International Journal of Disaster Risk Reduction 39 (2019) 101242

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction

journal homepage: www.elsevier.com/locate/ijdrr

Use of post-earthquake damage data to calibrate, validate and compare two T

seismic vulnerability assessment methods for vernacular architecture
Javier Ortegaa,*, Graça Vasconcelosa, Hugo Rodriguesb, Mariana Correiac, Tiago Miguel Ferreiraa,
Romeu Vicented
a
ISISE, Department of Civil Engineering, University of Minho, Guimarães, Campus de Azurém, 4800-058, Guimarães, Portugal
b
RISCO, School of Technology and Management, Polytechnic Institute of Leiria, Campus 2, 2411-901, Leiria, Portugal
c
CI-ESG Research Centre, Escola Superior Gallaecia, Vila Nova de Cerveira, Portugal
d
University of Aveiro, Aveiro, Portugal

ARTICLE INFO ABSTRACT

Keywords: The paper presents and discusses the application of two large scale seismic vulnerability assessment methods on
Seismic vulnerability assessment the island of Faial in Azores (Portugal). The two methods are specifically conceived to assess the seismic vul-
Vernacular architecture nerability of vernacular architecture. The first method follows a classical seismic vulnerability index approach
Stone masonry and is referred as SVIVA (Seismic Vulnerability Index for Vernacular Architecture). The second method is re-
Vulnerability index
ferred as SAVVAS (Seismic Assessment of the Vulnerability of Vernacular Architecture Structures) and it is a
1998 Azores earthquake
numerical tool intended to estimate the seismic capacity of vernacular buildings in terms of seismic load factors
associated with different structural damage limit states. The main reason behind the selection of Faial Island as a
case study was the availability of post-earthquake reports of the building stock after the 1998 Azores earthquake,
which allowed comparing the damage scenarios obtained using both methods with the post-earthquake damage
data and thus helped for the calibration and validation of the two methods. The application of both methods led
to a good fit between estimated versus observed damage grades, which validated their applicability as large-scale
first level approaches. Moreover, as the main outcome, the paper presents the novelties of the SAVVAS method,
which had not been applied before, and discusses its main advantages, namely: no need for calibration with
previous post-earthquake damage data, an enhancement of the prediction capabilities, a more individualized
evaluation of the buildings and the possibility to assess the seismic performance of the building in different
loading directions.

1. Introduction [5]. Typically, people tend to see vernacular construction technologies

as unsafe and eventually abandon and substitute them with modern
The increasing vulnerability of vernacular architecture has been ones. Subsequently, vernacular heritage, along with traditional building
already highlighted by different organisms, professionals and scholars knowledge, technologies and materials face the risk of disappearing due
[1–3]. The term vernacular commonly applies to non-engineered to this economic, cultural and architectural homogenization. Moreover,
buildings, typically self-constructed by the owner or the community, this progressive abandonment increases the vulnerability of vernacular
based on empirical knowledge and reflecting the tradition and life style architecture facing natural disasters, including earthquakes.
of a community, as part of a process that involves many people over Seismic vulnerability assessment methods for the built environment
many generations. The need for the valorization and the preservation of can play an important role on mitigating the risk faced by vernacular
vernacular architecture has been widely acknowledged because of structures against earthquakes. They are mainly aimed at estimating the
being a key-element for cultural identity and the fundamental expres- damage that a certain structure will suffer as a consequence of a seismic
sion of the culture of a community and its relationship with its territory event of a given intensity. In spite of the many seismic vulnerability
[2,4]. However, it is nowadays considered in many places as an ob- assessment methods existing in the literature, suitable for different
solete way of building and only valued as part of the region’s identity types of analysis and different goals, none has been specifically adapted

*
Corresponding author.
E-mail addresses: [email protected] (J. Ortega), [email protected] (G. Vasconcelos), [email protected] (H. Rodrigues),
[email protected] (M. Correia), [email protected] (T.M. Ferreira), [email protected] (R. Vicente).

https://doi.org/10.1016/j.ijdrr.2019.101242
Received 14 February 2019; Received in revised form 29 June 2019; Accepted 8 July 2019
Available online 11 July 2019
2212-4209/ © 2019 Elsevier Ltd. All rights reserved.
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

to the distinct characteristics of vernacular architecture. Based on this according to the approach followed to extract correlations between
gap in knowledge and intending to contribute to the awareness and damage and ground motion: (a) empirical methods are defined on the
protection of the vernacular heritage, two novel methods have been basis of post-earthquake damage data; (b) analytical methods define
previously developed by the authors [6] with vernacular structures as vulnerability functions on the basis of analytical and numerical studies;
their main target. The methods are particularly focused on the Portu- (c) expert-based methods rely on expert judgment; and (d) some
guese vernacular heritage, including stone masonry, fired brick ma- methods can be classified as hybrid, since they result from a combined
sonry, adobe and earthen constructions, which share many character- use of the previously described approaches.
istics with other vernacular constructions throughout the world. The Simplified seismic vulnerability assessment approaches aimed at
present paper intends to evaluate them through a practical application large scale analyses are typically empirical methods, relying on quali-
on a set of vernacular buildings in the island of Faial, in Azores (Por- tative data gathered from post-earthquake damage observation.
tugal). These two methods are: (a) Seismic Vulnerability Index for Correlations between damage and seismic motion are defined for dif-
Vernacular Architecture (SVIVA) method; and (b) Seismic Assessment ferent building typologies after observing the damage suffered due to a
of the Vulnerability of Vernacular Architecture Structures (SAVVAS) particular earthquake [9]. Even though there exist a great number of
method. empirical vulnerability functions in the literature developed from post-
Since both methods aim at the preservation of vernacular archi- earthquake damage data, there is a large variation in the procedures
tecture, they were conceived as large scale assessments, able to perform applied to collect the data (e.g. damage characterization, data quality,
analyses comprising a large number of buildings. The built vernacular etc.) or in the selected ground motion intensity [10]. There are also
heritage is rarely represented by single structures, but usually involves different ways of expressing this relationship. For instance, damage
a group of buildings and settlements within a rural region or within an probability matrices (DPM) can be formulated in a discrete form based
historical city center. The two methods are thus first level approaches on the concept that a particular structural typology has a similar
that can make use of simple more expedite inspections because they can probability of reaching a given damage state after an earthquake of a
rely on less detailed qualitative information related to a few para- given intensity. They were firstly proposed by Whitman et al. [11];
meters. This is another crucial matter given the typical lack of resources based on the damage caused by the 1971 San Fernando earthquake.
assigned to the study and preservation of the vernacular heritage. Afterward, more DPM were developed after the occurrence of different
Nevertheless, despite the expedited nature and ease of use of both earthquakes, using different intensity and damage scales [12–16]. An-
methods, they should be able to provide a robust estimation of the other possibility of describing the damage-motion relationship is
seismic capacity of vernacular buildings, as well as allow the individual through continuous vulnerability functions, first developed by Spence
assessment of the buildings. et al. [17]. The main problem to overcome for their derivation is that
After the brief introduction of the two methods, the paper presents both earthquake intensity and damage are typically expressed in a
Faial as the case study. The main objective of the present paper is the discrete form and not as continuous variables. However, different au-
calibration and validation of the two new seismic vulnerability assess- thors used different ways to describe the earthquake action and the
ment methods using a wide set of damage data collected after the 1998 damage in order to develop empirical vulnerability or fragility curves
Azores earthquake from Neves et al. [7] and Ferreira et al. [55]. The after post-earthquake surveys [18–21]. Empirical methods require a
data includes information of the existing traditional stone masonry large set of post-earthquake damage data which is usually not available.
building stock characteristic from the island and the damage survey Moreover, the obtained empirical correlations cannot always be ex-
carried out after the earthquake. The use of post-earthquake damage trapolated to other scenarios with a different building stock. None-
information allowed the comparison of the damage estimated after the theless, they are adequate for large scale analyses because they use
application of the two methods with the observed damage after the simple qualitative data that can be obtained from an expedite evalua-
earthquake, which led to the calibration and validation of both tion of the buildings. Another important limitation of these methods is
methods. This exercise was extraordinarily helpful for a better under- the difficulty of associating the damage observed to a single seismic
standing on the use of both methods to perform a seismic vulnerability event, since they are not able to take into account damage accumula-
assessment. tion induced by the subsequent occurrence of aftershock earthquakes
In the end, a detailed discussion of the advantages, drawbacks and [22].
limitations of each method is provided, showing a comparison of the Analytical methods use models representing buildings or building
performance of both methods. The evaluation of the applicability of the components and perform structural analysis to evaluate the seismic
methods for an efficient large-scale seismic vulnerability assessment of effect on the structures, in terms of damage. There are many methods
vernacular buildings is considered as the main contribution of the that can range different degrees of complexity depending on the type of
paper, since both methods had not been applied before. As a conclusion, model selected to simulate the structure and the analytical procedure
the paper discusses the potential of both methods to contribute to the adopted to perform the analyses. Analytical vulnerability curves can
preservation of the built vernacular heritage located in earthquake then be derived through regression analysis on the damage distribution
prone areas by evaluating the reliability of the methods in predicting data obtained after performing a large number of analyses on the
damage to vernacular buildings. The paper also focuses on evaluating models. Some common analytical methods existing in the literature are
the capability of the methods to identify the most vulnerable elements based on simplified mechanical models and limit state analysis [23] or
at risk and possible weaknesses and failure mechanisms of the building, kinematic limit analysis [24,25]. Others make use of more sophisticated
which is particularly important because they can eventually allow de- models and nonlinear static analysis procedures [26,27]. Many recent
fining and assessing appropriate structural retrofitting strategies at an studies use the equivalent frame model [28] and perform a high
urban or regional level as evidenced by Ferreira et al. [8]. number of nonlinear dynamic and static analyses in order to obtain
vulnerability curves for different masonry building typologies [29–31].
2. Overview of the two evaluated seismic vulnerability Analytical approaches are suitable to overcome the lack of post-earth-
assessment methods quake damage observations, but they require more detailed information
and a better understanding of construction details and materials to
The main components of seismic vulnerability assessment methods prepare the models. Thus, they can be very computationally expensive
are vulnerability curves of functions that express the probability of a to use on large-scale analysis comprising areas with buildings showing
building to suffer a certain degree of damage according to the earth- diverse construction characteristics. Moreover, they highly depend on
quake ground motion severity. Seismic vulnerability assessment the analytical model considered. For example, some of the mentioned
methods can be generally classified into four general categories equivalent frame models disregard the out-of-plane behavior of the

2
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

walls, which is a common failure mechanism for unreinforced masonry Table 1

buildings. Another important limitation is the simulation of the ground SVIVA vulnerability index formulation.
motion and, in the case of dynamic analysis, the record selection, which Parameter Class (Cvi) Weight (pi) Vulnerability index
highly influences the results of the seismic vulnerability assessment
[32]. On the other hand, the use of complex numerical modeling also A B C D
allows taking into account the effect of constructive and material P1. Wall slenderness 0 5 20 50 1.00 IV =
10
i = 1 Cvi × pi
P2. Maximum wall span 0 5 20 50 0.50
characteristics that cannot be typically considered in empirical P3. Type of material 0 5 20 50 1.50
methods, meaning that it is an appropriate tool to carry out parametric P4. Wall-to-wall 0 5 20 50 0.75
studies. It is noted that analytical methods should be always validated connections
with empirical observations. P5. Horizontal diaphragms 0 5 20 50 1.50 0 IV 500
P6. Roof thrust 0 5 20 50 0.50
Expert-based methods emerged as a result of the limited post-
P7. Wall openings 0 5 20 50 1.50
earthquake damage data in terms of different building typologies and P8. Number of floors 0 5 20 50 1.50 Normalized index
the high costs related to analytical approaches [33]. On the basis of P9. State of conservation 0 5 20 50 0.75 0 IV 100
expert opinion and previous knowledge, these methods estimate the P10. In-plane index 0 5 20 50 0.50
damage that a certain structure can suffer for a given seismic intensity
by analyzing the structural characteristics of the constructions and
classifying them into different building typologies [34,35]. Finally, 2.1. a. The SVIVA method
there are also hybrid methods that result from a combined use of the
previously described approaches, such as the vulnerability index The SVIVA method proposes a new formulation for the classical
method [36] and the macroseismic method [37], which are supported vulnerability index approach, firstly proposed by Benedetti and Petrini
by statistical studies of post-earthquake damage information, but also [36]; adapted to the characteristics of vernacular architecture. Ulti-
rely on expert opinion. mately, it consists of an adaptation of the hybrid approach followed by
This brief overview of existing seismic vulnerability assessment Vicente [41]; which combines the vulnerability index method and the
methods in the literature shows that recent works have applied analy- macroseismic method proposed by Giovinazzi and Lagomarsino [42].
tical approaches to derive fragility and vulnerability functions for dif- These approaches are all based on empirical post-earthquake damage
ferent building types and structures [25,38–40]. However, these ap- observation and expert opinion. In the case of the SVIVA formulation,
proaches have not been applied to vernacular masonry or earthen the quantification of each parameter’s influence on the seismic behavior
structures. Accurate numerical models with nonlinear material con- of vernacular buildings, which resulted in an updated definition of the
stitutive laws have not been yet applied to develop specific vulner- parameters' classes and weights were defined based on the previously
ability assessment methods for vernacular buildings. This gap in mentioned extensive numerical parametric campaign [6].
knowledge and the need to support traditional methods with analytical Vulnerability index methods provide a measure of the building
and numerical studies was therefore detected. The two methods eval- vulnerability under seismic loads through a dimensionless vulnerability
uated and applied in the present paper have been developed specifically index (IV) [43]. Table 1 shows the SVIVA vulnerability index formula-
for vernacular architecture using an analytical approach instead of an tion for vernacular architecture. As shown schematically in Fig. 1, the
empirical one. This process included an extensive numerical parametric method is composed of ten vulnerability parameters, which were se-
study based on detailed finite element modeling and nonlinear static lected based on existing vulnerability index formulations described in
(pushover) analysis intended to quantify the influence of a set of geo- the literature [44–48] and on the earthquake performance of verna-
metrical, structural, constructive and material parameters in the seismic cular constructions observed in past earthquakes [49,50]; [7,51,52].
response of vernacular buildings. A brief overview of the two seismic Each building is evaluated by providing a vulnerability class for
vulnerability assessment methods under evaluation is provided next, each of them. Four seismic vulnerability classes of increasing vulner-
but the reader is referred to Ortega [6] for an in-depth explanation of ability, from A (lowest) to D (highest), are defined for each parameter
their development. and associated with a qualification coefficient (Cvi). Following the

Fig. 1. Seismic vulnerability assessment parameters considered for the SVIVA and SAVVAS methods.

3
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

common vulnerability index formulations existing in the literature, the (Iv) and the macroseismic method (V) are different. Thus, following the
qualification coefficients are the same for all parameters. Class A is procedure described by Vicente [41]; IV had to be transformed into the
related to the lowest vulnerability class coefficient (Cvi = 0), while class vulnerability index used in the macroseismic method (V), using another
D is related to the highest vulnerability class coefficient (Cvi = 50). It analytical correlation:
should be noted that they have not been calibrated. The calibration
V= c + d × IV (2)
with the post-observation data takes effect over the weights assigned to
the parameters. This is intended to provide a formulation that is similar where c and d are again coefficients that can be calibrated for the type
to those existing in the literature and results in vulnerability index of buildings under evaluation based on post-earthquake observations.
values within a similar range. Thus, results can be comparable. Each The calibrating procedure for all coefficients a, b, c and d using existing
parameter is also associated to a weight (pi), reflecting its relative im- earthquake damage surveys is presented in section 4.1.
portance and ranging from 0.5 for the least important to 1.5 for the
most important ones. The vulnerability index (IV) is calculated as the 2.2. b. SAVVAS method
weighted sum of ten parameters using the equation shown in Table 1.
The value of IV ranges between 0 and 500 but, it is common, for ease of The SAVVAS method also makes use of a set of parameters related to
use, to normalize it to fall within a range between 0 (very low vul- geometrical, structural, constructive and material characteristics of
nerability) and 100 (very high). vernacular buildings shown in Fig. 1. However, this novel approach
The strategy of the parametric study used to define the parameters intends to estimate the maximum seismic capacity of buildings in
classes consisted of modifying a reference numerical model according quantitative terms. The results of the extensive numerical parametric
to the different parameters. Nonlinear static (pushover) analyses were analysis carried out to evaluate and quantify the influence of these
carried out for all the models constructed. The variations on the seismic parameters on the seismic response of vernacular buildings were com-
performance according to the variations in the parameters could be thus piled into a database [6]. Regression analysis was performed on the
analyzed and quantified. This procedure led to the definition of the database to extract correlations between the seismic capacity of the
seismic vulnerability classes. The definition of the parameters weight building and the key parameters shown in Fig. 1. As a result, the
was carried out by using statistical analysis. The results of the para- SAVVAS method is a numerical tool consisting of different formulations
metric study were assembled into a database. Multiple linear regression that allow defining the seismic capacity of the building through seismic
analysis led to assess the relative importance of the different parameters load factors expressed as accelerations (in terms of g) associated with
in defining the seismic performance of the buildings analyzed. different structural damage limit states, using as input simple variables
Performing a seismic vulnerability assessment requires an expres- based on the ten key seismic vulnerability assessment parameters. Thus,
sion that is able to correlate the estimated vulnerability index of the the SAVVAS method is intended to be an analytical approach developed
building (IV ) with the expected damage to be suffered for different using numerical and statistical analysis.
seismic inputs. As previously stated, the SVIVA method follows the The SAVVAS formulation and procedure is shown in Table 2. The
approach defined by Vicente [41]. Therefore, it uses the analytical first step of the SAVVAS method is partially common to the SVIVA
expression from the macroseismic method developed by Giovinazzi and method, namely the assignment of seismic vulnerability classes to the
Lagomarsino [42]: parameters. However, as shown in Table 2, while some of the para-
meters are defined by assigning a seismic vulnerability class from 1 to 4,
I+ aV b directly associated to the classification from A to D defined also for the
µD = 2.5 1 + tanh
Q (1) SVIVA method, others had to be defined through specifying different
quantitative attributes. For example, P2 (maximum wall span) can be
where I is the seismic input in terms of macroseismic intensity, V is the directly defined by the span (in m), instead of by the vulnerability class.
vulnerability index and Q is the ductility index, which is an empirically The same occurs for P1, defined by the wall slenderness ratio (λ = h/t)
defined index that takes into account the ductility of a determined and P8, defined by the number of floors (N) of the building. P10 refers
construction typology, typically ranging from 1 to 4 [48]. Coefficients a to the in-plane index (γi) and it is also defined quantitatively as the ratio
and b should be calibrated for the set of buildings under analysis when between the in-plane area of earthquake resistant walls in each main
post-earthquake damage data is available. This analytical expression direction (Awi) and the total in-plane area of the earthquake resistant
can be used to build vulnerability curves for the subsequent seismic walls (Aw): i = Awi / Aw . Parameter P7 refers to the amount and area of
vulnerability evaluation and estimation of losses. It should be noted walls openings and was further divided into two parameters, aiming at
that the vulnerability indexes used by the vulnerability index method distinguishing between: (1) P7a, ratio between the maximum area of

Table 2
SAVVAS formulation and procedure.
Step 1 Definition of the seismic vulnerability assessment parameters

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10
λ s P3 [1–4] P4 [1–4] P5 [1–4] P6 [1–4] P7a P7b N P9 [1–4] γi
Step 2 Calculation of the load factors associated to the limit states in each main direction i (in terms of g)
LS1i = e (1.97 0.06 0.1s 0.68 ln(P 3) 0.14P 4 0.28P5 0.39 ln(P6) 3.43P 7b 0.82 ln(N ) 2.27 ln(P 9) + 0.63P5P 7b) c
LS 2i = 0. 16 × LS1(g ) + 0. 78 × LS3(g )
LS3i = e (2.16 0.04 0.05s 0.24P3 0.16P 4 0.28P5 0.08P 6 + 0.3P 7a 2.79P 7b 0.37N 0.15P 9 + 0.74 i + 0.44P5P 7b)
Step 3 Calculation of the global load factors defining the limit states of the building (in terms of g)
LS1 = min (LS1i )
LS 2 = min(LS 2i)
LS3 = min (LS 3i)

4
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Fig. 2. Definition of the limit states according to the pushover curve.

openings in the walls perpendicular to the loading direction and the The expressions from Step 2 of the SAVVAS formulation that allow
total surface area of the walls; and (2) P7b, ratio between the area of calculating the load factors were obtained from the multiple linear re-
wall openings in all in-plane resisting walls and the total surface area of gression analysis performed on the database. These regression models
all in-plane resisting walls. The remaining parameters, including the obtained showed a good correlation between the seismic capacity of the
type of material (P3), the quality of the wall-to-wall connections (P4), building and the ten key parameters selected [6]. It should be also
the horizontal diaphragms (P5), the roof thrust (P6) and the previous noted that the load factors can be calculated for the four main direc-
structural damage (P9), are defined as a function of their class, in tions of the building (+/-X and ± Y). This is intended to provide a
qualitative terms. Thus, they are described in a discrete form, assuming more accurate description and understanding of the seismic behavior of
four countable numbers from 1 to 4. the evaluated vernacular buildings, as well as a better estimation of
With respect to the load factors defining the structural limit states their most vulnerable direction. However, in order to have a global
(LS1, LS2 and LS3), they are associated to specific damage levels ex- seismic assessment of the building, the minimum values for each LS
hibited by the structure. They were determined according to the push- obtained among the four resisting directions are given as the global
over (capacity) curves obtained from the parametric analyses, which is a load factors defining the seismic vulnerability of the building. This is
relation between the load factor (ratio between the horizontal forces at the last step of the procedure and, as a result, the SAVVAS method
the base and the self-weight of the structure) and the displacement at a provides an estimation of the minimum load that will cause the
control node (taken as the node where the highest displacements occur), building to reach the different limit states. Since the load factors related
see Fig. 2. They provide information of both load and deformation ca- with the different structural damage limit states are expressed as ac-
pacity of the building, in terms of stiffness and ductility. Nevertheless, celerations, they can be used in a straightforward way to eventually
the basis of comparison of the seismic capacity of the SAVVAS method is correlate the seismic action in terms of peak ground acceleration (PGA)
defined in terms of load capacity. Therefore, the limit states are estab- with the expected damage.
lished according to the seismic actions that can cause the building to
reach the different structural limit states. They are expressed as an ac-
3. Damage data after the 1998 Azores earthquake
celeration (in terms of g). LS1 can be associated to an Immediate Occu-
pancy Limit State. Before this limit, the structural behavior of the
The 1998 Azores earthquake struck the central group of the Azores
building remains in the elastic branch and the structure can be con-
Archipelago with a moment magnitude Mw = 6.2 , mainly striking Faial,
sidered as fully operational. LS1 thus corresponds to the formation of the
Pico and San Jorge islands. The earthquake reached high levels of de-
first cracks in the structure, characterizing the end of the elastic response.
struction and affected more than 5000 people, causing 8 fatalities and
LS2 is associated to a Damage Limitation Limit State, as it depicts the
leaving 1500 persons homeless [53]. A Modified Mercalli Intensity
transition between a point where the structure is still functional, re-
(MMI) scale distribution map for the Faial Island was proposed by
taining most of its original stiffness and strength, showing minor struc-
Zonno et al. [54] based on post-earthquake damage survey campaigns,
tural damage and cracks, and a state where significant damage is visible
see Fig. 3. Nevertheless, it is noted that the construction of this docu-
so that the building could not be used after without significant repair.
ment is subjected to uncertainties and Zonno et al. [54] argues that
LS3 can be referred as Life Safety Limit State and is defined by the load
some locations might have been subjected to higher intensities than
factor and displacement corresponding to the attainment of the building
those plotted on the map.
maximum resistance. As a result, the building has lost a significant
amount of its original stiffness, but is supposed to retain some lateral
strength and, in the case of masonry structures, they still may show a 3.1. a. Building stock characterization
large margin against collapse in terms of displacements. Nevertheless,
they should not be used after the earthquake. It is noted that LS4 is as- The seismic event was followed by the collection of extensive data
sociated to the Near Collapse Limit State, but was excluded because it on the effects of the earthquake on the building stock of the islands.
corresponds to the point where the building maximum strength is re- Neves et al. [7] focused on the detailed characterization of the buildings
duced 20%, thus being mathematically dependent on LS3. The load in the Faial Island and particularly presented a detailed study of the
factor associated to the collapse of the building is thus not defined ac- construction systems that characterize the traditional architecture of
cording to the pushover curve and was calibrated in a subsequent step the island, whose structure is mainly composed of stone masonry load
using post-earthquake damage data (see Section 4.2.). bearing walls, timber floor diaphragms and timber roof trusses. This is
particularly adequate, given that the two seismic vulnerability

5
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Fig. 3. MMI scale distribution map of Faial Island indicating the administrative subdivision of the island into the different districts (adapted from Ferreira et al. [55].

assessment methods proposed are mainly addressed for this structural the reconstruction process with information of the original and retro-
typology. Neves et al. [7] also proposed a detailed damage classification fitted structure (including plans, damage reports and photographs) to
for this traditional masonry building stock by identifying the main very limited information with barely a damage report fulfilled on-site or
damage patterns surveyed. Moreover, the earthquake also attracted a a couple of photographs.
significant amount of scientific research dedicated to the character-
ization of the mechanical properties of the traditional construction
techniques from the island [56–58]. This vast amount of information 3.2. b. Damage classification
gathered and produced on the seismic performance of traditional
Azorean masonry constructions after the 1998 earthquake makes this A general damage distribution of 3154 traditional masonry build-
case study very appropriate for the calibration of the two seismic vul- ings on Faial and Pico Islands was presented in Neves et al. [7]. The set
nerability assessment methods proposed. Actually, it has also been of 88 buildings in Faial selected for this study was meant to include
previously used to calibrate other seismic vulnerability assessments buildings presenting a wide variation in terms of the observed grade of
methods [7,55]. damage. The classification of the damage observed in each building was
The same set of 88 masonry buildings used by Ferreira et al. [55] carried out according to the EMS-98 European Macroseismic Scale [15]
was also selected for the application and calibration of the two methods and is presented in Table 3 as a reference. This damage classification
proposed in this work. This selection includes comprehensive in- was chosen because the macroseismic method [42] is based on the
formation on different representative traditional masonry construction EMS-98 macroseismic scale defined by Grünthal [15]. Thus, the mean
types scattered throughout various villages in Faial Island. Both rural damage grade (μD) estimated using this approach directly relates to the
and urban building types are present in the selection, see Fig. 4. The classification shown in Table 3. The same damage grade is also the main
reader is referred to Costa and Arêde [59] and Neves et al. [7] for a output of the SAVVAS method, allowing the direct comparison between
more detailed description of these buildings in terms of construction the results obtained using both methods.
systems and materials. The documentation available for each of these The buildings were thus classified in terms of damage using the data
88 buildings varied widely: from very detailed reports drafted during available. It is worth noting that a damage assessment is always sub-
jective and depends on the judgment of the evaluator. Besides, as

Fig. 4. Examples of typical traditional Azorean masonry construction types in the island of Faial present in the selection: (a) one-floor rural building; (b) two-floor
rural building; and (c) three-floor urban building.

6
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Table 3
Damage grades adopted for the study based on the EMS-98 [15].
Damage grade Description

0 No damage No observed damage

1 Negligible to slight damage No structural damage and/or slight non-structural damage: hairline cracks in very few walls, fall of small pieces of plaster, fall or loose stones
from upper parts of buildings in very few cases
2 Moderate damage Slight structural damage and/or moderate non-structural damage: cracks in many walls, fall of large pieces of plaster, partial collapse of
chimneys
3 Substantial to heavy damage Moderate structural damage and/or heavy non-structural damage: large and extensive cracks in most walls, roof tiles detach, chimneys
fracture at the roof line, failure of individual non-structural elements (partition or gable walls)
4 Very heavy damage Heavy structural damage and/or very heavy non-structural damage: serious failure of walls, partial structural failure of roofs and floors
5 Destruction Very heavy structural damage: total or near total collapse

previously stated, the existing information on the buildings is variable the opportunity of obtaining mid-values in between the 6 damage
and, in some cases, limited. Therefore, in order to minimize un- grades (e.g. 3.25), which allowed a better comparison with the damage
certainties and to have a more robust and reliable assessment, four values resulting from the two seismic vulnerability assessment methods
experts carried out the evaluation of the damage grades for the 88 that express damage as a continuous variable.
buildings independently. The results were then analyzed and compared. Fig. 5 shows several examples of buildings classified under the five
The final damage classification adopted for each building was the mean damage grades. None of the buildings in the set was considered as grade
value obtained from the four evaluations. This approach also provided 0, since all of them presented at least slight non-structural damage. The

Fig. 5. Examples of evaluated buildings belonging to each damage grade from the EMS-98 scale.

7
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

qualitative judgment of the person conducting the assessment, since

different persons may reach to different classifications.
The definition of the classes can be particularly difficult for para-
meters that are not easily evaluated from the exterior, such as the
quality of the wall-to-wall connections or the type of horizontal dia-
phragm. In this particular case, for example, it should be also noted that
interpreting the class for parameter P9, which refers to the previous
structural damage in the building, was very difficult, since all the pic-
tures available correspond to the state of the buildings after the
earthquake. Thus, it was decided to establish that all buildings fell
within class A for parameter P9, so that this parameter does not have a
relative influence in the results. As abovementioned, the information
available for some of the buildings barely consisted of a brief damage
report and a couple of photographs. For these buildings with limited
information, the data obtained from other buildings from the set with
more detailed information served as the basis for extrapolation. Also,
the detailed construction characterization of the masonry walls, timber
roofs and timber floors, conducted by Neves et al. [7]; was very helpful
for the determination of some parameters classes.
Fig. 6. Distribution of the evaluated buildings according to the estimated da-
mage grade.

4.1. a. Seismic Vulnerability Index for Vernacular Architecture (SVIVA)

examples are compared with reference drawing provided by the EMS- method
98 scale. Fig. 6 shows the distribution of the assessed buildings ac-
cording to their estimated damage level. Damage levels in the graph are The application of the formulation to the 88 buildings from Faial
used as thresholds and include all buildings that have not reached the resulted in the vulnerability index distribution presented in Fig. 7. The
following damage level (i.e. buildings whose damage grade was esti- mean value of the seismic vulnerability index (IV ) obtained is 43.22
mated as 3.5 are included within damage grade 3). The graph shows with a standard deviation value (STD) of 7.1, which results in a coef-
that the majority of the buildings (over 65%) did not reach damage ficient of variation (CoV) of 16%. The minimum and maximum values
grade 3 and, thus, did not present substantial structural damage. of IV are 21.5 and 55 respectively. The little variation within the index
shows clearly that most of the buildings assessed belong to similar
4. Calibrationa. Building stock characterization and validation of construction typologies. The main typological difference occurs be-
the methods tween the rural and urban buildings. However, even between both
construction types, the majority of the classification of the parameters
As previously mentioned, both seismic vulnerability assessment coincides.
methods that are evaluated in the present paper make use of geome- The expected mean damage grade ( µD ) can be estimated for each
trical, structural, constructive and material parameters to estimate the building using Eq. (1), as a function of the building vulnerability and
building vulnerability. Thus, the parameter survey (i.e. classification of the seismic input. It is noted that μD also refers to the damage classifi-
each parameter in terms of seismic vulnerability class or quantitative cation from EMS-98 shown in Table 3. Thus, the results obtained can be
attribute) is a crucial step for the application of the two proposed compared with the observed damage level on the buildings after the
methods. It is worth highlighting that, just as with the damage classi- earthquake (Fig. 6). The buildings were grouped by location and in-
fication, the damage data available for each building is not always tensity, following the district subdivision shown in the intensity map
complete enough to carry out a sound parameter survey. Therefore, (Fig. 3). The seismic input (I) from Eq. (1) is also expressed in terms of
some assumptions had to be made in order to decide the class for some the EMS-98. Following the recommendations of Musson et al. [60]; the
of the parameters. The parameter survey is much dependent on the degrees from the MMI scale depicted in the intensity map from Fig. 3

Fig. 7. (a) Vulnerability index (IV) distribution; and (b) parameter class distribution.

8
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

can be directly correlated with the degrees from the EMS-98 scale, The ductility index (Q) is empirical parameter and depends on the
acknowledging a certain degree of subjectivity involved within this construction typology evaluated. In this study, a value of 2.0 is assumed
assumption Ferreira et al. (2017), Thus, a scale V in MMI scale can be based on recommendations of other authors dealing with load bearing
associated to a scale V in the EMS-98 scale. masonry wall construction types [47,55]. This factor defines the slope
The initial mean damage grade estimated using the original for- of the vulnerability curve and the value of 2.0 adopted also proved to
mulation and the coefficients proposed by Vicente [41]; where a, b, c, d provide the most accurate approximation. The fitting process resulted
and Q were 6.25, 12.7, 0.56, 0.0064 and 3 respectively, did not match in a significant improvement in the correlation between the estimated
well the observed results (Fig. 8). Therefore, a curve-fitting process was and observed damage. Fig. 8 shows side by side plots of the mean da-
applied in order to find a better approximation between the observed mage grade observed ( µD ) versus the vulnerability index (IV ), with the
damage-vulnerability index point cloud and the vulnerability curves. corresponding vulnerability curves built using the original formulation
Thus, Eq. (1) and Eq. (2) had to be calibrated for the buildings under and the calibrated one, for the three different macroseismic intensities
analysis. The availability of post-earthquake damage data allows the registered in the island (VI, VII and VIII). It should be noted that, since
comparison between the estimated and the observed damage. The fit- only a few buildings within the set correspond to areas where the
ting process was carried out using CurveExpert Pro software [61]. This macroseismic intensity level registered was VIII, the improvement re-
software automatizes the process of finding the best fit allowing the sulting from the fitting process is less optimized (Fig. 8c). The sig-
definition of a custom regression model based on the analytical ex- nificant differences between both curves illustrate the inherent un-
pressions shown in Eq. (1) and Eq. (2). Subsequently, these two ana- certainty of these formulations to estimate damage, since they depend
lytical expressions could be calibrated to better represent the seismic on parameters that can only be calibrated with post-earthquake damage
behavior observed for this particular type of buildings, by means of data. This fact also highlights the importance of interpreting the results
varying the coefficients that define both expressions. The resulting ca- statistically and in comparative terms, as a first-level assessment that
librated expressions are shown below, highlighting in bold the updated highlights those buildings that are more vulnerable than others and
coefficients: require further more detailed evaluation. With regard to the partial
distributions of IV for each intensity level, a mean value of 41.1, 46.1
I+ 6.25V 12.7 and 41.2 were obtained for IEMS 98 = VI , VII and VIII , respectively. The
µD = 2.5 1 + tanh
Q (3) similar values obtained confirmed a construction typology homogeneity
of the set of buildings evaluated and showed that the fact that some
V= 0.46 + 0.012 × IV (4) buildings suffered a higher level of damage should be associated to the

Fig. 8. Observed damage versus mean damage grade estimated using the original and updated expressions for the construction of the vulnerability curves, grouped
by the different macrosesimic intensities.

9
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Fig. 9. (a) Predicted versus observed damage grades; and (b) residuals versus observed damage grades.

higher accelerations registered in those areas. 4.2. b. Seismic Assessment of the Vulnerability of Vernacular Architecture
The damage estimation achieved using this new proposed vulner- Structures (SAVVAS) method
ability index formulation was considered satisfactory. The estimated
versus observed damage plot is shown in Fig. 9a, while Fig. 9b presents The SAVVAS method was applied on the same 88 buildings, fol-
the residual versus observed damage. The value of the coefficient of lowing procedure specified in Table 2 and leading to the load factor
determination (R2) obtained reaches 0.605. This coefficient measures distributions shown in Fig. 10. The mean values of the load factors
how well the model fits the actual data. A value of 0.605 can be con- obtained are 0.13 g, 0.22 g and 0.25 g for LS1, LS2 and LS3 respectively,
sidered high for these simplified seismic vulnerability assessment with a standard deviation (STD) value of 0.06 g, 0.08 g and 0.09 g,
methods. The errors are also low, showing a maximum error in the which result in coefficients of variation (CoV) of 47%, 37% and 36%.
prediction of 2.24, but a Mean Absolute Error (MAE) value of 0.56 and These results show significantly greater variations than the ones ob-
a Root-Mean-Squared Error (RMSE) value of 0.71. The graph from tained from the vulnerability index method, which suggests that the
Fig. 9b shows that the level of damage is predicted within a maximum SAVVAS method is able to distinguish the capacity of the buildings that
difference of 1 level for the great majority of the buildings, with the previously had the same vulnerability index (IV ). Therefore, the
exception of a few cases. Acknowledging the uncertainties inherent to SAVVAS method seems to be able to detect more precisely the differ-
the whole prediction process, namely the attribution of the macro- ences in the seismic performance of the different buildings, even though
seismic intensities, the assignment of a level of damage and the selec- they belong to a very similar construction typology. It is noted that a
tion of the parameter classes to the different buildings, it should be detailed comparison between the results obtained using both methods is
highlighted that the results show a good prediction capability. The provided later.
model is able to recognize the most vulnerable constructions and pro- A first seismic assessment of the buildings can be carried out just by
vide a good estimate of the damage that each building might suffer for comparing the seismic load factors obtained with the seismic demand
earthquakes of different intensities. established by the code. For Faial Island, the value of reference peak

Fig. 10. Load factor distributions for the three limit states: LS1, LS2 and LS3.

10
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Table 4
Statistics from the parametric survey and the estimated load factors defining each limit state.
Variables Units Minimum Maximum Mean Median Mode STD CoV (%)

Parameters P1 λ 3.71 7.07 5.12 5.00 5.38 0.64 12.33

P2 m 2.85 17.40 7.38 6.52 4.50 3.30 44.74
P3 Class 2 4 3.47 4 4 0.66 19.04
P4 Class 1 4 2.90 3 3 0.67 23.00
P5 Class 2 4 3.68 4 4 0.49 13.42
P6 Class 1 4 1.33 1 1 0.92 69.43
P7a P7a 0 0.49 0.09 0.06 0 0.10 110.64
P7b P7b 0 0.36 0.07 0.06 0 0.06 96.43
P8 N 1 3 1.49 1 1 0.63 42.08
P9 Class 1 1 1 1 1 0 0.00
P10 γi 0.19 0.71 0.39 0.40 0.45 0.10 24.34
Load factor LS1 g 0.04 0.38 0.13 0.12 – 0.06 46.67
LS2 g 0.09 0.63 0.22 0.20 – 0.08 36.72
LS3 g 0.11 0.74 0.25 0.23 – 0.09 36.13

ground acceleration (PGA) is 0.25 g (NP EN1998-1 2010). About 60% which is associated mainly to the fact that many buildings are built in a
of the buildings present a load factor corresponding to LS3 below slope. Therefore, different sides of the buildings can present different
0.25 g, which means that their maximum capacity would be exceeded heights, which results also different values for this parameter within the
by the design load action of an earthquake with the characteristics same building.
defined by the code. This is a first indicator that reveals the vulner- As abovementioned, it should be here noted that in cases where
ability of the buildings in the island. Moreover, most of the buildings there is a limited amount of information available, some of the values
are prone to suffer structural damage. For 95% of the buildings eval- assigned to each parameter had to be inferred from a reduced set of
uated, the load factor corresponding to LS1 obtained is considerably pictures. The conditions observed in other buildings with more detailed
lower than 0.25 g (Fig. 10). information served as reference. However, there was no way to know if,
Table 4 shows the statistics obtained for the vulnerability para- for example, there were intermediate resisting walls that can reduce the
meters corresponding to the surveyed buildings. Table 4 also includes span value adopted for P2 or, if the condition of the wall-to-wall con-
the statistics from the computed global load factors defining the three nections was good. For these buildings, the analysis of the damage
limit states. Similarly to what we could observe in Fig. 7, the variations developed during the earthquake helped also to infer the classification
found for some parameters are very small, particularly for parameters of some of the parameters, taking into account that the damage is ty-
P1 (wall slenderness), P3 (type of material) and P5 (horizontal dia- pically associated to deficiencies of the building. As an example, some
phragms). Therefore, the majority of the buildings belong to a similar photographs depicted the collapse of some walls allow detecting defi-
construction type that consists of thick load bearing irregular masonry cient wall-to-wall connections otherwise impossible to detect by a vi-
walls with flexible timber horizontal diaphragms. The higher deviation sual survey from the outside of the building, see Fig. 11.
shown by the remaining parameters can be attributed to: (a) the The use of the real post-earthquake damage information available
parameters are classified differently for each main direction; and (b) from the 1998 Azores earthquake was in fact very useful to gain
parameters are more specifically classified and have a wider range of knowledge on how to carry out the parameter survey. The classification
variation. For example, the variation observed for parameter P6 is due of some parameters was not straightforward in many cases. Some as-
to the fact that, within the same building, some walls might be con- sumptions were considered in the present work that can be helpful for
sidered to receive the roof thrust while others do not. This is common the future application of the method, including: (1) the wall slenderness
when buildings have gable roofs (as is the case for most of the buildings might vary among the different walls of the building, the minimum
under analysis), where only two walls can receive the possible thrust observed was considered for all directions; (2) whenever walls showed
from the roof. Regarding parameter P2, walls get to span distances over different number of floors along their length because of being con-
15 m in several cases, which also confirms a clear trend for the build- structed in a slope, the maximum height was always considered; or (3)
ings in the island to be very slender in plan ( i > 0.75). The coefficient of the value of the in-plane index considered in all directions was always
variation (CoV) for the two parameters addressing wall openings is very the minimum calculated, unless the building presents a class A or B type
high because of the low value of the mean. However, the buildings of diaphragm (P5), able to redistribute the load to the earthquake re-
typically present few openings, with some exception of those located in sistant walls in the loading direction. These assumptions were always
urban areas, which can show facades with up to 49% of wall openings. aimed at taking into account the worst scenario.
With respect to the number of floors, there is also a greater variation, The next step after the application of the SAVVAS method consists

Fig. 11. Examples of collapsed buildings showing deficient wall-to-wall connections.

11
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Fig. 12. PGA maps computed by Zonno et al. [54] for the 1998 Azores earthquake assuming two different possible epicenters.

of the estimation of damage grade based on the EMS-98 scale, corre- damage observed after 1998 Faial earthquake (Fig. 12b).
lated with the calculated load factors associated to the three limit states After the definition of the seismic input, a correlation between
defined. In a first step, the SAVVAS method requires that the seismic seismic input, load factors (expressed in g) associated to the structural
input is expressed in terms of PGA instead of macroseismic intensity, so limit states and mean damage grade ( µD ) has to be defined. Results need
that it can be compared with the values of load factor. The existing data to be expressed in terms of the same EMS-98 damage grade scale in
for the 1998 earthquake included strong-motion records and a large order to enable the output of the SAVVAS method to be comparable
collection of post-earthquake damage in the building stock. Based on with other seismic vulnerability assessment methods, such as the
this information, Zonno et al. [54] prepared possible PGA maps for the macroseismic method. Fig. 13 shows the equivalence between the
earthquake, according to two possible epicenter locations (Fig. 12), but structural limit states defined from the pushover curve and EMS-98
stated that the second epicenter considered (Fig. 12b) best reproduced damage grades.
the observed effects of the Faial earthquake. Fig. 13 shows that damage grade 0 was removed from the scale. The
The previously shown MMI map (Fig. 3) used for the application of SAVVAS method does not detect non-structural damage. Grades 0 and 1
the SVIVA method was constructed based on the surveyed damage data, are the same and represent the starting point of the scale representing
as abovementioned. Subsequently, in order to have comparable results, no structural damage. The load factor defining LS1 delimits the point
the PGA values could be inferred from the values of MMI shown in the where the building reaches damage grade 2 and, thus, for values of PGA
map. There are many empirical relationships between seismic intensity higher than LS1, the building is assumed to start presenting slight
and acceleration. These expressions are typically derived based on data structural damage. Similarly, LS2 is associated to damage grade 3 and
from previous earthquakes in different locations and the macroseismic LS3 with damage grade 4. The correlation with the 5th damage grade
intensity is correlated with the logarithm of the ground shaking para- that refers to the total or near collapse of the structure was not
meter (such as the PGA). Table 5 shows some correlation relationships straightforward. An empirical factor was established to define a load
between macroseismic intensity and PGA existing in the literature, as that would cause the collapse of the building and could be related to
well as the results obtained when applied to the MMI map from Fig. 3. damage grade 5. This factor was calibrated using the damage data from
None of these expressions were derived based on previous earthquake the 1998 earthquake to fit better the collapse observed and was finally
data from Azores and all lead to different values of PGA, showing sig- set as 1.25 times the value found for LS3. The final damage values for
nificant scatter, clearly illustrated by the high coefficient of variation the ranges of PGA between limit states are obtained from simple linear
(CoV) shown in the table. Therefore, there is a great amount of sub- interpolation in order to provide a continuous variable.
jectivity of adopting one expression on top of another for the present Once this correlation was established, the level of damage was as-
study. As a result, this study adopts as a reference the PGA map com- sessed for the 88 buildings evaluated. The estimation of damage
puted by Zonno et al. [54] using the epicenter that fitted best the achieved using the SAVVAS method was deemed considerably accurate,
clearly outperforming the prediction capability of the SVIVA method.
Table 5 Fig. 14a gives the estimated versus observed damage plot, while
Intensity-PGA relationships from the literature. Fig. 14b presents the residual versus observed damage. The value of R2
obtained from the correlation between observed and predicted damage
PGA (g)
reaches 0.802 and was considered quite satisfactory. The errors are also
Reference Correlation I = VI I = VII I = VIII reduced, showing a maximum error in the prediction of 2.33 but a MAE
Murphy and O’Brien [62] 0.06 0.10 0.18 of 0.32 and a RMSE of 0.71. The graph from Fig. 14b shows that the
Guagenti and Petrini [63] 0.03 0.06 0.10
level of damage is predicted within a maximum difference of less than
Margotini et al. [64] 0.07 0.12 0.19
Theodulis and Papazachos 0.11 0.22 0.44 0.5 in the damage level for the great majority of the buildings.
[65] Fig. 15 presents the vulnerability curves constructed using the
Decanini et al. [66] 0.11 0.18 0.32 SAVVAS method as a function of the load factor defining LS3. The
Wald et al. [67] 0.13 0.24 0.44 observed damage-LS3 point cloud is also presented together for re-
Marin et al. [68] 0.02 0.05 0.14
ference. The curves fit well the points representing the damage ob-
Faccioli and Cauzzi [69] 0.05 0.18 0.57
Gómez Capera et al. [70] 0.08 0.12 0.19 served. Results are plotted for the three different PGA associated to the
Tselentis and Danciu [71] 0.09 0.17 0.33 three different macroseismic intensities registered in the island. It
Bilal and Askan [72] 0.03 0.06 0.11 should be noted that the estimation of damage depends not only on the
Gómez Capera et al. [73] 0.07 0.14 0.26
load factor defining LS3, but also on LS1 and LS2, and the relative
Zanini et al. [74] 0.06 0.15 0.42
Mean (CoV) 0.07 0.14 (44%) 0.28 (52%)
differences among them, as shown in Fig. 13. Thus, the vulnerability
(48%) curves constructed with the SAVVAS method may vary according to

12
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Fig. 13. Correlation between the seismic input (PGA), SAVVAS limit states and EMS-98 damage grades.

this. Four possible vulnerability curves are presented with varying (Fig. 17). Besides, the requirement of numerical values does not gen-
difference between LS1 and LS3, which define the slope of the curves erate an increment in the complexity of the application of the tech-
and can be related to the ductility of the structure. The mean difference nique, since the parameters are defined by simple ratios that are usually
between both limit states is 0.1g and that is why the vulnerability curve also required for the definition of the classes for the vulnerability index
constructed using this difference is highlighted in red. The majority of method. It should be noted that in addition to the higher value of R2
the points representing the damage observed lie within or very close to obtained, results are reliable because of the low errors obtained (MAE
these curves, which confirms the good prediction capability of the of 0.32) and the fact that there is not a systematic underestimation or
SAVVAS method. overestimation of the damage observed (Fig. 14b).
Another main advantage of the SAVVAS method is the fact that it
4.3. c. Comparison between the two methods does not require the calibration of the vulnerability curves performed
for the SVIVA method. The coefficients from the expression defined by
Both seismic vulnerability assessment methods are evidently related the macroseismic method (Eq. (1)) had to be redefined based on the
since they are based on the same parameters and were developed on the observed damage in order to establish Eq. (3) and Eq. (4). As shown
basis of a numerical parametric study [6]. The classes of the parameters during the assessment performed (Fig. 8), the discrepancies can be quite
are also common to both methods. Thus, a strong correlation between high from using the original formulation and the calibrated ones. This is
the vulnerability index (IV ) obtained with the SVIVA method and the an important limitation when performing a seismic vulnerability as-
load factors obtained with the SAVVAS method can be observed. Fig. 16 sessment where an initial calibration is not possible. The SAVVAS
shows the correlation between the vulnerability index and the load method was in this sense applied blindly and provided good results
factor corresponding to LS3 (IV LS3), as an example. However, it is from the beginning. In this method, just the factor of 1.25 defining the
noted that the SAVVAS method allows a more detailed seismic vul- damage grade associated to the collapse of the building was calibrated,
nerability assessment. The estimation of the numerical load factors but its definition does not have such an influence on the results, since it
based on numerical values adopted for the definition of some para- only affects one level of the damage scale. In fact, the definition of the
meters enables to have a greater variation on the load factors when collapse is acknowledged as the main weakness of the SAVVAS method.
compared with the vulnerability index. Fig. 16 shows clearly that, for Only the limit states LS1-LS3 are defined according to an extensive
some buildings with the same vulnerability index, the load factor de- numerical parametric study [6]. The last damage grade has been here
fining LS3 estimated with the SAVVAS method varies greatly. For in- defined using this empirically devised factor of 1.25 that has been va-
stance there is a building with IV = 39 and LS 3 = 0.21g and buildings lidated using this case study. Further research on the definition of the
with IV = 39 and LS 3 = 0.41g . In this particular example, for the same collapse for the SAVVAS method is recommended.
vulnerability index, the predicted maximum capacity of the building Another main difference among the SAVVAS and the SVIVA method
almost doubles. This example highlights the capability of the SAVVAS concerns the seismic input. While the SVIVA method requires the de-
method to provide more detailed results. finition of an earthquake scenario in terms of general macroseismic
The more detailed seismic vulnerability assessment obtained from intensities, the SAVVAS method is carried out using values of PGA to
the SAVVAS method results in a commonly higher accuracy in the define the seismic event. In the case study presented, the PGA scenario
prediction of damage, as previously reported, showing also a significant used is based on an already defined MMI scenario. However, this does
reduction of the deviation with respect to the damage observed not necessarily have to be always the case. A more detailed scenario can

Fig. 14. (a) Predicted versus observed damage grades; and (b) residuals versus observed damage grades.

13
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Fig. 15. Observed damage versus mean damage grade estimated using the SAVVAS method for the construction of the vulnerability curves as a function of LS3,
grouped by the different PGA.

deficiencies and strengths on the building stock under evaluation.

Results are therefore particularly valuable in comparative terms, as
they offer an expeditious and reliable evaluation on the buildings that
are more vulnerable within a set, which is very useful to define and
address structural retrofitting strategies at a regional or urban level.
Regarding this latter aspect, one advantage of the SAVVAS method
is that this method allows evaluating the seismic load factors of the
building in the four main directions of the building. Therefore, when
carrying out the seismic vulnerability assessment, results did not only
show a good correlation in terms of global damage, but were also able
in many cases to identify the failure mode suffered by the building, as
the most vulnerable direction identified matched the collapse observed
at the earthquake. The evaluation of one building from the dataset is
shown below as an example. Fig. 18 shows the building plan and ex-
terior and interior views depicting the damage suffered. Since the plans
of the building were available, the quantitative parameters could be
Fig. 16. Correlation between vulnerability index (IV) and LS3(g). properly identified, leaving some uncertainty only for the classification
of the qualitative parameters, namely the quality of the wall-to-wall
be defined based, for example, on the seismic microzonation of the area connection (P4), type of material (P3), level of roof thrust (P6) and
under study, which takes into account local effects. Therefore, the previous structural damage (P9). This case had sufficiently detailed
seismic vulnerability assessment can be carried out on the basis of more data to fulfill the parameter survey easily, see Table 6. Besides the good
detailed seismic hazard scenarios. Moreover, using the site response correlation between the damage predicted and observed, the method is
spectra and estimating the building natural frequency, the assessment able to detect that the most vulnerable direction is +Y direction, which
can be carried out using specific accelerations adapted to each building involves the gable wall that actually collapsed in reality (Table 6). The
and site. Further research is also recommended continuing with this building also suffered damage at the connection between the walls at
research line. the interior. The method also predicts that the building is prone to
In any case, besides these aspects related to the damage prediction suffer structural damage for low values of acceleration (LS1 = 0.14g ),
potential of both methods, the biggest advantage of this type of seismic matching the damage observed.
vulnerability assessment methods lies in their ability to detect possible

14
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

was possible to use that data to calibrate and test the two seismic
vulnerability assessment methods, which by itself is a valuable exercise
and a major contribution to this field of research.
The availability of post-earthquake damage data has contributed to
the main outcome of the paper, which was the calibration and the va-
lidation of two new methods as large scale seismic vulnerability as-
sessments for vernacular architecture. The calibration process was
particularly important for the SVIVA method because it led to the ad-
justment of the analytical expression that correlates the vulnerability
index with the mean damage grade. In the case of SAVVAS method, a
correlation between the seismic input in terms of PGA and the EMS-98
damage grades was established a priori and then validated using the
available damage data. The application of both methods led to very
good results in terms of predicted versus observed damage grades,
confirming the validity of both methods as first level approaches using
few input data, mostly qualitative.
The second main contribution is the first application of the SAVVAS
method, which has been recently developed, on a case study. The paper
thus has focused on presenting the advantages of this method with
respect to other existing methods. Among these advantages, the
SAVVAS method shows an enhanced prediction capability. First of all,
one of the main advantages of the SAVVAS method is the fact that the
correlation between damage and seismic input could be applied di-
Fig. 17. Comparison between predicted and observed damage grades obtained rectly, while the SVIVA method needed to be calibrated based on the
with the SVIVA vulnerability index method and the SAVVAS method evaluated. observed results to obtain a good accuracy. Secondly, results were very
accurate and showed very low deviations between estimated and ob-
5. Conclusions served damage. Since the data used for the application is slightly more
specific, it allows a significantly more detailed assessment. The SAVVAS
The present paper deals with the calibration and application of two method is able to detect more precisely the differences in the seismic
novel seismic vulnerability assessment methods: (a) Seismic performance of buildings belonging to the same construction typology
Vulnerability Index for Vernacular Architecture (SVIVA) method; and that were classified with the same vulnerability index according to the
(b) Seismic Assessment of the Vulnerability of Vernacular Architecture SVIVA method. Finally, the method calculates the vulnerability of the
Structures (SAVVAS) method. The calibration of the methods was car- building in different directions, which represents a great advantage in
ried out based on post-earthquake damage data on a set of 88 buildings accurately assessing the most vulnerable direction and thus detecting
located in the island of Faial, in Azores, taken from damage drawn up in the possible deficiencies of the building under evaluation. In several
the sequence of the 1998 earthquake. Since both the main structural cases, as in the one reported in Section 4.3, the method was indeed able
features of the buildings and the damage suffered by them are known, it to identify the failure mode suffered by the building.

Fig. 18. (a) Building plan and directions nomenclature; (b) main façade of the building; (c) collapsed gable wall; and (d) visible damage at the wall-to-wall
connections from the interior of the building.

Table 6
Parameter survey and results obtained per main resisting direction (PGA = 0.18 g).
Variables Damage

Method Dir. P1 P2 P3 P4 P5 P6 P7a P7b P8 P9 P10 LS1 LS2 LS3 Predicted Observed
SAVVAS +X 4.79 12.99 4 4 3 1 0.03 0.02 1 1 0.29 0.14 0.19 0.22 4.02 3.75
-X 4,79 12.99 4 4 3 1 0.30 0.02 1 1 0.29 0.14 0.21 0.24
+Y 4.79 3.96 4 4 4 1 0.04 0.15 2 1 0.29 0.13 0.16 0.18
-Y 4.79 3.96 4 4 4 1 0.00 0.15 1 1 0.29 0.24 0.24 0.26
SVIVA A D D D D A B C A D IV = 55 3.90

15
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

In summary, the paper validates the applicability of both methods on Greek earthquake damage data, Earthq. Eng. Eng. Vib. 10 (1) (2011) 129–141.
as large scale seismic vulnerability assessment methods. Both of them [17] R.J.S. Spence, A.W. Coburn, A. Pomonis, S. Sakai, Correlation of ground motion
with building damage: the definition of a new damage-based seismic intensity scale,
proved to be able to identify the buildings that are more vulnerable Proc. Of 10th Conference on Earthquake Engineering, 1992 (Madrid, Spain).
within the whole evaluated set. This is a key issue because this type of [18] M. Rota, A. Penna, C. Strobbia, Typological fragility curves from Italian earthquake
methods takes into account possible uncertainties related to the input damage data, Proc. Of First European Conference on Earthquake Engineering and
Seismology, 2006 (Geneva, Switzerland).
information collected at the expeditious inspection phase. Therefore, [19] F. Sabetta, A. Goretti, A. Lucantoni, Empirical fragility curves from damage surveys
detecting the most vulnerable elements at risk is essential in order to and estimated strong ground motion, Proc. Of 11th European Conference on
proceed with a more detailed assessment. It should be highlighted that Earthquake Engineering, 1998 (Paris, France).
[20] M. Colombi, B. Borzi, H. Crowley, M. Onida, F. Meroni, R. Pinho, Deriving vul-
the amount of information required to perform the seismic vulnerability nerability curves using Italian earthquake damage data, Bull. Earthq. Eng. 6 (3)
assessment using both methods is the same. However, the capability of (2008) 485–504.
the SAVVAS method to evaluate in more detail the seismic behavior of [21] H. Azizi-bondarabadi, N. Mendes, P.B. Lourenço, N.H. Sadeghi, Empirical seismic
vulnerability analysis for masonry buildings based on school buildings survey in
the buildings makes it particularly adequate for defining and optimizing
Iran, Bull. Earthq. Eng. 14 (11) (2016) 3195–3229.
possible structural retrofitting strategies at an urban or regional level. [22] L. Hofer, P. Zampieri, M.A. Zanini, F. Faleschini, C. Pellegrino, Seismic damage
The SAVVAS method do not only highlights the buildings where the survey and empirical fragility curves for churches after the August 24, 2016 Central
biggest efforts should be concentrated on, but also is able to identify Italy earthquake, Soil Dyn. Earthq. Eng. 111 (2018) 98–109.
[23] G.M. Calvi, A displacement-based approach for vulnerability evaluation of classes of
weaknesses in the buildings and possible failure mechanisms, which buildings, J. Earthq. Eng. 3 (3) (1999) 411–438.
makes it very useful for managing seismic risk on a city or region. [24] D.F. D'Ayala, E. Speranza, Definition of collapse mechanisms and seismic vulner-
ability of historic masonry buildings, Earthq. Spectra 19 (3) (2003) 479–509.
[25] P. Zampieri, M.A. Zanini, F. Faleschini, Derivation of analytical seismic fragility
Acknowledgments functions for common masonry bridge types: methodology and application to real
cases, Eng. Fail. Anal. 68 (2016) 275–291.
The work presented in this paper was partly financed by FEDER [26] ATC-40, Seismic Evaluation and Retrofit of Concrete Buildings, Applied Technology
Council (ATC), Redwood City, California, USA, 1996.
funds through the Competitivity and Internationalization Operational [27] P. Fajfar, Capacity spectrum method based on inelastic demand spectra, Earthq.
Programme – COMPETE and by national funds through FCT – Eng. Struct. Dyn. 28 (1999) 979–993.
Foundation for Science and Technology within the scope of the project [28] S. Lagomarsino, A. Penna, A. Galasco, S. Cattari, TREMURI program: an equivalent
frame model for the nonlinear seismic analysis of masonry buildings, Eng. Struct. 56
POC1-01-01-0145-FEDER-007633\. (2013) 1787–1799.
[29] M.A. Erberick, Generation of fragility curves for Turkish masonry buildings con-
Appendix A. Supplementary data sidering inplane failure modes, Earthq. Eng. Struct. Dyn. 37 (2008) 387–405.
[30] L. Pasticier, C. Amadio, M. Fragiacomo, Non-linear seismic analysis and vulner-
ability evaluation of a masonry building by means of the SAP2000 V. 10 code,
Supplementary data to this article can be found online at https:// Earthq. Eng. Struct. Dyn. 37 (2008) 467–485.
doi.org/10.1016/j.ijdrr.2019.101242. [31] M. Rota, A. Penna, G. Magenes, A methodology for deriving analytical fragility
curves for masonry buildings based on stochastic nonlinear analyses, Eng. Struct. 32
(2010) 1312–1323.
References [32] M.A. Zanini, L. Hofer, F. Faleschini, C. Pellegrino, The influence of record selection
in assessing uncertainty of failure rates, Ingegneria Sismica 34 (4) (2017) 30–40.
[1] M.R. Degg, J. Homan, Earthquake vulnerability in the Middle East, Geography 90 [33] K. Jaiswal, W. Aspinall, D. Perkins, D. Wald, K.A. Porter, Use of expert judgement to
(1) (2005) 54–66. estimate seismic vulnerability of selected building types, Proc. Of 15th World
[2] ICOMOS, Charter on the Built Vernacular Heritage, International Council of Conference on Earthquake Engineering, 2012 (Lisbon, Portugal).
Monuments and Sites (ICOMOS), ICOMOS 12th General Assembly, Mexico, 1999. [34] ATC-13, Earthquake Damage Evaluation Data for California, Applied Technology
[3] J. May, Handmade Houses & Other Buildings: the World of Vernacular Architecture, Council(ATC), Redwood City, California, USA, 1985.
Thames & Hudson, London, UK, 2010. [35] HAZUS, HAZUS Earthquake Loss Estimation Methodology: Technical Manual vol. 1,
[4] J. Ortega, G. Vasconcelos, H. Rodrigues, M. Correia, P.B. Lourenço, Traditional Federal Emergency Management Agency (FEMA), Washington D.C., USA, 1999.
earthquake resistant techniques for vernacular architecture and local seismic cul- [36] D. Benedetti, V. Petrini, Sulla vulnerabilità di edifici in muratura: proposta di un
tures: a literature review, J. Cult. Herit. 27 (2017) 181–196. metodo di valutazione, L’industria delle Costruzioni 149 (1) (1984) 66–74.
[5] M. Correia, Experiences from past for today's challenges (Chapter 6) – Traditional [37] S. Lagomarsino, S. Giovinazzi, Macroseismic and mechanical models for the vul-
and generational change, La fábrica, Fundación Contemporánea, The Road to nerability assessment of current buildings, Bull. Earthq. Eng. 4 (4) (2006) 415–433.
Sustainable Development, 2017 (Madrid, Spain). [38] D. D'Ayala, A. Meslem, D. Vamvatsikos, K. Porter, T. Rosetto, V. Silva, Guidelines
[6] J. Ortega, Reduction of the Seismic Vulnerability of Vernacular Architecture with for Analytical Vulnerability Assessment of Low/mid-Rise Buildings - Methodology,
Traditional Strengthening Solutions, Ph.D. thesis, University of Minho, Guimarães, Vulnerability Global Component Project, (2014).
Portugal, 2018. [39] K. Pitilakis, H. Crowley, A.M. Kaynia (Eds.), SYNER-G: typology definition and
[7] F. Neves, A. Costa, R. Vicente, C.S. Oliveira, H. Varum, Seismic vulnerability as- fragility functions for physical elements at seismic risk. Buildings, lifelines, trans-
sessment and characterization of the buildings on Faial Island, Azores, Bull. Earthq. portation networks and critical facilities. Series: Geotechnical, Geological and
Eng. 10 (1) (2012) 27–44. Earthquake Engineering (GGEE), vol. 27, Springer, 2014.
[8] T.M. Ferreira, R. Maio, R. Vicente, Analysis of the impact of large scale seismic [40] V. Silva, S. Akkar, J. Baker, P. Bazzurro, J.M. Castro, H. Crowley, M. Dolsek,
retrofitting strategies through the application of a vulnerability-based approach on C. Galasso, S. Lagomarsino, R. Monteiro, D. Perrone, K. Pitilakis, D. Vamvatsikos,
traditional masonry buildings, Earthq. Eng. Eng. Vib. 16 (2017) 329–348. Current challenges and future trends in analytical fragility and vulnerability mod-
[9] G.M. Calvi, R. Pinho, G. Magenes, J.J. Bommer, L.F. Restrepo-Vélez, H. Crowley, elling, Earthq. Spectra (2019).
Development of seismic vulnerability assessment methodologies over the past 30 [41] R. Vicente, Estratégias e metodologias para intervenções de reabilitação urbana,
years, ISET J. Earthq. Technol. 34 (472) (2006) 75–104. Ph.D. thesis, Universidade do Aveiro, Avaliação da vulnerabilidade e do risco
[10] T. Rossetto, I. Iaoannou, D.N. Grant, Existing Empirical Fragility and Vulnerability sísmico do edificado da Baixa de Coimbra (2008) (Aveiro, Portugal).
Functions: Compendium and Guide for Selection, (2015) GEM Technical Report [42] S. Giovinazzi, S. Lagomarsino, A macroseismic model for the vulnerability assess-
2015-1. ment of buildings, Proc. Of 13th World Conference on Earthquake Engineering,
[11] R.V. Whitman, J.W. Reed, S.T. Hong, Earthquake damage probability matrices, Canada, Vancouver BC, 2004.
Proc. Of the 5th World Conference on Earthquake Engineering, 1974 (Rome, Italy). [43] A. Barbat, M. EERI, F. Yépez Moya, J.A. Canas, Damage scenarios simulation for
[12] F. Braga, M. Dolce, O. Liberatore, A statistical study on damaged buildings review of seismic risk assessment in urban zones, Earthq. Spectra 12 (3) (1996) 371–394.
the MSK-76 scale, Proc. Of the Conference of the European Association of [44] M. Boukri, M. Bensaibi, Vulnerability index of algiers masonry buildings, Proc. Of
Earthquake Engineering, 1982 (Athens, Greece). 14th World Conference on Earthquake Engineering, 2008 (Beijing, China).
[13] G. Di Pasquale, G. Orsini, R.W. Romeo, New developments in seismic risk assess- [45] T.M. Ferreira, R. Vicente, H. Varum, Seismic vulnerability assessment of masonry
ment in Italy, Bull. Earthq. Eng. 3 (2005) 101–128. facade walls: development, application and validation of a new scoring method,
[14] M. Dolce, A. Masi, M. Marino, M. Vona, Earthquake damage scenarios of the Struct. Eng. Mech. 50 (4) (2014) 541–561.
building stock of potenza (southern Italy) including site effects, Bull. Earthq. Eng. 1 [46] V. Sepe, E. Speranza, A. Viskovic, A method for large-scale vulnerability assessment
(1) (2003) 115–140. of historic towers, J. Struct. Control Health Monit. 15 (2008) 389–415.
[15] G. Grünthal, European macroseismic scale 1998 (EMS-98), european seismological [47] M. Shakya, Seismic Vulnerability Assessment of Slender Masonry Structures, Ph.D.
commission, subcommission on engineering seismology, Working Group thesis, Universidade de Aveiro (2014) (Aveiro, Portugal).
Macroseismic Scales, Cahiers du Centre Européen de Géodynamique et de [48] R. Vicente, S. Parodi, S. Lagomarsino, H. Varum, J.A.R. Mendes da Silva, Seismic
Séismologie, 15 (1998). vulnerability and risk assessment: a case study of the historic city centre of Coimbra,
[16] A.K. Eleftheriadou, Karabinis, Development of damage probability matrices based Portugal, Bull. Earthq. Eng. 9 (4) (2011) 1067–1096.
[49] M. Blondet, G.M. Villa García, S. Brzev, A. Rubiños, Earthquake-resistant

16
J. Ortega, et al. International Journal of Disaster Risk Reduction 39 (2019) 101242

Construction of Adobe Buildings: a Tutorial, Earthquake Engineering Research with seismic intensity and other physical parameters, Bull. Seismol. Soc. Am. 67 (3)
Institute (EERI), Oakland, California, USA, 2011. (1977) 877–915.
[50] J. Bothara, S. Brzev, A Tutorial: Improving the Seismic Performance of Stone [63] E. Guagenti, V. Petrini, Il caso delle vecchie costruzioni: verso una legge danni-
Masonry Buildings, Earthquake Engineering Research Institute (EERI), Oakland, intensità, Proc. Of 4th Italian National Conference on Earthquake Engineering,
California, USA, 2012. 1989, pp. 145–153 (Milan, Italy).
[51] D. Gautam, J. Prajapati, K.V. Paterno, K.K. Bhetwal, P. Neupane, Disaster resilient [64] C. Margottini, D. Molin, L. Serva, Intensity versus ground motion: a new approach
vernacular housing technology in Nepal, Geoenviron. Dis. 3 (1) (2016). using Italian data, Eng. Geol. 33 (1992) 45–58.
[52] L. Sorrentino, L. Liberatore, D. Liberatore, R. Masiani, The behaviour of vernacular [65] N.P. Theodulis, B.C. Papazachos, Dependence of strong ground motion on magni-
buildings in the 2012 Emilia earthquakes, Bull. Earthq. Eng. 12 (5) (2013) tude distance, site geology and macroseismic intensity for shallow earthquake in
2367–2382. Greece: I, peak horizontal acceleration, velocity and displacement, Soil Dyn. Earthq.
[53] L. Matias, N.A. Dias, I. Morais, D. Vales, F. Carrilho, J. Madeira, J.L. Gaspar, Eng. 11 (1992) 387–402.
L. Senos, A.B. Silveira, The 9th of july 1998 faial island (Azores, north atlantic) [66] L. Decanini, C. Gavarini, F. Mollaioli, Proposta di definizione delle relazioni tra
seismic sequence, J. Seismol. 11 (3) (2007) 275–298. intensità macrosismica e parametri del moto del suolo, Proc. of 7th Convegno
[54] G. Zonno, C.S. Oliveira, M.A. Ferreira, G. Musacchio, F. Meroni, F. Mota-de-Sá, Nazionale di Igegneria Sismica in Italia, 1 1995, pp. 63–72.
F. Neves, Assessing seismic damage through stochastic simulation of ground [67] D.J. Wald, V. Quitoriano, T.H. Heaton, H. Kanamori, Relationships between peak
shaking: the case of 1998 Faial earthquake (Azores Islands), Surv. Geophys. 31 (3) ground acceleration, peak ground velocity and modified Mercalli intensity in
(2010) 361–381. California, Earthq. Spectra 15 (1999) 557–564.
[55] T.M. Ferreira, R. Maio, R. Vicente, Seismic vulnerability assessment of the old city [68] S. Marin, J.P. Avouac, M. Nicolas, A. Schlupp, A probabilistic approach to seismic
centre of Horta, Azores: calibration and application of a seismic vulnerability index hazard in metropolitan France, Bull. Seismol. Soc. Am. 94 (6) (2004) 2137–2163.
method, Bull. Earthq. Eng. 15 (7) (2017) 2879–2899. [69] E. Faccioli, C. Cauzzi, Macroseismic intensities for seismic scenarios, estimated from
[56] A. Costa, Determination of mechanical properties of traditional masonry walls in instrumentally based correlations, Proc. of First European conference on earthquake
dwellings of Faial Island, Azores, Earthq. Eng. Struct. Dyn. 31 (7) (2002) engineering and seismology, 2006 (Geneva, Switzerland).
1361–1382. [70] A.A. Gómez Capera, D. Alberello, P. Gasperini, Aggiornamento Relazioni fra
[57] A.A. Costa, A. Arêde, A. Costa, C.S. Oliveira, In situ cyclic tests on existing stone l'Intensità Macrosismica e PGA, Technical Report, Convenzione INGV-DPC 2004-
masonry walls and strengthening solutions, Earthq. Eng. Struct. Dyn. 40 (4) (2011) 2006, (2007).
449–471. [71] G.A. Tselentis, L. Danciu, Empirical relationships between modified Mercalli in-
[58] A.A. Costa, A. Arêde, A. Campos Costa, A. Penna, A. Costa, Out-of-plane behaviour tensity and engineering ground-motion parameters in Greece, Bull. Seismol. Soc.
of a full scale stone masonry façade. Part 1: specimen and ground motion selection, Am. 98 (4) (2008) 1863–1875.
Earthq. Eng. Struct. Dyn. 42 (2013) 2081–2095. [72] M. Bilal, A. Askan, Relationships between felt intensity and recorded ground-mo-
[59] A. Costa, A. Arêde, Strengthening of structures damaged by the Azores earthquake tion parameters for Turkey, Bull. Seismol. Soc. Am. 104 (1) (2014) 484–496.
of 1998, Constr. Build. Mater. 20 (4) (2006) 252–268. [73] A.A. Gómez Capera, M. Locati, E. Fiorini, P. Bazurro, L. Luzi, M. Massa, R. Puglia,
[60] R. Musson, G. Grünthal, M. Strucchi, The comparison of macroseismic intensity M. Santulin, D3.1. Macroseismic and ground motion: site specific conversion rules,
scales, J. Seismol. 14 (2) (2010) 413–428. (2015) DPC-INGV-S2 Project “Constraining observations into Seismic Hazard”,
[61] D.G. Hyams, CurveExpert Professional Documentation, Hyams Development, 2017 deliverable D3.1.
Release 2.6.4. [74] M.A. Zanini, L. Hofer, F. Faleschini, Reversible ground motion-to-intensity con-
[62] J.R. Murphy, L.J. O'Brien, The correlation of peak ground acceleration amplitude version equations based on the EMS-98 scale, Eng. Struct. 180 (2019) 310–320.

17