Tackling Non-linearity in Seismic Risk Estimation using Fuzzy Methods
J. Rub
´
en G. C
´
ardenas
1
,
`
Angela Nebot
2
, Francisco Mugica
2
and Helen Crowley
1
1
IUSS UME School Via Ferrata 45, Pavia, Italy
2
Soft Computing Group, Technical University of Catalonia, Jordi Girona Salgado 1-3, Barcelona, Spain
Keywords:
Fuzzy Sets, Risk Management, Natural Hazards, Vulnerability Index, Social Vulnerability, Seismic Vulnera-
bility, Inference System.
Abstract:
Traditional approaches to measure risk to natural hazards considers the use of composite indices. However,
most of the times such indices are built assuming linear interrelations (interdependencies) between the ag-
gregated components in such a way that the final index value is based only on an accumulative or scalable
structure. In this paper we propose the use of Fuzzy Inference Systems type Mamdami in order to aggregate
physical seismic risk and social vulnerability indicators. The aggregation is made by establishing rules (if-
then type) over the indicators in order to get an index. Finally a quantitative seismic risk estimation is made
though the convolution of these two main factors by means of fuzzy inferences, in such a way that no linear
assumptions are used along the estimation. We applied the fuzzy model over the city of Bogota Colombia.
We consider that this approach is a useful way to estimate a measure of an intangible reality such as seis-
mic risk, by assuming the urban settlement’s complexity where the interrelations between the associated risk
components are inherently non-linear. The proposed model possess a practical use over the risk management
field, since the design of the logic rules uses a smooth application of risk management knowledge following
a multidisciplinary approach, thus making the model easily adapted to a particular circumstance or context
regardless the background of the final user.
1 INTRODUCTION
Holism (from greek: all, whole, entire) is an episte-
mology position which postulate that complex sys-
tems cannot be completely understood by taking un-
der scope each of their components in a separate
way. The holism defines then, the basis for a non-
reductionism methodology for the study of systems.
The idea behind holism is ”the integration of the parts,
through its synergies, to understand the whole” (Car-
dona, 2001). According to a holistic approach, the
”whole” is more complex than the sum of its con-
stituent elements, therefore the total behavior of the
system cannot be derived from its fundamental com-
ponents without considering the trade off of informa-
tion (energy) between them. If we intent to frame risk
to natural hazards into a holistic or integral scheme,
we need to take into account the complexities over an
urban environment. In this terms, an important part of
the urban complexities can be considered as a result
of the non linear interrelationships between the mul-
titude of components conforming the urban system.
The physical risk, defined as the seismic risk com-
ponent that reflects the type of assets that can be dam-
aged because an earthquake occurrence (including
lost lives) have a solid framework of analysis and ex-
perimentation. Even if that the large majority of seis-
mic hazard, vulnerability and exposure models con-
siders a probabilistic approach, the engineering field
has the possibility to compare their results with ex-
perimental data, that can be obtained either from sim-
ulations or practical experiments. Although physical
risk models uses approximations, there is a mark of
reference to compare. At the other hand, how can we
estimate an intangible reality such as social vulnera-
bility?
Social vulnerability is a crucial aspect of risk man-
agement. There can be no analysis or management
without a social vulnerability understanding however,
vulnerability have embedded confuse concepts that
may leads towards many (and some times different)
conclusions. Nevertheless, most of the approaches
used to define social vulnerability, focuses over the
susceptibility and capacities of urban elements to act
against external influences, thus understanding vul-
nerability as a sort of detector capable to determine
the state of the system. Therefore, social vulnerabil-
ity becomes an essential source of information in or-
532
González Cárdenas J., Nebot À., Mugica F. and Crowley H..
Tackling Non-linearity in Seismic Risk Estimation using Fuzzy Methods.
DOI: 10.5220/0005577905320541
In Proceedings of the 5th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (MSCCES-2015), pages
532-541
ISBN: 978-989-758-120-5
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
der to implement suitable hazard and mitigation as-
sessments, reduction and disaster preparedness that
requires first of all, the identification of the vari-
ous dimensions of vulnerability over a society, either
economic, institutional structure or environmental re-
sources.
Carre
˜
no et al. (2012) proposed a seismic risk
model from a holistic perspective, by considering that
seismic risk is the result of physical risk (those el-
ements susceptible to be damage or destroyed) and
an aggravation coefficient that includes both, the re-
silience and the fragility of an urban environment.
By describing physical risk and social aggravation by
means of indices the final estimation of seismic risk is
made by means of the so called Moncho’s equation.
In this paper, we built a holistic seismic risk fuzzy
model considering Cardona-Carre
˜
no risk descriptors.
By establishing fuzzy logic rules between such de-
scriptors we were able to aggregate them all into a
single seismic risk index without assuming a linear
behavior between the descriptors. We found seismic
risk tendencies and spatial distributions patterns over
Bogota (Colombia) by performing a classical Mam-
dani fuzzy approach, supported by well established
fuzzy theory, which is characterized by a high expres-
sive power and an intuitive human-like manner.
2 CARRE
˜
NO’S MODEL
Taking as a base Cardona’s original model (2001),
Carre
˜
no et al. (2012) proposed a seismic risk model
considering an integral (holistic) approach, regarding
seismic risk as a function of the potential damage on
assets (considering hazard intensities) plus the socioe-
conomic fragilities and lack of resilience of the con-
text. In this view, seismic risk would be the result
of physical risk, aggravated by social conditions and
lack of resilience capacities. Carre
˜
no et al. model re-
lies in the use of descriptors for both: physical risk
(see the 8 physical risk descriptors of Figure 1) and
social aggravation (see the 11 aggravation descriptors
of Figure 1).
A conceptualization of Carre
˜
no’s seismic risk
model can be seen in Figure 1.
Carre
˜
no et al. (2012) obtained a seismic risk eval-
uation for Bogota city by means of indicators that
leads to the calculation of a total risk index. This is
obtained by direct application of Moncho’s equation
described in 1:
R
T
= R
Ph
(1 + F) (1)
where R
T
is the total risk, R
Ph
is the physical risk and
F is a aggravation coefficient.
Figure 1: Carre
˜
no et al. (2012) Holistic Seismic Risk
Model.
Thus, considering seismic risk as produced for
physical and an aggravation coefficient; the risk in-
dex provides an approximate vision of the state of the
social capital infrastructure.
The physical risk is evaluated by using the Equa-
tion 2,
R
Ph
=
p
i=1
w
R
Ph
i
F
R
Ph
i
(2)
where F
R
Ph
i
are the physical risk descriptors, w
R
Ph
i
are
their weights assessed by an analytic hierarchy pro-
cess (Carre
˜
no et al., 2007; Saaty and Vargas, 1991),
and p the total number of considered descriptors in
the estimation. The physical risk descriptors values
can be obtained from previous physical risk evalua-
tions (damage scenarios) already made at the studied
location.
The aggravation coefficient, F, depends on a
weighted sum of an aggravation descriptors set as-
sociated to socioeconomic fragility of the commu-
nity (F
SFi
) and lack of resilience of exposed context
(F
LR j
), according to Equation 3,
F =
m
i=1
w
SFi
F
SFi
+
n
j=1
w
LR j
F
LR j
(3)
where w
SFi
and w
LR j
are the assessed weights on
each factors and m and n the total number of descrip-
tors for fragility and lack of resilience, respectively.
TacklingNon-linearityinSeismicRiskEstimationusingFuzzyMethods
533
The descriptors values were obtained from existent
databases and statistical data of the studied area.
The use of descriptors conforms an indirect tech-
nique to estimate a quantitative measure of change.
The final aim is to describe intangible realities, hidden
trends or different classes of information in a com-
posite manner in order to present them all as quantifi-
able entities that can be compared across space and
time scales. Basically, descriptors are an encapsu-
lation of a more complex reality using a single con-
struct, and they can be used solely as independent en-
tities of measure, or they can be aggregated to form
indices. Since an index is intended to describe a par-
ticular attribute, the attribute will determine a sort of
causality structure between it and the descriptors that
can be either reflective (the attribute influences the de-
scriptors) or formative (the descriptors influence the
attribute). The main difference is based in the internal
correlation of the descriptors. In the case of Carre
˜
no’s
indices, there is a strong formative causality struc-
ture and therefore, the attributes can be assumed to
be interdependent. Therefore, the final outcome of
descriptor’s aggregation would estimate the attribute
considering only a linear influence between descrip-
tors. Even if this assumption may be valid in some
circumstances, the main objective of the index is lost
since in fact, there is no a real measure of conditional-
ity or causality between descriptors, since none of the
indicators can be diminished or amplified by another
indicator, and therefore there is no way to assess non-
linearities or feedbacks that do exist in real systems.
In the same way Moncho’s equation presents a
linear relation between its physical and aggravation
components. Since the main assumption behind Mon-
cho’s equation is to consider physical risk as the main
seismic risk ”driver”, a clear consequence of assum-
ing linearity would be that the final estimation of the
total seismic risk will rely in the existence, first, of a
significative physical risk value. Therefore, if a region
presents large social aggravation values, but the val-
ues for physical risk are small or medium for the same
area, the final risk estimation will be small. This as-
sumptions can lead to a underestimation in the final
risk estimation that can be misleading. For exam-
ple, an area with an important aggravation or social
vulnerability, might be severely affected by an earth-
quake of less intensity and the effects could be even
bigger and more spread, creating a series of unwanted
consequences that cannot be estimated using a linear
relationship between seismic risk components.
3 FUZZY SEISMIC RISK MODEL
The integral frame that we followed was built in the
believe that seismic risk can be viewed as the convo-
lution of two principal components: the social aggra-
vation and physical risk, which in turn forms the total
seismic risk of an urban center. The Fuzzy Seismic
Risk Model (FSRM) is divided in three main mod-
ules or sections: Social Aggravation, Physical Risk,
and Total Risk. Each one of them is conformed by dif-
ferent submodules. The main objective is to be able
to estimate seismic risk for an urban center consid-
ering social and physical aspects trough fuzzy infer-
ence modeling, therefore, not assuming a linear inter-
dependency between seismic risk components.
3.1 Aggravation Coefficient
We built an aggravation coefficient by re-defining
and grouping Carre
˜
no’s descriptors into three differ-
ent Fuzzy Inference Systems (FIS) called: resilience,
fragility and aggravation. Each subsystem is defined
by a set of rules involving all proper descriptors. A
conceptualization of the different steps followed to es-
timate the aggravation coefficient can be seen in Fig-
ure 2. The descriptors involved in each subsystem are
presented in the left hand side of Figure 2. FIS 1,
corresponds to the Fragility model and has as input
variables the Marginal Slums (MS), the Social Dis-
parity Index (SDI) and the Population Density (PD).
The output of FIS 1 is the level of fragility. On the
other hand, FIS 2 corresponds to the Resilience model
which have as input variables the Human Health Re-
sources (HHR), the Emergency Operability (EO) and
the Development Level (DL). The output of FIS 2 is
the resilience level. Finally, the Aggravation model
(FIS 3) takes as inputs the fragility and resilience lev-
els that are the output of FIS 1 and 2, respectively,
and infers the aggravation coefficient. All the fuzzy
inference systems proposed in this research are based
on the Mamdani approach (Mamdani and Assilian,
1975), since it is the one that better represents the
uncertainty associated to the inputs (antecedents) and
the outputs (consequents) and allows to describe the
expertise in an intuitive and human-like manner.
The original 11 aggravation descriptors, presented
in Figure 1, were reduced to six variables by consid-
ering a subjective method which is based in the as-
sumption that certain descriptors reflects similar at-
tributes in terms of aggravation formation. For exam-
ple, descriptors called: mortality rate and delinquency
rate, are linked since they reflect negative social con-
sequences produced by a social structure failure, i.e.
a lack of access to certain social advantages, such as
SIMULTECH2015-5thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
534
Figure 2: Mamdani fuzzy classical model structure to estimate Aggravation coefficient.
an efficient public health program, a strong marginal-
ization dynamics, no access to education or effec-
tive justice and law policies. Therefore, we consider
that these particular descriptors could be described us-
ing only the descriptor called social disparity index,
which is a fragility descriptor as well. In the case of
descriptors related to resilience we consider that de-
scriptors called: public space, hospital beds and emer-
gency personnel, are already reflected by the descrip-
tor named emergency operability, since the attributes
of the former descriptors are related with the capacity
of reaction when the emergency is being or has re-
cently occurred. Descriptors called: marginal slums,
population density, human health resources and de-
velopment level remain the same.
We used three linguistic labels defined to qual-
ify each descriptor: low, medium and high, along
their respective universe of discourse. However, for
the FIS outputs, i.e. resilience, fragility and aggrava-
tion, we decided to use five labels: low, medium-low,
medium-high, high and very-high. We think that 3
classes is enough to accurately represent input vari-
ables of resilience and fragility models. Moreover, a
reduced number of classes implies a more compacted
and reduced set of fuzzy rules. To improve model’s
sensibility, we design membership functions in order
to consider the data variability. With these member-
ship functions we build a set of fuzzy rules that could
infer the behavior of the aggravation coefficient us-
ing the three Mamdani fuzzy inference systems men-
tioned before (see Figure 2).
The development of the fuzzy rule base consid-
ers all possible combinations of descriptors linguis-
tic labels, therefore a total of 27 rules (3 descriptors)
characterized by 3 linguistic labels each) where re-
spectively used for infer fragility and resilience val-
ues. We think that the completeness of the fuzzy par-
tition is of great importance in this application. These
rules were intended to follow risk management litera-
ture which suggests possible outcomes when three of
these elements interact to form resilience or fragility.
At the other hand, the aggravation model, that has as
inputs resilience and fragility inferred values, char-
acterized by 5 classes each, is then composed of 25
fuzzy rules. In Table 1, the rules that conform the re-
silience FIS model are presented as an example. The
use of classical fuzzy systems, with well established
fuzzy inference theory, allow to build a solid model,
easily understandable by experts which leads towards
a deepest discussion over social vulnerability descrip-
tion and casuals non linear interrelations.
3.2 Physical Risk Coefficient
In the holistic model presented in Figure 1, 8 descrip-
tors are associated with physical risk formation. Nev-
ertheless, we consider important to include another
descriptor already used in previous studies (Cardona,
2001; Carre
˜
no, 2006) called damage in main roads,
due to its significance for the analysis of seismic risk.
We categorized these new collection of physical risk
descriptors into three different models called: Prop-
erty Damage, Life Lines Sources Damage and Net-
work Damage, each of those was later framed into a
TacklingNon-linearityinSeismicRiskEstimationusingFuzzyMethods
535
Table 1: Rules that compose the resilience FIS model, used
to estimate the level of resilience. HHR = Human Health
Resources; DL = Development Level; EO = Emergency Op-
erability; R = Resilience; VH = very-high; H = high; MH =
medium-high; ML = medium-low; L = low.
1. If (HHR is L) and (DL is L) and (EO is L) then (R is L)
2. If (HHR is M) and (DL is M) and (EO is M) then (R is ML)
3. If (HHR is H) and (DL is H) and (EO is H) then (R is VH)
4. If (HHR is M) and (DL is L) and (EO is L) then (R is L)
5. If (HHR is H) and (DL is H) and (EO is L) then (R is M)
6. If (HHR is L) and (DL is M) and (EO is L) then (R is ML)
7. If (HHR is M) and (DL is M) and (EO is L) then (R is MH)
8. If (HHR is H) and (DL is M) and (EO is L) then (R is H)
9. If (HHR is L) and (DL is H) and (EO is L) then (R is MH)
10. If (HHR is M) and (DL is H) and (EO is L) then (R is MH)
11. If (HHR is H) and (DL is H) and (EO is L) then (R is H)
12. If (HHR is L) and (DL is L) and (EO is M) then (R is ML)
13. If (HHR is M) and (DL is L) and (EO is M) then (R is MH)
14. If (HHR is H) and (DL is L) and (EO is M) then (R is H)
15. If (HHR is L) and (DL is M) and (EO is M) then (R is MH)
16. If (HHR is H) and (DL is M) and (EO is M) then (R is H)
17. If (HHR is L) and (DL is H) and (EO is M) then (R is MH)
18. If (HHR is M) and (DL is H) and (EO is M) then (R is H)
19. If (HHR is H) and (DL is H) and (EO is M) then (R is H)
20. If (HHR is L) and (DL is L) and (EO is H) then (R is MH)
21. If (HHR is M) and (DL is L) and (EO is H) then (R is H)
22. If (HHR is H) and (DL is L) and (EO is L) then (R is H)
23. If (HHR is L) and (DL is M) and (EO is H) then (R is H)
24. If (HHR is M) and (DL is M) and (EO is H) then (R is VH)
25. If (HHR is H) and (DL is M) and (EO is H) then ((R is VH)
26. If (HHR is L) and (DL is H) and (EO is H) then (R is H)
27. If (HHR is M) and (DL is H) and (EO is H) then (R is VH)
FIS structure. Each model contains as inputs three of
the original descriptors. The structure of the physical
risk model can be seen in Figure 3. FIS 1, corresponds
to the Property Damage model and have as input vari-
ables: Damage Area (DA), Dead People (DP) and In-
jured People (INJ). The output of FIS 1 corresponds
to the level of property damage. The FIS 2 represents
the Life Lines Sources Damage model and have as in-
put variables: Telephonic Substation Affected (TSA),
Electrical Substation Affected (ESBA), and Damage
in Water Mains (DWM). The output of FIS 2 is the
level of damage to life lines sources. The FIS 3 corre-
sponds to the Network Damage model and have as in-
put variables: Damage in Gas Network (DGN), Fallen
Length of Electrical Lines (FLEN) and Damage in
Mains Roads (DMR). The output of FIS 3 is the net-
work damage level. FIS 4 corresponds to the Physical
Risk model, which takes as inputs the outputs of all
previous models, i.e. FIS 1, FIS 2 and FIS 3, and then
infers the physical risk coefficient.
We decided to characterize each input variable
(descriptor) into 3 labels: low, medium and high, and
into five labels: low, medium-low, medium-high, high
and very-high, each FIS output, i.e. property dam-
age, life lines sources damage and network damage.
Therefore, as before, 27 fuzzy rules were obtained for
FIS 1, FIS 2 and FIS 3, and 25 for FIS 4. We also
designed the membership functions in order to reach
current data variability.
The rules of the Mamdani physical risk model are
Table 2: Rules that compose the physical risk FIS model,
used to estimate the physical risk coefficient. LLSD = Life
Lines Sources Damage; PD = Property Damage; ND = Net-
work Damage; PR = Physical Risk; VH = very-high; H =
high; MH= medium-high; ML = medium-low; L = low.
1. If (LLSD is L) and (ND is L) and (PD is L) then (PR is L)
2. If (LLSD is M) and (ND is M) and (PD is M) then (PR is MH)
3. If (LLSD is H) and (ND is H) and (PD is H) then (PR is VH)
4. If (LLSD is M) and (ND is M) and (PD is M) then (PR is ML)
5. If (LLSD is H) and (ND is L) and (PD is L) then (PR is MH)
6. If (LLSD is L) and (ND is M) and (PD is L) then (PR is ML)
7. If (LLSD is M) and (ND is M) and (PD is L) then (PR is MH)
8. If (LLSD is H) and (ND is M) and (PD is L) then (PR is H)
9. If (LLSD is L) and (ND is H) and (PD is L) then (PR is MH)
10.If (LLSD is M) and (ND is H) and (PD is L) then (PR is M)
11.If (LLSD is H) and (ND is H) and (PD is L) then (PR is H)
12.If (LLSD is L) and (ND is L) and (PD is M) then (PR is ML)
13.If (LLSD is M) and (ND is L) and (PD is M) then (PR is MH)
14.If (LLSD is H) and (ND is L) and (PD is M) then (PR is H)
15.If (LLSD is L) and (ND is M) and (PD is M) then (PR is MH)
16.If (LLSD is H) and (ND is M) and (PD is M) then (PR is H)
17.If (LLSD is L) and (ND is alto) and (PD is M) then (PR is MH)
18.If (LLSD is M) and (ND is H) and (PD is M) then (PR is H)
19.If (LLSD is H) and (ND is H) and (PD is M) then (PR is H)
20.If (DLLS is L) and (ND is L) and (PD is H) then (PR is MH)
21.If (DLLS is M) and (ND is L) and (PD is H) then (PR is H)
22.If (DLLS is H) and (ND is L) and (PD is H) then (PR is H)
23.If (DLLS is L) and (ND is M) and (PD is H) then (PR is H)
24.If (DLLS is M) and (ND is M) and (PD is H) then (PR is VH)
25.If (DLLS is H) and (ND is M) and (PD is H) then (PR is VH)
26.If (DLLS is L) and (ND is H) and (PD is H) then (PR is H)
27.If (DLLS is M) and (ND is H) and (PD is H) then (PR is VH)
presented in Table 2 as an example.
3.3 Total Risk Coefficient
The FIS called Total Risk will perform the convolu-
tion of all the previous FIS already developed, em-
bedded in one main structure that have as inputs the
variables representing the inferred values of physical
risk and social aggravation. In this case, both, inputs
and outputs, are characterized by the 5 labels men-
tioned in the previous section, obtaining a Mamdani
model composed of a set of 25 fuzzy rules to be used
in the inference process. As previously, the member-
ship functions are designed to represent the data vari-
ability. The Total Risk model structure can be seen in
Figure 4.
4 RESULTS: CITY OF BOGOTA
Colombia’s Capital is divided since 1992 into 20 ad-
ministrative Localities. However in our study we took
into account only 19 on these because the locality
called Sumapaz corresponds basically to the rural area
of the city. For the social aggravation coefficient esti-
mation on each district we used statistical and demo-
graphic data from 2001 (Carre
˜
no et al., 2012).
In Figures 5 and 6 we can see the aggravation val-
ues obtained by the proposed fuzzy model and the
index method, respectively. The general aggravation
SIMULTECH2015-5thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
536
Figure 3: Conceptualization of Mamdani fuzzy classical model to estimate Physical Risk coefficient.
Figure 4: Mamdani fuzzy classical model structure to estimate Total Risk coefficient.
level seems to be lower for the FIS model when com-
pared with the index model. However, the FIS spatial
pattern distributes the highest values of aggravation
at the South-West part of the city as reported by in-
dex method, corresponding to the districts of: Ciudad
Bol
´
ıvar, Bosa, Usme, and San Cristobal. The East
part of the city remains with medium-low, while the
North-West part of the city presents medium-high ag-
gravation values. The index method reaches a very-
high value at South-West part of the city while the
northern part presents mostly a medium-low aggra-
vation value. In these figures we can see that even
there is no correct total match among the two meth-
ods, all of them preserve quite the same order in terms
of higher and lower aggravation values.
The physical risk coefficient values are presented
in Figures 7 and 8 were the results of the fuzzy and
the index models are showed. Although the spatial
patterns changes, the highest values are encountered
in the north part of the city on both models, thus con-
taining the districts of: Usaquen, Suba Barrios Unidos
and Chapinero.
For the rest of the city, the fuzzy model esti-
mates homogeneous medium-high physical risk val-
ues, while in the index method map, this level of risk
is given only for the south part of the city.
The highest levels of physical risk are alike in the
two models. However, the change between higher and
lower physical risk values is more smooth in the fuzzy
model along 5 districts (Tunjuelito, Bosa, Ciudad
Kennedy, Fontib
´
on, Engativ
´
a and Antonio Nari
˜
no),
when compared with the index model. Last, the to-
tal risk values were obtained for the 19 districts of
Bogot
´
a. Figures 9, 10 and 12 show the results for the
totally fuzzy, the Moncho’s-Index and the Moncho’s-
Fuzzy methods, respectively.
The proposed fuzzy model estimate a higher total
risk values for the city of Bogota and a more homoge-
neous spatial pattern. However, the areas with highest
levels, correspond also to those areas with the highest
TacklingNon-linearityinSeismicRiskEstimationusingFuzzyMethods
537
Figure 5: Aggravation coefficients obtained by the proposed
Fuzzy Model: (1) Usaqu
´
en, (2) Chapinero, (3) Santa Fe,
(4) San Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8)
Ciudad Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba,
(12) Barrios Unidos, (13) Teusaquillo, (14) M
´
artires, (15)
Antonio Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18)
Rafael Uribe, (19) Ciudad Bol
´
ıvar.
Figure 6: Aggravation coefficients obtained by the Index
Model:(1) Usaqu
´
en, (2) Chapinero, (3) Santa Fe, (4) San
Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8) Ciudad
Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba, (12) Bar-
rios Unidos, (13) Teusaquillo, (14) M
´
artires, (15) Antonio
Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18) Rafael
Uribe, (19) Ciudad Bol
´
ıvar.
levels according to the index method, even if the cur-
rent values are different. As the physical risk values
obtained by the fuzzy model are higher for a number
of Bogota localities, comparing with the same physi-
cal risk values obtained by the index method (see Fig-
ure 7 vs. Figure 8), the result is a more higher total
risk when using the fuzzy model (see Figure 9 vs. Fig-
Figure 7: Physical Risk coefficients obtained by the pro-
posed Fuzzy Model: (1) Usaqu
´
en, (2) Chapinero, (3) Santa
Fe, (4) San Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8)
Ciudad Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba,
(12) Barrios Unidos, (13) Teusaquillo, (14) M
´
artires, (15)
Antonio Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18)
Rafael Uribe, (19) Ciudad Bol
´
ıvar.
Figure 8: Physical Risk coefficients obtained by the Index
Model: (1) Usaqu
´
en, (2) Chapinero, (3) Santa Fe, (4) San
Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8) Ciudad
Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba, (12) Bar-
rios Unidos, (13) Teusaquillo, (14) M
´
artires, (15) Antonio
Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18) Rafael
Uribe, (19) Ciudad Bol
´
ıvar.
ure 10).
The total risk levels using the Moncho’s-Fuzzy
model (see Figure 12), show the highest level of seis-
mic risk at the northern part of the city, which is con-
gruent with the results from the other two models.
However, a direct application of Moncho’s equation
SIMULTECH2015-5thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
538
Figure 9: Total Risk coefficients obtained by the proposed
Fuzzy models: (1) Usaqu
´
en, (2) Chapinero, (3) Santa Fe,
(4) San Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8)
Ciudad Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba,
(12) Barrios Unidos, (13) Teusaquillo, (14) M
´
artires, (15)
Antonio Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18)
Rafael Uribe, (19) Ciudad Bol
´
ıvar.
Figure 10: Total Risk coefficients obtained by the Moncho’s
equation using aggravation and physical risk Index models
(Moncho’s-Index): (1) Usaqu
´
en, (2) Chapinero, (3) Santa
Fe, (4) San Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8)
Ciudad Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba,
(12) Barrios Unidos, (13) Teusaquillo, (14) M
´
artires, (15)
Antonio Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18)
Rafael Uribe, (19) Ciudad Bol
´
ıvar.
gives an homogeneous medium-high level for almost
all the city. As it can be seen, a direct application of
the Moncho’s equation vanishes the effect of social
aggravation to the total risk.
Figure 11 shows the total risk trend line over 19
districts of the city of Bogot
´
a. The variability esti-
mated by the fuzzy model resembles the one given
by the Moncho’s-Index model, especially at the right
hand side of the plot. However, the districts of: Santa
Fe, San Cristobal and Chapinero, show an opposite
trend from the values estimated by the index method.
The homogeneous values given by the Moncho’s-
Fuzzy model are more clearly seen in this graph,
although the highest risk values looks similar when
comparing with the others two alternative models.
According to the previous analysis, with the use
of classical fuzzy inference system methodology it is
plausible to achieve similar results to those obtained
from a more analytical method such as indexes, in
terms of district classification, or in reproducing spa-
tial patterns of aggravation and physical risk. Fuzzy
logic inference capabilities can be exploited in a more
suitable way since the outputs from each FIS used in
the model are always fuzzy sets, giving the chance to
connect them trough a new FIS without loosing con-
sistency, allowing model completeness and avoiding
the assumption of interdependency between descrip-
tors, in order to calculate a final risk output. There-
fore, one of the main advantages of the model is the
assumption of connectivity to create a risk value that
actually reflects the result of the correlation (both lin-
ear an non linear) between the components that were
assumed to influence seismic risk. It is interesting to
note how by not using Moncho’s equation for esti-
mate total risk allows to have a more clear vision of
the drivers of seismic risk in a non linear way. For
example, Figures 6, 8 and 10 shows the three com-
ponents of seismic risk estimated by index method.
Moncho’s equation follows the idea that there can be
only a high seismic risk if there is a high physical risk
value. If not, no matter if there is a large area pre-
senting a ”very high” aggravation value (the south-
eastern part of the city in figure 6), the final index
will say that the total risk in that area is only be-
tween ”high and medium high”. The opposite behav-
ior is estimated by using fuzzy inferences (Figures
5, 7 and 9) where an area presenting high aggrava-
tions values, corresponds a proportional value of total
risk. As discussed, one of the steps needed to con-
form the designed FIS was in terms of the use of a
subjective methodology in order to reduce the orig-
inal dataset in order to avoid over correlation (dou-
ble counting) between variables. In more general risk
models, the number of variables involved, specially
in the social vulnerability part, can be in the order
of hundreds. Clearly a subjective scheme will not be
enough to reduce the number of variables and more
analytical methods are needed. The most common
way to reduce a data set relies in the use of statistical
TacklingNon-linearityinSeismicRiskEstimationusingFuzzyMethods
539
Figure 11: Total Risk coefficients tendency: (1) Usaqu
´
en, (2) Chapinero, (3) Santa Fe, (4) San Crist
´
obal, (5) Usme, (6)
Tunjuelito, (7) Bosa, (8) Ciudad Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba, (12) Barrios Unidos, (13) Teusaquillo, (14)
M
´
artires, (15) Antonio Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18) Rafael Uribe, (19) Ciudad Bol
´
ıvar.
Figure 12: Total Risk coefficients obtained by the Moncho’s
equation using aggravation and physical risk Fuzzy models
(Moncho’s-Fuzzy): (1) Usaqu
´
en, (2) Chapinero, (3) Santa
Fe, (4) San Crist
´
obal, (5) Usme, (6) Tunjuelito, (7) Bosa, (8)
Ciudad Kennedy, (9) Fontib
´
on, (10) Engativ
´
a, (11) Suba,
(12) Barrios Unidos, (13) Teusaquillo, (14) M
´
artires, (15)
Antonio Nari
˜
no, (16) Puente Aranda, (17) Candelaria, (18)
Rafael Uribe, (19) Ciudad Bol
´
ıvar.
approaches such as linear correlation, principal com-
ponent analysis, or factor analysis. Nevertheless, all
of these methods assume as well, a linear correlation
between variables. At the same time, at the moment
of aggregate descriptors in order to build a composite
index, most of the time linear assumptions are used
which leads to lose information coming from the non
linear interdependency over descriptors that, in fact,
are the most importants. Therefore, the development
of tools to perform variable selection that do not fol-
low strictly linear assumptions are most needed.
5 CONCLUSIONS
We obtain a inference fuzzy model to make an estima-
tion of social aggravation over Bogota city using the
descriptors proposed in (Carre
˜
no et al., 2012). Build-
ing inference compositional rules over the selected
descriptors, we were able to obtain a robust method
that resembles the identification of relevant aspects
and characteristics of seismic risk of cities already
achieved by other consolidated method. The pro-
posed model displays simplicity, flexibility and res-
olution capacities and does not assume linearity be-
tween the different components needed to obtain a fi-
nal outcome.
REFERENCES
Cutter, S. L., Boruff, B. J., Shirley, W. L., (2003) Social
Vulnerability to Environmental Hazards. Social Sci-
ence Quarterly,84.
Cardona, O. D., (2001) Holistic evaluation of the seis-
mic risk using complex dynamic systems (in Span-
ish), PhD Thesis Technical University of Catalonia,
Barcelona, Spain.
Cardona, O.D., (2003) The need for rethinking the concepts
of Vulnerability and Risk from an Hollistic Perspec-
tive: a necessary review a criticism for effective Risk
SIMULTECH2015-5thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
540
Management, Mapping Vulnerability: Disasters, De-
veloping and People Chapter 3 Earthscan Publishers,
London.
Carre
˜
no, M. L., Cardona, O. D., Barbat, A. H., (2007), Dis-
aster risk management performance index, Nat Haz-
ards 40 1-20.
Carre
˜
no, M. L., Cardona, O.D., Barbat, A. H., (2012) New
methodology for urban seismic risk assessment from
a holistic perspective Bull Earthquake Eng, 10, 547-
565.
Mamdani, E. H., Assilian, S., (1975) An experiment in lin-
guistic synthesis with a fuzzy logic controller. Intern.
J. of Man-Machine Studies,7(1), 1-13.
Marulanda, M. C., Cardona, O. D., Barbat, A. H., (2009)
Robustness of the holistic seismic risk evaluation in
urban centers using the USRi, Nat Hazards, 49, 501-
516.
Kumpulainen, K., (2006) Vulnerability Concepts in Hazard
and Risk Assessment Geological Survey of Findand,
42, 65-74.
Saaty, T. L., Vargas, L. G., (1991) Prediction, projection,
and forecasting: applications of the analytical hier-
archy process in economics, finance, politics, games,
and sports, Kluwer Academic Publishers
TacklingNon-linearityinSeismicRiskEstimationusingFuzzyMethods
541