Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

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

Dynamic soil-structure interaction models for fragility characterisation of

buildings with shallow foundations
Francesco Cavalieri a, *, Anto
�nio A. Correia b, Helen Crowley c, Rui Pinho d, e
a
European Centre for Training and Research in Earthquake Engineering (EUCENTRE), Via Adolfo Ferrata 1, 27100, Pavia, Italy
b
National Laboratory for Civil Engineering (LNEC), Av. do Brasil 101, 1700-066, Lisboa, Portugal
c
Seismic Risk Consultant, 27100, Pavia, Italy
d
University of Pavia, Civil Engineering Department, via Adolfo Ferrata 5, 27100, Pavia, Italy
e
Mosayk Ltd., via Fratelli Cuzio 42, 27100, Pavia, Italy

A R T I C L E I N F O A B S T R A C T

Keywords: Dynamic Soil–Structure Interaction (SSI), involving the coupling of structure, foundation and soil, is a crucial and
Fragility curves challenging problem, especially when soil nonlinearity plays an important role. This paper shows the impact of
Soil-structure interaction (SSI) adopting different SSI models on the assessment of seismic fragility functions. The linear substructure approach is
Shallow foundations
initially adopted by implementing two different models, the first of which is one-dimensional and includes,
Impedance functions
between the foundation node and the ground, a translational elastic spring and a dashpot, whose stiffness and
Lumped-parameter model
Nonlinear macro-element viscous damping are retrieved from the real and imaginary parts of the dynamic impedance at the first natural
Unreinforced masonry frequency of the structure. The second and more refined model is a Lumped-Parameter Model (LPM) accounting
for frequency dependence of the impedance. In order to explore the sensitivity of fragility functions to the
linearity assumption, an additional approach, including soil nonlinearities, is employed. A nonlinear footing
macro-element is adopted to model the near-field behaviour by condensing the entire soil-foundation system into
a single nonlinear element at the base of the superstructure. Energy dissipation through radiation damping is also
accounted for. The superstructure response is simulated in all approaches as a simple nonlinear single-degree-of-
freedom (SDOF) system. The comparison between the adopted approaches is evaluated in terms of their effects
on the characterisation of fragility functions for unreinforced masonry buildings (URM) on shallow foundations.

1. Introduction
Recent earthquake occurrence in northern Netherlands has been
attributed to gas production activity in the Groningen field, the largest of
which to date has been the Huizinge event of August 2012 with a
magnitude ML 3.59 (Mw 3.53: Dost et al., 2018) [1]. In response to this
induced seismicity, the operators of the field, NAM - Nederlandse Aar­
dolie Maatschappij B.V., have been developing a comprehensive seismic
hazard and risk model for the region, which comprises the entire gas
field plus a 5 km buffer zone onshore (van Elk et al., 2019) [2].
A key component of the risk assessment involves the definition of
fragility functions (which describe the probability of exceeding a given
damage or collapse state, conditional on the intensity of input ground
motion) for each building type that has been identified within the re­
gion, and included in the exposure model. At least one real
representative building from the region was found for each building type
(the so-called index building) and structural drawings were used to
develop a multi-degree-of-freedom (MDOF) numerical model of the
structural system that included also the predominant non-structural el­
ements (such as partition and external façade walls). Nevertheless,
running nonlinear dynamic analyses of many such numerical models,
each subjected to tens or hundreds of records, was deemed to be a too
large computational effort to allow fragility functions to be directly
developed from these analyses. A simplified single-degree-of-freedom
(SDOF) equivalent system approach, which considers dynamic Soil-
Structure Interaction (SSI), was thus used instead to analytically
derive the fragility functions for the structural system of each building
typology (Crowley et al., 2017 [3], 2019 [4]).
Khosravikia et al. (2018) [5] investigated the effect of SSI on seismic
risk, the latter interpreted as the probability distribution of seismic

* Corresponding author.
E-mail addresses: francesco.cavalieri@eucentre.it (F. Cavalieri), aacorreia@lnec.pt (A.A. Correia), helen.crowley78@gmail.com (H. Crowley), rui.pinho@unipv.it,
rui.pinho@mosayk.it (R. Pinho).

https://doi.org/10.1016/j.soildyn.2019.106004
Received 31 July 2019; Received in revised form 15 November 2019; Accepted 8 December 2019
0267-7261/© 2019 Elsevier Ltd. All rights reserved.

Please cite this article as: Francesco Cavalieri, Soil Dynamics and Earthquake Engineering, https://doi.org/10.1016/j.soildyn.2019.106004
F. Cavalieri et al. Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

monetary loss due to structural and nonstructural damage; the findings addresses the adopted methodology to retrieve fragility functions and
show that structures (with shallow foundations) on very soft soils are the comparison between the obtained fragility curves for the investi­
expected to experience reduced losses due to SSI, whereas the presence gated index buildings, whilst final discussion and conclusions are given
of moderately soft soils leads to considerable probability that SSI has in Section 7.
detrimental effects and thus increases the seismic losses. It confirms the
well-known fact that SSI can be beneficial, detrimental, or uninfluential 2. Investigated index buildings
on the seismic vulnerability, and thus risk, of structures, depending on
the soil and structural characteristics. By modelling SSI through a finite Ten different index buildings (Arup, 2017) [14], all typically con­
element soil-block (i.e. using the direct approach) and performing In­ structed with shallow foundations, have been considered herein, with
cremental Dynamic Analysis (IDA) to retrieve fragility curves of rein­ the characteristics summarised in Table 1. These residential buildings
forced concrete (RC) buildings, Pitilakis et al. (2014) [6] found that the are either detached or terraced (with units varying from 2 to 8) and,
consideration of SSI effects may significantly affect the expected per­ depending on their age, they are constructed with timber or concrete
formance of structures founded on soft soils producing an important floors, and solid or cavity URM walls. Fig. 1 shows screenshots of the
shift to the left of fragility curves, i.e. towards more fragile response, in numerical models for the buildings.
comparison to the fixed-base case. The crucial role of SSI under linear or As mentioned above, in all SSI models considered in this work, the
nonlinear soil behaviour in altering the expected structural performance superstructure is represented in a simplified way as a SDOF system,
and fragility curves of high-rise fixed-base RC structures was also whose behaviour is described in SeismoStruct (Seismosoft, 2019) [15]
pointed out by Karapetrou et al. (2015) [7], who concluded that the with the multi_lin model (Sivaselvan and Reinhorn, 1999) [16]. The
hypothesis of fixed-base structure may lead to unconservative results. latter, characterised by a polygonal hysteresis loop, can simulate the
Nevertheless, it should be recognised that SSI will, in general, reduce the deteriorating behaviour of strength and stiffness. The sixteen parameters
structural deformation demands at the cost of larger overall needed to fully define the response curve, for which an example is shown
displacements. in Fig. 2, are related either to the backbone curve or to the hysteretic
Lesgidis et al. (2017) [8] quantified the impact of the frequency rules.
dependence of the (linear) SSI on the fragility of RC bridges. By In order to calibrate this hysteretic model, and with the exception of
comparing the predicted vulnerability of a reference bridge using both a LNEC-BUILD3 (for which shake-table test results were employed), fixed-
conventional, frequency-independent, Kelvin–Voigt model and the base MDOF models for each index building were produced in LS-DYNA
lumped parameter formulation by the same authors, it was found that [10] or ELS [17] and were subjected to nonlinear dynamic analyses
the actual fragility curves of a bridge can be both underestimated or using 11 training records (see Arup, 2017 [18] for further details). The
overestimated by the simplified, frequency-independent approach, and maximum attic displacement of a given MDOF model under each
thus the latter may lead to a bridge behaviour significantly diverging training record was converted to the equivalent SDOF displacement (see
from the actual one. Crowley et al., 2019 [4]) and then compared with the displacement
An interesting study highlighting the importance of taking into ac­ obtained under the same records for the fixed-base SDOF model in
count soil nonlinearities in SSI was presented by Bolisetti et al. (2018) SeismoStruct. These displacements were plotted against the average
[9], with reference to risk assessment of nuclear structures. Results from spectral acceleration (AvgSa) of each record, defined as the geometric
the nonlinear time-domain SSI analysis in LS-DYNA (LSTC, 2013) [10] mean of ten spectral acceleration ordinates from 0.01 to 1 s, and the
for high intensity shaking were compared with those from an linear regressions, in log-log space, of each model were compared (Fig. 3
equivalent-linear analysis using a frequency-domain code, namely Sys­ shows such comparison for one of the structural models); the SDOF
tem for Analysis of Soil-Structure Interaction (SASSI) [11], finding that model was iteratively adapted until a reasonable match was obtained.
the equivalent-linear and nonlinear responses are significantly different: The adopted properties for the SDOF systems for each index building
it was concluded that ignoring the nonlinear effects, including gapping, are reported in Table 2. The symbol Heff denotes the effective height of
sliding and uplift, may lead to an unconservative prediction of the su­ the SDOF (see Crowley et al., 2019 [4]), representing the building
perstructure response and of its seismic risk. The relevance of a centroid height. The sixteen parameters of the multi_lin hysteretic model
nonlinear approach was also highlighted by Petridis and Pitilakis (2018) are defined as follows: EI is the initial stiffness (kN/m), PCP and PCN are
[12], who retrieved fragility curves for a set of RC moment resisting the positive and negative “cracking” force (kN), PYP and PYN are the
frames using the Beam on Nonlinear Winkler Foundation (BNWF) model positive and negative yield force (kN), UYP and UYN are the positive and
and lumped individual elastic springs. Rajeev and Tesfamariam (2012) negative yield displacement (m), UUP and UUN are the positive and
[13] using the BNWF model provided another evidence of the impact of negative ultimate displacement (m), EI3P and EI3N are the positive and
(nonlinear) SSI on the seismic vulnerability of RC frames. negative post-yield stiffness as percent of elastic, HC is the stiffness
The objective of the current work is to investigate the impact of degrading parameter, HBD is the ductility-based strength decay
adopting different SSI models of shallow foundations on the collapse parameter, HBE is the hysteretic energy-based strength decay param­
fragility functions for the unreinforced masonry (URM) building typol­ eter, HS is the slip parameter, and IBILINEAR is a model parameter equal
ogies found in the Groningen exposure model (Crowley et al., 2019) [4],
and, to this end, three different SSI modelling approaches were adopted. Table 1
The first two, namely a one-dimensional frequency-independent model Summary of the URM index buildings with shallow foundations.
and a Lumped-Parameter Model (LPM) accounting for frequency Index Building System Floor Wall Number of Mass
dependence of the impedance, belong to the linear substructure Name type type type storeys (tonnes)
approach, considering kinematic and inertial interaction effects by the
Zijlvest Terraced Concrete Cavity 2 þ attic 219
principle of superposition. Instead, the third approach relies on the Julianalaan Terraced Concrete Cavity 2 þ attic 252
adoption of a nonlinear macro-element and belongs to the class of E45 Terraced Concrete Cavity 2 þ attic 315
hybrid methods, combining the features of sub-domain decomposition Patrimoniumstraat Terraced Timber Cavity 2 þ attic 148
and finite element modelling, including soil nonlinearities. Kwelder Detached Concrete Cavity 1 þ attic 96
Badweg Detached Timber Cavity 1 þ attic 44
Sections 2 and 3 of this manuscript summarise the properties of the LNEC-BUILD3 Detached Timber Solid 1 þ attic 44
investigated index buildings and of the soil in Groningen, respectively. Dijkstraat Detached Timber Solid 2 þ attic 138
The two linear SSI models are described in Section 4, dedicated to the Solwerderstraat Detached Timber Solid 2 þ attic 106
substructure approach, while Section 5 presents the nonlinear SSI De Haver (house) Detached Timber Solid 2þ 159
mezzanine
approach with the developed footing macro-element. Section 6

2
F. Cavalieri et al.
Fig. 1. Screenshots of LS-DYNA or ELS models of URM index buildings with shallow foundations (Arup, 2017) [18].
Fig. 2. Example response curve for the multi_lin hysteretic model.
to 0 for trilinear model, 1 for bilinear model, and 2 for vertex-oriented
model.
The typical foundations of both detached and terraced building types
consist of a grid of continuous beams oriented in two orthogonal di­
rections, of either unreinforced masonry or concrete.
Fig. 4a shows the schemes of foundations considered for bearing
capacity calculation by Crux Engineering (2014) [18] for both unrein­
forced masonry and concrete foundations. The typical width is 600 mm,
whereas the foundation level ranges between 0.2 and 1 m.
A typical unreinforced masonry foundation is a strip foundation that
is achieved by a widening of the load bearing walls; a representative
foundation section is shown in Fig. 4b (Arup, 2015a) [19]. The inertia
characteristics of the foundation were evaluated considering a rectan­
gular section 660 mm wide, characterised by the same moment of inertia
of the section showed in Fig. 4b. The resulting height of the equivalent
3
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
Fig. 3. Comparison of displacements from a MDOF (transformed to SDOF) LS-
DYNA model and a corresponding SDOF SeismoStruct model with calibrated
multi_lin hysteretic model.
square section is equal to 221 mm. A typical reinforced concrete foun­
dation is a strip foundation with width 600 mm and height 330 mm; only
in some cases, for non-bearing walls of terraced buildings, a square cross
section 330 mm wide was considered, as shown in Fig. 4c.
For both unreinforced masonry and concrete, a foundation level at
600 mm depth was considered herein.
3. Soil characterisation in Groningen
In order to account for SSI it is first required to define representative
soil profiles that may be used for assessment of the input parameters of
the different models used (one-dimensional frequency-independent,
LPM, macro-element). The selection of representative soil profiles
takes advantage of the detailed microzonation carried out in recent
years for the Groningen region, resulting in maps of the site response
F. Cavalieri et al. Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Table 2
Adopted properties for the SDOF systems.
Index Building Name Mass (tonnes) Period (s) Heff (m) multi_lin hysteretic model parameters (in base units of kN and m)

EI PCP PYP UYP UUP EI3P

Zijlvest 219 0.34 3.75 75000 150 320 0.020 0.109 0.012
Julianalaan 248 0.15 4.01 448000 800 1300 0.034 0.050 1.00E-09
E45 308 0.24 4.99 202900 467 900 0.010 0.098 1.00E-09
Patrimoniumstraat 148 0.10 2.85 571429 400 800 0.007 0.017 0.018
Kwelder 96 0.08 2.75 600000 300 500 0.008 0.118 1.00E-09
Badweg 44 0.13 2.81 100000 150 151 0.002 0.020 0.054
LNEC-BUILD3 44 0.08 2.72 253906 138 246 0.014 0.052 1.00E-09
Dijkstraat 138 0.36 6.70 41366 400 401 0.010 0.154 1.00E-09
Solwerderstraat 106 0.30 5.40 46875 150 600 0.033 0.123 1.00E-09
De Haver 159 0.13 3.70 400000 400 900 0.010 0.061 1.00E-09

Index Building Name multi_lin hysteretic model parameters (in base units of kN and m) [continued from table above]

PCN PYN UYN UUN EI3N HC HBD HBE HS IBILINEAR

Zijlvest 150 320 0.020 0.109 0.012 1 0.001 0.001 1 0

Julianalaan 800 1300 0.034 0.050 1.00E-09 200 0.001 0.001 1 0
E45 467 900 0.010 0.098 1.00E-09 1 0.001 0.001 1 0
Patrimoniumstraat 400 800 0.007 0.017 0.018 1 0.001 0.001 1 0
Kwelder 300 500 0.008 0.118 1.00E-09 1 0.001 0.001 1 0
Badweg 150 151 0.002 0.020 0.054 1 0.001 0.001 1 0
LNEC-BUILD3 138 246 0.014 0.052 1.00E-09 200 0.001 0.001 1 0
Dijkstraat 400 401 0.010 0.154 1.00E-09 1 0.001 0.001 1 0
Solwerderstraat 150 600 0.033 0.123 1.00E-09 1 0.001 0.001 1 0
De Haver 400 900 0.010 0.061 1.00E-09 1 0.001 0.001 1 0

Fig. 4. a) Schemes of foundations for bearing capacity calculation; b) Typical masonry and c) concrete foundations for both detached and terraced buildings.

Amplification Factor (AF) for several spectral ordinates (Rodriguez-
Marek et al., 2017) [21]. The examination of AF distributions shows that
in general the patterns of high and low AF are well reflected by the
geological zonation model (Bommer et al., 2017) [22]. Therefore, the AF
represents well the soil behaviour of the shallow deposits, and it can be
considered a reliable parameter for the identification of representative
soil profiles.
The site response analysis study (carried out for ten levels of input
motion) was performed for a grid of about 1400 000 points homoge­
neously distributed in the Groningen area. Fig. 5a shows the distribution
of AFs for the highest input motion level spectral ordinates for a period
of 0.5 s. Due to the non-negative values of the AF, it was assumed that AF
follows a lognormal distribution.
A representative shear wave velocity (VS) profile was evaluated as
the mean of VS profiles around the median AF (equal to 2.25) consid­
ering all sites with AFs in an interval of amplitude equal to 0.2. The AFs
corresponding to the largest input motion level were considered.
Different levels of AF and/or input motion can be selected to define
alternative VS profiles in future works, with the median AF being the
most representative. The VS profile is not the only relevant parameter for
SSI, therefore a real stratigraphy, with the corresponding soil parame­
ters (strength, stiffness, etc.), needs to be identified. The simplest way to
perform this operation is to identify a real stratigraphy (i.e., one from
aforementioned 140 k sites considered for site response analyses)

4
F. Cavalieri et al.
Fig. 5. a) Histogram of AF at a period equal to 0.5 s and highest input motion level; b) Mean shear wave velocity profile around median AF vs best fit profile. Plots
were derived using data described in Kruiver et al. (2017) [23] and Rodriguez-Marek et al. (2017) [21].
compatible with the computed mean VS profile. This was done by
evaluating the deviation between the mean VS profile and each one of
the VS profiles in the interval of median AF considered. Fig. 5b shows the
comparison between the mean VS profile and the VS profile with mini­
mum deviation.
The upper 30 m of the selected soil deposit is constituted by an
alternation of fine sand and cohesive layers (i.e. clayey sand and sandy
clay). In the shallow part, which mostly affects the response of shallow
foundations, there is a 5 m thick layer of fine sand. The shallow water
table level implies that the computation of the footing seismic response
should be performed in undrained conditions, given the large velocities
of soil deformation due to seismic loading.
In the framework of the site response analysis carried out for the
Groningen region, and in addition to the VS profile and soil stratigraphy,
Fig. 6. Undrained shear strength measured/estimated in the shallow layers of
the representative soil profile.
5
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
a number of other geotechnical properties were considered, including a
set of geomechanical parameters important to describe the dynamic soil
behaviour such as the modulus reduction and damping curves (see
Kruiver et al., 2017 [23] and Rodriguez-Marek et al., 2017 [21]). Un­
fortunately, for the fine sand surficial layer, strength parameters are not
available; consequently, these were estimated based on existing litera­
ture, trying to constrain the selected values based on available infor­
mation (i.e. VS profile, coefficient of uniformity and D50 – diameter of
the particle with 50% of passage in the grain size distribution). In
particular, Fear and Robertson (1995) [24] proposed a framework for
estimating the undrained steady state shear strength of sand (su) from in
situ tests; the formulation combines the theory of critical state soil me­
chanics with shear wave velocity measurement. This undrained shear
strength was computed taking into account the drained conditions on
the distribution of soil stresses along the depth due to permanent loads
resulting from the weight of the soil above a given depth (the effect of
the relatively modest weight of the structures being studied was
considered in these calculations through engineering judgement and
considering all uncertainties involved, resulting in a slight increase of
the undrained shear strength at shallow depths). Fig. 6 shows the un­
drained shear strength profile in the shallow part of the selected
representative soil profile, used to compute the bearing capacity under
undrained conditions. These values are for the free-field case, since the
actual values at each building site will slightly differ when the building
weight is considered.
For the complete set of calculations and results obtained for the
characterisation of soil profiles in the Groningen region, interested
readers are referred to the report by Mosayk (2019) [25].
4. Substructure approach
As mentioned already, in this work SSI was initially analysed by the
substructure approach, which allows splitting kinematic and inertial
interaction in different sub-steps and considering their combined effects
using the principle of superposition (Mylonakis et al., 2006) [26]. A
substructure approach is typically subdivided in three sub-steps: (i) ki­
nematic interaction, where the effect of having a foundation with a
given geometry and stiffness in the soil is analysed for evaluating any
possible modifications to the input motion at the base of the structure in
comparison to the free-field motion, resulting in the Foundation Input
Motion (FIM); (ii) soil impedance determination, where the deform­
ability and overall dynamic characteristics of the soil layers are analysed
F. Cavalieri et al.
in order to include the soil compliance at the base of the structural
model; (iii) inertial interaction, where the structural mass is considered
and its effects on the overall response of the structure, subjected to the
FIM and on a compliant base, are determined.
Kinematic interaction modifications of the free-field motion on
shallow foundations are mainly attributed to three phenomena: (i) base
slab averaging of the motion underneath the footings; (ii) a ground
motion amplitude decrease at the foundation level, with the embedment
of the footings, and possible introduction of rocking motions at the
footings due to such variation along the depth of the foundation; (iii)
wave scattering effects at the edges of the footings, resulting in a reduced
amplitude of high-frequency components (Stewart, 2000 [27], Iovino
et al., 2019 [28], Brandenberg et al., 2015 [29], Di Laora, 2016 [30],
Conti et al., 2018 [31]). In practical applications, earthquake engineers
commonly neglect the effects of kinematic interaction (Dezi et al., 2010)
[32]. In the seismic assessment of the type of structures considered in
this study, Arup (2015a, 2015b) [20], [33] also considered that kine­
matic interaction could be assumed as being negligible. In fact, given the
relatively shallow depth of the footings of the index buildings considered
in this study, as shown in Fig. 4, the ground motion amplitude decrease
at the foundation level with respect to the free-field is assumed to be
negligible. Likewise, the base slab averaging and wave scattering effects
were considered to be unimportant for vertically propagating S-waves.
As a consequence, the free-field motion was used as input motion for the
nonlinear dynamic analyses in this study.
On the other hand, inertial interaction includes the dynamic
response of the coupled soil-foundation-structure system due to the
input motion and is characterised predominantly by a shift of structural
frequencies to lower values and by an increase of damping in the
coupled system. In the substructure approach, the soil is typically
replaced by a set of springs and dashpots (as well as masses, in some
cases) at the foundation level, representing the foundation dynamic
impedance (see Section 4.1). The latter is a complex-valued function,
whose real and imaginary parts vary with frequency and depend on the
stiffness and on the energy dissipation properties of the system,
respectively.
Two different models following the substructure approach were
implemented in SeismoStruct for derivation of fragility functions; they
are described in Sections 4.2 and 4.3 in relatively brief fashion, but
interested readers may refer to the report by Mosayk (2019) [25] for all
those details, and results, that could not be included here due to space
constraints.
4.1. Definition of impedance functions
Impedance functions were evaluated using the software DYNA6.1
(GRC, 2015) [34]. The foundations of the considered buildings consist of
a grid of continuous beams oriented in two orthogonal directions.
Conversely, the structural model used for definition of the fragility
curves is a SDOF system in which the contact with the soil is limited to a
single point. The geometry of the foundation system does not allow a
simple and unique definition of equivalent dimensions for impedance
function calculation; in fact, depending on the degree of freedom ana­
lysed (i.e. translational or rotational) or on the nature of the impedance
component under consideration (i.e. stiffness or damping), the charac­
teristics of the real foundation to be preserved are different (contact
area, inertia, etc.). For such reason, in order to properly consider the real
foundation geometry, the definition of the equivalent footing di­
mensions for impedance calculation made use of the calibration step
carried out for the macro-element analyses (see Sections 5.2 and 5.3),
which employs a 3D MDOF model of the buildings. For each building,
these equivalent dimensions were evaluated independently for stiffness
and damping, as well as for the translational and rotational degrees of
freedom, in order to reproduce the static stiffness and damping deter­
mined for the equivalent macro-element of the SDOF system described
in Section 5.3.
6
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
The impedance functions were computed considering a composite
medium (i.e. soil layer with limited depth on top of a half-space),
characterised by a linear shear wave velocity profile in the upper soil
layer and a constant value on the half space underneath. The layer
properties (e.g. thickness, shear wave velocity) were defined taking into
account the constraints imposed by the software used, which considers
fixed values of the ratio of layer thickness to the half-width of the
equivalent square footing. Moreover, the fitting of the shear wave ve­
locity profile was carried out for a ratio between the shear wave velocity
at the base of the footing and at the half-space equal to 0.6. Fig. 7a shows
an example of shear wave velocity profile fitting. Given the different
equivalent dimensions considered, the VS profile fitting needs to be
repeated for each of the four cases (stiffness and damping, translational
and rotational degrees of freedom) accounted for.
It is also noted that the employed software (DYNA6.1) considers
fixed values of material damping, equal to 0.03 for the upper layer and
0.05 for the half-space. Impedance functions were calculated consid­
ering a Poisson’s ratio ν equal to 0.45, i.e. corresponding to a quasi-
incompressible medium.
Based on the results of site response analysis, scaling factors (SF) for
the VS profile were defined to account for soil nonlinearity depending on
the strain level. A relationship between PGA and VS scaling factors was
obtained considering at different PGA levels the mean strain level and
shear modulus degradation in the fine sand layer, which is characterised
by two different degradation curves. Fig. 7b shows the G/Gmax scaling
factors for the two shear modulus degradation curves considered within
the fine sand layer. Five PGA levels ranging from 0.05 g to 0.43 g were
considered in the derivation of impedance functions, with different sets
of impedance functions being used in the fragility curve derivation at
different seismic intensities. Fig. 10 shows an example of impedance
functions computed using the input data described above.
4.2. One-dimensional frequency-independent model
The simplest SSI model employed in the fragility functions’ devel­
opment in this work is a one-dimensional frequency-independent model,
called SSI 1-D hereafter, having a lateral spring with stiffness kx and a
dashpot with viscous damping coefficient cx (see Fig. 8). The model
proposed by Maravas et al. (2014) [35] was taken as a reference. Such
model was simplified by considering only the translational spring and
dashpot (i.e., rotational spring and dashpot are not included) and
frequency-independent impedance.
The kx and cx parameters should be evaluated at the resonant fre­
quency of the compliant system, that is, the system including the
structural SDOF plus spring and dashpot. The fundamental period of
such system can be estimated as a function of the period of the fixed-base
SDOF and the ratio between the stiffness of the fixed-base SDOF and the
one of the lateral spring (Bilotta et al., 2015) [36]. This fundamental
period estimation requires an iterative process, in which the spring
stiffness at the first step is evaluated at the resonant frequency of the
fixed-base SDOF: for the ten considered index buildings, a few iterations
(up to three, depending on the building) were needed. The kx and cx
parameters were then evaluated at both resonant frequencies (i.e. of the
compliant and fixed-base systems), considering the impedance functions
derived at the highest PGA level, corresponding to a 100 k years return
period: at this intensity level, the real part of impedance attains the
lowest values and hence the highest period divergence (between the two
systems) occurs. The comparison carried out for the ten index building
showed small to negligible variations of both kx (zero or less than 1% for
eight buildings and around 5% and 10% for the remaining two) and cx
(always less than 1%). Such variations are not expected to alter, at least
significantly, the fragility curves. It is also noteworthy to highlight that
for a return period of 100 k years the soil behaviour is deemed to be
nonlinear, and thus the inaccuracies due to the use of elastic impedance
make the obtained small variations even less influential. On the other
hand, for lower intensity levels higher values of the real part of
F. Cavalieri et al. Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Fig. 7. a) Example of shear wave velocity profile fitting; b) G/Gmax scaling factors obtained from site response analysis for different levels of shear strain in the fine
sand layers, characterised by two shear modulus degradation curves.

4.3. Lumped-Parameter Model (LPM)

A Lumped-Parameter Model (LPM) accounting for frequency

Fig. 8. The adopted one-dimensional frequency-independent model.
impedance occur, leading to lower period shifts and, as a consequence,
even smaller kx and cx variations between the two systems.
For the above reasons and for simplicity, the values of the stiffness
and viscous damping coefficient were obtained by evaluating the
impedance functions derived for the Groningen field at the fundamental
frequency of the fixed-base SDOF.
The structural SDOF mass, stiffness and damping coefficient are
indicated with ms, ks and cs, respectively, in Fig. 8. The seismic excitation
is input to the system as an acceleration time history, a(t), applied to the
fixed support at the base.
Table 3 and Table 4 report the retrieved properties of the SSI 1-D
systems for two index buildings (one terraced and one detached), in
terms of stiffness and damping coefficient, for all five scaling factors (SF)
considered. For the complete set of SSI 1-D parameters, interested
readers are referred to the report by Mosayk (2019) [25].
dependence of the impedance functions was also simulated in Seis­
moStruct and used for the derivation of fragility functions.
Even though techniques are available to describe frequency depen­
dence of any type through a generalised LPM whose form is not known
in advance (Lesgidis et al., 2015) [37], this work adopted the simplest
LPM capable of describing approximately, over the frequency range of
interest, the features of two components of impedance, namely the
translational and rotational terms.
The LPM model proposed by Dezi et al. (2009) [38] and Carbonari
et al. (2011, 2012, 2018) [39–41], was taken as a reference. Such model
was simplified in order to neglect the rocking-sway coupling, because
the focus in this paper is on shallow foundations where such coupling is
not important in the linear range. The adopted system is shown in Fig. 9.
The crucial feature of this LPM is the introduction of a translational
fictitious (non-physical) mass mx at the interface node (representing the
foundation), linked to the ground by a translational spring (of constant
kx) and by a dashpot (of constant cx). This system is characterised by a
frequency-dependent response to an input and thus allows for an
approximate description of the frequency dependence of the impedance.
Expressing the equation of motion of the system without the super­
structure in the frequency domain, it can be easily seen that the real
component of the complex dynamic impedance decreases parabolically
(kx – mx ω2) with frequency, whereas the imaginary part increases lin­
early (cx ω) with frequency. In case the foundation mass is taken into
account, it is added to the fictitious mass in the same node.
In order to model the foundation rotation, the LPM also includes a
fictitious rotational mass mry at the interface node, linked to the ground
by a rotational spring (of constant kry) and dashpot (of constant cry).
The soil portion of the LPM is thus characterised by two independent

Table 3
Properties of the SSI 1-D system for Patrimoniumstraat (terraced) index building.
SF1 SF2 SF3 SF4 SF5

kx (kN/m) 4.805Eþ06 3.889Eþ06 3.225Eþ06 2.744Eþ06 1.997Eþ06

cx (ton/s) 8.754Eþ04 7.856Eþ04 7.056Eþ04 6.430Eþ04 5.438Eþ04

Table 4
Properties of the SSI 1-D system for Solwerderstraat (detached) index building.
SF1 SF2 SF3 SF4 SF5

kx (kN/m) 3.545Eþ06 2.804Eþ06 2.281Eþ06 1.918Eþ06 1.396Eþ06

cx (ton/s) 3.334Eþ04 2.978Eþ04 2.698Eþ04 2.478Eþ04 2.085Eþ04

7
F. Cavalieri et al.
Fig. 9. The adopted Lumped-Parameter Model.
Fig. 10. Sample fit of real and imaginary parts of two impedance components,
in the 0–10 Hz frequency range.
degrees of freedom. The mass matrix takes the form:
� � � � Following the parametric study by Pianese (2018) [47], the five
M11 M12 mx 0
M¼ ¼ (1)
M12 M22 0 mry
The stiffness and damping matrices, K and C, are written similarly.
The six diagonal terms of the matrices, namely M11, M22, K11, K22, C11,
C22, which are coincident with the parameters of the soil portion of the
LPM, are obtained by fitting the two components of impedance (i.e.,
translational and rotational) with parabolic and linear functions for the
real and imaginary parts, respectively.
Fig. 10 shows an example of such fit, for a structural SDOF with first
natural frequency of 7.6 Hz. In order to capture the inertial interaction
effects between the superstructure and the foundation, the superstruc­
ture mass is placed above the ground at the building centroid height,
Heff, and is connected to the interface node by a rigid link. In this way,
the rigid displacement of the superstructure mass due to the foundation
rotation θf, equal to Heff ⋅θf , is taken into account within the nonlinear
dynamic analyses, and then subtracted from the total displacement, as
8
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
well as the sliding translational displacement of the footing, resulting in
the structural displacement only. The seismic acceleration, a(t), is
actually input to the system as an inertia force history, f(t), applied to the
superstructure mass: this approach properly considers the inertial
components in the presence of the structure (structure and foundation
masses and their interaction), resulting in a response in terms of relative
displacements with respect to the ground motion.
Table 5 and Table 6 report the retrieved properties of the LPM sys­
tems for two index buildings (one terraced and one detached), in terms
of mass, stiffness and damping coefficients, for all five scaling factors
considered. For the complete set of LPM parameters, interested readers
are referred to the report by Mosayk (2019) [25].
5. Hybrid approach with nonlinear macro-element
5.1. Overview of the employed nonlinear macro-element
Inertial interaction in presence of nonlinear soil response can be
simulated through the use of soil-foundation macro-elements. SSI
analysis using macro-elements is frequently adopted in research studies
for a more reliable estimation of soil displacements, given that these
have been previously shown to be a cost-effective and reliable tool for
such type of analysis, since they suitably represent both the nonlinear
soil behaviour at the near-field and the ground substratum dynamic
characteristics at the far-field, as well as the interaction with the seismic
response of the structure (Correia, 2011, 2013) [42,43]. Hence, all as­
pects of elastic and inelastic behaviour of the foundation system are
encompassed into one computational entity and are described by the
behaviour of a single point at the centre of the foundation.
The footing macro-element model by Correia and Paolucci (2019)
[44] builds upon the concepts and formulations of the models by
Chatzigogos et al. (2011) [45] and by Figini et al. (2012) [46], incor­
porating improvements to address inconsistencies regarding the
formulation of the participating mechanisms and to extend their scope
to three-dimensional loading cases. Moreover, this macro-element in­
troduces an enhanced uplift model, based on a nonlinear elastic-uplift
response, which also includes a phenomenological model for the pro­
gressive degradation of the contact at the soil/footing interface due to
irrecoverable changes in its geometry as plastic deformations develop in
the soil. An amended bounding surface plasticity model, able to repre­
sent the ultimate plastic flow conditions and the transition between the
initial elastic and inelastic responses, was developed. Finally, improved
return mapping algorithms were adopted in order to reproduce a more
general and realistic behaviour, that correctly takes into account the
simultaneous elastic-uplift and plastic nonlinear responses. It was
implemented in SeismoStruct [15] and is used herein for the derivation
of fragility functions. Fig. 11 illustrates nonlinear responses obtained
with the macro-element.
calibration, model-specific, parameters of the macro-element became
well-constrained, allowing for the dynamic response to be obtained with
confidence. The remaining parameters correspond to: (i) the footing
dimensions; (ii) the six initial elastic frequency-independent values of
the diagonal impedance matrix, which can be easily obtained from
literature, and which represent the far-field response; (iii) the six
bearing capacity values, which can be derived from classical formulae,
and which represent the near-field failure conditions. In between these
two extreme types of response, the macro-element gradually evolves
from the initial elastic response to the plastic flow at failure through the
bounding surface plasticity model, incorporating the uplift and contact
degradation phenomena.
The adopted system for nonlinear dynamic analyses, as modelled in
SeismoStruct, composed of a nonlinear structural SDOF and a footing
macro-element, is shown in Fig. 12. As done for the LPM, in order to
capture the inertial interaction between the superstructure and the
foundation (with mass mf), the superstructure mass is placed above the
F. Cavalieri et al. Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Table 5
Properties of the LPM system for Patrimoniumstraat (terraced) index building.
SF1 SF2 SF3 SF4 SF5

mx (ton) 1.252Eþ03 9.953Eþ02 8.116Eþ02 6.779Eþ02 4.699Eþ02

mry (ton*m2) 1.743Eþ05 1.604Eþ05 1.501Eþ05 1.328Eþ05 9.618Eþ04
kx (kN/m) 9.649Eþ06 7.741Eþ06 6.366Eþ06 5.368Eþ06 3.816Eþ06
kry (kNm/rad) 7.653Eþ08 6.188Eþ08 5.127Eþ08 4.351Eþ08 3.147Eþ08
cx (ton/s) 8.619Eþ04 7.735Eþ04 6.967Eþ04 6.331Eþ04 5.369Eþ04
cry (ton*m2/s) 5.223Eþ04 4.894Eþ04 4.641Eþ04 4.416Eþ04 4.014Eþ04

Table 6
Properties of the LPM system for Solwerderstraat (detached) index building.
SF1 SF2 SF3 SF4 SF5

mx (ton) 1.762Eþ03 1.532Eþ03 1.360Eþ03 1.192Eþ03 8.420Eþ02

mry (ton*m2) 4.486Eþ04 4.394Eþ04 4.247Eþ04 4.063Eþ04 3.674Eþ04
kx (kN/m) 4.349Eþ06 3.506Eþ06 2.896Eþ06 2.457Eþ06 1.766Eþ06
kry (kNm/rad) 1.549Eþ08 1.253Eþ08 1.039Eþ08 8.847Eþ07 6.435Eþ07
cx (ton/s) 3.234Eþ04 2.889Eþ04 2.617Eþ04 2.403Eþ04 2.024Eþ04
cry (ton*m2/s) 2.047Eþ05 1.771Eþ05 1.567Eþ05 1.413Eþ05 1.170Eþ05

Fig. 11. Examples of macro-element nonlinear responses for an individual footing: vertical settlement due to an increasing vertical force (left); moment-rotation
dominated by uplift behaviour, when a small vertical force is applied (centre); and moment-rotation dominated by plasticity, when a large vertical load is
applied (right).

ground at the building centroid height, Heff. Similarly, the seismic ac­ proposed by Gazetas (1991) [48] for rectangular foundations (B < L are
celeration, a(t), is actually input to the system as an inertia force history, the semi-width and semi-length of the circumscribed rectangle) on ho­
f(t), applied to the superstructure mass. The three springs and dashpots mogenous half-space, the latter being characterised by VS, weight of soil
represented in the 2D view of Fig. 12 model the macro-element elastic unit volume γ and Poisson’s coefficient ν, equal to 190 m/s, 18.4 kN/m3
behaviour in the far-field. Their constants correspond to the stiffness and and 0.45, respectively. The corresponding initial elastic shear modulus
damping in the vertical direction (kV, cV), horizontal x-direction (kHx, Gmax was assumed equal to 67.7 MPa. As mentioned above, the
cHx) and rotational direction around the y-axis (kMy, cMy). For simplicity, macro-element requires as input the constant initial elastic stiffness and
the remaining three springs and dashpots are not visualised in the 2D radiation damping coefficients for the six degrees of freedom. With
scheme: however, such elements are present in the macro-element reference only to the three directions of interest for the analyses (i.e.
implementation and play an active role in the dynamic analyses, being vertical, horizontal x-direction and rotational direction around the
the macro-element behaviour fully coupled in the six directions. y-axis), the following expressions proposed by Gazetas (1991) [48] were
Since the structural model used for the computation of fragility used for the elastic stiffnesses:
curves is a SDOF model, the definition of the input parameters for a � � �0:75 �
representative macro-element requires a calibration step. Such calibra­ kV ¼
2 Gmax L
0:73 þ 1:54
B
tion was carried out in order to define the characteristics of the macro- 1 ν L
element equivalent to the real foundation system, which is composed of � � � � �0:85 �
0:2 Gmax L B 2 Gmax L B
a grid of foundation beams. The calibration step, described in Section kHx ¼ kHy 1 ; ​ with ​ kHy ¼ 2 þ 2:5
0:75 ν L 2 ν L
5.3, was carried out considering a MDOF model for each building, in
� �0:15
which each portion of a foundation beam between openings of the Gmax 0:75 L
structural walls was represented by a macro-element, whose charac­ kMy ¼ I 3
1 ν by B
teristics were defined as described in Section 5.2. (2)

5.2. Properties of the macro-element under a single foundation beam where Iby is the area moment of inertia about the y-axis of the soil-
foundation contact surface. Again, only for the directions of interest,
The input parameters of the macro-element include the foundation the following expressions proposed by Gazetas (1991) [48] were used
impedances and its bearing capacity. The derivation of both sets of in­ for the radiation damping coefficients:
formation was based on the representative soil stratigraphy defined in
cHx ffi ρVS Ab
Section 3. Values for the five model-specific parameters introduced � ​ ; ​ with ​ VLa ¼
3:4
V (3)
above are also required. cMy ¼ ρVLa Iby cry πð1 νÞ S
Since only a shallow depth is involved in the response of the footings,
the foundation impedances were determined using the relationships where ρ is the soil mass density, Ab is the soil-foundation contact surface

9
F. Cavalieri et al.
Fig. 12. The adopted system with footing macro-element
foundations).
area, VLa is the ‘‘Lysmer’s analog’’ wave velocity, defined as the
apparent velocity of propagation of compression–extension waves under
a foundation, while cry is plotted in Gazetas (1991) [48] as a function of
the non-dimensional frequency a0 ¼ ω B/VS. The circular frequency, ω
¼ 2π f, was evaluated at a frequency f equal to 1.67 Hz (i.e. period of 0.6
s). Such value was selected taking into account both the period of the
investigated index buildings and the AF trend with frequency, the latter
showing peaks between 0.4 and 1.67 Hz.
The vertical (Nmax), horizontal (Hmax) and rotational (Mmax) com­
ponents of the foundation bearing capacity were evaluated under un­
drained conditions, considering the undrained shear strength profile
shown in Fig. 6 and using the formulation proposed by EC7 (CEN, 2004)
[49]. In particular, the maximum centred vertical load capacity, Nmax ¼
qlim ⋅B, corresponding to the ultimate static bearing capacity of a foun­
dation characterised by a width equal to B, was evaluated by the stan­
dard superposition formula (e.g. Lancellotta, 2008 [50]):
qlim ¼ su Nc soc doc ioc boc goc þ q
where the bearing capacity coefficient Nc is a function of the angle of
shear resistance, the undrained shear strength su was assumed equal to
12 kPa at the depth of interest, and q is the surcharge acting on the
foundation level. For vertical centred load, the only correcting factor
different from the unity is the shape coefficient, equal to:
B were based on a 2D response). Most of the parameters were computed
soc ¼ 1 þ 0:2
L analytically starting from the foundation geometry and properties of the
The maximum base shear capacity, Hmax, and maximum base
moment capacity, Mmax, which can be calibrated based either on ma­
terial parameters (e.g. soil-foundation friction resistance) or on theo­
retical values, were obtained using the following expressions:
Hmax ¼ su ⋅Ab
Mmax ¼ 0:12⋅Nmax ⋅B
Finally, the five model-specific parameters were assigned values
consistent with the calibration performed in the work by Pianese (2018)
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
[47]. In particular:
� the uplift initiation parameter (α) is only dependent on the assumed
stress distribution of vertical stresses underneath the foundation and
its value can be determined from simple static considerations. In the
analyses, it was set equal to 3, thus assuming a linear distribution of
vertical stresses underneath the foundation for the soil at the
beginning of the analysis;
� the soil/footing contact degradation (dmg), taking into account the
decrease of the contact area due to inelastic rocking, is evaluated
based on calibration to experimental results. In the analyses, it was
set equal to 0.1;
� the reference plastic modulus (ho) was set equal to 0.2;
� the exponent for loading history in unloading/reloading (nUR) was
set equal to 1;
� the plastic potential parameter (χ g) was set equal to 2.
The scallop shape was assumed for the bounding surface, since the
dynamic analyses were performed under undrained conditions [44].
5.3. Properties of the equivalent macro-element
The employed footing macro-element models the soil under a single
footing or foundation beam. However, since a simplified SDOF system
approach was used to represent the structural system, the derivation of
an “equivalent” macro-element for an entire building was needed.
To this end, the first step was to build a MDOF model for each index
building. Fig. 13 shows the models built in SeismoStruct for both
(shallow
terraced and detached buildings. Given the similarity of geometric
properties for all the considered terraced index buildings, the same
model (Fig. 13a) was used for all of them, only changing the total mass
accordingly. For the same reason, the model for Badweg (Fig. 13c) was
used also for the shake-table test specimen LNEC-BUILD3, and the model
for Dijkstraat (Fig. 13d) was used also for Solwerderstraat. The models
for Kwelder and De Haver are shown in Fig. 13b and Fig. 13e,
respectively.
Masonry piers and spandrels were introduced as columns and beams,
respectively. The rigid RC slabs were modelled with rigid diaphragms
linking the column nodes at the floor levels. The total number of footing
macro-elements included at the base of the models, in correspondence to
the centroid of masonry piers, is 27 for the terraced buildings, 16 for the
modern detached house, 13 for the old detached houses, 8 for the
aggregate unit buildings and 28 for the farmhouse building. Reinforced
concrete foundation tie-beams connect the upper nodes of the macro-
elements. Both masonry and reinforced concrete were considered as
linear elastic materials, in the MDOF models, with their actual values for
the elastic modulus and mass density. The total masses of the models,
(4)
given by the superstructure mass plus the foundation mass, are
approximately equal to the actual total masses, which were used in the
derivation of the single macro-element properties as described in Section
5.2.
The equivalent macro-element calibration requires the computation
of the (elastic) stiffnesses, bearing capacity and damping coefficients
along the six directions (or three, in this case, since the fragility analyses
(5)
single macro-elements, while the remaining ones required the output
from the model. The model output parameters needed for the calibration
are the vertical reactions of the macro-elements and the base shear ca­
pacity in the horizontal direction x: the output results were obtained
from a pushover analysis, along x. The latter was carried out pushing the
(6) structure in load control with point forces located at the floor levels,
according to a triangular distribution.
The vertical stiffness, kV, and the horizontal stiffness, kHx, were ob­
tained by simply summing up the stiffness values of the single macro-
elements, assuming a rigid behaviour of the foundation plane. For the
10
F. Cavalieri et al.
Fig. 13. MDOF models in SeismoStruct (used for the definition of some of the input parameters of the SDOF’s equivalent SSI macro-element).
rotational stiffness, kMy, the lower bound would be simply the sum over
the single macro-elements, as done for the other stiffness components,
while adopting the upper bound would mean accounting for both the
rotational stiffness of each macro-element and their vertical stiffness
contribution for a rigid rotation of the foundation plane. For the case at
hand, it was decided to employ the rotational stiffness upper bound,
consistently with the rigid foundation plane assumption adopted for the
horizontal stiffness. Since the dynamic behaviour of buildings on
shallow foundations is driven more by sliding than by rocking, this
choice should not lead to important variations in the results. To verify
this, the fragility curves were also retrieved by using a reduced rota­
tional stiffness, in between the two extreme values. In particular, based
on expert judgement, rather than rigorous mechanical considerations,
one tenth of the upper bound was adopted, a value that is, of course,
larger than the lower bound. For all the considered buildings, this
reduced stiffness led to small to negligible variations in the fragility
curve with respect to the one obtained with the upper bound, as
expected.
Concerning the bearing capacity, the vertical component, Nmax, was
computed as the sum over the single macro-elements, while for the other
components the fully coupled behaviour of the macro-element in the six
directions was used for defining the size of its bounding surface. In
particular, the bearing capacity in the horizontal direction, Hmax, was
obtained as follows:
P
nME
QNH;i ⋅Hmax;i
Hmax ¼ i¼1
QNH
where nME is the total number of macro-elements in the model, Hmax,i is
the maximum horizontal capacity of each macro-element in the direc­
tion considered, and QNH and QNH,i are function of the applied vertical
load, for the equivalent macro-element and for each of the single macro-
elements, respectively. This function of the applied vertical load relates
the maximum horizontal capacity of the macro-element with its actual
horizontal shear capacity. Further details on the involved quantities can
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
be found in Correia and Paolucci (2019) [44].
In order to obtain the rotational bearing capacity, Mmax, the 3D
vertical-horizontal-rotational interaction surface for the capacity was
used to derive the following expression:
P
Ns
Fu;k ⋅hk
Mmax ¼ sk¼1
ffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�
ffiffiffiffiffi (8)
2
Hu
QNM 1
QNH ⋅Hmax
where Fu,k is the ultimate horizontal force at the k-th floor level, ob­
tained from a pushover analysis in the relevant horizontal direction and
considering a triangular distribution along the building height, Ns is the
number of storeys, hk is the height of the k-th floor level, and Hu is the
sum of Fu,k for all storeys and corresponds to the ultimate base shear
value. QNH was already defined, while QNM is also a function of the
applied vertical load that relates the maximum rotational moment ca­
pacity of the macro-element with its actual moment capacity for such
vertical load. The torsional capacity is of no interest for the 2D analyses
performed.
The damping constants modelling the radiation damping in the soil
along the six directions were computed by summing up the values of the
single macro-elements. For what concerns the rocking response this
corresponds to a lower bound assumption. Nonetheless, as mentioned
above, the response of the buildings considered is mainly dominated by
sliding and not by rocking, thus not being affected by this choice.
(7) The five model specific parameters, as well as the bounding surface
type (i.e. scallop shape), were set equal to those of the single macro-
elements.
Table 7 reports the retrieved properties of the equivalent footing
macro-elements for all the investigated terraced and detached index
buildings, in terms of initial stiffness, foundation capacity and radiation
damping equivalent dashpot coefficients, only along the directions of
interest for the analyses. Note that the macro-element properties for
Julianalaan, LNEC-BUILD3 and Solwerderstraat are the same as those
for Zijlvest, Badweg and Dijkstraat, respectively, given that, as
11
F. Cavalieri et al. Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx

Table 7
Properties of the equivalent macro-elements for all the index buildings.
Index Building Name kV (kN/m) kHx (kN/m) kMy (kNm/rad) Nmax (kN) Hmax,x (kN) Mmax,y (kNm) cHx (ton/s) cMy (ton*m2/s)

Zijlvest & Julianalaan 1.521Eþ07 1.167Eþ07 8.514Eþ08 4.653Eþ03 6.634Eþ02 2.188Eþ04 3.568Eþ04 7.356Eþ02
E45 1.521Eþ07 1.167Eþ07 8.514Eþ08 4.653Eþ03 6.474Eþ02 2.289Eþ04 3.568Eþ04 7.356Eþ02
Patrimoniumstraat 1.521Eþ07 1.167Eþ07 8.514Eþ08 4.653Eþ03 6.804Eþ02 2.451Eþ04 3.568Eþ04 7.356Eþ02
Kwelder 7.613Eþ06 5.664Eþ06 1.480Eþ08 2.429Eþ03 3.739Eþ02 7.219Eþ03 1.522Eþ04 4.427Eþ02
Badweg & LNEC-BUILD3 4.828Eþ06 3.478Eþ06 1.123Eþ08 1.425Eþ03 2.192Eþ02 4.709Eþ03 8.417Eþ03 4.364Eþ02
Dijkstraat & Solwerderstraat 6.374Eþ06 5.055Eþ06 1.585Eþ08 2.392Eþ03 2.967Eþ02 6.415Eþ03 2.027Eþ04 2.251Eþ03
De Haver 1.533Eþ07 1.147Eþ07 2.969Eþ08 5.145Eþ03 7.619Eþ02 3.582Eþ03 3.456Eþ04 3.352Eþ03

mentioned before, these sets of buildings share the same grid foundation
plan and have similar masses.
For the complete set of calculations and results obtained for the
definition of the SSI macro-element model, interested readers are
referred to the report by Mosayk (2019) [25].
6. Fragility functions
6.1. Methodology
For the development of fragility functions, which describe the
probability of reaching or exceeding a given damage or collapse state
under increasing levels of ground shaking intensity, a model for the
probabilistic relationship between ground motion intensity and the
nonlinear structural response of the SDOF system is needed. The ap­
proaches that are commonly used for estimating this probabilistic rela­
tionship include the cloud method (Jalayer, 2003) [51], (Cornell et al.,
2002) [52], the multiple-stripe method (Jalayer, 2003) [51], (Jalayer
and Cornell, 2009) [53] and Incremental Dynamic Analysis (IDA)
(Vamvatsikos and Cornell, 2002) [54]. Hazard-consistent record selec­
tion together with linear regression (typically used in the cloud method)
has been used herein. Indeed, whilst the selection of records conditional
on increasing levels of intensity could allow the multiple-stripe method
to be used, whereby the probability of damage/collapse threshold ex­
ceedance at each intensity measure level is calculated from the response
data and then maximum likelihood is applied to fit a fragility function to
the results, this has not been undertaken herein as the largest selected
ground motions do not always lead to sufficient numbers of damage
exceedance/collapse for many of the vulnerability classes.
Hazard-compatible records for the development of fragility functions
were selected through disaggregation of seismic hazard at four different
return periods (Tr ¼ 500, 2500, 10 k and 100 k years) at one of the
highest hazard locations in the field. Using the mean magnitude and
distance from the disaggregation together with the 2017 ground motion
prediction equation for the Groningen field (Bommer et al., 2017) [22],
the records were then selected from a large database, including Euro­
pean (Akkar et al., 2014) [55] and NGA-West records (Chiou et al.,
2008) [56]. The records were selected to match spectra conditioned on
four different levels of AvgSa (corresponding to the four return periods),
namely, using the ground motion selection procedure proposed by Baker
and Lee (2018) [57]. Plots of the time-histories of the selected records
are given in Fig. 14, whilst the corresponding response spectra are
shown in Fig. 15.
It is noted that AvgSa was adopted as the intensity measure in this
study not only because it has been shown to be sufficient (Kohrangi
et al., 2017) [58], but also because, unlike e.g. spectral acceleration at
the period of vibration of the structure (which can also constitute a
sufficient intensity measure), it allows a comparison between the
fragility functions obtained for the different structural systems consid­
ered (each of which has a different period of vibration).
Once the maximum nonlinear dynamic displacement response of a
given SDOF (Sd) is obtained from all n ground-motion records, each
response (sd,i) is plotted against a scalar/vector intensity measure (ln
(AvgSa) herein) and the statistical parameters corresponding to a fitted
lognormal distribution of Sd | ln(AvgSa) can be extracted. In particular,
the expected value, E[ln Sd|ln(AvgSa)], is modelled by a linear regres­
sion equation (Equation (9)) with parameters b0 and b1, whilst the
standard deviation or dispersion (Equation (10)) is estimated by the
standard error of the regression:
E½lnSd jlnðAvgSaÞ� ¼ lnηSd jlnðAvgSaÞ ¼ b0 þ b1 lnðAvgSaÞ (9)

Fig. 14. Time-histories of the selected records.

12
F. Cavalieri et al.
Fig. 15. Spectra of selected records and the conditional spectra (herein rep­
resented with the mean and � 2σ ) to which they have been matched.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
i ln sd; i
βSd jlnðAvgSaÞ � (10)
As mentioned above, the parameters b0 and b1 are the estimated
regression coefficients obtained by performing a linear regression. In
order to correctly treat the results of the nonlinear dynamic analyses
where the displacement response exceeds the expected ultimate
displacement capacity, a censored regression has been undertaken when
estimating the coefficients of Equation (9) (see Stafford, 2008 [59]). In
these cases, the value of displacement demand from the nonlinear dy­
namic analysis is not trusted (as the collapse displacement capacity has
been exceeded and hence the estimated displacement response is no
longer reliable), but it is known to exceed a given limiting value, and is
thus referred to as a censored observation; if all censored observations
were set to the limiting value, and a normal linear regression analysis
were to be applied as above, the fitted model would be biased. To obtain
an unbiased model, maximum likelihood technique is used (refer e.g. to
Crowley et al., 2019 [4] for the corresponding formulae).
It is noted that the collapse displacement capacity has been estimated
through the nonlinear dynamic analysis of the MDOF systems intro­
duced in Section 2 above, and taken as the mean between the maximum
attic displacement obtained for the records where global collapse does
not occur and the lowest displacement at which global collapse was
instead identified.
An example cloud data plot with censored regression is shown in
Fig. 16, where the censored observations have been plotted at the
limiting displacement capacity value (their original non-censored
values, exceeding the collapse displacement, are indicated with red
markers).
6.2. Proposed fragility functions and comparison
The obtained fragility curves for the collapse limit state and for the
ten investigated index buildings are shown in Fig. 17. Each subplot
displays the curves related to: i) the simple one-dimensional elastic SSI
case, ii) the LPM elastic SSI case, and iii) the nonlinear macro-element
SSI case. The curve for the fixed-base case is also displayed for refer­
ence. It can be noted that for most of these buildings with shallow
foundations the influence of SSI is small to negligible, and leads the
curves to be marginally shifted to the right with respect to the fixed-base
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
Fig. 16. Example cloud data plot with linear censored regression of the dy­
namic displacement responses of the SSI þ SDOF system (note: each vertical
stripe corresponds, from left to right, to the results obtained using the Tr ¼ 500,
2500, 10 k and 100 k year records, respectively).
case: this means that SSI may have a beneficial effect on the seismic
vulnerability of these buildings.
�
lnηSd jlnðAvgSaÞ
�2 A more obvious SSI influence can be noted for the Kwelder case-
study alone, which, as can be gathered from Table 2, features not only
n 2
a low period of vibration (0.08 s), but also a relatively high ultimate
structural displacement capacity (0.118 m). As this building is stiff (and
ductile, hence not prone to premature collapse), the response of the
relatively weak soil inevitably plays a more determinant role in the
overall fragility of the system, and hence the impact of SSI modelling
becomes evident. Further, the impact of modelling explicitly soil
nonlinearity is also particularly evident for this stiff and ductile build­
ing, for which higher ground motion levels are required to reach dam­
age/collapse limit states, with the macro-element fragility curve being
clearly shifted to the right of the curves obtained with elastic SSI models.
This also confirms that, as expected, considering soil nonlinearity be­
comes even more relevant in cases where seismic action is high.
This shift due to explicit modelling of soil nonlinearity, visible also,
albeit to a lesser extent, in all other buildings but Zijlvest, Dijkstraat and
De Haver, confirms that the consideration of inelastic SSI behaviour
effectively leads to additional energy dissipation and, consequently, to
smaller structural displacements.
7. Conclusions and future work
In recent years, the Groningen region (northern Netherlands) has
been affected by induced seismicity attributed to gas production activ­
ity. Within the seismic hazard and risk model for the region developed
by the operators of the field (NAM), the definition of fragility functions
for several URM index buildings is crucial. With reference to ten of these
representative buildings with shallow foundations, this paper investi­
gated the impact on the collapse fragility functions of adopting different
SSI modelling approaches. Two of such SSI models, namely the one-
dimensional frequency-independent and the LPM, are elastic, whereas
the remaining one adopts a nonlinear macro-element to encompass all
aspects of elastic (in the far-field) and inelastic (in the near-field)
behaviour of the foundation system.
The influence of SSI resulted to be non-negligible only for stiffer
buildings, and in general leads to fragility curves that are less unfav­
ourable with respect to the fixed-base case. Moreover, the results
showed that taking into account the inelastic behaviour of the soil-
foundation system may lead to smaller structural displacements and
hence to a lower vulnerability of the buildings.
13
F. Cavalieri et al.
Fig. 17. Proposed fragility curves for the investigated index buildings and the different SSI models. The grey dashed lines indicate the four considered levels of
AvgSa; 0.2 g (Tr ¼ 500 years), 0.34 g (Tr ¼ 2500 years), 0.5 g (Tr ¼ 10 k years) and 0.86 g (Tr ¼ 100 k years).
The above demonstrates that, in order to avoid the introduction of
conservative bias in the results of risk assessment exercises, it may be
important not only to include SSI effects in the development of the
fragility functions of the building stock, but also to do so through the
employment of a nonlinear SSI model, even more so when ground mo­
tion intensity levels are high. The latter, in the context of fragility
functions derivation, where hundreds or thousands of nonlinear dy­
namic analyses need to be run, must necessarily be computationally
effective like the macro-element, given that more refined approaches
(involving e.g. the development of a 3D elasto-plastic soil-block model)
have a computational cost that renders them unfeasible for such
applications.
Funding
This work was undertaken within the framework of the research
programme for hazard and risk of induced seismicity in Groningen
sponsored by the Nederlandse Aardolie Maatschappij BV (NAM).
Declaration of competing interest
None.
CRediT authorship contribution statement
Francesco Cavalieri: Methodology, Software, Writing - original
draft, Writing - review & editing, Formal analysis, Data curation.
14
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
Anto� nio A. Correia: Conceptualization, Methodology, Software,
Writing - original draft, Writing - review & editing, Formal analysis.
Helen Crowley: Methodology, Software, Writing - original draft,
Writing - review & editing, Formal analysis, Data curation. Rui Pinho:
Conceptualization, Supervision, Methodology, Software, Writing -
original draft, Writing - review & editing, Formal analysis.
Acknowledgements
The authors are particularly grateful to Pauline Kruiver, who kindly
provided access to the soil mechanical characterisation data and site
response analysis results for the Groningen region. The constructive
feedback of three anonymous reviewers, which led to the improvement
of the original version of the manuscript, is also gratefully
acknowledged.
References
[1] Dost B, Edwards B, Bommer JJ. The relationship between M and ML—a review and
application to induced seismicity in the Groningen gas field, The Netherlands.
Seismol Res Lett 2018;89(3):1062–74.
[2] van Elk J, Bourne SJ, Oates S, Bommer JJ, Pinho R, Crowley H. A probabilistic
model to evaluate options for mitigating induced seismic risk. Earthq Spectra 2019;
35(2):537–64.
[3] Crowley H, Polidoro B, Pinho R, van Elk J. Framework for developing fragility and
consequence models for local personal risk. Earthq Spectra 2017;33(4):1325–45.
[4] Crowley H, Pinho R, Cavalieri F. Report on the v6 fragility and consequence models
for the Groningen field. NAM platform. http://www.nam.nl/feiten-en-cijfers/o
nderzoeksrapporten.html; 2019. March 2019.
F. Cavalieri et al.
[5] Khosravikia F, Mahsuli M, Ghannad MA. The effect of soil–structure interaction on
the seismic risk to buildings. Bull Earthq Eng 2018;16(9):3653–73.
[6] Pitilakis KD, Karapetrou ST, Fotopoulou SD. Consideration of aging and SSI effects
on seismic vulnerability assessment of RC buildings. Bull Earthq Eng 2014;12(4):
1755–76.
[7] Karapetrou ST, Fotopoulou SD, Pitilakis KD. Seismic vulnerability assessment of
high-rise non-ductile RC buildings considering soil–structure interaction effects.
Soil Dyn Earthq Eng 2015;73:42–57.
[8] Lesgidis N, Sextos A, Kwon OS. Influence of frequency-dependent soil–structure
interaction on the fragility of R/C bridges. Earthq Eng Struct Dyn 2017;46(1):
139–58.
[9] Bolisetti C, Whittaker AS, Coleman JL. Linear and nonlinear soil-structure
interaction analysis of buildings and safety-related nuclear structures. Soil Dyn
Earthq Eng 2018;107:218–33.
[10] LSTC – Livermore Software Technology Corporation. LS-DYNA—a general-purpose
finite element program capable of simulating complex problems. 2013
[Livermore].
[11] Lysmer J, Ostadan F, Chin C. Computer program SASSI2000 – a system for analysis
of soilstructure interaction. Berkeley, California: University of California; 1999.
[12] Petridis C, Pitilakis D. Soil-Structure Interaction effect on earthquake vulnerability
assessment of moment resisting frames: the role of the structure. In: Proceedings of
16th European conference on earthquake engineering (16ECEE), Thessaloniki,
Greece; 2018.
[13] Rajeev P, Tesfamariam S. Seismic fragilities of non-ductile reinforced concrete
frames with consideration of soil structure interaction. Soil Dyn Earthq Eng 2012;
40:78–86.
[14] Arup. EDB V5 data documentation, Report n. 229746_052.0_REP2014. NAM
Platform; 2017. http://www.nam.nl/feiten-en-cijfers/onderzoeksrapporten.html.
December 2017.
[15] Seismosoft. SeismoStruct 2020 – a computer program for static and dynamic
nonlinear analysis of framed structures. available at: http://www.seismosoft.com.
[Accessed 29 January 2020].
[16] Sivaselvan MV, Reinhorn AM. Hysteretic models for cyclic behavior of
deteriorating inelastic structures. Report MCEER-99-0018, MCEER, University of
Buffalo; 1999.
[17] Applied Science International, LLC. Extreme Loading for Structures - 2D & 3D
Nonlinear Static & Dynamic Structural Analysis Software. 2019 [ASI].
[18] Arup. Typology modelling: analysis results in support of fragility functions—2017
batch results, Report n. 229746_031.0_REP2005. NAM Platform; 2017.
http://www.nam.nl/feiten-en-cijfers/onderzoeksrapporten.html. November 2017.
[19] Crux Engineering. Appendix C BV Reports (229746/032.0/REP102). 2014. 28 Apr
2014.
[20] Arup. Soil-structure interaction for nonlinear static analysis - Groningen
earthquakes - structural upgrading. Report n. 229746_032.0_REP118. Amsterdam,
The Netherlands: Arup; 2015. February 2015.
[21] Rodriguez-Marek A, Kruiver PP, Meijers P, Bommer JJ, Dost B, van Elk J,
Doornhof D. A regional site-response model for the Groningen gas field. Bull
Seismol Soc Am 2017;107(5):2067–77.
[22] Bommer JJ, Edwards B, Kruiver PP, Rodriguez-Marek A, Stafford PJ, Dost B,
Ntinalexis M, Ruigrok E, Spetzler J. V5 ground-motion model for the Groningen
Field. NAM Platform; 2017. http://www.nam.nl/feiten-en-cijfers/onderzoeksr
apporten.html. October 2017.
[23] Kruiver PP, van Dedem E, Romijn R, de Lange G, Korff M, Stafleu J, Gunnink JL,
Rodriguez-Marek A, Bommer JJ, van Elk J, Doornhof D. An integrated shear-wave
velocity model for the Groningen gas field, The Netherlands. Bull Earthq Eng 2017;
15(9):3555–80.
[24] Fear CE, Robertson PK. Estimating the undrained strength of sand: a theoretical
framework. Can Geotech J 1995;32:859–70.
[25] Mosayk. Calibration and verification of a nonlinear macro-element for SSI analysis
in the Groningen region. 2019. http://www.nam.nl/feiten-en-cijfers/onderzoeksr
apporten.html. NAM Platform, April 2019.
[26] Mylonakis G, Nikolaou S, Gazetas G. Footings under seismic loading: analysis and
design issues with emphasis on bridge foundations. Soil Dyn Earthq Eng 2006;26
(9):824–53.
[27] Stewart JP. Variations between foundation-level and free-field earthquake ground
motions. Earthq Spectra 2000;10(2):511–32.
[28] Iovino M, Di Laora R, Rovithis E, de Sanctis L. The beneficial role of piles on the
seismic loading of structures. Earthq Spectra 2019;35(3):1141–62.
[29] Brandenberg SJ, Mylonakis G, Stewart JP. Kinematic framework for evaluating
seismic earth pressures on retaining walls. J Geotech Geoenviron Eng 2015;141(7).
04015031.
[30] Di Laora R. Discussion of “Kinematic framework for evaluating seismic earth
pressures on retaining walls” by Scott J. Brandenberg, George Mylonakis, and
Jonathan P. Stewart. J Geotech Geoenviron Eng 2016;142(8). 07016013.
15
Soil Dynamics and Earthquake Engineering xxx (xxxx) xxx
[31] Conti R, Morigi M, Rovithis E, Theodoulidis N, Karakostas C. Filtering action of
embedded massive foundations: new analytical expressions and evidence from 2
instrumented buildings. Earthq Eng Struct Dyn 2018;47(5):1229–49.
[32] Dezi F, Carbonari S, Leoni G. Kinematic bending moments in pile foundations. Soil
Dyn Earthq Eng 2010;30(3):119–32.
[33] Arup. Soil-structure interaction for linear analysis - Groningen earthquakes -
structural upgrading. Report n. 229746_032.0_REP102. 2015. Arup, Amsterdam,
The Netherlands, February 2015.
[34] GRC – Geotechnical Research Centre of Western Ontario University. DYNA6.1 – a
program for the computation of the response of rigid foundations to all types of
dynamic loads. 2015 [Ontario, Canada].
[35] Maravas A, Mylonakis G, Karabalis DL. Simplified discrete systems for dynamic
analysis of structures on footings and piles. Soil Dyn Earthq Eng 2014;61–62:
29–39.
[36] Bilotta E, Sanctis LD, Di Laora R, D’Onofrio A, Silvestri F. Importance of seismic
site response and soil–structure interaction in dynamic behaviour of a tall building.
Geotechnique 2015;65(5):391–400.
[37] Lesgidis N, Kwon OS, Sextos A. A time-domain seismic SSI analysis method for
inelastic bridge structures through the use of a frequency-dependent lumped
parameter model. Earthq Eng Struct Dyn 2015;44(13):2137–56.
[38] Dezi F, Carbonari S, Leoni G. A model for the 3D kinematic interaction analysis of
pile groups in layered soils. Earthq Eng Struct Dyn 2009;38(11):1281–305.
[39] Carbonari S, Dezi F, Leoni G. Linear soil–structure interaction of coupled
wall–frame structures on pile foundations. Soil Dyn Earthq Eng 2011;31(9):
1296–309.
[40] Carbonari S, Dezi F, Leoni G. Nonlinear seismic behaviour of wall-frame dual
systems accounting for soil-structure interaction. Earthq Eng Struct Dyn 2012;41
(12):1651–72.
[41] Carbonari S, Morici M, Dezi F, Leoni G. A lumped parameter model for time-
domain inertial soil-structure interaction analysis of structures on pile foundations.
Earthq Eng Struct Dyn 2018;47(11):2147–71.
[42] Correia AA. A pile-head macro-element approach to seismic design of monoshaft-
supported bridges. Ph. D. thesis. In: European school for advanced studies in
reduction of seismic risk. Pavia, Italy: ROSE School; 2011.
[43] Correia AA. Recent advances on macro-element modeling: shallow and deep
foundations. In: Proceedings of Final workshop of project Compatible soil and
structure yielding to improve system performance (CoSSY), Oakland, USA; 2013.
[44] Correia AA, Paolucci R. A 3D coupled nonlinear shallow foundation macro-element
for seismic soil-structure interaction analysis. Earthq Eng Struct Dyn 2019
[Unpublished results].
[45] Chatzigogos CT, Figini R, Pecker A, Salençon J. A macroelement formulation for
shallow foundations on cohesive and frictional soils. Int J Numer Anal Methods
Geomech 2011;35(8):902–31.
[46] Figini R, Paolucci R, Chatzigogos CT. A macro-element model for non-linear soil-
shallow foundation-structure interaction under seismic loads: theoretical
development and experimental validation on large scale tests. Earthq Eng Struct
Dyn 2012;41(3):475–93.
[47] Pianese G. Non-linear effects on the seismic response of buildings with foundation-
structure interaction. PhD thesis. Milan, Italy: Politecnico di Milano; 2018.
[48] Gazetas G. Foundation vibrations. In: Fang HY, editor. Foundations engineering
handbook. second ed. New York: Van Nostrand Reinholds; 1991. p. 553–93
[chapter 15].
[49] CEN – Comit� e Europ�een de Normalisation. Eurocode 7: geotechnical design - Part
1: general rules, ENV 1997:1-1994. 2004.
[50] Lancellotta R. Geotechnical engineering. CRC Press; 2008.
[51] Jalayer F. Direct probabilistic seismic analysis: implementing non-linear dynamic
assessments. Ph.D. Dissertation. Stanford University; 2003.
[52] Cornell CA, Jalayer F, Hamburger RO, Foutch DA. Probabilistic Basis for 2000 SAC
Federal Emergency Management Agency Steel Moment Frame Guidelines. Journal
of Structural Engineering 2002;128(4):526–33.
[53] Jalayer F, Cornell CA. Alternative non-linear demand estimation methods for
probability-based seismic assessments. Earthq Eng Struct Dyn 2009;38(8):951–72.
[54] Vamvatsikos D, Cornell CA. Incremental dynamic analysis. Earthq Eng Struct Dyn
2002;31(3):491–514.
[55] Akkar S, Sandikkaya MA, Senyurt M, Azari Sisi A, Ay BO, Traversa P, Douglas J,
Cotton F, Luzi L, Hernandez B, Godey S. Reference database for seismic ground-
motion in Europe (RESORCE). Bull Earthq Eng 2014;12:311–39.
[56] Chiou B, Darragh R, Gregor N, Silva W. NGA project strong-motion database.
Earthq Spectra 2008;24(1):23–44.
[57] Baker JW, Lee C. An improved algorithm for selecting ground motions to match a
conditional spectrum. J Earthq Eng 2018;22(4):708–23.
[58] Kohrangi M, Bazzurro P, Vamvatsikos D, Spillatura A. Conditional spectrum-based
ground motion record selection using average spectral acceleration. Earthq Eng
Struct Dyn 2017;46(10):1667–85.
[59] Stafford PJ. Conditional prediction of absolute durations. Bull Seismol Soc Am
2008;98(3):1588–94.