Bulletin of Earthquake Engineering (2021) 19:6497–6519

https://doi.org/10.1007/s10518-020-00923-y

S.I. : RECENT ADVANCES IN SEISMIC FRAGILITY

AND VULNERABILITY

Explicit modelling of collapse for Dutch unreinforced

masonry building typology fragility functions

Damian N. Grant1 · Jamie Dennis2 · Richard Sturt3 · Giovanni Milan2 ·

David McLennan3 · Pedro Negrette2 · Rene da Costa2 · Michele Palmieri2

Received: 28 April 2020 / Accepted: 30 July 2020 / Published online: 27 August 2020
© Springer Nature B.V. 2020

Abstract
The development of seismic fragility functions for buildings generally relies on simplified
modelling methods and the use of indirect engineering demand parameters (EDPs) for the
determination of collapse or other damage states. The collapse response of real buildings,
particularly those that have not been specifically designed for seismic resistance, can often
be driven by local failures that may not be captured in simplified models. Furthermore, the
use of EDP thresholds to indicate damage states may not be consistent with multiple pos-
sible failure modes, which may be triggered by different characteristics of the ground shak-
ing or variations in model parameters. This paper demonstrates the use of non-linear finite
element models including explicit progressive collapse simulation for the development of
fragility functions. It presents an overview of the method developed and its application to
an unreinforced masonry (URM) building typology in the Groningen region of the Nether-
lands, where induced seismicity risk is currently being evaluated. Multiple index buildings
were selected to represent the variations in geometry, material properties, and connection
types found within the typology. For each index building, Latin Hypercube sampling was
used to generate batches of several hundred realisations of an LS-DYNA time-history anal-
ysis, each selecting from a set of 100 hazard-consistent ground motions, and varying mate-
rial properties and other uncertain variables according to pre-assigned probability distribu-
tions. Automation was used in model generation, running analyses and in post-processing
to allow the required computation with minimal analyst intervention. The main output
from each analysis was a normalised debris cover estimate, which describes the extent of
damage observed in the model and is correlated with life safety risk. Regression analyses
were carried out directly on threshold levels of debris cover identified building collapse.
Fragility functions were developed for the URM terraced house typology by combining
results from the individual index buildings together.

Keywords Fragility functions · Unreinforced masonry buildings · Debris cover · Explicit

collapse assessment · LS-DYNA

* Damian N. Grant
damian.grant@arup.com
Extended author information available on the last page of the article

13
Vol.:(0123456789)
6498 Bulletin of Earthquake Engineering (2021) 19:6497–6519

1 Introduction

Fragility and vulnerability functions are used in probabilistic seismic risk analysis to assess
the likelihood and consequences of damage or collapse in buildings. Fragility functions
can be developed on the basis of empirical data, analytical modelling or expert judgement
(Rossetto et al. 2015). Analytical methods are commonly used, and, if well-calibrated
against real building performance or laboratory testing, can provide robust estimates of
building damage and collapse probabilities. This is particularly useful for building typolo-
gies for which sufficient previous damage data is not available—e.g., for tall buildings,
retrofitted buildings, or in cases of induced seismicity.
Analytical methods for evaluating fragility functions vary in the modelling methodology
used (from inelastic single-degree-of-freedom systems to detailed finite element analysis),
the way in which epistemic uncertainty and aleatoric variability are taken into account, the
method for applying seismic forces or ground motions to the model (e.g., pushover analy-
sis, or non-linear response history analysis (NLRHA)), and the regression techniques used
(see, for example, Baker 2015). Most modelling methodologies have not been calibrated
to model full collapse, and therefore it is common to introduce proxy engineering demand
parameters (EDPs; e.g., interstorey drifts), and to assume that exceeding a threshold value
of the EDP corresponds to reaching a particular damage state, such as collapse (e.g. Vam-
vatsikos and Cornell 2002; Jalayer and Cornell 2009; ATC 2009; Bakalis and Vamvatsikos
2018). This assumes that threshold EDPs can be well-established based on laboratory
results, and that they do not depend on the different failure modes that the building model
may experience in the analysis (Bakalis and Vamvatsikos 2018).
Sometimes in parallel with EDP-based collapse prediction, collapse results are treated
“explicitly”, based on non-convergence of an implicit time-stepping algorithm (e.g.
Jalayer and Cornell 2009; Bakalis and Vamvatsikos 2018). This has two problems for
risk assessment. Firstly, non-convergence can indicate instability associated with a single
component in the model, but in reality, forces may be redistributed without collapse, or
only partial collapse may occur. This is usually difficult to assess from a model that has
experienced convergence problems, and at the very least requires engineering judgement.
Secondly, this does not allow the full consequences of the collapse to be assessed–how
much of the building is damaged (for repair cost and downtime assessment) and how
much of the occupied floor area is impacted by falling debris. For these reasons, conse-
quence functions (or “damage to loss functions”), which measure the consequences of
building damage states in terms of financial and human losses, are generally based on
empirical data, and are not explicitly tied to the results from the analysis (Coburn et al.
1992; So and Pomonis 2012).
These issues with the use of proxy EDP thresholds and model non-convergence as indi-
cators of various damage states and collapse are perhaps of less concern for engineered
frame-type structures, where extensive research data is available to pair analytical results to
damage. However, for non-engineered unreinforced masonry (URM) houses, where multi-
ple failure modes can contribute to collapse response, code drift limits are often very con-
servative when compared to the variable results from laboratory testing (e.g. Salmanpour
et al. 2013; Wilding and Beyer 2018; Beyer et al. 2019), and load paths for redistribution
are not always obvious, an approach based on explicit prediction of damage and collapse
may be preferable.
This paper describes the explicit collapse fragility prediction of a URM build-
ing typology in the Groningen region in the Netherlands. This work was carried out to

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6499

validate the fragility functions that have been calculated using a more traditional (EDP-
based) approach that has been used for probabilistic seismic risk assessment of 35 build-
ing typologies in the Groningen region (see Crowley and Pinho 2020; NAM 2020). The
work described in the paper covers analyses conducted on five index building models,
considered to represent two typologies for terraced houses in the region. Latin hyper-
cube sampling was used to describe the epistemic uncertainty on material parameters,
geometric variations, and modelling choices, and aleatoric record-to-record variability.
Fragility functions were developed for each index building, which provides insight into
the main parameters contributing to the fragility. Then typology-wide fragility functions
were developed by combining weighted results from individual index buildings (Arup
2019d, e).

2 Project background

2.1 Induced seismic risk in Groningen gas field

A seismic risk assessment is being carried out for induced seismicity in the Groningen
region in the north of the Netherlands, to investigate the “local personal risk” for occu-
pied buildings (van Elk et al. 2019; NAM 2020). The risk assessment comprises an expo-
sure model for approximately 260,000 individual buildings in the area (Arup 2019a).
Buildings in the field are grouped into typologies, whereby buildings of a similar struc-
tural system are collected together on the grounds that their seismic behaviour should be
comparable. To determine a fragility function for a typology, one or more representa-
tive index buildings from the typology is selected. Fragility functions were developed for
each typology on the basis of a simplified modelling approach, calibrated on the results
of deterministic finite element models of the index buildings subjected to suites of up to
11 ground motions (Crowley and Pinho 2020; Arup 2017, 2019b, c). Modelling uncer-
tainty and building-to-building variability are added later based on results from litera-
ture, judgement and comparison of blind predictions to test results (Crowley and Pinho
2020).

2.2 URM terraced house building typologies

One typology that was found to be relatively vulnerable consists of terraced house build-
ings with URM cavity walls, reinforced concrete (RC) floors and large window openings
in the front façade, referred to as URM4L in the reporting on the risk study (NAM 2020).
“Large” window openings was defined as a façade openings percentage greater than or
equal to 90% (measured based on total width of window and door openings divided by
total façade width). A similar building typology, URM3L, distinguished from URM4L
only by its smaller window openings (< 90%), was also studied. Given the importance of
these typologies, particularly the more fragile URM4L, for the overall risk results (based
both on numbers of such buildings in the region, and its relative vulnerability), a validation
of the overall fragility approach was required. The method used in this validation study is
summarised in this paper.

13
6500 Bulletin of Earthquake Engineering (2021) 19:6497–6519

2.3 Representative index buildings

In the original risk study, the URM4L typology was represented by a single index build-
ing; in this work, this index building (referred to herein as URM4L-1) was studied
in the most detail to allow direct comparison with the previous work. The URM4L-1
model represents a real two-storey, terraced house, with masonry cavity walls, concrete
first floor one-way spanning onto end walls, timber attic floor and roof. Cavity walls
are formed of calcium silicate (CaSi) inner leaf and clay brick outer leaf. See Fig. 1
for details of the URM4L-1 building and LS-DYNA modelling (described further in
Sect. 3).
Four additional index buildings were analysed as part of this study, to explore the
building-to-building variability in fragility throughout the URM4L and URM3L typol-
ogies. They are all URM terraced house buildings with concrete floors; one of them
(URM4L-2) has front façade openings that put it into the URM4L typology, and the
other three have smaller openings, and therefore fall in URM3L (URM3L-1, URM3L-2,
URM3L-3). The same LS-DYNA modelling strategy was employed for each. The four
additional index building models are shown in Fig. 2, and relevant properties of all five
(including URM4L-1) are summarised in Table 1.
Note that all index buildings are real houses in the Groningen area, all of which had
previously been assessed by Arup as part of building assessment work that is currently
taking place. In typology-wide fragility studies, it is common to analyse archetype
buildings that are considered to be representative (e.g. Porter et al. 2014; D’Ayala et al.
2015), but may not capture the full range of idiosyncrasies of real building performance.
It was particularly important to analyse buildings with as much of their real-life detail
as possible for this work, as it had previously been found that failure modes and seismic
resistance of non-seismically-detailed URM buildings are sensitive to detailing and con-
nectivity of components. Furthermore, index buildings were selected based on availabil-
ity of building models and representativeness with respect to the typology. Given the
complexities of the URM building response, it is difficult to assign a priori index build-
ings based on (say) “poor”, “typical” and “good” resistance as advocated (for example)
by D’Ayala et al. (2015).

Fig.1  Index building model, URM4L-1 (corner of model hidden to show internal details)

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6501

Fig.2  Additional index building models: a URM4L-2; b URM3L-1; c URM3L-2; d URM3L-3 (corner of

models hidden to show internal details)

3 Finite element modelling approach

3.1 LS‑DYNA finite element package

Analyses were carried out in LS-DYNA®, a versatile three-dimensional non-linear finite

element analysis program used for seismic analysis among many other applications. The
program has strong capabilities for modelling components of buildings, soils and soil-
structure interaction and is optimised for fast solution of large, complex models on multi-
processor distributed memory computer platforms. The explicit time integration scheme
suits the analysis of brittle materials that may abruptly crack, soften or fail–in these situa-
tions implicit time integration schemes may have difficulties with convergence. The prin-
cipal limitation of the explicit scheme in this context is the small timestep size required for
numerical stability; however, in this study, computing times remained manageable even for
batches of several hundred analyses.

3.2 Masonry material model

A User Material Model (*MAT_SHELL_MASONRY) has been implemented for seismic

analysis of URM walls, and is described in detail by Sturt et al. (2018); hence only a brief
summary is provided here. The material model is used with a relatively coarse mesh of
shell elements representing the composite behaviour of the bricks and mortar together.
It considers the orientation of bed (horizontal) and head (vertical) mortar joints, and the
effect on overall stress–strain behaviour of the interlocking of units. The following modes
of response and failure are taken into account [diagrams are provided in figures 1 to 4 of
Sturt et al (2018)]: (1) cyclic bed-joint crack opening/closing and sliding; (2) non-linear

13
 6502

13
Table 1  Summary of index buildings studied
Label URM4L-1 URM4L-2 URM3L-1 URM3L-2 URM3L-3

Construction year 1976 1966 1971 1961 1963

Building mass per 94 114 119 101 90
unit (t)
Total number of units 2 2 3 3 3
Front dimension (m) 10.80 12.15 18.74 19.07 18.88
Side dimension (m) 7.90 8.74 7.76 7.58 6.12
Roof gutter height (m) 5.40 5.35 5.32 5.25 5.35
Gable height (m) 2.80 2.41 2.35 2.11 2.23
Opening percentage (%) 95 90 68 68 61
First floor system NeHoBo 2-Way RC slab 1-Way RC slab 2-Way RC slab 2-Way RC slab
Attic floor system Timber 2-Way RC slab 2-Way RC slab 1-Way RC slab 2-Way RC slab
Internal walls Gravity only Masonry walls in trans- Masonry walls in trans- Masonry walls in trans- Masonry walls in transverse
verse direction of ground verse direction of ground verse direction of ground direction of ground and
and first storeys; all and first storeys; all and first storeys; all first storeys; all others non-
others non-load bearing others non-load bearing others non-load bearing load bearing partitions
partitions partitions partitions
Roof framing Timber purlins + sheathing Timber purlins + planks Timber purlins + planks Timber purlins + planks Timber purlins + planks
Approx. average wall tie 3.5 3.0 4.0 3.5 2.0
density (ties/m2)
Other comments Original building has 4 Floor slabs and inner leaf
units; only 2 units mod- continuous between units;
elled based on symmetry solid party walls
Bulletin of Earthquake Engineering (2021) 19:6497–6519
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6503

compressive response and crushing; (3) head-joint opening combined with bed-joint slid-
ing (stepped diagonal cracks); (4) anisotropic shear response (shear on a horizontal plane is
more flexible and weaker than shear on a vertical plane due to interlocking units). Element
deletion is used (with deletion criteria calibrated against experiments) to simulate full col-
lapse response as these failure modes develop.
The LS-DYNA material model (and other details of the modelling approach, further
described in the following subsections) was calibrated against a number of laboratory tests
conducted primarily in the EUCENTRE in Pavia, Italy (Graziotti et al. 2018) and TU Delft
in the Netherlands (Messali et al. 2017). Tests included in-plane and out-of-plane tests on
wall specimens and full-scale shake table tests of URM houses including different types of
masonry (clay bricks and calcium silicate) and different construction types (such as cav-
ity wall and solid wall, concrete floor slabs and timber diaphragms). Further studies indi-
cated that results were not unduly sensitive to analysis parameters such as element size and
timestep, although, where collapse occurs, the physical behaviour becomes more chaotic
and therefore an exactly repeatable result is not expected.

3.3 Connection details

A number of element types are included in the model to represent the non-linear behaviour
and failure of the connections between the various building components. These include
mechanical connections such as the nailed joints between the purlins and sheathing of the
roof and attic floors, and the wall ties between cavity walls. Approximate average wall tie
densities for each index building model are provided in Table 1. Friction-based connections
represent bearing of timber elements onto masonry walls. These bearing connections can
simulate the unseating of the timber members once sliding greater than the seating distance
has occurred. Further information about connection modelling is summarised in Table 2.

3.4 Fixed base boundary condition

All results presented in this paper are for fixed base analyses (i.e. with ground motions
applied directly at the top of the foundation level of the building). Sensitivity analyses
incorporating full soil-structure interaction, including variability of soil properties, foun-
dation type and modelling, and connectivity of foundations into the building were also
carried out on the URM4L-1 building, and results were shown to be relatively insensitive
(although there were difficulties in applying consistent ground surface level accelerations
to the model to allow ready comparison). Refer to Arup (2019d) for more details.

3.5 Progressive collapse and debris cover measurement

In order to model collapse explicitly, together with the accumulation of debris inside and
outside the building footprint, the model captures the impacts of falling elements on the
floors beneath, with the potential for progressive collapse induced by these impacts. Poten-
tial for contact was modelled between outer leaf and inner leaf, between party walls, and
between each floor or ground surface and all elements above them (to allow for debris
accumulation and estimation).
The fatality consequence model used in the risk assessment (Crowley and Pinho 2020)
gives fatality rates (number of expected fatalities normalised by total number of occupants)

13
 Table 2  Summary of connection modelling
6504

Connection type Sketch Modelling approach

13
Roof beam pocket connection Spring-like elements which model restraint provided by edge
girder
of pocket, friction on underside of timber element, and
possibility for unseating (loss of support) for excessive
relative movement
pocket

inner leaf

Cavity wall tie connections Nonlinear axial-only springs with force–displacement rela-
tion taken from lab tests. Element removal occurs when
ultimate displacement capacity is exceeded

Floor slab to wall connection inner leaf

Fixed connection between wall and floor shell elements. Ini-
outer leaf tial bond strength and any post-cracking sliding behaviour
topping slab taken into account within masonry model of wall shells

floor panel

Party wall anchors between purlins (when present) roof tile Nonlinear tension-only springs. Element removal (erosion)
roof plinth
sheathing occurs when ultimate displacement capacity is exceeded

timber plank
nails
purlin steel anchor

masonry
party wall

Nailed connections between timber floor elements Connection modelled by beam elements which represent
the nonlinear shear response of a nailed joint under cyclic
timber floor boards
cut-out loading. Element removal occurs when ultimate displace-
ment capacity is exceeded
nails
floor joists
Bulletin of Earthquake Engineering (2021) 19:6497–6519
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6505

as a function of debris cover (total floor area impacted by debris normalised by total floor
area). An algorithm was developed to estimate debris cover directly from the LS-DYNA
models. The floor elements that fall or are struck by debris during the analysis are recorded,
together with any that are projected to be struck by debris that is still falling at the end of
the analysis. Dividing the area of these affected floor elements by the total floor area gives
the debris cover ratio.
Floor elements were also extended for 5 m outside the building footprint at ground floor
level to monitor debris falling externally to the building. This was used to validate the
model used in the risk assessment for exterior fatality risk. Exterior debris is not covered in
this paper.

4 Simulation and treatment of variability

Fragility analysis should consider all sources of uncertainty and variability that apply to
the analysis models used and their interpretation. This section describes the approach used
to consider this uncertainty and variability in the analyses.

4.1 Latin hypercube simulation

Latin hypercube sampling (LHS) was used to generate hundreds of combinations of build-
ing parameters and ground motions to include the effects of epistemic uncertainty and alea-
toric variability in the fragility assessment. LHS generates random samples of parameter
values for any number of random variables. Each random variable is assigned a probability
distribution (which may be a continuous- or discrete-valued distribution), and each cumu-
lative distribution function (CDF) is divided into a number of equiprobable bands equal
to the number of analyses to be carried out. The number of simulations required does not
depend on the number of variables. The combination of values for the variables is set up
such that every band of the CDF is sampled exactly once (in the case of discrete variables,
each value is sampled a number of times in proportion to its probability mass). See Vam-
vatsikos (2014) and Gokkaya et al. (2016) for applications of LHS for fragility function
development.
The variables considered for the LHS are briefly described in the following.

4.2 Staged introduction of variability

The scope of the study was intended to cover the variability expected across two entire
building typologies, which include rather diverse variants of URM terraced house with RC
floors. Different sources of variability were introduced in stages in order to understand the
effect of each source on the outcome, and to be able to compare intermediate results to cor-
responding fragility functions in the original risk assessment (Crowley and Pinho 2020). It
was also useful for practical management of the model development process and efficient
quality control of the analysis models.
Each stage of analysis (in terms of variability included) is described (and labelled) in
the following. The initial stage carried out during model development was:

• DetBldg Variable ground motions, deterministic building (i.e., all continuous building
parameters set to mean values and discrete parameters set to modal values).

13
6506 Bulletin of Earthquake Engineering (2021) 19:6497–6519

Each of the aspects mentioned (ground motions, building parameters, etc.) is described
in the following sections. The DetBldg stage of analysis was carried out for all buildings,
but only results for URM4L-1 are reported in this paper.
The baseline level of variability–for which analysis results are reported in this paper and
compared for all five index buildings–was:

• VarBldg-Index Variable ground motions; variable building parameters including spatial

variation (see Sect. 6.1). Variabilities in building parameters are intended to represent
“within-index-building” uncertainty–i.e., only variations that are consistent with the
actual observed or measured properties, connectivity or layout are included.

A further stage was carried out on the URM4L-1 building model, to quantify the effect of
other variations across the typology, while keeping the overall building footprint and layout
of components the same:

• VarBldg-Typ Same as VarBldg-Index, but incorporating further variations in assumed

material properties, geometry (such as gable height and window opening size; see
Sect. 6.3), roof/attic material combinations, and other miscellaneous details. Material
property variations separated into “within-building” and “between-building” terms in
the simulation, to properly account for a high correlation within a single building.

Other sensitivity studies were carried out on the URM4L-1 index building on removal of
within-building spatial variation of material properties (VarBldg-Index-NoSpatialVar), and
inclusion of soil-structure interaction (SSI) effects through explicit finite element model-
ling. The influence on results compared to VarBldg-Index was small, which supported the
use of the VarBldg-Index level of variability as a baseline. See Arup (2019a, b, c, d, e) for
further information.

5 Ground motion inputs

Record-to-record variability was captured with a suite of 100 ground motions total, in two
“stripes” of 50 ground motions, provided by consultants responsible for the risk assess-
ment (Crowley, pers. comm.). The ground motions were selected to be compatible with
the hazard in Loppersum (approximately the highest seismic hazard in Groningen, and the
location of the URM4L-1 building), conditioned on return periods of 10,000 years and
100,000 years, respectively for the two stripes. Ground motions were selected to represent
the conditional distribution of spectral ordinates and significant durations (based on 5% to
75% Arias Intensity) conditioned on the 0.5 s spectral acceleration (the original intensity
measure used in the risk analysis for this building typology) (Baker 2011; Bradley 2010).
The specific ground motion used for each simulation is also treated as a discrete random
variable; e.g., when 300 analyses are carried out, each of the 100 ground motions is used
three times, along with variations of the other modelling parameters.
Subsequent versions of the risk model used an “average spectral acceleration” measure
(Kohrangi et al. 2017); a new suite of ground motions was conditioned on this measure,
and the fragility functions were also developed for it. This change has not been incorpo-
rated into the results in this paper, but fragility functions for Sa(0.5 s), Sa(1.5 s) and aver-
age spectral acceleration are included in Arup (2019e).

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6507

6 Variability of materials, connections, geometry and layout

The following subsections describe the parameters that were varied (following either con-
tinuous or discrete probability distributions) across the simulations. A total of 235 continu-
ous variables and 25 discrete variables were simulated in the VarBldg-Typ analyses (includ-
ing many duplicates that were sampled independently to simulate spatial variation—see
Sect. 6.1). As previously stated, an advantage of the Latin Hypercube approach is that such
a large number of variables can be considered without requiring an astronomical number of
simulations.

6.1 Material properties, DetBldg and VarBldg‑Index stages

The DetBldg and VarBldg-Index stages used Dutch code, NPR1998 (NEN 2020), mean
properties for Calcium Silicate (CaSi) and brick masonry. This decision was made because
the work was being used to validate the simplified fragility procedure (referred to in
Sect. 2.2), and it was important that mean properties were consistent with those previously
adopted. For the VarBldg-Typ stage on the other hand, experimentally-derived mean mate-
rial properties were used. See Table 3 for mean properties for the main masonry strength
and stiffness parameters.
The URM4L-1 index building had been relatively well studied and photos and informa-
tion from the demolition of the house were also available. Masonry material testing had
been carried out both in situ before demolition (Graziotti et al. 2014), and in the labora-
tory using samples retrieved from the demolition (e.g., Braam and Jafari 2015). Extensive
testing had also been carried out on other buildings in the area. Therefore, coefficients of
variation for properties for all index buildings were informed–where possible–by material
testing specifically carried out on samples from the URM4L-1 house, as well as on similar
houses from the area, and the collective judgement of the analysis team. Most coefficients
of variation were in the range 0.2–0.35, although more extreme coefficients of variation
were used for cohesive shear strength of calcium silicate (0.8) and brick (0.4) masonry.
Continuous-valued parameters were assumed to be normally-distributed. To ensure that
negative or extremely low unrealistic values of material parameters did not unduly influ-
ence the results, continuous probability distributions were truncated at 5% of the mean. For
these samples, the value associated with 5% of the mean was used (i.e. the probability mass
of the tail of the distribution was lumped at this value–a rectified distribution rather than a
truncated distribution).
Correlation coefficients between related masonry material properties were assigned
based on judgement. For each masonry material (calcium silicate and clay brick), the
following correlations were assumed: strong correlation (correlation coefficient = 0.8)
between Young’s modulus and compressive strength, moderate correlation (0.5)
between compressive strength and tensile or shear cohesive strength, and low correla-
tion (0.3) between Young’s modulus and tensile or shear cohesive strength, and between
tensile and shear cohesive strength. Parameters relating to tensile or shear fracture
energy were determined deterministically from values of other failure parameters to
avoid non-physical stress−strain behaviour. NeHoBo floor strength properties were fully
correlated by simulating a “strength scaling factor” and deterministically scaling the
properties from this factor.

13
6508 Bulletin of Earthquake Engineering (2021) 19:6497–6519

The material testing showed relatively significant variation in material properties

for different locations in the house. To study the effect of spatial variation of material
properties on building response, a “patchwork” approach was used, where a patch size
(varying between approximately 0.3 m and 1.5 m) and pattern was randomly assigned
to the model and then five sets of material properties (including correlations between
properties) were simulated, and then randomly assigned to patches in the URM walls.
The approach was also used to spatially vary the wall tie and nailed timber connection
properties. Initial results from this patchwork model (see Arup 2019d) showed a lack of
sensitivity to material spatial variation, and therefore it was not considered necessary
to take into account complicated effects that the construction sequence and inconsistent
weathering and degradation over the life of the building may have on the actual distribu-
tion of material properties within the building.
Other building parameters related to modelling assumptions. These were informed
based on expert judgement of Groningen-based engineers with experience in existing
building inspections and assessment, and in some cases based on lessons learned in the
calibration of the LS-DYNA modelling approach with laboratory testing.
A summary of the building parameters varied in the analyses is given in Table 4. See
Appendix A of Arup (2019d) for more details.

6.2 Material properties, VarBldg‑Typ stage

In the VarBldg-Typ stage of work, mean properties and coefficients of variation based on
the collected data were used in order to be more representative of the whole typology.
Mean masonry properties are summarised in Table 3. When sampling across the range
of parameters found in the whole typology, the overall variability for a given patch of
masonry is higher than the variability modelled for the single building in the VarBldg-
Index stage, but there is correlation amongst samples within a given building (i.e. some
houses have higher mean masonry strength than other houses). Therefore, there is
a “between-building variability” term that was sampled once per building in the Latin
Hypercube, and a “within-building variability” that was sampled separately for each patch.
Overall coefficients of variation were increased to account for this additional
between-variability term. Values were still typically in the range 0.2–0.35, although
more extreme coefficients of variation were used for tensile and cohesive shear strength
of calcium silicate (0.5 and 1.1, respectively) and brick (0.8 and 0.6, respectively)
masonry. See Appendix A of Arup (2019e) for more details.

Table 3  Mean masonry material properties used in analysis stages

DetBldg and VarBldg-Index VarBldg-Typ
Calcium silicate Clay Calcium silicate Clay

Young’s modulus (GPa) 3.5 6.0 7.2 7.8

Compressive strength (MPa) 7 10 10.4 15.8
Tensile strength (kPa) 100 200 177 230
Shear strength (kPa) 250 400 262 460
Shear friction coefficient 0.60 0.75 0.79 0.80

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6509

Table 4  Superstructure parameters varied

Inner leaf (CaSi)/outer leaf (clay) masonry Connections

Density, Young’s modulus and Poisson’s ratio
Compressive strength and fracture release energy
Tensile strength and fracture release energy
Shear strength and fracture release energy
Shear friction coefficient
Diagonal tensile strength
Failure modelling parameters
Degree of interlock between perpendicular ­walls*
Complete / incomplete fill of mortar joints
Concrete type floors
Wall tie peak tensile strength
Wall tie peak compressive strength
Wall tie tensile failure displacement
Wall tie compressive failure displacement
Nailed connection stiffness
Nailed connection shear strength
Nailed connection shear strain at failure
Timber–masonry overlap dimensions
Timber–masonry friction coefficient
Timber–masonry mortar bond strength
Timber–masonry pocket rotational stiffness
Timber floors/roof

NeHoBo† masonry tensile strength Timber beam strength

NeHoBo concrete tensile strength Timber beam Young’s modulus
NeHoBo reinforcement tensile strength Plywood sheathing yield strength
NeHoBo floor % reinforcement Plywood sheathing Young’s modulus
Kwaaitaal‡ floor % reinforcement

*Interlock modelled as either “strong” (fully composite connection at wall edges) or “weak” (equivalent to
a vertically-oriented bed joint). This was simulated independently in each “patch”, which meant that any
individual wall joint could be fully strong, fully weak, or a mix of the two
†
A composite masonry– reinforced concrete floor system found in the Netherlands
‡
A prefabricated reinforced concrete floor system found in the Netherlands

6.3 Geometric variations

The geometry of the studied index buildings was well defined, based on building inspec-
tions and in situ measurements. Therefore, geometric variations were not introduced in the
VarBldg-Index stage, which was intended to give only plausible simulations of the actual
represented buildings. In the VarBldg-Typ stage, geometric and layout variations were
introduced. (As noted in Sect. 4.2, the VarBldg-Typ stage was applied to URM4L-1 only).
The openings in the façade had previously been used as an indicator of potential vul-
nerability, as many modern terraced houses in Groningen have very large window open-
ings, and very little structural wall to resist lateral forces. In fact, the URM4L typology
was distinguished from the similar URM3L typology by the large openings in the façade.
The original URM4L-1 index building had a façade openings ratio of 95% (based on the
length of window divided by length of façade–a linear rather than areal measure of opening
sizes). Two other openings percentages were included in the LHS simulation: 62% open-
ings and 75% openings (discrete values based on the ease of incorporating into the existing
finite element mesh; see Fig. 3a). Discrete probabilities were assigned to these three values
based on the approximate populations within the full URM3L and URM4L data sets from
the Exposure Database (EDB) developed for the project (Arup 2019a).
The gable height is also very relevant for the out-of-plane stability of the gable. Again,
discrete values of gable height were sampled (1.1 m, 2.2 m, 2.4 m, 2.8 m, 3.6 m and

13
6510 Bulletin of Earthquake Engineering (2021) 19:6497–6519

Fig.3  URM4L-1 VarBldg-Typ model variations. a Openings percentage–red shells only included in 62%
and 75% openings models, orange shells only included in 62% model; b and c lower and upper bound gable
heights (1.1 m and 5.0 m, compared to 2.8 m in baseline model); d and e FE models for sheathing roof with
timber attic, and planks roof with concrete attic; f ground floor internal walls shown in blue

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6511

5.0 m), with probabilities based on URM3L/URM4L data from the EDB. The weighted
mean gable height was 2.8 m, which matches the value from the original baseline model
for URM4L-1. See Fig. 3b, c for lower and upper bound gable heights, respectively.

6.4 Roof/attic variations

The VarBldg-Typ stage also introduced variations in the attic floor and roof diaphragms. The
original URM4L-1 (and the actual house it was based on) had an attic floor and roof compris-
ing timber purlins/joists and timber sheathing. When data for URM3L and URM4L typolo-
gies from the EDB were collected and assessed, it became apparent that this was a relatively
rare system for buildings of this type. The most common combination was a concrete attic
floor and timber plank roof, and other common combinations included plank roof with timber
attic floor, and sheathed roof with concrete attic floor. See Fig. 3d, e for two of the combina-
tions. Each of these four combinations was included in the LHS, with probabilities assigned
based on the EDB data (the original URM4L-1 roof type was assigned a probability of only
1%, whereas the most common combination was assigned to 78% of simulations).

6.5 Miscellaneous details

The following other miscellaneous aspects of the URM4L-1 were also varied at the Var-
Bldg-Typ stage.

• Internal walls The URM4L-1 did not have internal lateral-load resisting walls, although
these had been observed in many buildings assigned to this typology (see Table 1).
Specific data on this aspect was not available in the EDB; for lack of other information,
lateral-load resisting internal walls were included in 50% of simulations. See Fig. 3f for
ground-storey internal walls added to the baseline URM4L-1 model.
• Party wall anchors The URM4L-1 did not have party wall anchors between roof pur-
lins, although these had been observed in many buildings assigned to this typology. For
the same reasons as above, the two options (with and without wall anchors) were each
assigned to 50% of simulations. See Table 2 for details of typical party wall anchors
used in the region.
• Corrosion of wall ties Wall ties connecting leaves in cavity walls have commonly been
observed to be corroded in building assessments in Groningen. In the finite element
modelling, corroded wall ties were assumed to resist no force. Based on the judgement
of experienced Dutch engineers, a probability of corrosion of 15% was assumed. This
was included within the spatial variation simulation; i.e. it was sampled individually for
each patch, such that most simulations had some amount of corrosion, but none were
fully corroded.

7 Results and fragility assessment

7.1 Summary of analyses and results

Based on initial studies, it was found that around 100 analyses were required to describe the
record-to-record variation in the fragility on the DetBldg model stage. For VarBldg-Index,

13
6512 Bulletin of Earthquake Engineering (2021) 19:6497–6519

around 300 analyses were required to give a stable fragility function (i.e., each ground
motion was used three times within the overall Latin Hypercube approach for varying
model parameters). For VarBldg-Typ, around 600 analyses were required. The variation in
number of analyses required reflects the amount of variability in model inputs, and there-
fore the number of simulations required to sample across this uncertainty.
Debris cover estimates were generated for each analysis using the automated pro-
cedure described previously. Debris cover for the index building models was typically
binary–either less than 5% debris for non-collapsing models or greater than 90% debris
for collapsing models (i.e. very limited failure or complete collapse). Collapse was gener-
ally associated with excessive soft-storey drift in the ground floor (due to the large win-
dow openings), for attic floor displacements (relative to the ground) beyond about 100 mm.
Figure 4 shows the development of a typical collapse mechanism in the URM4L-1 index
building. For the results presented subsequently, debris cover of greater than 90% is
reported as “collapse”; most full collapses give debris estimates close to 100%, but the
90% threshold was used to pick up a few cases where small parts of the building footprint
were not impacted by debris. This decision is specific to the buildings studied, and inter-
mediate partial collapse states may be required for other buildings.

7.2 Fragility assessment

7.2.1 Univariate regression of fragility functions

Maximum Likelihood regression was carried out to develop collapse fragility functions
for the models (e.g., Baker 2015). A lognormal CDF was used. Fragility functions were
developed for intensity measures of the spectral acceleration at 0.5 s (based on the original
risk analysis) and 1.5 s (which was found to be a more efficient intensity measure during
this study, due to the extreme softening response required to drive collapse). Since 1.5 s
spectral ordinates vary in each stripe (as the conditional spectrum was conditioned only on
0.5 s ordinates), each analysis was treated separately with a Bernoulli likelihood function,
in contrast to the Binomial likelihood function typically used on grouped data such as those
in Fig. 5.

7.2.2 Multivariate fragility functions

The large pool of analysis results with parametric variations also allowed explicit regres-
sion analysis to be carried out for fragility functions in terms of both spectral ordinates
(as above) and secondary hazard metrics (e.g., ground motion significant duration) or

Fig.4  Typical development of collapse in URM4L-1 model (red–impacted by debris)

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6513

(a) (b)
Fig.5  a Fragility functions from each analysis phase in terms of Sa(1.5s); b fragility functions from fixed
base analyses in terms of Sa(1.5s) and masonry inner leaf compression strength

model parameters (e.g., material strengths). A fragility function of the following form was
considered:
( )
c + ln (Sa) + c1 ln(x)
p(collapse) = Φ 0 (1)
𝛽

where Sa is the spectral acceleration metric, x is a secondary hazard or model parameter,

β is the logarithmic standard deviation of the fragility function, Φ() is the standard normal
cumulative distribution function, and c0 and c1 are coefficients determined using Maximum
Likelihood regression analysis. For discrete x parameters, ln(x) in Eq. (1) is replaced with
x.
The functional form of Eq. (1) is relatively limited, in that it assumes that any additional
parameter (beyond the spectral acceleration) has a linear effect on the mean fragility (and
only one additional parameter was included at a time). More complex functional forms,
including multiple parameters and possibly interactions between parameters would prob-
ably be needed to give robust predictions of the effect of parameters.

7.3 Comparison of results

7.3.1 Effect of including model uncertainty (URM4L‑1: DetBldg to VarBldg‑Index)

Results of univariate regression analysis for fragility functions in terms of 1.5 s spectral
acceleration on the URM4L-1 model are shown in Fig. 5a. Results for the three stages of
variability outlined in Sect. 4.2 are shown. Debris cover data from the 100 DetBldg-FB
analyses are also shown on the same axes (from 0 to 100% cover).
Introducing model variability in the URM4L-1 building (from DetBldg to VarBldg-
Index) is seen to increase the building fragility (i.e. decrease the median of the fragility
function, moving it to the left). This appears to be due to a combination of the following
effects: (1) some discrete variations introduced into the variable building model are biased
towards increasing building vulnerability (e.g., introducing imperfect brick interlock at
connections between orthogonal walls); (2) introducing variability means that some com-
ponents will be weaker and some stronger than the deterministic model, and collapse may

13
6514 Bulletin of Earthquake Engineering (2021) 19:6497–6519

be triggered by failure of the weaker components. This effect was consistently observed
also with the other index buildings analysed for this study (Arup 2019d, e) and has also
previously been observed in the research literature in other fragility development applica-
tions (e.g. Vamvatsikos 2014; Liel et al. 2009; Gokkaya et al. 2016).

7.3.2 Effect of additional model variations (URM4L‑1: VarBldg‑Index to VarBldg‑Typ)

Adding further material, geometric, connectivity and other miscellaneous variations to the
URM4L-1 model (VarBldg-Index to VarBldg-Typ; see Sects. 6.2 to 6.5) has the following
main effects (Fig. 5a):

• Fragility is decreased (the median of the fragility function is increased) for the results
with additional variations added. This may be at least partly due to the fact that mate-
rial properties were generally stronger, as described in Sect. 6.2. Other model varia-
tions would also be expected to affect the fragility, although sometimes the effect is not
self-evident.
• The variability (β value) is increased, reflecting the more variable range of buildings
modelled. This increase in β is used to inform typology-wide fragility function develop-
ment in Sect. 7.4.

Although not shown here, the debris data are significantly less binary than those for
DetBldg and VarBldg-Index–more intermediate values of debris cover between 0 and 100%
are observed. Modifications introduced at this stage sometimes changed the failure mecha-
nism from the full building collapse shown in Fig. 4 to one of several observed partial
mechanisms, including single-unit collapses and front façade collapse without loss of sup-
port of floors, giving these intermediate values of debris cover. Although the same 90%
debris cover threshold was used here for comparison with the previous results, other partial
collapse states could be introduced with fragility functions evaluated with respect to lower
debris thresholds. Alternatively, a continuous vulnerability model that does not require the
identification of discrete damage states could be fitted to the data (Grant, 2020).

7.3.3 Multivariate regression results (URM4L‑1 only)

Multivariate regression was carried out on the URM4L-1 index building. Statistical signifi-
cance of additional parameters was explored using a likelihood ratio test, at the 5% signifi-
cance level. The additional parameters identified in Table 5 (considered individually, and
not all together) were found to have a statistically significant effect on the fragility function
(based on 1.5 s spectral acceleration). The VarBldg-Index-NoSpatialVar model was used,

Table 5  Superstructure parameters of URM4L-1 index building found to be statistically significant (at
p = 5% level) when added to Sa (1.5 s) in Eq. (1)
Parameter

Masonry, inner leaf, compression strength
Masonry, inner leaf, shear strength
Masonry, inner leaf, tensile displacement at failure
Masonry, inner leaf, diagonal tension strength
Masonry, inner leaf, Young’s modulus
Initial bond strength between timber and masonry
Masonry, inner leaf, compression fracture energy
Masonry, inner leaf, tensile strength
Masonry, outer leaf, Young’s modulus

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6515

as it provided more direct assessments of the effect of material properties (the baseline
model incorporating spatial variation had variable properties throughout the model).
This only holds with respect to the fragility function form in Eq. (1)—e.g., other param-
eters may be significant, but may have a non-linear effect on the fragility. Also note that
correlation between masonry material parameters was modelled in the Latin Hypercube
sampling, so some of the dependency shown in Table 5 may be due to correlation with
one of the other parameters rather than indicating a causal effect on the fragility. Figure 5b
shows an example fragility curve for the mean value of masonry inner leaf compression
strength, and the 16th and 84th percentile values from the set of simulations.

7.3.4 Comparison across index buildings (all models; VarBldg‑Index results)

Fragility functions developed for each of the five index buildings (VarBldg-Index stage)
are shown in Fig. 6a. As before, the Sa(1.5 s) intensity measure is used, and full collapse
is based on debris cover exceeding 90%. The main conclusions from this comparison are:

• The original index building, URM4L-1, is the most fragile of the index buildings stud-
ied, followed by URM3L-2 (which is only marginally more fragile than URM3L-1).
• URM3L-3 is significantly less fragile than the other buildings, which motivated a series
of sensitivity studies. As noted in Table 1, URM3L-3 is unusual (with respect to the
other index buildings) in that floor slabs and inner leaf of façade walls are continuous
between house units, and party walls are solid (not cavity walls). The building also had
the lowest openings percentage. The sensitivity studies concluded that each of these
factors contributed to the extra resilience of URM3L-3.
• URM4L-2 is the second-least fragile of those studied; this result is unexpected because
the URM4L typology is considered to be more fragile than URM3L in the risk assess-
ment. This is partly the result of selecting 90% debris cover as a collapse threshold–
unique amongst the index buildings studied, URM4L-2 analyses showed several partial
collapses, with one unit collapsing and the other remaining intact (approximately 50%
debris cover). Although the units were almost identical, occupant modifications (widen-
ing of internal openings), common to buildings in the area, led to a vulnerability in one
unit that was not present in the other. Fragility functions developed for a lower debris
threshold (50%) showed that URM4L-2 was actually of a similar fragility to URM4L-1.
This shows the importance of considering multiple collapse states (as in Crowley and

(a) (b)
Fig.6  Fragility function results using Sa(1.5 s) intensity measure based on debris cover > 90%. a Compari-
son of all index buildings, and b typology-wide functions

13
6516 Bulletin of Earthquake Engineering (2021) 19:6497–6519

Pinho 2020) and also shows the value in considering multiple index buildings to under-
stand typology-wide fragility.

7.4 Combined typology‑wide fragility functions

7.4.1 Methodology

Typology-wide fragility functions were developed for URM3L and URM4L typologies
separately, as well as a combined function representing both typologies together, based
on the relative frequencies of buildings in each typology in the exposure database, and an
extra component of variability to take into account the results reported in Sect. 7.3.2.
The methodology for combining index building results into typology-wide fragility
functions was the following:

• For the individual typologies, each index building was given a uniform weighting (0.5
for the two URM4L index buildings; 0.333 for the three URM3L index buildings).
• For the combined URM3L/4L fragility function, each URM3L building was assigned
0.333 × 0.234 and each URM4L building is assigned 0.5 × 0.766, based on relative fre-
quencies of each typology in the EDB.
• A combined fragility function was estimated for each intensity measure value as the
sum of the weighted probabilities of collapse from each constituent fragility function.
This combination of multiple index building results accounts for the differences intro-
duced by layout, topology and building footprint.
• An additional uncertainty (βextra) term was added with a square root sum of squares
(SRSS) combination to represent the extra uncertainty due to typology-wide geometric
and material variations. The additional term represents the variability introduced by
geometric and structural system uncertainties, not included in the pool of index build-
ings. Its value was estimated partly based on the difference in the URM4L-1 results
shown in Fig. 5(a) (i.e. the difference between VarBldg-Index and VarBldg-Typ stages),
and was taken as 0.3. (Note that this approach is equivalent to adding the additional
uncertainty term to individual index building results and then combining; this was veri-
fied numerically).
• This gives a mixed formulation fragility function (i.e. one that does not follow a typical
lognormal cumulative distribution function (CDF), which is required for use in both
typical risk assessment software and in the specific risk calculation engine developed
for this project). Therefore, a lognormal CDF was estimated using least squares regres-
sion, including only up to the 50th percentile results on the fragility functions. Focusing
on the lower intensity/lower probability of collapse values in the regression ensured
that the lognormal function is appropriate for the range of intensities driving the risk
in the risk assessment model. The misfit between the mixed formulation and lognormal
CDF fragility functions was verified to be small in the region included in the regression
(i.e. up to the 50th percentile).

7.4.2 Results

Results for the typology-wide fragility functions are shown in Fig. 6b. As could be
expected from the methodology outlined in Sect. 7.4.1, the fragility functions have medi-
ans that are weighted averages of their constituents, and standard deviations higher than

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6517

then individual index building results, accounting for both the extra variability (βextra) and
the differences between the medians. The median of the combined URM3L/4L fragility
function is significantly skewed towards the results for the URM4L typology, due to the
higher weighting factor assigned based on relative numbers in the EDB.

8 Conclusions

In this paper, an application of the development of fragility functions for a whole typology
of URM buildings from analytical models of individual index building was summarised.
The analytical models incorporated explicit modelling of collapse, including the potential
for progressive collapse when falling components impact on the rest of the structure. This
also enabled an explicit and automated estimation of the debris cover inside the collapsing
building models for use in casualty assessment. Modelling and building parameters were
varied in the analysis using a Latin Hypercube approach.
To estimate the fragility of an entire typology, the extra variability (beyond that of a
single index building) was included in the study by decomposing into two components: (1)
variability that can be expressed as modifications to the original index building (including
gable heights, attic/roof system, openings percentage and a few other miscellaneous mod-
elling assumptions); (2) variability related to the overall layout, topology and connectiv-
ity which is not amenable to a parametric approach. The first component was investigated
by comparing results on the same index model incorporating different levels of variability
in the Latin Hypercube simulations. The second component was included by combining
results from the original index building with those from analyses of four additional index
buildings.
Finally, typology-wide fragility functions were developed based on weighted averages
of the results from individual index buildings, with additional uncertainty introduced to
reflect the parametric variations described above. Due to the availability of appropriate
analysis models for the fragility assessment, separate fragility functions were developed
for URM3L and URM4L typologies (both representing terraced houses with cavity walls
and concrete floors, but differing in terms of the façade openings percentage), as well as a
single fragility function representing the combined typology.
Main findings from the fragility analyses have been incorporated into the most recent
induced seismicity seismic risk assessment:

• Fragility assessment incorporating building model variability (on material properties

and uncertain connection details) led to both an increase in dispersion and an increase
in fragility (a decrease in the median of the fragility function).
• Fragility functions developed to represent whole typologies gave median fragilities that
were essentially weighted averages of those of the index buildings considered and sig-
nificantly increased dispersion.

Main findings from the fragility analyses have been incorporated into the most recent
induced seismicity seismic risk assessment (Crowley and Pinho 2020; NAM 2020).

Acknowledgements We would like to thank our client, Nederlandse Aardolie Maatschappij (NAM), for
their support in this study and for the permission to publish this paper. We would also like to acknowledge
the contributions made by Jeroen Uilenreef, Helen Crowley, Rui Pinho, Rinke Kluwer and Arup colleagues

13
6518 Bulletin of Earthquake Engineering (2021) 19:6497–6519

in regular technical conversations held during the course of this project. Finally, we would also like to thank
Jack Baker and one anonymous reviewer who provided very useful review comments.

References
ATC (Applied Technology Council) (2009) Quantification of building seismic performance factors, FEMA
P695, Federal Emergency Management Agency (FEMA), Washington DC, USA
Arup (2017) Typology modelling: Analysis results in support of fragility functions—2017 batch results,
229746_031.0_REP2005, NAM Platform
Arup (2019a) EDB v7 Cover Note, NAM Platform
Arup (2019b) Typology modelling: analysis results in support of fragility functions—2018–2019 batch
results, NAM Platform
Arup (2019c) Typology modelling: analysis results in support of fragility functions—2019 batch results,
NAM Platform
Arup (2019d) Typology modelling: explicit MDOF validation of fragility functions—URM3L/URM4L,
NAM Platform
Arup (2019e) Typology modelling: explicit MDOF validation of fragility functions—update, NAM Platform
Bakalis K, Vamvatsikos D (2018) Seismic fragility functions via nonlinear response history analysis. J
Struct Eng 144(10):04018181
Baker JW (2011) The conditional mean spectrum: a tool for ground motion selection. J Struct Eng
137(3):322–331
Baker JW (2015) Efficient analytical fragility function fitting using dynamic structural analysis. Earthq
Spectra 31(1):579–599
Beyer K, Wilding B and Rezaie A (2019) Drift capacity models for modern URM walls: background docu-
ment for the masonry chapter in EC8 Part 1, Report No. CEN/TC250/SC8N. 911, European Commit-
tee for Standardization (CEN), Brussels, Belgium
Braam CR, Jafari S (2015), Zijlvest house masonry: compressive strength, Research Report 25.5–15–01,
Delft University of Technology
Bradley BA (2010) A generalized conditional intensity measure approach and holistic ground-motion selec-
tion. Earthq Eng Struct Dynam 39(12):1321–1341
Coburn AW, Spence RJS, Pomonis A (1992), Factors determining casualty levels in earthquakes: mortality
prediction in building collapse, In: Proceedings of the 10th World conference on earthquake engineer-
ing, Madrid, Spain
Crowley H, Pinho R (2020) Report on the fragility and consequence models for the Groningen Field (Ver-
sion 7). NAM Platform, NAM
D’Ayala D, Meslem A, Vamvastikos D, Porter K, Rossetto T and Silva V (2015) Guidelines for analytical
vulnerability assessment of low/mid-rise buildings—methodology. Vulnerability Global Component
project. GEM Technical Report, Global Earthquake Model, Pavia, Italy
Gokkaya BU, Baker JW, Deierlein GG (2016) Quantifying the impacts of modeling uncertainties on the
seismic drift demands and collapse risk of buildings with implications on seismic design checks.
Earthq Eng Struct Dyn 45(10):1661–1683
Grant DN (2020) A mathematical form of probabilistic vulnerability model for loss and casualty ratios.
Earthq Spectra 36(2):700–717
Graziotti F, Penna A, Magenes G (2018) A comprehensive in situ and laboratory testing programme sup-
porting seismic risk analysis of URM buildings subjected to induced earthquakes. Bull Earthq Eng.
https​://doi.org/10.1007/s1051​8-018-0478-6
Graziotti F, Senaldi I, Peloso S, Rossi A (2014), Results of in-situ tests—building unit: Loppersum, Zijlvest
25, NAM Platform, EUCENTRE
Kohrangi M, Bazzurro P, Vamvatskios D, Spillatura A (2017) Conditional spectrum-based ground motion
record selection using average spectral acceleration. Earthq Eng Struct Dyn 46(10):1667–1685
Jalayer F, Cornell CA (2009) Alternative nonlinear demand estimation methods for probability-based seis-
mic assessments. Earthq Eng Struct Dyn 38(8):951–972
Liel AB, Haselton CB, Deierlein GG, Baker JW (2009) Incorporating modeling uncertainties in the assess-
ment of seismic collapse risk of buildings. Struct Saf 31(2):197–211
Messali F, Ravenshorst G, Esposito R, Rots J (2017) Large-scale testing program for the seismic characteri-
zation of Dutch masonry walls. In: 16th World conference on earthquake engineering, Santiago, Chile
NAM (2020) Seismic hazard and risk assessment groningen field update for production profile GTS—Ram-
ing 2020. In: van Elk J, Landman AJ, Uilenreef J, Doornhof D (eds) NAM platform, NAM

13
Bulletin of Earthquake Engineering (2021) 19:6497–6519 6519

NEN (2020) NPR 9998: Assessment of structural safety of buildings in case of erection, reconstruction and
disapproval—induced earthquakes—basis of design, actions and resistances
Porter K, Farokhnia K, Vamvatsikos D, Cho IH (2014) Guidelines for component-based analytical vulner-
ability assessment of buildings and nonstructural elements. GEM Technical Report 2014-13
Rossetto T, Ioannou I, Grant DN (2015) Existing empirical fragility and vulnerability relationships: com-
pendium and guide for selection. GEM Technical Report 2015-01 V1.0.0, 10.13117/GEM.VULN-
MOD.TR2015.01
Salmanpour AH, Mojsilovic N, Schwartz J (2013) Deformation capacity of unreinforced masonry
walls subjected to in-plane loading: a state-of-the-art review. Int J Adv Struct Eng. https​://doi.
org/10.1186/2008-6695-5-22
So EKM, Pomonis A (2012) Derivation of globally applicable casualty rates for use in earthquake loss
estimation models. In: Proceedings of the 15th World conference on earthquake engineering, Lisbon,
Portugal
Sturt R, Avanes C, Muriithi B, Bernardi M, Huang Y (2018) A masonry material model for seismic analysis
in LS-DYNA: implementation and validation. In: 16th European conference on earthquake engineer-
ing, Thessaloniki, Greece
Vamvatsikos D, Cornell CA (2002) Incremental dynamic analysis. Earthq Eng Struct Dyn 31(3):491–514
Vamvatsikos D (2014) Seismic performance uncertainty estimation via IDA with progressive accelerogram-
wise Latin hypercube sampling. J Struct Eng 140(8):A4014015
van Elk J, Bourne SJ, Oates SJ, Bommer JJ, Pinho R, Crowley H (2019) A probabilistic model to evaluate
options for mitigating induced seismic risk. Earthq Spectra 35:537–564
Wilding BV, Beyer K (2018) Analytical and empirical models for predicting the drift capacity of modern
unreinforced masonry walls. Earthq Eng Struct Dyn 47(10):2012–2031

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Affiliations

Damian N. Grant1 · Jamie Dennis2 · Richard Sturt3 · Giovanni Milan2 ·

David McLennan3 · Pedro Negrette2 · Rene da Costa2 · Michele Palmieri2
1
Arup, 13 Fitzroy Street, London W1T 4BQ, UK
2
Arup, Naritaweg 118, 1043 CA Amsterdam, The Netherlands
3
Arup, Blythe Valley Park Solihull, Birmingham B90 8AE, UK

13