Engineering Structures 197 (2019) 109381

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

A vulnerability index formulation for the seismic vulnerability assessment of T

vernacular architecture
Javier Ortegaa, , Graça Vasconcelosa, Hugo Rodriguesb, Mariana Correiac
⁎

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

ARTICLE INFO ABSTRACT

Keywords: The valorization and preservation of vernacular architecture, as well as traditional construction techniques and
Seismic vulnerability assessment materials, is a key-element for cultural identity. Conservation efforts are often mainly focused on historical
Vulnerability index constructions and monuments. Furthermore, more detailed and sophisticated seismic vulnerability assessment
Vernacular architecture approaches typically used for monumental buildings require time, cost and resources that are not commonly
Numerical analysis
assigned to the study of vernacular architecture. Earthquakes come unexpectedly, endangering in-use vernacular
Pushover parametric study
Multiple regression analysis
architecture and the population who inhabits it. That is why simplified methods for the seismic vulnerability
Expert opinion assessment of vernacular architecture are of paramount importance. The present paper presents a new for-
mulation for the vulnerability index method particularly adapted to the characteristics of vernacular archi-
tecture: Seismic Vulnerability Index for Vernacular Architecture (SVIVA). The vulnerability index method has
been used extensively in the literature using different formulations that were always defined based on empirical
knowledge acquired through post-earthquake damage observation and expert judgment. The SVIVA formulation
is developed by means of an analytical process instead of the traditional empirical approach. The process in-
cluded an extensive numerical modeling campaign that allows adapting the method to the characteristics of
vernacular architecture by gaining a deeper quantitative knowledge on their seismic behavior.

1. Introduction Vernacular architecture located in seismic prone areas can be par-

ticularly vulnerable to earthquakes due to a scarcity of resources in
Vernacular architecture is the result of a tight relation between generally poor communities, resulting in inadequate overall structural
humans and the environment. When local communities have to build layout of the construction, the use of poor materials, the lack of proper
their dwellings, they respond to their surrounding environment and constructive details and a poor maintenance. Earthquakes come un-
climate, through empirical knowledge acquired along generations. expectedly, endangering in-use vernacular architecture and the popu-
Vernacular architecture reflects the tradition and life style of a com- lation who inhabits it. There is a critical gap in knowledge regarding
munity, and the inhabitants’ bonding with the natural environment. As vernacular architecture earthquake preparedness, since research in
a result, vernacular architecture is extremely heterogeneous, re- vernacular architecture has predominantly been focused on building
sponding to local conditions. However, the use of technological and typologies and spatial organization. The study of the seismic behavior
standardized modern materials has homogenized the way of building and vulnerability of representative vernacular construction systems has
throughout the world, providing an architecture that can be observed in traditionally been ignored, and conservation efforts have been mainly
any geography, jeopardizing the local building culture and vernacular placed on historical architecture and monuments. Despite this, there
architecture. For this reason, the valorization and preservation of the has been a growing interest on the seismic behavior of vernacular ar-
vernacular heritage, as well as the traditional construction techniques chitecture in the recent years, involving: (a) the experimental char-
and materials is crucial, not only as a key element of cultural identity acterization of different vernacular construction systems [3–9]; (b) in-
and a witness of the past, but also as a privileged factor for local de- situ post-earthquake damage observations [10–13]; or (c) the seismic
velopment, boosting local economies [1,2]. assessment of different vernacular typologies using numerical analysis

Corresponding author.
⁎

E-mail addresses: [email protected] (J. Ortega), [email protected] (G. Vasconcelos), [email protected] (H. Rodrigues),
[email protected] (M. Correia).

https://doi.org/10.1016/j.engstruct.2019.109381
Received 20 December 2018; Received in revised form 3 July 2019; Accepted 4 July 2019
0141-0296/ © 2019 Elsevier Ltd. All rights reserved.
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 1. Examples of targeted vernacular architecture of the SVIVA formulation.

and analytical procedures [14–17], or empirical correlations between main objective of seismic vulnerability assessments is to measure the
damage and ground motion developed after the occurrence of different probability of a specific building to reach a given level of damage when
earthquakes for specific vernacular construction typologies [18–20]. subjected to a specific seismic action. Seismic vulnerability assessment
The seismic assessment of the built vernacular heritage requires a deep methods for the built environment, together with hazard analysis
knowledge and investigation of the place, traditional techniques and (evaluation of the probability of exceedance of a certain level of seismic
materials. However, the time, cost and resources required to obtain a intensity) and exposure data (inventory of the elements at risk), are the
sufficient in-depth level of information of the analyzed structure are not main components of earthquake loss models [22]. Earthquake loss
commonly assigned to the study of vernacular architecture. That is why models are meant to predict the consequences of an earthquake quan-
the development of an expedited method for the seismic vulnerability titatively, in terms of economic impact, repair cost and human ca-
assessment of vernacular architecture is of paramount importance, sualties. They are an essential tool for seismic risk mitigation because
since more detailed and sophisticated approaches are typically re- they can save lives and contribute for the preservation of the built
stricted for individual monumental buildings. heritage, through the evaluation of different mitigation policies and the
The present paper shows the development of a new formulation for preparation of immediate emergency response and disaster recovering
the vulnerability index method firstly proposed by Benedetti and plans.
Petrini [21] particularly adapted to the characteristics of vernacular
architecture and referred as: Seismic Vulnerability Index for Vernacular
Architecture (SVIVA). In particular, the targeted vernacular archi- 2.1. Brief overview of existing methods
tecture of the SVIVA formulation was initially the Portuguese verna-
cular heritage, including stone masonry, fired clay brick masonry, There exists a wide range of seismic vulnerability assessment
adobe masonry and rammed earth constructions (Fig. 1). However, in methods available in the literature, suitable for different types of ana-
spite of the peculiarities of Portuguese vernacular architecture, it shares lysis with different goals. The selection of a specific method depends on
many characteristics with other vernacular constructions throughout the scale of the analysis and on the level of detail required for the
the world, especially in the south Mediterranean region. This is parti- targeted buildings in the area of study [23]. Because of the typical lack
cularly evident at a structural level, since the structural system is in of resources (mainly economic) that can be assigned to the study of a
most cases conceptually the same and consists of load bearing walls as traditionally underestimated and precarious vernacular heritage, the
the main vertical resisting elements. Therefore, the SVIVA formulation method for the seismic vulnerability assessment of vernacular archi-
is expected to be also applicable to other similar structures outside the tecture should make use of mostly qualitative data that can be rapidly
Portuguese context. obtained from simple visual inspections of the buildings. These methods
Vulnerability index methods require the definition of: (a) a number are commonly referred as first level approaches and are particularly
of parameters representing features of vernacular buildings that influ- well suited for large scale analyses, e.g. urban or national scale, which
ence their seismic behavior; (b) seismic vulnerability classes for each comprise large numbers of buildings. First level methods are typically
parameter; and (c) weights for each parameter. The definition of these based on empirical post-earthquake surveys and expert judgment.
three aspects is addressed and presented sequentially within the present Empirical approaches extract correlations between damage and seismic
paper. Finally, the proposed SVIVA formulation for the vulnerability motion after observing the damage suffered by different building
index method is presented at the end of the paper as the main outcome. typologies due to a particular earthquake [24]. Thus, they require a
large set of post-earthquake damage data which is not always available.
Methods based on expert judgment emerged as a result of this data
2. Proposed seismic vulnerability assessment method for limitation [25]. On the basis of expert opinion and previous knowledge,
vernacular architecture they estimate the damage that a certain structure can suffer for a given
seismic intensity by analyzing the structural characteristics of the
The seismic vulnerability of a structure can be defined as its intrinsic constructions and classifying them into different building typologies
proneness to suffer damage as a result of a seismic event. Therefore, the [26,27].

2
J. Ortega, et al. Engineering Structures 197 (2019) 109381

The method proposed is an adaptation of the hybrid approach fol- present paper focuses solely on the definition of the updated vulner-
lowed by Vicente [23], which combines the vulnerability index method ability index formulation.
[21] and the macroseismic method [28]. Both methods are supported The paper firstly focuses on the selection of the key parameters for
by statistical studies of post-earthquake damage information, but also the SVIVA formulation, based on literature review. Secondly, a thor-
rely on expert opinion. The vulnerability index method is based on the ough parametric numerical simulation based on detailed finite element
identification of constructive aspects that are more influential in the (FE) modeling and pushover analysis was designed to support the
seismic structural behavior of the building, which results in the defi- analysis of the influence of the selected key parameters on the seismic
nition of several qualitative and quantitative parameters. The original behavior of vernacular buildings. The strategy consists of modifying a
formulation proposed by Benedetti and Petrini [21] accounted for a reference model according to the different parameters considered and
total of eleven parameters including, among others, the type of vertical analyzing the variations on the seismic performance of the structure.
structural system, the type of horizontal diaphragms, the plan config- This procedure led to the definition of the seismic vulnerability classes.
uration and the conservation state. Parameters are related to four Existing formulations for the vulnerability index method have defined
classes corresponding to increasing vulnerability, from A (lowest) to D vulnerability classes and weights on the basis of empirical knowledge
(highest), associated with a qualification coefficient (Cvi). A weight and expert judgment [21,30,32,34]. Few studies have combined ana-
factor (pi) is included to emphasize the relative importance of each lytical approaches, such as numerical analysis, to add robustness to
parameter. Each parameter is qualified individually, and the overall mostly empirical methods [33]. However, analytical approaches and
vulnerability of the building is calculated as the weighted sum of the complex numerical modeling are suitable to overcome the lack of post-
parameters, expressed through the vulnerability index (Iv). The vul- earthquake damage data. They are particularly appropriate to carry out
nerability index is typically normalized to fall within a range between 0 parametric studies that allow taking into account the effect of con-
(very low vulnerability) and 100 (very high) and can be understood as a structive and material characteristics that cannot be typically con-
measure of the building safety under seismic loads [29]. sidered in empirical methods. The use of an analytical procedure to
The combination of the two methods proposed by Vicente [23] define the SVIVA formulation is the main novelty of the present work
enables to: (a) use the vulnerability index formulation to estimate the and is intended to help existing simplified vulnerability index methods
seismic vulnerability of the building using predefined parameters, al- to gain in reliability.
lowing the individual evaluation of the buildings instead of using a Finally, the definition of the parameters weights that are necessary
general vulnerability class for a specific building typology; and (b) to complete the SVIVA formulation is carried out following two dif-
calculate the mean damage grade using the equation proposed by the ferent approaches: (1) statistical analysis; and (2) expert judgment. The
macroseismic method: first statistical approach arises from the idea of taking advantage of the
large amount of numerical data obtained from the parametric study
I + 6.25V 13.1 carried out for the definition of the seismic vulnerability classes. The
µD = 2.5 1 + tanh
Q (1) information of the numerical models, as well as the results from the
parametric analyses performed on each of them were organized and
This analytical expression correlates the expected mean damage
structured in a wide database. The weights are then defined through the
grade (μD) and the seismic input, as a function of the building vulner-
analysis of the large database previously assembled using multiple
ability, and allows the construction of vulnerability curves for the
linear regression, which led to assess the relative importance of the
subsequent seismic vulnerability evaluation and estimation of losses. I
different parameters. However, the importance of the expert opinion on
is the seismic input in terms of macroseismic intensity, V is the vul-
the evaluation of the weights should not be disregarded. The second
nerability index and Q is the ductility index, which is an empirically
approach for the definition of the weights consists of assigning the
defined index that considers the ductility of a determined construction
parameters weight based on expert judgment, which is the common
typology, typically ranging from 1 to 4 [30]. It should be noted that the
procedure used by existing methods. The expert opinion was collected
vulnerability index used by the two methods (Iv and V) are different.
by means of a questionnaire prepared and distributed among a group of
That is why Vicente [23] proposed another analytical expression re-
international experts in the field, from around the world. As a result,
lating both measures of vulnerability:
two sets of seismic vulnerability parameters weights are obtained fol-
V = 0.56 + 0.0064 × IV (2) lowing the two distinct approaches. A discussion comparing the results
obtained using each approach is thus provided. The final SVIVA for-
This approach combining the seismic vulnerability index formula- mulation is presented at the end of the paper.
tion with the macroseismic method has been recently implemented for
the seismic vulnerability assessment of Portuguese masonry structures 3. Definition of seismic vulnerability assessment parameters
in several historic city centers [10,30–32], obtaining useful and reliable
results as a first level approach. It has also been adapted for other Benedetti and Petrini [21] proposed a set of 11 parameters in their
particular structures, such as Nepalese pagoda temples [33]. original formulation. Other studies have adapted this formulation to
specific structural typologies or to the characteristics of local con-
2.2. Methodology for the definition of a new formulation for the structions in specific regions, identifying the most relevant parameters
vulnerability index method and discarding others that are not considered remarkably significant for
those building typologies [30,33,35–37]. Based on the work developed
The proposed SVIVA formulation follows the same structure from by these authors and on the earthquake performance of vernacular
the original formulation proposed by Benedetti and Petrini [21]. Thus, constructions observed in past earthquakes [10,11,13,38–40], a new set
three steps were necessary for its development: (a) to identify and de- of parameters is proposed for the targeted vernacular typologies de-
fine a number of parameters that represent distinctive characteristics of picted in Fig. 1. Ten parameters are finally selected according to the
vernacular buildings that influence their seismic behavior; (b) to obtain singular behavior of this structural type. They are listed in Table 1
seismic vulnerability classes for each parameter; and (c) to estimate the showing a brief description and the upper and lower bound of the range
weights for each parameter. Once the updated SVIVA formulation is of variation considered for each of them.
completed, the analytical expressions from Eqs. (1) and (2) can be used With respect to the ranges of variation shown in Table 1, it should
to perform a seismic vulnerability assessment. It is noted that the be noted that they were determined based on initial reference values

3
J. Ortega, et al. Engineering Structures 197 (2019) 109381

obtained from the detailed geometrical and constructive survey of ty-

pical Portuguese vernacular rammed earth constructions that can be
found in Correia [41] and Correia [42]. However, the range of variation

Negligible stiffness poorly connected to the walls ≤ P5 ≤ rigid properly

is made wider for all parameters in order to cover more vernacular

Non-thrusting roof ≤ P6 ≤ thrusting heavy roof with low inclination

No visible damage ≤ P9 ≤ severe damage with widespread cracking

typologies and for the study to be more comprehensive. A more detailed
description of the selected parameters is also provided next, which is
mainly intended to describe their structural role on the global seismic
behavior of vernacular buildings.
Adobe masonry ≤ P3 ≤ dressed stone masonry
Visible separation ≤ P4 ≤ efficient interlocking

3.1. Wall slenderness (P1)

The wall slenderness particularly affects the out-of-plane behavior

of walls and several authors have already used this parameter to assess
the seismic vulnerability of masonry walls [37,43,44]. Generally
speaking, vernacular buildings walls are rather thick, being rarely
connected to the walls

thinner than 0.45–0.5 m thick. The maximum wall height is more

Range of variation

variable, even though vernacular buildings tend to be rather compact

0.26 ≤ γi ≤ 0.79
2% ≤ IP ≤ 63%

and present small height dimensions. The range of variation considered

4 ≤ λ ≤ 22.5
4 ≤ s ≤ 12

1≤N≤4

for the wall slenderness varies between 4 and 22.5, which may exceed
typical values for vernacular constructions, but were adopted for the
study to be more comprehensive.
Ratio between the in-plane area of earthquake resistant walls in each main direction (Awi) and the total in-plane area

3.2. Maximum wall span (P2)

Ratio between the total area of wall openings in all earthquake resistant walls (i.e. walls parallel to the loading

Degree of deterioration existing in the load bearing walls of the buildings and the weakening signs (cracks and
Organization of the vertical structural system in terms of the level of connection between the orthogonal walls
Maximum length spanned by a wall prone to out-of-plane movements, i.e. wall without intermediate supports

Construction solutions and materials used to build the horizontal structural elements of vernacular buildings,

The maximum wall span is another geometrical parameter gov-

erning the out-of-plane response of the walls. Vulnerability index for-
mulations including this parameter propose a classification in terms of
focusing on the quality of their connection to the load bearing walls and their in-plane stiffness

span to thickness ratio [23]. However, since the wall thickness is al-
ready taken into account in the previous parameter, this parameter
considers simply the variation of the maximum wall span (s), measured
in meters.

3.3. Type of material (P3)

deformation) that may aggravate damage in the event of an earthquake
Ratio between the effective wall inter-story height (h) and its thickness

direction) in one direction and the total surface area of the same walls

With respect to the type of material, the constituent material of

vernacular constructions usually consists of stone, fired clay brick,
adobe and rammed earth. Masonry fabric typologies can present sig-
Possible thrust exerted by the roof to the load bearing walls
Nature of the material used to build the load bearing walls

nificant variations in terms of morphology of the masonry wall: (a)

type, shape and size of the masonry units (ashlar stone masonry, irre-
gular rubble stone masonry, roughly shaped stone masonry, fired clay
brick masonry, etc.); (b) masonry layout (irregular/regular horizontal
Number of floors of the building under study
Seismic vulnerability assessment parameters considered for the SVIVA formulation.

courses, presence of several leaves, lack of connection between the

leaves, etc.); or (c) type of mortar used, if any. These aspects determine
the quality of the masonry and, thus, the capacity of the building to
of earthquake resistant walls (Aw)

withstand horizontal forces resulting from the seismic load, as reported

by many authors after post-earthquake observations [45–47]. The me-
chanical properties of typical masonry and earthen materials used for
the construction of walls in vernacular architecture have been studied
widely and characterized experimentally on the literature. In order to
assess the influence of this parameter, the material properties of the
Description

walls are modified based on typical values for these traditional mate-
rials collected from the literature [6,48–50]. A series of models were
then defined varying the Young’s modulus (E), the compressive strength
(fc) and the tensile strength (ft), adopting values that can be associated
Symbol

to the different materials typically used in vernacular constructions.

P3
P4
P5

P6

P9
IP

γi
λ
s

3.4. Wall-to-wall connections (P4)

Wall-to-wall connections
Horizontal diaphragms
Maximum wall span

Damage observation after earthquakes has shown how the failure

Previous structural
Number of floors
Wall slenderness

Type of material

mode of the building is characterized many times by vertical cracks at

In-plane index
Wall openings

the wall intersections, leading to the out-of-plane overturning or

Roof thrust

bending failure of the walls. Thus, the quality of the wall-to-wall con-
damage

nections, namely at the building corner and at the connection between

Parameter

external and internal walls, is a key aspect regarding the seismic be-
Table 1

havior of the building. In order to evaluate the influence of this para-

P10
P1
P2
P3
P4
P5

P6
P7

P8
P9

meter in a simple way, the mechanical strength of the elements at the

4
J. Ortega, et al. Engineering Structures 197 (2019) 109381

corners was reduced to simulate weak connections that are more prone vulnerability induced by these elements. Vernacular buildings in rural
to fail, leading to the independent behavior of perpendicular walls. This areas generally present a reduced area of wall openings, but it is vari-
way of simulating weak connections at rammed earth buildings re- able and the area of wall openings can increase significantly if the
presents the difficulty of creating corners inside the frameworks and building is located in an urban environment.
poor joints with vertical recess solution [51]. For stone masonry
buildings, it represents the presence of vertical joints and, thus, the lack 3.8. Number of floors (P8)
of proper interlocking between orthogonal walls.
With respect to the number of floors, taller buildings tend to be
3.5. Horizontal diaphragms (P5) more vulnerable to earthquakes because the center of gravity is raised.
Thus, the overturning moment of the walls due to horizontal loading
Horizontal timber diaphragms have a critical role in transmitting also increases. This is particularly evident in unreinforced vernacular
the lateral earthquake loads to the vertical resisting elements of the masonry buildings whose mass is mainly concentrated at the walls in-
structure. The flexibility of traditional timber floors in unreinforced stead of at the floors. Vernacular buildings are typically not too high. In
masonry and earthen vernacular buildings leads to significant bending the rural environment, rammed earth constructions usually extend
and shear deformations under horizontal loads [52]. This excessive horizontally and are composed by a single story, but in the urban
deformability or lack of proper connections with the load bearing walls context they rarely present more than two stories. Stone masonry ver-
forces the walls to work independently, resulting in their local out-of- nacular buildings in the urban context can easily present up to four
plane failure when the building is subjected to earthquake loading. stories, particularly when arranged in aggregates, such as in most
However, when effective diaphragm-to-wall connections are ensured, European historical city centers.
and the in-plane stiffness of the diaphragm is enough to engage the
walls parallel to the seismic load, the seismic behavior of vernacular 3.9. Previous structural damage (P9)
buildings relies on the in-plane response of the walls. Thus, the seismic
response of vernacular buildings is strongly dependent on the char- Generally, a critical reason for the vernacular heritage to be so
acteristics of timber diaphragms [17]. Timber floor construction in vulnerable to earthquakes is its advanced state of deterioration, as a
vernacular architecture is usually very simple, consisting of wooden result of poor maintenance or abandonment, which results in previous
beams covered with cross boards directly nailed to the beams com- structural damage often going unrepaired. Existing cracks increase the
posing the sheathing. As a result, the behavior of the diaphragm is vulnerability of specific parts of the structure and can anticipate its
clearly different in the two orthogonal directions: perpendicular and failure. The previous structural damage was simulated by imposing an
parallel to the main beam axis. In terms of construction, there are dif- initial level of damage to the structure. The numerical models were
ferent ways of achieving a proper diaphragm-to-wall connection in both firstly loaded in one direction until reaching a certain degree of da-
directions. Three aspects were considered as the most critical in de- mage, and then the pushover analysis can be performed in the per-
fining the seismic behavior of horizontal diaphragms and were eval- pendicular direction. In this way, the initial level of damage imposed to
uated independently in the parametric study: (a) level of connection the structure is not directly related to the resisting mechanism that will
between timber beams and the perpendicular walls; (b) stiffness of the be later activated when performing the pushover analysis. In this work,
diaphragm; and (c) level of connection between the whole diaphragm the level of damage was defined following the classification proposed
and the perimeter walls. by Masciotta et al. [53] according to the maximum crack size: (1) slight
damage, corresponding to hairline and fine cracks (with crack
3.6. Roof thrust (P6) width < 1 mm); (2) moderate damage, corresponding to a crack width
between 1 and 5 mm; and (3) severe damage, corresponding to large
There are particular roofing structural systems that exert lateral cracks impairing functionality (with crack width > 5 mm). Taking this
thrust and can anticipate the out-of-plane collapse mechanism of the classification into account, a range of variation of the initial level of
load bearing walls supporting the roof. Some roof structural types do damage imposed was established in order to perform the parametric
not exert lateral thrust simply because of their geometry or because of study.
the addition of specific structural elements. Thrust-exerting roof types
are mainly composed by rafters with no intermediate support, whose 3.10. In-plane index (P10)
feet are fixed at a wall plate but are not properly connected among
them at the ridge. Thus, rafters subjected to vertical loads push the The in-plane index ratio provides an estimation of the shear strength
supporting walls outwards at their top. That is why the type of roof and in each orthogonal direction and can be an indicator of the in-plan ir-
its ability to exert or not thrust onto the supporting walls is a key aspect regularity of the building. This parameter gives a measure of the in-
regarding the seismic behavior of buildings. The level of thrust exerted plane stiffness of the structure in each main direction and, thus, it can
by thrust-exerting roof types depends on: (a) the span covered by the be considered as an indicator of the feasible seismic performance of the
roof; (b) the load of the roof; and (c) its inclination. The variation of building [44]. For values close to 0.5 the walls are well-balanced. Va-
these features inducing different levels of roof thrust was considered in lues of in-plane index deviating from 0.5 show that the building has a
the parametric study to evaluate the influence of this parameter. weaker direction. The seismic capacity of a building may be jeopardized
when it presents an unbalanced area of resisting walls in the two or-
3.7. Wall openings (P7) thogonal directions. This index can be very variable in vernacular
buildings, indicating very different plan configurations.
The presence of openings in earthquake resistant walls mainly af-
fects their in-plane behavior, compromising their in-plane resistance. 4. Definition of seismic vulnerability classes
This is particularly significant when the building is prone to suffer in-
plane damage, such as when sufficiently stiff diaphragms well-con- The methodology adopted for the definition of the seismic vulner-
nected to the walls are able to avoid premature out-of-plane collapses. ability classes for the SVIVA formulation is presented. The definition of
Damage patterns observed after earthquakes show that crack lines often the classes for the parameter related to the maximum wall span (P2) is
follow the distribution of the façade openings, revealing the provided as an example of the methodology that was consistently

5
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 2. Steps 1 and 2 from the methodology adopted for the definition of the seismic vulnerability classes using P2 (maximum wall span) as an example.

Table 2
Mechanical properties adopted for the different materials used in the reference models.
Material E (MPa) ν fc (MPa) Gfc (N/mm) ft (MPa) GfI (N/mm) W (kN/m3)

Stone masonry 1500 0.2 1.5 2.4 0.15 0.012 20

Rammed earth 300 0.3 1 1.6 0.1 0.012 20
Timber 10,000 0.2 – – – – 6
Timber diaphragms 200 0.3 – – – – 7.5

followed for the ten key parameters. The complete process followed for deformability [52]. When modeling a proper beams-to-wall connection,
the definition of the classes of the ten parameters can be found in the beams are considered fully embedded within the wall, going
Ortega [54]. through the whole thickness. When modeling a good diaphragm-to-wall
connection, the nodes at the connections between the board sheathing
4.1. Step 1: preparation of reference numerical models and the walls are assumed to share all degrees of freedom. The roof is
modelled as distributed load along the walls and, when expected to
The reference finite element (FE) models prepared were initially exert thrust to the walls, a distributed horizontal load is also applied at
based on typical Portuguese vernacular stone masonry and rammed the top of the walls. The displacements of the walls elements at the base
earth constructions, such as those shown in Fig. 1. The reference models are fully restrained.
are prepared in a generic way to easily accommodate the variations Different materials are considered for the walls according to the
required to assess the influence of the different parameters. For the discussion from Section 3.3. Timber is used for the lintels and floor
evaluation of each parameter, different reference models were prepared construction elements. Only the materials used for the walls are con-
assuming different initial conditions and combinations of the remaining sidered to present nonlinear behavior, adopting standard isotropic Total
parameters. This is meant to provide a more comprehensive under- Strain Rotating Crack Model (TSRCM). The model describes the tensile
standing of the influence of each parameter on buildings showing dif- and compressive behavior of the material with one stress-strain re-
ferent characteristics. Fig. 2 shows the two reference models prepared lationship and assumes that the crack direction rotates with the prin-
for the parametric analysis aimed at defining the seismic vulnerability cipal strain axes. The tension softening function selected is exponential
classes according to the maximum wall span (P2): (1) one-floor rammed and the compressive function selected to model the crushing behavior is
earth building with flexible diaphragm (RE1F); and (2) two-floor parabolic. It was selected because of its robustness and simplicity and
rammed earth building with a sufficiently rigid diaphragm well-con- has been already successfully applied in previous analysis of complex
nected to the walls able to activate their in-plane response (RE2Fd1). stone masonry and earthen structures [6,56–58]. The material proper-
The use of two differentiated building typologies as reference models ties adopted varied for the different models constructed and were based
allows understanding the influence of the maximum wall span when the on data collected from different authors [48,50,51,59,60]. The parti-
building is prone to show an out-of-plane failure mode and when is cular mechanical properties adopted for the models shown in Fig. 2 are
prone to present in-plane collapse mechanisms. shown in Table 2 as an example.
The software selected to perform the numerical parametric analysis
was DIANA software [55]. Walls are simulated with ten-node isopara- 4.2. Step 2: preparation of the set according to the variations defined for
metric 3D solid tetrahedron elements (CTE30), with four-point in- each parameter
tegration scheme over the volume. The walls have at least two elements
within the thickness. When modeled, the diaphragms are assumed to be On the basis of the reference model selected, the second step con-
composed by: (a) timber beams simulated using three-node beam ele- sists of preparing the rest of the models according to the variations
ments (CL18B); and (b) cross-board sheathing modeled using six-node defined for each parameter. A range of variation was determined for
triangular shell elements (CT30S), aiming at simulating the in-plane each parameter (Table 1). Thus, a set of models is prepared departing

6
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 3. Steps 3 and 4 from the methodology adopted for the definition of the seismic vulnerability classes using P2 (maximum wall span) as an example.

from each reference model according to this range, constituting the 4.5. Step 5: building of four-linear capacity curves
base of the parametric analysis. For example, in the same case of
parameter P2, the range of maximum wall span established went from 4 The pushover analysis allows describing the seismic response of the
to 12 m. The maximum wall span of the reference models was 7 m. The structure in terms of the capacity or pushover curve. The pushover
span was thus decreased and increased by 1 m until covering the whole curve is given as relation between the base shear coefficient or load
range defined. As a result, two sets of 9 models were prepared (Fig. 2). factor (i.e. the ratio between the horizontal forces at the base and the
self-weight of the structure, expressed as an acceleration in terms of g)
and the displacement at the control node (usually taken as the node
4.3. Step 3: pushover analysis
showing the highest displacements). It should be noted that this node
usually varies according to the collapse mechanism obtained, which
Once all the models are constructed, a pushover analysis is per-
differs between buildings. Thus, the curves are representative of the
formed on each of them. The distribution of lateral loads in the push-
global structural behavior of the different buildings subjected to hor-
over analysis was considered to be directly proportional to the mass, as
izontal loading, not individual structural elements composing the
generally recommended for the seismic assessment of existing masonry
buildings.
buildings [61]. Pushover consists of increasing the loads monotonically
In order to have a common quantitative basis of comparison of the
until reaching the collapse of the structure, which allows determining
seismic capacity of the buildings, four structural limit states (LS) asso-
the ability of the building to resist the characteristic horizontal loading
ciated to specific damage levels exhibited by the structure are identified
caused by the seismic action taking into account the material nonlinear
from the pushover curve, following recommendations available from
behavior. Convergence was checked at each load step using the internal
the literature [62–64]:
energy norm criteria with a tolerance of 10-3.
LS1: Represents the onset of cracking and the end of the elastic
The direction selected to perform the pushover analyses depends on
behavior. Until this limit, the behavior of the building is essentially
the expected response and failure mode of the buildings. Each set of
elastic and the structure can be considered as fully operational. The
models is tested in the same direction, which is commonly the one in
beginning of cracking is assumed to start when there is a reduction of
which the buildings are assumed to be more vulnerable. However, in
the initial stiffness of the global response of the building up to 2%. This
some cases, the models are tested in the direction in which the para-
reduction was defined after observing that the first cracks in the nu-
meter under evaluation is supposed to have a greater influence. For
merical models are visible after this reduction of the initial stiffness,
example, continuing with the example above, parameter P2 evaluates
characterizing the end of the elastic behavior. It is noted that the value
the variations in the response of the building when the maximum length
is relatively low and is related with the low tensile strength of the
of a wall prone to out-of-plane movements varies. Thus, the direction
materials considered in this study.
selected for the pushover analyses had to be perpendicular to the walls
LS2: This limit state tries to depict the transition between a point
whose span is being modified. The direction is marked in red in Fig. 3.
where the structure is still functional and retains most of its original
stiffness and strength, showing minor structural damage and cracks,
4.4. Step 4: analysis of variations in the damage patterns and failure and a state where significant damage is visible so that the building
mechanisms could not be used after without significant repair. The definition of the
point depicting LS2 in the pushover curve is made by satisfying two
This step is primarily aimed at obtaining a better understanding of energy criteria: (1) the first energy criterion assumes that the area
the seismic behavior of vernacular buildings. The pushover analyses below the three-linear curve formed by LS1, LS2 and LS3 coincides with
carried out on each model from the set allow performing a comparison the area below the pushover curve from LS1 to LS3; and (2) the second
among them in terms of variations in the crack patterns and failure criterion assumes that the LS2 point is on the slope associated to the
mechanisms. This step helps to understand how the seismic response of secant stiffness corresponding to 70% of the maximum strength.
the building changes according to the variations in the parameter under LS3: Defined by the load factor and displacement corresponding to
evaluation. The failure mode of each model is analyzed in terms of: (a) the attainment of the building maximum strength. The building shows
maximum total displacements; and (b) crack pattern; see Fig. 3.

7
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 4. Step 5 from the methodology adopted for the definition of the seismic vulnerability classes involving the transformation of the pushover curves into
equivalent four-linear capacity curves, using P2 (maximum wall span) as an example.

significant structural damage and has lost a significant amount of its LS 3(g )i
LS3(%)i =
original stiffness. It retains some lateral strength and margin against LS3(g )max (5)
collapse, but it cannot be used after the earthquake.
LS4: This ultimate limit state corresponds to the point where the Continuing with the example of P2, for both set of models, the
building resistance deteriorates below an acceptable limit, which is set building with s = 4 m showed the maximum capacity and the highest
at the 80% of the maximum strength. Repairing the building after values of load factor defining LS1, LS2 and LS3. Therefore, these three
reaching this limit state may be neither possible nor economically values of load factor are used for the normalization of the load factors
reasonable. The pushover analyses were not continued after reaching obtained for the remaining models of the set. Three curves can be
this limit state. constructed showing the variation of the load factor defining each LS as
Step 5 thus consists of transforming the pushover curves into four- a function of the wall span, see Fig. 5. The curves show in a clear
linear capacity curves according to the points associated to each LS. manner the influence of the parameters in the global seismic behavior
These four-linear capacity curves describe the seismic behavior of each of the buildings.
building through four equivalent static horizontal loads (load factors)
that the buildings can withstand before reaching each LS. Since they
also provide information about the deformation capacity of the struc- 4.7. Step 7: definition of seismic vulnerability classes
ture, they allow an easier and quantitative comparison between the
structural response of the models from each set in terms of capacity, The seismic vulnerability classes are defined according to the var-
stiffness and ductility, see Fig. 4. iation of the load factor corresponding to the attainment of the max-
imum capacity of the building (LS3). The criterion followed for the
4.6. Step 6: Analysis of load factor variations definition of the typical four classes of increasing vulnerability (A, B, C
and D) consists of dividing equally the total range of variation
The sixth step consists of comparing the values of load factor cor- (LS3(%)max LS3(%)min ) within each set into four parts. Each interval is
responding to LS1, LS2 and LS3 for the different models within each set. associated with a vulnerability class and the buildings are classified
LS4 is not included because it is directly determined from LS3. The load according to the interval they lie within. Fig. 6 illustrates this process
factor variations for each LS can be expressed in terms of percentage using as an example the definition of the classes for the seismic vul-
normalized by dividing the value of load factor obtained for each model nerability parameter P2. It is noted that the ranges of variation obtained
of the set (i) by the maximum value of load factor obtained among the for each set can differ, resulting in differences in the definition of the
buildings analyzed: seismic vulnerability classes. As an example, in the definition of the
classes for P2, within the RE1F set of models, the building with s = 5 m
LS1(g )i classifies as A, whereas within the RE2Fd1 set of models, the building
LS1(%)i =
LS1(g )max (3) with s = 5 m classifies as B. The final classification is made taking into
account these discrepancies by adopting the most unfavorable class, in
LS 2(g )i order to be on the safe side. Thus, buildings with s = 5 m are considered
LS 2(%)i =
LS 2(g )max (4) as class B.

8
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 5. Step 6 from the methodology adopted for the definition of the seismic vulnerability classes that analyzes the variation of the load factors defining each LS
according to the variations in the parameters, using P2 (maximum wall span) as an example.

Fig. 6. Final step 7 from the methodology adopted for the definition of the seismic vulnerability classes using P2 (maximum wall span) as an example.

4.8. Summary the seismic behavior of vernacular buildings. The seismic vulnerability
classes defined for each parameter are shown in Table A1 in Appendix
The extensive numerical parametric study included the preparation A. It provides: (1) quantitative ranges of values delimiting each para-
of 277 numerical models and concluded with the definition of the meter class, whenever possible; and (2) qualitative description of the
seismic vulnerability classes for the ten key parameters. The study classes intended to serve as a reference when doing the assessment in
helped also to validate that the selected parameters are influential in terms of simple visual inspections.

9
J. Ortega, et al. Engineering Structures 197 (2019) 109381

5. Definition of seismic vulnerability assessment parameters several input variables. The relationship between variables is often very
weight complex and the simplest approach consists of fitting a multilinear
equation to the data, which reads:
As previously mentioned, the vulnerability index (Iv) is calculated as
Y= + 1 x1 + 2 x2 + + k xk + (6)
the weighted sum of the seismic vulnerability assessment parameters. 0

The weights (pi) are the coefficients that multiply the vulnerability class where Y is the output variable and k is the number of input explanatory
numeric value (Cvi) and, therefore, indicate the relative importance of variables (xk). The parameters βk are called the regression coefficients
each parameter in estimating the overall building seismic vulnerability. and ε is the error. The line defined by the regression equation (Eq. (6))
Parameters weights have been commonly assigned based on expert describes how the response changes according to the explanatory
opinion. The present work proposes to specify the weights that are variables. Each regression coefficient is a slope of the line and re-
necessary to complete the SVIVA formulation following two different presents the expected change in the response per unit change in the
approaches: (1) statistical analysis; and (2) expert judgment. input variable, when the remaining variables are held constant. For this
reason, since they provide a measure of the amount of change, they can
5.1. Definition of the parameters weight based on statistical analysis be used to compare the relative strength of the various predictor vari-
ables in the prediction of the dependent variable. This information
The pushover parametric study performed to obtain the seismic precisely contributes to understand which parameters are more influ-
vulnerability classes required the construction of 277 numerical models ential on determining the seismic vulnerability of vernacular buildings,
with varying geometrical, construction, material and structural char- which allows sorting the ten parameters by importance. Weights can
acteristics. Since most models were analyzed in the two orthogonal thus be attributed to each parameter according to this arrangement,
directions, more than 400 pushover analyses were finally performed. which enables their numerical definition. Fig. 7 summarizes the steps
The results of the numerical study were organized within a database. followed for parameters weights definition.
Multiple linear regression (MR) analysis was applied to analyze the
database, which has a large number of variables and complex and un- 5.1.1. Data analysis and extension of the database
clear relationships among them. The use of MR is ultimately intended to The first step for the definition of the parameters weights (pi) for the
investigate if the parameters are able to predict the seismic response of SVIVA formulation dealt with the organization of the database. Thus,
buildings, in terms of the load factor associated to LS3, which was data was structured in a database composed of 11 attributes. After the
chosen because it was considered as the most determining when eval- definition of the seismic vulnerability classes for the ten parameters,
uating the seismic behavior of buildings. It represents the attainment of each model could be defined in terms of the class assigned to each
the maximum capacity of the building and has a very clear physical parameter. Since there are only four classes of increasing seismic vul-
meaning. nerability, the ten input variables associated to the parameters are ac-
Regression analysis is a popular statistical method used to in- cordingly expressed in a discrete form, assuming only four countable
vestigate the dependence of one output variable on one or more other numbers from 1 to 4, associated to the classes A to D, respectively. On
input variables. In the present case, the ten parameter variables related the other hand, the output variable (the load factor defining LS3) is
to the ten key seismic vulnerability assessment parameters were the expressed as a continuous variable expressed in g, whose value typically
input variables, while LS3 was the output variable. The objective was to ranges from 0 to 1.
define and quantify the relationship among them through a mathema- After a preliminary data analysis performed, the extension of the
tical model. A multiple regression model is required because there are database was not deemed enough to define a robust regression model

Fig. 7. Methodology adopted for the definition of the seismic vulnerability assessment parameters weights.

10
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Table 3
Statistical measures of the variables.
Variables Minimum Maximum Mean Median Mode STD Skewness

Input P1 1 4 1.59 1 1 0.95 1.48

P2 1 4 2.1 2 1 1.06 0.17
P3 1 4 3.36 4 4 1.04 −1.25
P4 1 4 1.44 1 1 0.94 1.90
P5 1 4 2.71 3 4 1.38 −0.27
P6 1 4 1.24 1 1 0.70 3.04
P7 1 4 1.64 1 1 1.04 1.29
P8 1 4 2.39 3 1 1.21 −0.09
P9 1 4 1.29 1 1 0.77 2.65
P10 1 4 2.68 3 2 0.97 −0.08
Output LS3 (g) 0.03 1.24 0.42 0.41 0.41 0.21 0.62

that allows a confident estimation of the parameters weights.

Considering the pattern of variability within each parameter variable
(i.e. the distribution of buildings belonging to each class), a lack of
balance (or asymmetry) was detected. Due to the use of reference
models showing similar initial conditions for the definition of the
classes, some seismic vulnerability classes in certain parameters are
more frequent than others. For example, in the case of parameter re-
lated to the wall slenderness (P1), the height and the thickness of the
walls of most of the models constructed did not vary when assessing
other parameters and most buildings from the database belong to class
A.
Therefore, additional numerical models were built in order to en-
large the database. The main criterion applied for the enlargement of
the database was to ensure that there are a minimum of 25 models
representing each parameter class, in order to contain a meaningful
statistical amount for all of them. The new models were then prepared
by randomly combining the parameters classes. In total, 30 numerical
models were prepared and 112 analyses were performed on the new set
of models prepared for the enlargement of the database. The precise
information on the process of extension of the database is detailed in
Ortega [54]. This led to a final database composed of 530 results ob-
tained from pushover analyses performed on the FE models. Despite the
limitations of the database, which certainly cannot cover all possible
cases, it is considered to be sufficiently representative. Fig. 8. Predicted versus observed LS3 values for the regression model MR0.
Table 3 provides a deeper insight into the data by showing some
statistical measures of the variables. It is noted that there is still an lines. This indicates that the error between the predicted and the ob-
asymmetry in the distribution of the data in most parameters, despite served value obtained is lower than 0.1 g for most cases. The behavior
the additional models constructed. Nevertheless, an asymmetry was of the model is deemed very satisfactory, considering that the model
expected since some classes are more typical than others in vernacular predicts quite accurately the maximum capacity of the building using as
constructions. Thus, continuing with the previously mentioned para- input merely the parameters classes.
meter P1 shows a clearly skewed distribution towards lower values of The performance of the model can be measured in terms of errors
wall slenderness. The vast majority of the values lie within class A and and coefficient of determination (R2). The errors relate to the difference
B, which implies values of slenderness λ < 9. This is reasonable given between the predicted LS3 and the numerical (observed) LS3. The
that vernacular buildings rarely present values of wall slenderness maximum error observed (εmax) is 0.405 g, which is large, but the Mean
greater than 9. Absolute Error (MAE) and the Root-Mean-Squared Error (RMSE) are
0.075 g and 0.095 g respectively, being both of them below 0.1 g, which
5.1.2. Multiple regression analysis for the definition of vulnerability index
weights Table 4
The multiple regression analysis was performed using R software Regression coefficients from MR0 model in absolute value and parameters
[65]. The multiple linear regression model (MR0) obtained measures weight definition.
the capacity of the ten parameters variables to predict LS3 adopting the
β |β| Vulnerability index weight (pi)
simplest function as the structure:
(7) P1 0.058 0.100 1.00
MR0: LS 3 P1 + P 2 + P 3 + P 4 + P 5 + P 6 + P 7 + P 8 + P 9 + P10
P2 0.017 0.029 0.50
which leads to the following regression equation: P3 0.087 0.150 1.50
P4 0.040 0.070 0.75
LS3 = 1.681 0.058 × P1 0.017 × P 2 0.087 × P 3 0.04 P5 0.096 0.165 1.50
P6 0.032 0.056 0.50
× P4 0.096 × P5 0.032 × P 6 0.088 × P 7 0.093 P7 0.088 0.152 1.50
× P8 0.042 × P 9 0.026 × P10 (8) P8 0.093 0.160 1.50
P9 0.042 0.072 0.75
Fig. 8 illustrates the predicted versus observed LS3 values. All values P10 0.026 0.045 0.50
are close to the 45° line and most of them lie within the 45° ± 0.1 g

11
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 9. Vulnerability index parameters definition using normalized regression coefficients (|β|).

can be considered small. The coefficient of determination, R2, is 0.772 expert opinion was collected through a survey questionnaire carefully
and measures how well the model fits the actual data. It is noted that R2 prepared. However, there is always a challenge in transforming sub-
here refers to the adjusted R-squared. The performance of the model in jective data and opinions expressed in terms of preferences and com-
terms of adjusted R2 is deemed satisfactory (R2 = 0.772). parisons into quantitative results that can be used for the numerical
The results from the regression analysis confirmed the ability of the definition of the parameters weight. The questionnaire was prepared in
multiple regression equation to estimate LS3. Subsequently, the ten key order to be later processed using the Analytical Hierarchy Process
parameters selected are confirmed to be relevant predictors of the (AHP) proposed by Saaty [67]. It has been employed for similar studies
maximum seismic capacity of the building. Thus, the regression equa- in other fields of civil engineering that required the definition of
tion (Eq. (8)) could be used for the estimation of the parameters weights parameters weights based on expert judgment [68,69].
through the regression coefficients (βk). It is worth highlighting that all
coefficients are negative because an increase in the seismic vulner- 5.2.1. Questionnaire survey and analytical Hierarchy process (AHP)
ability parameter class leads to a decrease in the output variable. In this methodology
case, understanding a unit increase in a predictor is straightforward The questionnaire survey was prepared using the Google Forms
because it directly corresponds to an increase in the vulnerability class, digital platform. The survey was web based, which facilitated the dis-
and all of the predictors are defined within the same range of classes, tribution among experts throughout the world. The AHP is based on
from 1 to 4. For example, a unit increase in the seismic vulnerability establishing pairwise comparisons among the parameters under eva-
class of P1 leads to a decrease in the maximum capacity of the building luation in order to judge their relative importance in pairs. Thus, the
of 0.058 g, but a unit increase in the class of P5 leads to a decrease in survey asks the respondents to compare the relative importance of the
LS3 of 0.096 g, which is almost the double (Eq. (8)). The higher the parameters in a scale from 1 to 9, according to the fundamental scale
absolute values of the regression coefficient of a specific parameter, the defined by the AHP [67]. In this scale, 1 means that the factors under
stronger its influence in determining LS3. comparison have equal importance, while 9 means that one factor is
Table 4 compiles the absolute values of the regression coefficients extremely more important than the other. Thus, a question of the
obtained. The table also shows the normalized absolute values (|β|) so survey was composed of two parts: (1) a pairwise comparison intended
that they all sum to unity, which are used for the numerical definition to indicate which parameter is more important, if any; and (2) a rating
of the parameters weights. Vulnerability index methods available in the scale to compare the relative importance of the two parameters ac-
literature [21,23,33,36,66] typically use weights for the parameters cording to the AHP scale. An example of the questions reads like:
ranging from 0.25 to 1.50 in 0.25 intervals. Therefore, the criterion
followed for the definition of the weights consisted of approximating (1) Please, indicate which of the following parameters you consider has
the normalized regression coefficients to multiples of 0.025 establishing a greater influence on the seismic behavior of vernacular buildings:
0.050 and 0.150 as minimum and maximum respectively. Five ranges o TYPE OF MATERIAL: Type of material used for the structural
could be defined, directly associated to different weights: (a) elements (load bearing walls) of the building (e.g. rammed earth,
< 0.0625 pi = 0.50 ; (b)0.0625 < < 0.0875 pi = 0.75; (c) irregular stone masonry, dressed stone masonry, etc.)
0.0875 < < 0.1125 pi = 1.00 ; (d) 0.1125 < < 0.1375 pi = 1.25; o HORIZONTAL DIAPHRAGMS: Type of horizontal diaphragm
and (e) > 0.1375 pi = 1.50 . The final weights are also displayed in (floors and roofs) and its connection to the load bearing walls
Table 4. Fig. 9 shows the graphical representation of the coefficients for o Equal importance
a better visualization of their relative importance and the criterion used (2) Compare their relative importance in the provided scale from 1 to
for the definition of the weights. 9:

5.2. Definition of the parameters weight based on expert opinion

1 2 3 4 5 6 7 8 9
Equal impor- o o o o o o o o o Extreme importance of one
As aforementioned, besides the analytical procedure, the present tance parameter over the other
paper also introduces the expert point of view into the understanding
and definition of the parameters weight for the SVIVA formulation. The Two different surveys were prepared with a different set of 25 questions

12
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 10. Respondents profile in terms of: (a) assumed expertise according to academic status; and (b) country of origin.

in each other, which were deemed enough to apply the AHP, which survey among specialists from seismic prone regions all over the world,
requires certain amount of redundancy in the answers. The objective of so that the particularities of the seismic behavior of vernacular con-
using the AHP was to generate a pairwise comparison matrix of order n structions from different areas of the world are indirectly included
(where n is the number of parameters), from which the parameters can within the survey. Experts from countries such as Italy, Greece, Mexico,
be arranged hierarchically in terms of importance for the definition of Peru, Iran, and Nepal were surveyed (Fig. 10b).
the weights. Since the matrix is reciprocal and the value of importance
assigned to one parameter i when compared with parameter j is the 5.2.2. Definition of the weights
reciprocal value assigned to j when compared with i (i.e. aji = 1 aij ), the Following the AHP, the weights were derived using the eigenvector
number of judgments required is n(n-1)/2. This would result in pre- method. This method establishes that the attribute weights are equal to
paring 45 questions to consider the 10 parameters. However, the the components of the normalized principal eigenvector of the pairwise
number of questions was reduced in order to avoid that the survey comparison matrix resulting from the survey. Thus, for each respondent
becomes repetitive. The problems of the incompleteness of the matrix of the survey, a matrix was composed using the answers provided and
could be tackled based on the method proposed by Harker [70], which completing the missing elements of the matrix with the above-
allows completing the matrix and thus deriving the weights without the mentioned method proposed by Harker [70]. From each matrix A, the
need of including the complete set of pairwise comparisons in the principal eigenvector (w) can be computed as:
survey.
The survey was distributed and collected from a total of 50 experts Aw = max w (9)
working in the field of seismic engineering and historical and verna-
where max is the largest or principal eigenvalue of A. The principal
cular masonry structures. The experts mainly belong to the academic
eigenvector (w) obtained using Eq. (9) is then normalized. Finally, ac-
environment, including professors, post-doctoral researchers, Ph.D and
cording to the AHP the components of the normalized principal ei-
master’s degree students. They were grouped in three categories based
genvector (w ) can be directly associated with the parameters weights
on the assumed level of knowledge in the field: expert, high and regular
( pi ):
knowledge. Professors and post-doctoral researchers were considered to
have expert knowledge, Ph.D researchers were classified as high wi,1 = pi (10)
knowledge and the master’s degree students as regular knowledge
(Fig. 10a). A particular emphasis was also placed on distributing the This procedure was followed for the total of 50 answers collected,
calculating the vector of relative weights for each answer. Table 5

Table 5
Relative weights considering the mean values from all answers.

w mean σ C.o.V.

P1 0.113 0.061 53.96%

P2 0.076 0.046 59.97%
P3 0.087 0.085 97.44%
P4 0.153 0.079 51.45%
P5 0.128 0.073 57.43%
P6 0.059 0.045 77.11%
P7 0.069 0.050 72.64%
P8 0.122 0.096 78.87%
P9 0.101 0.081 80.03%
P10 0.092 0.064 70.21%

13
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Fig. 11. (a) Parameters weights according to the assumed level of knowledge of the respondents; and (b) different scenarios considered for the definition of the
weights taking into account the levels of knowledge.

shows the mean value of the relative weights obtained from all the considering a 50% variation between the knowledge levels. The com-
answers. There is a great variety in the results, as shown by the standard parison of the results shows that there are not significant differences
deviation (σ) and coefficient of variation (C.o.V.), which reaches values among them.
close to 1 for some parameters (such as P3). This depicts a wide dis- According to the opinion of experts, the parameters that are more
crepancy among the experts in identifying the most influential para- relevant on the seismic behavior of vernacular buildings are: P1 (wall
meters and thus emphasizes the difficulty in assigning the weights slenderness), P4 (wall-to-wall connections), P5 (horizontal diaphragms)
based solely on empirical observations. and P8 (number of floors). Fig. 12 shows the frequency of the different
Fig. 11a presents the mean values of the parameters weights cal- parameters to be selected by the different respondents as the most and
culated independently for each level of knowledge assumed for the the least influential. This confirms that the majority of the respondents
respondents, namely expert, high and regular. There are differences understand that the connection between perpendicular walls, the type
between the weights assigned by experts with different level of of horizontal diaphragms and the number of floors are crucial in de-
knowledge but, with the exception of P3, the relative difference among fining the seismic behavior of a building. It is worth highlighting that
the parameters weights is similar for the three levels of knowledge. The the type of material is considered by the majority of the respondents to
weights shown in Table 5 were calculated assuming the same im- be the least influential parameter in the seismic behavior, while around
portance in the answers for the different levels of knowledge. However, 10% consider this parameter to be the most important. This is in
in order to refine the definition of the weights taking into account the agreement with the great variability shown in Table 5, recognizing a
assumed level of knowledge of the respondents, two different scenarios notable lack of consensus over the influence of this parameter. There is
were additionally considered and shown in Fig. 11b: (1) increase in a quite uniform understanding that the maximum wall span (P2), the
20% the importance of the answers from the Expert level and decrease roof thrust (P6) and wall openings (P7) are among the least influential
in the same proportion the importance of the answers obtained from the parameters.
Regular level respondents; and (2) the same as in the previous case but

Fig. 12. Frequency of the parameters being selected as: (a) the most; and (b) the least influential.

14
J. Ortega, et al. Engineering Structures 197 (2019) 109381

openings have a smaller influence. On the other hand, most experts

judged as critical the importance of the connection among orthogonal
walls (P4), whereas the numerical results granted it a more moderate
influence in comparison with other parameters. This difference is pos-
sibly due to the fact that a bad connection leading to an out-of-plane
wall collapse is a very common post-earthquake damage pattern. Nu-
merical results showed that damage usually arises at the connection
between walls but the simulation of the reduction of the wall-to-wall
integrity appears not to have such a strong effect on the seismic capa-
city of the building when compared with other parameters [54].
Finally, the definition of the weights for the vulnerability index
formulation adapted for vernacular architecture is finally based on the
numerical results, using the correlation shown in Table 4 after the
statistical analysis. The main idea behind the definition of a new SVIVA
formulation was to use an analytical procedure in order to add ro-
bustness to existing formulations, which are traditionally based on
Fig. 13. Comparison among the relative weights calculated based on expert
empirical approaches and expert opinion. The expert opinions obtained
opinion and numerically.
are mainly intended to serve as a comparison, not to disregard its im-
portance on the evaluation of the weights.
5.3. Comparison between the statistical and expert-based approaches As a reference, the weights calculated numerically are compared
with values proposed in the literature by other authors. Table 6 shows
Despite the variations in the answers to the questionnaire, results this comparison, acknowledging that the parameters are not exactly the
show consistency and are relevant, providing a good insight of the same for all the methods, neither have the same exact physical
expert opinion on this matter. Nevertheless, there are important dif- meaning. There is an expected high variability among most of the
ferences with the results obtained from the numerical analysis. Fig. 13 weights, such as P4, P5 and P6. There is a noticeable disagreement in
shows a comparison among: (a) the weights obtained from the statis- the case of wall openings (P7), which existing methods confer a
tical approach (Table 4); and (b) the weights calculated based on expert minimal importance, but the parametric study proved that it can be a
judgment, finally established considering the scenario where the im- decisive parameter in some cases, particularly when coupled with rigid
portance of the answers from the respondents assumed to have an ex- diaphragms [54]. Opposite to what was observed in Fig. 13 where the
pert level of knowledge is increased 50% (Fig. 11b). survey respondents did not consider much influential the type of ma-
Both experts and numerical results led to overall similar weights for terial (P3), there is a general consensus among the methods in the lit-
some parameters. For example, P5 is considered among the most in- erature on the importance of parameter P3, which particularly con-
fluential parameters. The weights obtained for P1, P6, P8 and P9 are siders the mechanical properties and strength of the material. There is
also well-matched. However, there is a remarkable difference for other also agreement in giving moderate importance to the roof thrust (P6)
parameters. For example, the type of material (P3) and the wall and the previous structural damage in the building (P9). All methods
openings (P7) have a significant influence according to the numerical also agree in conferring a reduced importance to the maximum wall
results, while the survey respondents did not give them such an im- span (P2).
portance. In the case of the type of material, the discrepancy among the
experts has already been discussed. Regarding wall openings, the dif-
6. Seismic vulnerability index for vernacular architecture (SVIVA)
ference may lie in the fact that the classes defined in the numerical
formulation
results take into account the effect of wall openings when the building
presents rigid diaphragms that activate the in-plane response of the
Table 7 shows the proposed final SVIVA vulnerability index for-
walls, which is the scenario for which this parameter has a greater in-
mulation for vernacular architecture as the main outcome of the paper.
fluence. Vernacular constructions many times lack this positive dia-
Vulnerability index methods provide a measure of the lack of seismic
phragm effect and seismic behavior mainly depends on the out-of-plane
safety of the buildings through a vulnerability index (IV). In the SVIVA
resistance of the walls, on which the area of wall openings has a
case, the building is evaluated by providing a vulnerability class for
minimum influence [54]. Post-earthquake damage observations
each of the ten parameters. Four classes of increasing vulnerability are
showing out-of-plane collapses may lead to the assumption that wall
defined and associated with a qualification coefficient (Cvi). Following

Table 6
Comparison among the parameters weights proposed by different authors.
Vulnerability index weight (pi)

Parameter Value adopted Benedetti and Petrini [21] Vicente [23] Boukri and Bensaibi [36] Ferreira [65] Shakya [33]

P1: Wall slenderness 1.00 – – – 0.50 –

P2: Maximum wall span 0.50 0.25 0.50 – 0.50 –
P3: Type of material 1.50 1.50 1.50 1.50 0.75 1.50
P4: Wall-to-wall connections 0.75 1.00 1.00 1.00 0.50 1.50
P5: Horizontal diaphragms 1.50 1.00 1.00 0.25 0.50 0.50
P6: Roof thrust 0.50 1.00 1.00 0.25 0.50 –
P7: Wall openings 1.50 – 0.50 – 0.50 1.00
P8: Number of floors 1.50 – 1.50 – – –
P9: Previous structural damage 0.75 1.00 1.00 1.00 0.75 1.00
P10: In-plane index 0.50 0.50 0.75 0.50 – 1.00

15
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Table 7 parameters, an extensive parametric study was performed to evaluate

Vulnerability index formulation for vernacular architecture. and quantify numerically their influence on the seismic performance of
Parameter Class (Cvi) Weight (pi) Vulnerability index vernacular buildings. The influence of the different parameters was thus
validated through numerical analysis and the results were used to
A B C D provide a quantitative definition of four seismic vulnerability classes for
each parameter. The vulnerability classes proposed are overall in line
P1. Wall slenderness 0 5 20 50 1.00 10
P2. Maximum wall span 0 5 20 50 0.50
IV = i = 1 Cvi × pi
with the classifications available in the literature, but are adapted to the
P3. Type of material 0 5 20 50 1.50 evaluation of the seismic vulnerability of vernacular buildings.
P4. Wall-to-wall 0 5 20 50 0.75 The completion of the vulnerability index formulation also required
connections the determination of the parameters weights, which indicate the re-
P5. Horizontal diaphragms 0 5 20 50 1.50 0 IV 500
lative importance of each parameter in estimating the overall seismic
P6. Roof thrust 0 5 20 50 0.50
P7. Wall openings 0 5 20 50 1.50 vulnerability of the building. Two approaches were followed for the
P8. Number of floors 0 5 20 50 1.50 Normalized index definition of the weights: (a) statistical; and (b) expert judgment. With
P9. State of conservation 0 5 20 50 0.75 0 IV 100 respect to the statistical approach, all models built for the parametric
P10. In-plane index 0 5 20 50 0.50
analysis composed an extensive database that was used to perform a
regression analysis. This allowed obtaining regression coefficients that
were used to compare the relative importance of the parameters to
the common vulnerability index formulations existing in the literature,
estimate the seismic behavior of the models. These coefficients were
the qualification coefficients are the same for all parameters. Thus, class
then associated with the weights of the seismic vulnerability assessment
A, related to the lowest vulnerability, has a qualification coefficient
parameters. The second approach was mostly intended to perceive the
(Cvi = 0 ), while class D, related to the highest vulnerability has a qua-
overall view of experts from different seismic prone regions throughout
lification coefficient (Cvi = 50 ). Each parameter is also associated to a
the world on the topic, and to compare their empirical judgment with
weight (pi), reflecting its relative importance and ranging from 0.5 for
the results obtained from the numerical campaign. The high scatter
the least important to 1.5 for the most important ones. The vulner-
observed in their answers confirmed the difficulty of assessing the in-
ability index (IV) is calculated as the weighted sum of ten parameters
fluence of the different parameters based only on expert judgment and
using the equation shown in Table 7. The value of IV ranges between 0
post-earthquake damage observations. This fact also justified the use of
and 500 but, it is common, for ease of use, to normalize it to fall within
an analytical-based approach for enhancing the comprehension of the
a range between 0 (very low vulnerability) and 100 (very high). Ulti-
contribution of the parameters to the structural response of vernacular
mately, as previously discussed, using the approach proposed by
buildings under seismic loading. In the end, a comparison between the
Vicente [23] allows transforming the vulnerability index (IV) into the
weights obtained through the statistical and expert judgment ap-
vulnerability index from the macroseismic method (V). Then, the ana-
proaches was provided and discussed. The final weights were also
lytical expression from the macroseismic method (Eq. (1)) can be used
compared with the weights already used by similar methods available
to estimate the mean damage grade (μD) that the building would suffer
in the literature.
for different seismic inputs and perform a complete seismic vulner-
The results obtained highlighted the complexity of the matter and
ability and loss assessment.
the difficulty to explain the seismic behavior of vernacular buildings
only by considering the influence of ten independent parameters. The
7. Conclusions
relationship between parameters is more complex than the one pro-
posed by vulnerability index methods, where the influence of the
The main objective of the present paper was to show the develop-
parameters is considered individually. Nevertheless, the approach fol-
ment of a new formulation for the seismic vulnerability index method
lowed for the development of the SVIVA formulation was considered
that targets specifically vernacular architecture: the SVIVA formulation.
satisfactory, since the proposed seismic vulnerability assessment
Using the framework of vulnerability index methods, the SVIVA for-
method is intended to be an easy-to-apply tool. Moreover, it also con-
mulation is intended to be an expedited simplified approach. It provides
tributed to gain a better understanding of the structural behavior of
the possibility of performing a primary seismic safety assessment by
many representative Portuguese vernacular architectural heritage
obtaining an indicator of the seismic performance of a building or group
typologies under seismic loading, acknowledging that Portuguese ver-
of buildings based on expedited surveys that can be carried out by
nacular buildings share many characteristics with other vernacular
means of simple visual inspection. As every vulnerability index method
constructions throughout the world. Thus, the results obtained in this
existing in the literature, it correlates the vulnerability of a structure
work can also be representative for other similar structures outside
with different parameters related to simple geometric and constructive
Portugal.
characteristics.
The development of the proposed method was based on obtaining a
comprehensive understanding of the seismic behavior of vernacular
Acknowledgments
architecture, by means of detailed finite element modeling and non-
linear static analysis. The use of an analytical procedure to define the
The work presented in this paper was partly financed by FEDER
proposed vulnerability index formulation and the classes is considered
funds through the Competitivity and Internationalization Operational
as the main novelty of the present method and helps to strengthen the
Programme – COMPETE and by national funds through FCT –
reliability of this simplified method, since existing formulations rely
Foundation for Science and Technology within the scope of the project
solely on empirical results obtained from post-earthquake surveys and
POCI-01-01-0145-FEDER-007633.
expert judgment. A set of geometric, material and construction para-
meters were selected based on literature review and, according to those

Appendix A

See Table A1.

16
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Table A1
Seismic vulnerability classes proposed for the ten key parameters.

(continued on next page)

17
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Table A1 (continued)

(continued on next page)

18
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Table A1 (continued)

References [20] Rosti A, Rota M, Penna A. Damage classification and derivation of damage prob-
ability matrices from L’Aquila (2009) post-earthquake survey data. Bull Earthq Eng
2018;16(9):3687–720.
[1] Fernandes J, Mateus R. Energy efficiency principles in Portuguese vernacular ar- [21] Benedetti D, Petrini V.Sulla Vulnerabilità Di Edifici in Muratura: Proposta Di Un
chitecture. In: Proc of international conference BSA, Porto, Portugal; 2012. Metodo Di Valutazione, L’industria delle Costruzioni 1984;149(1): 66–74.
[2] Correia M. Experiences from past for today’s challenges, in: The road to sustainable [22] Coburn AW, Spence RJS, Pomonis A. Vulnerability and risk assessment, Disaster
development. Traditional and generational change, La fábrica, Fundación Management Training Programme (DMTP) of the United Nations Development
Contemporánea, Madrid, Spain; 2017 [chapter 6]. Programme (UNPD). Cambridge, U.K.: Cambridge Architectural Research Limited;
[3] Vasconcelos G, Lourenço PB. Experimental characterization of stone masonry in 1994.
shear and compression. Constr Build Mater 2009;23(11):3337–45. [23] Vicente R. Estratégias e metodologias para intervenções de reabilitação urbana.
[4] Vasconcelos G, Lourenço PB. In-plane experimental behavior of stone masonry Avaliação da vulnerabilidade e do risco sísmico do edificado da Baixa de Coimbra,
walls. J Struct Eng (ASCE) 2009;135(10):1269–78. Ph.D. thesis. Aveiro, Portugal: Universidade do Aveiro; 2008.
[5] Varum H, Figueiredo A, Silveira D, Martins T, Costa A. Outputs from the research [24] Calvi GM, Pinho R, Magenes G, Bommer JJ, Restrepo-Vélez LF, Crowley H.
developed at the university of Aveiro regarding the mechanical characterization of Development of seismic vulnerability assessment methodologies over the past 30
existing adobe constructions in Portugal. Revista Informes de la Construcción years. ISET J Earthq Technol 2006;34(472):75–104.
2011;63(523):127–42. [25] Jaiswal K, Aspinall W, Perkins D, Wald D, Porter KA. Use of expert judgement to
[6] Miccoli L, Oliveira DV, Silva R, Müller U, Schueremans L. Static behaviour of estimate seismic vulnerability of selected building types. In: Proc of 15th world
rammed earth: experimental testing and finite element modelling. Mater Struct conference on earthquake engineering, Lisbon, Portugal; 2012.
2014;48(10):3443–56. [26] ATC-13. Earthquake damage evaluation data for California, Applied Technology
[7] Miccoli L, Drougkas A, Müller U. In-plane behaviour of rammed earth under cycling Council (ATC), Redwood City, California, USA; 1985.
loading: experimental testing and finite element modelling. Eng Struct [27] HAZUS. HAZUS earthquake loss estimation methodology: technical manual, Vol. 1,
2016;125:144–52. Federal Emergency Management Agency (FEMA), Washington D.C., USA; 1999.
[8] Aktas YD. Seismic resistance of traditional timber-frame Himis structures in Turkey: [28] Giovinazzi S, Lagomarsino S. A macroseismic model for the vulnerability assess-
a brief overview. Int Wood Prod J 2017;8:21–8. ment of buildings. In: Proc of 13th world conference on earthquake engineering,
[9] Preti M, Neffati M, Bolis V. Earthen masonry infill walls: use of wooden boards as Vancouver BC, Canada; 2004.
sliding joints for seismic resistance. Constr Build Mater 2018;184:100–10. [29] Barbat A, Eeri M, Yépez Moya F, Canas JA. Damage scenarios simulation for seismic
[10] Neves F, Costa A, Vicente R, Oliveira CS, Varum H. Seismic vulnerability assessment risk assessment in urban zones. Earthq Spectra 1996;12(3):371–94.
and characterization of the buildings on Faial Island, Azores. Bull Earthq Eng [30] Vicente R, Parodi S, Lagomarsino S, Varum H, Mendes da Silva JAR. Seismic vul-
2012;10(1):27–44. nerability and risk assessment: a case study of the historic city centre of Coimbra,
[11] Sorrentino L, Liberatore L, Liberatore D, Masiani R. The behaviour of vernacular Portugal. Bull Earthq Eng 2011;9(4):1067–96.
buildings in the 2012 Emilia earthquakes. Bull Earthq Eng 2013;12(5):2367–82. [31] Ferreira TM, Vicente R, Mendes da Silva JAR, Varum H, Costa A. Seismic vulner-
[12] Sayin E, Yön B, Calayir Y, Karaton M. Failures of masonry and adobe buildings ability assessment of historical urban centres: case study of the old city centre in
during the June 23, 2011 Maden-(Elazig) earthquake in Turkey. Eng Fail Anal Seixal Portugal. Bull Earthq Eng 2013;11(5):1753–73.
2013;34:779–91. [32] Ferreira TM, Maio R, Vicente R. Seismic vulnerability assessment of the old city
[13] Gautam D, Prajapati J, Paterno KV, Bhetwal KK, Neupane P. Disaster resilient centre of Horta, Azores: calibration and application of a seismic vulnerability index
vernacular housing technology in Nepal. Geoenviron Disast 2016;3(1). method. Bull Earthq Eng 2017;15(7):2879–99.
[14] D’Ayala D. Correlation of fragility curves for vernacular building types: houses in [33] Shakya M. Seismic vulnerability assessment of slender masonry structures Ph.D.
Lalitpur, Nepal and Istanbul, Turkey. In: Proc of the 13th world conference on thesis Portugal: Universidade de Aveiro, Aveiro; 2014.
earthquake engineering, Vancouver BC Canada; 2004. [34] GNDT. Scheda di esposizione e vulnerabilità e di rilevamento danni di primo livello
[15] Barros R, Rodrigues H, Varum H, Costa A, Correia M. Seismic analysis of a e secondo livello (muratura e cemento armato), GNDT (Italian Group for the
Portuguese vernacular building. J Archit Eng 2017;24(1). Defense against Earthquakes), Rome, Italy; 1994.
[16] Carabbio R, Pieraccini L, Silvestri S, Schildkamp M. How can vernacular con- [35] Sepe V, Speranza E, Viskovic A. A method for large-scale vulnerability assessment of
struction techniques sustain earthquakes: the case of the Bhatar buildings. Front historic towers. J Struct Control Health Monit 2008;15:389–415.
Built Environ 2018;4:18. [36] Boukri M, Bensaibi M. Vulnerability index of Algiers Masonry buildings. In: Proc of
[17] Ortega J, Vasconcelos G, Rodrigues H, Correia M. Assessment of the influence of 14th world conference on earthquake engineering, Beijing, China; 2008.
horizontal diaphragms on the seismic performance of vernacular buildings. Bull [37] Ferreira TM, Vicente R, Varum H. Seismic vulnerability assessment of masonry
Earthq Eng 2018. facade walls: development, application and validation of a new scoring method.
[18] Kouris L. Seismic fragility curves for timber-framed masonry structures based on Struct Eng Mech 2014;50(4):541–61.
empirical damage data. Int J Sustain Mater Struct Syst 2016;2:233–49. [38] Blondet M, Villa García GM, Brzev S, Rubiños A. Earthquake-resistant construction
[19] Gautam D, Rodrigues H. Seismic vulnerability of urban vernacular buildings in of adobe buildings: a tutorial. Oakland, California, USA: Earthquake Engineering
Nepal: case of Newari Construction. J Earthquake Eng 2018.

19
J. Ortega, et al. Engineering Structures 197 (2019) 109381

Research Institute (EERI); 2011. 2011.

[39] Bothara J, Brzev S. A tutorial: improving the seismic performance of stone masonry [56] Lourenço PB, Krakowiak F, Fernandes FM, Ramos LF. Failure analysis of Monastery
buildings. Oakland, California, USA: Earthquake Engineering Research Institute of Jerónimos, Lisbon: how to learn from sophisticated numerical models. Eng Fail
(EERI); 2012. Anal 2007;14:280–300.
[40] Ortega J, Vasconcelos G, Rodrigues H, Correia M, Lourenço PB. Traditional earth- [57] Lourenço PB, Karanikoloudis G, Mendes N, Corallo C. Assessment of the South aisle
quake resistant techniques for vernacular architecture and local seismic cultures: a in Canterbury. In: Proc of the 2nd international conference on preservation,
literature review. J Cult Herit 2017;27:181–96. maintenance and rehabilition of historical buildings and structures, Porto, Portugal;
[41] Correia M. A habitação vernácula rural no Alentejo, Portugal, in: Memorias del IV 2015.
Seminario Iberoamericano sobre vivienda rural y calidad de vida en los asenta- [58] Lourenço PB, Karanikoloudis G, Greco F. In situ testing and modeling of cultural
mientos rurales, Santiago del Chile, Chile; 2002. heritage buildings in Peru. In: Proc. of the 10th International conference on struc-
[42] Correia M. Taipa no Alentejo. Lisboa, Portugal: Argumentum; 2007. tural analysis of historical constructions (SAHC 2016), Leuven, Belgium; 2016.
[43] Spence RJS, D’Ayala D. Damage assessment and analysis of the 1997 Umbria- [59] Lourenço PB. Recent advances in masonry structures: micromodelling and homo-
Marche earthquakes. Struct Eng Int 1999;9(3):229–33. genization. In: GalvanettoAliabadi Ferri, editors. Multiscale modeling in solid me-
[44] Lourenço PB, Oliveira DV, Leite JC, Ingham JM, Modena C, Porto F. Simplified chanics: computational approaches. London, UK: Imperial College Press; 2009. p.
indexes for the seismic assessment of masonry buildings: International database and 280–300.
validation. Eng Fail Anal 2013;34:585–605. [60] Gomes MI, Lopes M, de Brito J. Seismic resistance of earth construction in Portugal.
[45] Giuffrè. Sicurezza en conservazione dei centri storici: Il caso Ortigia, Editrice Eng Struct 2011;33(3):932–41.
Laterza, Bari, Italy; 1993. [61] Lourenço PB, Mendes N, Ramos LF, Oliveira DV. Analysis of masonry structures
[46] Binda L, Penazzi D. Classification of masonry cross sections and of typologies of without box-behavior. Int J Architect Heritage Conserv Anal Restor
historic buildings, Technical Report, Book of Commission RILEM MMM; 2000. 2011;5(4–5):369–82.
[47] Binda L, Saisi A. State of the Art of Research on Historic Structures in Italy, [62] Rota M, Penna A, Magenes G. A methodology for deriving analytical fragility curves
Department of Structural Engineering, Politecnico di Milano, Milan, Italy; 2001. for masonry buildings based on stochastic nonlinear analyses. Eng Struct
[48] NTC. Norme Tecniche per le Costruzioni. D.M. 14 gennaio 2008, S.O. No. 30, G.U. 2010;32:1312–23.
No. 29/2008, Rome, Italy; 2008. [63] Ferreira TM, Vicente R, Varum H, Costa A, Costa AA. Out-of-plane seismic response
[49] Tarque N. Seismic risk assessment of adobe dwellings, M.Sc. thesis, Università degli of stone masonry walls: experimental and analytical study of real piers. In: Proc of
Studi di Padova, Padua, Italy; 2008. the 15th world conference on earthquake engineering, Lisbon, Portugal; 2012.
[50] Gallego R, Arto I. Evaluation of seismic behaviour of rammed earth structures. In: [64] Mouyiannou A, Rota M, Penna A, Magenes G. Identification of suitable limit states
Mileto Vegas, Soriano,Cristinieditors. Earthen architecture: past, present and future. from nonlinear dynamic analyses of masonry structures. J Earthq Eng
London, UK: Taylor & Francis Group; 2014. 2014;18(2):231–63.
[51] Angulo-Ibáñez Q, Mas-Tomás Á, Galvañ-Llopis V, Sántolaria-Montesinos JL. [65] R Development Core Team. R: A language and environment for statistical com-
Traditional braces of earth constructions. Constr Build Mater 2012;30:389–99. puting, R Foundation for Statistical Computing, Vienna, Austria; 2008.
[52] Mendes N, Lourenço PB. Seismic vulnerability of existing masonry buildings: non- [66] Ferreira TM. Avaliação da vulnerabilidade sísmica das paredes de fachada de
linear parametric analysis. In: Psycharis, Pantazopoulou, Papadrakakis, editors. edifícios em alvenaria M.Sc. thesis Aveiro, Portugal: Universidade de Aveiro; 2009.
Seismic assessment, behavior and retrofit of heritage buildings and monuments. [67] Saaty RW. The analytic hierarchy process – what it is and how it is used. Math
Computational methods in applied sciences, vol. 37, Springer; 2015. Model 1987;9(3–5):161–76.
[53] Masciotta MG, Ramos LF, Lourenço PB, Matos JAC. Development of key perfor- [68] Liu YC, Chen CS. A new approach for application of rock mass classification on rock
mance indicators for the structural assessment of heritage buildings. In: Proc of 8th slope stability assessment. Eng Geol 2007;89(1–2):129–43.
European workshop on structural health monitoring (EWSHM 2016), Bilbao, Spain; [69] Pinheiro M, Sanches S, Miranda T, Neves A, Tinoco J, Ferreira A, et al. A new
2016. empirical system for rock slope stability analysis in exploitation stage. Int J Rock
[54] Ortega J. Reduction of the seismic vulnerability of vernacular architecture with Mech Min Sci 2015;76:182–91.
traditional strengthening solutions Ph.D. thesis Portugal: Universidade do Minho, [70] Harker PT. Incomplete pairwise comparisons in the analytic hierarchy process.
Guimarães; 2018. Math Model 1987;9(11):837–48.
[55] TNO. DIsplacement method ANAlyser. User’s manual, release 9.4.4., Netherlands;

20