Empirical Fragility

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Bulletin of Earthquake Engineering (2021) 19:3057–3076

https://doi.org/10.1007/s10518-020-00845-9

S.I. : SEISMIC RISK ASSESSMENT IN ITALY

Empirical fragility curves for Italian URM buildings

Annalisa Rosti1 · Maria Rota2 · Andrea Penna1,2

Received: 16 December 2019 / Accepted: 9 April 2020 / Published online: 22 April 2020
© Springer Nature B.V. 2020

Abstract
This paper illustrates the derivation of an empirical fragility model for residential unrein-
forced masonry (URM) buildings, calibrated on Italian post-earthquake damage data and
compatible with the key features of the Italian national seismic risk platform. Seismic vul-
nerability is described by fragility functions for three vulnerability classes, then refined
based on the building height. To this aim, a clustering strategy is implemented to merge
predefined building typologies into vulnerability classes, based on the similarity of the
observed seismic fragility. On the other side, a specific procedure is built up to determine
the vulnerability composition of the exposed URM building stock, starting from national
census data. The empirically-derived model was implemented into the national seismic
risk platform and used, together with other vulnerability models, for assessing seismic
risk in Italy. The results presented in this paper, consisting of refined typological fragility
curves and fragility curves for vulnerability classes, can be also exploited for estimating
both expected seismic damage and risk in sites with similar seismic hazard and building
inventory.

Keywords Fragility curves · Seismic vulnerability · Seismic risk · Post-earthquake damage


data · Clustering · Unreinforced masonry buildings

1 Introduction

Large-scale seismic risk assessment is very important for post-earthquake emergency man-
agement, implementation of strategies for risk mitigation and preparedness, prioritization
of retrofit measures to be implemented in the building stock. Accurate quantifications of
the socio-economic losses due to future earthquakes are also of concern for insurance and
reinsurance industries and for the allocation of financial resources. Considering its mani-
fold practical implications, seismic risk was assessed in several studies, at the urban (e.g.
Erdik et al. 2003; Dolce et al. 2003, 2006; Mouroux and Le Brun 2006; Vicente et al. 2011;

* Maria Rota
maria.rota@eucentre.it
1
Department of Civil Engineering and Architecture, University of Pavia, Via Ferrata 3,
27100 Pavia, Italy
2
Department of Buildings and Infrastructures, EUCENTRE Foundation, Via Ferrata 1,
27100 Pavia, Italy

13
Vol.:(0123456789)
3058 Bulletin of Earthquake Engineering (2021) 19:3057–3076

Marulanda et al. 2013; Eleftheriadou et al. 2016; Smerzini and Pitilakis 2018) and national
scales (e.g. Lucantoni et al. 2001; Di Pasquale et al. 2005; Tyagunov et al. 2006; Crowley
et al. 2009; Rota et al. 2011; Chaulagain et al. 2015; Silva et al. 2015, 2018; NDCP 2018),
with the aim of evaluating the spatial distribution of the expected seismic damage and loss
in a given area.
Clearly, potential damage and losses due to future earthquakes can be effectively
reduced only if reliable and accurate risk estimates are available. In this context, the main
difficulty is that seismic risk assessment requires the involvement of different disciplines,
being the outcome of the convolution of seismic hazard, exposure and seismic vulnerabil-
ity. A further component (i.e. damage to loss functions) could be then included to relate
damage to consequences, such as social or economic loss. Representing the proneness of
buildings to be damaged by an earthquake, seismic vulnerability is the risk component
that mostly attracts the engineering community and it can be estimated through different
approaches, typically classified into empirical, analytical, expert-judgmental and hybrid.
The selection of one rather than another method, each exhibiting different benefits and
downsides, depends on the nature of the available data and purpose of the study. Observa-
tional damage data can be affected by several issues, which require careful consideration
and accurate processing for a reliable quantification of the seismic vulnerability (e.g. Ros-
setto et al. 2013). Nevertheless, thanks to their statistical nature, empirical approaches are
an effective tool for seismic vulnerability description, particularly for territorial applica-
tions (e.g. Braga et al. 1982; Rossetto and Elnashai 2003; Rota et al. 2008; Karababa and
Pomonis 2011; Eleftheriadou and Karabinis 2013; Del Gaudio et al. 2017; Rota and Rosti
2017; Rosti and Rota 2017).
This paper presents an empirical fragility model for residential masonry buildings, to
be used for territorial seismic risk assessment. The fragility model was developed with
the need of complying with the key features of the Italian national seismic risk platform
(IRMA, Faravelli et al. 2019; http://irma.eucen​tre.it/irma/web/home), where it was then
implemented. This platform allows to collect and process the information on seismic fra-
gility produced by the scientific community, providing a unique global tool for getting seis-
mic risk maps and scenarios, taking advantage of the calculation engine OpenQuake v3.2.0
(Pagani et al. 2014). Reference seismic hazard is the official Italian hazard model MPS04
(Stucchi et al. 2004, 2011) whereas exposure derives from national census data (ISTAT
2001, 2011).
While the hazard component cannot be edited, the use of the platform allows one to
import a set of cumulative lognormal fragility curves in terms of PGA and defined for
vulnerability classes, and to associate these fragility functions to different building cate-
gories drawn from building census. Hence, seismic vulnerability description, in terms of
fragility curves, and vulnerability classification of the exposed building stock constitute
the core of the outlined methodology. The general approach on which the proposed fra-
gility model rests, in terms of data source and main assumptions for the identification of
a robust damage database, is discussed in Sect. 2. Key ingredients for deriving fragility
curves and modelling the vulnerability composition of the building stock are respectively
examined in Sects. 3 and 4. Starting from the typological classification of the unreinforced
masonry building stock, empirical fragility curves are derived for three vulnerability
classes, of decreasing vulnerability, then refined based on the building height, to comply
with the IRMA platform, which indeed requires the definition of fragility functions for vul-
nerability classes. In this context, a novel aspect of this work is the implementation of an
agglomerative clustering procedure allowing to objectively identify vulnerability classes,
starting from the observed seismic fragility of the predefined building typologies (Sect. 3).

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3059

In IRMA, the fragility curves for vulnerability classes are then associated to building cat-
egories, identified consistently with the exposure provided by national building census, by
defining the percentages of these building categories belonging to the considered vulner-
ability classes. To meet this requirement, an ad-hoc procedure, using empirically-derived
fragility curves as a target, is developed, allowing to determine the composition of the
exposed building stock in terms of the considered vulnerability classes (Sect. 4). Con-
clusions and future developments of this work are summarised in Sect. 5. Although spe-
cifically designed for meeting the main features of the national seismic risk platform, the
results presented in the paper, i.e. parameters of the empirical fragility functions derived
at different methodological steps and vulnerability classification of the exposed building
stock, can be applied elsewhere for forecasting seismic damage and risk.

2 General approach and source of damage data

This paper proposes an empirically-derived fragility model obtained by statistical process-


ing damage data collected after past Italian seismic events. Damage data were retrieved
from the web-based platform Da.D.O. (Dolce et al. 2019), including damage databases of
nine Italian earthquakes occurred in the time window 1976–2012. This study makes use
of the 1980 Irpinia and of the 2009 L’Aquila post-earthquake damage data, which were
selected for the availability of complete post-earthquake field surveys and of shakemaps,
derived by a rigorous procedure specifically developed for the Italian territory (Michelini
et al. 2008), as discussed in the following.
The incompleteness of post-earthquake field surveys represents one of the possible
sources of uncertainty affecting observational damage data. This issue typically arises in
sites more distant from the earthquake source, where it is likely that post-earthquake sur-
veys were limited to buildings for which the inspection was explicitly requested by the
owner. As a consequence, resulting damage distributions will be biased by the fact that
the number of undamaged buildings is underestimated, as it is often the case that only the
owners of damaged buildings would request an inspection. To tackle this issue, literature
studies either assumed that non-inspected buildings were undamaged (e.g. Karababa and
Pomonis 2011; Eleftheriadou and Karabinis 2013) or selected only those municipalities
with a percentage of inspected buildings exceeding a preselected threshold, denoted as
threshold of completeness (e.g. Sabetta et al. 1998; Goretti and Di Pasquale 2004; Rota
et al. 2008; Rosti et al. 2018).
Figure 1 shows the geographical location of the epicentres of the selected seismic events
together with the spatial distribution of the surveyed municipalities (solid hatch). As indi-
cated by Braga et al. (1982), all the municipalities reported in the Irpinia damage database
were completely surveyed (Fig. 1a). Following the L’Aquila seismic event, 129 munici-
palities were surveyed in the Abruzzi region (solid hatch in Fig. 1b), but only those with
assigned macroseismic intensity (MCS) exceeding VI were completely inspected (Dolce
and Goretti 2015). In this work, the L’Aquila municipalities exhibiting completeness ratio
(i.e. number of inspected buildings over the total number of buildings retrieved from
national census) higher than 90% were selected for creating an unbiased damage dataset
(solid hatch with black dashed lines in Fig. 1b).
Considering that non-surveyed buildings in less affected areas were most probably
undamaged, the L’Aquila dataset was then integrated by considering the negative evidence
of damage in the municipalities characterised by lower PGA levels, in line with existing

13
3060 Bulletin of Earthquake Engineering (2021) 19:3057–3076

Fig. 1  Identification of the municipalities surveyed after the 1980 Irpinia (a) and 2009 L’Aquila (b) earth-
quakes

studies (e.g. Karababa and Pomonis 2011). This means that buildings located in non-sur-
veyed and partially-surveyed (completeness ratio lower than 10%) municipalities, mostly
with an estimated level of PGA lower than 0.05 g, were assumed to be undamaged and
added to the dataset. Under these hypotheses, the considered non-surveyed and partially-
surveyed (with completeness ratio lower than 10%) municipalities for L’Aquila event
resulted to be 176 and 49, respectively, with a total number of masonry buildings equal
to 136’396 (non-surveyed municipalities) and 38′775 (partially-surveyed), estimated from
national census (ISTAT 2001).
Taking advantage of the metrical and typological building data collected in the post-
earthquake survey forms, statistics of the building datasets were carried out for different
building attributes, such as number of storeys, construction age, type of vertical and hori-
zontal structures. Classification of the Irpinia and L’Aquila data, based on the number of
storeys and construction age, is shown in Figs. 2, 3, respectively. Low-rise (i.e. 1–2 storeys)

Fig. 2  Classification of the Irpinia (a) and L’Aquila (b) masonry buildings based on the number of storeys

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3061

1 1

(a) (b)
0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
n 00 4 3 6 2 71 71 n 9 5 1 1 1 1 1 1
ow 19 1- 4- 3- 19 ow 91 9-4 6-6 2-7 2-8 2-9 2-0 200
nk
n 0 4 6 > nk
n <1 1 94 96 97 98 99 >
U 19 19 19 U 1 9 1 1 1 1 1

Construction age Construction age

Fig. 3  Classification of the Irpinia (a) and L’Aquila (b) masonry buildings based on construction age

masonry buildings constitute 89 and 56% of the Irpinia and L’Aquila datasets, respectively.
The majority of the Irpinia masonry buildings was built before 1900 (36%) and a signifi-
cant fraction of L’Aquila constructions dates back prior to 1919 (57%). Figure 4 classifies
masonry buildings based on the type of vertical and horizontal structures, according to the
information reported in the survey forms. Stone masonry buildings with wooden (36%) and
steel (19%) diaphragms represent the majority of the Irpinia dataset (Fig. 4a). More than
half of the L’Aquila masonry buildings exhibit irregular layout or poor-quality materials
and flexible diaphragms (i.e. 57%), followed by regular and good-quality masonry build-
ings with rigid diaphragms (20%). It is observed that different survey forms were used for
damage and usability assessment of residential buildings in the aftermath of the two seis-
mic events, involving a subsequent process of data interpretation oriented to classify data
according to the same damage and building classification systems.

Fig. 4  Classification of the Irpinia (a) and L’Aquila (b) masonry buildings based on the type of vertical and
horizontal structures

13
3062 Bulletin of Earthquake Engineering (2021) 19:3057–3076

3 Derivation of empirical fragility curves

Seismic vulnerability was described in terms of empirically-derived fragility curves,


correlating the ground motion severity with the expected seismic damage. The deriva-
tion of fragility curves required the characterisation of seismic input at damage loca-
tions (Sect. 3.1), the definition of discrete levels of damage (Sect. 3.2) and the adoption
of a typological classification system (Sect. 3.3). After accurate data processing, a sta-
tistical model was fitted to observational damage data (Sect. 3.4), allowing to provide a
continuous description of seismic vulnerability as a function of the ground shaking.

3.1 Seismic input characterisation

Seismic input characterisation requires the identification of a ground motion inten-


sity measure capable of representing the ground motion severity experienced at dam-
age locations (e.g. Rosti et al. 2020). In this study, the peak ground acceleration was
selected, consistently with the framework of the national seismic risk platform. PGA
values were derived from shakemaps produced by the Italian National Institute of Geo-
physics and Volcanology (INGV), based on Michelini et al. (2008). Although INGV
shakemaps are published online only for seismic events occurred after 2008, shakemaps
of some older events, such as the Irpinia earthquake, can be found in Michelini et al.
(2008). Figure 5 shows the shakemaps used in this study, in terms of PGA (i.e. larger of
the two horizontal ground motion components). Consistently with the USGS procedure
(Wald et al. 2006), the INGV shakemaps rely on estimates of local site amplifications
based on the average S-wave velocity in the uppermost 30 m (Vs,30). A Vs,30 classifica-
tion, based on the 1:100′000 Italian geological map, opportunely calibrated against the
observed in situ velocities, is specifically used. PGA was estimated at the municipality
level, in case of the Irpinia dataset, and at the building level, in case of the L’Aquila
earthquake, given the availability in this case of accurately geo-referenced damage data.

Fig. 5  Irpinia (a) and L’Aquila (b) PGA shakemaps used in this study. Adapted from Michelini et al. (2008)
and http://shake​map.rm.ingv.it/shake​/index​.html

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3063

3.2 Damage classification

Damage classification consists of consistently assigning a global level of damage to each


building in the dataset. In the framework of empirical approaches, a suitable damage con-
version rule needs to be established between the damage description reported in the survey
form and discrete damage levels of a preselected scale. Different procedures can be used
for defining a global damage state from the observed damage detected on different build-
ing components, either based on the average or maximum observed seismic damage (Rosti
et al. 2018). In this work, five damage levels (i.e. DS1: slight or negligible damage, DS2:
moderate damage, DS3: severe damage, DS4: very heavy damage and DS5: collapse) plus
the absence of damage (DS0) were defined, consistently with the European Macroseis-
mic Scale (EMS-98, Grünthal 1998). Damage states were defined as the maximum level
of damage observed among the vertical structure, intermediate diaphragms and roof (Rota
et al. 2008, Rosti et al. 2018). The conversion of the damage reported in the survey form in
the considered damage levels was carried out by applying the Dolce et al. (2019) and Rota
et al. (2008) damage conversion rules, for the Irpinia and L’Aquila damage data, respec-
tively (Table 1). The adoption of two different damage conversion rules was unavoidable,
as different survey forms were used for post-earthquake inspections after the two seismic
events. Figure 6 shows the distribution of damage data in the different damage levels for
the 1980 Irpinia (a) and the 2009 L’Aquila (b) earthquakes.

3.3 Typological classification of unreinforced masonry buildings

Building data were allocated to eight building typologies representative of the Italian built
environment, based on the typological classification proposed by Rota et al. (2008). Build-
ing typologies were identified starting from the building attributes reported in the survey
forms and mainly impacting the seismic vulnerability of masonry constructions, i.e. texture
and quality of the masonry fabric (i.e. irregular layout or poor-quality masonry, regular layout
and good-quality masonry), in-plane flexibility of diaphragms (i.e. flexible, rigid), presence
(or absence) of connecting devices, such as tie-rods and tie-beams. 43′718 masonry buildings

Table 1  Damage conversion EMS-98 Damage Description


rules adopted for the 1980
Irpinia (Dolce et al. 2019) and Irpinia (1980) L’Aquila (2009)
the 2009 L’Aquila (Rota et al.
2008) seismic events. The extent Damage level Extent
of each damage level is specified
in the AeDES form used for DS0 No damage No damage (D0)
L’Aquila and evaluated as the DS1 Insignificant Slight (D1) e < 1/3
spread of the corresponding
Slight Slight (D1) 1/3 < e < 2/3
damage level, as a fraction of the
entire building extent Slight (D1) e > 2/3
DS2 Considerable Medium-severe (D2-D3) e < 1/3
Serious
DS3 Very serious Medium-severe (D2-D3) 1/3 < e < 2/3
Medium-severe (D2-D3) e > 2/3
DS4 Partially-collapsed Very heavy (D4-D5) e < 1/3
Very heavy (D4-D5) 1/3 < e < 2/3
DS5 Collapsed Very heavy (D4-D5) e > 2/3

13
3064 Bulletin of Earthquake Engineering (2021) 19:3057–3076

Fig. 6  Damage classification of the Irpinia (a) and L’Aquila (b) masonry dataset-frequency of occurrence of
the different damage levels (DS)

resulted from processing of post-earthquake data based on the abovementioned building


attributes. The adopted building taxonomy is reported in Table 2, where the total number of
buildings falling into each typology is also indicated.
As discussed in Sect. 2, the damage dataset was then enlarged by including buildings sited
in non-surveyed and partially-surveyed municipalities of the Abruzzi region, which could be
reasonably assumed to be undamaged. For each one of these municipalities, the number of
residential masonry constructions was retrieved from the national building census (ISTAT
2001) and then distributed among the predefined building typologies (Table 2) based on the
representativeness (i.e. frequency of occurrence) of each structural typology within the con-
sidered damage dataset.

3.4 Statistical model and fitting procedure

A continuous representation of the seismic vulnerability as a function of ground motion shak-


ing was obtained by imposing a statistical model to observational data points, through a suit-
able fitting procedure. The ingredients of a statistical model are the random component, being
the conditional probability distribution of the response variable given the explanatory variable,
and the systematic component, representing the mean of the response variable, as a function
of the explanatory variable. In this study, damage and the selected ground motion intensity
measure (i.e. PGA) represent the response and the explanatory variables, respectively.
In line with existing studies (e.g. Charvet et al. 2014), the random component was
described via the multinomial distribution. The repartition of buildings in the different
damage states, nij, given the jth PGA threshold is hence defined as:


nDS
Nj ! ( )n
nij ∼ P ds = DSi |PGAj ij (1)
n !
i=0 ij

with Nj being the total number of buildings associated with the jth PGA threshold and
P(ds = DSi|PGAj) is:

13
Table 2  Data classification based on the adopted taxonomy. Number of buildings is reported into brackets
Irregular texture or poor-quality masonry (IRR) Regular texture and good-quality masonry (REG)

Flexible diaphragms (F) Rigid diaphragms (R) Flexible diaphragms (F) Rigid diaphragms (R)
Bulletin of Earthquake Engineering (2021) 19:3057–3076

W/o connecting W/connecting W/o connecting W/connecting W/o connecting W/connecting W/o connecting W/connecting
­devicesa (NCD) devices (CD) devices (NCD) devices (CD) devices (NCD) devices (CD) devices (NCD) devices (CD)

IRR-F-NCD IRR-F-CD (4694) IRR-R-NCD (2537) IRR-R-CD (2457) REG-F-NCD REG-F-CD (1445) REG-R-NCD REG-R-CD (4732)
(24131) (2451) (1271)
a
Connecting devices: tie-rods/tie-beams
3065

13
3066 Bulletin of Earthquake Engineering (2021) 19:3057–3076

⎧ 1 − P�ds ≥ DS �PGA �
P ds = DSi �PGAj = ⎨ P�ds ≥ DSi �PGAj � − P ds ≥ DSi+1 �PGAj
i=0
� �⎪ � i+1 � �j �
⎪ P ds ≥ DSi �PGAj
0 < i < nDS
i = nDS

(2)
The fragility curve, i.e. the probability of reaching or exceeding a given damage state
(DS), was described by the cumulative lognormal distribution. This choice was neces-
sary to get fragility functions complying with the main characteristics of the national
seismic risk platform, although alternative distributions could be adopted, such as the
normal (e.g. Karababa and Pomonis 2011) or the exponential distributions (e.g. Ros-
setto and Elnashai 2003). Based on the cumulative lognormal distribution, the probabil-
ity of reaching or exceeding damage state DS is given by:
[ ( )]
P ds ≥ DSi |PGAj = 𝛷
( ) log PGAj ∕𝜃i
(3)
𝛽

where Φ[·] is the cumulative standard normal distribution, θi is the median value of the
fragility function corresponding to damage level DSi and β is the logarithmic standard
deviation. A unique constant value of dispersion (β) was assumed for all damage states of a
given building typology, to ensure the ordinal nature of damage and avoid crossing fragil-
ity functions (e.g. Lallemant et al. 2015). The unknown optimal parameters of the fragility
model (i.e. θ, vector of median values and β, logarithmic standard deviation, common to
all damage states) were derived by maximising the logarithm of the likelihood, defined as:
[ (nPGA nDS )]
∏ ∏ Nj ! ( )nij
(𝜽, 𝛽) = arg max[log (L(𝜽, 𝛽)] = arg max log P ds = DSi |PGAj
n !
j=1 i=0 ij
(4)
The abovementioned statistical model was fitted to observational damage data to
derive typological fragility curves (Fig. 7). Observation of Fig. 7 shows that masonry
buildings with irregular layout or poor-quality materials are significantly more vulner-
able than those with regular texture and good-quality masonry, whereas the presence of
aseismic devices or of rigid diaphragms reduces seismic vulnerability. The parameters
of the lognormal typological fragility curves are reported in Table 3.
Being better constrained in the lower and upper ground motion range, the obtained
typological fragility curves (Fig. 7) represent a significant improvement with respect
to previously published empirical curves (e.g. Rota et al. 2008). Indeed, the approach
used for defining the dataset of post-earthquake data, together with the selected fitting
procedure, provides curves which have a more standard lognormal shape and physically
meaningful values of the corresponding lognormal parameters (Table 3), hence leading
to more realistic results, especially in the low ground motion range. The fragility esti-
mate in this low ground motion range has a very strong impact on the seismic risk and
loss estimates. This depends on the fact that seismic risk is computed as a convolution
of seismic hazard curves and seismic fragility curves. Since hazard curves provide a
higher contribution in terms of annual frequency of exceedance in the low PGA range,
the effect of fragility curves in this range is amplified by the corresponding hazard factor
and, hence, it provides a large contribution to the risk estimate. This is particularly true
for low-medium seismicity areas, which constitute the majority of Italy. For this reason,

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3067

Fig. 7  Typological empirically-derived fragility curves. Numbers in the legend indicate the sample size

Table 3  Parameters (i.e. median and logarithmic standard deviation) of the typological fragility curves
Building typology θDS1 (g) θDS2 (g) θDS3 (g) θDS4 (g) θDS5 (g) β (–)

IRR-F-NCD 0.110 0.169 0.234 0.312 0.540 0.747


IRR-F-CD 0.134 0.234 0.308 0.441 0.792 0.898
IRR-R-NCD 0.141 0.271 0.360 0.473 0.796 0.823
IRR-R-CD 0.201 0.416 0.524 0.760 1.350 0.969
REG-F-NCD 0.219 0.459 0.610 0.913 1.717 1.078
REG-F-CD 0.315 0.868 1.130 1.699 3.727 1.170
REG-R-NCD 0.324 0.756 1.043 1.539 2.415 1.107
REG-R-CD 0.553 1.544 2.216 3.072 5.118 1.217

having fragility curves better constrained, and hence more representative, in the low
PGA range, can be seen as a significant improvement towards reliable risk estimates.

3.5 Attribution of building typologies to vulnerability classes and definition


of the corresponding fragility curves

The use of the national seismic risk platform for territorial risk applications requires the
definition of fragility curves for vulnerability classes. This raised the need of defining vul-
nerability classes starting from the typological classification of the masonry building stock.
Several existing works focused on the definition of vulnerability classes starting from
post-earthquake survey data (e.g. Goretti and Di Pasquale 2004; Di Pasquale et al. 2005;
Dolce et al. 2006; Dolce and Goretti 2015). The association of structural type to vulner-
ability class established by those studies however follows the relations proposed by Braga
et al. (1982), based on the best agreement between the 1980 Irpinia damage data and the
MSK-76 scale. A different relationship between building typologies and vulnerability
classes is embedded in the EMS-98, considering the different degree of belonging of each
typology to multiple vulnerability classes. This implicit probabilistic relation between

13
3068 Bulletin of Earthquake Engineering (2021) 19:3057–3076

building typologies and vulnerability classes is expressed by linguistic terms, handled by


the macroseismic method (Lagomarsino and Giovinazzi 2006) through the use of the fuzzy
set theory.
In this study, the building typologies identified in Table 2 were objectively merged
into three vulnerability classes of decreasing vulnerability (i.e. A: high vulnerability, B:
medium vulnerability, C1: low vulnerability) based on the similarity of the observed seis-
mic fragility, using a hierarchical agglomerative clustering technique (e.g. Day and Edels-
brunner 1984). The similarity between pairs of building typologies was quantified in terms
of a distance metric, dij, defined as:

√ 5 ( )
√∑
d =√
2
ij ij
𝛿k (5)
k=1

where 𝛿k is the difference in terms of area between the fragility curves of the kth damage
ij

level of the ith and jth typologies in the relevant PGA range, quantified as:
� ⎡ � � � �
0.5g � log PGA∕𝜃 i ⎤ ⎡ j ⎤��
� ⎢ k ⎥ ⎢ log PGA∕𝜃 ⎥��

k
ij
𝛿k = �Φ⎢ ⎥ − Φ⎢ ⎥�dPGA (6)

� ⎢ 𝛽 i
⎥ ⎢ 𝛽 j
⎥��
� ⎣ ⎦ ⎣ ⎦�
PGA=0

where 𝜃ki and βi are the median and logarithmic standard deviation describing the fragil-
ity curve of damage state k of the ith building typology, respectively, and θjk and βj are the
median and logarithmic standard deviation describing the fragility curve of damage state k
of the jth building typology.
The adopted clustering strategy is built on the construction of a distance matrix among
pairs of clusters (i.e. building typologies in this specific context), considering the dis-
tance metric defined in Eq. (5). Damage data of the two clusters with shortest distance are
merged into a unique cluster and the distance matrix among pairs of clusters is updated.
Again, the following closest pairs of typologies are unified and the procedure is iteratively
repeated until the required three classes are formed.
Table 4 shows the attribution of building typologies to vulnerability classes resulting
from the implemented strategy. Figure 8 depicts the cumulative lognormal fragility curves
for vulnerability classes A, B and C1, further refined based on the building height, i.e. L:
low-rise (1–2 storeys) and MH: mid-/high-rise (> 2 storeys). It can be observed that vulner-
ability class A exhibits highest vulnerability, as expected, whereas vulnerability class C1
is the less vulnerable. Also, the impact of the number of stories on seismic vulnerability is

Table 4  Definition of vulnerability classes from the implemented clustering strategy


Horizontal structure Irregular texture or poor-quality Regular texture and good-quality
masonry (IRR) masonry (REG)
w/o connecting w/connecting w/o connecting w/connecting
­devicesa (NCD) devices (CD) devices (NCD) devices (CD)

Flexible diaphragms (F) A A B C1


Rigid diaphragms (R) A B C1 C1
a
Connecting devices refer to tie-rods and tie-beams

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3069

Fig. 8  Empirically-derived fragility curves for vulnerability classes and building classes of height. Numbers
in the legend indicate the sample size

evident, particularly for the vulnerability class B. Median PGAs and logarithmic standard
deviations defining the fragility curves for vulnerability classes are collected in Table 5.

4 Vulnerability classification of the exposed building stock

After deriving fragility functions for the considered vulnerability classes, a procedure for
defining the composition of the exposed building stock in terms of these vulnerability
classes was developed.
IRMA uses national building census data to define the exposure. The exposed build-
ing stock is hence subdivided into macro-categories identified by construction material
(masonry/reinforced concrete/other), number of stories and construction period. Within
the IRMA framework, the vulnerability of the different macro-categories is described as a

Table 5  Parameters (i.e. median and logarithmic standard deviation) of the lognormal fragility curves for
the three considered vulnerability classes
Vulnerabil- Class of height θDS1 (g) θDS2 (g) θDS3 (g) θDS4 (g) θDS5 (g) β (–)
ity class

A L 0.116 0.185 0.261 0.346 0.583 0.754


MH 0.113 0.176 0.226 0.314 0.584 0.821
B L 0.230 0.509 0.664 0.988 1.727 1.029
MH 0.174 0.331 0.426 0.624 1.212 0.997
C1 L 0.484 1.350 1.928 2.742 4.707 1.222
MH 0.418 1.073 1.444 2.118 3.818 1.199

13
3070 Bulletin of Earthquake Engineering (2021) 19:3057–3076

combination of the fragility curves of the considered vulnerability classes (Sect. 3.5), i.e.
assuming that each macro-category can be subdivided into fractions of buildings belonging
to the different vulnerability classes.
Post-earthquake damage data were re-examined to associate each inspected building to
one of the macro-categories, focusing on the L’Aquila dataset, for which the subdivision
into construction age classes is consistent with the one considered in national census data.
Twelve macro-categories were defined by considering two classes of building height (i.e.
L: 1–2 storeys and MH: > 2 storeys), consistently with the previous part of this work, and
six classes of construction age (i.e. < 1919, 1919-45, 1946-61, 1962-71, 1972-81, > 1981),
accounting for building code evolution.
For each macro-category, observational damage data were approximated by fitting log-
normal cumulative distributions, according to Sect. 3.4. Fragility curves of low-rise and
mid-/high-rise masonry macro-categories are respectively depicted in Figs. 9 and 10,
whereas corresponding parameters (i.e. median PGA values and logarithmic standard devi-
ation) are reported in Table 6. Observation of Figs. 9 and 10 shows the reduction of seismic
vulnerability with construction age. Also, for a given construction age, low-rise buildings
are less vulnerable than the corresponding mid-/high-rise ones.
The fragility curve of each damage state of a given macro-category was approximated
by linearly combining the fragility curves previously defined for the considered vulner-
ability classes. The coefficients of the linear combination represent the percentages of
each macro-category belonging to the predefined vulnerability classes and are reported in
Table 7. They were derived through an optimisation problem, by minimising the difference
with the empirically-derived fragility curves of a given macro-category (Figs. 9, 10).
Based on the adopted procedure, the fragility curves of masonry buildings built in a
given time span can be obtained as a function of the parameters of the fragility curves

Fig. 9  Empirically-derived fragility curves for masonry low-rise macro-categories. Numbers in the legend
indicate the sample size

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3071

Fig. 10  Empirically-derived fragility curves for masonry mid-/high-rise macro-categories. Numbers in the
legend indicate the sample size

Table 6  Parameters (i.e. median and logarithmic standard deviation) of the lognormal fragility curves for
masonry macro-categories
Construction age Class of height θDS1 (g) θDS2 (g) θDS3 (g) θDS4 (g) θDS5 (g) β (–)

< 1919 L 0.154 0.228 0.276 0.377 0.656 0.856


MH 0.130 0.189 0.227 0.309 0.579 0.826
1919-45 L 0.198 0.317 0.382 0.561 0.983 0.905
MH 0.162 0.280 0.351 0.524 0.880 0.866
1946-61 L 0.283 0.520 0.684 1.023 1.876 1.023
MH 0.216 0.396 0.516 0.730 1.238 0.916
1962-71 L 0.437 0.972 1.301 1.956 3.071 1.141
MH 0.326 0.738 1.004 1.516 2.228 1.051
1972-81 L 0.553 1.365 1.973 2.705 4.934 1.134
MH 0.467 1.145 1.472 2.040 3.551 1.084
> 1981 L 0.626 1.292 1.560 2.081 3.809 1.175
MH 0.490 1.034 1.271 1.680 2.854 1.098

of the vulnerability classes A-L, B-L and C1-L or A-MH, B-MH and C1-MH, depend-
ing on the building height, and as a function of the percentages reported in Table 7.
For instance, the fragility curves of low-rise masonry buildings built between 1946 and
1961 are defined by combining the curves of vulnerability class A, vulnerability class
B and vulnerability class C1, with combination coefficients equal to 9%, 59% and 32%,
respectively.

13
3072 Bulletin of Earthquake Engineering (2021) 19:3057–3076

Table 7  Composition of the masonry building stock in terms of percentages of macro-categories belonging
to the different vulnerability classes
Class of Vulner- < 1919 (%) 1919-45 (%) 1946-61 (%) 1962-71(%) 1972-81 (%) > 1981 (%)
height ability
class

L A 86 45 9 5 0 0
B 0 44 59 4 0 0
C1 14 11 32 91 100 100
MH A 97 22 0 0 0 0
B 0 78 75 18 0 0
C1 3 0 25 82 100 100

For each macro-category, in Fig. 11, the accuracy of the adopted methodology is
demonstrated by graphically comparing the fragility functions resulting from the
weighted mean of the fragility curves of the different vulnerability classes (continuous
line) with those obtained by directly fitting observational data points (dashed line). It
can be observed than the linear combination of the fragility curves of the vulnerability
classes generally leads to a good approximation of the empirically-derived ones.

Fig. 11  Comparison of fragility curves of masonry macro-categories, derived from linear combination
of the fragility curves of vulnerability classes (continuous lines) and from direct fitting of empirical data
points (dashed lines)

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3073

Fig. 12  Geographical distribution of masonry buildings belonging to vulnerability classes A, B and C1 in


the municipalities of the Abruzzi region

It is worth recalling, once more, that it was not possible to use fragility curves directly
fitted on empirical data for macro-categories, since IRMA requires the use of fragility
curves for vulnerability classes. On the other hand, the fragility curves previously defined
for vulnerability classes (Sect. 3.5) were not directly assigned to buildings from national
census data, to avoid preassigning macro-categories to given vulnerability classes only
based on expert judgement. The proposed approach, instead, makes use of empirical fragil-
ity curves for macro-categories as a target for defining optimal coefficients representing the
degree of belonging of each macro-category to the considered vulnerability classes.
Figure 12 depicts the vulnerability classification of the exposed masonry building stock
(from national census) with reference to the Abruzzi region, selected as an example. It
shows, for each municipality, the percentage of masonry buildings falling into the three
considered vulnerability classes, by making use of the association between macro-catego-
ries and vulnerability classes reported in Table 7.

5 Conclusions

The aim of this study is to develop a fragility model for residential masonry buildings,
consistent with the framework of the Italian national platform for large-scale seismic risk
assessment. The presented fragility model is empirically-derived by statistically process-
ing a robust database including damage data from the Irpinia (1980) and L’Aquila (2009)
earthquakes. The outlined methodology includes a description of seismic vulnerability in
terms of fragility curves and the vulnerability classification of the exposed buildings, i.e.
the identification of the fractions of masonry buildings belonging to the considered vulner-
ability classes, starting from building attributes of the national census.
The first original contribution of this work consists in a more refined and updated
description of the seismic vulnerability of masonry typologies typical of the Italian build-
ing stock. With respect to previous research works on empirical fragility curves, the pro-
posed curves are the result of a more accurate and sound processing of post-earthquake
data, properly accounting for the issue of survey completeness and the negative evidence
of damage. As a result, the curves are much better constrained, especially in the lower and
upper ground motion range, hence overcoming some of the limitations typical of empirical

13
3074 Bulletin of Earthquake Engineering (2021) 19:3057–3076

fragility/vulnerability curves available in the literature (e.g. Rota et al. 2008; Ioannou et al.
2018).
The main restrictions dictated by the use of the national platform concerned the deriva-
tion of fragility curves for vulnerability classes, the selection of peak ground acceleration
as ground motion intensity measure and the adoption of the cumulative lognormal distri-
bution for the definition of fragility curves. Bearing in mind these constraints, empirical
lognormal fragility curves are derived for three vulnerability classes and two classes of
building height, adopting the PGA from shakemap as seismic intensity measure.
Novelty of this work is the use of a hierarchical agglomerative cluster strategy for objec-
tively defining vulnerability classes based on the similarity of the observed seismic fragil-
ity of predefined building typologies. On the other side, the vulnerability classification of
the exposed building stock is obtained by deriving the percentages of masonry building
categories, drawn from building census attributes, pertaining to the predefined vulnerabil-
ity classes. At this stage of the work, the vulnerability composition of the building stock is
assumed to be uniformly representative of all the Italian municipalities. Possible differen-
tiations based on demography, geographical context and/or rural versus urban character of
the municipalities could be envisaged in the future.
Additional future developments of this work will concern the derivation of fragil-
ity curves for other seismic intensity measures, aimed at investigating the ground motion
parameter better correlated with the observed seismic damage. Intensity measures will also
be selected based on their availability from both shakemaps and seismic hazard studies, to
make the fragility model usable for seismic risk applications. In this context, the opportu-
nity of accounting for the ground motion uncertainty in the fragility evaluation will be also
investigated and the deriving bias eventually removed (e.g. according to the procedure pro-
posed by Ader et al. 2018). The possibility of further enlarging the adopted damage dataset
by adding empirical damage data from other seismic events will be also explored, provid-
ing that both shakemaps and post-earthquake survey forms are available.
Although the presented vulnerability model was derived for complying with the struc-
ture of the national seismic risk platform, the results reported in this paper, i.e. param-
eters of fragility curves and vulnerability classification of the exposed building stock, can
be exploited for assessing seismic vulnerability and risk in territories with similar seis-
mic hazard and exposed buildings. A further advantage of this study is the availability of
the parameters of fragility functions derived at all intermediate methodological steps (i.e.
fragility curves for building typologies, vulnerability classes, macro-categories), allowing
others to select the set of fragility functions which is more appropriate for their adopted
building taxonomy and specific objective.

Acknowledgements This work was developed under the financial support of the Italian Department of Civil
Protection, within the ReLUIS-DPC 2018 Research Project, which is gratefully acknowledged.

References
Ader T, Grant DN, Free M, Villani M, Lopez J, Spence R (2018) An unbiased estimation of empirical log-
normal fragility functions with uncertainties on the ground motion intensity measure. J Earthq Eng.
https​://doi.org/10.1080/13632​469.2018.14694​39
Braga F., Dolce M., Liberatore D. (1982) A statistical study on damaged buildings and an ensuing review
of the M.S.K.-76 scale. In: Proceedings of the 7th European conference on earthquale engineering,
Athens, Greece
Census Data (2001) http://dawin​ci.istat​.it/jsp/MD/dawin​ciMD.jsp

13
Bulletin of Earthquake Engineering (2021) 19:3057–3076 3075

Charvet I, Ioannou I, Rossetto T, Suppasri A, Imamura F (2014) Empirical fragility assessment of build-
ings affected by the 2011 Great East Japan tsunami using improved statistical models. Nat Hazards
73(2):951–973
Chaulagain H, Rodrigues H, Silva V, Spacone E, Varum H (2015) Seismic risk assessment and hazard
mapping in Nepal. Nat Hazards 78:583–602
Crowley H, Colombi M, Borzi B, Faravelli M, Onida M, Lopez M, Polli D, Meroni F, Pinho R (2009) A
comparison of seismic risk maps for Italy. Bull Earthq Eng 7:149–180
Day WHE, Edelsbrunner H (1984) Efficient algorithms for agglomerative hierarchical clustering meth-
ods. J Classif 1(1):7–24
Del Gaudio C, De Martino G, Di Ludovico M, Manfredi G, Prota A, Ricci P, Verderame GM (2017)
Empirical fragility curves from damage data on RC buildings after the 2009 L’Aquila earthquake.
Bull Earthq Eng 15(4):1425–1450
Di Pasquale G, Orsini G, Romeo RW (2005) New developments in seismic risk assessment in Italy. Bull
Earthq Eng 3:101–128
Dolce M, Goretti A (2015) Building damage assessment after the 2009 Abruzzi earthquake. Bull Earthq
Eng 13(8):2241–2264
Dolce M, Masi A, Marino M, Vona M (2003) Earthquake damage scenarios of the building stock of
Potenza (Southern Italy) including site effects. Bull Earthq Eng 1:115–140
Dolce M, Kappos A, Masi A, Penelis G, Vona M (2006) Vulnerability assessment and earthquake dam-
age scenarios of the building stock of Potenza (Southern Italy) using Italian and Greek methodolo-
gies. Eng Struct 28:357–371
Dolce M, Speranza E, Giordano F, Borzi B, Bocchi F, Conte C, Di Meo A, Faravelli M, Pascale V
(2019) Observed damage database of past Italian earthquakes: the DaDO WebGIS. Boll Geofis
Teor Appl 60(2):141–164
Eleftheriadou AK, Karabinis AI (2013) Evaluation of damage probability matrices from observational
seismic damage data. Earthq Struct 4(3):299–324
Eleftheriadou AK, Baltzopoulou AD, Karabinis AI (2016) Urban seismic risk assessment: statistical
repair cost data and probable structural losses based on damage scenario-correlation analysis. Int J
Adv Struct Eng 8(2):133–150
Erdik M, Aydinoglu N, Fahjan Y, Sesetyan K, Demircioglu M, Siyahi B, Durukal E, Ozbey C, Biro Y,
Akman H, Yuzugullu O (2003) Earthquake risk assessment for Istanbul metropolitan area. Earthq
Eng Eng Vibr 2(1):1–25
Faravelli M, Polli D, Quaroni D, Onida M, Pagano M, Di Meo A, Borzi B (2019) Italian platform for
seismic risk and damage scenario evaluation. In: 7th ECCOMAS thematic conference on compu-
tational methods in structural dynamics and earthquake engineering, 24–26 June, Crete (Greece)
Goretti A, Di Pasquale G (2004) Building inspection and damage data for the 2002 Molise, Italy, earth-
quake. Earthq Spectra 20(S1):S167–S190
Grünthal G (ed.), Musson RMW, Schwarz J, Stucchi M (1998) European Macroseismic Scale. Cahiers
du Centre Européen de Géodynamique et de Séismologie, Vol 15-European Macroseismic Scale
1998. European Centre for Geodynamics and Seismology, Luxembourg
Ioannou I, Bessason B, Kosmidis I, Bjarnason JO, Rossetto T (2018) Empirical seismic vulnerability
assessment of Icelandic buildings affected by the 2000 sequence of earthquakes. Bull Earthq Eng
16:5875–5903
Karababa FS, Pomonis A (2011) Damage data analysis and vulnerability estimation following the
August 14, 2003 Lefkada Island, Greece, Earthquake. Bull Earthq Eng 9:1015–1046
Lagomarsino S, Giovinazzi S (2006) Macroseismic and mechanical models for the vulnerability and
damage assessment of current buildings. Bull Earthq Eng 4:415–443
Lallemant D, Kiremidjian A, Burton H (2015) Statistical procedures for developing earthquake damage
fragility curves. Earthq Eng Struct Dyn 44:1373–1389
Lucantoni A, Bosi V, Bramerini F, De Marco R, Lo PT, Naso G, Sabetta F (2001) Seismic risk in Italy.
Ing Sismica, XVII 1:5–36 (in Italian)
Marulanda SMC, Carreño ML, Cardona OD, Ordaz MG, Barbat AH (2013) Probabilistic earthquake risk
assessment using CAPRA: application to the city of Barcelona. Nat Hazards 69:59–84
Michelini A, Faenza L, Lauciani V, Malagnini L (2008) ShakeMap implementation in Italy. Seismol Res
Lett 79(5):688–697
Mouroux P, Le Brun B (2006) Presentation of the RISK-UE Project. Bull Earthq Eng 4(4):323–339
National Department of Civil Protection (NDCP, ed., 2018) National risk assessment: overview of the
potential major disasters in Italy http://www.prote​zione​civil​e.gov.it/docum​ents/20182​/82380​3/
Docum​ento+sulla​+Valut​azion​e+nazio​nale+dei+risch​i/57f33​7fd-a421-4cb0-b04c-234b6​1997a​2f

13
3076 Bulletin of Earthquake Engineering (2021) 19:3057–3076

Pagani M, Monelli D, Weatherill G, Danciu L, Crowley H, Silva V, Henshaw P, Butler L, Nastasi M, Panz-
eri L, Simionato M, Vigano D (2014) Openquake engine: an open hazard (and risk) software for the
global earthquake model. Seismol Res Lett 85(3):692–702
Rossetto T, Elnashai A (2003) Derivation of vulnerability functions for European-type RC structures based
on observational data. Eng Struct 25(10):1241–1263
Rossetto T, Ioannou I, Grant DN (2013) Existing empirical vulnerability and fragility functions: Compen-
dium and guide for selection. GEM Technical Report 2013-X, GEM Foundation, Pavia, Italy
Rosti A, Rota M (2017) Comparison of PSH results with historical macroseismic observations at different
scales. Part 2: application to South-East France. Bull Earthq Eng 15(11):4609–4633
Rosti A, Rota M, Penna A (2018) Damage classification and derivation of damage probability matrices from
L’Aquila (2009) post-earthquake survey data. Bull Earthq Eng 16(9):3687–3720
Rosti A, Rota M, Penna A (2020) Influence of seismic input characterisation on empirical damage probabil-
ity matrices for the 2009 L’Aquila event. Soil Dyn Earthq Eng 128:105870. https​://doi.org/10.1016/j.
soild​yn.2019.10587​0
Rota M, Rosti A (2017) Comparison of PSH results with historical macroseismic observations at different
scales. Part 1: methodology. Bull Earthq Eng 15(11):4585–4607
Rota M, Penna A, Strobbia CL (2008) Processing Italian damage data to derive typological fragility curves.
Soil Dyn Earthq Eng 28(10):933–947
Rota M, Penna A, Strobbia C, Magenes G (2011) Typological seismic risk maps for Italy. Earthq Spectra
27(3):907–926
Sabetta F, Goretti A, Lucantoni A (1998) Empirical fragility curves from damage surveys and estimated
strong ground motion. In: Proceedings of the 11th European conference earthquake engineering,
Balkema, Rotterdam
Silva V, Crowley H, Varum H, Pinho R (2015) Seismic risk assessment for mainland Portugal. Bull Earthq
Eng 13(2):429–457
Silva V, Amo-Oduro D, Calderon A, Dabbeek J, Despotaki V, Martins L, Rao A, Simionato M, Viganò
D, Yepes-Estrada C, Acevedo A, Crowley H, Horspool N, Jaiswal K, Journeay M, Pittore M (2018)
Global Earthquake Model (GEM). Seismic Risk Map (version 2018.1), https​://doi.org/10.13117​/gemgl​
obal-seism​ic-risk-map-2018
Smerzini C, Pitilakis K (2018) Seismic risk assessment at urban scale from 3D physics-based numerical
modelling: the case of Thessaloniki. Bull Earthq Eng 16:2609–2631
Stucchi M, Akinci A, Faccioli E, Gasperini P, Malagnini L, Meletti C, Montaldo V, Valensise G, (2004)
Mappa di Pericolosità sismica del territorio Nazionale http://zones​ismic​he.mi.ingv.it/docum​enti/rappo​
rto_concl​usivo​.pdf (in Italian)
Stucchi M, Meletti C, Montaldo V, Crowley H, Calvi GM, Boschi E (2011) Seismic hazard assessment
(2003–2009) for the Italian building code. B Seismol Soc Am 101:1885–1911
Tyagunov S, Grünthal G, Wahlström R, Stempniewski L, Zschau J (2006) Seismic risk mapping for Ger-
many. Nat Hazards Earth Syst 6:573–586
Vicente R, Parodi S, Lagomarsino S, Varum H, Mendes Silve JAR (2011) Seismic vulnerability and risk
assessment: case study of the historic city centre of Coimbra, Portugal. Bull Earthq Eng 9:1067–1096
Wald DJ, Worden CB, Quitoriano V, Pankow KL (2006) S ­ hakeMap® manual, technical manual, users guide,
and sofware guide. http://pubs.usgs.gov/tm/2005/12A01​/pdf/508TM​12-A1.pdf

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

13

You might also like