Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/258354407

Typological fragility curves form Italian


earthquake damage data

Conference Paper · January 2006

CITATIONS READS

19 318

3 authors:

Maria Rota Andrea Penna


European Centre for Training and Research i… University of Pavia
49 PUBLICATIONS 550 CITATIONS 123 PUBLICATIONS 1,354 CITATIONS

SEE PROFILE SEE PROFILE

Claudio Luciano Strobbia


realtimeseismic
91 PUBLICATIONS 974 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Sigma Project View project

Contribution in the preparation of the Italian Norms for Construction View project

All content following this page was uploaded by Maria Rota on 03 June 2014.

The user has requested enhancement of the downloaded file.


First European Conference on Earthquake Engineering and Seismology
(a joint event of the 13th ECEE & 30th General Assembly of the ESC)
Geneva, Switzerland, 3-8 September 2006
Paper Number: 386

TYPOLOGICAL FRAGILITY CURVES FROM ITALIAN EARTHQUAKE


DAMAGE DATA

Maria ROTA1, Andrea PENNA2 and Claudio STROBBIA2

SUMMARY

The data obtained from post-earthquake surveys carried out in Italy during the last 30 years have
been used to derive typological fragility curves for typical building classes. The adopted building
classification has been derived from the one presented in the RISK-UE project, consistent with the
EMS98 macroseismic scale, and adapted to the information available from the different survey
forms. The experimental damage probability matrices have been processed in order to infer the
parameters of lognormal distributions, deriving fragility curves relating damage probability to
PGA, for five damage states. Nonlinear regression procedures have been implemented to obtain
the curve parameters, fitting the experimental points and considering the relative reliability of each
point, depending on the sample size and quality. Curve parameters, reported in the paper, allow
their practical use in seismic risk analysis and scenarios and possible relative comparisons
between different typologies.

1. INTRODUCTION

Building vulnerability is the crucial aspect of earthquake risk evaluation and reduction: for this reason a lot of
effort has been devoted in the past in order to develop vulnerability prediction models and many researchers still
focus their activity on this topic. The commonly used tool to characterise seismic vulnerability is the fragility
curve, derived using different methods and available data. These methods can be classified on the basis of the
main source of information used: judgement-based, analytical, empirical and hybrid methods [Rossetto and
Elnashai, 2003]. Judgement-based fragility curves are derived from statistical treatment of the information
provided by a team of experts [ATC-13, 1985]. Advantages of this approach consist in the possibility of
obtaining vulnerability curves for any type of structure, ground motion severity and so on, easily including all
the factors affecting the seismic response of different structures. However, the reliability of judgement-based
curves is questionable due to their dependence on the individual experience of the experts. Analytical methods
derive vulnerability curves from statistical elaboration of the results of numerical structural analyses. For large
scale vulnerability analyses, usually some simplified analytical methods are used [e.g. Calvi, 1999; FEMA,
1999; Milutinovic and Trendafiloski, 2003] since refined models would be computationally too expensive. An
important issue, especially when simplified models are used, concerns the similarity of the model to the real
structure, which strongly influences the reliability of the results and which is affected by the modelling
capabilities.
Empirical methods use as the main source of information the data collected during post-earthquake surveys and
elaborate these data according to statistical procedures. It is evident that real observed data are the most realistic
source of information, being able to account for any characteristic of the building stock and of the ground
motion. One of the limitations of this method is related to the heterogeneity of the empirical data and of the post-
earthquake surveys from which data are obtained. Data should be as much as possible homogeneous both in time
1
European School for Advanced Studies in Reduction of Seismic Risk, via Ferrata 1, Pavia, Italy
Email : mrota@roseschool.it
2
European Centre for Training and Research in Earthquake Engineering, via Ferrata, 1, Pavia, Italy
Email: andrea.penna@eucentre.it, claudio.strobbia@eucentre.it

1
and space, due to the significant variations of construction technique occurring during time and from place to
place. Another limitation is related to the amount of available data: in order to obtain realistic vulnerability
curves, a large amount of data is required, to reduce the scatter in the results. The use of small data sets, in fact,
is likely to produce biased results, which are not significant from a statistical point of view.
Hence the choice becomes whether to use data coming from a single event [e.g. Orsini, 1999; Braga et al., 1982],
leading to a limited data set, or combining data from different events [e.g. Rossetto and Elnashai, 2003; Spence
et al., 1992], with problems of heterogeneity.
The current study has the enormous advantage of having a very large amount of empirical data, all derived from
Italian post-earthquake surveys. This implies a significant increase of the data homogeneity, even if there can
still be some regional differences in the surveyed building stocks. Furthermore, the abundance of data allows to
carry out severe choices on the quality of the information derived from the survey, on the homogenisation of the
data and on the allocation of data to classes of ground motion severity. Moreover, building typologies have been
accurately evaluated based on a priori information and on the analysis of data. In particular the sample accuracy,
the differences in the observations and the necessity and possibility of aggregating or simply disregarding some
data have been taken into account.
Finally, hybrid methods make use of different sources of information combined together, in order to overcome
some of the main limitations of the various methods previously described, which are based on a single type of
information. An example can be the macroseismic methods, which are based on both empirical data and expert
judgement [e.g. Giovinazzi and Lagomarsino, 2004; Bernardini, 2004].

2. MAIN INGREDIENTS OF FRAGILITY CURVES

Fragility curves describe the conditional probability of a class of buildings reaching or exceeding a specific level
of damage, for a given level of ground motion. Thus, the development of fragility curves requires the
characterization of the ground motion, the identification of the different levels of damage, which are of interest,
and the choice of the typologies of buildings for which fragility curves are being developed.

2.1 Ground motion characterization

One of the basic ingredients of a vulnerability curve is the seismic demand, which is represented on the
horizontal axis of the curve. Many different parameters have been proposed in the literature to characterize the
severity of ground motion, but none of them have been universally accepted. The most commonly used
parameters for describing the ground motion are macroseismic intensity and peak ground acceleration (PGA).
Both of them have advantages and drawbacks and there is a rather wide discussion on the literature on which one
is more suitable for use in the derivation of vulnerability curves [e.g. Faccioli et al., 1999, Rota, 2006].
In the present work, PGA has been chosen for representing ground motion. The main reason for this choice is
that the Italian hazard map [Gruppo di Lavoro, 2004] is represented in terms of PGA and hence vulnerability
curves in terms of PGA are needed in order to develop risk scenarios. Each municipality has been characterized
by a unique value of PGA, to be considered as a mean value, evaluated using the attenuation relationship
proposed by Sabetta and Pugliese [1987], with the magnitude and epicentre of the corresponding earthquake.
The mean value of the attenuation relationship has been used, with the PGA evaluated on rock. This approach
has been validated through a comparison of the measured PGA, when available, with the PGA interval in which
the predicted value falls. It is well known that the value of PGA on rock may be amplified due to stratigraphical
and/or topographical effects. Since risk is obtained from the convolution of hazard, vulnerability and exposure,
site effects can be either incorporated in the hazard part or in the vulnerability. In this work, local site
amplification effects are included in the vulnerability definition, since they are intrinsically accounted for by the
empirical data, which are then elaborated to derive the vulnerability curves.

2.2 Damage scale

In order to construct fragility curves, it is necessary to establish a damage scale, with well defined levels of
damage. The damage scale used in this work actually corresponds to the one defined in the European
Macroseismic Scale [Grunthal, 1998] and it includes five levels of damage (plus the case of no damage):
negligible to slight damage (L), moderate damage (M), substantial to heavy damage (S), very heavy damage (V)
and destruction (D).
Post-earthquake surveys were carried out using forms varying in time and hence referring to different damage
scales from earthquake to earthquake. In order to derive vulnerability curves based on data from all the available

2
events, it has been necessary to convert the different scales into a unique one. Only structural damage has been
considered, ignoring at this stage non structural damage. In particular, damage to vertical bearing structure,
horizontal structure and roof has been used. The information regarding non structural damage has been used for
those cases in which there was no indication about the structural damage, to separate cases of no damage, from
cases of incompletely filled forms.

2.3 Building typologies

In order to develop a vulnerability model, it is necessary to introduce a building classification, to ensure a


uniform interpretation of the results. Classes of buildings should be defined by grouping together structures that
are expected to have a similar seismic behaviour.
Most studies of earthquake damage have concluded that the type of vertical and horizontal bearing structure is
the most important factor affecting earthquake damage [e.g. Coburn and Spence, 2002]. However, also the date
or period of construction of a building is a crucial factor influencing the vulnerability evaluation, since it
determines constructive details and design criteria enforced in a given area. It is well known that the number of
storeys plays an essential role in defining the seismic behaviour of a structure, being directly related to the height
of the building and hence to the period of vibration. Hence, the number of storeys has a strong effect on both
structural capacity and demand. Many other aspects have been shown to have an influence over buildings
performance in earthquakes, but it would not be easy to include all of them in a typological classification,
especially because finer and finer subdivisions of the building classification would require correspondingly more
vulnerability relationships to be defined, and quantitative measures of the separate influence of these factors are
difficult to obtain.
Nevertheless, the choice of the building typologies to be adopted will depend not only on the characteristics
which are expected to influence the earthquake performance of the structure, but also on the extent of data
available. In particular, in this study, building typologies have been identified with reference to the RISK-UE
[Milutinovic Z.V. and Trendafiloski, 2003] building typology matrix, modified on the basis of the information
available from all the databases (each one corresponding to a different post-earthquake survey). The typologies
adopted are reported in Table 1.
Some comments about the adopted buildings typologies of Table 1 are needed. Mixed structures refer to
buildings with vertical bearing structure constituted by elements of both masonry and reinforced concrete.
For what concerns reinforced concrete structures, they have been distinguished according to whether they are
seismically designed or not, based on the period of construction and the year of seismic classification of the
municipality in which the building is located. In particular, buildings have been considered to be seismically
designed when constructed after 1975, in a municipality that was already classified as “seismic” at the time of
construction. The year 1975 has been selected, according also to Di Pasquale et al. [2005], since it corresponds
to the first applicative decree of the seismic Italian law No. 64 of 1974, which introduced a new seismic zonation
and specific regulations for constructions in seismic areas. For what concerns masonry, the type of horizontal
structure has been considered, distinguishing between rigid floors (reinforced concrete slabs and vaults) and
flexible floors (wood, steel, etc.). Moreover, the layout and quality of masonry has been considered, identifying
buildings with a regular layout of good quality masonry and buildings with irregular layout or bad quality
masonry. The reference year of construction used to identify masonry buildings with tie rods and tie beams, in
those databases where this information was not directly available, is 1909, the year after the terrible Reggio and
Messina earthquake, corresponding to the first seismic classification of the Italian territory. Finally, data
concerning steel structures were not actually available from all the earthquakes and were quite limited in
number, so that no further subdivision has been introduced for this typology of structure.

3
Table 1: Adopted building typological classification
No. of No. of
Label Description Label Description
storeys storeys
Masonry – irregular layout – rigid
MX1 Mixed 1 -2 IMA7 ≥3
floors – with tie rods or tie beams
Masonry – irregular layout – rigid
MX2 Mixed ≥3 IMA8 ≥3
floors – w/o tie rods or tie beams
Reinforced concrete – seismic Masonry – regular layout – flexible
RC1 1-3 RMA1 1-2
design floors – tie rods or tie beams
Reinforced concrete – no Masonry – regular layout – flexible
RC2 1-3 RMA2 1-2
seismic design floors – w/o tie rods
Reinforced concrete – seismic Masonry – regular layout – rigid
RC3 ≥4 RMA3 1-2
design floors – with tie rods
Reinforced concrete – no Masonry – regular layout – rigid
RC4 ≥4 RMA4 1-2
seismic design floors – w/o tie rods
Masonry – irregular layout – Masonry – regular layout – flexible
IMA1 1-2 RMA5 ≥3
flexible floors – with tie rods floors – with tie rods
Masonry – irregular layout – Masonry – regular layout – flexible
IMA2 1-2 RMA6 ≥3
flexible floors– w/o tie rods floors – w/o tie rods
Masonry – irregular layout – Masonry – regular layout – rigid
IMA3 1-2 RMA7 ≥3
rigid floors – with tie rods floors – with tie rods
Masonry – irregular layout – Masonry – regular layout – rigid
IMA4 1-2 RMA8 ≥3
rigid floors - w/o tie rods floors – w/o tie rods
Masonry – irregular layout –
IMA5 ≥3 ST Steel all
flexible floors – with tie rods
Masonry – irregular layout –
IMA6 ≥3
flexible floors –w/o tie rods

3. AVAILABLE DATA

Vulnerability curves have been derived starting from a dataset of observed data composed of almost 164.000
surveyed buildings (163479). These data have been collected during post-earthquake surveys, carried out after
the main recent Italian earthquakes. In particular, data concerning the following events have been used: Irpinia
(1980), Abruzzo (1984), Umbria-Marche (1997), Pollino (1998) and Molise (2002). For what concerns the
earthquake of Umbria-Marche, it is important to know that only data observed in the Marche region are included
in this work, since data of the Umbria region are still being processed.
Different forms have been used for collecting data in the post-earthquake survey following each one of the above
events. Moreover, in each case, data have been computerised in a different way. For these reasons a very
complex and time-consuming work has been necessary, in order to first understand the meaning of each field in
each database, and then to homogenize the information, to construct uniform damage scales, building typologies
etc., common to all databases. This process has required several interpretative assumptions for each database,
which are described in Rota [2006].
Due to the very rapid nature of the survey, not all the forms were adequately compiled: in some cases, some very
important data were missing and hence the corresponding buildings have been disregarded. After processing of
the databases, data concerning 9713 buildings have been disregarded. This corresponds to less than 6% of the
total number of available data. The 153766 buildings considered at this stage of the study are subdivided, for
what concerns the typology and the PGA, as indicated in Figure 1.
To derive vulnerability curves, it is necessary to obtain, for a population of buildings exposed to a certain PGA,
the frequency of different damage states. For a given PGA, the number of buildings reaching a given level of
damage has to be normalised to the total number of buildings in the population. This information is usually
unknown, due to the nature of post-earthquake surveys, which are often carried out only on damaged structures,
hence disregarding some buildings, which are actually part of the total building stock. Incomplete surveys can
lead to biased data, because the distribution of damage of non surveyed buildings is likely to be different from
the one of surveyed buildings. It is also likely that all damaged buildings have been surveyed, and that the lost
observations refer to undamaged buildings; hence, considering incomplete survey data could bring to an
overestimation of the probability of damage, because of the normalisation with respect to an underestimated
sample. Hence, for each considered municipality, the number of surveyed buildings is compared to the total

4
number of buildings according to national census data. This information, together with the number of surveyed
buildings, is then used to define a threshold for deciding whether a municipality is completely surveyed or not. It
should be noticed that the definition of building can be however slightly different in the two approaches.
Two different samples have been defined, including data of different levels of completeness of the survey. The
comparison between the results of the two datasets has shown that there is no systematic difference between
them. After these considerations, the data set that has been analysed consists of approximately 100.000 buildings
(91394). The amount of these data coming from each single earthquake, for each PGA interval and for each
vertical bearing structure typology, is shown in Figure 2 (left and right, respectively).

Figure 1: Subdivision of data among building typologies and experienced PGA

50000 80000
45000 Molise
70000 Molise
40000 Pollino
Pollino
60000
Number of buildings

Number of buildings

35000 Marche
Marche
30000 Abruzzo 50000
Abruzzo
25000 Irpinia
40000 Irpinia
20000
30000
15000
20000
10000
5000 10000
0
0
0 - 0.1 0.1 - 0.2 0.2 - 0.3 0.3 - 0.4 0.4 - 0.5
RC IRR MAS REG MAS MIXED STEEL
PGA intervals [g]
Type of vertical structure

Figure 2: Subdivision of data among experienced PGA (left) and building typologies (right) with
identification of data coming from the different earthquakes

4. DATA PROCESSING

For each building typology and for each PGA, the relative frequency of the different damage states has been
computed, obtaining a damage probability matrix (DPM). Figure 3 shows an example of some DPMs, for five
PGA classes and for the three typologies of RC2, MX1 and IMA2 (see Table 1 for the identification of the
typologies). Notice that what is represented in Figure 3 is the probability of reaching a given damage state
without exceeding it and that also the frequency of no damage (DS0) is plotted.
The probability of reaching or exceeding each damage state can be hence obtained simply cumulating the
experimental frequencies from the highest to the lowest level of damage. Then, for each damage state, the data
can be easily rearranged, to show the experimental frequency as a function of PGA. An example of this is given
in Figure 4, which shows, for a given typology (IMA2), the experimental data for the 5 considered damage
states. Notice that, in this case, the no damage is not represented, since its cumulative value is always equal to 1.

5
(0.0-0.1)g (0.1-0.2)g (0.2-0.3)g (0.3-0.4)g (0.4-0.5)g

0.6 0.6 0.6 0.6 0.6


RC2

0.4 0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2 0.2

0 0 0 0 0
ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2ds3ds4ds5

0.6 0.6 0.6 0.6 0.6


MX1

0.4 0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2 0.2

0 0 0 0 0
ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2ds3ds4ds5

0.6 0.6 0.6 0.6 0.6


IMA2

0.4 0.4 0.4 0.4 0.4

0.2 0.2 0.2 0.2 0.2

0 0 0 0 0
ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2 ds3ds4ds5 ds0ds1ds2ds3ds4ds5

Figure 3: DPMs, for 3 building typologies: RC2, MX1 and IMA2

0.8
probability [-]

0.6

0.4

0.2

0
0 0.2 0.4 0 0.2 0.4 0 0.2 0.4 0 0.2 0.4 0 0.2 0.4
pga [g] pga [g] pga [g] pga [g] pga [g]

Figure 4: Experimental frequency of each damage state as a function of PGA – building typology IMA2

The experimental data have been hence processed to get the parameters of an analytical function describing, for
each damage state, the probability as a function of PGA: the adopted model is the functional form of the
lognormal cumulative distribution, which has been often used for fragility curves [e.g. FEMA, 1999; Milutinovic
and Trendafiloski, 2003]. The cumulative lognormal distribution does not exist in a simple closed formula, and is
computed as the integral of the lognormal probability density function:
X − ( ln x − μ )2

∫ f (x )dx
1
F(X ) = where: f ( x) = e 2σ 2
(1)
0 xσ 2π
Considering the experimental points of a single typology and a single damage state, the two parameters μ and σ
of the lognormal distribution are inferred by fitting the curve to the experimental data. A nonlinear regression is
needed to solve this optimisation problem, and an iterative linearised least-squares algorithm has been used for

6
this task. The partial derivatives necessary to build the Jacobian matrix are computed with a first order finite
difference at each of the few iterations performed to reach convergence.
However, this simple approach can pose several problems since the data quality is not homogenous. The total
number of surveyed buildings is subdivided in 23 building typologies and in 10 PGA classes, hence reducing the
size of some samples and consequently the reliability of the estimated damage distribution. Since some of the
low quality points can be outliers and can strongly affect the inferred curve, it is very important to evaluate the
uncertainty associated to each experimental point and consequently to the estimated probability of damage. The
bootstrap technique [Efron and Tibshirani, 1994] has been implemented and performed for each building
typology data set, resampling with substitution to estimate a standard deviation for each experimental
probability. Then the regression is performed using, as weights, the inverse of the estimated standard deviations.
Examples of the fragility curves obtained using this procedure are shown in Figure 5, which illustrates the fitting
and the effect of weights for the two typologies IMA2 and RMA2, differing only for the type of layout. It can be
noticed that the difference between non weighted and weighted curves decreases as the sample size and quality
of the data increase. In the right part of Figure 5, in particular, it can be observed that the curves of the highest
damage states are strongly affected by the weights, which determine a change of curvature with respect to the
non weighted case. The observation of the results obtained also allows a comparison between different
typologies: for the two cases of Figure 5, it can be noticed that, being all other parameters the same, a regular
layout of masonry (RMA2) is significantly less vulnerable than an irregular one (IMA2).

IMA2 23952 buildings RMA2 7611 buildings


100 100 100 100
non weighted weighted
90 90 90 90
80 non weighted 80 weighted 80 80
70 70 70 70
DS1
60 60 60 DS2 60
DS3
50 50 DS4
%

50 50
%

%
DS5
40 40 40 40
30 30 30 30
20 20
20 20
10 DS1 10
DS2 10 10
0 DS3 0
0 0.2 0.4 DS4 0 0.2 0.4 0 0
PGA [g] DS5 PGA [g]
0 0.2 0.4 0 0.2 0.4
PGA [g] PGA [g]

Figure 5: comparison of non weighted and weighted curves for IMA2 (left) and RMA2 (right)
typologies

5. DISCUSSION OF RESULTS

The parameters of the lognormal distributions, obtained for the different building typologies and for the five
considered damage states, are summarised in Table 2. Notice that not all the building typologies described in
Table 1 are reported in Table 2. In particular, the parameters concerning the typologies RC3 and ST are missing,
due to the limited amount of data which has not made possible to derive reliable parameters of the lognormal
distribution. Similarly, for few typologies, the parameters of DS5 are missing, since the corresponding
observations were very few and did not allow a reliable statistical elaboration.
As well known, the bootstrap technique, being based on the random generation of samples, produces values of
the estimated standard deviation that can vary from one run to the other. Since the weights used in the regression
can affect the final estimate of the lognormal parameters, the number of bootstrap samples has been increased
until a standard deviation of the resulting lognormal parameters was smaller than 0.01, which is visually
imperceptible on the curve.

7
Table 2: Parameters of the lognormal distribution for the adopted building typologies

Label DS1 DS2 DS3 DS4 DS5


μ σ μ σ μ σ μ σ μ σ
MX1 -4.83 4.73 0.81 4.28 3.31 5.24 2.76 3.33 2.07 2.24
MX2 -5.05 5.11 -0.43 2.32 -0.15 1.74 0.97 2.05 1.56 1.83
RC1 -0.17 2.80 0.64 1.42 0.38 1.08 0.23 0.85 - -
RC2 -1.34 2.06 -0.18 0.98 -0.26 0.64 0.31 0.91 0.67 0.93
RC4 -1.95 0.86 -1.16 0.44 -1.04 0.40 -0.82 0.43 -0.47 0.52
IMA1 -10.48 8.01 0.04 10.81 2.13 4.99 3.06 3.81 2.68 2.51
IMA2 -9.96 5.95 -3.02 5.30 -1.12 4.77 -0.21 2.95 0.04 1.84
IMA3 -4.20 4.74 1.12 4.72 2.54 4.09 3.73 3.52 1.93 1.89
IMA4 -7.32 5.74 -1.48 3.60 -0.22 3.41 1.12 3.21 2.03 2.65
IMA5 -10.61 7.45 -2.50 8.20 -0.51 2.98 0.74 2.46 - -
IMA6 -9.35 5.35 -2.91 4.36 -1.36 3.05 -0.41 2.10 -0.15 1.34
IMA7 -3.64 2.89 -1.48 2.59 -0.75 1.86 0.67 2.16 1.99 2.00
IMA8 -6.89 4.26 -2.39 2.08 -1.67 1.67 -0.69 1.64 0.53 1.79
RMA1 -5.13 6.82 6.28 8.44 5.18 5.11 3.70 2.90 - -
RMA2 -10.76 9.84 1.30 9.89 5.48 9.33 5.42 6.18 3.53 3.41
RMA3 -1.12 5.67 5.55 5.54 4.97 4.16 3.90 2.94 - -
RMA4 -6.66 12.63 4.57 7.17 2.92 3.96 3.44 3.22 2.33 1.94
RMA5 -4.77 6.34 0.98 6.34 1.79 4.17 0.46 1.45 - -
RMA6 -11.12 10.53 -0.67 8.99 0.80 5.13 1.66 3.48 4.13 3.54
RMA7 -2.02 2.51 -0.18 1.78 -0.01 1.48 0.13 1.11 0.11 0.55
RMA8 -6.03 9.12 -0.82 3.43 -0.35 2.33 -0.36 1.29 3.49 2.83

A scalar value of mean damage has been calculated combining the probabilities of exceeding the different levels
of damage. In particular, the expected mean damage has been computed using the following coefficients for the
various damage states: 1 for DS5, 0.8 for DS4, 0.6 for DS3, 0.4 for DS2 and 0.2 for DS1. This scalar parameter
is a function of PGA and allows relative comparisons among the different typologies. It is evident that, in order
to associate this parameter to an actual loss estimate, it can be necessary to recalculate it using different
coefficients, which take into account the realistic contribution of each damage level to the economic loss of the
building [e.g. Di Pasquale et al., 2005]. In Figure 6, the average value is plotted for some of the considered
typologies, in the PGA range of main interest for Italy. From the observation of this figure, some preliminary
conclusions on the relative vulnerability of different building typologies can be easily obtained. Notice that all
typologies plotted in Figure 6 are low rise buildings (≤ 2 storeys for masonry and mixed structures, ≤ 3 storeys
for concrete). Just as an example, it can be noticed that the expected mean damage of masonry with irregular
layout and without tie rods and tie beams (IMA2), at a PGA of 0.2 g, is almost 6 times higher than that of
reinforced concrete designed according to seismic regulations (RC1). Mixed structures (MX1) appear to be less
vulnerable than buildings of masonry with irregular layout, but more vulnerable than those in reinforced concrete
or masonry with regular layout.

0.6
MX1
0.5
RC1
mean damage

0.4
RC2

0.3 IMA1

IMA2
0.2
RMA3
0.1
RMA4
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
PGA [g]

Figure 6: value of expected mean damage, as a function of PGA, for some of the considered building
typologies

8
In order to allow a rapid relative comparison between different building typologies, other scalar parameters have
been calculated. Figure 7 shows a comparison among different typologies, using as a parameter the average
value of each damage state probability in three different ranges of PGA. This corresponds to assuming a uniform
hazard in each PGA interval, which is obviously not the real case, but which can be considered acceptable for the
purpose of a rough relative comparison between typologies. Note that the extreme values of the ranges
correspond to the PGA of Italian seismic zones 3, 2 and 1. To give an idea of the data quality, the number of
surveyed buildings for each typology is also plotted in the right part of the figure. It can be noticed that only the
typologies of Table 2 are shown in Figure 7. It can be observed that the distribution of damage among the
different damage states varies even significantly from one building typology to the other.

Figure 7: comparison between different building typologies, based on mean values over three different
ranges of PGA (colours correspond to damage states, as in Figure 3). In the right part, number of
buildings for each typology.

6. CONCLUSIONS

Fragility curves, derived from Italian post-earthquake surveys damage data for several building typologies are
presented. They are obtained from a very large data set, consisting of homogeneous data, all coming from Italian
observations. This is a significant advantage over other datasets used in the literature, which are either small [e.g.
Braga et al., 1982] or obtained assembling data from earthquakes all over the world [e.g. Rossetto and Elnashai,
2003].
The curves have been obtained by fitting the experimental data with a lognormal distribution, after resampling
using the bootstrap technique, to estimate the standard deviation associated to each experimental probability. The
inverse of this standard deviation is then used as a weight for the nonlinear regression. In this way, the relative
reliability of each experimental point, depending on the sample size and quality, is accounted for.
The curves obtained using this robust procedure are consequently very significant, at least for what concerns
Italy, and can be used in order to derive risk scenarios at different scales. It is important to underline once more
that the derived curves do not pretend to accurately represent the behaviour of a single building; however it is
believed that they represent in a reliable way the mean values of vulnerability for selected typological classes of
buildings. Moreover, it should be pointed out that they consider all the elements contributing to damage: they
include hence, in addition to the mean value of site effects and to the contribution of structural elements which
would not be considered in simplified analyses, also the uncertainty due to errors occurring during the surveys
and which, setting aside any technical consideration, contribute in modifying the damage distribution.
Since they are obtained from relationships between observed damages and PGA, estimated through an
attenuation law, these curves are perfectly compatible both with a probabilistic value of PGA derived from

9
probabilistic analyses and with deterministic simplified hazard studies, characterised by a value of magnitude
and distance.

7. ACKNOWLEDGEMENTS

The authors would like to express sincere thanks to the Italian Civil Protection Department (National Seismic
Survey) for providing the data collected during the post-earthquake surveys and in particular to Dr. Giacomo Di
Pasquale for his very precious suggestions in interpreting those data. The valuable help of Mauro Onida in the
development of the visual basic codes necessary to manage the databases is greatly acknowledged. The first
author would also like to thank the Regione Lombardia for providing partial funding for this study. Last but not
least, it is important to recall that this work has been originally hinted by Prof. Gian Michele Calvi and Dr.
Barbara Borzi of EUCENTRE.

8. REFERENCES

ATC-13 (1985), Earthquake Damage Evaluation Data for California, Applied Technology Council, FEMA
Contract No. EMW-C-0912, Redwood City, California.
Bernardini, A. (2004) Classi macrosismiche di vulnerabilità degli edifici in area veneto-friulana, Atti del XI
Congresso Nazionale “L’ingegneria Sismica in Italia”, Genova, Italy
Braga, F., Dolce, M. and Liberatore, D. (1982), A statistical study on damaged buildings and an ensuing review
of the MSK76 scale, Proceedings of 7th European Conference on Earthquake Engineering, Athens.
Calvi, G.M. (1999), A displacement based approach for vulnerability evaluation of classes of buildings, Journal
of Earthquake Engineering, 3. n° 3, 411-438.
Coburn, A. and Spence, R. (2002), Earthquake Protection (Second Edition), John Wiley and Sons Ltd.,
Chichester, England
Di Pasquale, G., Orsini, G. and Romeo, R.W. (2005), New developments in seismic risk assessment in Italy,
Bulletin of Earthquake Engineering, 3, 101-128.
Dolce, M., Zuccaro, G., Kappos, A. and Coburn, A.W. (1994), Report of the EAEE Working Group 3:
Vulnerability and risk analysis, Proceedings of the 10th European Conference on Earthquake Engineering,
Vienna, 4, 3049-3077.
Efron, B. and Tibshirani, R.J. (1994) An Introduction to the Bootstrap, Chapman & Hall/CRC.
Faccioli, E., Pessina, V., Calvi, G.M. and Borzi, B. (1999), A study on damage scenarios for residential buildings
in Catania city, Journal of Seismology, n° 3, 327-343.
FEMA (1999), HAZUS Earthquakes Loss Estimation Methodology, US Federal Emergency Management
Agency, Washington.
Giovinazzi, S. and Lagomarsino, S. (2004) A macroseismic model for the vulnerability assessment of buildings,
Proceedings of the 13th World Conference on Earthquake Engineering, Vancouver, Canada.
Grünthal, G. (ed.) (1998), European Macroseismic Scale 1998 (EMS 1998), Council of Europe, Cahiers du
Centre Européen de Géodynamique et de Sismologie, 15.
Gruppo di lavoro (2004). Redazione della mappa di pericolosità sismica prevista dall’Ordinanza PCM 3274 del
20 marzo 2003. Rapporto conclusivo per il Dipartimento della Protezione Civile, INGV, Milano-Roma,
aprile 2004, 65 pp. + 5 appendici
Milutinovic, Z.V. and Trendafiloski, G.S. (2003), RISK-UE: An advanced approach to earthquake risk scenarios
with applications to different European towns, WP4: Vulnerability of current buildings.
Orsini, G. (1999), A Model for Buildings’ Vulnerability Assessment Using the Parameterless Scale of Intensity
(PSI), Earthquake Spectra, 15, n° 3, 463-483.
Rossetto, T. and Elnashai A. (2003), Derivation of vulnerability functions for European-type RC structures based
on observational data, Engineering structures, 25, 1241-1263.
Rota, M. (2006) Fragility curves from post-earthquake survey data, Individual Study, European School for
Advanced Studies in Reduction of Seismic Risk (ROSE School), Pavia, Italy – in preparation
Sabetta, F. and Pugliese A. (1987), Attenuation of peak horizontal acceleration and velocity from Italian strong-
motion records, Bulletin of the Seismological Society of America, 77, n° 5, 1491-1513.
Spence, R.J.S, Coburn, A.W. and Pomonis, A. (1992), Correlation of ground motion with building damage: the
definition of a new damage-based seismic intensity scale, Proceedings of the 10th World Conference on
Earthquake Engineering, Madrid, Spain, 1, 551-556.

10

View publication stats

You might also like