue of adequate inter-
faces needs to be addressed (see also Chap. 7).
The source characterization logic trees contain the complete earthquake defini-
tions (magnitude range, style of faulting, depth distribution, dip angles, magnitude-
area scaling). The ground motion experts need to consider these definitions when
evaluating the candidate ground motion models and address any significant differ-
ences between the source and ground motion characterization. Other interface
issues to be addressed are, for example, distance conversion relationships, focal
depth and mechanisms, which can be related to software implementations.
To improve the consideration of the ground motion and site-response interface,
this interface should be supported by a dedicated discussion or workshop. The rock
(VS30) to be used for the rock ground motion calculation can be based on the site-
specific shear-wave velocity profile at the site. The common understanding of this
interface layer, its depth, seismological and soil mechanical properties needs to be
developed.
Any interface issues with the end user engineers or PSA analysts should be eval-
uated at the very beginning of the project in order to ensure that all necessary param-
eters are addressed and captured. This is especially true with respect to any additional
ground motion parameter (e.g. duration, peak ground velocity, average spectral
acceleration over a given frequency band) that will be needed, e.g. for the fragility
calculations.
SIGMA has tried to consistently address these interface issues in order to improve
the hazard estimation. As will be discussed later in the report, SIGMA has, for
example, investigated and compared the different approaches to estimate the so-
called control point and to assess the impact of the choice of time histories for the
site response.
2.7 Common Required Outputs forSeismic Hazard Results
The required outputs depend on the specific intended use of the PSHA.There are
different perspectives that deserve to be mentioned and the study and associated
output should always be discussed and defined among all stakeholders at the very
beginning. First, the purpose of the PSHA could be for design or verification pur-
poses, or for input to a seismic probabilistic safety assessment. Second, the approach
will depend on if the output is applied to a new or existing structure, system or
component. Thus, the provided requirements below have to be understood as the
outcome of best practice and they are only indicative (see also IAEA SSG-9).
2.7Common Required Outputs forSeismic Hazard Results 19
2.7.1 Basic Definitions andRequirements
2.7.1.1 Control Point
For a site-specific hazard assessment, the ground motion is often presented for the
site-specific soil condition at different depths below the surface. Soil hazard curves
can, for example, be given at ground surface and foundation level of the turbine
building or reactor building, in the case of a nuclear power plant. Commonly, the
ground motions at depth are given as outcropping motion. The definition of the
project specific elevation levels should be made together with the engineers so as to
be consistent with their requirements (e.g. for soil-structure interaction
computations).
2.7.1.2 Components ofMotion
Usually, the hazard is computed for the geometric mean of the two horizontal com-
ponents and for the vertical component. This is because most GMPEs use this defi-
nition for the horizontal motion. The standard deviation of the horizontal
component-to-component variability, which can be used for the development of
time histories, can be added back in afterwards if necessary.
2.7.1.3 Vibration Frequencies forHazard Analysis
As the hazard computation is made for given frequencies, the project needs to define
the frequencies required to define the uniform hazard spectrum (UHS). The choice
of the relevant frequencies should be made together with the engineers who will
subsequently use the results of the PSHA for design purposes or for probabilistic
safety assessments. To compute the rock hazard results over a representative fre-
quency range, the following nine spectral frequencies can, e.g., be defined: 0.5 Hz,
1 Hz, 2.5 Hz, 5 Hz, 10 Hz, 20 Hz, 33 Hz, 50Hz and 100Hz (which can generally
be interpreted as PGA).
The soil hazard should be computed at the nine spectral frequencies given above
for the rock hazard plus one or more additional frequencies, so that the site-specific
soil resonance is adequately represented. Depending on the number of frequencies
that have been used to determine the soil amplification, it may be worthwhile calcu-
lating the soil hazard at more frequencies in order to capture the peaks and valleys
of the soil response spectrum.
2.7.1.4 Lowest Annual Probability ofExceedance
Today, in order to satisfy the requirements and needs of modern probabilistic safety
assessments, the hazard curves should be defined down to an annual probability of
exceedance (APE) of 107/year (or 108/year). The annual rate of exceedance of
107/year stated here is a common reference value, e.g. if seismic PSA is necessary
for the plant. The actual underlying hazard computation provide much lower prob-
abilities, but the display of results can stop at this lowest annual probability of
exceedance for most end users. Nevertheless, it should be noted that the smallest
annual frequency of exceedance of interest will depend on the eventual use of the
PSHA and the 107/year have to be understood as indicative in the following.
2.7.2 Common Hazard Results
2.7.2.1 Seismic Hazard Curves forReference Rock Site Condition
The reference rock site seismic hazard for the horizontal components for each fre-
quency at a site should be supplied for ground motion levels, e.g. between 0.01 and
10 g and also higher accelerations until the mean annual hazard level falls below
107/year in order to achieve a reasonable sampling (e.g. by at least 30 values) of the
hazard curve down to low annual probabilities of exceedance.
The range (epistemic uncertainty) in the reference rock site hazard curves should
be presented for the horizontal and vertical components for each frequency in plots
of the 5%, 16%, 50%, 84% and 95% fractiles and the mean hazard. Numerically,
tables of the fractiles at 99 equally-weighted levels should be provided in order to
serve as adequate input for the probabilistic safety assessment. The fractile curves
are depicted (extended to small annual probabilities of exceedance respectively)
until they reach the PGA level where the mean annual hazard has the annual prob-
ability of exceedance of 107/year. As mentioned above, depending on the final use,
the very low annual probabilities of exceedance and fractiles may not be needed
(e.g. when defining design values).
2.7.2.2 Seismic Hazard Curves fortheSoil Site Condition
The soil site hazard for the horizontal and vertical components for each frequency
at each plant should be supplied at the same ground-motion levels as specified in the
section above until the mean annual hazard level falls below 107/year.
2.7Common Required Outputs forSeismic Hazard Results 21
2.7.2.3 Epistemic Uncertainty intheSoil Hazard Curves
The range (epistemic uncertainty) in the soil site hazard curves should be presented
for the horizontal and vertical components for each frequency and each plant as fol-
lows: Plots of the 5%, 16%, 50%, 84% and 95% fractiles and the mean hazard (sup-
ported by tables of the fractiles).
2.7.2.4 Uniform Hazard Spectra
Uniform hazard spectra (UHS) for the horizontal and vertical components for the
soil site condition are usually computed and plotted for the following annual exceed-
ance frequencies, provided that no extrapolation of the hazard curves to these levels
is necessary: 102/year, 103/year, 104/year, 105/year, 106/year and 107/year. The
value of 2.1 103/year (= return period of 475 years) can be added if there is an
interest to compare the hazard value with that given in the national building code.
The ground motions for the UHS are determined by linear interpolation in log-log
space between the defined ground motion levels. The range (epistemic uncertainty)
in the UHS is shown in terms of the mean and the 5%, 16%, 50%, 84%, and 95%
fractiles.
2.7.2.5 Disaggregation
To be consistent with the representation of the UHS, the mean horizontal compo-
nent rock hazard should be disaggregated in terms of magnitude, distance, and
(number of standard deviations) at the following levels of annual exceedance fre-
quency: 102/year, 103/year, 104/year, 105/year, 106/year and 107/year. The dis-
aggregation plots (Sect. 6.5) are generated for the following representative
frequencies 1 Hz, 5 Hz, 10Hz and 100Hz. A disaggregation for additional frequen-
cies can be made on a structure-specific basis. The disaggregation is used to deter-
mine the controlling earthquakes in terms of magnitude, distance and , which is
usually used as guidance to select or develop time histories for engineering pur-
poses. In order to consistently select or develop time histories that are representative
of the hazard, the notion of is important, as the median prediction is not necessar-
ily the one that contributes the most. Based on the engineers will know if an aver-
age earthquake (thus, a mean prediction with = 0) should be assumed for the site
or if the very unusual earthquake (thus, above average with > 0, e.g. 2 or 3)
dominates.
2.7.3 Additional Parameters
2.7.3.1 Ground Motion Duration
Ground-motion durations need to be defined if different earthquake scenarios have

to be developed for engineering purposes. There are many available duration mod-
els (including global models for large magnitudes and region-specific models for
small magnitudes). The models applicable in the context of the study need to be
determined with respect to applicability to large magnitudes in the study region
(thus, usually with magnitudes up to 8) and consistency with the used ground
motion prediction models. Today, the GMPEs used to estimate spectral acceleration
(the most commonly used intensity parameter) do not provide a duration as explicit
parameter, but they are build including implicitly a duration model based on the
earthquakes used to develop the GMPE.The end user of a GMPE expressing spec-
tral acceleration usually overlooks this additional parameter and it has been seen in
the past that the regional duration models have a strong effect on the resulting
response spectra. Thus, for the sake of consistency between the proposed duration
model and the study results, specific thoughts on this issue need to be devoted from
the very beginning of a study.
2.7.3.2 Peak Ground Velocity
In the past, there were few ground motion models for peak ground velocity (PGV),
so simplified scaling relations between PGV and spectral acceleration were often
used to estimate PGV.Many, but not all, of the new ground motion models include
models for PGV.The PGV hazard can be computed on the request of the engineers
if necessary, but it requires a full re-computation of the hazard for this intensity
measure.
2.7.3.3 Cumulative Absolute Velocity
A typical value adopted for the minimum magnitude of the hazard integral is Mw =
5.0 (assumption commonly used for nuclear plants). Based on the observation that
the Cumulative Absolute Velocity (CAV) is a parameter much better correlated with
the observed damages than other ground motion parameters, the CAV-filtering
approach was introduced by EPRI (2006) to consider, in the ground motion assess-
ment, only the contribution of seismic sources having a significant influence on the
structures. When such an approach is applied, a lower minimum magnitude is con-
sidered (e.g. Mw 4) and the CAV filtering is applied for sources with Mw between 4
and 5.5.
References 23
References
Bazzurro P (1998) Probabilistic seismic demand analysis. PhD thesis, supervised by C.A.Cornell,
Stanford University, Palo Alto, CA
Bazzurro P, Cornell CA (1999) Disaggregation of seismic hazard. Bull Seismol Soc Am
89(2):501520
Bazzurro P, Cornell CA, Pelli F (1999) Site- and soil-specific PSHA for nonlinear soil sites. In:
Proceedings of 2nd international symposium on earthquake resistant engineering structures.
Proc. ERES99, 1517 June, Catania, Italy
Cornell CA (1968) Engineering seismic risk analysis. Bull Seismol Soc Am 58(5):15831606.
Erratum: 59(4), 1733
Cornell CA (1971) A probabilistic analysis of damage to structures under seismic loads. In:
Howells DA etal (eds) Chapter 27 of dynamic waves in Civil engineering. Wiley, London
Der Kiureghian A, Ang AH-S (1975) A line source model for seismic risk analysis, University of
Illinois Technological Report, UILU-ENG-75-2023. University of Illinois, Urbana. 134p
EPRI (Electric Power Research Institute) (2006) Use of Cumulative Absolute Velocity (CAV) in
determining effects of small magnitude earthquakes on seismic hazard analyses,
EPRI-TR-1014099. Electric Power Research Institute, Palo Alto
lations, Specific Safety Guide SSG-9. International Atomic Energy Agency, Vienna
Kulkarni RB, Youngs RR, Coppersmith KJ (1984) Assessment of confidence intervals for results
of seismic hazard analysis. In: Proceedings of 8th world conference on earthquake engineering,
vol 1, San Francisco, pp263270
McGuire RK (1976) FORTRAN computer program for seismic risk analysis, U.S. Geological
Survery, Open File Report 7667, 69. U.S. Department of the Interior, Geological Survey,
Menlo Park
McGuire RK (1978) FRISK: computer program for seismic risk analysis using faults as earth-
quake sources, U.S.Geological Survery, Open File Report 78-1007. U.S.Department of the
Interior, Geological Survey, Menlo Park
McGuire RK (1995) Probabilistic seismic hazard analysis and design earthquakes: closing the
loop. Bull Seismol Soc Am 85:12751284
McGuire RK, Silva WJ, Costantino CJ (2002) Technical basis for revision of regulatory guidance
on design ground motions: hazard- and risk-consistent ground motion spectra guidelines,
NUREG/CR-6728. U.S.Nuclear Regulatory Commission, Washington, DC
NRC (1997) Recommendations for probabilistic seismic hazard analysis: guidance on uncertainty
and the use of experts, NUREG/CR-6372. NRC, Washington, DC
NRC (2012) Practical implementation guidelines for SSHAC level 3 and 4 hazard studies,
NUREG-2117, Rev. 1. NRC, Washington, DC
Parzen E (1962) Stochastic processes. Holden-Day Publishers, San Francisco
Rodriguez-Marek A, Cotton F, Abrahamson N, Akkar S, Al Atik L, Edwards B, Montalva GA,
Dawood H (2013) A model for single-station standard deviation using data from various tec-
tonic regions. Bull Seismol Soc Am 103(6):31493163
Walling MA (2009) Non-ergodic probabilistic seismic hazard analysis and spatial simulation of
variation in ground motion. PhD thesis, University of California, Berkeley
Walling MA, Abrahamson NA (2012) Non-ergodic probabilistic seismic hazard analyses. In:
Proceedings of the 15th world conference on earthquake engineering 15 WCEE, Paper 1627,
Lisboa
Chapter 3
Seismic Source Characterization
Seismic sources characterization (SSC) relies on the interpretation and integration

of a variety of parameters and data primarily collected in a geological, geophysical,
geotechnical and seismological database (D4-41, Carbon etal. 2012). It has three
fundamental objectives:
The identification (location and geometry) of all seismic sources contributing to
the total hazard at the site of interest. According to the seismotectonic context of
the site, and to the considered hazard return period, this involves developing the
database within a radius of a few hundred kilometres around the site and to con-
sider different scales of analysis especially when individual faults are
considered.
The characterization of the seismic activity with a large number of uncertain
parameters: the maximum magnitude of these seismic sources, their activity rate
and the models for seismicity distribution.
The consideration of all inherent epistemic and aleatory uncertainties and the
influence of the different uncertainties in the hazard results.
The main objectives of the SIGMA project were to better understand the sources
of uncertainties and their impact on the seismic hazard at the regional scale and at
the site scale, and to identify the actions to strengthen the confidence in the mean
and median hazard estimates and to reduce the dispersion in the measure of the
uncertainty, e.g. in the distance between the lower (5%, 16%) and upper fractiles
(84%, 95%) of the hazard curves resulting from the logic-tree.
While the SSC is often seen or considered as the sole responsibility of the geolo-
gists, an important lesson gained from the SIGMA project is that the management
of the interfaces among the different disciplines is of particular importance to
develop a robust and reliable seismic hazard assessment and is a significant indica-
tor of the degree of confidence in the results of the PSHA.
The two PSHA case studies for France and Italy have demonstrated that the
development of a complete seismic hazard assessment model requires the interven-
tion of specialists from various disciplines (geology, quaternary geology, seismology,

26 3 Seismic Source Characterization
geophysics, numerical simulation, statistics, soil dynamics, and mathematics).

From the identification and characterization of the seismic sources to the modelling
of site response, the communication and exchanges between the different specialists
and actors involved at each step of the PSHA are not always easy, and the integra-
tion of all the components and outputs of the different steps of a PSHA into a coher-
ent model, dealing in an appropriate manner with the uncertainties (e.g. identifying
the uncertainties without double or multiple counting and avoiding inclusion of
unnecessary branches in a logic-tree), remains a real challenge.
The SIGMA project has clearly identified this issue and was organized in differ-
ent work-packages with identification of the interactions between them. A signifi-
cant number of R&D actions were conducted within WP-1 to provide the inputs for
the SSC.Some of them were launched with the objective to be directly transferred
into operational application (site-specific hazard assessment) while others were
more dedicated to academic work with the objective to improve scientific knowl-
edge, to better characterize parameters required in a seismic hazard analysis or to
measure their impact on the ground-motion assessment.
The lessons learned during the SIGMA project are considered in this chapter,
under two perspectives:
What are the actions, tasks, efforts that should be considered when conducting a
PSHA after SIGMA compared with a PSHA developed before SIGMA; and
How to identify and manage the uncertainties that influence the hazard results
given the objectives of the PSHA.
3.1 Pre-requisites toDevelop theSSC Models
At the onset of the SIGMA project, all the R&D tasks of WP-1 and WP-5 were
identified by a panel of institutions, research centres and consulting engineers, and
implemented by several research teams without possible consideration of the sim-
plifications needed for a PSHA calculation model. These simplifications of the vari-
ous interpretations and geological models developed by the geologists, provided in
the hazard input documents used by the hazard calculation team, are needed to
render the PSHA calculations operational.
While it is obvious that not all the research work can be easily transferred into
practical application, a site-specific PSHA must rely, especially when developed for
highly critical facilities, on unambiguous parameters that describe and reflect the
models elaborated by the team in charge of the SSC development. To avoid any
misunderstanding between this SSC team and the other teams (Ground motion and
hazard calculation teams), the interaction process between them should focus sys-
tematically on the preparation of the relevant sections of the hazard input docu-
ments with the aim of having a common understanding of the seismic source models
to be used for the hazard computation, making sure that there is a common under-
standing of the seismic source models and of obtaining confidence that the
3.1 Pre-requisites toDevelop theSSC Models 27
c alculation team adequately describes the models and ingredients in a complete and
unambiguous way.
To facilitate the interaction between the different components of the PSHA and
the consideration of the interfaces, as well as to determine which level of effort has
to be devoted to the development of the model, several requirements can be identi-
fied in carrying out the future PSHA, in particular:
Clearly define the objective of the PSHA and the deliverables that are expected
to be used by others. The end users are not the developers of the PSHA but most
often structural engineers in charge of the design of a new building or the retrofit
of existing buildings. If the needs of the end-users are explained, this may orien-
tate the work and tasks to be considered in the development of the models. For
instance, if it is known that the structural periods of interest are large, care will
be taken by the team in charge of the SSC to focus on the delineation and char-
acterization of seismic sources (and specifically of identified faults) that control
the long period content of the ground motion at specific return periods. The
objectives of the PSHA study are also a guide to the level of effort in the identi-
fication and propagation of uncertainties and in the scale to consider. This is
because the treatment of the uncertainties will not be the same whether the aim
is to provide ground motion estimates for low or for high probability of exceed-
ance or because the level of efforts will be different if the objective is to produce
regional hazard maps or to conduct a site-specific assessment for design pur-
poses. In the last case, the acquisition of data to identify fault sources at the site-
vicinity scale, the consideration of local effects (soil amplification, directivity),
the GMPEs adjustments, and single-station sigma become standard tasks that are
not considered in a regional study.
Identify the main relevant interfaces between the different components of the
PSHA.The interfaces between the various disciplines should be considered at an
early stage and at the various steps of the PSHA model development. This is the
case between the SSC (Seismic Source Characterization) and GMC (Ground
motion characterization) where the actors of SSC must provide the parameters
required in the selected GMPEs and vice versa the GMPEs developers must
seek, in the development or adjustment of their models, to account for the seis-
mic sources characteristics identified in the SSC model. SSC and GMC develop-
ers must also interact with the hazard calculation team to verify that their models
are properly included in the PSHA software.
Conduct preliminary sensitivity analyses to identify the seismic sources and
parameters that control the hazard at the site of interest, given: (1) the objectives
of the PSHA and, (2) the location and distance between the site and the identified
seismic sources. This allows focusing on the resolution of the most relevant com-
ponents of the PSHA, on the acquisition of data to characterize the controlling
sources and generally to simplify the model. As the calculation tools nowadays
allow managing complex logic trees, a key lesson learned from SIGMA is that
the quality of a future PSHA will greatly benefit from sensitivity analyses and
tests that are conducted at the beginning of the PSHA to identify which elements
of the models exert the most significant control on the hazard results, at the speci-
fied return periods (and sometimes spectral periods) of interest.
3.2 D
atabase, Earthquake Catalogue, Magnitude
Conversions, Uncertainties onMetadata
Prior to the data integration and to the development of the SSC logic tree, the com-
pilation of all existing data and models that are relevant to characterize the seismic
context of the site or region of interest, constitutes the first stage of the PSHA.The
objective of the compilation is to provide a record and traceability of all the data
used during the project, and a description on how the data are interpreted and inte-
grated to develop the SSC models.
As described in D4-41 and D1-27, different scales must be considered from
regional to site scale, consistent with the rules or regulation adopted for the project.
The database is usually developed at four scales (regional, near regional, site vicin-
ity, and site) as described in the IAEA (2010)SSG-9 guide. When the purpose is a
site-specific assessment for NPs or other critical facilities, new data should be
acquired or generated through site investigations such as seismic profiles, local seis-
mic or accelerometric networks and geotechnical investigations. For other purposes,
new data acquisition would also represent a significant added value for the project.
The SSC model development is based on the coherent interpretation (integration
phase) of the data compiled in this database (e.g. existing bibliographic information,
databases already developed at regional scales, international publications, Ph.D.
theses and new data collected for the project).
A comparison of the PSHA conducted in the two regions considered within
SIGMA (South-East quarter of France and the Po Plain in Italy) is relevant to
emphasize the impact on the logic trees of new data: the more data are available or
newly generated (in case of the Po plain, benefiting from an extensive fault data-
base, see D1-27 and D1-67, dense seismic networks, see D2-72, and in situ mea-
surements), the more the logic tree can be refined and the more new generation
approaches (non-ergodic models, rupture simulations, seismic sources modelled as
faults rather than as area sources, host-to-target adjustments and single-station
sigma) can be implemented, see D4-94. One direct consequence is that the more
efforts are paid to the generation of new data, the more it becomes possible to better
capture/constrain the uncertainties in the seismic sources and more the predictive
models are accurate.
For nuclear sites the seismic hazard assessment must be regularly updated over
time and the operator must keep, for audit and safety purposes, a full traceability of
all the data (from regional to site scales) collected and generated during the life time
of the installation. Consequently, the database should be progressively enriched
with the inputs of more local data, like those retrieved during possible additional
3.2 Database, Earthquake Catalogue, Magnitude Conversions, Uncertainties 29
field investigations, and with the data acquired from monitoring activities during the
plant life.
The modern way to develop the database is to use a GIS database, where all the
data are georeferenced and all the objects included in the database described by a
series of attributes characterizing the geological, geophysical, geotechnical and
seismological information.
Keeping in mind the interfaces between the disciplines already mentioned in
Sect. 3.1, the database development should not be the sole responsibility of the SSC
team but must also include the information pertaining to the ground motion predic-
tion and all parameters that should be deemed necessary by the hazard calculation
team.
A description of such a database is available in D4-41. If the information is pri-
marily used to identify any earthquake-related geohazards that may affect the region
and to develop the seismic sources models, it must also describe the uncertainties
associated with each parameter, to allow the hazard calculation team to quantify the
influence of the uncertainties on the hazard estimates.
The databases developed in the two pilot regions (D4-41, D4-94) distinguish the
uncertainties respectively ascribed to area sources and to fault sources, which should
be considered in any PSHA.For area sources the uncertainties associated with the
following are considered:
The earthquake parameters;
The boundaries delimiting volumes of the earth crust showing a homogeneous
deformation pattern under the present stress field;
Activity rate/km2;
The nature of deformation/style of faulting, considering when necessary single
or multiple mechanism of deformation;
The thickness of the seismogenic crust;
The maximum magnitude; and
The parameters of the frequency-magnitude distribution in each source.
When considering fault sources additional uncertainties are considered (D4-94,
Ameri etal. 2015) and associated to:
The 3D geometry of the faults;
The rupture scenario (segmented/unsegmented);
The slip rates;
The characteristic magnitude; and
The background activity prevailing in the area source where the fault is located.
Assessing the completeness of the database available to develop the SSC models
is basically an indicator of the strategy that can be adopted to develop the logic tree
and of the uncertainties that can be introduced in the PSHA model. Again, the com-
parison between the two case studies indicates that the amount of data (existing or
specifically collected) is a significant parameter to determine the degree of refine-
ment that it is possible to introduce in the model. In the French case, it was very
clear that the limitation of the fault database did not allow developing a fault model
covering the entire South-East region with a sufficient level of confidence. The strat-
egy was much more to introduce epistemic uncertainties in area source models con-
sidering different criteria to delineate the seismic sources (structural heritage, major
domains of deformation or tentative identification of fault systems). On the con-
trary, the Italian fault database [DISS, INGV; Burrato etal. (D1-27); Valensise (D1-
67)] and the new information acquired after the recent Emilia earthquake sequence
of May and June 2012, were used to develop a reliable model including fault sources
and paying less attention to the uncertainties associated with the area sources.
Among the data, the preparation of the earthquake catalogue (including different
time scales) is identified as a crucial step of the SHA because the data are used at
several stages of the analysis (determination of maximum observed and assumed
magnitude, distribution of activity rates, deformation pattern and so forth). Such a
step relies on the compilation, interpretation and integration of multiple sources,
including pre-historical, historical and instrumental seismicity.
The strategy deployed between the initial and the refined PSHA in the two case
studies was significantly different, even if the common objective was to develop
more reliable estimates of Mw. A significant difference in the two regions is that a
number of earthquakes of magnitudes between 4.0 and 6.0 occurred in Italy since
the 1960s and hence benefit from direct determinations of Mw, while in France most
of the earthquakes in this range of magnitudes are only known from macroseismic
observations.
For the Po plain, the catalogue used at the beginning of the project was CPTI11
(Rovida etal. 2011), covering earthquakes up to 2007, developed for Italy as the
national parametric earthquake catalogue. Sensitivity analyses were conducted in
the second version of the PSHA (improvement of the magnitude conversions) and
the decision was taken to only consider Mw magnitudes above a minimum cut-off
level of Mw 4.5 to avoid magnitude conversions for small events, deemed less suit-
able to calculate the Gutenberg-Richter parameters.
In the French case, no homogenized catalogue was available at the beginning of
the project, each institute developing its own rationale to convert the historical
earthquakes defined in MSK intensities and the instrumental earthquakes into Mw.
The initial catalogue was developed based on conversions among magnitude scales
and encompassed Mw above 2.0in order to include small magnitudes from which to
calculate the Gutenberg-Richter parameters in low activity areas. For the updated
model, a new catalogue is considered taking benefit from the national instrumental
Si-Hex catalogue in Mw (D1-110; Denieul etal. 2015) and from the homogenized
Mw derived from the exploitation of macroseismic intensities and new empirical
macroseismic predictive equations (D1-147).
An important lesson from SIGMA (French case) was that the effort for the devel-
opment of a homogenized catalogue from scratch combining instrumental and mac-
roseismic data was significantly underestimated, the catalogue being only finalized
at the end of the project. Significant progress has been accomplished but it remains
as an objective to develop a unique catalogue homogenized in Mw, and prepared
(with consistency between instrumental Mw and Mw derived from macroseismic
3.2 Database, Earthquake Catalogue, Magnitude Conversions, Uncertainties 31
data, completeness periods assessment and declustering) to refine the PSHA in the
region of interest. More specifically:
Source parameters have been significantly improved for the instrumental events
with the joint effort of the Si-Hex project: EOST developed a relocation and a Mw
assessment based on the coda of recorded events (D1-110, Denieul PhD, 2014;
Denieul etal. 2015), ISTERRE and CEA made a specific analysis based on ceps-
tral analysis and genetic inversion of teleseismic records to improve the depth
estimate of significant events (D1-50, Letort PhD; Letort etal. 2014, 2015). The
French catalogue is composed of approximately 40,000 events for which both
location and Mw are more accurate. Declustering of the catalogue according to
Marsan and Lenglin (2008, 2010) is valuable and can be used as complemen-
tary to more standard methods (Gardner and Knopoff 1974; Reasenberg 1985)
commonly adopted.
A preliminary set of source parameters defined for 15 pre-instrumental events in
the period 19001972 were investigated (D1-129) based on waveform inversion.
While the Mw magnitude can be better constrained, the method does not allow for
the determination of a robust depth estimate and significant expert judgment is
still necessary to infer a best estimate of Mw and depth. The results on Mw repre-
sent valuable information on events typically studied only through macroseismic
data.
Exploratory works were conducted on the determination of isoseismal areas
using automatic assessment either based on kriging techniques (D1-31 and
D1-128) or on manual determination of isoseismals (D1-148). While promising
these two alternatives to the method based on calibrated intensity prediction
equations, still present limitations (high sensitivity to the parameters adopted in
the kriging approach, bias introduced by the expert interpretation of the intensity
data points (IDP), no systematic quantification of magnitudes/depth); conse-
quently, it was decided to adopt a more classic and robust approach to refine the
historical earthquake catalogue. A preliminary work was also conducted for
revisiting the analysis and contextualisation of historical documents (D1-60).
This action should certainly be more systematically conducted in order to control
the IDP interpretations of the most significant events and to verify that no bias
was introduced in the current French historical database, developed by a single
analyst since 20 years.
The definition of macroseismic attenuation models using macroseismic data
points to derive Mw for the historical events (D1-108) has required the develop-
ment of a robust set of well calibrated events including data from the neighbour-
ing regions. A new earthquake catalogue has been developed and its impact on
the seismic hazard is still ongoing.
The sensitivity analyses in the Gutenberg-Richter parameters estimates and in
the hazard results, conducted using different conversions between Mw and other
magnitude scales, as well as on different strategies in the evaluation of Mw from
macroseismic data (D1-108), demonstrate that in regions of low to moderate activ-
ity these sources of uncertainty in the determination of the correlated parameters a
and b of the G-R magnitude frequency distribution significantly impact the hazard
estimates at low annual frequencies of exceedance. This is not observed in the more
active Italian region where the G-R parameters are more stable and mainly con-
trolled by the instrumental Mw values. The consolidation of the Mw catalogue in the
French area appears as one of the tasks that may significantly contribute to a reduc-
tion of the hazard estimate dispersion of the future site-specific PSHA conducted in
this region. More efforts should be made in such contexts to collect additional
description on the effects of historical earthquakes, to enrich the macroseismic data-
base and to revise the definition of the intensity data points.
3.3 Seismic Source Models
3.3.1 D
iffuse Seismicity VersusIdentified Seismogenic
Structures
Within the SIGMA project, the following two types of seismic sources were
considered:
Area sources where earthquakes occur on faults that are not identified or not
identifiable using our current knowledge and understanding of the seismogenic
mechanisms. In this case the future earthquakes are assumed to be distributed
throughout the entire area source.
Fault sources where earthquakes only occur on identified seismogenic structures
defined by static parameters (geometry of the fault) and kinematic parameters
(mechanism of rupture).
To identify the fault sources, it is necessary to exploit appropriate data to develop
models representative of a physical process. Since any tectonic earthquake is gener-
ated by a causative fault, the concept of area sources of diffuse seismicity zones
reveals our inability to identify the causative faults, due either to the impossibility
to understand the seismogenic process or to a lack of studies or efforts in acquiring
the relevant data. This implies that when fault sources are considered, the seismic
hazard calculation focus on the development of models of a physical phenomenon
that is constrained by boundary conditions, while when considering diffuse seismic-
ity areas, we model a conceptual process for which the boundary conditions are
much looser.
3.3 Seismic Source Models 33
3.3.2 Seismic Source Characterization Framework
One of the objectives of SIGMA was to identify the sources of uncertainty in the
seismic source models, to incorporate these uncertainties in the hazard calculations
and to quantify their impact and influence on the PSHA outputs.
In consideration of the purposes of the SIGMA project, the conceptual SSC
framework was driven by different considerations:
While a number of R&D tasks have been conducted within the work packages 1
and 4, the driver was much more to focus on individual components of the seis-
mic source models, rather than obtaining new SSC models under SIGMAs own-
ership. This choice was made because the main priority was given to the
identification of the influence of the uncertainties, rather than to the development
of new models. This is because the development of these models was considered
as requiring a project itself. The method was based on comparisons between
PSHA results obtained at the beginning of the project using existing SSC models
(D4-29, D4-41), and PSHA results at the end of the project obtained after
improvement of components of the SSC models (D4-94, D4-138) on which the
R&D efforts focused. Different metrics were defined to measure how the hazard
estimates were improved and how much the uncertainties were reduced. This
was performed using comparisons between the mean and median of the ground-
motion predictor variables as well as ratios between selected fractiles at selected
spectral periods and annual probabilities of exceedance of the hazard curves.
Between the initial and final PSHA runs, refinements of existing models were
considered to improve the SSC logic tree. In the French case, a new area source
model was developed with the purpose of better accounting for fault systems that
were the object of a specific task within SIGMA (Belledonne fault system and
faults in Provence) and a fault source model was introduced covering the
Provence sub-region. In the Italian case a new area source was identified to
account for deeper events and a new composite faults model was introduced.
The new conceptual models were introduced to consider alternative future spa-
tial distributions of earthquakes (geometry of seismic sources) and the occur-
rence processes and hazard computation approaches (Poisson using the doubly
truncated exponential model and the characteristic model, elapsed time model,
renewal model and non-ergodic model the last applied only in the Po Plain
region).
Epistemic uncertainties in the SSC models were initially identified and quanti-
fied adopting a process where the teams in charge of the seismic hazard assessment
(WP 4) developed their own assessment, based on existing seismic source zonations
at national or regional scales, and using as extensively as possible outputs from
WP1 as well as inputs produced in other scientific work. The SSC models and iden-
tified uncertainties considered in the initial PSHA (D4-29; Faccioli etal. 2012; and
D4-41) represent the decision taken and choices by the analysts when integrating
the data available at the beginning of the project, developing their interpretations
and having alternative solutions in mind. The models were reviewed by independent
experts from the scientific committee and interaction with the reviewers and with
other experts occurred through the Scientific Committee meetings and critical
review of the deliverables. As such, the models initially developed by the two haz-
ard teams were significantly improved in the course of the project (D4-94; Faccioli
etal. 2015; D4-138 and D4-169) benefiting from the feedback of sensitivity analy-
ses and the introduction of new alternatives and new outputs resulting from the
research tasks based on the recommendations and advice by the Scientific
Committee.
3.3.3 Area Source, Fault Sources, Gridded Seismicity
Most of the SIGMA models consider both area source zones and fault sources. The
case of fault sources was treated in more detail in the Italian test zone where more
geological and seismological data exist. The PSHA models consider, to a lesser
extent, also a zone-less approach where the seismicity is spatially smoothed. In such
a case the degree of spatial smoothing is controlled by the size and distance of the
kernel function. This approach is an alternative to the classical seismotectonic mod-
els that consider a well-defined geometry of the seismic sources, and it is of signifi-
cant interest when there are large uncertainties in the geological model. One
drawback of the approach is that the seismicity rates are a function of the kernel
distance, which is a difficult parameter to determine. This may lead to overestimat-
ing or underestimating the hazard, especially at large return periods, which is a
reason why the weights in the logic tree may be different at short and long return
periods (D4-170).
3.3.3.1 Area Sources
In the more active Italian area of interest the project principally focused on the
development of a composite seismic source model, for which a single area source
model was considered as a starting point, i.e. the so-called ZS9 (Meletti etal. 2008)
adopted for the official Italian seismic hazard map. This model was refined (D4-29,
Faccioli 2013) in the geometry of some of the area sources and, more notably, by
introducing a deep dipping zone, subduction-like, that describes the slab under the
Po Plain dipping towards the Tyrrhenian Sea, which is justified by the analysis of
the earthquake focal depths.
In the French case study, three area source models were considered to capture the
variability of the activity rates in a more stable region. All models integrate static
and dynamic parameters defined in the GIS database, but consider different strate-
gies to delineate the seismic sources. One model gives more emphasis to the inher-
ited geological structure (Fig.3.1, D4-41). A second model gives more emphasis to
known fault systems and to the seismic activity, as identified by the distribution of
Fig. 3.1 Area source model 1 of the French case study based on the combined interpretation of
static and dynamic parameters characterizing the earth crust (D4-41)
historical and instrumental earthquakes. This model considers smaller areas. The
third model is more controlled by the identification of a coherent deformation pat-
tern and includes larger area sources.
3.3.3.2 Fault Sources
The consideration of fault source models is of significant interest for sites where
fault sources may control the hazard at specific ranges of spectral periods (low or
high), when the purpose is to assess the hazard at long return periods, or when direc-
tivity effects can significantly affect the ground motion.
For fault sources, the style of faulting and the strike and dip of ruptures may be
single or composite. In this case, the logic tree should encompass all types of style
of faulting, with weights adding up to 1 (reverse, normal, strike-slip, unknown). The
Fig. 3.2 Composite seismogenic fault sources considered in the Po plain (D4-94). Orange sources
are the fault sources considered for the hazard assessment at three sites (purple triangles)
strike and the dip are a function of the style of faulting and of the level of approxi-
mation with which the fault is identified from the data.
The reliability of fault source models in low to moderate activity regions is, how-
ever, controversial as data to characterize all the parameters required as input to the
PSHA model are often missing. Hence, the approaches adopted in the two SIGMA
case studies were significantly different.
Composite fault sources were considered in the Italian logic tree (D4-94) with
seismic activity described by the characteristic earthquake model for large magni-
tudes and by the usual Poisson truncated exponential model for the background
seismicity. The existing national and regional fault database (DISS Version 3.11,
INGV) was used and completed by Burrato etal. (D1-27) and Burrato and Valensise
(D1-67) by introducing the data generated by the recent sequence of May 2012, to
refine the composite source model (SSC) of the Po Plain (Faccioli et al. D4-94,
Fig. 3.2). Because of the limitations of GMPEs to capture site-specific ground-
motion properties at short distances from the fault ruptures (like near-field and
directivity effects), an innovative approach was introduced by the Italian team. They
compare and check, at two sites, the results using a simulation of the ground motion,
in which the ground-motion propagation between the ruptures and the site results
from deterministically simulated ground motion in replacement of the empirical
GMPEs. Finite-fault stochastic simulations were carried out using the EXSIM soft-
ware (Motazedian and Atkinson 2005) which considers the (1D) local site response,
and the generalized description of ground-motion attenuation called GAF (Faccioli
2013, and Faccioli etal. D4-94) was applied to the same composite source model as
CAS site, RP 2475 yr
GAF stress drop 50

GAF stress drop 30
1 AB11(49-50-51 CSS)
Ita13(49-50-51 CSS)
NTC2008
0.8
UH SA [g]
0.6
0.4
0.2
0
10-1 100
T[s]
Fig. 3.3 Po plain case study (presentation slides by Faccioli, Lyon SC meeting 14/11/2013).
Comparison of UHS for CAS study site at 2475 years return period, obtained with Composite
Seismogenic Sources (CSS) treated as simple area sources and characteristic earthquake behav-
iour, and finite-fault stochastic simulations (GAF, green lines), using two different GMPEs (AB11
and Ita13). Black dotted line corresponds to the standard spectrum of the NTC2008 Italian code for
ground type C
the one developed considering the characteristic model. Comparison at one of the
study sites, in terms of UHS, demonstrated the interest of such an approach to
enhance site-specific peculiarities that impact the predicted ground motion, due to
the site location in its seismo-tectonic context or soil conditions (Fig.3.3). As in any
simulation, one difficulty is to determine the appropriate range of values for all
parameters required to run the model (fault distribution layout, slip distribution,
path effects and distance dependence of the signal amplitude, stress drop and kappa),
which remains sensitive to each parameter. Such an approach requires good site-
specific data as is the case for the Po plain application, where for the CAS site
(shown in Fig.3.3) the site spectral amplification function used in the GAF approach
was based on observations from a vertical array. This may be considered representa-
tive of future PSHA evolution when the ergodic assumptions introduced at different
steps of the classic approach do not allow capturing the potential site-specific
effects.
On the French side, the main actions conducted to improve the fault characteriza-
tion focused on:
Geomorphological and tectonic analysis of the Belledonne fault system (D1-66),
tentatively considered as an area source of small extension in the preliminary
SSC model;
Tentative fault database in South-East quarter of France (D1-127); and
Geomorphological and topographic marker analysis in Provence and low Rhone
Valley (D1-149).
As findings of these works were not sufficiently conclusive to be integrated in a
fault model within the schedule of the project, the fault model introduced in the final
PSHA relies on a model developed of the Provence region for the CEA.Between
the initial and final PSHA, the software code was changed to better model the rup-
ture geometry and to improve the calculation of the distance metrics, which were
questioned when comparing the PSHA codes (D4-140 and Chap. 6). The results of
the sensitivity analyses demonstrate that in the French context, the influence of the
fault model on the hazard results is only significant at large return periods and for
sites close to the faults. This would confirm and underline the necessity of near-
regional and site-vicinity scale investigations for critical infrastructures.
3.3.3.3 Gridded/Smoothed Seismicity
Smoothed seismicity representations, which usually do not account for the geologi-
cal or tectonic settings, represent a valuable alternative strategy to area or fault
source models. This approach is used in nuclear and non-nuclear projects in the US
(CEUS USGS Moschetti and Petersen 2012; CEUS-SSC EPRI, DOE, NRC
2012) and was tested in both SIGMA case studies. The main basic assumptions to
implement the smoothed seismicity approach (as applied in the French case study)
are the following:
Future earthquakes are more likely to occur close to past earthquakes. This
allows the development of a spatial-likelihood function (also called kernel func-
tions or smoothing kernels) to predict the location of future events.
Different types of scaling can be used to define the smoothing kernels, such as:
(1) A fixed size kernel; (2) A magnitude-dependent kernel for which the smooth-
ing width is proportional to the magnitude (the larger the earthquake magnitude,
the wider the kernel function); (3) A spatially adaptive kernel according to
Helmstetter etal. (2007), such that the smoothing width at a given point is equal
to the distance to the n-th closest earthquake. In highly active areas, the smooth-
ing width is much shorter that in weakly active areas. This model is a density-
dependent function.
The smoothed gridded seismicity rates are calculated from a declustered, Mw,
earthquake (EQ) catalogue using the data within the completeness periods.
The smoothed gridded seismicity rates calculated at each Mw bin are analysed to
evaluate the Gutenberg-Richter (GR) parameters at each grid point.
Super-domains may be used to define different Mmax and to spatially constrain the
smoothed seismicity (leaky/strict boundaries)
A different approach was adopted in the Italian case study, as will be seen in the
following. Other assumptions that may or may not be introduced are:
The maximum magnitude and thickness of the seismogenic crust are defined
consistently with the zonation approach and a strategy can be applied to attribute
a tectonic style to each grid point.
Even if the smoothing seismicity models do not formally account for the seis-
motectonic characteristics as is done for a seismotectonic zonation, it remains
possible to introduce super-domains to consider strong boundaries delimiting
significant deformation patterns. This issue (related to the strict/leaky boundary
effects on the kernel definition), however, was not addressed during the project.
In the French SIGMA case study, alternative smoothing models were imple-
mented (D4-138, D4-170). A zoneless approach was included in the logic tree also
for the Italian case study, where a higher weight was assigned to the model-based
branches (area sources and Fault sources + Background seismicity) with respect to
the gridded ones (0.6 vs 0.4), for return periods = 2475 and 10,000 years, but equal
weights for RP = 475 years. Equal weights were assigned for 475 years also in the
French case study, while the weight was decreased to 0.1 for the smoothed seismic-
ity branch for the 10,000 year return period. For the Po Plain, the HAZGRID
smoothed seismicity model (Akinci 2010, updated with the 2011 CPTI11 cata-
logue), based on Poisson occurrences (time-independent) was applied; this operates
on a regular grid of point sources, 0.1 0.1 in size, with smoothing performed
spatially by a 2-D Gaussian function with 25km correlation length, and a constant
G-R b-value = 1.26 determined on a nation-wide basis. In the French case 2-D,
isotropic, Gaussian smoothing kernels were used considering spatially adaptive,
radii kernels. The maximum magnitude and hypocentral depth distributions were
defined consistently with one of the area source models.
On average, it was observed that the mean spectral acceleration distribution asso-
ciated to the gridded seismicity branch is lower than that associated to the area or
fault source models. From the sensitivity analyses, results were shown to be sensi-
tive to the choice of the minimum magnitude above which the seismicity rates are
derived and quite large hazard variability was caused by the adopted smoothing
adaptive kernels. A specific investigation on the benefit of using strict boundary
conditions rather than leaky conditions may help to reduce the edge effects included
in the spectral acceleration distribution. When the variations of seismic activity
within the large region considered around the site of calculation is significant, the
use of isotropic smoothing kernels may be questionable. In this matter, the use of
strict boundaries within super-domains should be considered in regions where
abrupt changes in the activity rates are observed. This, however, was not checked
during the project.
3.3.4 Lessons Learned Related toSeismic Source Models
Several lessons were learnt from SIGMA through the work carried out and the
review process, namely:
Epistemic uncertainty in the area source boundaries must be introduced espe-
cially when sites are close to those boundaries. This is because the contrast in
seismicity rates between the host zones and neighbouring zones significantly
affects the estimated hazard. This can be done by considering alternative SSC
models and considering several types of models in the logic-tree (Area/Fault/
Smoothed seismicity).
The zoneless and gridded seismicity approaches are of significant interest for
testing the standard PSHA approach and to counterbalance the remaining uncer-
tainty associated to the boundaries of the seismic sources. In areas with moderate
seismicity the earthquake sample is, however, not representative of the long-term
seismicity and does not include large/rare events, close to the characteristic or
maximum magnitude, which are needed for the predictions of ground motion at
low annual frequency of exceedance. The validity of the zoneless approach is
more questionable at those frequencies. When included in the logic-tree, the
weighting scheme must consider the objective of the PSHA (e.g. the return peri-
ods of interest) and the robustness of the earthquake catalogue. A higher weight
was adopted in the Italian case study, because the gridded model had the best
ranking for a RP of 1000 years, and a middle score for a RP of 475 years accord-
ing to the ranking procedure discussed in Albarello etal. (2013).
While extensively used when fault sources are considered, the characteristic
model requires parameters (slip rates and recurrence of the Characteristic mag-
nitudes), the definition of which suffers from a scarcity of geological data and
earthquake catalogue records in zones of moderate seismicity. Significant efforts
to acquire new data remain necessary to introduce fault models and, potentially,
fault rupture simulation in future PSHA.While the composite source model for
Italy is developed using a quantity of local and recent data, which justify its
introduction in the logic tree, the same model is more questionable for France
except where data support the identification of the fault geometry and allow for
the characterization of its activity.
Ground-motion simulations based on fault rupture modelling are seen as poten-
tial future methods that can bring significant information when ergodic models
present too strong limitations in predicting the ground motions. This is especially
the case when sites are located close to identified active faults. This also provides
a way to reduce the uncertainties in a context where the seismic sources that
control the seismic hazard in the regions of interest are located at short distances,
where near-fault ground motions (long-period pulses, permanent ground dis-
placements and directivity effects) may significantly differ from ground motions
predicted by conventional GMPEs.
A preliminary fault database was developed within SIGMA for SE France, but
such a database should contain all the parameters required in the calculation
3.4 Occurrence Processes 41
process: fault geometry (length, width, strike and dip), direction of slip (rake
angle), the time history of fault slip (slip-time function), rupture initiation, rup-
ture velocity, stress drop, slip ratio and so forth. It was recognized that more
interaction is necessary between the geologists in charge of the database devel-
opment and the team in charge of the ground-motion assessment and hazard
estimation.
Arriving at the completion of the project, a large number of new data and ingre-
dients were developed within WP1. However, the adopted strategy to identify the
uncertainties in the SSC and test their influence on the hazard results did not allow
the development of new SSC models for France within the time frame of the project.
A significant benefit of the actions conducted during SIGMA to improve the geo-
logical and seismological database would consist in integrating the new database
and methods to develop new seismic source models. This should be considered in
an extension of the project.
3.4 Occurrence Processes
Among the models used within the SIGMA project, the magnitude distribution or
magnitude probability density function that describes the relative number of earth-
quakes that occur in a given seismic source between Mmin and Mmax was essentially
defined in the final models considering:
the Poisson model with a truncated exponential PDF; and
the characteristic model.
In the Italian case study (Phase I, see D4-29) alternative models, such as the so-
called renewal model, were also checked. The time dependence is modelled by a
renewal process with a Brownian Passage Time (BPT) distribution. In this model,
the occurrence of large earthquakes is assumed to present a recurrence and the time
of a past strong earthquake is considered through a conditional probability that an
earthquake occurs in the next years given that it has or not occurred already in the
last years. A key parameter in applying a renewal model is the periodicity of the
earthquake recurrence interval, which requires either a complete earthquake cata-
logue or sufficient paleo-seismological studies to constrain the input parameters to
the model.
3.4.1 Poisson Model
The Poisson model is preferably applied in areas of low to moderate activity,

because the behaviour of individual faults is not known in sufficient detail to estab-
lish a reliable fault-specific activity model. In this case, all earthquakes within the
seismic source are assumed to be independent and the probability distribution of the
small to moderate earthquakes is exponential. The most common approach is the
doubly truncated exponential distribution (Gutenberg and Richter 1956). The spa-
tial variation of recurrence parameters is considered as uniform or varying. In the
latter case, when a smoothed/gridded seismicity representation is adopted, a spatial
smoothing is applied considering a kernel function or an adaptive kernel function
that accounts for the spatial density of the epicentres or the magnitude (see
Sect.4.3).
3.4.2 Characteristic Model
Fault sources tend to obey a different principle (Youngs and Coppersmith 1985),
whereby individual faults or fault segments generate ruptures of similar size at
recurrent intervals, representative of a characteristic event. The model is typically
assumed to follow a truncated normal distribution to account for variability in the
characteristic magnitude. To allow small and moderate magnitudes to occur on the
fault, a composite model is considered, based on a combination of the truncated
exponential model up to the characteristic magnitude (or to a somewhat smaller
magnitude) and the characteristic magnitude for large earthquakes. The characteris-
tic model requires the definition of fault slip rates and characteristic magnitude Mchar.
3.4.3 Time-Dependent Seismicity Models
It is worth mentioning that in D4-29 a sensitivity analysis on the impact of time-

dependent seismicity models was performed and that within the Italian DPC-INGV
S2 project both stationary and non-stationary earthquake time occurrence models
were tested for the evaluation of the seismicity of Italy. These examples showed, in
some zones, significantly higher hazard, when the time since the last occurrence is
long, than in the classic approach based on Poisson occurrence applied to seismic
source zones, in moderate or low seismicity regions.
3.5 Maximum Magnitude andRecurrence Parameters
3.5.1 Maximum Magnitude
In areas of low to moderate seismicity, the determination of the maximum magni-

tude (Mmax) is one of the most controversial issues in the definition of seismic source
parameters required in a PSHA, especially for long return periods for which the
hazard is governed by events with large magnitudes. The methods of determination
of Mmax were significantly different in the two case studies.
3.5 Maximum Magnitude andRecurrence Parameters 43
Epistemic uncertainties associated with Mmax are characterized by alternative

models generally involving a magnitude range between a lower and an upper bound
and a probability density function.
In the initial model for the Po plain, (D4-29) the maximum magnitudes were
inspired by the results of the European SHARE (2013) project. Mmax was taken as
the maximum observed magnitude in each superzone of SHARE(2013), imposing
a lower limit of 6.5. The uncertainties associated to the maximum magnitude were
between 0.3 and 0.5 and were used to propagate the uncertainties in Mmax as a ran-
dom variable with truncated uniform distribution. In the same initial model, the
mean Mmax values were also made to depend on depth, by decreasing them down to
a minimum of 5.0 in association with the lowest depth used for the area sources
(4 km). In the second version of the model (D4-94), the same assumptions were
maintained for Mmax and its dependence on depth. For the composite source model
all faults able to generate Mw 5.5 or higher were considered. The method of the char-
acteristic model was used with Mchar between 5.5 and 6.0, with an uncertainty of 0.3.
In the French case, the maximum magnitude was initially defined as a uniform
distribution between lower and upper bounds of Mmax (D4-41). The lower Mmax limit
was defined based on an increment applied to the maximum observed magnitude in
zones of similar deformation and the upper bound by the maximum fault dimen-
sions in each area source.
In D4-138 and D4-170 a careful sensitivity analysis was performed in order to
test different methods for the definition of the Mmax distribution (Fig.3.4), including
Grenoble Valence Marseille

0.01 0.01 0.01
2012 SIGMA
0.001 0.001 0.001 Model 1
Model 2
Model 3
APE
0.0001 0.0001 0.0001
1E-005 1E-005 1E-005
1E-006 1E-006 1E-006

0 400 800 1200 0 400 800 1200 0 400 800 1200
PGA [cm/s2] PGA [cm/s2] PGA [cm/s2]
0.01 0.01 0.01
0.001 0.001 0.001

APE
0.0001 0.0001 0.0001
1E-005 1E-005 1E-005
1E-006 1E-006 1E-006
0 400 800 1200 0 400 800 1200 0 400 800 1200

PSA T=1s [cm/s2] PSA T=1s [cm/s2] PSA T=1s [cm/s2]
Fig. 3.4 Effect of different epistemic Mmax distributions on the PGA and PSA (T = 1s) mean haz-
ard curves at three sites. The epistemic uncertainties on Mmax have a notable impact on seismic
hazard at low annual probabilities of exceedance
the EPRI (1994) Bayesian approach that provides a framework to handle, in a trans-
parent and reproducible way, the definition of Mmax in low-seismicity areas (Ameri
etal. 2015). The approach was applied by distinguishing two areas with very low
seismicity and areas of moderate seismicity in the SHARE superzones (area source
model and earthquake catalogue). Two prior Mmax distributions have thus been
developed. The likelihood functions and the posterior Mmax distribution are calcu-
lated for three domains of the French South-East quarter, characterized by different
level of seismic activity, and are applied in the final PSHA.It was found that the
lower bound was significantly increased (toward maximum observed magnitude,
except for one model) while the upper bound is consistent with the initial estimate.
The shape of the posterior Mmax distribution is defined on a more rational basis than
the assignment of weights through expert judgement.
For fault sources, Mmax was estimated from the fault dimensions and scaling
laws. Possible fault segmentation was considered to determine different alterna-
tives. Using the fault segmentation approach requires having available sufficient
data to identify geometric discontinuities that may stop the ruptures. This is a key
issue when developing fault models as it impacts both the definition of the maxi-
mum magnitude and the characteristic magnitude and it benefits from appropriate
data to weight the models considering multiple segments or unsegmented faults.
3.5.2 Recurrence Parameters
The activity rates of the different magnitude levels between Mmin and Mmax charac-
terize the magnitude density function representative of a seismic source.
The estimation of the activity rates based on the seismic catalogue requires fit-
ting the truncated exponential model and paying attention to: the earthquake sample
deemed to be statistically representative of the seismicity, the homogenized magni-
tude scale of the catalogue, the completeness periods of the catalogue and the statis-
tical independence of the events.
The common approach used in the two SIGMA case studies was to derive the
activity rates and the b-value through the maximum likelihood method (Weichert
1980), while other methods were applied in sensitivity analysis, such as least-
squares and weighted least-squares (D4-128).
For the calculation of G-R parameters, some significant differences between the
two case studies are noticeable. The minimum magnitude adopted was significantly
higher, Mw 4.5, in the Italian case (thus corresponding to the minimum magnitude
of the hazard integral), while it was Mw 2.5in the French case due to more sparse
seismicity. More specific attention was paid in the French case to propagating the
uncertainties of the frequency magnitude distribution considering the uncertainties
in the magnitude estimates and running Monte Carlo simulations. In the Italian
case, however, the influence of adopting a lower bound magnitude Mw 2.5 (as in the
initial study, D4-29) or Mw 4.5 (as in the final version, D4-94) for the estimation of
3.5 Maximum Magnitude andRecurrence Parameters 45
Casaglia site, RP 475 yr Casaglia site, RP 2475 yr Casaglia site, RP 10000 yr

2.5 2.5
4 old par
new par
2 2 3.5
1.5 1.5
2.5
UH SA [g]
UH SA [g]
UH SA [g]
2
1 1
1.5
1
0.5 0.5
0.5
0 0 0
-1 0 -1 0 -1 0
10 10 10 10 10 10
T[s] T[s] T[s]
Fig. 3.5 UHS for three return periods (RP) at the Italian CAS (Casaglia) study site: shown are the
initial study D4-29 results (blue curves, old par) using Mwmin 2.5in the calculation of G-R param-
eters, and those of the final study D4-94 (green curves, new par) using Mwmin 4.5, for which no
magnitude scale conversions were needed in the sample catalogue
the G-R parameters was discussed in detail and shown to have a moderate influence,
as shown in Fig.3.5.
In the final version of the SE France model, and following the recommendations
of the scientific committee, the uncertainties in the G-R parameters for each zone
were quantified by directly propagating the uncertainties on the earthquake cata-
logue in terms of earthquake Mw and completeness period, via a Monte Carlo
approach. In practice, many synthetic catalogues were generated by sampling
uncertainties on Mw and completeness periods and, for each realization, a G-R
model was fitted to the calculated rates, to obtain a series of correlated a and b val-
ues introduced as branches of the logic tree (Fig.3.6, D4-138).
3.5.3 Lessons Learned
The GR parameters a and b are not independent, and the implementation of the
associated uncertainty propagation in the PSHA software must be clearly
explained considering the correlation between the parameters.
The estimation of the G-R parameters (a and b) is quite stable in regions of
moderate-to-high seismicity. This was demonstrated within the Po Plain model
where the final G-R parameters are calculated considering a high cut-off Mmin
(Mw > = 4.5) threshold from which the uncertainties related to the magnitude
homogenization substantially decrease because the moment magnitude estima-
tion is mostly provided by moment tensor solutions rather than by using
magnitude-conversion equations. On the contrary, the estimation of G-R param-
eters in low seismicity areas remains a challenging task. Within the French
model, the minimum Mw used for the fitting of the G-R correlation was quite
zone name: 4010; a=2.4061 b=0.90239 1

zone name: 4010
1 10
10
0 0
10 10
-1 -1
10 10
annual rate M+
annual rate M+
-2 -2
10 10
-3 -3
10 10
-4 -4
10 10
-5 -5
10 10
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5
Mw Mw
zone name: 4010 zone name: 4010
4 3
3.5 2
3 1
epsilon a-value
a-value
2.5 0
2 -1
1.5 -2
1 -3
0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 -3 -2 -1 0 1 2 3
b-value epsilon b-value
Fig. 3.6 example of assessment of G-R parameters for a seismic source. Upper-left: G-R derived
based on the observed rates in the catalogue. Vertical bars represent the 16-84 confidence intervals.
Upper-right: G-R models (in grey) derived from 500 synthetic catalogue. Red symbols represent
the rates for each synthetic catalogue and blue symbols the observed rate for the initial catalogue.
Lower-left: a and b correlated values of the 500 G-R models. Lower-right: plot of epsilon
a = (a a)/a and of epsilon b = (b b)/b
variable and typically much smaller than the one used for the Italian region (as
low as Mw = 2in the most stable areas). This implies that during the magnitude
homogenization process, Mw was estimated from other magnitude types based on
conversion equations that typically carry large uncertainties. Moreover, because
the number of earthquakes is limited, the estimated activity rates are affected by
large uncertainties, which require conducting more sensitivity analyses and alter-
native methods to capture the uncertainties in the activity rates and recurrence
parameters.
Especially in regions of low to moderate activity, the maximum magnitude esti-
mates have a significant impact when the objective is to provide ground-motion
assessment at annual frequencies of exceedance lower that 104 (Fig.3.4). The
selection of values just because they have been published in scientific journals or
scientific projects is not acceptable when the purpose of such a project was to
conduct seismic hazard assessments for annual probability of exceedance much
3.6 Logic-Tree Implications 47
higher than the annual probability of exceedance considered in a site-specific

hazard assessment (e.g. for nuclear projects). While no specific task to develop a
methodology and to assess the Mmax distribution and its uncertainties was con-
ducted within the SIGMA project, the sensitivity analyses performed for differ-
ent sites showed that the epistemic uncertainties on Mmax have a notable impact
on seismic hazard and that a variety of defendable approaches can be employed
to assess the Mmax distribution (Fig.3.4). Although it is recognized that the CEUS
context and west European contexts in terms of seismotectonics and seismic his-
tory is very different, the adaptation of the EPRI (1994) Bayesian approach to the
French context is one of the recommended approaches because it limits the use
of expert judgement and its effect on the hazard estimates. Another approach to
imposing a constraint to the possible range of Mmax values is to compare the
deformation rates measured from geodetic data and those derived from geologi-
cal deformation along longer geological times or implied by the G-R model with
a certain range of Mmax for a specific zone. The objective of such comparisons is
to determine whether the deformation rate implied by a certain geological model
is consistent with the observed ones (geodetic and seismological). Such
approaches may provide, in the future, a constraint to the possible Mmax values,
particularly in low seismicity areas. Hence, more efforts should be paid to these
subjects in the future.
Scaling relationships can also be used to estimate an upper bound of Mmax con-
sidering the maximum dimensions of known faults in each seismic source.
3.6 Logic-Tree Implications
3.6.1 Logic Tree Approaches
The SSC logic tree represents the various interpretations that the analysts have con-
sidered as being credible in identifying and locating the seismic sources and charac-
terizing their seismic activity.
In the two regions of interest the seismic sources are composed by area or fault
sources and the logic trees typically consist of branches for alternative conceptual
models of seismic source boundaries, occurrence models, activity rate parameters,
style of deformation and maximum magnitudes. All of these parameters have an
influence on the hazard curves that are the final products of a PSHA model.
Assigning weights to the alternative SSC introduced at each node of the logic
tree is an imposed exercise to quantify the degree of confidence that the considered
alternative represents reality. As a general requirement, the sum of the weights at
each node should be one and there is a requirement that the values on the branches
of the tree be mutually exclusive and that at a node all the branches be collectively
exhaustive (Bommer and Scherbaum 2008). The sum to 1.0 indicates that the
branches represent the collectively exhaustive set of options. This requirement is
general seismicity source model GMPEs weight

description
(0.4) 0.04
Fea10
0.08
(0.4) 0.04
AS AB11
0.08
model-
(0.5)
based (0.33)
(0.2) ergodic 0.02
seismicity ITA13
non-ergodic 0.04
(0.67)
(0.6) 0.04
0.08
(0.5)
0.04
FS + BG GMPEs 0.08
RESULT 0.02
0.04
(0.4) 0.0533
0.1667
gridded
HAZGRID (new) GMPEs 0.0533
seismicity 0.1667
0.0267
0.0533
Fig. 3.7 Example of logic tree adopted for the PSHA conducted in the Po Plain (D4-94).
Conceptual models are considered early in the logic tree, then, more specific assessments are con-
sidered as sub branches, the sum at each node being 1.0.
not always easy to fulfil through lack of data or due to insufficient efforts to collect
and generate the appropriate data. Hence, it is sometimes difficult to be able to dem-
onstrate that the logic tree complies with this requirement.
In general, a logic tree is structured such that more general assessments occur as
the first branches of the tree (like conceptual models: area sources/fault sources/
smoothed seismicity) and more specific assessments are included later in the tree as
sub-branches of the main branches (see the example of Fig.3.7).
It was not the main objective of SIGMA to develop site-specific results at a site
(although in the Po Plain this was done), but rather to treat the uncertainties and
appreciate their impact on the hazard. In building the logic trees, the rationale and
criteria for identifying the first branches were a function of the data available to
conduct the analysis and of the objective of the analysis. The three mentioned con-
ceptual models (area sources, faults and gridded seismicity) compose the first
branches. Then the suite of branches includes all relevant parameters that character-
ize the activity of the identified sources given the conceptual model adopted and the
data at hand (maximum magnitude, variation of recurrence parameters). For this
reason, there is no generic logic tree but only specific logic trees that are consistent
with: (1) The objective of the analysis (regional map vs. site-specific assessment,
high annual probability of exceedance (AFE) vs. low AFE), (2) The level of efforts
to acquire the data and, (3) The complexity of the seismotectonic environment.
In the Italian case both area and fault sources were considered (D4-94), while in
the French model area sources were preferably used to model the spatial distribution
of the earthquakes because the faults were not accurately known or because of lack
of data to characterize their activity. Fault sources were, however, introduced in the
Provence region in the final run (D4-170). In both cases, the first step was to develop
a set of models identifying the geometry of the sources. In fact, the area sources
were defined as a volume, i.e. in addition to the delineation of each area on a map,
the range of depths where the sources are located were specified. Although the dis-
tribution of the seismic sources in the volume obeys a probability distribution, epis-
temic uncertainties characterize the thickness of the seismogenic crust.
In the composite fault source model of the Po plain, where faults have a multi-
planar geometry and ruptures are distributed along the fault plane considering a
uniform or non-uniform distribution, area sources were included to account for
earthquakes that occur off of the faults (background seismicity which is considered
as diffuse seismicity). A similar approach was adopted in the Provence fault model.
3.6.2 E
fficient Tools fortheLogic Tree Conception
andWeights Assignment
An important lesson of the SIGMA project and a good recipe for the development
of a SSC logic tree with the objective to reduce uncertainties is to adopt a phased
approach based on the introduction of calculations at an earlier stage of the project.
Several tools were identified during the project to assist in identifying hazard-
significant issues and topics that need to be understood and resolved during the
PSHA study:
The development of a pilot model (at a very early stage);
Hazard disaggregation;
Sensitivity analyses; and
Testing of the PSHA results using observations.
3.6.2.1 P
ilot Model foranInteraction andInterface Management
BetweenComponents ofthePSHA
Any PSHA should be conceived in such a way as to develop an initial logic tree and
then to iterate during the progression of the project in order to improve the logic tree
(add branches, logic-tree trimming, weight branches, etc.), to focus on the relevant
parameters that control the hazard at the site and the influence of epistemic
uncertainties.
Each individual site belongs to a site-specific tectonic environment for which the
treatment of uncertainties may be complex. There is a need to capture the technical
issues at an earlier stage of any project; this would help to appreciate if all the sur-
veys, investigations and tasks are developed with the objective of collecting the
appropriate level of knowledge on the seismic sources and of identifying the
a ctivities needed for improvement of the parameters that control the seismic hazard
at the site.
Within SIGMA such pilot models were introduced early (D4-29, D4-41) to iden-
tify the seismic sources and the parameters or components of the models that have
a significant impact on the hazard results, at specified return periods of interest and
at specific spectral periods of interest.
Such a pilot model should be preferably developed using the existing and avail-
able data (earthquake catalogues, published documentation, pre-existing models)
without waiting for the integration task and even if a full identification and charac-
terization is not available yet. Such a model is specifically helpful to identify, among
the different actions, the priorities and where to concentrate efforts (in acquiring
new data, in adopting a method rather than another and so forth) to reduce the uncer-
tainties and to anticipate the conception of the logic tree.
3.6.2.2 Disaggregation
In the same way, hazard disaggregation of the pilot model is a tool that should be
used not just at the end of the PSHA, but all along the development of the seismic
source model, again to offer quantitative indicators on the contribution of each indi-
vidual source to the total hazard.
3.6.2.3 Sensitivity Analyses
Assigning weights of logic tree branches is seen as a subjective approach whatever

the complexity of the PSHA approach. This is because the weights are often pro-
vided by experts, who assign a relative weight to each branch and are treated as a
probability in the calculations, which expresses the experts degree of belief in the
model or parameter value being appropriate.
One of the requirements is that the technical assessments that underlie the iden-
tification of the nodes of the trees, the alternative branches at a node, and the weights
assigned to each branch must be justified and documented to provide the necessary
material to other experts to make a review or a critical analysis of the PSHA
outputs.
The technical bases for the selection of branches included in the logic tree, the
assignment of weights to different models in the final distribution or the exclusion
of models and methods that are discarded from the general SSC model can be more
easily justified by conducting sensitivity analyses.
Within the SIGMA project and for both case studies, different sensitivity analy-
ses were conducted, some early in the project, others during its course, and an inte-
grated exercise was finally conducted to quantify the reduction of uncertainties at
the end of SIGMA.They constitute a tool to investigate the origins of uncertainties,
to understand their influence on the hazard results and to guide better the improve-
ment of the model in the course of the project. Because this feedback allows
Fig. 3.8 Example of sensitivity analyses to the different components of the PSHA model: Tornado
diagrams at 475 years return period for three test sites in SE France showing the sensitivity of mean
hazard to Source Model, Earthquake Catalogue and GMPE at three spectral periods (PGA, 0.2s
and 1s) (D4-138).
easuring the sensitivity of the seismic hazard to the different parameters/inputs

m
(e.g. which one affects most the results), and evaluating the contribution of the
uncertainties in the input parameters to the total hazard uncertainties, they also pro-
vide justifications in the selection of the branches of the logic tree and in the assign-
ment of weights.
The sensitivity analyses were performed for three different sites in each area of
interest, characterized by different levels of seismicity and/or their location at
boundaries of seismic sources (areas or faults).
Different metrics can be used to measure the impact of the uncertainties, i.e.:
The distance or ratio between percentiles hazard curves (e.g., 595 or 1684%)
at specified return period and spectral period (D4-94 and D4-138);
The distance between mean hazard curves of two models (D4-138);
The variability in UHRS at specified return periods (D4-94); and
Tornado plots as shown in Fig.3.8, from D4-138.
A lesson from SIGMA is that the extent to which the uncertainties impact the
hazard estimates at targeted annual frequency of exceedance, given the tectonic
environment of the site, is site-specific. A number of the basic influencing factors
should be, however, systematically addressed, such as:
How the uncertainties due to the Mw homogenization of the catalogue (via mag-
nitude conversions or via exploitation of macroseismic data) affect the results;
How the geometry and boundaries of seismic sources affect the activity rates;
Whether the methods in the determination of the G-R parameters lead or not to
substantial variability;
How uncertainty in the thickness of the seismogenic crust impacts the ground
motion; and
How the Mmax uncertainty and distribution of the Mmax values impact the hazard
curves.
Other factors are a function of the site location within the seismic source context
and/or the occurrence models adopted to characterize the seismicity distribution,
such as:
3D geometry of the fault sources;
Fault segmentation; and
Stationary versus non-stationary (time-dependent) occurrence models.
Site-specific assessments, developed for nuclear sites, require the development
of an up-to-date site-specific earth science database, in which the studies and inves-
tigations are conceived and designed to provide increasing level of specificity and
increasing effort in reducing the epistemic uncertainties, moving from the regional
scale to the site-specific location. Feedback from the sensitivity analyses provides
information and insights at the different stages of the PSHA.This is used to take
decisions in the development of the logic tree or in the optimization of new investi-
gations or approaches required to better control and reduce the uncertainties.
3.6.2.4 T
esting theBranches oftheLogic Tree Using Data
andObservations
One of the recurrent observations from the SIGMA reviewers was that the justifica-
tion of the weights adopted at each node of the logic tree was rarely explained
clearly by the hazard team. Documentation in this context is key in order for third
parties to understand the choices.
It was recognized that sensitivity analyses are appropriate tools to compare
among alternatives and to provide indicators for branch trimming; it was also
pointed out that they do not measure if the tested parameter or branches of the logic
tree represent a more reliable hazard estimate than another branch considered in the
model.
An interesting task of SIGMA was the application of a Bayesian approach to
update the PSHA logic tree based on observed exceedances of a certain acceleration
threshold (D4-139). This approach uses the comparison between the observed and
predicted number of exceedances and a Bayesian framework to modify the logic
tree weights. In practice, the a priori logic tree weights of the branches that are in
better agreement with the observations are increased while the others are lowered to
obtain a posterior distribution of the weights; the sum still being 1.0. Tests were
conducted on the hazard estimates to identify incompatibilities with observations
References 53
and to boost or invalidate components of the hazard model or parts of the hazard
model.
While still requiring some development, the approach is seen as a promising tool
to complement the definition of weights by expert judgment and it was recognized
by the SC that the consolidation of the method was worth pursuing.
3.6.3 Verification andQuality Assurance (QA)
Another important lesson of the SIGMA project stressed by the review process
concerns the requirement for including a verification process and quality assurance
at various steps of the PSHA model development.
While most of the PSHA studies carried out for conventional buildings are rarely
peer-reviewed, the QA requirements for nuclear applications become more strin-
gent in the post-Fukushima context and QA verification and validation procedures
are mandatory for any nuclear project.
The different hypotheses adopted at each node of the logic tree tend to make the
number of combinations extremely large and the identification of the components
that control the hazard at a site very complex. This is the reason why the develop-
ment of the SSC (and of course GMC) logic trees must be elaborated following
rules that are now commonly accepted in the nuclear community. Such a process
was introduced before the final run of the PSHA for the French area. It consisted in
preparing hazard input documents that describe the interfaces between the SSC
models elaborated by geologists and seismologists, and the parameters and models
introduced by the hazard calculation team into the PSHA software. This was an
iterative process where the SSC analysts and the hazard calculation specialists
closely interacted to verify that the hazard calculation model was as much as pos-
sible representative of the SSC models, and also that the models and ingredients
were implemented in the PSHA in a complete and unambiguous way. The hazard
input document for the French South-East quarter was validated by the scientific
committee before the hazard computations.
Finally, it is a crucial requirement that the computational tool selected be a docu-
mented, verified and validated software, but at the same time sufficiently flexible to
allow for the introduction of new techniques or methods that may be required during
the project. A discussion on the software is provided in Chap. 6.
References
Akinci A (2010) HAZGRIDX: earthquake forecasting model for ML 5.0 earthquakes in Italy
based on spatially smoothed seismicity. Ann Geophys 53(3):5161
Albarello D, DAmico V, Peruzza L (2013) D6.1 report on model validation procedures. DPC-
INGV- S2 project 20122013. https://sites.google.com/site/ingvdpc2012progettos2/deliver-
ables/d6
Ameri G, Baumont D, Gomes C, Le Dortz K, Le Goff B, Martin C, Secanell R. (2015) On the

choice of maximum earthquake magnitude for seismic hazard assessment in metropolitan
France insight from the Bayesian approach. 9ime Colloque Nat. AFPS, Marne-La-Valle
Bommer JJ, Scherbaum F (2008) The use and misuse of logic trees in probabilistic seismic hazard
analyses. Earthquake Spectra 24(4):9971009
Carbon D, Drouet S, Gomes C, Leon A, Martin Ch, Secanel R (2012) Initial probabilistic seismic
hazard model for Frances Southeast . SIGMA deliverable n D441
Denieul M (2014) Seismic moment magnitude and Crustal Wave Coda. EOST PhD thesis (Institut
de Physique du Globe de Strasbourg)
Denieul M, Sbe O, Cara M, Cansi Y (2015) Mw estimation from crustal coda-waves recorded on
analog seismograms. Bull Seismol Soc Am 105(2A):831849
EPRI (Electric Power Research Institute) (1994) The earthquakes of stable continental regions
volume 5: seismicity database programs and maps, Report EPRI TR-102261-V5. Electric
Power Research Institute, Palo Alto
Faccioli E (1992) Selected aspects of the characterization of seismic site effects, including some
recent European contributions. In: Proceedings of international symposium on the effects of
surface geology on seismic motion (ESG 1992), vol I, pp 6596, March 2527, 1992, Odawara,
Japan
Faccioli E, Vanini M, Villani M, Galadini F (2012) Preliminary PSHAs at selected sites, based on
state of the art earthquake source models and attenuation relationships, considering both out-
cropping bedrock and soil conditions. SIGMA deliverable n D429
Faccioli E (2013) Recent evolution and challenges in the seismic hazard analysis of the Po Plain
region, Northern Italy. Bull Earthq Eng 11(1):533
Faccioli E, Paolucci R, Vanini M (2015) Evaluation of probabilistic site-specific seismic hazard
methods and associated uncertainties, with applications in the Po Plain, Northern Italy. Bull
Seismol Soc Am 105(5):27872807
Gardner JK, Knopoff L (1974) Is the sequence of earthquakes in southern California, with after-
shocks removed, poissonian? Bull Seismol Soc Am 64(15):13631367
Gutenberg B, Richter CF (1956) Seismicity of the earth and associated phenomena, 2nd edn.
Princeton University Press, NewYork, p310
Helmstetter A, Kagan YY, Jackson DD (2007) High-resolution time-independent grid-based fore-
cast for M #5 earthquakes in California. Seismol Res Lett 78:7886
lations. In: Specific Safety Guide SSG-9. International Atomic Energy Agency, Vienna
Letort J, Vergoz J, Guilbert J, Cotton F, Sebe O, Cano Y (2014) Moderate earthquake teleseismic
depth estimations: new methods and use of the comprehensive nuclear-test-ban treaty organiza-
tion network data. Bull Seism Soc Am 104(2):593607
Letort J, Guilbert J, Cotton F, Bondar I, Cano Y, Vergoz J(2015) A new, improved and fully auto-
matic method for teleseismic depth estimation of moderate earthquakes (4.5<M<5.5): applica-
tion to the Guerrero subduction zone (Mexico). Geophys JInt 201(3):18341848
Marsan D, Lenglin O (2008) Extending earthquakes reach through cascading. Science
319:10761079
Marsan D, Lenglin O (2010) A new estimation of the decay of aftershock density with distance to
the mainshock. JGeophys Res 115(B09302):116
Meletti C, Galadini F, Valensise G, Stucchi M, Basili R, Barba S, Vannucci G, Boschi E (2008) A
seismic source model for the seismic hazard assessment of the Italian territory. Tectonophysics
(online version) 450:85108
Moschetti M, Petersen M (2012) Smoothing methods for background seismicity, CEUS in 2014
National Seismic Hazard Map. In: CEUS workshop 2223 Feb 2012, Memphis, TN
Motazedian D, Atkinson GM (2005) Stochastic finite-fault modeling based on a dynamic corner
frequency. Bull Seismol Soc Am 95(3):9951010
NRC (2012) Practical implementation guidelines for SSHAC Level 3 and 4 hazard studies,
NUREG-2117, Rev. 1. NRC, Washington, DC
References 55
Reasenberg P (1985) Second-order moment of central California seismicity, 19691982. JGeophys

Res 90(B7):54795495
Rovida A, Camassi R, Gasperini P, Stucchi M (eds) (2011) CPTI11, the 2011 version of the para-
metric catalogue of Italian earthquakes. Milano, Bologna, Italy. http://emidius.mi.ingv-it/CPTI
SHARE (2013) Seismic hazard harmonization in Europe. http://share-eu.org
Weichert DH (1980) Estimation of the earthquake recurrence parameters for unequal observation
periods for different magnitudes. Bull Seismol Soc Am 70(4):13371346
Youngs RR, Coppersmith KJ (1985) Implications of fault slip rates and earthquake recurrence
models to probabilistic seismic hazard estimates. Bull Seismol Soc Am 75:939964
Chapter 4
Rock Motion Characterization
4.1 Empirical Models andPoint Source Stochastic Models
As stated in Chap. 2, rock (as well as soil) ground motion characterization for PSHA
requires that both the median response spectral acceleration and its standard devia-
tion (aleatory variability) be estimated by appropriate algorithms, such as GMPEs
or stochastic models. In Sect. 2.5 of Chap. 2 the logic tree approach was introduced
for handling epistemic uncertainty, pointing out that the dynamic characteristics of
the earthquake source and wave propagation near the site are typical sources of
uncertainty in ground-motion prediction. Some implications for the logic tree treat-
ment of epistemic uncertainty of rock ground motion will be discussed at the end of
this chapter.
It is also recalled from Chap. 2 that the horizontal ground motion component is
described as the geometric mean of the two orthogonal horizontal components: the
aleatory variability of these components with respect to the geometric mean is
incorporated in the set of acceleration time histories eventually selected to represent
ground motion on rock. The vertical motion component can be estimated in differ-
ent ways, see Sect. 4.5.
Ground-motion models can be developed through empirical GMPEs, point-
source stochastic simulations, finite fault simulations, and the hybrid empirical
method, but since only the first two of these were addressed in SIGMA, the attention
here will be focused on them.
4.1.1 Empirical Ground Motion Attenuation Models
Empirical GMPEs have the advantage of being best fits of recorded data sets, often
global ones. However, although their number has increased rapidly in the last few
years, in some cases these published equations may not be well adapted to the

58 4 Rock Motion Characterization
Table 4.1 GMPEs for 0.05 damped response spectral ordinates selected for the final SIGMA
PSHA South-East quarter of France (Ameri 2015)
Region Mw Distance range Period Site Fault
GMPE (dataset) range (km) range (s) description mechanism
Ameri Europe- 3.07.6 RJB,EPI = 0200 PGA; 4 site Yes
(2014) middle East + 0.013 classes
(D2131) French and
Swiss data
Drouet and French data 38a REPI = 1250 0.013 Rock or No
Cotton (Stochastic hard rock
(2015) model)
Akkar et al. Europe- 4.07.6 RJB,EPI, HYP = 0200 PGA; Function of Yes
(2014a, b) middle East 0.014 VS30
Bindi etal. Europe- 4.07.6 RJB,HYP = 0300 PGA; Function of Yes
(2014) middle East 0.023 VS30
Boore etal. Global 3.58.5 RJB = 0400 PGA; Function of Yes
(2014) (mostly 0.0110 VS30
California)
Cauzzi Global 4.57.9 RRUP = 0150 PGA; Function of Yes
etal. (mostly 0.0110 VS30
(2015) Japanese)
a
This GMPE is developed from weak motions and extended to large magnitudes with numerical
simulations
magnitude and distance ranges of most interest for hazard analysis with, e.g., low
annual probabilities of exceedance. This can happen for low or moderate seismicity
regions, like much of France; for regions where hazard is dominated by specific
source and propagation path combinations, like in the sedimentary plains of
Northern Italy; or for hard-rock site conditions that may prevail in SCRs and moun-
tainous zones in ASCRs. To bypass at least in part these limitations, regionally-
oriented prediction equations were developed in SIGMA for:
(1) the French context, with special attention devoted to small magnitude earth-
quakes, see D292 and D2131;
(2) Northern Italy, with focus on the dominant combination of thrust fault sources
and the deep sedimentary basin that characterizes the Po Plain, see D253 and
D2133, and
(3) hard rock conditions in ASCRs (D3150).
In this Chapter items (1) and (3) will be discussed, among others. In connection
with (1), the GMPEs used in the SIGMA final PSHA of the South-East quarter of
France on exposed rock are shown in Table4.1, with the salient features associated
to each equation. Trellis plots of the spectra estimated by these equations for differ-
ent magnitudes and distances (see Fig.4.1, where the Joyner and Boore distance,
4.1Empirical Models andPoint Source Stochastic Models 59
Fig. 4.1 Median response spectra predicted by the GMPEs of Table4.1 for several M-RJB distance
scenarios for VS30 = 800 m.s1 and strike-slip fault. An equivalent rupture distance (Rrup) was calcu-
+ ( 0.3046 + ztor ) as proposed
2
lated for the Cauzzi etal. (2015) model following log Rrup = 2 RJB2
by Cauzzi etal. (2015) for Mw > 5.5, where ztor is the depth to the top of fault rupture. ztor = 4km
and =1km is used for Mw = 6 and Mw = 7, respectively. For Mw = 4 and 5, Rrup is taken equivalent
to the hypocentral distance assuming a point-source at 10km depth (From Ameri 2015)
RJB, is used) are instructive, as they display that the variability decreases from a
factor of 3 at small magnitudessmall distances to less than a factor of 2 at M 7.0 for
any distance. In the selection of the GMPEs of Table4.1 special consideration was
given to models that: (a) use French data in some sense, (b) were recently developed
for the pan-European region, (c) use a worldwide dataset (to better constrain ground
motions from large magnitude events), or (d) have a functional form well suited for
area source models.
Concerning the characterization of rock sites, Fig.4.2 shows the distribution of
VS30 (the time-averaged shear-wave velocity of the top 30 m) values available in the
dataset extracted from the RESORCE 2013 European database (see D2-91) that was
used to derive the Ameri 2015 (D2-131)), Akkar etal. (2014a, b), and Bindi etal.
(2014) GMPEs of Table4.1.
Fig. 4.2(top) Distribution of the VS30 values associated to records in the updated (2013) European
database (see SIGMA D291), which is the basis of the GMPEs developed for the French context
in D2131. This corresponds to roughly 40% of the total records in the database. (bottom) Same
for the KiK_net stations on hard ground/rock used in D3-150: on the left, the number of surface
stations corresponding to the selected VS30 values and, on the right, the number of DH stations
belonging to the ranges of the VS values measured at the depth of the instrument, denoted as VShole
(from D3-150)
The recording sites with values VS30 800 m.s1, ascribable to the category of
rock sites (ground type A in the EC8 classification), are a small fraction of the total.
As a further example, the DBN2 dataset used to develop GMPEs for Northern Italy
(see D2-72) includes 95 sites in category A, with 610 records, but only at 4 sites
(with 32 records) were measured VS30 values available, while all other sites of the
same category have been classified according to a geological description, of admit-
tedly low quality (denoted as A*). Relatively small proportions of rock-site records
are found in practically all recent GMPEs datasets, including those used in the so-
called integrated exercise of SIGMA, described in D4-153; most are included in
Table4.1. As an example, representative for the four ASCR GMPEs applied in the
integrated exercise, the Cauzzi etal. (2015) model used a global database where
rock records with a measured VS30 are less than 5% of the total.1
Thus, the estimation of ground motions at surface rock sites from GMPEs is, as
a rule, more weakly constrained by data than at soil sites of various categories.
Moreover, the fraction of surface rock records in ASCR databases tends to vanish as
VS30 approaches the limit of 1500 m.s1, which the U.S. standards define as the lower
1
A reliable attribution to the rock site category requires measured VS30, as attributions based on
field survey or geological maps are likely to be inaccurate.
4.1Empirical Models andPoint Source Stochastic Models 61
threshold for hard rock. On the other hand, exposed bedrock formations in SCRs
often belong to the very hard rock site category, with VS30 > 2000 m.s1, and even
in those among such regions where monitoring of seismic activity is reasonably
good, like Eastern North America (ENA), sites with measured S-wave velocities are
scarce. For instance, for the Mw 5.4 2005 Riviere du Loup, Qubec earthquake, one
of the best recorded SCR events ever, S-wave velocities are not available for the
more than 20 stations that recorded the ground motion within 100km of the source
(Assatourians and Atkinson 2010).
To estimate ground motion at hard rock sites in ASCRs, one can follow the path
referred to in previous item (3), which exploits data recorded at depth. Some GMPEs
had been developed in earlier studies from such recordings, notably those of the
KiK-net, as the stations of this network feature both a surface and a down-hole
3-component accelerometer on the same vertical profile (e.g., Rodriguez-Marek
et al. 2011). The down-hole (DH) KiK-net recordings were jointly exploited in
SIGMA and in SINAPS@, see D3-150, leading to a significant enlargement of the
database for ground motions observed on rock, as illustrated in the bottom graphs of
Fig. 4.2. However, DH records do not represent incident motions (as they are
affected by interference of waves reflected at the free surface) and do not account
for the free-surface effect; hence, they require a correction to make them suitable for
SH applications. Laurendeau etal. (D3-150) corrected the KiK-net DH recordings
as indicated by Cadet et al. (2012a, b), and were thus able to extend the ground
motion estimates across the VS30 range shown in the bottom right graph of Fig.4.2,
as further discussed in Sect. 4.3. To remedy in part the deficiency of hard rock
records, other approaches have been devised, which are discussed in more detail in
Sect. 4.3.
In low-seismicity regions, like much of France, events capable of generating
significant ground motions over a sizable area (with Mw > 5) are rare and the need
arises to use data from low-magnitude events, down perhaps to 3.03.5 (see
Table4.1). This allows, in principle, a validation against the PSHA results at short
return periods (typically of the order of 30 years), comparable with the time inter-
vals for which some accelerometer stations have been in operation.
Since many published GMPEs have a lower threshold magnitude of 4.0 or 4.5,
attenuation models extending to lower magnitudes were developed in SIGMA-WP4,
as illustrated in D-92 and D-131, benefitting from the significant amount of small
magnitude French and Swiss earthquake data present in RESORCE 2013. To
account for observed correlations of the prediction residuals with the Brune stress
drop () of the generating events, a stress parameterdependent term was added in
the prediction equations, strongly affecting the high frequency portions of the spec-
tra for Mw 5.0, and constrained to vanish for Mw > 6.0. This term accounts for
regional differences in stress parameter and decreases the standard deviation of the
GMPE.Obviously, for application to different French regions average stress param-
eters (and their uncertainty) are needed, taking into account that available estimates
span across more than two orders of magnitude and that uncertainties at play are
typically quite large. Perhaps the effect of the stress parameter could be handled in
a more standard way using style-of-faulting terms instead of , considering that

the different regions have different dominant mechanisms.
In closing this section, it is noted that at sufficiently long structural periods the
response acceleration ordinates estimated on rock by physically sound GMPEs
should exhibit a T2 dependence, at least to some degree, so that when they are con-
verted to displacement response ordinates, constant values are obtained. The corner
period beyond which a roughly constant spectral displacement should be observed
strongly depends on magnitude. A number of GMPEs still in use do not comply
with this requirement.
4.1.2 Point Source Stochastic Models
The attenuation models derived from point source stochastic simulations are typi-
cally built in two steps, as was done in the SIGMA application to the French context
(see D2-71). First, synthetic acceleration waveforms are computed using a stochas-
tic model simulation tool, such as SMSIM (Boore 2003). Second, the synthetic data
are used to build a GMPE by regression analysis, assuming a functional form.
The point source stochastic model is the simplest numerical simulation tool
available that is consistent with seismological theory. The main input parameters
include earthquake magnitude, stress drop (), regional attenuation parameters
[chiefly geometrical spreading exponent (in r-) and quality factor Q], crustal
velocity and density profiles or a site transfer function, and the near-site high-
frequency attenuation factor . Region-specific models of stress drop and attenua-
tion are determined empirically from recordings of smaller earthquakes in the
regions of interest.
Alps, Pyrenees, and Rhine Graben tectonic provinces were thus assigned differ-
ent values in the simulations for the French context, where regional values of
increasing with magnitude were adopted for magnitudes 5.0, and a single constant
value (2.5, 5.0, and 10 MPa) for larger magnitudes. The regional geometrical
spreading parameters are all close to the unit value of body wave attenuation, and
quality factors range from about 300 (Alps) to about 800 (Pyrenees and Rhine
Graben). While it is common to have a term in GMPEs to account for the different
style-of-faulting types, in the France application the simulations did not include
such a term.
Two types of sites are considered: one corresponding to standard rock with
VS30 = 800 m.s1 and = 0.03 s, and the other to hard rock with VS30 = 2000 m.s1
and = 0.01 s. In order to quantify the uncertainty in the predictions derived from
the point-source model, all the input parameters to the stochastic simulations for
France were considered as random variables with normal or log-normal distribu-
tions, and the uncertainty on such parameters was propagated to the synthetic
ground motions, allowing for a sensitivity analysis to be carried out to evaluate the
influence of the uncertainty on each input parameter to the total GMPE uncertainty.
4.2Model Selection andCriteria 63
The major contributors to the total uncertainty are the stress parameter model and
the site model (both site amplification and kappa), while the uncertainties on the
attenuation parameters have a second order influence, and the remaining ones are
negligible.
The GMPEs coefficients of the attenuation model derived from the stochastic
simulations for France are available for different distance metrics (Repi, Rhypo, RJB,
and Rrup). The point-source simulations work well for small earthquakes and at large
distances, where they are comparable to finite fault simulations, but for large earth-
quakes and close to the fault, some adjustments to the single distance used in the
point-source simulation were introduced in D2-71 to mimic near-source effects
such as the saturation at close distances.
In carrying out the study for France much effort went into the characterization of
the models for large earthquakes and large distances, although it is known that the
seismic hazard in France will be mostly coming from small to medium events at
distances between 25 to 50km. A test in a full PSHA of France is not currently
available.
4.2 Model Selection andCriteria
Careful decisions are required in the selection of applicable GMPEs for PSHA stud-
ies, notably in low seismicity regions with few or no strong motion records avail-
able, and limited seismological data. A preliminary requirement is that of peer
review: attenuation equations that have not been published in a peer-reviewed jour-
nal should in principle be discarded, but flexibility is needed, e. g. to admit the use
of robust equations derived for specific projects or regions and published as a tech-
nical report. The screening of suitable GMPEs should satisfy different sets of crite-
ria; to simplify the criteria, these can be grouped in the following three types
(assuming that the selection will consider only updated models published in the last
few years):
Modelling criteria;
Tectonic consistency; and
Site-conditions consistency.
4.2.1 Modelling Criteria
The requirements on the functional forms chosen for the GMPEs and on the tech-
niques used for regressing the datasets are extensively discussed in Cotton et al.
(2006) and Bommer etal. (2010). Based on the GMPE development work carried
out in SIGMA, attention will be focused here on the predictor variables used in the
empirical models, principally the magnitude, the source-to-site distance, a descrip-

tor of the near-surface geological material, and the style of faulting.
For magnitude, the scale employed in the equation should be consistent with that
used to define earthquake activity rates in the seismic source model. Moment mag-
nitude, Mw, is a standard descriptor in current GMPEs: if magnitude conversion
relations such as ML Mw, MS Mw, or mb Mw are used, typically because Mw is not
available for smaller magnitude earthquakes (e. g. <~4.5), the variability associated
with these relationships should be propagated into the sigma value of the GMPE for
which the adjustment is made. An approach that would remove the excess variabil-
ity introduced by conversions is to use GMPEs that rely on datasets where no con-
versions are made, e.g. where ML (or mb) is retained for the records of the smaller
events (if available) and Mw for the larger ones, and proceed in the same way in the
definition of activity rates. The no-conversion strategy was adopted to develop
GMPEs for Northern Italy (in D2-72 and D2-133), where the resulting between-
event variability is larger (especially for T > 0.2 s) in the case with conversion.
Similar considerations apply for the different distance metrics: Repi, Rhypo, RJB,
and Rrup. The key here is to perform the PSHA with a hazard code capable of model-
ling extended seismic sources in such a way that each of the foregoing distance
measures can be computed and each equation used with its native distance measure.
In practice, it is important to recall that Repi can be assimilated to RJB and Rhypo to Rrup
for Mw < ~5.5. This was done in D2-72, while in the French application of D2-131
preference was given to Repi alone.
In most current GMPEs a descriptor of the near-surface geological materials is
also included, either as VS30 or as an EC8 site category. SIGMA-based empirical
GMPEs for France and Italy use the EC8 site categories, allowing the inclusion in
the dataset of a sufficient number of data points. Regarding the contribution to
uncertainty, using the EC8 site categories is equivalent to using a VS30 characteriza-
tion, see D2-92. The situation is different for the French stochastic model (D2-71),
which is applicable only to rock sites (both firm and hard).
The influence of the style-of-faulting (SOF) predictor, invariably introduced
through classes of fault mechanisms (normal, strike-slip, and reverse), is significant
only for vibration periods up to roughly 1 s; it is rather limited for strike-slip mecha-
nisms and can account for an increase (reverse w.r.t. strike slip) up to about 50% in
the predicted spectrum ordinates. Note that if the SOF is not used as a predictor in
a given GMPE, the prediction itself will not reflect an undifferentiated faulting
mechanism but the proportion of the different SOF types that are present in the sup-
porting event dataset.
4.2.2 Tectonic Consistency
As stated in Bommer etal. (2010) the first basis for exclusion of a model is that it
is from a tectonic region that is not relevant to the location of the site for which the
PSHA is being conducted, considering also that there is no strong evidence for
4.2Model Selection andCriteria 65
Fig. 4.3 Excerpt of the seismotectonic map for GMPE selection in the Euro-Mediterranean area,
developed by the SHARE project (http://diss.rm.ingv.it/share-edsf/SHARE_WP3.2) 1: SCR,
shield (a) and continental crust (b); 2: oceanic crust; 3: Active Shallow Crustal Regions:
compression-dominated areas (a) including thrust or reverse faulting, associated transcurrent fault-
ing (e.g. tear faults), and contractional structures in the upper plate of subduction zones (e.g. accre-
tionary wedges), extension-dominated areas (b) including associated transcurrent faulting, major
strike-slip faults and transforms (c), and mid oceanic ridges (d); 4: subduction zones with contours
at 50km depth interval of the dipping slab; 5: areas of deep-focus non-subduction earthquakes; 6:
active volcanoes and other thermal/magmatic features
persistent regional differences in ground motions amongst tectonically comparable

areas, at least in the range of moderate-to-large magnitude earthquakes. At play
here is the key distinction between ASCRs, SCRs and subduction zones which, in
the broad European context, translates into a map as shown in Fig.4.3, expressly
developed in the SHARE (2013) project as a basis for choosing GMPEs for hazard
analysis.2
GMPEs derived for SCRs and ASCRs differ in several respects, crucial among
which is that owing to the paucity of observations even at the global scale many
of those applicable in SCRs were derived for Central and Eastern North America
(CENA) from stochastic models either for point- or finite-fault sources, with limited
data from small and moderate earthquakes. Broadly speaking, response spectral
ordinates estimated by a SCR GMPE are larger and attenuation of their amplitude
with distance considerably slower (so that, for some of the models, the range of
applicability extends as far as 1000 km), which makes GMPEs for one type of
For continents other than Europe one can consult the maps in EPRI, Vol. 5.
2
region not interchangeable with those for the other. Criteria that can be applied for
the different types of crustal region in Europe can be found in Delavaud etal. (2012),
although today the choice of the specific GMPEs is possibly worth updating, as it
has been acknowledged that in most cases a GMPE needs to have a host-to-target
adjustment, as it is not directly applicable to the study region. Since in most parts of
France (i.e. excluding the Alps, Pyrenees, and the Rhine Graben), as well as in most
of Spain and the British Isles, the continental crust is believed to be of the extended
type (i.e. with reduced thickness with respect to older continental regions), using a
mixture of GMPEs for ASCR and SCR would seem reasonable for these regions.
The coupling of tectonic regionalization with different rock conditions prevailing at
the surface is discussed in the following.
4.2.3 Site-Conditions Consistency
It is fairly clear that the choice of the soil profile category (or VS30) in a GMPE
should be consistent with that of the site of application. The need for this check of
consistency typically arises with respect to the identification of the rock category
present at the application site. In the simplest terms, the empirical models that
include the generic rock category with VS30 = 800 m.s1, like those developed in
SIGMA, should not be applied without correction to hard rock sites with VS30 >
1500 m.s1, see Fig. 4.2. The so called host-to-target corrections that may be
required to make the application possible are discussed in detail in the next section.
While geological field assessments can be useful for the classification of rock sites,
they should in general not replace in-situ wave speed measurements for a correct
determination of VS30.
4.3 Corrections or Modifications ofPublished Models
In many stable regions, such as large parts of northern Europe, no strong motion
data are available and, hence, no specific GMPEs exist for these regions. Moreover,
hard rock sites occur frequently in mountainous areas of ASCRs. The associated
high VS30 values are not consistent with those of most rock sites where records are
available or with the rock definition used in Eurocode 8. Therefore, some adjust-
ments, or corrections, to ground-motion predictions are needed to account for the
different rock site conditions. Hence, either some adjustment to ground-motion pre-
dictions should be introduced to account for the different rock site properties, or
recourse to other instrumental observations must be sought. The extent of the adjust-
ments depends on the rock type, hardness and erosion of the in-situ formations. We
consider in the following two alternative approaches, both used (and the second
even developed) in SIGMA.The first approach consists of applying a standard type
of correction, based on the high-frequency attenuation parameter and VS30 and
4.3Corrections or Modifications ofPublished Models 67
quantified through numerical simulations, while the second one (which became
available only at the end of SIGMA) relies on recorded data, used as a basis for
direct ground motion estimations through GMPEs.
4.3.1 -VS30 (Simulation-Based) Correction
In the PEGASOS Refinement Project (Biro and Renault 2012) and the Thyspunt
Nuclear Siting Project (Rodriguez-Marek etal. 2014), empirically predicted motions
have been corrected by a theoretical adjustment factor depending only on the two
site parameters, and VS30; the same method has been used in the SIGMA Integrated
Exercise (case study of a site-specific probabilistic model), see D4-153. governs
the high-frequency decay of the ground acceleration Fourier amplitude spectrum for
frequencies higher than a specific frequency fe, in the form exp(f) (Anderson
and Hough 1984), assuming that the effective quality factor (i.e., the overall attenu-
ation) is frequency independent. To account for the observed dependence of on the
source-to-site distance, R, Anderson and Hough (1984) suggested the linear rela-
tion, in units of time:
= 0 + R R ( s ) (4.1)

Ktenidou etal. (2013) observe that the intercept 0 at zero distance corresponds
to the attenuation S-waves encounter when travelling vertically through the shallow
geology, and the slope of the trend (R) corresponds to the incremental attenuation
due to predominantly horizontal S-wave propagation through the crust, and report
estimates of 0 from rock site records in different regions from 0.02 to 0.04 s.
To interpret the features of Fourier amplitude spectra corrected through -scaling,
it is useful to derive a simple analytical expression of the correction. Let for this
purpose the amplification function at the surface of a rock site be expressed as:
AFR = AR ( f ) exp ( R f ) (4.2)

where AR is the amplification factor for waves propagating through the velocity and
the density profile assumed to be representative for the site at hand, such as that of
Atkinson and Boore (2006, their Table4) with VS30 = 760 m.s1, coupled with =
0.03 s [average in the range indicated by Ktenidou etal. (2013)]. This profile fea-
tures VS = 3700 m.s1 at the average earthquake focal depth of 1015km. Likewise,
for a hard/very hard rock site, HR, one can write
AFHR = AHR ( f ) exp ( HR f ) (4.3)

where AHR is the corresponding amplification factor, for example that associated
with the very hard rock profile of Boore and Joyner (1997, their Tables 2 and 4)
which has a VS30 = 2780 m.s1. Coupled to this value is = 0.005 s, mean of a
Fig. 4.4 Rock/very hard rock response spectral correction factor as a function of structural period,
from different sources discussed in text. The green curve, corresponding to Eq. (4.4), depicts a
Fourier amplitude ratio, to be read as a function of 1/f
uniform distribution between 0.002 s and 0.008 s in Atkinson and Boore (2006).
Dividing (4.2) by (4.3) the rock/hard rock correction factor for the Fourier ampli-
tude spectrum is obtained in the form
AR ( f ) ( R HR ) f
C R / HR ( f ) = e (4.4)
AHR ( f )

Taking for AR (f) and AHR (f) the foregoing amplification factors and for thevalues
just introduced, the green curve labelled as theoretical in Fig. 4.4 is obtained,
which tends to zero as T 0, i. e. f .
The red curve and associated variability band plotted in Fig.4.4 as a function of
the structural period T represent (generic rock)/(hard rock) correction factors for
response spectra, and were derived by Van Houtte etal. (2011), using a VS30- cor-
relation from Japanese Kik-Net data and the hybrid adjustment method of Campbell
(2003), assuming the host and target regions to be the same but with different rock
site conditions. The host region is the one for which a GMPE exists, and which can
be described by seismological parameters (stress drop, quality factor, , etc.). The
same set of parameters must be available for the target region, for which a GMPE is
being sought. Then using a simulation tool such as SMSIM, mentioned in Sect.
4.1.2, synthetic ground-motions from the seismological parameters can be com-
puted for the two regions and response spectra derived. The ratios between the spec-
tra of these synthetic predictions for given magnitudes and distances are used to
adjust the original GMPE to the target conditions.
To produce the smooth envelopes in Fig.4.4 Van Houtte etal. (2011) adopted
different magnitude-distance combinations and used for the host region the
Campbell (2003) GMPE and related ENA seismological parameters, i.e. 2000 <
VS30 < 2800 m.s1 and 0.002 0 0.012 s; the target generic rock site parameters
were VS30 = 800 m.s1 and 0.02 0.05 s. The theoretical correction curve in
Fig.4.4, although a ratio of amplification functions in the Fourier amplitude domain
and not a response spectral ratio,3 for T > 0.025 s appears to explain qualitatively the
salient features of the correction factor derived from host-to-target numerical simu-
lations. In essence, the simulation-based correction factor of Fig.4.4 tells us that at
high frequencies, a hard rock site should amplify more than a normal rock site due
to its lower attenuation. However, this indication is associated with large uncertain-
ties, which may have large effects on the resulting ground motions. A strong reason
is that observed transfer functions of stiff surface sites characterized by VS30 > 500
m.s1 (and even as high as 850 m.s1) exhibit amplification peaks, especially at high
frequencies, related to local effects that are not considered in the adjustment factor
computation (D3-150). The consequences of this are illustrated in the next
sub-section.
Figure 4.4 shows two additional curves, labelled Method 1 and Method 2 SSS,
borrowed from the SIGMA Integrated Exercise (D4153). These curves represent
ratios of the 10,000 year spectrum on generic rock (VS30 = 800 m.s1), estimated by
the single-station sigma (SSS) approach, with respect to the spectrum on very hard
rock at the same site (with VS30 = 3200 m.s1), by the two different methods. Both
curves, resulting from the combined use of four different GMPEs, fit for the most
part within the Van Houtte etal. bounds. Thus, if we believe the numerical simula-
tions, these bounds seem to encompass most of the spread caused by the epistemic
uncertainty in the spectrum adjustment between different representative rock sites.
However, VS profiles at real rock sites may depart significantly from those used in
the simulations so that the amplification factor ratio in (4.4) may tend to dominate
the kappa-scaling factor and to attain values significantly greater than unity (gener-
ated by amplification of rock with respect to hard rock) at high frequency.
Kappa-scaling of attenuation models from one rock type (host) to another (tar-
get) can be managed in a more site-specific way, as was done in the SIGMA
Integrated Exercise, using the two methods described in D2-130 (Bora method) and
in Al Atik et al. (2014), respectively. The first one takes a frontal approach and
works by developing GMPEs from a given record database directly in the Fourier
domain, applies the desired host-to-target correction in such domain, and then
converts the result into the response spectra domain via RVT.The method by Al
Atik, simpler to apply, differs basically in the first step, in that it converts host
response spectra into the Fourier amplitude spectrum domain via inverse RVT tools,
while the following two steps are conceptually similar to those of the Bora method.
The ratio of the initial (host) to the final, modified response spectrum represents the
scaling factor. In the SIGMA integrated exercise, the application of the Bora
3
The main difference being of course that for T 0 the response spectral ratio tends to the PGA
ratio, where T is the structural period.
Table 4.2 Values of , in s, estimated for generic rock sites (VS30 = 800 m.s1) from the indicated
GMPEs: AB11 = Atkinson and Boore (2011), ZEA06 = Zhao etal. (2006), CEA14 = Cauzzi etal.
(2015), AM14 = Ameri (2014) in D2-131
AB11 ZEA06 CEA14 AM14
0.0400.044 0.0510.054 0.0250.026 0.035
method has been partial, because it could be correctly used only with the GMPEs
developed from the RESORCE strong motion database, while databases underlying
the other GMPEs where not directly available.
The so-called adjustment method 2, by Al Atik etal. (2014) does not require
the selection of a number of the parameters involved in method 1, that are not con-
strained by data, and is easier to apply. An asset of the method is that it does not
require seismological models for the stochastic parameters (stress drop, whole-path
attenuation, etc.) of the host and target regions; the ground motion duration is a criti-
cally influent parameter in applying the RVT.The response spectra used as a point
of departure for the adjustment procedure are generated with the selected GMPEs
as scenarios at representative magnitudes, distances, and ground motion durations4;
it is, therefore, important that the host values estimated from such scenarios be rea-
sonably stable, in order for the kappa-scaling to be meaningful. Instability in the host
estimates may point to metadata of insufficient quality in the dataset underlying the
selected GMPE. Table 4.2 shows the host values estimated by the first steps of
Method 2 from a few GMPEs, with scenario magnitudes 5 and 6 and Repi 10 and
20km. The spread of values in the table is rather large; part of it may depend on
how the rock site attribution is handled by the GMPE, for example AB11 and
CEA14 directly use the assigned VS30 value, while ZEA06 and AM14 use differently
defined site classes. Moreover, the underlying datasets may reflect different domi-
nant features of rock formations.
However, since average 0 values estimated for normal rock sites by different
researchers vary between 0.02 s and 0.04 s (Ktenidou etal. 2013, their Figure14),
and since the effect of R in (4.1) is probably negligible within source-to-site dis-
tances of few tens of km (Ktenidou etal. 2013), widely used GMPEs like AB11 and
ZEA seem to reflect near-site attenuation effects that are actually outside the
expected rock site range, while a model like CEA14 sits near the middle of that
range. These differences should also be considered vis--vis their logic tree
implications.
4
The inverse of the corner frequency can be taken as duration at short source distances (<10 km),
while for larger distances a path duration equal to 0.05 (epicentral distance), in s, can be added,
see Al Atik etal. (2014).
4.3.2 Data-Based Predictions forHard Rock
Laurendeau et al. (D3-150), working with carefully selected subsets of KiK-net

data, have produced tools for approaching the estimation of GM at sites with VS30 >
1500 m.s1 in a different way, i.e. by extending the available datasets of surface
recorded data with recordings at depth obtained in vertical arrays. Specifically, they
derived empirical GMPEs from the two separate KiK-net data subsets depicted in
the two bottom graphs of Fig.4.2 with respect to their Vs distribution: at the left is
shown the distribution of the surface records, for which 500 m.s1 < VS30 < 1500
m.s1. The downhole (DH) sites, corresponding to the bottom right graph in Fig.4.2,
are characterized by their velocity at depth, denoted as VS hole, which allows expand-
ing the velocity distribution to more than 3000 m.s1. The distribution median of
velocities is around 650m s1 at the surface and around 1900m s1 at depth. All the
records were generated by shallow crustal events in Japan with a focal depth less
than 25 km, excluding offshore earthquakes.
Laurendeau etal. (D3-150) developed, among other things, two separate GMPEs
of simple analytical form for the two previous data subsets, with indicative applica-
bility ranges 4.5 Mw < 7.0 and 5km Rrup 200km. The DH response spectra of
the records effectively used in the regression, 1040in number, were subjected to the
correction devised by Cadet etal. (2012a, b) to account for the free surface effect,
and to remove the effect of destructive interference between upward- and downward
propagating waves. This manifests itself at the destructive frequency fdest = VS/4H
where H is the downhole sensor depth and VS the mean shear wave velocity of the
upper layers, and gives rise to a notable peak, usually visible both in the Fourier and
the response spectra surface/DH ratio. The reliability of the spectra corrected in this
way was carefully checked by different criteria, and it was assumed that wave prop-
agation at depth under the recordings sites is of a predominant 1D nature. For the
regression of the natural surface records the same dataset was selected.
One can thus use the Laurendeau etal. (D3-150) GMPEs to estimate for different
magnitude-distance scenarios the mean response spectra (and associated sigma) for
the representative (hard rock) host value VS30 = 2400 ms1 and for the target value
(rock) of 800 m.s1, associated with the envelope in Fig. 4.4. Two pairs of such
spectra are illustrated for two magnitude-distance scenarios in Fig.4.5.
The spectra of Fig.4.5 highlight the amplifying response of surface rock sites
with respect to hard rock sites in the KiK-net throughout the 0.03 to 2.0 s period
range, with peak amplification at around 0.08 s. This is at variance with the correc-
tion factor in Fig.4.4, which shows strong amplification of hard rock with respect
to rock sites at very short period and moderate de-amplification at longer periods.
The discrepancy is likely caused by the near-surface features of the VS profiles at
real rock sites in the KiK-net, typically exhibiting a low velocity layer of weathered
rock, with VS ~ 500 m.s1 in the upper 15m or so, but still compatible with a VS30 of
the order of 1000 m.s1 or more, as shown in Fig.4.6.
M 6.5 R=15 km M 5.5 R=10 km

10 10
1 1
SA(T)
SA(T)
0.1 0.1
0.01 rock (mean) 0.01
hard rock (mean)

0.001 0.001
0.01 0.1 1 0.01 0.1 1
T (s) T (s)
Fig. 4.5 Acceleration response spectra on exposed Rock sites (VS30 800 m.s1, solid red curves)
and Hard Rock sites (VS30 2400 m.s1, solid green curves) for the indicated magnitude rupture
distance (Rrup) combinations, calculated by the GMPEs of Laurendeau etal. (D3-150). These cor-
respond to regressions of the natural surface (rock) record set and of the corrected downhole (hard
rock) record set, having the S wave velocity distributions shown in the bottom graphs of Fig.4.2,
respectively. Dashed curves represent the mean 1 spectral levels
Thus, amplification effects at rock sites in the KiK-net seem to dominate over the
influence of . Laurendeau etal. (D3-150) found a mean DH 0 = 0.011 s 0.007 s,
but argue that, based on the similarity of mean spectral ratios obtained from empiri-
cal, generalized inversion and theoretical (1D) approaches at individual sites, the
correction associated with 1D propagation suffices to capture the whole station
specificity at the KiK-net vertical array scale (at most few hundreds of m from the
surface) in the frequency range 0.5 to about 12 Hz (or perhaps 15 Hz), without
requiring a 0 correction. This indication cannot yet be extrapolated to higher fre-
quencies, because of variability of the scaling of the QS quality factor with fre-
quency, which remains unresolved.
The lesson to be drawn from the comparison of the two approaches just pre-
sented is rather problematic: relying only on simulation-based, smooth envelopes of
the type shown in Fig. 4.4 to perform host-to-target ground motion adjustments
between different classes of rock sites may not be advisable, in view of the evidence
shown in Fig. 4.6. In ASCRs, at target rock sites with VS30 values significantly
exceeding those of EC 8 ground category A, direct spectrum predictions should be
independently derived, e.g. with the Laurendeau etal. (D3-150) GMPEs and, at the
same time, adjustments factors should be obtained from 1D propagation analyses
with realistic, site-specific VS profiles.
In closing this overview, one should not forget that a number of parameters, such
as stress drop, strong motion duration, or Q, as part of the initial dataset, are usually
implicitly built in into an attenuation equation and there is no straightforward way
to correct them to make the equation more site-specific. Thus, users may have to
decide whether to adapt an existing GMPE or to develop a completely new one.
4.4Standard Deviation ofModel Predictions; Truncation 73
Vs (m/s)
0 500 1000 1500 2000 2500 3000
0
10
20
30
40
z (m)
50
60
70
80
90
100
Fig. 4.6 Illustration of the different shallow-depth features of real vs. smooth shear wave velocity
profiles at rock sites. Shown are: average of 9 KiK-net Vs profiles, with VS30 = 1290 m.s1 (red
curve); smooth profiles derived by Cotton etal. (2006) from the Boore and Joyner (1997) generic
rock profiles, for VS30 = 1200 m.s1 (green curve) and 1500 m.s1 (blue curve); average of 5 profiles
(from cross-hole measurements) at Italian rock sites (Faccioli 1992), merged with the Cotton 1500
profile at 17m depth (black curve)
4.4 Standard Deviation ofModel Predictions; Truncation
PSHA studies have traditionally been using the attenuation equations in an ergodic
fashion, i. e. associating to the median log predictions of the ground motion param-
eters (log Y) their standard deviation, logY, derived from datasets where many differ-
ent earthquakes and different sites are present. When applied to a single site, the
ergodic assumption implies that the variability across different sites is interchange-
able with that resulting from many different events, which may lead to overestimat-
ing logY.
Residual analyses of the spectral accelerations predicted by GMPEs, applied to
extensive regional datasets, have actually shown that when one considers individual
sites with recorded data and the associated (non-ergodic) statistical measures of
variability, the range of the key uncertainties at play may diminish with respect to
the ergodic case.
For ease of reference, Table4.3 summarizes terms and notation pertinent to this
topic, starting from the total residual Rij, i.e. the difference between the ground
motion parameter observed in earthquake i at station j and the corresponding value
predicted by a GMPE.While for detailed definitions of the meaning of the different
terms the reader is referred to Rodrguez-Marek etal. (2011) and Rodrguez-Marek
etal. (2013), due attention should be paid to the site term S2S, which is one of
thetwo components of the within event residual Wij. This term, in the words of
Table 4.3 Components of total residuals of GMPE predictions and of their and standard deviations
Residual Standard deviation
components Notationa components Notation
Total residual Rij = Bi + Wij Total standard
deviation = 2 +2
Between-event Bi Between-event
residual standard deviation
Within-event Wij Within-event
residual standard deviation
Site term S2Sj Site-to-site s2s
variability
Event and site W0,ij = Wij S2S Event corrected ss
corrected single- station
residuals standard deviation
Event corrected ss,s
single-station
standard deviation at
individual site
Total single-station
standard deviation SS = 2 + SS2
Total single-station
standard deviation at SS,S = 2 + SS2 ,S
individual site
After Rodriguez-Marek etal. (2013)
a
Indices i and j denote earthquake and site, respectively
AlAtik etal. (2010) represents the systematic deviation of the observed amplifica-
tion at this site from the median amplification predicted by the model using simple
site classification such as the average shear-wave velocity in the uppermost 30m at
the site, VS30.
The term W0,ij, on the other hand, describes the record-to-record variability of
the response at site j for earthquake i. The standard deviation, ss, of the residuals
W0,ij associated with a given dataset, is commonly referred to as single-station
sigma; it displays limited variation with respect to magnitude and distance across
widely different regional datasets and tectonic environments (Chen and Faccioli
2013; Rodrguez-Marek etal. 2013). ss is generally smaller than the ergodic within-
event component of the sigma of the prediction of a GMPE based on the same
dataset. On the other hand, the standard deviation of the between-event residuals
is significantly source-and-path-dependent; it is less easily constrained, but it can to
some extent be assimilated to the between-event variability of a regionally based
GMPE (if it exists).
In a (partially) non-ergodic sigma approach the aleatory variability and the epis-
temic uncertainty are separated, by assuming that the site term can be independently
calculated (through site response analysis and -scaling, if needed), and by associat-
ing an epistemic uncertainty S2S to it via either logic tree (Rodrguez-Marek etal.
2014), or engineering judgment (as in Faccioli etal. 2015). Rodrguez-Marek etal.
(2014) termed the resulting approach as semi-ergodic and stipulated the following
key requirements for its application in PSHA: (1) the median S2S should be
properly estimated, and both (2) the epistemic uncertainty S2S and (3) the epistemic
uncertainty in the single-station sigma ss should be taken into account. If, due to
lack of recorded data, the site term is independently assessed and S2S is explicitly
taken into account, the two remaining variability measures ss and can be com-
bined into a (total) single-site sigma ss, as shown in the last column of Table4.3.
Insight on variability of ground motion at rock sites, applicable in single-station
sigma approaches, was gained in SIGMA through the analyses carried out in the Po
Plain, Northern Italy, and (on a more limited scale5) at the Euroseistest site in
Greece. For the Po Plain, specific empirical attenuation models were developed,
benefitting from the vast increase of data generated by the damaging Emilia earth-
quake sequence of May and June 2012.
Figure 4.7 displays a map of the central Po Plain, containing the area most
affected by the 2012 earthquakes (and by the smaller, Mw 5.5, 1996 Reggio earth-
quake), with the accelerometer stations, subdivided into (deep) soil and rock catego-
ries. These include many of those used for developing the Emilia-specific GMPEs,
used in the PSHA for Po Plain sites discussed in D4-94. For geological reasons,
there are no rock stations at distances <50 km from the epicentres of the 2012
events, which were generated by thrust-fault ruptures.
As an example for possible PSHA applications, illustrated here are the compari-
sons of Fig.4.8, derived from D4-94, and Faccioli etal. (2015). The graph on the
left of this figure shows that the regional single-station sigma on rock is nearly
coincident with that of the Rodriguez-Marek etal. (2013) constant model, derived
from datasets of different regions. In the graph on the right, one can note that the
mean total single-station sigma (ss) on rock is significantly smaller than the sigma
of the regional GMPE used to compute the residuals (denoted as ITA13in Fig.4.8).
The regional ss and ss shown in Fig.4.7, with their estimated variability, were
derived from the records of 21 accelerometer stations on category A sites, identified
by solid orange circles in Fig.4.7, lying within 120km of the 2012 epicentral area
and with at least 5 records. Since for all of these sites the category A attribution does
not rely on in-situ velocity measurements, it is indispensable that the associated
uncertainty (i.e., the sigma of the sigma) be taken into account in PSHA studies.
The sigmas in Fig. 4.8 quantify the ground motion prediction uncertainty on
exposed bedrock in an area where actual bedrock lies under a sedimentary cover
that is from 100m to some thousands of metres deep and, hence, hazard assessment
at surface sites requires a second step of site response analysis. The variability
ranges shown in Fig.4.7 reflect the predominance of the source-to-site paths dic-
tated by the 2012 events,6 but they sample the contribution of the sources that con-
trol the hazard in the Emilia sector of the Po Plain. Hence, they are believed to be
5
Because only two borehole stations on rock are present in the Euroseistest array.
6
Updated evaluations, benefitting from recent (20132015) recordings of N Apennines earth-
quakes with ML 4 (posterior to the SIGMA analyses) have substantially lessened the single-path
dependence of earlier data and exhibit substantial stability in the S2S and ss,s values shown
herein (presentation by Lanzano etal. at SIGMA final Symposium, November 2015).
Fig. 4.7 DEM of Po Plain portion with accelerometer stations on soft soil and ground type
A.Depth contours are for the base of early Pliocene (M-P1) from Bigi etal. (1992). Rupture area
projections of Reggio 1996 and two Emilia 2012 main shocks are portrayed as red rectangles (from
Faccioli etal. 2015)
indicative of the extent at which uncertainty in strong ground motion prediction can
be reduced by good regional data (Fig.4.8).
A different facet of the ground-motion variability on rock sites is illustrated by
the behaviour of the regional site terms, shown in Fig.4.9; while their mean value
is noticeably close to zero, as it should be, there is an overall tendency for the factors
of the sites to the north (MLC, MTRZ) to exhibit predominantly positive site terms,
and the opposite for sites to the south (ZCCA, BSZ) lying in the Apennines. This
reflects the predominant influence of the 2012 events (but see footnote 6), which
generated stronger shaking to the north of the sources, probably as a result of
stronger propagation in the Po Plain sedimentary basin, see Fig.4.7. The main indi-
cation here is the level of the site-to-site variability, S2S, somewhat on the high side.
4.4.1 Sigma Truncation
Based on the knowledge that the observation residuals tend to be log-normally dis-
tributed, the error () affecting the median logarithmic prediction of a ground motion
parameter obtained through a GMPE is assumed to be a normally-distributed ran-
dom variable with standard deviation logY. The error term is integrated out,
together with the magnitude and the distance terms, in applying the total probability
theorem by which the hazard calculations are carried out, see Fig. 2.1c). The
Ground type A sites r<120km (nrec>5) Ground type A sites (r=120km, nrec>5)
0.5
mean +\-1 std log10 ITA13
median 0.5
mean
0.4 Rodriguez-Marek2013 (constant model) upper\lower
0.4
0.3
mean,u,l
0.3
ss
ss
0.2
0.2
0.1 0.1
0 0
0 1 2 3 4 0 1 2 3 4
T (s) T (s)
Fig. 4.8(left) Mean 1 std. dev. range, in decimal log scale, of regional single-station (event cor-
rected) sigma (ss) for rock sites, based on the 120km Po Plain (see Fig.4.4) dataset restricted to
sites with a minimum of 5 records, compared with Rodriguez-Marek etal. (2013) constant model
(thick dotted line). (right) Total single-site sigma range, labelled as ssmean u,l, for the same rock site
dataset, compared with the ergodic standard deviation of the ITA13 GMPE (heavy line with sym-
bols), the attenuation model developed in D2-72 (with variability analysed in D2-133) and used for
all the residual computations
Fig. 4.9 Site terms S2S r = 120 km (nrec>5)

for the 21 stations on rock 1
with at least 5 records and ZCCA
RNC
lying within 120km BSZ
distance from the MTRZ
epicentral area of the 2012 0.5 MLC
TGG
Emilia, N Italy, earthquake mean +/-1st.dev.
sequence, with mean 1

S2S
band (shaded). The 0

location of most of these
stations is shown in
Fig.4.7
-0.5
-1
0 1 2 3 4
T (s)
question has often been posed whether a truncation (sigma truncation, generally
two sided) should be introduced in integrating over the -distribution in the hazard
integral. If a sigma truncation is introduced, one must also specify a truncation level
in units of sigma, e. g. a 3 truncation.7 In the design of nuclear installations, the
preference seems to be for no Sigma Truncation; according to the US NRC 1.208
(2007) Care should be taken in choosing a value (of the number of standard devia-
tions defining the truncation level) large enough such that natural aleatory variabil-
7
Note that, in the US NRC RG 1.208, the number of standard deviations chosen for the sigma
truncation level is denoted as epsilon.
ity in ground motions is adequately addressed. A study conducted by EPRI and the
U.S.Department of Energy (DOE) () found no technical basis for truncating the
ground motion distribution at a specified number of standard deviations () below
that implied by the strength of the geologic materials. Considering the shear
strength of intact rock materials, it seems unlikely that a limit to ground motions
that can be transmitted to the ground surface during an earthquake may result from
this strength limit.
In case truncation is opted for, there seems to be an understanding that it should
not be at less than 3 level, as in the case of PSHA for the South-East quarter of
France (see D4-170). Considering that logY values for response spectrum values are
typically between 0.3 and 0.4 for most current GMPEs, the +3 truncation level
corresponds to spectral ordinates increased or decreased from 6.0 to 7.5 times w.r.t.
the median. The argument should also be considered that there may not be enough
available observations residuals beyond the 3 level to validate the normal distribu-
tion assumption.
4.5 Approaches fortheVertical Ground Motion Component
Presumably because of the limited role of the vertical seismic action in conventional
earthquake resistant design, only a few of the recent GMPEs include independently
determined coefficients also for the vertical ground motion parameters (e.g. Cauzzi
and Faccioli 2008, Bindi et al. 2011). These indicate that the scaling of vertical
motion measures with respect to the main predictor parameters (magnitude and dis-
tance) is similar to that observed for horizontal motions, and that the standard devia-
tion of the predictions is also similar.
Among recently published equations that include only coefficients for the pre-
diction of horizontal spectral ordinates are those developed in SIGMA for France
(D2-131) and for Northern Italy (D2-53 and D2-133), as well as those derived from
the RESORCE database. In such cases, to estimate the vertical response spectrum
one can either use a simplified envelope for the vertical/horizontal (V/H) response
spectral ratio, or rely on independent GMPEs for the V/H ratio. An example of the
first approach is found in Eurocode 8 (Part 1), in which a value is recommended for
the V/H peak acceleration ratio (depending on magnitude) and different spectral
shape parameters are assumed for the vertical spectrum with respect to the horizon-
tal one.
The simplified envelopes rest on the empirical observation that at very short
periods the vertical and horizontal spectra have comparable amplitudes, while at
intermediate and long periods the level of the vertical spectrum tends to be a nearly
constant fraction of the horizontal one.8 In the intermediate period range a linearly
8
In this respect, it should be mentioned that the often used 2/3 factor for the V/H ratio may under-
estimate the vertical ground motion, according to recent measurements and evaluations (see e.g.
Edwards etal. 2011; Poggi etal. 2012; Nagashima etal. 2014).
4.6Logic Tree Implications 79
decreasing transition can be assumed. Cauzzi and Faccioli (2008) proposed for rock
sites an example of simplified envelope of this kind for the ratio in question, based
on independent prediction equations for the vertical and the horizontal spectrum. In
the simplified envelope approach the standard deviation of the vertical spectrum
estimation can be assumed to be the same as for the horizontal spectrum because the
epistemic uncertainty will already be captured to a large degree in the horizontal
motion.
In the more sophisticated approach relying on the use of independent GMPEs for
the V/H ratio, such as proposed by Bommer et al. (2011) and Glerce and
Abrahamson (2011), the epistemic uncertainty in the prediction of the V/H ratios
should also be addressed (e. g. considering more than one model for each GMPE
branch in the LT) since the total standard deviation of the log(V/H) prediction is
non-negligible (about 0.2). Nevertheless, the interface needs to consider the poten-
tial effect of double counting uncertainties and, thus, the standard deviation should
be partitioned transparently.
4.6 Logic Tree Implications
In addition to the choice of GMPEs, logic tree options of specific relevance for rock
motion characterization concern mainly the treatment of uncertainties in:
1. the implementation of the single-station sigma approach in two-step (hybrid)
hazard assessment, in which the first step defines a bedrock spectrum and the
second one deals with seismic site response analysis; and
2. the adjustments that may be needed among rock sites of different classes.
Concerning item 1, epistemic uncertainty should be considered for both the
median single-site sigma model and for the associated standard deviation.
Uncertainty is controlled by the within-event (ss), and the between-event () stan-
dard deviations of Table4.3. As already pointed out, ss displays limited variability
in different crustal regions (typically from about 0.20 to 0.25 in log10 scale, see
Fig.4.8), while for the between-event standard deviation components of regional
GMPEs could be used, such as given in D2-131 for France and in D2-133 for
Northern Italy. The treatment of uncertainties via logic tree in the single-station
sigma approach is discussed in detail in Rodriguez-Marek etal. (2014) and Faccioli
etal. (2015).
For item 2, the existing approaches lead to quite different results, and it seems
advisable to consider both the data-based adjustment of Sect. 4.3.2 (applicable to
ASCRs only and for frequencies 1215 Hz) and the simulation-based VS30- cor-
rection of Sect. 4.3.1. In applying the former, the similarity between the site-specific
VS profiles for the study and those of KiK-net should be considered for guidance
(see e. g. Fig.4.6), with their variability. The problem with the VS30- host-to-target
correction, as currently applied, derives in essence from the lack of supporting data:
the smooth velocity profiles associated to sites qualitatively described as generic,
hard, or very hard appear to lead to downplaying the amplification response of

rock with respect to hard rock, and to overemphasizing the influence of the -scaling,
for reasons that are still in part unclear.
4.7 Lessons Learned fromtheSIGMA Project
Salient issues in the characterization of ground motion on rock have been singled
out and discussed in this section. The following points are a summary of the most
important among them:
Estimation of ground motion at rock sites through empirical models is affected
by higher uncertainty than at soil sites, because accelerometer stations on rock
with a measured velocity profile are few, and erroneous site class attribution from
geological map inspection or field surveys is not infrequent.
In low or moderate seismicity regions, like much of France, the absence of
recorded data has led to developing rock motion estimation models from numeri-
cal simulation results, using point or finite-fault sources; such models require the
specification of source (e. g. stress drop) and attenuation (e. g. quality factor),
parameters typically affected by large epistemic uncertainty and which are
regionally dependent.
GMPEs adopted for hazard assessment on rock must be compatible with the
seismo-tectonic setting, i.e. they should, as a minimum, respect distinctions
between ASCRs with seismic activity present in the uppermost 2030 km, sub-
duction zones, and SCRs with a thick crust (up to 5060 km) and seismic activity
throughout.
The magnitude scale employed in the attenuation models should be consistent
with that used to derive earthquake activity rates in the seismic source model; if
smaller magnitude (e.g. <~4.5) earthquakes are used, for which Mw is not avail-
able, and magnitude conversion relationships are introduced, the variability asso-
ciated with these relationships should be propagated into the sigma value of the
GMPE for which the adjustment is made (for a discussion on the influence of
magnitude conversions in PSHA, see Chap. 3 of D4-94). It was shown in SIGMA
that GMPEs with lower inter-event variability can be obtained by avoiding mag-
nitude conversions (e.g. of ML into Mw) in the reference dataset.
Adjustments of GM estimations among different classes of rock sites, e.g. among
sites of EC 8 class A and hard rock sites (with VS30 > 1500 m.s1) are still a par-
tially unresolved issue, and two different approaches leading to notably different
results have been presented. Current VS30 corrections seem to overemphasize
the influence of and lead to hard rock spectra with high frequency peaks that
are not supported by the observational evidence of the KiK-net records on hard
rock (appropriately corrected for DH effects), at least up to about 15Hz.
The single-site sigma, or SSS, approach was extensively exploited in SIGMA,
and crucially important lessons were learned from its application, especially in
References 81
the Po Plain region where site-specific data are abundant. In this approach, the
influence of the site-to-site variability (S2S term) that affects the median ground
motion estimation is as important as that of the single-site sigma (ss,s). In par-
ticular, the site term can strongly vary from site to site even in geologically
homogeneous zones and it can lead to local highs/lows in the predicted spectra.
The application of SSS in zones without data should be supported by adequate
2D or 3D physically-based simulations, to constrain the relevant parameters up
to frequencies of a few Hz.
References
Akkar S,Sandkkaya MA, Bommer J (2014a) Empirical ground-motion models for point-and
extended-source crustal earthquake scenarios in Europe and the Middle East, Bull Earthq Eng,
12(1), 359387
Akkar S,Sandkkaya MA, Senyurt M, Azari Sisi A,Ay B, Traversa P,Douglas J, Cotton F, Luzi
L,Hernandez B, Godey S (2014b) Reference database for seismic ground-motion in Europe
(RESORCE), Bull Earthq Eng, 12(1), 311339.
Al Atik L, Abrahamson N, Bommer J (2010) The variability of ground-motion prediction models
and its components. Seismol Res Lett 81:659801
Al Atik L, Kottke A, Abrahamson N, Hollenback J(2014) Kappa () scaling of ground-motion
prediction equations using an inverse random vibration theory approach. Bull Seismol Soc Am
104(1):336346
Ameri G (2015) Progress Report Preliminary Hazard Input Document (HID) for the SIGMA
PSHA for the Frances southeastern , Doc. GTR/ARE/0915-1364, 29 September 2015,
GEOTER, Clapier, France
Anderson J, Hough S (1984) A model for the shape of the Fourier amplitude spectrum of accelera-
tion at high frequencies. Bull Seismol Soc Am 74:19691993
Assatourians K, Atkinson G (2010) Verification of engineering seismology toolbox processed
accelerograms: 2005 Riviere du Loup, Quebec earthquake. Available at www.seismotoolbox.ca
Atkinson G, Boore D (2006) Earthquake ground-motion prediction equations for eastern North
America. Bull Seismol Soc Am 96(6):21812205
Atkinson G, Boore D (2011) Modifications to existing ground-motion prediction equations in light
of new data. Bull Seism Soc Am 101(3):11211135
Bigi G, Bonardi G, Catalano R, Cosentino D, Lentini F, Parotto M, Sartori R, Scandone P, Turco E
(eds) (1992) Structural model of Italy and gravity map 1:500,000. CNR Progetto Finalizzato
Geodinamica. Sottoprogetto Modello Strutturale tridimensionale. Quaderni della Ricerca
Scientifica 114, 3. Firenze, Italy
Bindi D, Pacor F, Luzi L, Puglia R, Massa M, Ameri G, Paolucci R (2011) Ground motion predic-
tion equations derived from the Italian strong motion database. Bull Earthq Eng
9(6):18991920
Bindi D, Massa M, Luzi L, Ameri G, Pacor F, Puglia R, Augliera P (2014) Pan-European ground-
motion prediction equations for the average horizontal component of PGA, PGV, and
5%-damped PSA at spectral periods up to 3.0 s using the RESORCE dataset. Bull Earthq Eng
12(1):391430
Biro Y, Renault P (2012) Importance and impact of host-to-target conversions for ground motion
prediction equations in PSHA.In: Proceedings of 15 world conference on earthquake engineer-
ing, Lisboa, Portugal
Bommer JJ, Douglas J, Scherbaum F, Cotton F, Bungum H, Fh D (2010) On the selection of

ground-motion prediction equations for seismic hazard analysis. Seismol Res Lett
81(5):783793
Bommer JJ, Akkar S, Kale (2011) A model for vertical-to-horizontal response spectral ratios for
Europe and the Middle East. Bull Seismol Soc Am 101(4):17831806
Boore D (2003) Simulation of ground motion using the stochastic method. Pure App Geophys
160:635675
Boore D, Joyner W (1997) Site amplifications for generic rock sites. Bull Seismol Soc Am
87(2):327341
Boore DM, Stewart JP, Seyhan E, Atkinson GM (2014) NGA-West2 equations for predicting PGA,
PGV, and 5% damped PSA for shallow crustal earthquakes. Earthq Spectra 30(3):10571085
Cadet H, Bard P-Y, Duval AM, Bertrand E (2012a) Site effect assessment using KiK-net data: Part
2: Site amplification prediction equation based on f0 and VSZ. Bull Earthq Eng 10(2):451489
Cadet H, Bard P-Y, Rodriguez-Marek A (2012b) Site effect assessment using KiK-net data: Part 1.
A simple correction procedure for surface/downhole spectral ratios. Bull Earthq Eng
10(2):421448
Campbell K (2003) Prediction of strong ground motion using the hybrid empirical method and its
use in the development of ground-motion (attenuation) relations in Eastern North America.
Bull Seismol Soc Am 93(3):10121033
Cauzzi C, Faccioli E (2008) Broadband (0.05 to 20 s) prediction of displacement response spectra
based on worldwide digital records (2008). JSeismol 12(4):453475
Cauzzi C, Faccioli E, Vanini M, Bianchini A (2015) Updated predictive equations for broadband
(0.0110 s) horizontal response spectra and peak ground motions, based on a global dataset of
digital acceleration records. Bull Earthq Eng 13(6):15871612
Chen L, Faccioli E (2013) Single-station standard deviation analysis of 20102012 strong-741
motion data from the Canterbury region, New Zealand. Bull Earthq Eng 11:16171632
Cotton F, Scherbaum F, Bommer JJ, Bungum H (2006) Criteria for selecting and adjusting ground-
motion models for specific target applications: applications to Central Europe and rock sites.
JSeismol 10(2):137156
Delavaud E, Cotton F, Akkar S, Scherbaum F etal (2012) Toward a ground-motion logic tree for
probabilistic seismic hazard assessment in Europe. JSeismol 16(3):451473
Drouet S, Cotton F (2015) Regional stochastic GMPEs in low-seismicity areas: scaling and alea-
tory variability analysisapplication to the French Alps. Bull Seism Soc Am
105(4):18831902
Edwards B, Poggi V, Fh D (2011) A predictive equation for the vertical-to-horizontal ratio of
ground motion at rock sites based on shear-wave velocity profiles from Japan and Switzerland.
Bull Seism Soc Am 101(6):29983019
EPRI (Electric Power Research Institute) (1994) The earthquakes of stable continental regions
volume 5: seismicity database program and maps. Report EPRI TR-102261-V5, Palo Alto
Faccioli E (1992) Selected aspects of the characterization of seismic site effects, including some
recent European contributions. In: Proceedings of international symposium on The effects of
surface geology on seismic motion (ESG 1992), vol I, 2527 March 1992, pp6596, Odawara,
Japan
Glerce Z, Abrahamson NA (2011) Site-specific spectra for vertical ground motion. Earthq Spectra
27(4):10231047
Ktenidou O, Glis C, Bonilla L-F (2013) A study on the variability of kappa () in a borehole:
implications of the computation process. Bull Seismol Soc Am 103(2A):10481068
Nagashima F, Matsushima S, Kawase H, Sanchez-Sesma FJ, Hayakawa T, Satoh T, Oshima M
(2014) Application of horizontal-to-vertical spectral ratios of earthquake ground motions to
identify subsurface structures at and around the K-NET site in Tohoku, Japan. Bull Seism Soc
Am 104(5):22882302
References 83
Poggi V, Edwards B, Fh D (2012) Characterizing the vertical-to-horizontal ratio of ground motion

at soft-sediment sites. Bull Seism Soc Am 102(6):27412756
Rodriguez-Marek A, Montalva GA, Cotton F, Bonilla F (2011) Analysis of single-station standard
deviation using the KiK-net data. Bull Seismol Soc Am 101(3):12421258
Rodriguez-Marek A, Cotton F, Abrahamson N, Akkar S, Al Atik L, Edwards B, Montalva GA,
Dawood H (2013) A model for single-station standard deviation using data from various tec-
tonic regions. Bull Seismol Soc Am 103(6):31493163
Rodriguez-Marek A, Rathje E, Bommer J, Scherbaum F, Stafford PJ (2014) Application of single-
station sigma and site-response characterization in a probabilistic seismic-hazard analysis for a
new nuclear site. Bull Seismol Soc Am 104(4):16011619
SHARE (2013) Seismic Hazard Harmonization in Europe. http://share-eu.org
US Nuclear Regulatory Commission (2007) RG. 1.208: A performance-based approach to define
the site-specific earthquake ground motion, Washington, DC
Van Houtte C, Drouet S, Cotton F (2011) Analysis of the origins of (kappa) to compute hard rock
to rock adjustment factors for GMPEs. Bull Seismol Soc Am 101(6):29262941
Zhao J, Zhang J, Asano A, Ohno Y, Oouchi T, Takahashi T, Ogawa H, Irikura K, Thio HK,
Somerville PG, Fukushima Y (2006) Attenuation relations of strong ground motion in Japan
using site classification based on predominant period. Bull Seism Soc Am 96(3):898913
Chapter 5
Site Response Characterization
It is well recognized that the seismic response of a site is strongly dependent on the
local geological and geotechnical features of the ground profile. Several approaches
are available to include site response effects in the hazard assessment. They are
detailed in the following paragraphs but all of them require more or less in-depth
knowledge of the geotechnical characteristics. Such knowledge can only be acquired
through site investigations. Therefore, a considerable amount of efforts has been
devoted in SIGMA to investigating the reliability of several site investigation tech-
niques, whose level of complexity depends on the choice of the site effect evaluation
method; characterization is also mandatory to get the minimum information to
choose the method itself.
5.1 Soil Characterization
In seismic site response analysis, a key role is played by the shear-wave velocity
model of the site since shear-wave propagation controls the ground-motion amplifi-
cation. Several building codes, Eurocode 8 among them, require the definition of the
VS30, defined as the average velocity in the topmost 30 m, for the definition of soil
classes. The use of GMPEs also often requires the VS30 parameter, and numerical
methods rely on 1D, 2D or 3D spatial distributions of the shear-wave velocity.
Another parameter that proves to be useful in the determination of the site amplifi-
cation is the natural frequency of the soil deposit, f0. Such a parameter can be deter-
mined from the H/V ratio. The shear-wave velocity profile can be retrieved either
with invasive tests, such as crosshole (CH), down-hole (DH) or Suspension Logging
tests (PSSL), or non-invasive methods, such as spectral analysis of surface waves
(MASW).

86 5 Site Response Characterization
5.1.1 Determination oftheProfile Natural Frequency f0
Regardless of the method used for the evaluation of the site effect, measurement of
the profiles natural frequency is highly recommended. It can easily be obtained
with Ambient Vibration H/V Measurements (AVM). This method consists in mea-
suring the ambient noise in continuous mode with velocity meters (not accelerom-
eters) and then computing the ratio between the horizontal and vertical Fourier
amplitude spectra (Nakamura 1989). Guidelines were produced by the SESAME
research programme (SESAME 2004) to implement this technique, which is now
reliable and robust. H/V measurements can provide the fundamental frequency of
the studied site (but not the associated response amplitude) but can also be used to
assess the depth to bedrock and its possible lateral variation when the technique is
implemented along profiles. However, in this case, care should be taken when inter-
preting along the edge of basins, where the bedrock is significantly sloping, because
1-D geometry is assumed in the interpretation of measurements.
5.1.2 D
etermination oftheShear-Wave Velocity Profile
andSite Class
Invasive methods are considered more reliable than non-invasive ones because they
are based on the interpretation of local measurements of shear-wave travel times
and provide good resolution. However, these methods require drilling of at least one
borehole, making them quite expensive. Non-invasive techniques provide cost effi-
cient alternatives. In the last decades the methods based on the analysis of surface
wave propagation are getting more and more recognition (Foti etal. 2014). These
methods can be implemented with a low budget without impacting the site. However,
they need processing and inversion of the experimental data, which should be car-
ried out carefully. The surface-wave inversion is indeed non-linear, ill-posed and it
is affected by solution non-uniqueness. This leads sometimes to strongly erroneous
results causing a general lack of confidence in non-invasive methods in the earth-
quake engineering community.
In this framework, the project InterPACIFIC (Intercomparison of methods for
site parameter and velocity profile characterization), part of the SIGMA project,
compares the main techniques for surface wave methods (intra-methods compari-
son) and provides, as well, a comparison between non-invasive techniques and inva-
sive ones (inter-methods comparison) in order to evaluate the reliability of the
results obtained with different techniques. In the InterPACIFIC project, three sites
were chosen in order to evaluate the performance of both invasive and non-invasive
techniques in three different subsoil conditions: soft soil, stiff soil and rock. At all
sites, at least two boreholes were available to perform the in-hole measurements.
Both active and passive surface wave data were collected, all of them located in the
vicinity of the boreholes for a better comparison between the results from invasive
5.1Soil Characterization 87
a Rayleigh all modes -GRE b Rayleigh all modes -GRE

500
450
400 103
Phase Velocity [m/s]
Phase Velocity [m/s]

350
300
250
200
0 10 20 30 40 50 60 70 80 90 102
100 101 102
Frequency [Hz] Frequency [Hz]
Fig. 5.1 Comparison of the dispersion curves for the Grenoble site: (a) linear scale (b) log scale
and non-invasive methods. Ten different teams of engineers, geologists and seis-
mologists were invited to take part in the project in order to perform a blind test: the
same experimental non-invasive datasets and very little information about the sites
were provided to all teams and then the results were compared. As far as the inva-
sive methods are concerned, different techniques were used by different companies
in order to assess the repeatability of this kind of measurements.
The main conclusions drawn from this benchmark can be stated as follows:
The results show that, as far as the surface wave methods are concerned, the
determination of the dispersion curve is much less critical than the inversion
process. The dispersion curves provided by the participants were in very good
agreement with each other (Fig.5.1 presents one example for the Grenoble site
in France). Nevertheless, the VS profiles obtained by the inversion show a quite
high variability and some features are not uniquely identified or not identified at
all, like for example a low velocity layer at one site. When the velocity profiles
are considered within realistic depth ranges (i.e. those consistent with the maxi-
mum wavelength available with the used acquisition geometry), the results are
more satisfactory than initially expected (Fig.5.2). The standard deviation of VS
values at a given depth is still, by and large, higher for non-invasive techniques
(coefficients of variation (cov) = 0.10.15) than for invasive ones (cov < 0.1) as
shown in Fig.5.3 for the three tested sites (Garofalo etal. 2016). It is important
to note that, since it was a blind test, no a-priori information was provided to the
teams.
Within the project, the same in-hole measurements were tested by different com-
panies in an effort to assess the repeatability of such methods. The results show a
surprising dispersion (even if the dispersion on velocity profiles appears signifi-
Fig. 5.2 Left: VS profiles for Grenoble site; right: zoom of top 100m
Fig. 5.3 Comparison of invasive and non-invasive VS COV values as a function of depth at
Mirandola (MIR; a, b), Grenoble (GRE; c) and Cadarache (CAD; d)
Fig. 5.4 InterPACIFIC subproject: Comparison among the VS profiles obtained with invasive
methods (in green) and non-invasive methods, related to the analysis of active and passive seismic
data (in red) and only passive seismic data (in blue). Sites: Mirandola (MIR, left), Grenoble (GRE,
middle) and Cadarache (CAD, right)
cantly lower than for non-invasive methods, see Fig.5.4 where one green curve is
obviously an outlier and was not considered in the analysis).
Nevertheless, it is interesting to note that VS30, from which the site class can be
assessed, measured with invasive and non-invasive techniques compare favour-
ably to each other as shown in Fig.5.5, where results of the InterPACIFIC bench-
mark are compared to other studies. Moreover, the standard deviation in VS30 is
comparable between both methods (and even lower for non-invasive methods at
one site).
The PSSL method produced robust results (within the errors bars of cross-hole
and down-hole measurements). PSSL is widely used in USA and Japan, but not
yet in France and Italy. It allows performing in-situ measurements within a single
hole to rather large depths (several hundreds of meters) where cross-hole and
down-hole measurements are no longer reliable.
As a conclusion of this benchmark, for a complete characterization of a site it is
recommended to use both invasive and non-invasive methods, as they are comple-
mentary. Furthermore, it is mandatory to supplement the geophysical surveys with
geotechnical boreholes for soil classification (grain size distribution, Atterberg lim-
its, moisture content and so forth).
Fig. 5.5 Relation between VS30 values estimated with invasive and non-invasive methods
Non-invasive methods have a low vertical resolution. For example, they were not
able to identify with a sufficient vertical resolution some feature like the low v elocity
layers, found at 17m and 25 m, at the Grenoble test site. Nevertheless, for average
parameters (like VS30) or even for 1D transfer function estimations based on mea-
sured velocity profiles, they provide robust and reliable results. In addition, one or
several profiles (if 2D or 3D model are needed) based on non-invasive methods can
be implemented for a better evaluation of the spatial variability. Non-invasive meth-
ods do not have real depth limitations if the chosen arrays are large enough. Hence,
they can be a useful complement for sites where the bedrock is too deep to be
reached with invasive measurements at reasonable costs. Guidelines have been
established (deliverable D3-134) to increase the reliability and minimize the risk of
errors or misinterpretation in performing site investigations with non-invasive tech-
niques. These guidelines cover the data acquisition and processing, the parameter-
ization of the velocity profile and the inversion process.
It is recommended that at least one cross-hole (with three aligned boreholes for
more reliable results) be performed down to 30 or 50m. For NPPs a cross-hole test
is traditionally extended down to approximately 100m below the reactor building.
In addition, one of the boreholes can be extended to larger depth (if possible down
to bedrock) and be used for PSSL measurements to measure the bedrock velocity,
which is an important parameter for site response analysis. If this velocity is not
measurable beneath the site, additional measurements should be conducted where
the bedrock is outcropping, taking into account in the interpretation that the bedrock
may be weathered near the surface and that the velocity may increase with depth.
For a less complete site characterization, in relation with the method retained for
the evaluation of the site amplification, non-invasive techniques can be preferred,
for instance to define the site class or VS30.
5.1.3 Seismic Instrumentation
Seismic instrumentation is understood as installation of instruments that allow

recording, on site or in the vicinity, ground motions induced by real earthquakes.
Based on these freefield records, empirical approaches can be implemented that
could allow to:
Measure the site to reference transfer function or other quantification of this
quantity (e.g. coda and duration);
Consider the use of methods like empirical Greens functions;
Evaluate the high frequency attenuation of seismic signals (characterized by the
kappa parameter, see Chap. 4); and
Understand the potential for seismic activity on nearby faults.
While these approaches are readily applicable in areas of high seismicity (e.g.
Japan and Greece), where the data are both numerous and correspond to earth-
quakes of moderate to high magnitude, they are more difficult to implement in areas
of low seismicity, as in mainland France. Indeed, there are few records in these areas
and they correspond to distant earthquakes (medium to high magnitude earthquakes
far away from the site), or regional and local earthquakes, but with low to moderate
magnitudes.
Through continuous measurements made with both velocity meters and acceler-
ometers at several French sites it has been demonstrated that the number of useable
events (events with a good signal-to-noise ratio, S/N) obtained during a rather short
observation period is high in comparison with what was expected before installing
the instruments. At one French site (with moderate seismicity in a rather low noise
context) very robust amplification functions were determined with more than one
hundred usable events in the intermediate frequency range (Fig.5.6). On this site, it
was also possible to evaluate the 0 parameter of Eq. (4.1), which is more difficult to
assess because it needs a high-frequency analysis, where the S/N is usually lower.
Even if instrumentation can provide important information about site response,
it must be said that it cannot alone solve the whole site effect issue. Indeed, the
events that can be expected within a reasonable time interval have usually quite
large epicentral distances, which imply that the incidence angle of the wave field is
almost vertical. It is, however, worth mentioning that vertical wave incidence is
commonly assumed in site response calculations. Furthermore, the recordable
events produce (very) weak motions and, hence, they cannot address the non-
linearity issues. Nevertheless, instrumental data appear essential to producing
empirical amplification functions that are useful when validating numerical simula-
tions for a given range of incidence angles, source sizes and motion amplitudes.
Once the simulations are validated with empirical measurements, other scenarios
involving extended sources, non-linearity etc. can be explored.
Recommendations have been formulated in Deliverable D3-152 to collect valu-
able information from the seismic instrumentation; this involves use of velocity
Fig. 5.6 Evaluation of signal to noise ratio for 101 events recorded with velocity meters during
231 days (South-East of France); each subplot presents the analysis at a given frequency
meters, proper choice of the reference (rock) stations locations and the need for
continuous recording as opposed to triggered records.
5.1.4 Characterization ofNonlinear Soil Properties
With the presently available technology, nonlinear soil properties can only be
obtained from laboratory tests carried out under well-defined and constrained envi-
ronmental conditions. Needless to say, reliable results can only be obtained if undis-
turbed samples are used. Retrieving undisturbed samples from the ground in
cohesionless soils is a huge challenge that will not be addressed here.
Usually, engineers consider that the nonlinear behaviour is uniquely determined
from knowledge of the variation of the secant shear modulus G and equivalent
damping ratio with shear strain amplitude: the so-called G/Gmax = f() and = g()
curves. However, it must be realized that real soil behaviour involves a coupling
between shear strain and volumetric strain and that, even under pure 1-D analysis
(vertical propagation of shear waves in a horizontally-layered profile), settlements
in dry soils or pore pressure build-up in saturated sands may take place. Therefore,
5.2Hazard Assessment at theGround Surface 93
a complete description of the soil constitutive model requires data not only on the
shear behaviour but also on the volumetric behaviour. Only the linear and equivalent
linear models do not require knowledge on the volumetric behaviour because pure
shear (vertical propagation of shear waves in a horizontally-layered profile) induces
only shear strains.
Even if one focuses only on the shear behaviour, several pitfalls exist that should
be properly handled:
A common mistake in the characterization of the shear behaviour is to measure
the G/Gmax curves in the lab under a given confining pressure and to consider that
the same curve applies at any depth in the soil profile provided the material does
not change. It is well known, however, that not only Gmax but also the shape of the
curve depends on the confining pressure (Ishihara 1996). To overcome such a
difficulty, and to keep the number of tests to a reasonable number, the correct
representation is to normalize not only the modulus but also the strain
G/Gmax = f(/r), where r is a reference shear strain (Hardin and Drenvich 1972).
A difficulty for a complete definition of the shear behaviour, faced during the
Prenolin benchmark (Sect. 5.3.3), is the extrapolation of the G/Gmax curve back
to small strains; typically, the laboratory equipment that was used to measure the
soil properties was a cyclic triaxial apparatus, which led to inaccurate results for
strains smaller than about 104. How to reconcile the shear modulus at = 104
to the elastic modulus, Gmax, calculated from in situ geophysical measurements,
was a matter of debate and no unique, satisfactory solution was found. This issue
is important because it governs the shape of the stress-strain curve in the inter-
mediate strain range, which, in turn, affects the site response. Alternative choices
for the extrapolation contribute to an increase of the epistemic uncertainty. This
situation can obviously be improved by combining cyclic triaxial tests with reso-
nant column tests, which are able to handle smaller strains.
5.2 Hazard Assessment at theGround Surface
Seismic hazard at the ground surface may cover different aspects: vibratory ground
motion, surface faulting, induced hazards like slope instability, liquefaction and so
forth. The SIGMA project only addressed the vibratory ground motion aspect and,
therefore, the other topics are not covered in this document.
The most straightforward approach to define the hazard at the ground surface
consists of making use, in the PSHA, of generic GMPEs that, through a given proxy,
take into account the site characteristics. However, this approach will not give full
credit to the peculiar characteristics of the site. One alternative is to start from the
rock hazard and to define site amplification functions (SAF), typically described as
a function of frequency, which modify the rock hazard spectrum. These site ampli-
fication functions are defined by the response spectral ratio of the response at the
site divided by the corresponding response at the ideal outcropping bedrock
GENERIC SITE-SPECIFIC
Surface spectra
Surface
PSHA Amplification Factors, AF,
horizon
Site property model Site Response

GMPEs Vs,G/Gmax(),D() Input motions
Analyses
Disaggregation
Baserock Response
Spectra
Baserock
PSHA
horizon
Sigma model
Target GMPEs
(fss, )
Target Host
Vs, K0+ Vs-K Adjustment Factors Vs, K0+
Host GMPEs
Fig. 5.7 Schematic representation of generic and site-specific hazard calculations for a considered
site (Modified from Rodriguez-Marek etal. 2014)
Table 5.1 Classes of approaches to account for site effect in PSHA

Hybrid probabilistic/deterministic Fully probabilistic
Generic site Site-specific (HyS) Generic site Site-specific (FpS)
(HyG) (FpG)
PSHA at PSHA at rock + SAF PSHA based PSHA at rock + convolution
rock + SAF based on site-specific on site- with SAF conditioned to rock
based on seismic response analyses or specific ground motion.
norms measurements GMPEs Typically based on site response
(single-station analyses
sigma)
Modified from Faccioli etal. (2015)
(reference station). These SAFs can be either determined experimentally, from

observations, or numerically from site response analyses. While in the numerical
calculations the availability of a reference station is not generally of concern, since
response at an ideal rock site can generally be determined (Sect. 6.7), experimental
evaluations of SAFs are in most cases limited by the lack of suitable reference sta-
tions in the vicinity of the site.
The two approaches in the ground motion characterization for a site-specific and
generic PSHA are illustrated in Fig.5.7. Note that this figure only considers SAFs
evaluated from numerical calculations.
Following Cramer (2003), Bazzurro and Cornell (2004b) and Perez etal. (2009),
the classes of approaches to account for seismic site response within a PSHA can be
broadly classified as summarized in Table5.1.
Hybrid approaches are typically based on the results of a PSHA at a rock site,
where site response effects are superimposed by multiplying the UHS at rock by a
suitable SAF.The latter may be defined either by the spectral amplification factors
for generic sites introduced typically by local norms or guidelines (approach HyG),
or by a site-specific SAF, calculated in most cases by considering the mean amplifi-
cation function from 1D linear-equivalent seismic wave propagation analyses for
the specific soil profile (approach HyS). In such analyses, time-history calculations
are typically carried out by considering a suite of real accelerograms, satisfying the
response spectrum compatibility with the target PSHA spectrum on rock. While
HyG is the approach implicitly outlined by seismic norms, approach HyS is fre-
quently used for site-specific seismic hazard analyses of important facilities so that
it may be considered as the reference approach. Although sound, and easy to under-
stand, from an engineering point of view, a limitation of the hybrid approach is that
it may provide estimates of the exceedance rates at the site that are not consistent
with the corresponding ones on rock, as noted by Bazzurro and Cornell (2004a, b).
The above mentioned limitations can be overcome by following fully probabilis-
tic approaches, which may be broadly subdivided in terms of their range of applica-
tion, either for a generic site (FpG) or for a specific site (FpS). The FpG approach is
based on the standard application of PSHA, where the site response is summarized
within a period-dependent site correction factor to modify the expression of the
considered GMPE. Such correction factors are provided by practically all recent
GMPEs (www.gmpe.org.uk), either in terms of broad soil categories or in terms of
soil classes related to seismic norms, or of other related engineering parameters/
proxies such as VS30. The drawback of such an approach is that it may not provide
reliable results when dealing with site-specific response evaluations. In the latter
case, a site-specific GMPE could be used (Ordaz etal. 1994), if a sufficient number
of strong-motion records are available at the site for a reliable GMPE to be con-
structed, but this is seldom the case. Finally, an FpS approach may be followed,
such as described in Sect. 2.5.3.1 and in particular proposed by Bazzurro and
Cornell (2004a, b), involving the calibration of conditional SAFs, i.e., of the site-
specific ground motion amplification values at a specific vibration period, condi-
tioned to the exceedance of a given level of ground motion on rock.
Although this fully probabilistic site-specific approach allows a formally correct
incorporation of seismic site response into the PSHA, it suffers from several limita-
tions that were addressed in Deliverable D3-54, namely:
the probability distribution of the conditioned amplification function is based on
1D numerical simulations of vertically propagating plane waves at a nonlinear
soil site with uncertain properties: this assumption is expected to deeply affect
not only the median amplification function, but also its standard deviation;
observed amplification at the site may also be affected by source-to-site azimuth
and directivity, especially in near-source regions (e.g., because of different angles
of incidence of waves, or because of larger/smaller onset of surface waves
depending on the relative position of the source with respect to the basin). This
is neglected by 1D approaches; and
no validation is available against a sufficiently wide set of strong-motion records,

especially to quantify the values.
The Bazzurro and Cornell (2004a, b) approach has been followed in recent proj-
ects like PEGASOS and PEGASOS Refinement Project (Renault 2009). However,
it has not been implemented within the framework of the SIGMA project and, con-
sequently, it will not be presented in detail herein. The interested reader may refer
to the original work of Bazzurro and Cornell (2004a, b) or to Deliverable D3-96in
which it is briefly outlined. Within the framework of SIGMA the three other
approaches, HyG, HyS and FpG, have been tested by the French and Italian teams.
The Italian team applied the method to the Casaglia site in the Po Plain (Deliverables
D3-96 and D4-94, a further application to the Mirandola site in Po Plain is illus-
trated in Faccioli etal. 2015) and the French team applied it to two sites: Grenoble
(France) and EuroseisTest (Greece) (Deliverable D4-153).
5.2.1 D
irect Evaluation fromGround Motion Prediction
Equations (FpG)
Instead of computing the rock hazard and then transferring it to the ground surface
with approaches that will be detailed in the following paragraphs, relevant GMPEs
can be used in the PSHA to directly compute the hazard at the ground surface. This
approach assumes that: (a) the soil conditions at the site resemble those at the sta-
tions in the database considered for the development of the GMPEs used for the
hazard estimation and (b) the site response is assumed to be correctly captured by
the site model included in the adopted GMPEs. To be valid, the GMPE must be
representative of the site conditions, i.e. include a proxy that is deemed to represent
the ground conditions; some GMPEs may also include a nonlinear term for high-
amplitude motions. The most commonly used proxies are the VS30 and the site fun-
damental frequency f0 or the site class (Eurocode or NEHRP classification);
examples of relevant GMPEs are given in Deliverable D3-152. This approach is
crude because each site has its own peculiarities like, for instance, interbedded lay-
ers with a high stiffness contrast, marked subsurface topography and so forth and
use of a single proxy cannot account for all the site-specific features. Nevertheless,
it can be argued that empirical GMPEs are established from large databases, which
also certainly contain peculiar sites. The main advantage of the approach is that a
full probabilistic analysis is possible with propagation of all (certainly overesti-
mated) uncertainties. Note that even in this simplified approach, a good character-
ization of the soil profile is needed (VS30, soil class, f0). As mentioned previously,
this approach has been implemented for the Grenoble, Po Plain, and EuroseisTest
sites. Results will be compared to the other approaches in Sect. 5.3.2.
An alternative site-specific approach in the same broad category (FpG) consists
in applying data-based corrections to the GMPEs median values and in replacing
their standard deviations. This approach can be implemented when a sufficient
number of records is available at the target site. It is also desirable that the recorded
earthquakes span a sufficient range of magnitudes, distances and azimuth angles.
Following this approach the GMPEs median predictions are modified through site-
specific correction factors (S2S) and the GMPEs sigma is replaced with the single-
station sigma (ss,s), as explained in Sect. 4.3. The S2S factor can, in first
approximation, be considered as an intrinsic characteristic of the site and is used to
modify the predictions by a GMPE, in a very simple way. More precisely, the site
correction term modifies the GMPE median prediction ( GMPE (T)) as follows:
corrected ( T ) = GMPE ( T ) .10

S 2 S( T )
(5.1)

This approach has been implemented by the Italian team for the Casaglia (CAS)
and Novellara (NVL) deep soil sites in the Po Plain. A generic, regional GMPE has
been established (in Deliverable D2-72) to a large extent from records on sites in the
Po plain with similar characteristics to the Casaglia and Novellara sites; the median
predictions from this GMPE were then modified with the S2S term established
from the acceleration records obtained at both sites. Likewise, the original standard
deviation of the GMPE has been replaced by the single-station sigma derived from
the same records. Results for the Casaglia site are compared to the standard approach
(direct use of a GMPE for the relevant soil class) in Fig.5.8.
The figure below has been established for just a single branch of the PSHA logic
tree (one source model, one GMPE) and compares the standard approach (in blue)
and the site-specific approach with the above correction (in green). The other curves
in Fig.5.8 correspond to different approaches that are presented in the following
paragraphs, whereas the implications of the present approach on the terms are
discussed in Sect. 5.4. As seen from the figure, the CAS example strongly indicates
that the soil surface spectra can decrease significantly when using results from local
site analyses (single-station coefficients) instead of the site coefficient of the
GMPEs. This behaviour is however site-specific and does not occur for instance at
NVL.
5.2.2 Generic Site Specific Approaches (HyG)
This approach assumes that the hazard calculated with GMPEs, either at rock or at
the surface, is a first-order model for the target site response but that some site-
specificity must be introduced in order to provide a better description of the site
response. A correction factor can be developed and applied, as a post-processing, to
the computed hazard spectrum.
Fig. 5.8 Casaglia (Po plain) site. Site specific median ground surface spectrum for 2475-years
return period
5.2.2.1 Correction toRock Spectrum
The most commonly accepted meaning for this approach consists in starting from
the UHS at the bedrock and applying a site amplification factor (SAF), based on
some site characteristics, to compute the surface spectrum. SAFs can be given by
norms, based on the site class (e.g. Eurocode or NEHRP), or can be measured
experimentally on site (ideally in a vertical array) if the latter is located in a reason-
ably active area.
As neither Italy nor France have a sufficient amount of data from a single site, the
approach could not be fully tested. However, taking advantage of the Japanese KiK-
net data (http://www.kyoshin.bosai.go.jp/), Paolucci etal. (Deliverable D3-96) cal-
culated the SAFs from stations exhibiting a VS soil profile similar to those observed
in the Po Plain, i.e., deep soil sites with VS30 values in the range 200400 m.s1.
From their study at 21 stations and considering only shallow events (depth < 15 km)
with PGA > 10 cm.s2 they reach the following conclusions:
observed variability of SAFs at KiK-net deep soil sites is generally limited in
spite of the wide range of magnitude and distances encompassed by records
(Fig.5.9);
Wave velocity (m/s)

0 500 1000 1500 2000 2500 3000
0
S-wave
P-wave
40
80
Depth (m)
120
160
200
Fig. 5.9 Unconditioned SAFs (left side) and Vs profiles (right side) for one deep soil site station
(NIGH11) in the Kik-net. The Mw-Repi distribution for the considered records is also shown
conditioned SAFs (i.e. spectral amplifications as a function of the spectral

pseudo-velocity on rock, PSVR) do not show a significant, if any, dependence on
the intensity of motion at bedrock, suggesting, for the stations considered, no
clear evidence of nonlinear site response, in spite of the relatively soft soil condi-
tions (Fig.5.10). This is at variance with results from Rgnier etal. (2013) who
found significant evidence for the shift of the predominant peak frequency with
increasing amplitude of motion at the borehole station. Reasons for such discrep-
ancy are not yet understood;
log10 computed from the conditioned SAFs ranges typically between 0.04 and
0.08; furthermore, no clear evidence is found of a dependence of log10 with
period.
A similar finding of a limited aleatory variability of SAFs from low intensity
records at deep soil sites has been confirmed by the surface-to-borehole spectral
ratios observed from a selected set of records at Casaglia. The interesting remark on
these records is that the observed SAF remains nearly unchanged, irrespective of
earthquake magnitude and distance, both for local weak events and for distant stron-
ger earthquakes. However, in none of the selected cases, was the ground-motion
amplitude high enough to predict any significant nonlinear effect.
T=0.1 s T=0.2 s T=0.3 s

101 101 101
PSA surf/PSA depth
PSA surf/PSA depth
PSA surf/PSA depth

100 100 100
10-1 10-1 10-1

10-2 10-1 100 101 102 10-2 10-1 100 101 102 10-2 10-1 100 101 102
PSVR[cm/s] PSVR[cm/s] PSVR[cm/s]
T=0.5 s T=0.75 s T=1 s

101 101 101
PSA surf/PSA depth
PSA surf/PSA depth
PSA surf/PSA depth

100 100 100
10-1 10-1 10-1

10-2 10-1 100 101 102 10-2 10-1 100 101 102 10-2 10-1 100 101 102
T=1.5 s T=2 s T=3 s

101 101 101
PSA surf/PSA depth
PSA surf/PSA depth
PSA surf/PSA depth

100 100 100
10-1 10-1 10-1

10-2 10-1 100 101 102 10-2 10-1 100 101 102 10-2 10-1 100 101 102
Fig. 5.10 Conditioned SAFs for different vibration periods for station NIGH11, with data grouped
by Magnitude. Blue dots: M < 4, green: 4 < M < 5; magenta: 5 < M < 6; red: M > 6
5.2.2.2 Correction toGround Surface Spectrum
The approach suggested here consists of calculating first the ground surface response
spectrum with the relevant GMPEs (Sect. 5.2.1), and then of applying a site correc-
tion factor on the rationale that the GMPEs do not reflect all site peculiarities, for
instance a significant 2D sub-topography (basin effect). The correction factor can be
either derived from statistical relationships (Site Amplification Prediction Equation,
SAPE) or calculated with reference to a 1D analysis.
SAPE was introduced by Cadet etal. (2012a, b), to define empirical prediction
of site amplification as a function of a few parameters (VSZ, with z equal to 5, 10, 20
and 30m and f0) derived from Japanese strong-motion data (KiK-net). The amplifi-
cation factor was estimated from the ratios between the surface and down-hole hori-
zontal response spectra, corrected for the varying depths and impedance of the
down-hole locations. The amplification factors were then correlated with site
parameters. The results showed that the best performance in predicting site amplifi-
cations was obtained by the coupled parameters VS30 f0, while the best single
parameter proved to be f0. The hazard spectrum should then be multiplied by the
amplification function. However, care must be taken in the combination of hazard
spectrum and amplification function in order not to double count the site effect in
5.3Completely Site Specific Approaches (HyS) 101
both GMPEs and SAPE.For example, a SAPE based on f0 can be applied to a haz-
ard spectrum calculated for the selected VS30 value in the adopted GMPEs, assuming
that VS30 and f0 are poorly correlated and thus account for different origins of site
amplification. On the contrary if a SAPE based on both VS30 and f0 is used, then the
hazard spectrum should be calculated using a reference rock VS30 (e.g., VS30 = 800m.
s1) in the GMPEs.
The advantage of this approach is to allow considering the amplification at the
fundamental frequency of the site, which is usually neglected in the generic approach
because only few GMPEs adopt the fundamental frequency as a site-effect proxy.
The main limitation in the application of SAPE is related to the database, consisting
exclusively of Japanese data, which may emphasize systematic differences in the
shallow site amplification in Japan with respect to other regions. Due to this limita-
tion, use of SAPE is not recommended.
To account for sites prone to 2D or 3D amplification (basin effect, above surface
topography) much larger than implied in the generic GMPEs, the notion of aggrava-
tion factor (AG) was introduced by ChavezGarcia and Faccioli (2000) to reflect the
contribution in site effect due to a complex local geometry with respect to 1D geom-
etry. It is defined as the ratio between 2D/3D and 1D calculated response spectra.
AGs may be obtained from 2D/3D calculations with different soil constitutive mod-
els (linear, equivalent linear, fully nonlinear) and using accelerograms consistent
with the hazard level. Ratios between 2D/3D and 1D response spectra are then
computed and the final aggravation factor (and associated standard deviation) is
deduced. The AG is then applied to a reference hazard spectrum in which the 1D
response of the soil is assumed to be accounted for. AGs are based on ratios of com-
putations: as a first approximation, it is assumed that a change in the model descrip-
tion (e.g. a velocity change) will affect more or less similarly both 1D and 2D/3D
computations. The overall results are then less sensitive to changes (inaccuracies) in
the model description. AGs have also been proposed, based on an extensive number
of calculations and simplified typologies, by the NERA project (http://www.nera-
eu.org).
An example of aggravation factor calculated for the EuroseisTest site is depicted
in Fig.5.11 (Deliverable D3-152).
5.3 Completely Site Specific Approaches (HyS)
The site-specific approaches are based on a detailed consideration of the local site
response and relative uncertainties. The site amplification is estimated using more
or less sophisticated numerical or experimental (Sect. 5.2.2.1 for one example)
methods depending on the characteristics of the site. In order to solve this problem
several steps need to be considered (see Fig.5.7):
definition of an input motion at the base of the soil profile. This may require the
calculation of the rock hazard for large values of VS30 (hard-rock conditions) that
Fig. 5.11EuroseisTest: Left Calculated amplification factor; Right suggested aggravation

factors
may be outside the domain of validity of the adopted GMPEs (i.e., VS30 larger
than 12001500 m.s1). In this case the GMPEs need to be adjusted for such
hard-rock conditions in order to correctly represent the input motion for the site
response analyses (see Sect. 4.3).
choice of acceleration time histories when numerical simulations are foreseen.
Selection of relevant time histories is an important step, especially for nonlinear
analyses and represents an important task by itself; this aspect is covered in Sect.
6.7. Within the framework of SIGMA several different strategies have been used
to choose the accelerograms based on the UHS: scaled natural records over the
whole frequency range or over two different frequency ranges (2 sets of accelero-
grams) and spectrally-matched accelerograms.
geometric definition of the soil profile: 1D (i.e., horizontally layered strata), 2D
(i.e., alluvium valley or topographic ridge), or fully 3D.
rheological characterization of the soil layers: linear viscoelastic, viscoelastic
linear equivalent or fully non-linear behaviour.
definition of the incident wave field: plane wave with vertical or oblique inci-
dence, or surface wave (typically corresponding to a remote source); for a poten-
tial source close to the site, it is preferable to include the source in the
computational model; this can be either a point source or an extended source, the
latter one being more realistic for a source-site distance smaller or comparable to
the fault size, and being preferable for earthquakes with magnitude larger than 6.
choice of a (or several) software package to make the calculations; the most com-
monly used codes are listed in Sect. 6.3.2.
In SIGMA, almost all options were tested but most of the calculations were
restricted to vertical incident plane waves.
Fig. 5.12 Comparison between different attenuation models for 1D 2D 3D linear computa-
tions for the Grenoble test-site. Solid line: results with model 1 taking into account standard
scaling to define Q factors (e.g. QS = VS/10). Dashed lines: results with model 2, using the damp-
ing factor computed with 1D non-linear computations
5.3.1 Linear Numerical Analyses
From a rheological point of view, the linear viscoelastic assumption is the simplest
constitutive model that can be used: it simply requires the definition of the dilata-
tional VP and shear VS wave velocities and of attenuation (quality factor) for each
wave, QP and QS.1 While appropriate techniques exist for measuring the wave veloc-
ities (see Sect. 5.1), there are presently no established ones for measuring the atten-
uation; a typical, very crude rule of thumb is to use Q = VS/10 (m.s1). Furthermore,
no distinction is made between QP and QS. Sensitivity analyses carried out during
the project have shown that the results (ground surface motions) are very sensitive
to the choice of Q. Consequently, this parameter significantly increases the epis-
temic uncertainty of the analyses. To illustrate this point, Fig.5.12 presents for the
Grenoble site the surface response spectra calculated with two assumptions for
attenuation.
The main advantage of linear constitutive models is their simplicity that theoreti-
cally allows considering 2D and, possibly 3D, geometries. However, in practice, 3D
calculations are not feasible for frequencies higher than about 4Hz because of the
1
In geotechnical earthquake engineering the equivalent damping ratio, , is more commonly used;
it is related to the attenuation by 2 = Q1
restrictions posed on the element mesh size (h /10, with the wavelength) and
above all because it is impossible to characterize the soil medium at such a small
scale over large areas. Except for low amplitude rock motions, the linear viscoelas-
tic analyses are not recommended as it is recognized that soils may exhibit nonlin-
ear behaviour from small strains, although some counterexamples exist [see for
instance Fig.5.10 which documents linear behaviour up to PSV = 20 cm.s1, or the
Mirandola site studied in Faccioli etal. (2015)]. Linear calculations may, however,
be of value for calibrating the velocity profiles with the results of small recorded
earthquakes or validating the computational codes in the linear range.
5.3.2 Equivalent Linear Numerical Analyses
Given the complexity of the nonlinear behaviour of soils, equivalent linear models
represent a good compromise between engineering practice and scientific knowl-
edge; they presently constitute the state of practice in site response analyses. Many
models and codes have been developed and are currently used for such simulations
(see Sect. 6.3.2). However, these models have some limitations: it is generally
accepted that they are valid for shear strains smaller than 0.10.3%. Since the upper
limit for the shear strain depends on the soil plasticity index and confining pressure
(depth), it is better to relate it, rather than to an absolute value, to a reference shear
strain r defined as max = r. Gmax; a value r = 2 is suggested. Despite their relative
simplicity and the limited number of parameters required (wave-velocity profiles,
variation of the properties with shear strain), the equivalent linear models are also
subject to large uncertainties. As noted in Sect. 5.1.4, the variations of properties
can only be measured in the laboratory on undisturbed samples, and are not always
obtained under the relevant stresses, especially for large depths. In view of these
difficulties, often the curves G/Gmax = f() and = g() are chosen from published
results in the literature. This uncertainty in the definition of the nonlinear properties
is strongly reflected in uncertainties in the calculations. One such example is pre-
sented in Figs.5.13 and 5.14, obtained for the Casaglia site (Deliverable D3-96)
with different sets of published curves.
Equivalent linear analyses have been implemented both by the Italian (Casaglia
site) and the French teams (Grenoble and EuroseisTest) using 1D calculations.
Results from the Casaglia site are presented in Fig.5.8 introduced previously. The
two red solid curves correspond to two different strategies for the selection of the
input motions: either natural, moderately scaled, or spectrally-matched records. The
correction of the input signals for spectral matching does not affect values beyond
about 0.2 s, where a close agreement with the NTC2008 norms (Italian code spec-
trum for Class C soil) is attained. For T < 0.2 s, on the other hand, the input correc-
tion slightly affects the results; in particular, the PGA value of the spectrum. More
significant, however, is that the use of both SAFs leads to decreasing the values of
the site specific spectra below the bedrock values at periods 0 < T < 0.2 s, mostly
Fig. 5.13 Modulus reduction and damping curves for clay soils data at 9.6m depth (from D3-96)
due to nonlinear effects. Also worth noting is that the non-ergodic PSHA spectrum
from the GMPE (green curve) is not far from the rock SAF spectra. On the
whole, the Casaglia example strongly indicates that the soil surface spectra may
decrease significantly when using results from local site analyses (such as SAFs or
single site coefficients and sigmas) instead of the site coefficient of the GMPEs.
Fig. 5.14 Five percent damped response spectra obtained by 1D propagation analyses performed
using the soil degradation curves summarized in Fig.5.13 (From D3-96)
This is caused by the notable deamplification indicated by negative S2S values and
does not necessarily hold true for all sites.
Results obtained by the French team for the Grenoble site shown in Fig.5.15 are
consistent with those from Casaglia. The SAF applied to the rock spectrum are
derived from 1D linear, equivalent linear analyses after the bedrock motion has been
corrected for hard rock conditions (VS = 3500 m.s1). The surface spectra do exhibit
the same marked reduction from the spectrum calculated with the generic GMPE
(black solid line) when equivalent linear models are used and predict a much larger
amplification when using a linear model. Again, the set of time histories selected by
the Italian team (IT) or by the French team (FR) does not appear to have a strong
impact on the results.
Additional calculations were carried out for 2D and 3D geometries and linear
soil behaviour. The results, not presented herein (see Deliverable D3-152), show
that the aggravation factor, which intends to reflect the impact of the sub-surface
topography by correcting the surface spectrum leads to a surface spectrum that is
much higher (except at high frequencies >20 Hz) than the spectrum directly calcu-
lated from 3D, and even 2D, linear analyses; use of aggravation factors therefore
appears to be very conservative. Note, however, that due to inherent limitations of
3D calculations (see Sect. 5.3.1) they could not have been carried out above 4Hz
and the 3D results at larger frequencies have been extrapolated based on the SAFs
Fig. 5.15 Response spectra computed for the Grenoble test-site with different approaches (all
target levels are for a return period of 10,000 years). Comparison between generic mean UHS
(black) and site-specific mean GMRS obtained using linear (blue, red), equivalent-linear (magenta)
and nonlinear (light blue, green) site amplification factors and Vs-kappa adjustment. The mean
UHS for bedrock conditions is also shown by the dashed black curve
of the 2D calculations; this may explain the somewhat inconsistent results, which
are nevertheless also predicted by 2D linear calculations.
5.3.3 Nonlinear Numerical Analyses
When the shear strain exceeds the amplitudes previously indicated (i.e., above 0.3
0.5%, or better > 2r) or, in other words, for very soft soils and/or very severe
loading, a complete nonlinear modelling would theoretically be required, with an
appropriate constitutive relationship and its associated soil parameters. These mod-
els oscillate between two poles: relatively simple constitutive models with few
parameters, which can hardly reproduce all possible loading/unloading paths, and
more complex models with many parameters (sometimes exceeding ten), which
may succeed in describing all possible paths, but whose determination remain
largely beyond experimental capabilities. The validation of nonlinear approaches is
a major issue, which has not, up to now, received satisfactory scientific answers; the
Prenolin international benchmark was launched to address these issues. Twenty
three world wide teams participated in the benchmark (D3-114 and D3-149) with
the major objective of assessing the actual epistemic uncertainties associated with
nonlinear calculations, emphasizing those associated with the software (constitutive
100 A-0 G-0 L-1 Q-0 Y-0

(kPa)
-100
100 B-0 H-0 M-0 R-0 Z-1
(kPa)
-100
100 C-0 J-0 M-1 S-0 F-0
(kPa)
-100
100 D-0 J-1 M-2 T-0 -5 0 5
(%)
(kPa)
-100
100 E-0 K-0 N-0 U-0
(kPa)
-100
-5 0 5 -5 0 5 -5 0 5 -5 0 5
(%) (%) (%) (%)
Fig. 5.16 Cyclic stress-strain loops for a soil element with shear strength 65 kPa subjected to a
sinusoidal input seismic motion of 10 s. Predictions produced by the different constitutive
models
models and numerical scheme), those associated with the translation of raw field
and laboratory data into design values for the nonlinear model parameters, and the
actual deviation of numerical predictions from observations (bias and scatter). Two
sites have been chosen in Japan and extensive 1D nonlinear calculations have been
performed. One of the main lessons from this project is the importance of the choice
and calibration of the constitutive model; this implies numerical testing of the model
over the whole strain range and comparison with laboratory tests. Figure5.16 pres-
ents results of the numerical tests of the constitutive models used in Prenolin deemed
to represent the cyclic behaviour of the same soil element for which the same
mechanical characteristics were provided to all participants.
Many results have been produced by this benchmark; their post-processing is
still ongoing and cannot be summarized here. An illustration of comparisons
between computed nonlinear site response and that observed is depicted in Fig.5.17
for four input motions; the weakest input motion is #9 and the two strongest ones
are #1 and #2.
Preliminary conclusions from the benchmark indicate that:
a very careful characterization of the soil profile is needed to achieve adequate
modelling with a combination of both laboratory and site measurements imple-
mented with different techniques (invasive vs. non-invasive);
use of several non-linear codes operated by different teams is desirable to assess
the epistemic uncertainty associated with the constitutive models; and
Fig. 5.17 Comparison between the predicted transfer function at KiK-net site KSRH10 (1684
percentiles, colour shaded areas) and the observations (solid lines with corresponding colours),
for input motions 9, 5, 2 and 1
despite the large differences in the soil constitutive models and numerical
schemes, the epistemic uncertainty of the site-specific nonlinear calculations is
smaller than the value of the within-event variability of GMPEs for the site.
Nonlinear numerical analyses were also tested for the three sites: Casaglia,
Grenoble and EuroseisTest. Only one nonlinear constitutive model was retained for
each case and the results were compared to those of linear and equivalent linear
calculations. Results are presented only for one site (Grenoble) in Fig.5.18; how-
ever, the same conclusions were reached for the two other sites (see Fig. 5.8 for
Casaglia):
Linear calculations produce the highest amplification; as discussed previously,
this is due in a large part to the inaccurate definition of damping but, neverthe-
less, some sites may exhibit a more linear response than others;
Linear equivalent and nonlinear models tend to predict SAFs of the same order
of magnitude except for large amplitude motions (top diagram in Fig.5.18 cor-
responding to a large return period); and
SAFs are not very sensitive to the set of accelerograms used for the
calculations:
Fig. 5.18 Spectral amplification factor computed in 1D analyses for the Grenoble test-site, with
accelerometers from set 1 (top), set 2IT (middle) and set 2FR (bottom)
Set 2FR: accelerograms representative of a smaller motion corresponding to

a return period of 10,000 years and a PGA of 0.15 g, proposed by the French
team; and
Set 2IT: spectrally matched accelerograms proposed by the Italian team for
the same target as above.
In terms of ground response surface spectra Fig.5.15 shows that, as noted previ-
ously for the Casaglia site, the estimated soil surface spectra decrease significantly
5.4Treatment ofUncertainties 111
when using results from local site analyses (such as SAFs) instead of the site coef-
ficient of the GMPEs. Furthermore, except for very large input motions, complex
nonlinear analyses are not really warranted and equivalent linear analyses appear to
be accurate enough for engineering purposes.
5.4 Treatment ofUncertainties
The general framework for the treatment of uncertainties has been presented in Sect.
4.4 for the rock hazard. This paragraph explains how the uncertainties for site haz-
ard were handled in the SIGMA project.
5.4.1 Fully Probabilistic Generic Site Approach (FpG)
For site hazard, the same scheme applies when the FpG approach is implemented
with the site term S2S and its site-to-site variability S2S related to the soil site. The
total single-station standard deviation at an individual site is then given by the last
line of Table 4.3. As already cited in Sect. 5.2.1 the methodology has been applied
to 12 sites in the Po Plain by Faccioli etal. (2015) to calculate the standard deviation
of S2S and ss,s, the event corrected single-station standard deviation at individual
sites. The epistemic uncertainty on ss,s, was estimated from its standard deviation
across many stations (in this case, the Po Plain 12-station set), thereby implicitly
assuming ergodicity in the variance (not in the mean). Such standard deviation (std
(ss,s)) was found equal to 0.08, and used to define upper and lower bound estimates
of ss,s according to:
( )
std (ss ,s ) + 2
2
ss ,s = ss ,s (5.2)

is the inter-event variability component of the considered GMPE (ITA13, see
deliverable D2-72), and its uncertainty was neglected.
The foregoing variability estimates are illustrated in Fig.5.19, which also shows
that the ITA13 GMPE sigma represents a kind of average ss,s of the three sites.
Introducing the variability bounds through ss,s is intended to accommodate, at least
in part, the uncertainty caused by multipathing, which is only partially accounted
for in the available data because of the predominance of the 2012 Emilia sequence
records both at the study sites and in the ITA13 GMPE (but see, on this aspect,
footnote 6in Chap. 4).
The previous median (Eq. 5.1) and sigma (Eq.5.2) formulations have been com-
bined in a simple logic tree for seismic hazard estimation. Comparison of predic-
tions with data from actual records (Reggio Emilia 1992 and Emilia 2012
earthquakes) shows that spectra are mostly within the spread spanned by the logic-
a 1 b 0.8
MRN MRN
NVL NVL
S2S +/-S2S,epistemic
T0821 T0821
0.5 0.6
log10 ITA13
mean,u,l
0 0.4
ss,s
-0.5 0.2
-1 0
0 1 2 3 4 0 1 2 3 4
T (s) T (s)
Fig. 5.19(a) Site terms S2S for the three study sites with S2S;epistemic bands (T0821 coincides
with CAS). (b) Total single-site sigma ss;s for the same sites (lines), with variability bands enclosed
within the upper and lower limits estimated from Eq. (5.2); the ITA13 GMPE standard deviation
(log 10 ITA13) is also shown in this plot. The variability estimates are in both cases derived from a
12-site Po Plain data subset
tree branches for a return period of 475 years, and the spectral shapes are reasonably
similar. The code spectra are also consistent with the UHS at 475 years.
5.4.2 Hybrid Site Specific Approach (HyS)
Two main sources of uncertainties are involved in the hybrid-approach calculation

of site-specific hazard, each one containing an aleatory and an epistemic
component:
uncertainties in the rock (or hard-rock) hazard. Most of the ground motion uncer-
tainties are accounted for in the standard deviation of the GMPEs and in host-to-
target adjustment, if applied.
uncertainties in the site amplification function.
Apart from the aleatory uncertainty related to nonlinear soil behaviour, most of
the other sources of uncertainty are already, to a large extent, included in the alea-
tory variability of the rock GMPEs. For this reason, except for cases where the vari-
ability of nonlinear behaviour is significant, aleatory variability in site response
must not be included to avoid double counting of aleatory uncertainties in rock and
site hazard.
For a given target spectrum on rock, the sources of uncertainties in site response
evaluation can be identified as related to: (a) selection of input motions, (b) dynamic
properties of soil profile, (c) selection of the method of analysis for site response
evaluation, and (d) modelling of soil behaviour. To avoid double counting of alea-
tory uncertainties, only epistemic contributions will be referred to in the following,
so that the small scale random variability of soil properties is disregarded. As a mat-
ter of fact, the impact of the small scale random variability in the mechanical
5.4Treatment ofUncertainties 113
p roperties of the propagation path already can be considered as entering into the ss
uncertainty component of the partially non-ergodic analysis, by which the bedrock
UHS were determined.
The impact of the various sources of epistemic uncertainties in site response has
been illustrated in SIGMA with the studies reported in Faccioli etal. (2015) on one
site in the Po plain, i.e. the accelerometer station Mirandola (MRN).
5.4.2.1 Uncertainty Related toSelection ofInput Motions
It is shown that the uncertainty linked to the choice of input motions is considerably
reduced, to become almost negligible in the total uncertainty, when broad-band
spectral matching of the records to the target spectrum is achieved (see Sect. 6.7.1).
5.4.2.2 Uncertainty inVS Profiles
As pointed out in Sect. 5.1, this uncertainty can be reduced by combining experi-
mental measurements, measurements of velocity profiles by different techniques
including invasive and non-invasive techniques. For the Mirandola site studied by
Faccioli etal. (2015), the standard deviation of the site response due to alternative
VS profiles, Vs log10, was found to be of the order of 0.050.08 (Fig.5.20).
5.4.2.3 Uncertainty inSoil Behaviour Modelling
The examples presented either for the Casaglia site (Fig.5.14) and the Grenoble site
(Fig.5.18) have shown that the largest epistemic uncertainty resides in the choice of
the constitutive model (linear, equivalent linear or nonlinear) and of the associated
a 0.4 b 0.5
with 0-1 s scaling
with 0-5 s scaling 0.25
0.3 unscaled
0.2
SA out (g)
log10
0.2 0.15
0.1
0.1
0.05
0 0
0 1 2 3 4 0 1 2 3 4
T (s) T (s)
Fig. 5.20(a) Average acceleration response spectra resulting from different VS profiles and differ-
ent sets of input motions, either unscaled or scaled in the 01 s or 05 s period ranges. (b)
Corresponding sigma in a log 10 scale. RP = 475 years, equivalent-linear analyses
parameters (degradation curves of Fig.5.13). Although the Italian team found that
for the deep soil sites in the Po plain the linear model predicted the closest response
to the recorded motions, this conclusion cannot be generalized to any situation: it
obviously depends on the soil material, the intensity of the rock motion, the reli-
ability of the soil parameters and so forth.
5.4.2.4 Combination ofEpistemic Uncertainties
Faccioli etal. (2015) proposed that the resulting TOT associated with the average
site-specific response spectra for a given return period, including both the epistemic
uncertainties of the site-response analysis and the total (aleatory + epistemic) uncer-
tainties carried by the PSHA at exposed bedrock (PSHA rock) be evaluated as

TOT = max
T
( V2s + soil
2
_ model + PSHA _ rock ; Kik net
2
) (5.3)
where Kiknet is the standard deviation of SAFs of 21 deep soil sites considered from
the KiK-net (Sect. 5.2.1). The latter values of , which are in general related to the
combination of a wider set of seismic site amplification factors, rather than 1D
effects alone, can reasonably be considered as a lower bound for evaluations of site-
specific variability of results.
5.5 Lessons Learned fromtheSIGMA Project
The major lessons in site response analyses learned from the SIGMA project have
been highlighted in the present chapter. The following bullet points are simply a
summary of the most important ones:
Consideration of site response in SHA requires more or less detailed character-
ization of the soils, depending on the level of sophistication of the analyses; this
characterization should be obtained from tests (field and/or laboratory tests) and
can be advantageously complemented with site instrumentation. The minimum
required information should include:
Description of the geometry of softer layers, with respect to underlying bed-
rock, and especially, in cases where topographic effects are important, deter-
mination of the aspect ratio (ratio between height and width of a basin);
position of the considered site with respect to the basin border;
VS and VP profiles; and
shear-wave velocity contrast between basin (or more generally the soft layers)
and bedrock.
Whenever possible the single-station sigma approach should be used with a view
to decreasing the level of uncertainty in the ground surface response.
5.6Additional Topics inGround Surface Hazard Assessment 115
The Hybrid site specific approach (HyS) is the most versatile approach and rep-
resents a good compromise between the simple fully probabilistic generic
approach (FpG), which tends to overestimate the ground surface response, and
the fully probabilistic site-specific approach (FpS), which is very demanding and
does not necessarily address all aspects of site response.
In the HyS approach the main source of epistemic uncertainty resides in the soil
behavior modelling; several models are available from the simple viscoelastic
linear model to the fully nonlinear model. The most appropriate one is still a mat-
ter of debate and comparison with field records gives contradictory results.
Furthermore, characterization of the constitutive model parameters is prone to
large uncertainties. At the present stage, it seems that equivalent linear models
present a sufficient degree of accuracy and recourse to more sophisticated non-
linear models is not warranted, except maybe at very long return periods.
Based on the experience gained during the SIGMA project, guidelines to con-
sider site response in PSHA have been prepared as deliverable D3-152. The main
objective of the guidelines is to propose a gradual approach based on the HpG, FpG
and HyS methods for considering site response; the choice between the three
approaches depends on the level of characterization of the input data and the esti-
mated importance of site response.
5.6 Additional Topics inGround Surface Hazard Assessment
The following topics, although important for seismic hazard assessment, have delib-
erately not been considered in SIGMA.Therefore, they are not discussed in detail
and only general concepts are presented in the following paragraphs.
5.6.1 Vertical Ground Motion
The number of GMPEs for vertical motions is, by far, fewer than those for horizon-
tal motions. Therefore, among the approaches listed in Table 5.1, the most com-
monly implemented are the HyG, HyS and FpS approaches. HyS and FpS are based
on numerical analyses. HyG is usually implemented after having calculated the
horizontal UHS at the ground surface (by any of the four approaches) and applying
V/H ratios, i.e. ratios between the vertical and horizontal response spectra at the
ground surface. Statistical frequency-dependent relationships have been developed
for such ratios. In this approach, attention must be paid to large amplitude motions
that may induce significant nonlinearities in the soil under horizontal loading. Under
vertical loading, mainly caused by the vertical propagation of dilatational waves,
nonlinearities are much more limited; it is, therefore, important that the V/H rela-
tionships reflect this characteristic, otherwise the vertical motion may be
underestimated. Few such relationships exist; one can refer to Glerce and
Abrahamson, (2011).
Another approach for estimating the vertical ground motion may be based on
physics: for shallow profiles, especially those composed of water-saturated materi-
als, the wave lengths are much longer than the profile thickness because the soil is
nearly incompressible; the dilatational wave velocity exceeds 1500 m.s1 (velocity
of sound in water). It can, therefore, be argued that the vertical rock motion is barely
amplified at the ground surface.
Finally, if the ground surface response spectrum is calculated from 1D site
response analyses (HyS or FpS) attention must be paid to the soil characteristics.
Usually the vertical motion is calculated independently of the horizontal one and the
question arises of the proper soil characteristics to enter in the model, at least for
medium-to-large-amplitude motions. The state of practice is to use the properties
(VS) retrieved from the horizontal motion calculation (the strain compatible charac-
teristics in equivalent linear analyses) and to convert them into VP. In saturated soils
P waves travel through the water; the bulk modulus of the soil skeleton may be
slightly affected by the induced shear strain but the overall bulk modulus, which is
the sum of the soil skeleton bulk modulus and of the water bulk modulus, will be
almost unaffected. In dry soils, the propagation of P waves is controlled by the skel-
eton properties:
4
Vp 2 = K + G (5.4)
3
G and K are influenced by the shear strain but the bulk modulus K to a much lesser
extent than the shear modulus G. It seems, therefore, appropriate to calculate VP for
the vertical analyses assuming that K is strain independent, equal to its elastic value,
and that G is equal to the strain compatible value calculated in the horizontal equiva-
lent linear runs. Applying the same reduction to VP as to VS to account for nonlinear
effects is certainly unrealistic; however, any reduction in-between the one proposed
above, with no reduction on K, and the full reduction coming from VS is debatable
and contributes to the epistemic uncertainty. Obviously, if full nonlinear analyses
are carried out for the horizontal motion, the vertical motion shall be input simulta-
neously to the horizontal one and the constitutive model will account for the (time)
variations of both wave velocities.
5.6.2 Maximum Ground Motion: Truncation
Recent studies to assess very long-term seismic hazard in the United States and in
Europe have brought the issue of upper limits on earthquake ground motions into
the arena of problems requiring attention from the engineering seismological com-
munity. Few engineering projects are considered sufficiently critical to warrant the
use of annual frequencies of exceedance so low that ground-motion estimates may
References 117
become unphysical if limiting factors are not considered, but for nuclear waste
repositories, for example, the issue is of great importance. The definition of upper
bounds on earthquake ground motions also presents an exciting challenge for
researchers in the area of seismic hazard assessment (Bommer etal. 2004).
The maximum ground motions that can be experienced at the ground surface are
controlled by three factors: the most intense seismic radiation that can emanate from
the source of the earthquake; the interaction of radiation from different parts of the
source and from different travel paths; and the limits on the strongest motion that
can be transmitted to the surface by shallow geological materials. As seismic waves
propagate to the Earths surface, other factors act to limit the maximum amplitude
of the motion. These factors are associated with the failure of surface materials,
which are usually weaker than the underlying rock, under the loading conditions
generated by the passage of seismic radiation. The principle is similar to that of a
fuse: once failure is reached at a given depth within the soil profile, the high fre-
quency components of the incident motion are filtered and accelerations larger than
those reached at the failure stage cannot be transmitted to the upper strata; however,
this is counterbalanced by larger displacements.
The first and most obvious tool for exploring upper bounds on ground motions is
the ever-increasing databank of strong-motion accelerograms. Another approach is
to define the maximum rock motion, if any, and to calculate, based on the strength
of the overlying soil materials, the maximum motion that can be transmitted either
from simplified analytical models (Pecker 2005) or from nonlinear site response
analyses with increasing amplitudes at the bedrock. The simple procedure of trun-
cating at a specified number of logarithmic standard deviations above the logarith-
mic mean of the GMPE is unlikely to be an adequate solution (Bommer etal. 2004),
see also Sect. 4.4.1.
As a conclusion of this brief discussion, the issue of maximum ground motions
is still an open question requiring further attention.
References
Bazzurro P, Cornell CA (2004a) Ground-motion amplification in nonlinear soil sites with uncer-
tain properties. Bull Seismol Soc Am 94(6):20902109
Bazzurro P, Cornell CA (2004b) Nonlinear soil-site effects in probabilistic seismic-hazard analysis.
Bommer JJ, Abrahamson N, Strasser FO, Pecker A, Bard PY, Bungum H, Cotton F, Fh D, Sabetta
F, Scherbaum F, Studer J(2004) The challenge of defining upper bounds on earthquake ground
motions. Seismol Res Lett 75(1):8295
Cadet H, Bard P-Y, Rodriguez-Marek A (2012a) Site effect assessment using KiK-net data: Part 1.
A simple correction procedure for surface/downhole spectral ratios. Bull Earthq Eng
10(2):421448
Cadet H, Bard P-Y, Duval AM, Bertrand E (2012b) Site effect assessment using KiK-net data: Part
2: Site amplification prediction equation based on f0 and VSZ. Bull Earthq Eng
10(2):451489
Chavez GF, Faccioli E (2000) Complex site effects and building codes: making the leap. JSeismol
4(1):2340
Cramer CH (2003) Site seismic-hazard analysis that is completely probabilistic. Bull Seismol Soc
Am 93(4):18411846
Foti S, Lai CG, Rix GJ, Strobbia C (2014) Surface wave methods for near-surface site character-
ization. CRC Press, Boca Raton, p487, ISBN:9780415678766
Garofalo F, Foti S, Hollender F, Bard PY, Cornou C, Cox BR, Dechamp A, Ohrnberger M, Perron
V, Sicilia D, Teague D, Vergniault C (2016) InterPACIFIC project: comparison of invasive and
non-invasive methods for seismic site characterization, Part II: Inter-comparison between
surface-wave and borehole methods. Soil Dyn Earthq Eng 82:241254
Glerce Z, Abrahamson NA (2011) Site-specific spectra for vertical ground motion. Earthquake
Spectra 27(4):10231047
Hardin BO, Drnevich VP (1972) Shear modulus and damping in soils: design equations and
curves. JSoil Mech Found Div ASCE 98(SM6):667692
Ishihara K (1996) Soil behaviour in earthquake geotechnics, Oxford Engineering Science Series.
Oxford University Press, Oxford, p342
Nakamura Y (1989) A method for dynamic characteristics estimation of subsurface using micro-
tremor on the ground surface, Quarterly Report Railway Technical Research Institute, 301,
2533, Tokyo, Japan
Ordaz M, Singh SK, Arciniega A (1994) Bayesian attenuation regressions: an application to
Mexico City. Geophys JInt 117:335344
Pecker A (2005) Maximum ground surface motions in probabilistic seismic hazard analyses.
JEarthq Eng 9(4):125
Perez A, Jaimes MA, Ordaz M (2009) Spectral attenuation relations at soft sites based on existing
attenuation relations for rock sites. JEarthq Eng 13(2):236251
Rgnier J, Cadet H, Bonilla LF, Bertrand E, Semblat JF (2013) Assessing nonlinear behaviour of
soils in seismic site response: statistical analysis on KiK-net strong-motion data. Bull Seismol
Soc Am 103(3):17501770
Renault P (2009) PEGASOS/PRP Overview. Joint ICTP/IAEA advanced workshop on earthquake
engineering for nuclear facilities. Abdus Salam International Center for Theoretical Physics,
Trieste, Italy
Rodriguez-Marek A, Rathje E, Bommer J, Scherbaum F, Stafford PJ (2014) Application of single-
station sigma and site-response characterization in a probabilistic seismic-hazard analysis for a
new nuclear site. Bull Seism Soc Am 104(4):16011619
SESAME (2004) European research project WP12, Guidelines for the implementation of the H/V
spectral ratio technique on ambient vibrations measurements, processing and interpretation
Deliverable D23.12. European Commission Research General Directorate Project No.
EVG1-CT-2000-00026 SESAME
Chapter 6
Seismic Hazard Computation
This section deals with the seismic hazard computation process. It should be noted
that, in cases where the seismic hazard to be calculated includes site-specific soil
amplifications, this process may include two steps: hazard calculation for reference
rock conditions and subsequent site response analysis to combine the local soil
response with the rock hazard (see Sect. 5.2).
6.1 Basic Requirements
The following basic requirements apply to the seismic hazard computation.

In large projects, the teams in charge of the development of the Ground Motion
(GM) models (SIGMA WP2) and of the Seismic Source (SS) models (SIGMA
WP1 and WP4) are often different from the Hazard Calculation Team (HCT).
For this reason, it is vital for the projects success to establish good communica-
tion and interaction between the HCT and the model developers (Seismic Source
Characterization- SSC, and Ground Motion Characterization- GMC, teams) to
ensure full agreement between the model finally implemented in the PSHA soft-
ware and the models developed by the SSC and GMC teams.
It is desirable, and of great benefit for the project, that the SSC and GMC teams
produce Hazard Input Documents (HIDs) to be provided to the HCT (Ameri
2015). The HID provides a list of input parameters required for the PSHA calcu-
lation within the project. The final HID should provide sufficient information to
the HCT to build the input to the SSC and GMC logic tree in the hazard code for
PSHA calculations, and to avoid any non-unique interpretation. Several interac-
tions between the GMC-SSC teams and the HCT may be required in order to
refine the HID and to verify that consensus is reached on all the necessary input
parameters to the software.

120 6 Seismic Hazard Computation
Software verification and validation (V&V) documentation: in recent years, the

requirements related to software qualification were formally included in interna-
tional guides (e.g. IAEA, SSG-9, USNRC RG 1.208). Thus, it is crucial to
develop verification and validation plans to guarantee the users and end-users
that the calculations performed with the software are reliable. Such Quality
Assurance (QA) documents can also be available from previous studies for par-
ticular software (e.g., Thomas etal. 2010). It is, in any case, good practice to
identify the capabilities of the selected software and the potential limitations in
the application to the current project. Such limitations shall be identified early in
the project to allow adopting another software if necessary and to verify that the
software is flexible enough for integration of new calculation methods or for
reducing the calculation time. If documentation on the software V&V is not
available or judged insufficient, simple tests should be performed in order to
verify the results for typical applications of concern for the project.
6.2 Interfaces andBoundary Conditions
The different steps involved in a PSHA (SSC, GMC and SRC) are often perceived
as separate tasks that are finally combined into a unique logic tree at the stage of the
hazard calculations. However, this perspective should be avoided as a number of
obvious interfaces exist between all these tasks, and because the true inputs from
SSC, GMC and SRC should be correctly reflected in the hazard calculation model.
These interface issues are addressed in Sect. 2.6 and Chap. 7.
6.3 Software Packages
6.3.1 PSHA Software
Several PSHA software packages have been developed in the last decades. Some of
them are licensed products and others are completely open-source. The choice of
the PSHA software is not a trivial issue and it should be oriented on the basis of the
specific needs of the project. Moreover, with such a large number of available soft-
ware, the issues related to the reliability and accuracy of hazard results and to the
Quality Assurance protocol become of great importance (e.g., Bommer etal. 2013).
Another factor which might strongly influence the choice of a PSHA software is the
acceptance by an independent reviewer or a regulatory body. Some of the available
codes have an already accepted nuclear QA programme (e.g. by US NRC), while
some others are mainly used in the academic environment, without appropriate
V&V documentation.
6.3Software Packages 121
FRISK
Features CRISIS EQRM 88M MoCaHAZ MRS NSHMP OHAZ OpenSHA SEISRISK SEISHAZ
Monte Monte
PSHA approach classic classic classic classic classic classic classic classic
Carlo Carlo
Seismic Source Types

Area Yes No Yes No
Fault Line [2D] Yes Yes Yes

Limited Yes No No Yes
Volume [3D] Yes No Yes Yes
No
Point Yes No Yes Yes No Yes Yes
Allow to assign a depth
Yes No Yes
distribution to each source
Allow to assign a style-of
Limited Yes No Yes No Yes
faulting to each source
Rupture length & width
Yes No Yes
Modelling
Magnitude Frequency Distribution [MFD]
Gutenberg-Richter Yes
Gaussian Yes No Yes No No Yes No Yes No Yes

Customize MFD intervals No Yes No No No Yes No Yes Yes No
Ground Motion Prediction Equations [GMPE]

Built-in Yes No Yes
User Defined Yes No Limited Yes No No Limited No Yes No

Allow to assign different
GMPEs per seismic source Yes No Yes No Yes
type
Max Ground
Yes No Yes No
Truncation Motion Value
No. of Sigma Yes
Variability Yes
Customizable units Yes No Yes
LOGIC TREE
Allows to define a logic tree Yes No No Yes No Yes No Yes
OUTPUT
Hazard Curves Yes No Yes
Hazard Maps Yes No Yes

Uniform Hazard Spectra Yes No Yes No Yes No Yes
M-R Yes Yes
Dis- Yes Yes Yes
M-R-epsilon Yes No No
aggregation No
Geographic No Yes No No
Eq. Scenario /Shake Maps Yes Yes No Yes No
Fig. 6.1 Example of skill matrix for a set of software packages (From Danciu etal. 2010)
Recent studies have investigated and compared a number of available software in

terms of their functionalities and features (e.g., Danciu etal. 2010) and in terms of
hazard calculation results (e.g., Thomas etal. 2010; Bommer etal. 2013). A useful
product of the study by Danciu etal. (2010) is what we may call a skill matrix of the
PSHA programs (Fig.6.1). This type of matrix may be explored to assess whether
a particular code has the capabilities to address a specific issue of interest for the
project (e.g., use of leaky or strict source boundaries in which a fault rupture initiat-
ing within a source is allowed to extend outside the source perimeter). In large
projects, such as SIGMA, the PSHA software should be flexible enough to incorpo-
rate the results provided by the model developer on R&D actions initiated by the
project or by the scientific community.
Important results are also provided by studies aiming at comparing the perfor-
mance and accuracy of hazard results calculated using different codes for standard
cases (see, e.g., Thomas etal. 2010). Such tests are extremely useful because they
provide a well-defined benchmark for testing new codes and to evaluate the perfor-
mances of existing codes for different situations. In the study by Thomas et al.
(2010) the hazard curves calculated for a number of test cases using different soft-
ware packages are compared among each other and, where available, with analyti-
cal solutions. In particular, the comparison with analytical solutions is extremely
important, especially at very low probabilities of exceedance where numerical prob-
lems may arise in the codes.
The consistency of hazard results among different programs was only partially
considered in SIGMA by performing a comparative study between Crisis and
OpenQuake (D4-140). This study has the clear merit of pointing out that different
hazard codes may use different modelling assumptions that result in different haz-
ard estimates. It is thus important to verify that such assumptions are consistent with
the PSHA model that is intended to be implemented. For example, the simulation of
virtual ruptures within area source zones is clearly a topic in which different, and
maybe equally reliable, models can be implemented (e.g., elliptical vs. rectangular
ruptures). Furthermore, the study also pointed out the importance of the validation
and verification protocol especially in terms of implementation of GMPEs. The
approach followed in D4-140 is a good example of the strategy that should be fol-
lowed when comparing hazard programs. The first step focuses on a comparison of
the implementation between SIGMA and OpenQuake of the selected GMPEs for
the French area. The second step concerns the comparison of hazard results obtained
with CRISIS and OpenQuake for simple test cases. Finally, the CRISIS and
OpenQuake hazard results are compared for a simplified version of the SIGMA
PSHA logic tree.
6.3.2 Site Response Analysis Codes
Site response analyses are usually carried out under the assumption of one-
dimensional wave propagation. Equivalent linear ground response modelling is by
far the most commonly utilized procedure in practice (Kramer and Paulsen 2004) as
it requires the specification of well-understood and physically meaningful input
parameters (shear-wave velocity, unit weight, modulus reduction, and damping).
Nonlinear ground response analyses provide a more accurate characterization of the
true nonlinear soil behaviour, but they are seldom used in engineering practice
because they require the specification of input parameters that lack an obvious asso-
ciation with fundamental soil properties and because the sensitivity of the site
response results to these parameters is not well understood. The nonlinear codes
differ essentially by the soil constitutive relationships built in the code and to a
lesser extent by the numerical scheme: finite element (FEM), finite difference
(FDM) or spectral method (SP).
The equivalent linear codes are robust and have been used extensively since the
early seventies. Their main limitation lies in their inability to handle large amplitude
motions because the embedded constitutive relationship (the equivalent linear
6.4Sensitivity Analysis 123
model) is valid only for limited shear strains, typically of the order of twice the
reference shear strain (see Sect. 5.1.4). Codes belonging to this category are
SHAKE91 (Idriss and Sun 1992), EERA (Bardet etal. 2000) and CYBERQUAKE
(Modaressi and Foerster 2000).
It is almost impossible and also not meaningful to list all possible nonlinear soft-
ware. Several of them, belonging to all the categories listed above (FEM, FDM and
SP), have been tested in the PRENOLIN benchmark (Deliverable SINAPS@-2015-
V2-A2-T3-1) and in other benchmarks (Stewart etal. 2008). The most commonly
used codes are DEEPSOIL (Hashash etal.), FLAC (ITASCA), SUMDES (Li etal.
1992), DYNAFLOW (Prevost 2010), OPENSEES (http://opensees.berkeley.edu/),
NL-DYAS (Gerolymos and Gazetas 2005) and DMOD 2000 (Matasovi and
Ordez 2007). Only the last two softwares belong to the finite difference category.
Once again, the merit of each code relies essentially on the capability of the soil
constitutive model to reproduce the main features of the soil behaviour under cyclic
loading but also, as evidenced in the PRENOLIN benchmark, on the qualification
and experience of the users. An important lesson learned from SIGMA is the abso-
lute necessity, before launching a large number of nonlinear calculations, of testing
and validating the constitutive model against laboratory tests.
6.4 Sensitivity Analysis
During a project sensitivity analyses are used to test the impact of the different alter-
natives and to identify, for example, which parameters have a strong impact on the
hazard at the site and may deserve further investigation. In practice, sensitivity anal-
ysis is often used during the course of a project to test alternative branches of the
logic tree or to highlight the part of the model to which the SSC or the GMC teams
should dedicate their efforts. Sensitivity analyses can be conducted to evaluate if a
particular hypothesis should be included in the final logic tree or if its impact on the
hazard at the site is negligible. In this latter case, the initial scientific logic tree can
be simplified, thereby reducing the computational task for the project. Sensitivity
analyses shall be done at a sufficiently early stage of the project to fully take advan-
tage of the results. A sensitivity analysis generally aims at the following
objectives:
measuring the sensitivity of the seismic hazard to the different parameters/inputs
(i.e., those that affect most the results); and
evaluating the contribution of the uncertainties in the input parameters to the
total hazard uncertainties.
Sensitivity analyses were performed for the French and Italian areas of interest
in SIGMA (D4-18, D4-41, D4-29 and D4-94). A dedicated study is documented in
D4-138 for the French area.
The uncertainties in the seismic hazard and the sensitivity of the hazard to the
different parameters of the model can be measured using different metrics
(D4-138):
distance between hazard curve percentiles (e.g., 1684%): this can be used, for
instance, to compare the hazard from the whole logic tree among different sites
or for different return periods. This metric can also be used to compare the haz-
ard from a reduced logic tree, for example by comparing the distance between
percentiles provided by the logic tree when using two area source models.
ratio between percentiles (e.g. 16/50; 50/84) to appreciate the shape (symmetry)
of the distribution.
distance between mean hazard curves: this can be used, for instance, to compare
the mean hazard curves obtained for different GMPEs.
The sensitivity analysis should be conducted for several return periods and sev-
eral spectral periods (of interest for the project) in order to investigate the variation
of the hazard distribution as a function of the annual frequency of exceedance at
spectral periods consistent with the natural periods of the structure. This is done
because typically the parameters that contribute to the hazard are not the same for
short and long return or structural periods. Similarly, in low to moderate seismic
regions, some parameters, e.g., the maximum earthquake magnitude, Mmax, affect
the seismic hazard more in the low-frequency range than in the high-frequency
range; use of different spectral periods (e.g., PGA and SA at 0.2 s and 1 s) will serve
that purpose.
In D4-138 the results of the sensitivity analysis are presented as Tornado dia-
grams, already introduced in Sect. 3.6.2. These diagrams constitute a compact
graphical representation to summarize the results for a number of tests and for the
two above mentioned metrics. Tornados are constructed first by fixing the return
period of interest (e.g., 10,000 years) and finding in the mean hazard curves the
acceleration corresponding to that return period, then the hazard variations can be
appreciated in two ways:
in terms of Annual Frequency of Exceedance (AFE) compared to the AFE of the
mean acceleration; and
in terms of spectral acceleration compared to the mean spectral acceleration. In
this case the variability can be quantified as percentage variation around the
mean acceleration value (Fig.6.2).
Important information can also be obtained by examining the contribution of
each source zone or fault included in the SSC logic-tree to the hazard at the site (this
is also sometimes called disaggregation by source, although it is not a hazard disag-
gregation, strictly speaking). An example of hazard curves obtained from each
source zone and their contribution to the total hazard at a site is presented in Fig.6.3.
This analysis allows highlighting which seismic sources carry the largest contribu-
tion to the hazard and thus may help to focus on better characterization of, e.g.,
seismicity of the zone, deformation mechanism and so forth.
6.4Sensitivity Analysis 125
Fig. 6.2 Example of Tornado plot
Fig. 6.3 Example of hazard contribution by source: hazard curves for each area source (left) and
percentage contribution to the total hazard for two return periods of interest (right)
Fig. 6.4 Example of M-R-Epsilon (number of sigmas) hazard disaggregation for spectral accel-
eration at 0.5Hz (left) and 20Hz (right)
6.5 Hazard Disaggregation
The disaggregation of seismic hazard is an effective way to identify combinations

of magnitude, distance, and values of other random variables (e.g., epsilon = the
number of standard deviations of the ground motion distribution) defining seismic
events that contribute to a selected seismic-hazard level (McGuire 1995; Bazzurro
and Cornell 1999). The disaggregation process is carried out by evaluating the fre-
quency of occurrence of seismic events with different combinations of magnitude,
distance, and other random variables that contribute to the selected target seismic-
hazard level (McGuire 1995) or that lead to exceedance of the selected target
seismic-hazard level (Bazzurro and Cornell 1999). A higher frequency of occur-
rence of seismic events defined by a particular combination of values of the random
variables describing possible seismic events is an indication that such identified
events are the most likely to contribute to the selected target seismic-hazard level.
Therefore, the identified events can be used as scenario events for assessing both
new and existing engineered structures and infrastructures. For example, if the tar-
get seismic-hazard level of interest represents a value of the structural design
response spectrum at the fundamental natural vibration period of a designed struc-
ture, the identified scenario events can be used as an aid in selecting ground-motion
records for checking the adequacy of the seismic design (see Sect. 6.7).
Magnitude-distance-epsilon disaggregation is routinely performed in PSHA
studies and many software programs include such functionality. Figure6.4 shows
an example of M-R-epsilon disaggregation. The disaggregation is used to determine
the controlling earthquakes in terms of magnitude, distance and , which is usually
used as guidance to select or develop time histories for engineering purposes and at
appropriate annual frequencies of exceedance. Epsilon disaggregation provides
information on the contributing ground motions at the site in terms of the standard
6.6Additional Engineering Output Parameters 127
deviation of the GMPEs. This information may be useful to evaluate the coherence
and consistency of the earthquake scenarios when deterministic seismic hazard
assessment in conducted to control the PSHA results. It can also be used to depict
the contribution of significant outliers to the total hazard, when the hazard appears
to be controlled by the higher values.
6.6 Additional Engineering Output Parameters
Traditionally, PSHA is performed in terms of PGA and spectral acceleration

(defined using the response of elastic single-degree-of-freedom (SDOF) systems).
Alternative intensity measures (IMs) are proposed in the literature that attempt to be
more closely related to the damage potential of an earthquake.
Some of them are amplitude-based (Peak Ground Velocity), duration-based
[Arias Intensity, Cumulative Absolute Velocity (EPRI 1988), Standardized
Cumulative Velocity (EPRI 1991)] or frequency response-based [Effective Peak
acceleration (ATC 1978), Acceleration Spectral Intensity (Von Thun etal. 1988),
Averaged pseudo-spectral intensity (Bojorquez and Iervolino 2011)].
In WP5, the SIGMA project investigated the respective merits and performance
of existing and new proposed IMs for estimating the damage to structures and to
non-structural components (NCS). A new simple and efficient intensity measure
accounting for non-linear structural behaviour (De Biasio etal. 2014) and for proba-
bilistic assessment of the acceleration demand of non-structural components (De
Biasio etal. 2015) was proposed.
In a first step, a new IM, named Relative Average Spectral Acceleration (ASAR),
is defined as a simple function of the fundamental frequency of the structure, of the
reduced fundamental frequency of the damaged structure and of the spectral pseudo-
acceleration between the two frequencies. Extensive tests, comprehensive compara-
tive studies, statistical analyses based on recorded data from the RESORCE database
(SIGMA WP2) and finite element simulations demonstrate that this new IM, with a
reduction value R of 40% (ASA40), performs well compared to most of existing
IMs and constitutes a promising engineering parameter to conduct performance-
based seismic design. Based on a similar approach, a second parameter, namely
Equipment-Relative Average Spectral Acceleration (E-ASA67) is proposed by De
Biasio etal. (2015) to efficiently predict the acceleration demand to non-structural
components.
With the objective to further implement the ASAR parameter in probabilistic risk
assessment, a GMPE appropriate for the prediction of ASA40 was developed using
the RESORCE strong ground motion database. Koufoudi etal. (2015) proposed a
ASA40 prediction equation, with a domain of validity, while limited in the low
magnitude range (Mw from 5.5 to 7.6), covers an appropriate distance range (<200
km) and spectral periods (0.014 s), that make this new GMPE a worthy candidate
for future probabilistic seismic hazard and seismic risk analysis. However, these
aspects were not implemented during the project.
6.7 Selection ofTime Histories
6.7.1 Selection Based onUHS
Proper selection of input ground motions for linear and non-linear seismic analyses
of structures and soil systems has become one of the leading topics of research in
earthquake engineering. According to recent trends, as also recognized by the ASCE
recommendations both for buildings and other structures (ASCE/SEI 7-10) and for
nuclear power plants (ASCE/SEI 43-05), the set of input accelerograms should have
some basic properties that may be summarized as follows:
they should come from records of real earthquake events approaching, in terms
of magnitude, distance and site conditions, the conditions that mostly affect seis-
mic hazard for the specific return period;
the average value of response spectra of the input accelerograms should closely
approach the target UHRS (within a range that ASCE 43-05 recommends to be
from 10% to +30%);
a moderate scaling of accelerograms is generally accepted to improve the spec-
tral matching with the UHRS; and
such spectral matching in terms of average spectrum is recommended over a suf-
ficiently wide period range to constrain not only the spectral ordinate at the fun-
damental period of the structure, but the spectral shape as well (e.g., Baker and
Cornell 2006; Haselton etal. 2011).
For structural systems (see e.g., ASCE/SEI 7-10) the latter requirement is typi-
cally introduced to account for the sensitivity of nonlinear structural response to
spectral ordinates at periods larger than the fundamental one (lengthening of the
periods due to damage development), and for the higher modes contribution at
shorter periods. For systems possibly involving nonlinear soil structure interaction
effects, the requirement may even be more stringent, because soil response is gov-
erned by a wide range of frequencies and not by a narrow band around the funda-
mental period of the structure. In the latter case the ASCE/SEI 43-05 recommends
considering the whole frequency range 0.225Hz.
There is no standard method to select acceleration time histories to fit a pre-
scribed target spectrum; the ideal case being the selection of real accelerograms
recorded in seismic conditions close to the target ones, e.g., based on the magnitude,
distance and epsilon coming from the disaggregation of the PSHA at rock. Different
approaches have been proposed for such an optimum selection, although in practi-
cally all cases the accelerograms initially selected are more or less significantly
modified by scaling procedures to improve the fit with the target spectrum.
In D3-96 for the site response analyses performed for sites in the Po plain, the
approach for time-histories selection relies on the following ingredients (Smerzini
etal. 2014):
6.7Selection ofTime Histories 129
a high-quality strong motion database, containing digital records of earthquakes

relevant for seismic hazard studies in Italy, obtained in stations with good seis-
mic site characterization, in most cases provided by the VS30; and
a software tool for automatic selection of strong-motion records compatible on
average with a target spectrum, within a prescribed period range and a prescribed
tolerance.
In a second complementary step of the input motion selection procedure, a
spectrally matched set of accelerograms is created. Starting from the previous
selection, a scaling procedure in the frequency domain is applied. Namely, a correc-
tion factor for the Fourier spectrum of a prescribed accelerogram is calculated based
on the ratio of the response spectrum (RS) of the accelerogram, with respect to the
target one. Such a correction factor is applied to scale iteratively the amplitude of
the Fourier spectrum of the accelerogram, while keeping the same phase as the
original one, until its RS fits the target one. The procedure is similar to that proposed
by Shahbazian and Pezeshk (2010).
Contrary to other spectral matching approaches, since the seed real accelero-
grams have a broadband average spectral compatibility and are also reliable in the
long period range, the frequency scaling does not usually imply a significant modi-
fication of the original record, neither in terms of acceleration, nor of velocity or
displacement.
In D4-153 different sets of time-histories were selected for linear, equivalent
linear and nonlinear site response analyses at two selected sites. Two methods are
applied for the selection and modifications. One set is selected following the
approach employed in D3-96 above. The other procedure is much simpler and does
not involve any spectral matching. The motions are searched within the RESORCE
database (Akkar etal. 2014) according to magnitude-distance hazard disaggrega-
tion and site conditions. Then a linear scaling is performed such that five motions
are scaled to match, in an average sense, the low-frequency (LF) part (above 0.20.3
s) of the target UHS and five are scaled to match the high-frequency (HF) part
(below 0.20.3 s). The scaling factors are between 0.2 and 2.
As indicated in Sect. 5.4.2.1, the spectral matching technique has the advantage
of reducing the uncertainty when broadband spectral matching of the records to the
target spectrum is achieved. Furthermore, results obtained in SIGMA seem to indi-
cate that when using SAFs to calculate ground surface response spectrum the choice
of the set of time histories is not of paramount importance (Sect. 5.3) at least for the
HyS approach.
6.7.2 Selection Based onConditional Spectra
In recent years, ground-motion selection based on Conditional Spectra (Baker 2011;

Jayaram et al. 2011; Carlton and Abrahamson 2014), has gained a growing rele-
vance. In the framework of Seismic Probabilistic Safety Analysis (S-PSA) for
nuclear power plants, the selection of relevant ground motions is the key step for
defining the seismic load input for structural analysis. A classical method for struc-
tural analysis in S-PSA is the use of UHS, through generation of spectrum-
compatible synthetic time histories. The conditional spectrum (CS) approach aims
to avoid the conservatism of the UHS method. The use of such an approach has been
investigated in D5-141.
From the point of view of the seismic hazard computation it is important to know
which approach will be used to select the time histories in order to produce all the
necessary results for its application. The main outputs typically provided are the
UHS at the selected return periods and the magnitude-distance hazard disaggrega-
tion for the selected return periods and intensity measures (e.g., PGA and SA at 1
s). Hazard disaggregation is necessary to select time histories that are consistent
with the source and propagation properties for the earthquakes that control the haz-
ard at the site. If the CS approach is employed as in D5-141, the HCT needs to
provide also hazard curves for each of the GMPEs used in the logic tree. This infor-
mation is used, together with the GMPEs weights in the logic tree, to calculate dis-
aggregation weights for each GMPE that correspond to the fractional contribution
of that model to the total hazard for a given spectral period and hazard level.
Given the considerable amount of work involved in the CS method, compared to
the other selection methods, and in view of the small sensitivity of the site responses
(SAFs) to the choice of time histories, at least in the HyS methodology imple-
mented in SIGMA, use of the CS method was not further pursued in the framework
of SIGMA.
References
Akkar S,Sandkkaya MA, Senyurt M, Azari Sisi A,Ay B, Traversa P,Douglas J, Cotton F, Luzi
L,Hernandez B, Godey S (2014) Reference database for seismic ground-motion in Europe
(RESORCE). Bull Earthq Eng 12(1):311339
Ameri G (2015) Progress Report - Preliminary Hazard Input Document (HID) for the SIGMA
PSHA for the Frances southeastern , Doc. GTR/ARE/0915-1364, September 29, 2015,
GEOTER, Clapier, France
Ameri G, Baumont D, Gomes C, Le Dortz K, Le Goff B, Martin C, Secanell R (2015) On the
choice of maximum earthquake magnitude for seismic hazard assessment in metropolitan
France insight from the Bayesian approach. 9ime Colloque Nat. AFPS, Marne-La-Valle,
December 2015
American Society of Civil Engineers ASCE SEI 43-05 (2005) Seismic design criteria for struc-
tures, systems and components in nuclear facilities, NewYork, USA
American Society of Civil Engineers ASCE SEI 7/10 (2010) Minimum design loads for buildings
and other structures, NewYork, USA.
ATC (1978) Tentative provisions for the development of seismic regulations for buildings. ATC-
3-06 Report, Applied technology council, Redwood City, CA
Baker JW (2011) Conditional mean spectrum: tool for ground motion selection. J Struct Eng
137(3):322331
Baker JW, Cornell CA (2006) Spectral shape, epsilon and record selection. Earthq Eng Struct Dyn
35(9):10771095
References 131
Bardet JP, Ichii K, Lin CH (2000) EERA a computer program for equivalent-linear earthquake
site response analyses of layered soil deposits, University of Southern California, Department
of Civil Engineering
Bazzurro P, Cornell CA (1999) Disaggregation of seismic hazard. Bull Seism Soc Am
89(2):501520
Bojorquez E, Iervolino I (2011) Spectral shape proxies and nonlinear structural response. Soil Dyn
and Earthq Eng 31:9961008
Bommer JJ, Strasser FO, Pagani M, Monelli D (2013) Quality assurance for logic-tree implemen-
tation in probabilistic seismic-hazard analysis for nuclear applications: a practical example.
Seismol Res Lett 84(6):938945
Carlton B, Abrahamson N (2014) Issues and approaches for implementing conditional mean spec-
tra in practice. Bull Seism Soc Am 104(1):503512
Danciu L, Monelli D, Pagani M, Wiemer S (2010) GEM1 hazard: review of PSHA software, GEM
Technical Report 20102. GEM Foundation, Pavia
De Biasio M, Grange S, Dufour F, Allain F, Petre-Lazar I (2014) A simple and efficient intensity
measure to account for non-linear structural behavior. Earthquake Spectra 30(4):14031426
De Biasio M, Grange S, Dufour F, Allain F, Petre LI (2015) Intensity measures for probabilistic
assessment of non-structural components acceleration demand. Earthq Eng Struct Dyn
44(13):22612280
EPRI (Electric Power Research Institute) (1988) A criterion for determining exceedance of the
operating basis earthquake EPRI NP-5930. Palo Alto, CA
EPRI, (Electric Power Research Institute) (1991) Standardization of the Cumulative Absolute
Velocity. EPRI TR-100082, Palo Alto, CA
Gerolymos N, Gazetas G (2005) Constitutive model for 1-D cyclic soil behaviour applied to seis-
mic analysis of layered deposits. Soils Found 45:147159
Haselton C, Baker JW, Liel AB, Deierlein GG (2011) Accounting for ground motion spectral shape
characteristics in structural collapse assessment through an adjustment for epsilon. JStruct Eng
137(3):332344
Hashash YMA, Musgrove MI, Harmon JA, Groholski D, Phillips CA, Park D.DEEPSOIL V6.0.
Urbana, IL
Idriss IM, Sun JI (1992) SHAKE91: a computer program for conducting equivalent linear seismic
response analyses of horizontally layered soil deposits. Program modified based on the original
SHAKE program published in December 1972 by Schnabel, Lysmer and Seed, Center of
Geotechnical Modeling, Department of Civil Engineering, University of California, Davis, CA
lations. Specific safety guide SSG-9, Vienna, Austria
Jayaram N, Lin T, Baker JW (2011) A computationally efficient ground-motion selection algo-
rithm for matching a target response spectrum mean and variance. Earthquake Spectra
27(3):797815
Koufoudi E, Ktenidou OJ, Cotton F, Dufour F, Grange S (2015) Empirical Ground-Motion Models
adapted to a new intensity measure ASA40 (2015). Bull Earthq Eng 13(12):36253643
Kramer SL, Paulsen SB (2004) Practical use of geotechnical site response models. In: Proceedings
of international workshop on uncertainties in nonlinear soil properties and their impact on
modeling dynamic soil response, PEER Center Headquarters, Richmond, CA
Li XS, Wang ZL, Shen CK (1992) SUMDES: a nonlinear procedure for response analysis of
horizontally-layered sites subjected to multi-directional earthquake loading. Department of
Civil Engineering, University of California, Davis
Matasovi N, Ordez G (2007) D-MOD2000, Geomotins, LLC, Computer software
McGuire RK (1995) Probabilistic seismic hazard analysis and design earthquakes: closing the
loop. Bull Seism Soc Am 85:12751284
Modaressi H, Foerster E (2000) CyberQuake: modelling soil responses to seismic tremors. BRGM
(http://www.brgm.eu/scientific-output/scientific-software/scientific-software)
NRC (2007) Regulatory Guide 1.208, A performance-based approach to define the site-specific
earthquake ground motion, Washington, DC
Prevost JH (2010) DYNAFLOW: a nonlinear transient finite element analysis program, Princeton
University, NJ
Shahbazian A, Pezeshk S (2010) Improved velocity and displacement time histories in frequency-
domain spectral-matching procedures. Bull Seism Soc Am 100(6):32133223
Smerzini C, Galasso C, Iervolino I, Paolucci R (2014) Ground motion record selection based on
broadband spectral compatibility. Earthq Spectra 30(4):14271448
Stewart J, On-Lei Kwok A, Hashash YMA, Matasovic N, Pyke R, Wang Z, Yang Z (2008)
Benchmarking of nonlinear geotechnical ground response analysis procedures, PEER report
2008/04, Richmond, CA
Thomas P, Wong I, Abrahamson N (2010) Verification of probabilistic seismic hazard analysis
computer programs. Pacific Earthquake Engineering Research Center; PEER 2010/106,
Berkeley, CA
Von Thun JL, Rochim LH, Scott GA, Wilson JA (1988) Earthquake ground motions for design and
analysis of dams, earthquake engineering and soil dynamics II recent advances in ground-
motion evaluation (GSP 20). ASCE, NewYork, pp463481
Chapter 7
Interfaces Between Subprojects
7.1 SSC andGMC Interfaces
Interface issues between the seismic source characterization and the ground motion
characterization are usually implicitly handled by the hazard computation.
Nevertheless, the most relevant interface topics, for example the depth distribution
and the distance metric, should be discussed among experts from both sides of the
interface. In the following sections, the interface issues identified as the most impor-
tant for the hazard are discussed briefly.
The key generic interface issues for the SSC and GMC are listed and summa-
rized below:
Consistency of earthquake catalogue with ground-motion models:
The magnitude scale used by the GMPEs (at present mainly Mw) needs to be
consistent with the magnitude definition used in the earthquake catalogue and the
magnitude-frequency distributions in order to avoid additional conversions with
associated uncertainties. The magnitudes in the earthquake catalogue are mainly
based on models of the attenuation of macroseismic intensities. The compilation
of an intensity and ground-motion dataset for earthquakes with both types of data
can be useful for this. Furthermore, although this approach has not been imple-
mented in SIGMA, empirical correlations between the European Macroseismic
Scale (EMS) and spectral acceleration (Sa) can be used to evaluate the regional
variations in the EMS-Sa relations and provide an idea of the underlying uncer-
tainties, which are part of the earthquake catalogue.
Magnitude conversions:
Modern earthquake catalogues use magnitude conversions to convert ML into Mw
and MS into Mw or others. Attention must be paid to the consistency of Mw (and
depth) derived from individual macroseismic data points (for historical earth-
quakes) with Mw derived from records (for instrumental earthquakes). This can
be done from a limited set of calibration events in order to reduce the uncertainties

134 7 Interfaces Between Subprojects
related to magnitude conversions and to limit the bias possibly introduced in the
calculation of the activity rates. These conversions should also be checked
against the conversion (if any) used in developing the selected GMPEs. More
recent GMPEs are usually based on Mw so a conversion within the project is not
needed, but there may have been conversions made to develop the databases for
the GMPEs.
Rupture dimensions:
The magnitude-area scaling relations are used in the source characterization and
in the finite fault simulations (FFS). The GMPEs are based on a subset of earth-
quakes with a given area(M) distribution. The area(M) scaling used by the SSC
and GMC should be consistent. Modern databases include estimates of the rup-
ture dimensions and can be used to quantify the area-magnitude relation implicit
in the GMPEs. The area(M) scaling for the NGA dataset is consistent with the
classical area(M) models, but the area(M) model used in the finite fault simula-
tion is inconsistent with larger areas in the FFS; e.g. some FFS approaches have
unconventional area(M) models which should be discussed between the SSC
and GMC teams.
Distance conversions:
Distance conversions may be necessary depending on the source-to-site distance
metric used by the selected GMPEs. Ideally, a distance conversion is avoided or
is directly addressed in the hazard code using the native distance measures of the
ground-motion models. For the reliability of the hazard estimates it is very
important to have a consistent definition of distance metrics (see also Bommer
and Akkar 2012). Today, with some GMPEs making use of the depth to the top
of the rupture, this has become even more relevant as the hazard computed from
near-field sources is sensitive to differences in the distance measures.
Distance extrapolation:
Even if rarely the case in European regions, it may happen that the SSC includes
sources out to 500 km, while ground-motion models were usually only defined
up to 200300km. In consequence, the ground-motion models need to be defined
or extrapolated out to 500km. Most of the time, this does not have a significant
effect on the hazard, but should be addressed for consistency.
Style-of-faulting and dip definitions:
The SSC defines the style-of-faulting categories by dip whereas the GMC usu-
ally uses rake. These different definitions of style-of-faulting and dip need to be
consistent. The range of dips in the ground-motion dataset should be checked for
consistent with the dips for the SSC style-of-faulting classes. Experts can assign
different weights to the different fault styles for each seismic source. Usually, dip
angles are associated with these faulting styles in these particular seismic sources
and uncertainty in this parameter can be introduced in the form of alternative dip
angles. For example, it should be made clear whether composite fault mecha-
nism (e.g., transtension) can be handled by the hazard software and how this is
done. In the presence of zones with unknown style-of-faulting it should be
decided how this will be modelled in the GM logic tree.
7.2 GMC andSRC Interfaces 135
Reducing Mmin:
In the past, most hazard calculations were done using a Mmin of 5. Depending on
the project requirements, the hazard calculation may want to use a minimum
magnitude of 4.5 or even lower when a CAV filtering approach is applied or to
check the PSHA outputs against observations in regions with moderate activity.
For the SSC, there is no issue with reducing Mmin. For the GMC, the available
ground-motion models may not extrapolate well to smaller magnitudes. In this
case, the ground-motion models will need to be adjusted for smaller magnitudes
using the local ground-motion data as a constraint. The latest GMPEs [like NGA-
West2 and some GMPEs developed within SIGMA WP2: Ameri (D2-131) and
Drouet and Cotton (2015)] have been developed with data down to magnitude 3
and thus, can often be directly used, but the characteristics of the local ground-
motion data should always be compared to the lower magnitude data used by the
GMPE developers in order to identify potential regional differences.
Mmax:
In a limited number of European seismic regions, the Mmax values from the SSC
can be as high as magnitude 8. The ground-motion models need to be checked
regarding appropriate extrapolation to magnitude 8.
Depth Distribution of Earthquakes in the Study Region
The earthquake focal depth distributions in the study region should be deter-
mined, as they can be very different depending on the tectonic setting. This, of
course, requires acquisition of high quality data over a sufficiently long time
period. Depth is an important parameter for those GMPEs using rupture dis-
tance, as it affects the distance computation. Thus, this parameter should be dis-
cussed among the experts in charge of the SSC and GMC in order to have all the
necessary background information to adopt a consistent definition.
For the modelling, different magnitude-dependent depth distributions might
be adopted, implicitly reflecting the uncertainty of this parameter. Thus, the
hypocentral depth distribution can be: trapezoidal, triangular (a special form of
trapezoidal distribution), truncated/non-truncated normal and multiple uniform
(histogram) distributions. Similarly, a magnitude-dependent depth distribution
may be appropriate where the largest events cannot occur in the shallower part of
the seismogenic layer.
7.2 GMC andSRC Interfaces
Special attention should be paid to the treatment and transfer of uncertainties

between the ground motion characterization on rock and the site/soil response char-
acterization (SRC). This interface becomes relevant once the hazard evaluation is
split into the evaluation of GMPEs only up to the bedrock level and then adding the
site-specific soil response on top to reach a sub-surface layer or the surface. In case
the GMPEs are evaluated directly up to the surface this interface, of course, does not
exist in the described manner. In SIGMA both approaches have been investigated
and one of the conclusions was that there is a potential benefit in separating the
GMC on rock and the site-specific soil amplification in order to reduce in each area
the associated uncertainties. The key issues that have to be addressed in order to
avoid double counting of uncertainties are:
Partition of aleatory variability;
Host-to-target conversions and consistency with site amplification models;
Maximum ground motion; and
V/H models.
It should be mentioned that the maximum ground motion truncation models are
handled differently when separating the rock and soil ground motions. For both, the
same empirical database might be evaluated and used for the development of the
individual models. In general, there are no problems. It is hard to define an upper
limit to rock ground motion, especially in very hard rock conditions. On the other
hand, there is more empirical data available for soil and based on the geotechnical
soil properties, physical limits exist.
In the following, the common key interface issues between GMC and SRC are
briefly discussed.
A clear definition of the reference bedrock level and properties is necessary, as
this is the handover between the rock hazard and the soil response. Seismologists
see the bedrock as the geological base rock (generally the bottom of the sedimen-
tary column), while engineers are used to the practical engineering bedrock as
defined in construction standards. Furthermore, there are different interpretations
of rock (e.g. soft, stiff, hard and very hard), which can be clarified by introduc-
ing, e.g., VS30 as a criterion. This also implies checking the consistency of the
rock velocity profiles at the target site with the rock profiles for the GMPE data-
sets (see Chap. 4).
For soft rock conditions, the ratio of spectral acceleration to PGA will reach 1 at
about 4050Hz as opposed to hard rock sites, where this ratio is reached at about
100Hz.
Reducing Mmin to 4.5 or lower requires the ground-motion models to be appli-
cable through adjustments for M < 5 and that the extrapolation of non-linear
effects (if any) to 4.5 or lower is checked.
Avoiding double counting of site amplification variability due to input rock
ground motions and uncertainty in the soil: The aleatory variability of site ampli-
fication (between sites with the same VS30 and within a single site for different
input motions) is included in the GMC on rock if the traditional standard devia-
tions for GMPEs are used. It is also included in the epistemic uncertainty of the
site profile for the SRC and in the randomness of amplification from different
input time histories. The focus here is on the variability due to the use of input
time histories for rock ground motions applied at the bedrock level. There are
some alternative approaches to address this potential double counting, which
have been investigated and applied for SIGMA (see Sect. 5.4). Nowadays, there
is a trend for having the GMC remove the epistemic part by switching to a
7.3 Single-Station Sigma 137
s ingle-station sigma approach and having the SRC not add the aleatory variabil-
ity from different input time histories for the linear range of input ground motions
and to apply just the median amplification for site response. Furthermore, as
done also in SIGMA, there is also the alternative to use time histories tightly
matched to the rock UHS.Vertical aleatory variability: The GMC includes addi-
tional aleatory variability for the vertical ground motion components. The SRC
needs to evaluate the value of this uncertainty so that site response experts can
evaluate the effect on the vertical aleatory models. As the vertical component
was not part of SIGMA, the latter has not been an issue within the project.
Other interface issues to be considered are
The site response models incorporate non-linear effects. The ground-motion
models used by the GMC also include non-linear effects. These non-linear
effects should be consistent with the application. In case the reference rock con-
ditions are greater than 700 m/s, the non-linearity in the ground-motion model
will be very small. The differences in the non-linear models between GMC and
SRC will, therefore, not have a significant effect.
There are 2D and 3D effects in the site response models. Some of the variability
due to 2D and 3D effects is captured in the GMC aleatory terms. It is not clear
how much double counting exists due to this issue. This should be explored
depending on the topographic and near-surface geological configurations in the
study region.
Frequency content of rock ground motions: The site response usually uses times
histories to define the input rock motion. The consistency of the frequency con-
tent (spectral shape) of the Fourier spectra of the input rock motions used by
SRC with that from the GMC model needs to be checked.
Host-to-target parameter (Kappa) for input rock motion: The frequency content
characterized by the GMC for the input rock motion implies some Kappa values.
The site response input motions should be developed accordingly (see Sect. 4.3).
7.3 Single-Station Sigma
In terms of the interface between GMC and SRC, the use of single-station sigma
removes the effect of site-to-site differences in the average amplification from the
rock sigma, but the variability of the soil amplification (due to different rock input
time histories, for example) is still included in the single-station rock sigma values.
The common empirical datasets used to estimate sigma are dominated by ground
motions in the linear range, which originated mainly from weak motions. So, the
soil site amplification variability for linear response is mainly included in the rock
site single-station sigma. When separating the evaluation of rock ground motions
from soil (surface) ground motions the effects, if there is potential nonlinearity in
the soil, need to be addressed. The site amplification variability for cases with highly
nonlinear soil site response is not represented in the rock site single-station sigma.
Thus, if high levels of shaking within the soil are expected for the given site, an
additional aleatory variability on top of the epistemic considerations can be included
in the SRC part to cover those effects, or it can be explicitly covered in the SRC
through consideration of nonlinear models.
The use of single-station sigma requires that the epistemic uncertainty in the site-
specific amplification be captured in the SRC logic tree model through alternative
models for the VS profile and the non-linear properties, as illustrated already in
Chap. 5.
This concept applies to both the horizontal and the vertical ground motion
components.
7.4 V/H Models forRock andSoil
Even though in SIGMA the vertical component of hazard was not treated, some
considerations and issues are briefly summarized here for the interested reader.
In the GMC, the rock V/H ratios are evaluated and weighted separately from the
horizontal median models. In the rock site hazard calculation, the V/H ratios are
applied to the combined fractiles of the hazard curves for the horizontal
component.
In the SRC, two approaches can be used to provide V/H ratios and amplification
factors, which scale the horizontal rock ground motion to vertical soil ground
motion: (1) V/H ratios for soil and amplification factors for horizontal motion are
developed by the SRC experts and are combined, and (2) V/H ratios for rock by the
SRC experts and amplification factors for vertical motion by the SRC experts are
combined. For the first approach, the different soil V/H ratios are combined on a
per-model basis with the model-specific horizontal motion amplification. For the
second approach, the V/H rock ratios of all models are combined with the SRC
vertical motion amplification factors. Both approaches cover horizontal-to-vertical
motion scaling and amplification. The results of both approaches are combined into
a single soil input for the target site. This soil input is used to compute the vertical
soil motion hazard on the basis of the horizontal rock motion hazard, which is
always the full (all fractiles and complete SSC and GMC models) horizontal
motion rock hazard.
The second approach mentioned above is similar to the computation of the verti-
cal motion rock hazard in that the full horizontal motion rock hazard is combined
with all GMC V/H models, i.e. there is no correlation between (but full combination
of) the GMC-specific rock hazard subsets and the GMC expert-specific V/H rock
ratios.
References 139
References
Bommer JJ, Akkar S (2012) Consistent source-to-site distance metrics in ground-motion predic-
tion equations and seismic source models for PSHA.Earthquake Spectra 28:115
Drouet S, Cotton F (2015) Regional stochastic GMPEs in low-seismicity areas: scaling and alea-
tory variability analysis application to the French Alps. Bull Seismol Soc Am
105(4):18831902
Chapter 8
Probabilistic Seismic Testing andUpdating
ofSeismic Hazard Results
Considering the high uncertainties in PSHA and the importance of PSHA results for
seismic design and retrofit, it is pertinent to focus on the issue of consistency check-
ing of the PSHA results. In the last decade several approaches to test PSHA results
have been published (e.g. Stirling and Petersen 2006; Stirling and Gerstenberger
2009, 2010; Stirling 2012; Mucciarelli etal. 2008; Humbert and Viallet 2008; Labb
2010; Anderson etal. 2011; Mezcua etal. 2013; Selva and Sandri 2013; Gribovszki
etal. 2013). In addition, several recent opinion papers are encouraging hazard ana-
lysts to carry out tests of PSHA results (e.g. Stein etal. 2011). During the SIGMA
period an international workshop was held in Pavia (NEA 2015) on Testing PSHA
results and Benefit from Bayesian Techniques for Seismic Hazard Assessment
which concluded that a state-of-the-art PSHA should include a testing phase against
any available observation, including any kind of observation and any period of
observation, including instrumental seismicity, historical seismicity and paleoseis-
micity data if available. It should include testing not only against its median hazard
estimates but also against their entire distribution (percentiles).
Testing applications have been conducted in different countries (United States,
New Zealand, Italy and France). The techniques rely on various statistical assump-
tions but any testing technique must deal with the available observation time win-
dow. Considering the observation time window in seismology testing (approximately
80 years for instrumental networks in California and 45 years in Europe, and several
centuries for historical data), and the return periods of interest in engineering seis-
mology (in the order of several hundreds to thousands of years), the comparison of
observations and predictions is challenging.
Recent literature shows that PSHA models can be evaluated using different types
of observations, such as intensities, synthetic accelerations (converted from intensi-
ties or predicted from an earthquake catalogue and a ground-motion model), and
recorded accelerations at instrumented sites.
In the framework of SIGMA, WP4 specific actions were undertaken to develop
new methods for comparing the ground-motion distribution with observations, in

142 8 Probabilistic Seismic Testing andUpdating ofSeismic Hazard Results
AFPS 2006 47 RAP stations

25
20
N SITES 15
10
0
23 30 40 50 60 70 80 90 100 110 120 134
Accelerations (cm/sec2)
Fig. 8.1 Testing the probabilistic seismic hazard model against accelerometric data in France:
predicted and observed number of exceedance. Blue curves: median and percentiles 2.5 and 97.5
of the predicted distributions, red stars: observed number of sites. Results for the AFPS2006 PSH
model, considering a range of acceleration thresholds (47 rock sites, total time window 485 years)
(From D4-75)
particular with the objective of implementing and testing methods to clearly under-
stand their potential as well as their limits and accuracy.
Overall, the testing of PSHA models and results is an exercise that is worth
pursuing.
8.1 P
SHA Testing Using Acceleration andMacroseismic
Intensity Data
The topic of testing probabilistic seismic hazard models against observations by

using accelerometer data is discussed in SIGMA Deliverable D4-75. Three hazard
models for France are tested against a set of French accelerometric records (RAP
network, http://rap.resif.fr/). Two of the models (MEDD2002 and AFPS2006) pro-
vide hazard estimates for the whole metropolitan territory, while the third
(SIGMA2012, D4-41) focuses on Southern France. The method is based on com-
paring predicted exceedance rates for a set of ground motion levels with observed
ones at a set of sites (see Fig.8.1). In order to reduce the influence of site conditions,
which might considerably change the observed ground motion level, only record-
ings from rock sites are included in the analysis. Models are tested against different
datasets, true-recorded accelerations and synthetic amplitudes based on an earth-
quake catalogue combined with a GMPE.
The PSHA testing using recorded accelerometer data presented in D4-75 is
promising, although it suffers from a limited applicability to the French context due
to the relatively moderate seismicity, which does not allow for testing significant
8.1 PSHA Testing Using Acceleration andMacroseismic Intensity Data 143
ground-motion thresholds (i.e. low frequencies of exceedance). Another relevant

issue is related to site response, which is typically not considered in the testing pro-
cess by selecting only rock sites. However, the site classes of several French stations
are not well known, which increases the uncertainties in the process.
The authors conducted an application of the same approach to Turkey in order to
test PSHA for larger ground-motion levels (see Tasan et al. 2014). The SHARE
hazard model (Woessner etal. 2015) is tested against Turkish strong-motion data.
Two critical issues are identified: (1) the completeness of the recorded data at each
station and the need to identify potential gaps in the recordings, and (2) the fact that
most accelerometer stations are on soil sites while the hazard results are for rock
sites. To solve this problem, PGA is converted into rock PGA using empirical site-
amplification functions based on VS30 at the site. This process carries, however, large
uncertainties that are not considered because VS30 is only a proxy of the real site
amplification and only average site amplification can be estimated. The results
show that the observed rates of exceedance at different sites are well within the
bounds of the predicted distribution for accelerations between ~0.1 and 0.4 g. For
higher levels no conclusion can be drawn. Despite the previous uncertainties, the
method offers a rationale to assess a PSHA model at a 475-year return period.
In order to overcome the limitations related to the short lifetime of accelerometer
stations, macroseismic intensity observations can be considered as an alternative, as
they are available on a much longer period of time. Moreover, since macroseismic
intensities are not directly used as input in standard PSHA models issues related to
the independence between the model and the data used to test it may be avoided.
This approach is pursued in D4-118 and the follow-up study D4-154 that pro-
poses a method to compare PSHA results with historical information on damage
obtained from macroseismic intensity observations. The method consists in con-
verting the intensities into PGA by means of fragility curves to obtain estimates of
the PGA corresponding to the level of observed damage. Thus, the method requires
a procedure to convert records of historical observations into equivalent PGA esti-
mates. Such a conversion process carries uncertainties that are taken into account by
means of a logic tree approach (see Fig. 8.2). The empirically-derived rates of
exceedance are compared to the model derived estimates. The method was applied
to south-eastern France.
The approach proposed in D4-118 suffers from very large uncertainties in the
process of inverting the fragility functions. Such large variability in empirical PGA
exceedance estimates prevents, in fact, the results of this methodology from being
of practical use. In that respect, it should be mentioned that the methodology pro-
posed by Labb (2010) does not require any conversion from macroseismic intensi-
ties into accelerations nor does it require inversion of the fragility curves; however,
this approach was not tested in SIGMA.A possible improvement, which was imple-
mented in SIGMA, consists of including recorded accelerations in the procedure to
better constrain the PGA distributions. In particular, this will improve the quality of
estimation for lower values of PGA, whose rates of occurrence are likely to be
underestimated by operating on the macroseismic database. Other important con-
cerns are: (a) the treatment of the incompleteness of the catalogue, and the fact that
Fig. 8.2 Sketch of the proposed methodology in D4-118
large amount of damage is caused by induced events such as slope failures, and (b)
the need to integrate site response in this method.
The limitations found in D4-118 were considered in D4-154. This continuation
study was intended to respond to the criticisms and to explore several lines of devel-
opment, the most important one being the use for comparison of the mean damage
ratio (Lagomarsino and Giovinazzi 2006) rather than the ground motion itself;
although this latter paper does not address directly PSHA testing it is essential as it
ensures consistency between different fragility curves expressed as function of PGA
or macroseismic intensity. The implementation of the revised method to the south-
east quarter of France, and a complete re-visitation of the macroseismic database
were also included in this work. The method is applied to single sites, aggregated
sites, and at a regional scale. At each scale comparisons with historical data are car-
ried out.
Results for single sites are not promising, as already discussed in D4-118.
However, by using appropriate criteria in the selection of sites to satisfy the assump-
tion of ergodicity (independence, and common distributions), the authors demon-
strate that better comparisons are achieved for aggregated sites and for the regional
scale.
The method proposed in D4-154 is judged to be promising. It will benefit from
additional work to confirm its actual value as a tool to check the consistency of
PSHA analyses. To gain confidence in the methodology, it was suggested in the
project to consider a case study undertaken for a region with more abundant build-
ing damage data.
References 145
In conclusion, it is worth noting that the procedure proposed by Labb (2010)

does not require any conversion of macro seismic intensities into accelerations, not
does it require inversion of the fragility curves.
8.2 Bayesian Update ofPSHA
A different perspective in using observed data to test PSHA models is presented in

D4-139. A Bayesian approach is applied to update the PSHA logic tree based on
observed exceedances of a certain acceleration threshold. This approach differs
from the previous ones because its objective is not to compare observed and pre-
dicted number of exceedances but to use such comparisons in a Bayesian frame-
work in order to modify the logic tree weights. In practice, the logic tree weights of
the branches that are in better agreement with the observations are increased.
The study performed in D4-139 presents a first simplified application of this
approach. It provides the basis for the method and represents an initial step in exam-
ining ways of using additional information to update PSHA results in the French
environment. The strategic choice of concentrating on updating the weights of the
logic tree is appealing and could be made applicable to real cases; but it still requires
additional research and development to reach such a stage.
Further improvements should be considered to extend this approach to a real and
complete case study. In its present form, the following limitations and improve-
ments are identified:
The method considers only one ground-motion threshold level, therefore one set
of new LT weights, to update the entire hazard curve. To be rigorous, since the
set of updated weights changes with the value of the threshold, the updated set of
weights should be calculated for every discrete value of the ground-motion
threshold in the entire range of interest.
The application of the method requires defining criteria to make an appropriate
selection of the prior set of alternative branches, and an appropriate set of prior
weights to the branches.
The updating method should be applied in the spectral acceleration domain and
not only to PGA.This would allow an appreciation of the variability of weights
in the spectral domain and to update the full response spectrum.
References
Anderson JG, Brune JN, Purvance M, Biasi G, Anooshehpoor A (2011) Benefits of the use of
precariously balanced rocks and other fragile geological features for testing the predictions of
probabilistic seismic hazard analysis. In: Faber MH, Kehler J, Nishijima K (eds) Applications
of statistics and probability in civil engineering. Taylor & Francis, London, pp744752
Gribovszki K, Bokelmann G, Szeidovitz Gy, Paskaleva I, Brimich L, Kovcs K, Mnus P (2013)

Comprehensive investigation of intact, vulnerable stalagmites to estimate an upper limit on
prehistoric horizontal ground acceleration. Recent advances in earthquake engineering and
structural dynamics (VEESD conference), 2830 August, Vienna, Austria
Humbert N, Viallet E (2008) A method for comparison of recent PSHA on the French territory with
experimental feedback. In: Proceedings of 14th world conference on earthquake engineering,
1217 October, Beijing, China
Labb P (2010) PSHA outputs versus historical seismicity; Example of France. In: 14th European
conference on earthquake engineering, August 30September 3, Ohrid, FYROM
Lagomarsino S, Giovinazzi S (2006) Macroseismic and mechanical models for the vulnerability
and damage assessment of current buildings. Bull Earthq Eng 4(4):415443
Mezcua J, Rueda J, Garca Blanco RM (2013) Observed and calculated intensities as a test of a
probabilistic seismic-hazard analysis of Spain. Seismol Res Lett 84(5):772780
Mucciarelli M, Albarello D, DAmico V (2008) Comparison of probabilistic seismic hazard esti-
mates in Italy. Bull Seismol Soc Am 98:26522664
NEA (Nuclear Energy Agency) (2015) Workshop on testing probabilistic seismic hazard analysis
results and the benefits of Bayesian techniques, Pavia, Italy, 46 February, NEA/
CSNI/R(2015)15. www.oecd-nea.org
Selva J, Sandri L (2013) Probabilistic seismic hazard assessment: combining Cornell-like
approaches and data at sites through Bayesian inference. Bull Seismol Soc Am
103(3):17091722
Stein S, Geller R, Liu M (2011) Bad assumptions or bad luck: why earthquake hazard maps need
objective testing. Seismol Res Lett 82(5):623626
Stirling MW (2012) Earthquake hazard maps and objective testing: the hazard Mappers point of
view. Seism Res Lett 83(2):231232
Stirling M, Gerstenberger M (2009) Development of tests for long-term earthquake ground motion
forecasts in New Zealand, GNS science consultancy report 2009/140. p20, Lower Hutt, NZ
Stirling M, Gerstenberger M (2010) Ground motionbased testing of seismic hazard models in
New Zealand. Bull Seismol Soc Am 100(4):14071414
Stirling M, Petersen M (2006) Comparison of the historical record of earthquake hazard with
seismic-hazard models for New Zealand and the continental United States. Bull Seismol Soc
Am 96:19781994
Tasan H, Beauval C, Helmstetter A, Sandikkaya A, Guguen P (2014) Testing probabilistic seismic
hazard estimates against accelerometric data in two countries: France and Turkey. Geophys
JInt 198(3):15541571
Woessner J, Laurentiu D, Giardini D, Crowley H, Cotton F, Grnthal G, Valensise G, Arvidsson R,
Basili R, Dermicioglu MB, Hiemer S, Meletti C, Musson RW, Rovida AN, Sesetyan K, Stucchi
M (2015) The SHARE consortium, the 2013 European seismic hazard model: jey components
and results. Bull Earthq Eng 13(12):35533596
Chapter 9
Summary andWay Forward
During 5 years SIGMA has not only achieved significant steps forward in the meth-
ods for evaluating seismic hazard at a site, but it also contributed to establishing a
strong network of academic institutions, researchers and engineering companies
that will survive after the project ends. The overall organisation of the project fos-
tered very lively and fruitful discussions between all participants, including with the
members of the Scientific and Steering Committees.
Despite the amount of effort devoted to SIGMA, significant endeavours remain
to be made to increase the reliability of seismic hazard predictions and to better
understand and constrain the associated uncertainties. In fact, not all topics related
to SHA have been covered during the course of the project, either because they were
discarded after its start or because new topics emerged during the 5 year programme.
Without repeating all the achievements and lessons learned, SIGMA clearly stressed
the importance of collecting data to feed and constrain the models: a uniform seis-
mic catalogue is essential to better characterize the seismic source models, the
RESORCE database is useful for developing GPMEs adapted to the seismotectonic
context of the considered regions and site characterization through field instrumen-
tation is mandatory to classify the site, to evaluate the site response and to imple-
ment the single-station sigma approach. All these improvements in empirical
knowledge contribute to a better estimation of the hazard and characterization of the
epistemic uncertainties.
Aside from these achievements, the following topics were identified as important
and warranting more studies in the future.
9.1 Seismic Source Characterization
More data and studies are needed to better characterize the source models and more
specifically:

148 9 Summary andWay Forward
In France fault characteristics should be studied to assess their geometries (dip

angle, rake, dimensions) and to estimate their activity rates, in a way perhaps
similar to the Italian DISS fault database;
In areas of low to moderate seismicity the recurrence Gutenberg-Richter param-
eters, a and b, carry large uncertainties that should be better constrained, espe-
cially for magnitudes derived from macroseismic observations;
The maximum magnitude associated with each source zone, which has a large
impact on the hazard for long return periods, should be studied from different
perspectives: a physically-based approach relying on scaling relationships related
to fault dimensions, a statistically-based approach like the EPRI Bayesian
method and a model-based approach in connection with the G-R parameters
used to reflect the constraints imposed on the deformation rates. Ideally, these
different approaches must be reconciled.
9.2 Ground Motion Characterization
While an essential product of the SIGMA project has been the development of
RESORCE, this database should be maintained, updated and enriched with addi-
tional parameters such as: fault rupture parameters, natural frequency of record-
ing site f0, , presence of basin effects and so on;
Efforts to develop GMPEs specifically adapted to the European context should
be pursued taking advantage of the RESORCE database;
More research is needed to identify the most appropriate rock/hard rock adjust-
ments to GMPEs, which is a problem typically encountered on exposed rock
sites in SCRs;
9.3 Site Response Characterization
Investigations on the nonlinear behaviour of soils under cyclic loading are needed
from both the experimental and numerical viewpoint to answer critical questions
such as: is nonlinear behaviour supported by experimental evidence? Beyond
which threshold amplitude is it necessary to take it into account? What are the
prerequisites and validation steps for a nonlinear constitutive model?
The range of applicability of 2D geometric models for site response analyses
must be ascertained;
The spatial variability in soil characteristics needs to be considered in 2D site
response evaluations;
With the increasing number of available earthquake records, the selection of time
histories for site response analyses should be guided in more detail;
The evaluation of the vertical component of motion at the ground surface is an
issue that deserves further investigation.
9.5 Risk Assessment 149
9.4 Hazard Calculation
Utmost efforts should be devoted to testing of PSHA outputs. Testing should be

performed with instrumental data for short return periods (~30 years), with his-
torical seismicity in terms of macroseismic intensities for return periods up to
some hundreds of years, and with paleoseismicity data for very long return peri-
ods (thousands of years). Testing can be made at a site scale where the hazard
should be consistent with the local observations, and at a regional scale where
the hazard should be consistent with the average regional hazard;
Updating of PSHA using Bayesian approaches should be investigated, possibly
beyond the simple updating of weights in the logic trees; however, even in this
simple form, it helps assigning objective weights to the logic tree branches in
place of the experts judgment;
Based on progress made concerning knowledge of faults, a totally different
approach for hazard evaluation based on fault numerical simulations needs to be
investigated. Such finite fault simulations can be developed independently of
GMPEs and are mainly physics-based. Today, GMPEs are still sometimes used
in this context for reference comparison or scaling.
9.5 Risk Assessment
These aspects have barely been touched upon in SIGMA and warrant additional
studies.
Identification of ground motion intensity parameters like ASA40 that could be
good predictors of structural damage and subsequent development of associated
GMPEs that would allow direct probabilistic risk assessment of critical
facilities;
Development of hazard consistent fragility curves for different typologies of
structures (e.g. buildings, industrial facilities), systems and components.
Obviously, the issues listed above are far from representing an exhaustive list of
all topics of interest, but they constitute open questions the answers to which would
greatly contribute to improving hazard estimates and reducing the epistemic uncer-
tainties attached to them.
Annexes
Annex 1: List ofCommittees Members
Steering Committee
Pierre LABBE, Chairman EDF

Catherine BERGE-THIERRY CEA
Paolo CONTRI ENEL
Jean Pierre RZEPKA CEA
Jean Franois SIDANER AREVA
Senior Hazard Analysis Review Panel-SHARP
Yoshimitsu FUKUSHIMA International Atomic Energy Agency (Japan)

Robert KENNEDY RPK Structural Mechanics Consulting: (USA)
Roger MUSSON British Geological Survey (UK)
Alain PECKER Chairman of Scientific Committee (France)
Rui PINHO GEM Foundation (Italy)
Scientific Committee
Alain PECKER, Chairman French Association for Earthquake Engineering

(France)
Pierre Yves BARD ISTerre Institut des Sciences de la Terre, Grenoble
(France)
David BAUMONT FUGRO (France)

Geological and Earthquake Engineering 42, DOI10.1007/978-3-319-58154-5
152 Annexes
Olivier BELLIER CNRS/Universit Paul Czanne, Aix-En-Provence

(France)
Thierry CAMELBEECK Royal Observatory of Belgium (Belgium)
John DOUGLAS1 BRGM Bureau de Recherche Gologique et Minire
(France)
Ezio FACCIOLI Politecnico di Milano (Italy)
Michel GRANET EOST Ecole et Observatoire des Sciences de la Terre,
Strasbourg (France)
Aybars GURPINAR Consultant (Turkey)
Peter MOCZO University of Bratislava (Slovakia)
Marco MUCCIARELLI University of Potenza (Italy)
Francesca PACOR INGV Instituto Nazionale di Geofisica e Vulcanologia,
Milano (Italy)
Roberto PAOLUCCI Politecnico di Milano (Italy)
Philippe RENAULT SWISSNUCLEAR (Switzerland)
Jean SAVY SRC Savy Risk Consulting (USA)
Frank SCHERBAUM University of Postdam (Germany)
Gianluca VALENSISE INGV Instituto Nazionale di Geofisica e Vulcanologia,
Rome (Italy)
Gordon WOO Consultant (UK)
SIGMA Work Package Leaders
Gloria SENFAUTE Project manager EDF (France)

Mirko CORIGLIANO Deputy project manager ENEL (Italy)
WP1: Seismic Source
Leader: Kevin MANCHUEL (EDF)
Co-leader: Mirko CORIGLIANO (ENEL)
WP2: Prediction models
Leader: Paola TRAVERSA (EDF)
Co-leader: Fabrice COTTON (ISTerre)
WP3: Site effects
Leader: Fabrice HOLLENDER (CEA)
Co-leader: Deborah SICILIA (EDF)
WP4: Hazard Model
Leader: Jean-Michel THIRY (AREVA)
Co-leader: Gloria SENFAUTE (EDF)
WP5: Exploitation of Seismic Ground Motion
Leader: F.ALLAIN (EDF)
Co-leader: Stephane GRANGE (3SR, University Joseph Fourier of Grenoble)
Now at University of Strathclyde, Glasgow, UK

1
Annexes 153
SIGMA Partners
Basilicata University Potenza: http://www.unibas.it

BGS (British Geological Survey): www.bgs.ac.uk
BRGM (Bureau de recherches gologiques et minires): www.brgm.fr
C E R E G E (Centre Europen de Recherche et dEnseignement): www.cerege.fr
Comenius University in Bratislava: http://www.uniba.sk
EMSM (European Mediterranean Seismological Centre): www.emsc-csem.org
EOST (Institut de Physique du Globe de Strasbourg): www.eost.u-strasbg.fr
EUCENTRE (European Centre for Training and Research in Earthquake): http://
www.eucentre.it/
GEOAZUR: https://geoazur.oca.eu
GEOTER (Gologie Tectonique Environnement et Risques): www.geoter.fr
GEM (Global Earthquake Model), Italy: http://www.globalquakemodel.org/
IAEA (International Atomic Energy Agency): www.iaea.org
INGV (Istituto Nazionale di Geofisica e Vulcanologia, Milano: www.mi.ingv.it
INGV (Istituto Nazionale di Geofisica e Vulcanologia, Roma): www.ingv.it
IUSS (Institute for Advanced Study of Pavia): http://www.iusspavia.it/
ISTerre (Institut des Sciences de la Terre): www.isterre.fr
Laboratoire 3S-R (laboratoire Sols, Solides, Structures Risques): www.3s.hmg.
inpg.fr
METU (Middle East Technical University): http://www.metu.edu.tr/
ORB (Observatoire Royal de Belgique): http://www.astro.oma.be
POLIMI (Politecnico di Milano): www.polimi.it
RPK (Structural Mechanics Consulting)
SRC (Savy Risk Consulting): www.savyriskconsulting.com
SWISSNUCLEAR, Switzerland: http://www.swissnuclear.ch
UJF (Universit Joseph Fourier): www.ujf-grenoble.fr
Universit de Versailles Saint-Quentin-en-Yvelines: http://www.uvsq.fr
Universit Paul Czanne Aix Marseille: www.univ-cezanne.fr
University of California, Berkeley: http://www.berkeley.edu
University of Potsdam: http://www.uni-potsdam.de
Stphane Drouet: Earthquake Ground-Motion Consultant
Virginia Tech University, USA: www.vt.edu
154 Annexes
Annex 2: List ofPublications
Publications inInternational Journals withPeer Review
Title Authors Review/Journal

Reference database for seismic Akkar S., Sandkkaya M., Bull. Earthquake Eng.,
ground-motion Senyurt M., Azari Sisi A., Ay (2014), 12, 311339
in Europe (RESORCE) B.O., Traversa P., Douglas J.,
Cotton F., LuziL., Hernandez
L, Godey S.
Taxonomy of kappa: a review of Ktenidou O., Cotton F., Seismol. Research Letters
definitions and estimation Abrahamson N., and Anderson J. (2014), 85(1), 135146.
approaches targeted to
applications.
*
Comparisons among the five Douglas J., Akkar S, Ameri G., Bull. Earthquake Eng.
ground-motion models developed Bard PY, Bindi D., Bommer J., (2014), 12, 341358
using RESORCE for the Singh Bora S., Cotton
prediction of response spectral F.,Derras B., Hermkes
accelerations due to earthquakes M.Kuehn N.M.,Luzi L.,
in Europe and the Middle East Massa M., Pacor F., Riggelsen
C., Sandkkaya A., Scherbaum
F.,Peter J., Stafford, Traversa P.
Testing Probabilistic Seismic Tasan H., Beauval C., Geophys. J.Int. (2014),
Hazard estimates against Helmstetter C A., Sandikkaya 198, 15541571
accelerometric data in two A., Guguen Ph.
countries: France and Turkey.
*
Towards fully data driven Derras B., Bard P.Y.Cotton F. Bull. Earthquake Eng.
ground-motion prediction models (2014), 12, 495516
for Europe
Site Amplification: Dependency Laurendeau A., Cotton F., Bull. Seism. Soc. Am.
on VS30 and 0. Bull. Seism. Soc. Ktenidou O., Bonilla F., (2013), 103 (6),
Am. Hollender F. 31313148
Moderate Earthquake Teleseismic Letort J., Vergoz J., Guilbert J., Bull. Seism. Soc. Am.
Depth Estimations: New Methods Cotton F., Sebe O., Cano Y. (2014), 104 (2), 593607.
and Use of the Comprehensive
Nuclear-Test-Ban Treaty
Organization Network Data
A simple and efficient intensity De Biasio M., Grange S., Earthq. Spectra, (2014),
measure to account for non-linear Dufour F., Allain F., Petre- 30(4), 14031426
structural behaviour Lazar I.
*Pan-European Ground-Motion Bindi D., Massa M., Luzi L., Bull. Earthquake Eng.,
Prediction Equations for the Ameri G., Pacor F., Puglia R., (2014), 12, 391530
Average Horizontal Component Augliera P.
of PGA, PGV, and 5%-Damped
PSA at Spectral Periods up to
3.0s using the RESORCE dataset
Annexes 155

The Ferrara thrust earthquakes of Priolo E., Romanelli, Barnaba Annals of Geophysics,
May-June 2012: preliminary site C., Mucciarelli M., (2012), 55(4), 591597
response analysis at the sites of Laurenzano G., DallOlio L.,
the OGS temporary network Abu Zeid N., Caputo R.,
Santarato G., Vignola L., Lizza
C., Di Bartolomeo P.
Recent evolution and challenges Faccioli E. Bull Earthquake Eng
in the Seismic Hazard Analysis (2013), 11(1), 533
of the Po Plain region, Northern
Italy
*Simultaneous Quantification of Hermkes M., Kuehn N.M., Bull. Earthquake Eng.,
Epistemic and Aleatory Riggelsen C. (2014), 12, 449466.
Uncertainty in GMPEs using
Gaussian Process Regression
*
Fourier Spectral and Duration Bora S.S., Scherbaum F., Bull. Earthquake Eng.,
Models for the Generation of Kuehn N., Stafford P. (2014), 12, 467493
Response Spectra which are
Adjustable to Different Source
Propagation and Site Effects
A New Approach to Include Path Dawood H., Rodriguez-Marek A. Bull. Seism. Soc. Am.,
Effects in Ground Motion (2013), 103(2B),
Prediction Equations Using the 13601732.
Mw 9.0 Tohoku Earthquake
Aftershocks.
Sensitivity of coda waves to Mayor J., Margerin L., Geophysical Journal
lateral variations of absorption Calvet M. International, (2014),
and scattering: radiative transfer 197(2), 121.
theory and 2-D examples
Source of scatter in : analysis of Van Houtte C., Ktenidou O.J., Bull. Seism. Soc. Am.
Christchurch data and Larkin T., Holden C. (2014), 104 (4),
comparison with local ground 18991913.
motion prediction models
Regional Stochastic GMPEs in Drouet S., Cotton F. Bull. Seism. Soc. Am.
Low-Seismicity Areas: Scaling (2015), 105(4),
and Aleatory Variability 18831902
AnalysisApplication to the
French Alps
A Bayesian methodology applied Keller M., Pasanisi A., Quality Reliability and
to the estimation of earthquake Marcilhac M., Yalamas T., Engineering Intern.
recurrence parameters for seismic Secanell R., Senfaute G. (2014), 30(7), 921933.
hazard assessment
A new, improved and fully Letort J., Guilbert J., Cotton F., Geophys. J.Int. (2015),
automatic method for teleseismic Bondar I., Cano Y., Vergoz J. 201(3), 18341848
depth estimation of moderate
earthquakes (4.5<M<5.5):
application to the Guerrero
subduction zone (Mexico).
Empirical Ground-Motion Koufoudi E., Ktenidou O.-J., Bull. of Earthq. Eng.
Models adapted to a new Cotton F., Dufour F., Grange S. (2015), 13 (12),
intensity measure (ASA40) 36253643
156 Annexes

Uncertainty in PSHA related to Mucciarelli M. Nat. Hazards Earth Syst.
the parametrization of historical Sci. (2014), 14,
intensity data 27612765.
Intensity Measures for De Biasio M., Grange S., Earthq. Eng. & Struct.
Probabilistic Assessment of Dufour F., Petre Lazar I., Dyn. (2015), 44(13),
Non-Structural Components Allain F. 22612280.
Acceleration Demand
3D numerical simulations of Chaljub E.; Maufroy E.; Geophys. J.Int. (2015)
earthquake ground motion in Moczo P.; Kristek J.; Hollender 201, 90111
sedimentary basins: testing F.; Bard P.-Y.; Priolo E.; Klin
accuracy through stringent P.; De Martin F.; Zhang Z.;
models Chen X.
Earthquake ground motion in the Maufroy E., Chaljub E., Bull. Seism. Soc. Am.
Mygdonian basin, Greece: the Hollender F., Kristek J., Moczo (2015), 105 (3),
E2VP verification and validation P., Klin P., Priolo E., Iwaki A., 13981418
of 3D numerical simulation up to Iwata T., Etienne V., De Martin
4Hz F., Theodoulidis N.P.,
Manakou M., Guyonnet-
Benaize C., Pitilakis K.,
Bard P.-Y.
Mw estimation from crustal Denieul M., Sbe O., Cara M., Bull. Seism. Soc. Am.
coda-waves recorded on analog Cansi Y. (2015), 105(2A),
seismograms 831849.
*
SiHex: Nouveau catalogue de Cara M., Cansi Y., Schlupp A., Bull. Soc. gol. France
sismicit instrumentale pour la Arroucau P., Bthoux N. etal (2015), t. 186, no 1, 319
France mtropolitaine.
Evaluation of probabilistic Faccioli E., Paolucci R., Bull. Seism. Soc. Am.
site-specific seismic hazard Vanini M. (2015), 105 (5),
methods and associated 27872807
uncertainties, with applications in
the Po plain, Northern Italy
Tectonic and geomorphic Billant etal. Tectonophysics (2015),
analyses of the Belledonne 659, 3152
border fault and its extension,
Western Alps
On the Relationship between Bora S.S., Scherbaum F., Submitted for publication
Fourier and 1 Response Spectra, Kuehn N., Stafford P. Bull. Seism. Soc. Am.
2 Implications for the Adjustment
of Empirical Ground-Motion and
3 Prediction Equations (GMPEs)
Development of a Response 1 Bora S.S., Scherbaum F., Bull. Seism. Soc. Am.
Spectral Ground-Motion Kuehn N., Stafford P., (2015), 105(4),
Prediction Equation, 2 (GMPE) Edwards B. 21922218.
for Seismic Hazard Analysis
from Empirical Fourier Spectral
and 3 Duration Models
The role of site effects in the Gallipoli M.R., Chiauzzi L., Bull. Earthquake Eng.
comparison between code Stabile T.A., Mucciarelli M., (2014), 12, 22112230
provisions and the near field Masi A., Lizza C., Vignola L.
strong motion of the Emilia 2012
earthquakes
Annexes 157

Understanding the physics of Ktenidou O.-J., Drouet S., Geophys. J.Int. (2015)
kappa and correlating it with site Cotton F., Abrahamson N.N. 203, 678691
characterisation parameters: an
application to Euroseistest.
The Finite-difference Modelling Moczo P., Kristek,J., Galis M. ISBN 978-1-107-02881-4
of Earthquake Motions: Waves Cambridge University
and Ruptures. Press
Squeezing kappa () out of the Ktenidou O.J., Silva W., Submitted for publication
Transportable Array An Darragh R., Abrahamson N.N., Bull. Seism. Soc. Am.
example using band-limited data Kishida T.
from Southern Arizona
International benchmark on Regnier J. etal Submitted for publication
numerical simulations for 1D, Bull. Seism. Soc. Am.
non-linear site response
(PRENOLIN): verification phase
based on canonical cases.
*
Semi-automated procedure for Kishida T., Ktenidou O.-J., Pacific Earthquake
windowing time series and Darragh R., Silva W. Engineering Research
computing Fourier amplitude Center, PEER report
spectra (FAS) for the NGA- 2016/02.
West2 database.
*A methodology for the Ktenidou O.-J., Abrahamson Pacific Earthquake
estimation of kappa () for large N.N., Silva W., Darragh R. Engineering Research
datasets. Example application to Center, PEER report
rock sites in the NGA-East 2016/01.
database and implications on
design motions
InterPACIFIC project: Garofalo F., Foti S., Hollender J. of Soil Dyn. and
comparison of Invasive and F., Bard P.Y., Cornou C., Cox Earthq. Eng. (2016), 82,
noninvasive methods for seismic B.R., Ohrnberger M., Sicilia 222240.
site characterization. Part I: D., Asten M., Di Giulio G.,
Intra-comparison of surface wave Forbriger T., Guillier B.,
methods Hayashi K., Martin A.,
Matsushima S., Mercerat D.,
Poggi V., Yamanaka H..
InterPACIFIC project: Garofalo F., Foti S., Hollender J. of Soil Dyn. and
comparison of Invasive and F., Bard P.Y., Cornou C., Cox Earthq. Eng. (2016), 82,
non-invasive methods for Seismic B.R., Dechamp A., Ohrnberger 241254.
site characterization. Part II: M., Perron V., Sicilia D.,
inter-comparison between Teague D., Vergniault C..
surfacewave and borehole
methods.
Simulation of the Basin Effects Dujardin A., Causse M., Pure and Applied
in the Po Plain during the Courboulex F., Traversa P. Geophysics (2016), 173,
Emilia-Romagna Seismic 19932010.
Sequence (2012) using empirical
Greens functions
*Publications using the SIGMA data production or in connexion with scientific SIGMA actions
but the research is not directly funded by SIGMA
158 Annexes
Full Papers inPeer-Reviewed Conferences
Title Authors Symposium/Workshops

Relation between seismic De Biasio M., Grange Symposium SMiRT22, USA,
ground motions and S., Dufour F., August 2013
Structural damage Petre-Lazar I., Allain F.
Nonstationary Stochastic Laurendeau A., Cotton Symposium 15 WCEE, Lisboa,
Simulation of Strong F., Bonilla F. 2012
Ground-Motion Time
Histories: Application to the
Japanese Database
Validation of site-effect Maufroy E., Bard P.-Y., Symposium 15 WCEE, Lisboa,
numerical predictions: New Chaljub E., Hollender 2012
results from the Euroseistest F., Kristek J., Moczo
Verification and Validation P., Theodoulidis N.
Project (E2VP)
Topographic effects in strong Rai M., Rodriguez- Symposium 15 WCEE, Lisboa,
ground motion Marek A., Yong A. 2012
A methodology for Bosco B., Thiry J.M., Symposium 15 WCEE, Lisboa,
measuring the quality of Rota M., Penna A., 2012
probabilistic seismic hazard Magenes G., Senfaute
predictions in the SIGMA G., Martin C.
project framework
Correlation between seismic De Biasio M., Grange Symposium: Second European
ground motions and floor S., Dufour F., Conference on Earthquake
acceleration demand for Petre-Lazar I., Allain F. engineering and Seismology. Istanbul,
frame and structural-wall 2014
buildings
Engineering application of Van Houtte C., New Zealand Society for Earthquake
high frequency ground Ktenidou O.-J., Larkin engineering Technical Conference and
motions in Christchurch T., Holden C. AGM, Wellington, 2628 Apr., 2013.
0 estimates for rock stations Ktenidou, O.-J., 10th US National Conference on
in the NGA-West2 project Kishida T., Darragh R., Earthquake engineering, Anchorage,
Silva W., Abrahamson 2125 July, 2014
N.N.
Annexes 159
PhD Theses
PhD student Title Institution Supervisors

Alain Seismic Ground Motion Geoazur (CNRS, Courboulex F.
Dujardin Prediction: site-specific hybrid Universit de Nice (Geoazur) Causse
approach Sophia Antipolis) M. (ISTerre)
Aurore Sub-set of near-source ISTerre Cotton F. (ISTerre)
Laurendeau earthquake record Hollender F. (CEA)
Hilal Oksuz Testing probabilistic seismic ISTerre Beauval C.
hazard estimations against (ISTerre) Guguen
observations, application to P. (ISTerre)
France
Jean Letort Characterization of moderate ISTerre Cotton F. (ISTerre)
earthquake depths using
teleseismic data.
Jrmy Billant Study of a mega-structure: the CEREGE (Universit Bellier O.
Southwestern alpine front Aix-Marseille) (CEREGE)
Belledonne fault Hippolyte J.C.
(CEREGE)
Marylin Seismic Moment Magnitude EOST (Institut de Cara M. (EOST)
Denieul and Crustal Wave Coda Physique du Globe de Sbe O. (CEA)
Strasbourg)
Thibault tude de la vulnrabilit des Universit Quenet G. (UVSQ)
Fradet populations anciennes aux Observatoire de
tremblements de terre en Versailles Saint
France 16501850 Quentin en Yvelines
Jessie Mayor Regional variation of seismic IRAP (CNRS, Margerin L. (IRAP)
attenuation in Metropolitan Universit Toulouse Calvet M. (IRAP)
France, Observation and III, Paul Sabatier)
Modeling
Marco De Ground Motion Intensity 3SR Grenoble Grange S. (3SR
Biasio Measures for Seismic Grenoble) Calvet
Probabilistic Risk Analysis M. (IRAP)
Haitham Improving the understanding Virginia Tech, USA Rodriguez Marerk
Dawood and methodologies of ground A. (Virginia Tech,
motion variability USA)
160 Annexes
List ofDeliverables
D2-15 RESORCE: Seismic Motion Databank: 2011 S.Akkar (METU),

Version Including Improvement of Meta-Parameters F.Cotton (GFZ),
S.Godey (CSEM),
M.Senyurt, P.Traversa,
(EDF)
D2-16 Pre Selection of Ground Motion Prediction S.Drouet, (Geoter)
Equations for France- S.
D2-17 Methodology and Testing of GMPEs Based on E.Delavaud, (ETH
Ground Motion Data from France and Surrounding Zurich)
Countries
D4-18 Final Preliminary Probabilistic Hazard Map for D.Carbon, C.Gomes,
Frances Southeast Ch. Martin, (Geoter)
D4-19 Methods for Testing PSHA Against Observations H.Oksuz, (ISTerre)
(Accelerometer Data)
D1-27 Review of Active Faulting in the Po Plain P.Burrato, F.Maesano,
G.Valensise, (INGV)
D4-29 Preliminary PSHAs at Selected Sites, based on E.Faccioli, (Polimi)
State of the Art Earthquake Source Models and
Attenuation Relationships, Considering Both
Outcropping Bedrock and Soil Conditions
D2-33 Development of Regional Ground Motion S.Drouet, (Geoter)
Prediction Equations Covering a Wide Magnitude
Range Based on Stochastic Simulations
(Application to the Alps, Pyrenees and Rhine
Graben Regions in France)
D2-34 The Use of Ground Motion Data and Macroseismic E.Delavaud, (Geoter)
Intensities to Evaluate the Applicability of GMPEs
to France
D4-35 Methodological Approach for Measuring the J.M.Thiry, B.Bosco,
Quality of Probabilistic Seismic Hazard Predictions (Areva)
D3-36 Toward the Definition of Reference Ground-Motion L.Laurendeau, (Isterre)
D3-38 Evaluation of The Relevance of the Numerical E.Maufroy, (Isterre)
Simulation Approach for Site Effect Estimation
D3-39 Inventory of Approaches to Account for Site C.Lacave, M.Koller,
Effects: Methodologies and Regulations (Resonance Ing. Sa)
D4-41 Initial Probabilistic Seismic Hazard Model for D.Carbon, S.Drouet,
Frances Southeast 1/4 C.Gomes, A.Leon, Ch.
Martin, R.Secanel,
(Geoter)
D1-49 Seismic Moment Magnitude and Crustal Wave M.Denieul, (Eost)
Coda
D1-50 Characterization of Earthquake Depth Using J.Letort, (Isterre)
Teleseismic Data Developing New Methods for
Moderate Events and Application to the Po Plain
Earthquakes
Annexes 161
D1-51 Preliminary Earthquake Recurrence Models with M.Mucciarelli,

Extensive Use of Site Data (Universit della
Basilicata)
D5-52 Relation Between Seismic Ground Motion and M.De Biasio, (3SR
Structural Damage and Function-Loss Part 1: Grenoble)
Structural Wall Buildings Preliminary Results
D2-53 Ranking of Available GMPEs from Residual F.Pacor, (INGV Milano)
Analysis for Northern Italy and Definition of
Reference GMPEs
D3-54 Evaluations of Seismic Ground Motion Variability R.Paolucci, (Polit.
at Soft Sites by 3d-1d Propagations Models, Milano)
Including Christchurch and Selected Sites in The Po
Plain
D1-66 Morphostructural Analysis of the Belledone Fault J.Billant, (Cerege)
System (Western Alps)
D1-67 Preliminary Earthquake Source Models of the Po G.Valensise, (INGV
Plain, Based on Full 3d Definition of Active Faults Roma)
D1-68 Methodological Approach in Analysing Historical T.Fradet, (Universit St
Sources: A Way to Improve Vulnerability Studies Quentin)
D2-71 Stochastic GMPEs for France S.Drouet, (Drouet
Consulting)
D2-72 Calibration of GMPEs for Po Plain Region F.Pacor, (INGV Milano)
D2-73 On the Magnitude Dependency of Ground Motion A.Dujardin,
Decay with Distance (Geosciences Azur)
D3-74 Preliminary Study on Empirical Approaches for Site M.Mucciarelli,
Effect Characterization (Including in Situ (Universit Della
Measurements) Basilicata)
D4-75 Development of Methods for Testing Probabilistic H.Oksuz, (Isterre)
Seismic Hazard Estimates against Observations
D1-89 Crustal Tomography 3D vs. 1D Localisation in T.Theunissen,
the Pyrenees S.Cheverot,
M.Sylvander, (OMP)
D2-91 Databank of Seismic Motion for Europe S.Akkar (METU),
RESORCE, 2013 Version: Improvement of S.Godey (CSEM),
Metaparameters/Inclusion of Weak Motion Data/ J.Douglas (BRGM),
Internet Portal/Perspectives and Related European L.Luzi (INGV),
Initiatives P.Traversa, (EDF)
D2-92 Metadata Uncertainties and Sensitivity of Empirical G.Ameri, (Geoter)
Ground Motion Prediction Equations Adapted to the
French Context
D2-93 Ground Shaking Scenarios in the Po Plain with R.Paolucci, (Polit.
Special Emphasis on the Area Affected by the Milano)
Earthquake Sequence of May 2012
D4-94 Site Specific Probabilistic Study for Po Plain Area E.Faccioli, M.Vanini,
R.Paolucci, (Polimi)
D3-96 Approaches to Account for Sites Effects in the R.Paolucci, E.Faccioli,
PSHA of Selected Sites in The Po Area C.Smerzini, M.Vanini,
(Polimi)
162 Annexes
D3-97 Characterization of Classes of Sites with a Large P.Moczo, (University of

Potential to Cause Site-Effects Taking into Account Bratislava)
the Geological Heterogeneities (Methodological
Approach)
D1-109 Revision of the Po Plain Earthquake Catalogue in M.Mucciarelli,
Probabilistic Terms (University of Potenza)
D1-110 Seismic Moment Magnitude and Crustal Coda M.Denieul, M.Cara,
Waves (Eost)
D2-111 Improving the Understanding and Methodologies of A.Rodriguez-Marek,
Ground Motion Variability (Virginia Tech)
D2-112 Kappa () Physical Understanding and Estimation O.Ktenidou, (Isterre)
Approaches Targeted to Applications
D2-113 Spatial Variations of Seismic Attenuation in J.Mayor, (Irap)
Metropolitan France from Observation and
Modelling of the Seismic Coda
D3-114 Prediction of Non-Linearity Using Numerical J.Regnier, (Cerema)
Simulation: Results of the Verification Phase of the
PRENOLIN Program
D3-115 Characterizing Site Metadata of Accelerometric A.Dechamp, (Cea)
Network Stations: Results and Methodological
Feedback from a RAP Station Survey
D3-116 Site Instrumentation Usefulness: Example of V.Perron, (Cea)
Application of Empirical Site Effect Estimation in
Low Seismicity Context and Implementation
Recommendations
D4-118 Estimation of Seismic Risks Using Structural M.Rota, (Eucentre)
Fragility Functions and Macroseismic Intensity
Data
D5-119 Relation Between Seismic Ground Motion and M.Di Biasio, (3SR
Structural Damages & Function-Loss Grenoble)
D1-127 Seismotectonic and Geodynamic Synthesis of SE X.Sengelen, O.Bellier
France (Cerege), L Bollinger
(Cea), K.Manchuel,
(EDF)
D1-128 Study of the Relation Isoseismal Areas vs. A.Schlupp,
Depth-Magnitude, Based on Instrumental Data M.Durrenberger, C.Sira,
(EOST)
D1-129 Study of Instrumented Earthquakes that Occurred J.Benjumea, M.Cara,
during the First Part of the 20th Century L.Rivera, (EOST)
(19051962)
D2-130 Development of Response Spectral Ground-Motion S.Sigh Bora, (GFZ)
Prediction Equation (GMPE) from Empirical
Fourier Spectral and Duration Models
D2-131 Empirical Ground Motion Model Adapted to the G.Ameri, (Geoter)
French Context
D2-132 Ground Motion Uncertainty and Variability in O.Ktenidou, (Isterre)
Euroseistest, Greece
D2-133 Ground Motion Variability in the Po Plain Region F.Pacor, (INGV Milano)
Annexes 163
D3-134 Intercomparison of Methods for Site Parameter and S.Foti, F.Garofalo

Velocity Profile Characterization (INTERPACIFIC (Polito)
Project): Results and Learnings from the Blind-Test
Stage and Suggestions for a Good Practice Guide
for Non-Invasive Methods
D3-137 3D Ground Motions Simulations for Site Effects E.Maufroi, (Isterre)
Assessment: Learnings from Euroseistest
Verification and Validation Project
D4-138 Integration of Sigma Improvement for PSHA and G.Ameri, (Geoter)
Sensitivity Studies (Intermediate Results)
D4-139 Methodological Exercise for Testing a Statistical R.Secanell, C.Martin,
Approach to Update a Probabilistic Seismic Hazard (Geoter)
Assessment
D4-140 Comparative Study on the PSHA Estimates Using M.Pagani (GEM
Crisis and Openquake Foundation Advisor),
L.Rodriguez (UME
School Pavia)
D5-141 Application of the Conditional Spectra (CS) M.Marcilhac,
Method K.Macocco, (Phimeca)
D1-147 Relation of Intensity Attenuation and D.Baumont, (Fugro),
Methodologies to Determine Seismological J.Bonnet, (UNS Nice
Parameters of Historic Earthquakes. University)
D1-148 Catalogue of Isoseismal Areas for XXth Century J.Lambert, D.Monfort-
French Historical Earthquakes (Io>VI) Climent, O.Bouc,
(BRGM)
V2A2T3-1 PRENOLIN: Results from the Verification and J.Rgnier, (Cerema),
(D3-149) Validation Phases & Present Recommendations for F.Bonilla (IFSTTAR)
the Use of Non-Linear Simulation
V1A1T3-1 Correction of Surface Records of their Site Effect F.Hollender, (CEA),
(D3-150) before Developing GMPE: An Alternative Approach L.Foundotos,
to Get Reference Incident Ground Motion A.Laurendeau, (CEA),
(Application to Kik-Net Data) PY.Bard, (Isterre)
D3-151 Identification of Key Site Features for Site Effect P.Moczo, (Bratislava
Evaluation: Extensive Numerical Sensitivity Studies University)
of Sedimentary Basin Structures
D3-152 Good Practices Guidelines for Site Response
Evaluation
D4-153 SIGMA Integrated Exercise for Site-Specific PSHA G.Ameri, (Geoter)
D4-154 Estimation of Seismic Risks Using Structural A.Rosti, M.Rota,
Fragility Functions and Macroseismic Intensity E.Fiorini, A.Penna,
Data P.Bazzurro, G.Magenes,
(Eucentre)
D1-168 Calibration of Macroseismic Attenuation Models K.Manchuel, P.Traversa,
for France (EDF), D.Baumont
(Geoter)
D3-169 Good Practices Guidelines for Site Characterization
D4-170 Probabilistic Seismic Hazard Assessment for C.Martin, G.Ameri,
South-Eastern France (Geoter) D.Baumont
(Fugro)
Bibliography
Akkar S,Sandikkaya MA, Bommer J (2014a) Empirical ground-motion models for point- and
extended-source crustal earthquake scenarios in Europe and the Middle East Bull Earthq Eng
12(1):359387
Akkar S,Sandikkaya MA, Senyurt M, Azari Sisi A,Ay B, Traversa P,Douglas J, Cotton F, Luzi
L,Hernandez B, Godey S (2014b) Reference database for seismic ground-motion in Europe
(RESORCE) Bull Earthq Eng 12(1):311339
Al Atik L, Kottke A, Abrahamson N, Hollenback J(2014) Kappa () scaling of ground-motion
prediction equations using an inverse random vibration theory approach. Bull Seismol Soc Am
104(1):336346
American Society of Civil Engineers ASCE SEI 43-05 (2005) Seismic design criteria for struc-
tures, systems and components in nuclear facilities. American Society of Civil Engineers,
NewYork
American Society of Civil Engineers ASCE SEI 7/10 (2010) Minimum design loads for buildings
and other structures. American Society of Civil Engineers, NewYork
Anderson J, Hough S (1984) A model for the shape of the Fourier amplitude spectrum of accelera-
tion at high frequencies. Bull Seismol Soc Am 74:19691993
Assatourians K, Atkinson G (2010) Verification of engineering seismology toolbox processed
accelerograms: 2005 Riviere du Loup, Quebec earthquake. Available at www.seismotoolbox.ca
ATC (1978) Tentative provisions for the development of seismic regulations for buildings, ATC-
3-06 Report. Applied Technology Council, Redwood City
Atkinson G, Boore D (2006) Earthquake ground-motion prediction equations for eastern North
America. Bull Seismol Soc Am 96(6):21812205
Baker JW (2011) Conditional mean spectrum: tool for ground motion selection. J Struct Eng
137(3):322331
Baker JW, Cornell CA (2006) Spectral shape, epsilon and record selection. Earthq Eng Struct Dyn
35(9):10771095
Bardet JP, Ichii K, Lin CH (2000) EERAA computer program for equivalent-linear earthquake
site response analyses of layered soil deposits. University of Southern California, Department
of Civil Engineering, Los Angeles
Basili R, Valensise G, Vannoli P, Burrato P, Fracassi U, Mariano S, Tiberti MM, Boschi E (2008)
The database of individual seismogenic sources (DISS), version 3: summarizing 20 years of
research on Italys earthquake geology. Tectonoph 453:2043, database open for consultation
at http://diss.rm.ingv.it/diss/
Bindi D, Pacor F, Luzi L, Puglia R, Massa M, Ameri G, Paolucci R (2011) Ground motion predic-
tion equations derived from the Italian strong motion database. Bull Earthq Eng
9(6):18991920

166 Bibliography
Bindi D, Massa M, Luzi L, Ameri G, Pacor F, Puglia R, Augliera P (2014) Pan-European ground-
motion prediction equations for the average horizontal component of PGA, PGV, and
5%-damped PSA at spectral periods up to 3.0 s using the RESORCE dataset. Bull Earthq Eng
12(1):391430
Biro Y, Renault P (2012) Importance and impact of host-to-target conversions for ground motion
prediction equations in PSHA.In: Proceedings of 15 world conference on earthquake engineer-
ing, Lisboa, Portugal
Bojorquez E, Iervolino I (2011) Spectral shape proxies and nonlinear structural response. Soil Dyn
Earthq Eng 31:9961008
Bommer JJ, Akkar S (2012) Consistent source-to-site distance metrics in ground-motion predic-
tion equations and seismic source models for PSHA.Earthq Spectra 28:115
Bommer JJ, Douglas J, Scherbaum F, Cotton F, Bungum H, Fh D (2010) On the selection of
ground-motion prediction equations for seismic hazard analysis. Seismol Res Lett
81(5):783793
Bommer JJ, Akkar S, Kale (2011) A model for vertical-to-horizontal response spectral ratios for
Europe and the Middle East. Bull Seismol Soc Am 101(4):17831806
Bommer JJ, Strasser FO, Pagani M, Monelli D (2013) Quality assurance for logic-tree implemen-
tation in probabilistic seismic-hazard analysis for nuclear applications: a practical example.
Seismol Res Lett 84((6)):601, 938945
Boore D (2003) Simulation of ground motion using the stochastic method. Pure Appl Geophysics
160:635675
Boore D, Joyner W (1997) Site amplifications for generic rock sites. Bull Seismol Soc Am
87(2):327341
Campbell K (2003) Prediction of strong ground motion using the hybrid empirical method and its
use in the development of ground-motion (attenuation) relations in Eastern North America.
Carlton B, Abrahamson N (2014) Issues and approaches for implementing conditional mean spec-
tra in practice. Bull Seismol Soc Am 104(1):503512
Cauzzi C, Faccioli E (2008) Broadband (0.05 to 20 s) prediction of displacement response spectra
based on worldwide digital records (2008). JSeismol 12(4):453475
Cauzzi C, Dalguer L, Baumann C, Giardini D (2015a) Anatomy of near-field ground-shaking
generated by dynamic rupture simulations. In: Conference present at SSA2015, Pasadena, CA,
USA.Seismol Res Lett 862(2B):601
Cauzzi C, Faccioli E, Vanini M, Bianchini A (2015b) Updated predictive equations for broadband
(0.0110 s) horizontal response spectra and peak ground motions, based on a global dataset of
digital acceleration records. Bull Earthq Eng 13(6):15871612
Comit Europen de Normalisation CEN (2004) Eurocode 8: design of structures for earthquake
resistancePart 1: General rules, seismic actions and rules for buildings. Comit Europen de
Normalisation CEN, Brussels
Cotton F, Scherbaum F, Bommer JJ, Bungum H (2006) Criteria for selecting and adjusting ground-
motion models for specific target applications: applications to Central Europe and rock sites.
JSeismol 10(2):137156
CPTI Working Group (2004) Catalogo parametrico dei terremoti italiani. Istituto Nazionale di
Geofisica e Vulcanologia (INGV), Bologna, Italy. http://emidius.mi.ingv.it/CPTI04
Danciu L, Monelli D, Pagani M, Wiemer S (2010) GEM1 hazard: review of PSHA software, GEM
Technical Report 2010-2. GEM Foundation, Pavia
De Biasio M, Grange S, Dufour F, Allain F, Petre-Lazar I (2014) A simple and efficient intensity
measure to account for non-linear structural behavior. Earthq Spectra 30(4):14031426
De Biasio M, Grange S, Dufour F, Allain F, Petre LI (2015) Intensity measures for probabilistic
assessment of non-structural components acceleration demand. Earthq Eng Struct Dyn
44(13):22612280
Delavaud E, Cotton F, Akkar S, Scherbaum F etal (2012) Toward a ground-motion logic tree for
probabilistic seismic hazard assessment in Europe. JSeismol 16(3):451473
Bibliography 167
DISS Working Group (2010) Database of Individual Seismogenic Sources (DISS), Version 3.1.01:
a compilation of potential sources for earthquakes larger than M 5.5in Italy and surrounding
areas. http://diss.rm.ingv.it/diss/, INGV 20092010 -Istituto Nazionale di Geofisica e
Vulcanologia
Douglas J (2011) Ground-motion prediction equations 19642010. www.gmpe.org.uk
EPRI (Electric Power Research Institute) (1988) A criterion for determining exceedance of the
operating basis earthquake, EPRI NP-5930. Electric Power Research Institute, Palo Alto
EPRI, (Electric Power Research Institute) (1991) Standardization of the cumulative absolute
velocity, EPRI TR-100082. Electric Power Research Institute, Palo Alto
Gerolymos N, Gazetas G (2005) Constitutive model for 1-D cyclic soil behaviour applied to seis-
mic analysis of layered deposits. Soils Found 45:147159
Haselton C, Baker JW, Liel AB, Deierlein GG (2011) Accounting for ground motion spectral shape
characteristics in structural collapse assessment through an adjustment for epsilon. JStruct Eng
137(3):332344
Hashash YMA, Musgrove MI, Harmon JA, Groholski D, Phillips CA, Park D (2015) DEEPSOIL
V6.0. University of Illinois, Urbana
Idriss IM, Sun JI (1992) SHAKE91: A computer program for conducting equivalent linear seismic
response analyses of horizontally layered soil deposits. Program modified based on the original
SHAKE program published in December 1972 by Schnabel, Lysmer and Seed. Davis, Center
of Geotechnical Modeling, Department of Civil Engineering, University of California
ITASCA.Consulting Group. FLAC, Explicit continuum modelling of nonlinear material behav-
iour. ITASCA, Minneapolis. http://www.itascacg.com/
Jayaram N, Lin T, Baker JW (2011) A computationally efficient ground-motion selection algo-
rithm for matching a target response spectrum mean and variance. Earthq Spectra
27(3):797815
Johnston AC, Coppersmith KJ, Kanter LR, Cornell CA (1994) The earthquakes of stable continen-
tal regions assessment of large earthquake potential, Electric Power Research Institute
(EPRI), TR-102261-V1, 21-98, Palo Alto, CA
Kottke A, Rathje E (2008) Technical manual for Strata, Report 2008/10. Pacific Earthquake
Engineering Research (PEER) Center, Berkeley
Koufoudi E, Ktenidou OJ, Cotton F, Dufour F, Grange S (2015) Empirical ground-motion models
adapted to a new intensity measure ASA40 (2015). Bull Earthq Eng 13(12):36253643
Kramer SL, Paulsen SB (2004) Practical use of geotechnical site response models. In: Proceedings
of international workshop on uncertainties in nonlinear soil properties and their impact on
modeling dynamic soil response. PEER Center Headquarters, Richmond
Ktenidou O, Glis C, Bonilla L-F (2013) A study on the variability of kappa () in a borehole:
implications of the computation process. Bull Seismol Soc Am 103(2A):10481068
Li XS, Wang ZL, Shen CK (1992) SUMDES: a nonlinear procedure for response analysis of
horizontally-layered sites subjected to multi-directional earthquake loading. Department of
Civil Engineering, University of California, Davis
Matasovi N, Ordez G (2007) D-MOD2000, Geomotins, LLC, Computer software
Modaressi H, Foerster E (2000) CyberQuake: modelling soil responses to seismic tremors. BRGM.
http://www.brgm.eu/scientific-output/scientific-software/scientific-software
Nagashima F, Matsushima S, Kawase H, Sanchez-Sesma FJ, Hayakawa T, Satoh T, Oshima M
(2014) Application of horizontal-to-vertical spectral ratios of earthquake ground motions to
identify subsurface structures at and around the K-NET site in Tohoku, Japan. Bull Seismol
Soc Am 104(5):22882302
Poggi V, Edwards B, Fh D (2012) Characterizing the vertical-to-horizontal ratio of ground motion
at soft-sediment sites. Bull Seismol Soc Am 102(6):27412756
Prevost JH (2010) DYNAFLOW: a nonlinear transient finite element analysis program. Princeton
University, Princeton
Rodriguez-Marek A, Montalva GA, Cotton F, Bonilla F (2011) Analysis of single-station standard
deviation using the KiK-net data. Bull Seismol Soc Am 101(3):12421258
168 Bibliography
Secanell R, Martin C, Viallet E, Senfaute G (2015) A Bayesian methodology to update the proba-
bilistic seismic hazard assessment. In: CSNI workshop on testing PSHA results and benefit of
Bayesian techniques for seismic hazard assessment, 46 February 2015, Eucentre Foundation,
Pavia, Italy
Shahbazian A, Pezeshk S (2010) Improved velocity and displacement time histories in frequency-
domain spectral-matching procedures. Bull Seismol Soc Am 100(6):32133223
Smerzini C, Galasso C, Iervolino I, Paolucci R (2014) Ground motion record selection based on
broadband spectral compatibility. Earthq Spectra 30(4):14271448
Stewart J, On-Lei KA, Hashash YMA, Matasovic N, Pyke R, Wang Z, Yang Z (2008) Benchmarking
of nonlinear geotechnical ground response analysis procedures, PEER Report 2008/04. Pacific
Earthquake Engineering Research Center, Richmond
Thomas P, Wong I, Abrahamson N (2010) Verification of probabilistic seismic hazard analysis
computer programs, Pacific Earthquake Engineering Research Center; PEER 2010/106. Pacific
Earthquake Engineering Research Center, Berkeley
Van Houtte C, Drouet S, Cotton F (2011) Analysis of the origins of (kappa) to compute hard rock
to rock adjustment factors for GMPEs. Bull Seismol Soc Am 101(6):29262941
Von Thun JL, Rochim LH, Scott GA, Wilson JA (1988) Earthquake ground motions for design and
analysis of dams. In: Earthquake engineering and soil dynamices II Recent advances in
ground-motion evaluation (GSP 20). ASCE, NewYork, pp463481
Zhao J, Zhang J, Asano A, Ohno Y, Oouchi T, Takahashi T, Ogawa H, Irikura K, Thio HK,
Somerville PG, Fukushima Y (2006) Attenuation relations of strong ground motion in Japan
using site classification based on predominant period. Bull Seismol Soc Am 96(3):898913
Index
A Depth distribution, 135

Acceleration time histories, 102 Disaggregation, 11, 12, 21
scaled natural records, 102 conditional spectra
spectrally-matched, 102 scenarios, 12
Adjustments, 63 Dispersion curve, 87
host-to-target, 66 Distance, 63
Amplification factor, 15 conversion, 134
Annual probability of exceedance, 9, 20 metrics, 63
Attenuation parameters, 62 Down-hole, 61
attenuation equations, 63
high-frequency, 62
E
Earthquake catalogue, 133
B Equivalent linear analyses, 104
Bayesian approach, 145 Casaglia site, 104
Bayesian framework to modify the logic tree EuroseisTest, 104
weights, 145 Grenoble, 104
Ergodic, 73
European Macroseismic Scale (EMS), 133
C Experts, 13, 1617
CAV. See Cumulative Absolute Velocity evaluator, 17
(CAV) expert elicitation, 16, 17
Coefficients of variation, 87 expert judgement, 17
Consistency, 63 expert opinion, 17
site-conditions, 63 informed technical community, 17
tectonic, 63 proponent, 17
Control point, 18, 19 resource expert, 17
outcropping motion, 19 technical defensible interpretation, 17
Correction, 66
host-to-target, 66
Cumulative Absolute Velocity (CAV), 22 F
Fault, 64
mechanisms, 64
D normal, 64
Database, 59 reverse, 64
European, 59 strike-slip, 64

170 Index
Finite fault simulations, 134 Logic Tree, 1217, 70

Focal depth, 135 branch, 13
Fractile, 15, 20 collectively exhaustive, 14
Fragility, 18 mutually exclusive, 14
Frequency magnitude distribution, 44 node, 13
G M
GMPE. See Ground motion prediction Magnitude conversion, 133
equation (GMPE) Maximum ground motion, 117, 136
Greens functions, 91 Maximum magnitude, 8, 42
Ground motion characterization, 7, 135 Method to compare PSHA results with
Ground motion component, 57 historical information, 143
horizontal, 57 Minimum magnitude, 10, 135
vertical, 7879 Models, 57
Ground motion prediction equation (GMPE), area source, 59
9, 57, 93 point source, 62
median amplitude, 9 stochastic, 57
median prediction, 97
nonlinear term, 96
proxy, 96 N
site-specific correction factors (S2S), 97 Non-linear, 137
standard deviations, 96 Nonlinear Numerical Analyses,
Ground motion prediction model, 7 107111
Ground motions, 62 Casaglia, 109
duration, 22, 70 epistemic uncertainties, 107
estimates, 61 EuroseisTest, 109
synthetic, 62 Grenoble, 109
Gutenberg-Richter, 9 Japan, 108
numerical testing of the model, 108
Prenolin international benchmark, 107
H Nonlinear soil properties, 92
Hazard curve, 8, 11, 20 equivalent damping ratio, 92
Host-to-target, 137 Prenolin benchmark, 93
conversions, 136 reference shear strain, 93
Kappa, 137 secant shear modulus, 92
shear behaviour, 93
volumetric behaviour, 93
I
Incident wave field, 102
Interfaces, 18, 133138 O
Occurrence Processes, 4142
characteristic model, 42
J Poisson model, 41
Japan, 71
KiK-net, 71
P
Parameter, 61
L site, 67
Lessons learned, 26, 27 stress, 61
management of the interfaces Peak ground acceleration (PGA), 9, 136
interfaces between the various Peak ground velocity (PGV), 18, 22
disciplines, 27 Probabilistic Seismic Hazard Assessment
Linear Numerical Analyses, 103104 (PSHA), 5, 812
Grenoble test-site, 103 Probability of exceedance, 15
Index 171
Profiles natural frequency, 86 active shallow crustal region, 6

PSHA at a rock site, 95 area sources, 32
database, 28
GIS database, 29
Q depth distribution, 18, 133
Quality factor, 67, 103 dip angle, 18
distance metric, 133
fault sources, 32
R composite seismogenic fault
Rate of occurrence, 8, 11 sources, 36
Recurrence, 8 focal depth, 18
Reference bedrock, 136 management of the interfaces, 25
Reference rock, 14 magnitude-area scaling, 18
Regional Study, 67 QA verification and validation
Residual, 73 procedures, 53
between-event, 74 seismic sources, 5
prediction, 61 spatial smoothing, 6
within-event, 79 seismotectonic, 5
Results, 1822 smoothed seismicity approach, 38
design, 18, 19 source geometry, 8
probabilistic safety assessments, 1820 stable continental region, 5
soil-structure interaction, 19 style of faulting, 18
verification, 18 subduction region, 6
Rock, 57 Sensitivity, 14
hard, 58 Sensitivity analyses, 27
sites, 59 pilot model, 50
very hard, 61 Shear-wave velocity, 85
weathered, 71 Sigma, 64
Rock hazard, 93 truncation, 7378
Rupture dimension, 134 Single-station, 69
Single-station sigma, 7, 16, 97,
137138
S Site amplification functions, 93
Seismic hazard computation process, experimental, 98
119, 126 Japanese KiK-net data, 98
acceleration time histories, 128 norms, 98
alternative engineering output Site Amplification Prediction Equation,
parameters, 127 SAPE, 100
basic requirements applying to hazard site class, 98
computation, 119 Site characterization, 85
Conditional Spectra, 129 Site class, 89
disaggregation of seismic hazard, 126 Site conditions, 58
Magnitude-distance-epsilon hard-rock, 58
disaggregation, 126 Site investigation, 85, 86
PSHA software, 120 invasive tests, 85
sensitivity analysis, 123 crosshole, 85
Site Response Analysis Codes, 122123 down-hole, 85
V&V documentation, 120 Suspension Logging, 85
Seismic instrumentation, 91 guidelines, 90
accelerometers, 91 InterPACIFIC benchmark, 89
empirical amplification functions, 91 non-invasive, 85
signal-to-noise ratio, 91 Ambient Vibration H/V
velocity meters, 91 Measurements, 86
Seismic sources characterization (SSC), SESAME, 86
6, 29, 36 spectral analysis of surface waves, 85
172 Index
Site response, 1416 epistemic uncertainties in the SSC

damping, 14 models, 33
hybrid approaches (HyS, HyG), 95 2D or 3D amplification, 101
fully probabilistic approaches aggravation factor, 101
(FpS, FpG), 95 EuroseisTest, 101
impedance contrast, 14 NERA, 101
non-linearity, 14
resonance, 14
shear-wave velocity, 18 U
soil, 14 UHS. See Uniform Hazard Spectra (UHS)
soil amplification, 15 Uncertainties, 6, 1217, 20, 57, 136, 138
Site response analysis, 85 aleatory variability, 1213, 57, 112
Site specific, 67 double counting, 16, 112, 136
Site term, 73 epistemic uncertainty, 1213, 57
Site to reference, 91 ergodicity, 111
Soil constitutive models, 101 event corrected single-station standard
equivalent linear, 101 deviation, 111
fully nonlinear, 101 inter-event variability, 111
linear, 101 site-to-site variability S2S, 111
linear viscoelastic, 102 standard deviation, 9, 15
Soil response characterization (SRC). total single-station standard
See Site response deviation, 111
Spatial variability, 90 Uniform Hazard Spectra (UHS), 21
Spectra, 67 epistemic uncertainty, 1213
Fourier, 67
response, 68
Spectral acceleration, 9, 57 V
median response, 57 V/H Models, 138
standard deviation, 57 Variability, 57
SSC logic tree, 47 aleatory, 57
Bayesian framework to modify the logic record-to-record, 74
tree weights, 52 Velocity, 59
Style-of-faulting, 62, 134 profiles, 62
shear-wave, 59
Vertical motions, 115
T 1D site response analyses, 116
Testing probabilistic seismic hazard models physics, 116
against observations, 142 V/H relationships, 115
PSHA testing using recorded
accelerometer data, 142
Transfer function, 62 W
Treatment of the uncertainties, 27 Weight, 8, 14, 15

10.1007/978 3 319 58154 5

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

10.1007/978 3 319 58154 5

Caricato da

Copyright:

Formati disponibili

Geotechnical, Geological and Earthquake Engineering

Editorial Advisory Board

An Overview of the SIGMA

ISSN 1573-6059 ISSN 1872-4671(electronic)

Library of Congress Control Number: 2017940276

Springer International Publishing AG 2017

Printed on acid-free paper

This Springer imprint is published by Springer Nature

3.3 Seismic Source Models....................................................................... 32

5.2 Hazard Assessment at theGround Surface......................................... 93

9 Summary andWay Forward.................................................................... 147

ASCR Active shallow crustal region

1.1 Overview oftheProject Organisation

The main objective of the research programme was to establish a framework to be

Springer International Publishing AG 2017 1

Produce accepted and validated methods and calculation tools;

Fig. 1.2 Management flowchart

As mentioned previously, a large number (30) of worldwide academic, research

1.2 Object oftheDocument

2.1 Development ofaSeismotectonic Framework forPSHA

Springer International Publishing AG 2017 5

Active shallow crustal region (ASCR); and

2.3 Site Specific vs. Regional Study

State-of-the-art seismic hazard studies calculate ground motion exceedance proba-

can produce. Usually, magnitude is characterised in the moment magnitude

Haz ( a * ) = vi GA|m , r ( a*;m;r ) f M ( i ) ( m ) f R ( i )| M ( i ) ( r;m ) dm dr (2.1)

i and faults, respectively. The integration on magnitude in the equation considers

2.5 Logic Tree Approach andTreatment ofUncertainties

2.5.1 Epistemic Uncertainty vs. Aleatory Variability

2.5.2 Logic Tree Methodology

assumed to be a mutually exclusive and collectively exhaustive set of alternative

2.5.3 Site Response

2.5.3.1 Methods forImplementing Site Response Effects

2.5.4 Use ofExperts

2.5.4.1 Guidance Regarding Expert Elicitations

Comprehensive guidance on processes to be followed for expert elicitations has

2.6 Interface Issues Between Work Packages

A PSHA is fundamentally a multidisciplinary study that involves various experts

2.7 Common Required Outputs forSeismic Hazard Results

2.7.1 Basic Definitions andRequirements

2.7.1.1 Control Point

2.7.1.2 Components ofMotion

2.7.1.3 Vibration Frequencies forHazard Analysis

2.7.1.4 Lowest Annual Probability ofExceedance

2.7.2 Common Hazard Results

2.7.2.1 Seismic Hazard Curves forReference Rock Site Condition

2.7.2.2 Seismic Hazard Curves fortheSoil Site Condition

2.7.2.3 Epistemic Uncertainty intheSoil Hazard Curves

2.7.2.4 Uniform Hazard Spectra

2.7.3 Additional Parameters

2.7.3.1 Ground Motion Duration

Ground-motion durations need to be defined if different earthquake scenarios have

2.7.3.2 Peak Ground Velocity

2.7.3.3 Cumulative Absolute Velocity

Seismic sources characterization (SSC) relies on the interpretation and integration

Springer International Publishing AG 2017 25

geophysics, numerical simulation, statistics, soil dynamics, and mathematics).

3.1 Pre-requisites toDevelop theSSC Models

3.3 Seismic Source Models

3.3.2 Seismic Source Characterization Framework

3.3.3 Area Source, Fault Sources, Gridded Seismicity

3.3.3.1 Area Sources

3.3.3.2 Fault Sources

3.3 Seismic Source Models....................................................................... 32

5.2 Hazard Assessment at theGround Surface......................................... 93

9 Summary andWay Forward.................................................................... 147

1.1 Overview oftheProject Organisation

1.2 Object oftheDocument

2.1 Development ofaSeismotectonic Framework forPSHA

2.3 Site Specific vs. Regional Study

2.5 Logic Tree Approach andTreatment ofUncertainties

2.5.1 Epistemic Uncertainty vs. Aleatory Variability

2.5.2 Logic Tree Methodology

2.5.3 Site Response

2.5.3.1 Methods forImplementing Site Response Effects

2.5.4 Use ofExperts

2.5.4.1 Guidance Regarding Expert Elicitations

2.6 Interface Issues Between Work Packages

2.7 Common Required Outputs forSeismic Hazard Results

2.7.1 Basic Definitions andRequirements

2.7.1.1 Control Point

2.7.1.2 Components ofMotion

2.7.1.3 Vibration Frequencies forHazard Analysis

2.7.1.4 Lowest Annual Probability ofExceedance

2.7.2 Common Hazard Results

2.7.2.1 Seismic Hazard Curves forReference Rock Site Condition

2.7.2.2 Seismic Hazard Curves fortheSoil Site Condition

2.7.2.3 Epistemic Uncertainty intheSoil Hazard Curves

2.7.2.4 Uniform Hazard Spectra

2.7.3 Additional Parameters

2.7.3.1 Ground Motion Duration

2.7.3.2 Peak Ground Velocity

2.7.3.3 Cumulative Absolute Velocity

3.1 Pre-requisites toDevelop theSSC Models

3.3 Seismic Source Models

3.3.2 Seismic Source Characterization Framework

3.3.3 Area Source, Fault Sources, Gridded Seismicity

3.3.3.1 Area Sources

3.3.3.2 Fault Sources

3.3.3.3 Gridded/Smoothed Seismicity

3.3.4 Lessons Learned Related toSeismic Source Models

3.4 Occurrence Processes

3.4.1 Poisson Model

3.4.2 Characteristic Model

3.4.3 Time-Dependent Seismicity Models

3.5 Maximum Magnitude andRecurrence Parameters

3.5.1 Maximum Magnitude

3.5.2 Recurrence Parameters

3.5.3 Lessons Learned

3.6 Logic-Tree Implications

3.6.1 Logic Tree Approaches

3.6.2.3 Sensitivity Analyses

3.6.3 Verification andQuality Assurance (QA)

4.1 Empirical Models andPoint Source Stochastic Models

4.1.1 Empirical Ground Motion Attenuation Models

4.1.2 Point Source Stochastic Models

4.2 Model Selection andCriteria

4.2.1 Modelling Criteria

4.2.2 Tectonic Consistency

4.2.3 Site-Conditions Consistency

4.3 Corrections or Modifications ofPublished Models

4.3.1 -VS30 (Simulation-Based) Correction

4.3.2 Data-Based Predictions forHard Rock

4.4 Standard Deviation ofModel Predictions; Truncation

4.4.1 Sigma Truncation

4.5 Approaches fortheVertical Ground Motion Component

4.6 Logic Tree Implications

4.7 Lessons Learned fromtheSIGMA Project

5.1 Soil Characterization