incorporating the lcia concept into fuzzy risk assessment as a tool for environmental impact...
TRANSCRIPT
ORIGINAL PAPER
Incorporating the LCIA concept into fuzzy risk assessmentas a tool for environmental impact assessment
Kevin Fong-Rey Liu • Chih-Yuan Ko •
Chihhao Fan • Cheng-Wu Chen
Published online: 23 August 2012
� Springer-Verlag 2012
Abstract Environmental impact assessment (EIA) is a
procedural tool for environmental management that iden-
tifies, predicts, evaluates and mitigates the environmental
impact of development proposals. In the process of EIA,
EIA reports, prepared by developers, are expected to
delineate the environmental impact, but in practice they
usually determine whether the amounts or concentrations
of pollutants comply with the relevant standards. Actually,
many analytical tools can improve the analysis of envi-
ronmental impact in EIA reports, such as life cycle
assessment (LCA) and environmental risk assessment
(ERA). Life cycle impact assessment (LCIA) is one of
steps in LCA that takes account of the causal relationships
between environmental hazards and damage. Incorporating
the concept of LCIA into an ERA as an integrated tool for
the preparation of EIA reports extends the focus of the
reports from the regulatory compliance of the environ-
mental impact, to determine the significance of the envi-
ronmental impact. Sometimes, when using integrated tools,
it is necessary to consider fuzzy situations, because of a
lack of sufficient information; therefore, so ERA should be
generalized to a fuzzy risk assessment (FRA). Therefore,
this paper proposes the integration of a LCIA and a FRA as
an assessment tool for the preparation of EIA reports,
whereby the LCIA clearly identifies the causal linkage for
hazard–pathway–receptor–damage and then better explain
the significance of the impact; furthermore, a FRA copes
with fuzzy and probabilistic situations in the assessment of
pollution severity and the estimation of exposure proba-
bility. Finally, the use of the proposed methodology is
demonstrated in a case study of the expansion plan for the
world’s largest plastics processing factory.
Keywords Fuzzy risk analysis � Life cycle impact
assessment � Fuzzy logic � Environmental impact
assessment
1 Introduction
Environmental risk assessment (ERA) is a widely used
analytical tool in environmental management. ERA is
founded on the concepts of hazard and risk. An environ-
mental hazard is an object, event or situation with the
potential to cause damage to physical surroundings,
resources, ecosystems, humans, etc. An environmental risk
refers to the severity of the damage and the likelihood that
the damage will actually occur. Five stages have been pro-
posed for a wide-ranging ERA (DEFRA 2011; RSC 2008) as
follows (see Fig. 1). Firstly, problem formulation, some-
times also known as hazard identification, typically involves
the identification of the causal linkage for hazard–pathway–
receptor-damage. Secondly, release assessment determines
the severity of a hazard, based on a consideration of its
K. F.-R. Liu (&) � C. Fan
Department of Safety, Health and Environmental Engineering,
Ming Chi University of Technology, New Taipei City 24301,
Taiwan, ROC
e-mail: [email protected]
C. Fan
e-mail: [email protected]
C.-Y. Ko
SG Development Environmental Consultants Ltd.,
51591 Changhua, Taiwan, ROC
e-mail: [email protected]
C.-W. Chen
Institute of Maritime Information and Technology, National
Kaohsiung Marine University, Kaohsiung 80543, Taiwan, ROC
e-mail: [email protected]
123
Stoch Environ Res Risk Assess (2013) 27:849–866
DOI 10.1007/s00477-012-0621-x
magnitude, spatial extent and temporal duration. Thirdly,
exposure assessment has two components: the probability
that the hazard will occur and the probability or degree of the
receptors being exposed to the hazard. Fourthly, dose–
response assessment considers the probability or degree of
damage that results from exposure to standardized hazards
(hazards that have standard values). The final important step
is risk characterization, which simultaneously evaluates the
significance of a risk by considering the likelihood that the
hazard will occur and the severity of the hazard.
Fuzzy risk assessment (FRA) during an ERA deals with
situations where some assessments are performed using
fuzzy information. For example, an assessment of hazard
severity can be a subjective decision-making process which
is usually modeled by fuzzy logic (Zadeh 1996). The
evaluation of the probability of a receptor becoming
exposed to a hazard or the assessment of the probability of
damage resulting from exposure to a standardized hazard
can involve precise numbers or probability distributions.
However, these numbers and distributions may be assigned
through expertise or experience, if information is insuffi-
cient. Such cases are usually fuzzy and can be converted
into possibility distributions (Zadeh 1978).
Life cycle assessment (LCA) is another widespread ana-
lytical tool for environmental management. The term LCA is
generally reserved for the analytical procedure or method
that includes the compilation and evaluation of the inputs
and outputs and the potential impact of a product or process
throughout its life cycle (ISO 14040 2006). One important
step in LCA is the life cycle impact assessment (LCIA),
which considers the causal relationships between environ-
mental hazards and damage and devises a methodology to
assess the level of damage. Thus, incorporating the LCIA
concept into an ERA (or FRA) can help to identify the causal
linkage for hazard–pathway–receptor–damage, in problem
formulation. It can also aid a better understanding of the
environmental significance, during risk characterization.
Indeed, the combination of a LCIA and an ERA (or FRA) is a
beneficial tool for environmental management.
Environmental impact assessment (EIA) is a procedural
tool which involves the processes of identification, predic-
tion, evaluation and mitigation of the biophysical, social and
other relevant effects of development proposals, before
major decisions and commitments are made (Petts 1999).
Development proposals for which there is a concern of
adverse impact on the environment should prepare EIA
Hazard i
Spatial extent (Ei)Hazard severity (Si)
S=f1(M, E, D) Severitytransformation
(STi)
Frequency of hazard occurrence (Fi)
Magnitude (Mi)
Temporal duration (Di)
Direct or indirecteffect j
Standard value (SVi)
Standard severity (SSV-i)SSV=f1(ST, E, D)
Receptor k
Probability of receptor being exposed to effect (P1-j)
Probability of damage resulting from exposure to effect (P2-jkl)
Risk of damage l (Rjkl)R=f2(ST, F, P1, P2)
Damage l
Level of damage resulting from exposure to hazard
Dose-response assessment
Level of receptor being exposed to hazard
Exposure assessment
Risk characterization
Individual receptor
Hazard
Pathway
Receptor
Damage
Problemformulation
Release assessment
Magnitude
Spatial extent
Temporal duration
A group of receptors
Probability or frequency of hazard occurrence
Risk characterization
Probability of receptors being exposed to hazard
Probability of damage resulting from exposure to hazard
Exposure assessment
Hazard
Pathway
Receptor
Damage
Problem formulation
Release assessment
Magnitude
Spatial extent
Temporal duration
Significance of risk
Risk characterization
Probability or frequency of hazard occurrence
Probability or degree of a receptor being exposed to hazard
Exposure assessment
Dose-response assessment
Probability or degree of damageresulting from exposure to
a standardized hazard
Fig. 1 Framework for environmental risk assessment
850 Stoch Environ Res Risk Assess (2013) 27:849–866
123
reports. These reports should then be forwarded to compe-
tent authorities for review. EIA reports are expected to
evaluate environmental impact, but in practice they usually
only detail the amounts or concentrations of pollutants and
ensure that these comply with the relevant standards. Link-
ing a LCIA to an ERA (or FRA) as an assessment tool for the
preparation of EIA reports extends the focus of the report
from the regulatory compliance of the environmental
impact, to assess type and degree of damage that develop-
ment projects can cause. This is helpful to review commit-
tees or stakeholders, when determining the significance of
the environmental impacts. In summary, this study proposes
an integrated framework of a LCIA and a FRA, to take
account of fuzzy conditions in the framework and to use this
as a new analytical tool to help estimate significance in an
EIA. Finally, the expansion plan for the world’s largest
plastics processing factory is used as a case study in order to
demonstrate the use of the tool.
2 Literature review
2.1 Applying fuzzy methods to environmental risk
The relevant studies for the application of fuzzy methods to
environmental risks fall into three categories. The first uses
fuzzified multiple criteria decision making, such as a fuzzy
analytic hierarchy process, fuzzy synthetic evaluation, or a
fuzzy ranking method, to select the best drilling waste
discharge option (Sadiq et al. 2004), to determine the health
risk associated with disinfection by-products (Sadiq and
Rodriguezb 2004), to estimate the aggregated risk of vari-
ous environmental activities, pollution sources, or routes for
a given process (Sadiq and Husain 2005), to select drilling
fluid for offshore oil and gas operations (Tesfamariam and
Sadiq 2006), to rank contaminated sites (Zhang et al. 2009),
or to rank weighted alternatives in watershed ecological risk
management (Hao and Chen 2010) and to assess the water
quality–quantity-ecosystem (WQQE) for a River basin (Liu
et al. 2011). The second category of study views the eval-
uation of environmental risk as a process of subjective
judgment, so fuzzy rule-based methods are used to assess
the risk to human health from radioactive materials in water
(Shakhawat et al. 2006), the cancerous and non-cancerous
risks associated with disinfection by-products in drinking
water supplies (Sadiq et al. 2007), the risk of groundwater
contamination (Li et al. 2007) and the risk of the accidental
release of eco-toxic substances in hazardous plants (Darbra
et al. 2008). The third category fuzzifies probability meth-
ods to evaluate environmental risk. For example, Chen et al.
(2003) integrated fuzzy and stochastic modeling methods to
assess the environmental risk of contaminated groundwater
systems. Guyonnet et al. (2003) combined Monte Carlo
random sampling of probability distribution functions with
fuzzy calculus to estimate human exposure, via vegetable
consumption, to Cadmium in the surficial soils of an
industrial site in the north of France. Kentel and Aral (2004,
2005) proposed fuzzy Monte Carlo analysis to allow the use
of incomplete information with expert judgment in health
risk assessment. Karimi and Hullermeier (2007) employed
fuzzy set theory to complement probability theory, to
express the fuzzy likelihood of natural hazards. Li et al.
integrated fuzzy and stochastic approaches to assess the risk
of petroleum contamination (Li et al. 2003) and suscepti-
bility to asthma due to air pollution (Li et al. 2008). Fuzzy
process capability indices were proposed by Kaya and
Kahraman (2009), to determine the risk levels associated
with the air pollutants in Istanbul. Rehana and Mujumdar
(2009) developed an imprecise fuzzy waste load allocation
model that simultaneously addressed randomness and
imprecision, for water quality management in a river sys-
tem, subject to the uncertainty that had arisen due to partial
ignorance. Mofarrah and Husain (2011) incorporated a
technique for the probabilistic risk of contamination and
fuzzy set theory in a study to assess the risk to human health
risk from selected heavy metals that were being discharged
into a marine environment, because of petroleum opera-
tions. Qin (2011) proposed a fuzzy parameterized proba-
bilistic analysis method, which integrates environmental
transport modeling, fuzzy transformation, probabilistic risk
assessment and fuzzy risk quantification into a general risk
assessment framework, to assess risks associated with
problems of environmental pollution-control.
2.2 Combining LCA and risk assessment
The similarities and differences between a LCIA and a
HERA were thoroughly discussed by Udo de Haes et al.
(2006) and Bare (2006). In short, they noted that specific
HERA studies are usually restricted to one single substance
in a particular site, whereas specific LCIA studies deal with
hundreds of chemicals at a global level. Benetto et al.
(2007) proposed three methods for the combination of LCA
and ecological risk assessment in mineral waste reuse
scenarios: synthesis of the results of LCA and ERA into the
original categories in the LCA; substitution of the LCA
results with ERA results, within the categories of the LCA,
and a definition of the new impact categories that incor-
porate them. Most recent studies have focused on the
second method. For example, Khan et al. (2002) developed
a risk-based LCA framework for process plant design,
which substituted some original impact categories with
additional human health, ecological and security risks.
Sonnemann et al. (2003) and Bare (2006) replaced the
human toxicity impact category with a risk assessment that
considered the pollution dispersion model and population
Stoch Environ Res Risk Assess (2013) 27:849–866 851
123
density. Socolof and Geibig (2006) replaced the non-car-
cinogenic toxicity and ecological toxicity impact categories
in the LCA with the concept of a hazard quotient (HQ).
Nishioka et al. (2002) employed a LCA to assess the
environmental impact of residential insulation and used
pollution dispersion models and epidemiological statistics
to perform an ERA. Carpenter et al. (2007) used the LCA
software, PaLATE, to assess the risk caused by the use of
recycled materials in roadway construction and the
groundwater contaminant transport software, Hydrus2D, to
predict pollution concentrations, in order to improve
PaLATE.
3 Materials and methods
The integrated framework that combines the LCIA concept
with a FRA encompasses the following steps: (1) the use of
the LCIA concept to identify the causal linkage for hazard–
pathway–receptor–damage; (2) the use of fuzzy logic for
release assessment; (3) the use of severity transformation
(ST) to compare with standard values; (4) the estimation of
the frequency of hazard occurrence; (5) the estimation of
the probability that the receptors will be exposed to mid-
point effects; (6) the evaluation of the probability that a
receptor will be exposed to standardized hazards, (7) the
use of the vertex method to compute the risk of damage and
(8) the use of the distance method to defuzzify the risk.
3.1 Use of the LCIA concept to identify hazard–
pathway–receptor–damage
In practice, EIA reports usually detail the amounts or
concentrations of pollutants and ensure that these comply
with the relevant standards. Their scope should be extended
from regulatory compliance to the evaluation of the envi-
ronmental impact, and further to the interpretation of sig-
nificance. Environmental significance refers to whether the
midpoint and endpoint effects caused by pollutants are
important for stakeholders. Initially, the evaluation of sig-
nificance recognizes the possible midpoint effects and
damage (endpoint effects) caused by a hazard and deter-
mines their importance. Existing LCIA methods provide
this means to identify the cause-effect relationship between
hazards, pathways, receptors and damage. This study pro-
poses a three-step procedure to identify the cause–effect
relationship for hazard–pathway–receptor–damage, as
follows.
1. Identification of hazards. In accordance with the
characteristics of the factories studied, some hazards
(pollutants) are selected, as shown in part A of Fig. 2.
For example, the IPCC has identified six greenhouse
gases that cause climate change: carbon dioxide (CO2),
from fossil fuel combustion, forest clearing, cement
production, etc.; methane landfills (CH4), from the
production and distribution of natural gas and petro-
leum, fermentation from the digestive systems of
livestock, rice cultivation, fossil fuel combustion, etc.;
nitrous oxide (N2O), from fossil fuel combustion,
fertilizers, nylon production, manure, etc.; hydroflu-
orocarbons (HFCs), from refrigeration gases, alumi-
num smelting, semiconductor manufacturing, etc.;
perfluorocarbons (PFCs), from aluminum production,
the semiconductor industry, etc., and sulfur hexafluo-
ride (SF6), from electrical transmissions and distribu-
tion systems, circuit breakers, magnesium production,
etc.
2. Identification of pathways. Prior to determining sig-
nificance, a diagram of the causal relationships
between hazards and receptors that includes all of
the relevant pathways allows stakeholders to under-
stand the midpoint effects of a pollution emission, as
shown in part B of Fig. 2. A hazard may cause several
midpoint effects. For example, N2O can simulta-
neously induce climate change and stratospheric ozone
depletion. However, N2O, CO2, CH4, PFCs or SF6 can
also cause a rise in temperature and, may result in a
rise in sea level and flooding.
3. Identification of receptors and their potential for
damage. The type of damage (endpoint effects)
brought about by a midpoint effect depends upon the
receptors, as shown in parts C and D of Fig. 2. For
example, climate change could possibly lead to human
malnutrition, infectious diseases, heat stress and the
loss of biodiversity in ecosystems, resulting in a
decrease in the production of crops and wood.
3.2 Use of fuzzy logic for release assessment (S)
Release assessment estimates the severity of a particular
hazard (S); this is determined by its magnitude (M), spatial
extent (E) and temporal duration (D). The magnitude of a
hazard refers to the concentration of its pollution source,
usually denoted by ppm, mg/L or mg/m3. The geographical
scale of a hazard often extends considerably beyond the
boundaries of the source of the hazard. Failure to consider
the spatial extent of damage may result in the scope of the
risk assessment being too limited (DEFRA 2011). The
spatial extent is expressed as the radius of an area where
the concentrations of pollutants are either higher than the
standard values, or more than double the average concen-
trations in neighboring counties, if the former condition
does not exist. The temporal scale is also an important
852 Stoch Environ Res Risk Assess (2013) 27:849–866
123
aspect in release assessment, because damage may be so
prolonged that it can be assumed to be permanent and the
environment beyond recovery (DEFRA 2011). The tem-
poral factor is measured by the duration of emission of
pollution over 1 year.
The appraisal of the severity of a hazard can be a sub-
jective decision-making process. This type of appraisal is
eminently suited to the use of fuzzy logic (Zadeh 1996).
Fuzzy logic is a tool that computes with words, when
modeling qualitative human thought processes, in the
analysis of complex systems and decisions. In fuzzy logic,
qualitative perception-based reasoning is represented by
‘IF-THEN’ fuzzy rules. A rule set for evaluation is shown
in Table 1, where ‘magnitude’, ‘spatial extent’, ‘temporal
duration’ and ‘severity’ are linguistic variables (Zadeh
1975) and ‘very low’, ‘low’, ‘medium’, ‘moderate’, ‘high’
and ‘very high’ are their possible fuzzy values, which are
defined by triangular fuzzy sets, as shown in Fig. 3. A
triangular fuzzy set can be expressed as a 3-tuple (l, m, r),
where l, m, r are the locations of the left, middle and right
vertices of the triangle, respectively. For example, ‘low’
magnitude is expressed as (0, 0, 125), in Fig. 3.
To evaluate the severity of a hazard, 19 rule bases,
containing 513 fuzzy rules, were produced. These 19 rule
bases and their corresponding membership functions were
constructed based on expertise. The fuzzy logic systems are
implemented with the MATLAB Fuzzy Logic Toolbox.
Three factual statements (i.e., Fact 1: NOx magnitude is
48.09 ppm; Fact 2: NOx spatial extent is 2.8 km; Fact 3:
NOx temporal duration is 1 year) are fed into this inference
mechanism and subjected to fuzzy logic (Zadeh 1975). The
theory of fuzzy reasoning is detailed in Appendix 1, but it is
easily explained by a graphical representation as shown in
Figs. 4 and 5. The four major steps to reaching a conclusion
using fuzzy logic, in these figures, are described as follows.
Step 1: Computing compatibilities. Compatibility des-
ignates the similarity of an antecedent. It refers to a fact
having the same linguistic variable or the suitability of a
specific rule regarding several facts that corresponds to the
respective antecedents. For Rule 6, the compatibility of
Fact 1 with ‘NOx magnitude is low’ is C6-1 = 0.770
(Fig. 4a); for Fact 2 with ‘NOx spatial extent is medium’,
the compatibility C6-2 is 0.767 (Fig. 4b); for Fact 3 with
‘NOx temporal duration is high’, the compatibility C6-3 is
TSP
Nuclides
VOCs
CO
NOX
SOX
NH3
BOD
PO43-
PFCs
N2O
CO2
CH4
SF6
HFCs
Pesticide
Heavymetals
Noise
Vibration
A. Hazard(Pollutant)
Increased radiative forcing
Climate change (temperature rise)
FloodingSea level rise
Increased chlorine content of stratosphere
Stratosphericozone depletion
Increased UV-B radiation
Direct exposure
Photochemicalozone formation Increased tropospheric ozone
concentration
Smog episodes (urban)
Conversionreleasing protons
Exposure of leaves
Deposition on soil or water
Decreasing pH
Eutrophication of aquatic systems
Increased algal growth Sedimentation
of dead algae
Reduced light input
Oxygen depletion near bottom
Altered species composition
Human direct exposure or intake
Ecosystem direct exposure
ground-, fresh- or marine water
agricultural or natural soil
Fish or animal meat
Vegetation crop
Exuding Al3+
Reducing nutrient
Direct exposure
Loss of habitats
Ionising radiation
Human
Human
Human
Human
Human
Human
Human
Human
Human
Fish
Crops and woods
Wildlife
Wildlife
Ecosystem
Wildlife
Ecosystem
Ecosystem
Ecosystem
Wildlife
Wildlife
Wildlife
Crops and woods
Crops and woods
Crops and woods
Crops and woods
Fish
Fish
Ecosystem
egamaD.DyawhtaP.B(midpoint effect) (endpoint effect)
Disappearance of species
Loss of biodiversity
Loss of fish catch
Loss of productivity of crops and woods
Malnutrition
Infectious diseases
Heat stress
Immunosuppression
Cataract
Cardiovasculardisease
Respiratory diseases
Human toxicity
Cancer
Psychasthenia
Sleep disorders
ST, F P1 P2C. Receptor
Fig. 2 Overview of the causal relationships between hazards, pathways, effects and possible damage (modified from Udo de Haes et al. 1999)
Stoch Environ Res Risk Assess (2013) 27:849–866 853
123
1.000 (Fig. 4c). It should be noted that ‘product’ is chosen
as the t-norm operator, rather than using another more
widely used t-norm operator, ‘min’, because the t-norm
operator, ‘product’, makes the conclusion sensitive to
every input; whereas, only one input controls the conclu-
sion in the case of the t-norm operator, ‘min’. The overall
compatibility of Rule 6 C6 with the three facts is
0.770 9 0.767 9 1.000, so C6 = 0.591 (Fig. 4d). The
compatibilities of other rules are calculated in the same
way.
Step 2: Scaling conclusions. Once the compatibility for
each rule has been calculated, the degree to which the
antecedents have been satisfied for each rule is known. As
shown in Fig. 4d, a new conclusion for Rule 6 is then
inferred by scaling the triangular conclusion, whose height
is C6. The use of the implication operator ‘product’ results
in the scaling of each conclusion.
Step 3: Aggregating scaled conclusions. Inferred con-
clusions with the same linguistic variable must be aggre-
gated. As shown in Fig. 5, the triggered rules are Rules 3,
6, 12 and 15. Aggregation is the process by which the fuzzy
sets representing the scaled conclusions of triggered rules
are combined into a single fuzzy set. In Fig. 5, the final
conclusion is a combination of two triangles, and it is
aggregated as the union of all of the scaled conclusions.
Step 4: Defuzzifying the overall conclusion. In many
cases, the final output of an inference system is a single
number. Defuzzification is a method to justifiably convert a
fuzzy set into a precise value. This study utilized the
center-of-gravity method, which takes the centroid of the
area under the curve of the membership function of a fuzzy
set as the answer. Figure 5 shows that the score of severity
for NOx is 29.5 (S). An analysis of the sensitivity of
operators in fuzzy logic is detailed in Appendix 2.
3.3 Use of ST to compare with standard values (ST)
All outputs of fuzzy logic are linearly transformed so that
their lower bounds (5.23) correspond to 0.0 and the outputs
of standard values (94.8) correspond to 100.0 as shown in
Fig. 6. For example, the standard value of NOx in the
manufacturing processes is 250 ppm, so fuzzy logic infers
a value of 94.8. The result of ST is (29.5 - 5.23)/(94.8 -
5.23) 9 100 = 27.1.
3.4 Estimation of the frequency of hazard occurrence
(F)
The frequency of a hazard occurrence is defined as the
number of occurrences per year, which can be a precise
number, a probability distribution or a possibility distri-
bution. If historical records are insufficient and a precise
frequency or a probability distribution over possible fre-
quencies is not available, the frequencies may be assigned
using expertise or experience. These values are usually
fuzzy and can be converted into possibility distributions
(Zadeh 1978). When the methodology is applied to an EIA,
the frequency is estimated as ‘1’ for a continuous release of
pollutants from a factory.
3.5 Evaluation of the probability of a receptor being
exposed to a midpoint effect (P1)
Further investigation is not required, if no actual or potential
pathway exists between a hazard and the receptor (DEFRA
2011). For example, heavy metal contamination of a soil
poses no risk to humans if there are no residents near the
site. The evaluation of the probability of a receptor being
Table 1 Fuzzy rules for the evaluation of severity
Rule
no.
IF part THEN part
Magnitude Spatial
extent
Temporal
duration
Severity
1 Low Low Low Very low
2 Low Low Medium Very low
3 Low Low High Low
4 Low Medium Low Low
5 Low Medium Medium Low
6 Low Medium High Low
7 Low High Low Low
8 Low High Medium Low
9 Low High High Slightly
low
10 Medium Low Low Very low
11 Medium Low Medium Slightly
low
12 Medium Low High Moderate
13 Medium Medium Low Slightly
low
14 Medium Medium Medium Moderate
15 Medium Medium High Moderate
16 Medium High Low Moderate
17 Medium High Medium Moderate
18 Medium High High High
19 High Low Low Moderate
20 High Low Medium High
21 High Low High High
22 High Medium Low Moderate
23 High Medium Medium High
24 High Medium High Slightly
high
25 High High Low High
26 High High Medium Slightly
high
27 High High High Very high
854 Stoch Environ Res Risk Assess (2013) 27:849–866
123
exposed to a midpoint effect (P1) can yield a precise number
or a probability distribution, if sufficient information is
available; otherwise, it can be assigned using expertise or
experience, which is usually fuzzy and expressed by a
possibility distribution. For example, NOx can increase
tropospheric ozone concentration and the probability of the
receptors being exposed to the effect is subjectively esti-
mated as ‘approximately 0.1’, which is represented as a
triangular fuzzy set of the 3-tuple (0.0, 0.1, 0.2).
3.6 Assessment of the probability of damage resulting
from exposure to a standardized hazard (P2)
The probability of damage (endpoint effect) resulting from
exposure to a standardized hazard (P2) is defined as the
percentage of humans, ecosystems, crops and woods,
wildlife or fish production that sustains damage when
pollution reaches standard values. Even when exposed to
the same midpoint effect, the likelihood of damage is
probabilistic and depends on the likely susceptibility of an
individual receptor to the effect. Assessing P2 is an extre-
mely complicated task, which is riddled with uncertainty,
because the relevant knowledge of toxicology, epidemiol-
ogy and ecology is still incomplete. Therefore, it will
become a precise number or a probability distribution, once
the related knowledge is available; otherwise, it can be
assigned subjectively using expertise or experience as a
fuzzy number. For example, NOx, SOx, VOCs or CO can
increase tropospheric ozone concentration and further
cause human respiratory diseases. Their standard values for
the outlet of an emission pipe are 250, 650, 100 and
2000 ppm, respectively. The probability of respiratory
diseases resulting from exposure to the pollution that has
reached standard values is subjectively assessed as
‘approximately 0.3’, which is expressed as a triangular
fuzzy set of the 3-tuple (0.2, 0.3, 0.4). The P2 value of
‘approximately 0.3’ denotes that about 30 % of human
exposure to increased tropospheric ozone concentration
caused by the standard values of the relevant pollutants
induces respiratory diseases.
3.7 Use of the vertex method to compute the risk
of damage (R)
The risk of damage (R) is a function of four variables, ST,
F, P1 and P2:
R ¼ fðST; F; P1; P2Þ ð1Þ
The vertex method was proposed by Dong and Shah
(1987) to compute functions of fuzzy variables and is
applied herein to compute R in Eq. (1). The vertex method
uses an a-cut and the interval analysis technique. Using a-
cut, each fuzzy variable characterized by a convex
membership function is converted into a group of
intervals with various a values. Intervals with the same avalue are processed by interval analysis, which results in an
interval function with the a value. At the a-cut level, the
interval function is denoted as follows:
Ra ¼ f STa;Fa; Pa1; P
a2
� �ð2Þ
where
Ra ¼ RaL;R
aR
� �; STa ¼ aa
1; ba1
� �; Fa ¼ aa
2; ba2
� �;
Pa1 ¼ aa
3; ba3
� �; Pa
2 ¼ aa4; b
a4
� � ð3Þ
The interval computation is equivalent to solving a
minimization problem for the lower bound and a
maximization problem for the upper bound as follows:
1
(x)
Low hgiHmuideM
ppm
(a) Magnitude
1
(x)
Low hgiHmuideM
km
(b) Spatial extent
1
(x)
Low hgiHmuideM
year
(c) Temporal duration
0 7 7.365321
0 10070605040302010
1
(x)
x
(d) Severity (NOx)
80 90
Verylow Moderate
VeryhighHigh
Slightlyhigh
SlightlylowLow
0 25020015010050
4
0 10.80.60.40.2
Fig. 3 Membership functions of fuzzy values for linguistic variables
a magnitude, b spatial extent, c temporal duration and d severity
Stoch Environ Res Risk Assess (2013) 27:849–866 855
123
RaL ¼ min f2 st; f; p1; p2ð Þ; Ra
R ¼ max f2 st; f; p1; p2ð Þ ð4Þ
such that st [ [a1a, b1
a], f [ [a2a, b2
a], p1 [ [a3a, b3
a],
p2 [ [a4a, b4
a].
In statistics, the notion of risk is often modeled as the
expected value of an undesirable outcome. Therefore, the
risk of damage is defined as:
R ¼ ST� F� P1 � P2 ð5Þ
That is, R is considered as the fuzzy expected value of the
percentage of humans, ecosystems, crops and woods,
wildlife or fish production that sustains damage. Equation
(5) becomes
RaL ¼ aa
1 � aa2 � aa
3 � aa4; Ra
R ¼ ba1 � ba
2 � ba3 � ba
4 ð6Þ
For example, the ST of NOx is 27.1; F is estimated as ‘1’
for a continuous release of NOx; P1 is subjectively estimated
as ‘approximately 0.1 (0.0, 0.1, 0.2)’ and P2 is
‘approximately 0.3 (0.2, 0.3, 0.4)’. This gives RL0? = 27.1 9
1 9 0.0 9 0.2 = 0.000, RR0? = 27.1 9 1 9 0.2 9 0.4 =
2.168, RL1 = RR
1 = 27.1 9 1 9 0.1 9 0.3 = 0.813 and so
on. As shown in Fig. 7, the result for R is not exact, but is very
Step 1: Computing compatibilities Step 2: Scaling conclusions
1Medium
Rule 6:
250
1 Low
(a)12548.09
C6-1= 0.770
7.3
(b)3.652.8
C6-2= 0.767
1High
C6-3= 1.000
(c)0.5 1
0010
1Low
(d)
C6 = 0.770X0.767X1= 0.591
33.3
Scaled conclusion
Fig. 4 Computing compatibilities and scaling conclusions in fuzzy logic. a magnitude (ppm), b spatial extent (km), c temporal duration (year)
and d severity
Step 3: Aggregating scaled conclusions
Rule 3: Rule 6:
Rule 12: Rule 15:
0010
1Low
sC3 = 0.179
33.3 0010
1Low
sC6 = 0.591
33.3
1
0010
Moderate
sC12 = 0.090
33.3 66.7 0010
1
sC15 = 0.295
Moderate
33.3 66.7
29.5
Step 4:Defuzzifyingthe overall conclusion
0 100
1
s
Severity
0.5910.295
ModerateLowCentroid
Fig. 5 Scaling and aggregating conclusions in fuzzy logic
Output of fuzzy logic Severity transformation
Lower bound=5.230
100Standard value = 94.8
Example value = 29.5
100
0
27.1
Fig. 6 Severity transformation
856 Stoch Environ Res Risk Assess (2013) 27:849–866
123
similar to a triangular fuzzy number and can be
approximately represented as a triangular fuzzy set (0,
0.813, 2.168), which gives the fuzzy expected value of the
percentage of humans that suffer respiratory diseases due to
increased tropospheric ozone concentration.
3.8 Use of the distance method to defuzzify risk
The last step is to defuzzify the risk, R, in order to ultimately
obtain a precise number. The center of gravity or distance
methods (Cheng 1998; Chu and Tsao 2002) are widely used
for defuzzification. The two isosceles triangles, A(50, 75,
100) and B(60, 75, 90) shown in Fig. 7 should result in dif-
ferent levels of risk, but they are not distinguishable by either
the center of gravity method or the distance method, because
they have the same centroid, AC(75.000, 0.333), and the
same distance of 75.001 from the centroid to the origin. In
addition, for the purpose of being conservative, the right
wing of the possibility distribution of a risk should be given
more emphasis than the left wing. Therefore, the left wing of
the possibility distribution of risk, R, is reduced to half; that
is, it is multiplied by a weight of 0.5. The weights for the right
and left wings of the possibility distribution of a risk can be
determined by a panel of experts. After scaling down, the left
wing of R becomes a new fuzzy number, R0, with a centroid,
R0C(1.098, 0.295), and the distance, dR0, from the centroid to
the origin, is 1.137, as shown in Fig. 7. The distance, dR0,
indicates the expected value of the percentage of humans that
suffer from respiratory diseases due to increased tropo-
spheric ozone concentration.
4 Case study
4.1 Case description
A plastics factory, established in 1958, covers about
178.9 ha in an industrial zone of Yunlin County, Taiwan. It
is the world’s largest plastics processing factory, generat-
ing plastic products, petrochemical raw materials, elec-
tronic materials, polyester fiber products, etc. In 2009, its
output reached 3.71 million tons and its turnover was up to
US$5.4 billion. In response to market demand, the com-
pany wished to increase the supply of raw materials to 5.03
million tons of products and proposed a US$28.63 billion
expansion plan.
An environmental impact statement (EIS) was submitted
for review, in December 2009. According to the EIS, the
major air pollutants were SOx, NOx, VOCs, CO, TSP and
noise and the primary water pollutants in the treated
wastewater were BOD and PO43-. The emission details are
listed in Table 2. Before the expansion, the emissions of
SOx, NOx, VOCs, CO and TSP were, respectively, 838.6,
886.4, 291.2, 3,047.9 and 272.5 tons/year, which resulted
in concentrations in the emission pipes of 54.35 ppm,
48.09 ppm, 46.48 ppm, 432.31 ppm and 29.59 mg/m3,
respectively. After the expansion, the emissions were pre-
dicted to be 942.5, 1073.2, 416.9, 3047.9 and 341.0 tons,
respectively, and the concentrations in emission pipes were
predicted to be 61.09 ppm, 58.23 ppm, 66.53 ppm,
432.31 ppm and 37.02 mg/m3, respectively. Noise was
forecasted to increase slightly from 65.95 to 66.15 dB,
after the expansion. The treated wastewater was discharged
into the sea at the rate of 187,638 CMD, before the
expansion, but it was forecast to reach 257,638 CMD, after
the expansion. The level of BOD and PO43- were all
maintained within the standards (30 and 4 mg/L). The
relevant magnitudes, spatial extents and temporal durations
are summarized in the third to fifth columns of Table 2.
4.2 Results
The inferred severities of all pollutants (S) are determined
by their magnitude, spatial extent and temporal duration,
using fuzzy logic. These are listed in the second to last
column of Table 2. Using ST, the comparisons of the
inferred severities of all pollutants with the standard values
are shown in the last column of Table 2. Before and after
the expansion, the STs for noise, BOD and PO43- are very
high (denoted by italic values in Table 2), because their
magnitudes are very close to the standard values; the
remainder of the STs are acceptable. After the expansion,
the magnitudes of all pollutions are larger and their spatial
extents are wider, making their STs higher; in particular,
VOCs increase by 29.3 % (from 52.9 to 68.4, denoted by
single-underline in Table 2) and TSP increases by 22.6 %
(from 46.1 to 56.5, denoted by double-underline in
Table 2).
The frequency (F) of occurrence of an environmental
pollutant is defined as the number of occurrences per year
and is ‘1’ for a continuous release. No damage occurs if no
(x)
Risk of damage
0
10075502.1680.813
1 R
(1.098, 0.295)
'RA
B
60 90
µ
Fig. 7 Result for R using the vertex method
Stoch Environ Res Risk Assess (2013) 27:849–866 857
123
receptor is exposed to effects. The probabilities of recep-
tors being exposed to midpoint effects (P1) and the prob-
abilities for all damage resulting from exposure to
standardized hazards (P2) are derived from the real situa-
tion, in Fig. 2, and then assigned by experts, in the form of
possibility distributions rather than probability distribu-
tions, due to the lack of sufficient information, as shown in
Table 3.
The damage (endpoint effects) caused by pollutants
through various midpoint effects is summarized in the first
two columns of Table 4. In this study, the risk of damage
(R) is defined as the product of ST, F, P1 and P2. The vertex
method is thereby used to compute R, when any of its
factors are fuzzy. The output is also a fuzzy number, which
is not exact, but is very similar to a triangle and can be
approximately represented as a triangular fuzzy set of the
3-tuple (l, m, r), as shown in Table 4. R is then defuzzified,
in order to obtain a final precise result. The distance
(d) from the centroid (x, y) of R0 (with a scaled left wing of
R) to the origin is employed as the defuzzification method
in this study and is interpreted as the percentage of humans,
ecosystems, crops and woods, wildlife or fish production
that sustains damage. The fuzzy risks of damage and their
defuzzifications are shown in Table 4.
Before and after the expansion, the loss of fish resulting
from BOD and PO43- is severe, because of the high STs, as
denoted by double-underlined values in Table 4. Never-
theless, the risks resulting from VOCs and TSP show the
greatest increases (29.1 and 22.4 %) after the expansion, as
denoted by the bold numbers in Table 4. However, the
greatest absolute increase in the risk of damage is in the
risk of respiratory diseases caused by TSP, which increases
from 14.675 to 17.980, an increase of 3.306; the second
highest absolute increase of 1.594 represents the loss of
productivity of crops and woods caused by VOCs; the third
highest absolute increase is 1.110 for the disappearance of
species caused by VOCs. These absolute increases are all
indicated by single-underlines in Table 4. It should be
noted that although they are the second and third ranked
absolute increases, they represent increases of as much as
29.1 %.
4.3 Discussion
4.3.1 Comparisons with health risk assessment and a LCIA
In the context of an EIA, risk assessment usually charac-
terizes the nature and magnitude of health risks to humans
and ecological receptors from chemical contaminants and
other stressors, that may be present in the environment (US
EPA 2012). Recently, one study reported an assessment of
the health risk from air pollution for the same case study
(Hsiao 2009). It concluded that the average HQ of SO2,
NO2, CO and VOCs at the neighboring Tai-Si township,
from 2007 to 2008, were 1.52 9 10-2, 4.77 9 10-2,
1.35 9 10-2 and 4.76 9 10-2, respectively; the cancer
risk due to benzene (a carcinogenic VOC) is 3.80 9 10-5.
Compared with the results provided by Hsiao, this study’s
method not only assesses the risks of health effects (car-
diovascular disease, psychasthenia, sleep disorders, respi-
ratory diseases and human toxicity), but also assesses the
risks to the ecosystem (loss of biodiversity and disap-
pearance of species) and natural resources (loss of pro-
ductivity of crops, woods, and fish catch), owing to the
addition of a LCIA framework. On the other hand, another
study (Chiu 2011) used a LCIA method (Eco-indicator 99)
to improve the EIA report for the same case study and
found that the damage due to carcinogens, respiratory
organics, respiratory inorganics, climate change, ozone
layer, ecotoxicity and acidification/eutrophication was
1.03 9 10-1 (daly), 2.79 (daly), 3.26 9 10-3 (daly),
1.42 9 10-4 (daly), 1.83 9 10-2 (daly), 6.76 9 10-6
(paf m2 year) and 1.29 9 10-8 (pdf m2 year), respec-
tively. Compared with the results provided by Chiu, this
study’s method further considers the probabilities of the
occurrence of a hazard, of a receptor being exposed to a
hazard, and of damage resulting from exposure to a stan-
dardized hazard, because of the addition of a FRA.
4.3.2 Calculation of joint risks
Various pollutants may cause the same damage through
different midpoint effects. For example, NOx, VOCs and
CO can cause the formation of photochemical ozone and
further cause respiratory diseases. TSP can also directly
cause respiratory diseases. For example, the risks of
respiratory diseases from the four pollutants, after the
expansion, are 1.255 (3.2 km), 2.787 (8.6 km), 1.207
(1.8 km) and 17.980 (8.3 km), respectively, as shown with
a italic values in Table 4. Their joint risk, under the
assumption of independence, is 22.211 within 1.8 km;
21.267 between 1.8 and 3.2 km; 20.266 between 3.2 and
8.3 km; and 17.980 between 8.3 and 8.6 km, as shown in
Table 5. In Table 5, any risk higher than 10.0 is shown
with a italic values and the risks shown with a bold values
indicate values higher than 20.0. Loss of fish catch is as
much as 40.0 within 2.01 km, both before and after the
expansion, because the area is very close to the point of
discharge into the sea. Respiratory diseases also present a
risk of damage, after the expansion, because the joint risk
exceeds 20.0. Respiratory diseases show the greatest
absolute increase (3.750), which implies an increase in the
expected value for the percentage of humans that are the
subject of respiratory diseases.
858 Stoch Environ Res Risk Assess (2013) 27:849–866
123
4.3.3 Influence of fuzziness in risk assessment
Fuzziness in probabilities occurs because of a lack of a
complete knowledge of the variability in the exposure of
receptors (P1) and their responses (P2) and it can cause an
extra risk of damage. The wider the right and left wings of
the possibility distribution of a risk, the more the fuzziness
and greater the extra risk represent. Precise probabilities in
risk assessment result in lower risks, as shown in the last
column of Table 5. For example, the risk of respiratory
diseases within 1.8 km after the expansion is 22.211; but
this risk is reduced to 20.159, if no fuzziness exists in P1
and P2.
4.3.4 Reduction of the environmental impact
The combination of a LCIA and a FRA as an assessment
tool for the preparation of EIA reports provides more
information and assists review committees or stakeholders
in understanding the significance of the environmental
impact. Appropriate environmental management plans can
be proposed only after the significance of the environ-
mental impact has been ascertained. For example, if there
is goal to reduce the risk of respiratory diseases, after the
expansion, from 3.750 to below 2.000, by cutting down one
of the four associated pollutants (NOx, VOCs, CO and
TSP), this goal is impossible, even if the emission of NOx
or CO is completely removed. However, it is possible if the
risk of respiratory diseases resulting from VOCs can be
reduced from 2.787 (see Table 4) to 0.594, or the risk of
respiratory diseases resulting from TSP can be reduced
from 17.980 (see Table 4) to 16.135. Obviously, the latter
is easier and less expensive. Accordingly, the ST of TSP,
after the expansion, must be reduced from 56.5 (see
Table 2) to 50.7; its severity (S) must be reduced from 55.8
(see Table 2) to 50.6; its magnitude (M) must be reduced
Table 2 Emission details for the case study and the severity evaluation using fuzzy logic
Pollutant Emission Magnitude Spatial extent
(km)
Temporal duration
(year)
S ST (%)
Standard
SOx Outlet of emission pipes;
manufacturing process;
TSP for 10,000–20,000 N m3/min
650 (ppm) 11.3 1.00 94.8 100
NOx 250 (ppm) 7.3 1.00 94.8 100
VOCs 100 (ppm) 10.8 1.00 94.8 100
CO 2,000 (ppm) 4.5 1.00 94.8 100
TSP 73 (mg/m3) 12.2 1.00 94.8 100
Noise Category VI for factory plant 80 (dB) 1.5 1.00 94.8 100
BOD Discharge point 30.00 (mg/L) 2.35 1.00 94.8 100
PO43- 4.00 (mg/L) 2.35 1.00 94.8 100
Before the expansion
SOx 838.6 (ton/year) 54.35 (ppm) 2.8 1.00 22.2 18.9
NOx 886.4 (ton/year) 48.09 (ppm) 2.8 1.00 29.5 27.1
VOCs 291.2 (ton/year) 46.48 (ppm) 7.1 1.00 52.6 52.9
CO 3,047.9 (ton/year) 432.31 (ppm) 1.8 1.00 31.1 28.9
TSP 272.5 (ton/year) 29.59 (mg/m3) 7.3 1.00 46.5 46.1
Noise – 65.95 (dB) 1.406 1.00 79.3 82.7
BOD 187,638 CMD 30.00 (mg/L) 2.01 1.00 89.7 94.3
PO43- 4.00 (mg/L) 2.01 1.00 89.7 94.3
After the expansion
SOx 942.5 (ton/year) 61.09 (ppm) 3.0 1.00 22.9 19.7
NOx 1073.2 (ton/year) 58.23 (ppm) 3.2 1.00 32.2 30.1
VOCs 416.9 (ton/year) 66.53 (ppm) 8.6 1.00 66.5 68.4
CO 3,047.9 (ton/year) 432.31 (ppm) 1.8 1.00 31.1 28.9
TSP 341.0 (ton/year) 37.02 (mg/m3) 8.3 1.00 55.8 56.5
Noise – 66.15 (dB) 1.407 1.00 79.4 82.8
BOD 257,638 CMD 30.00 (mg/L) 2.23 1.00 93.0 98.0
PO43- 4.00 (mg/L) 2.23 1.00 93.0 98.0
Stoch Environ Res Risk Assess (2013) 27:849–866 859
123
from 37.02 mg/m3 (see Table 2) to 32.73 mg/m3; its spa-
tial extent (E) must be reduced from 8.3 km (see Table 2)
to 7.7 km. In summary, the emission of TSP must be
reduced from 341.0 ton/year (see Table 2) to 301.4 ton/
year, which can be accomplished by the installation of
more dust collectors.
4.3.5 Difficulties encountered
Several difficulties were encountered in integrating the con-
cept of a LCIA and a FRA as an assessment tool for the
preparation of EIA reports. Further work is still required to
overcome these difficulties. Firstly, the probabilities of
midpoint effects (e.g. climate change) resulting from envi-
ronmental hazards (e.g. CO2 emission) must be considered.
This type of probability was neglected in this study, because
some are still the subject of debate in the scientific community.
Secondly, gathering sufficient epidemiological studies to
determine the probability of damage resulting from exposure
to standardized hazards proved difficult, so subjective judg-
ment was used to assign the associated probabilities. The third
difficulty arose because of the need to calculate the joint risk of
damage resulting from various midpoint effects and the
aggregation of risks required an assumption of independence.
If these difficulties could be overcome, this model would
prove very beneficial for an EIA.
Table 3 Probabilities of receptors being exposed to midpoint effects (P1) and the probabilities of damage resulting from exposure to stan-
dardized hazards (P2)
Receptor Effect P1 Damage P2
Human Climate change (0.6, 0.7, 0.8) Malnutrition (0.0, 0.1, 0.2)
Infectious diseases (0.2, 0.3, 0.4)
Heat stress (0.4, 0.5, 0.6)
Ozone depletion (0.3, 0.4, 0.5) Cancer (0.1, 0.2, 0.3)
Immunosuppression (0.1, 0.2, 0.3)
Cataract (0.3, 0.4, 0.5)
Ionising radiation (0.1, 0.2, 0.3) Cancer (0.1, 0.2, 0.3)
TSP (direct effect) (0.4, 0.5, 0.6) Cardiovascular disease (0.0, 0.1, 0.2)
Respiratory diseases (0.5, 0.7, 0.9)
Noise and vibration (direct effect) (0.1, 0.2, 0.3) Psychasthenia (0.3, 0.4, 0.5)
Sleep disorders (0.5, 0.6, 0.7)
Photochemical smog (0.1, 0.2, 0.3) Respiratory diseases (0.3, 0.4, 0.5)
Increased tropospheric ozone concentration (0.0, 0.1, 0.2) Respiratory diseases (0.2, 0.3, 0.4)
Acidification (0.1, 0.2, 0.3) Human toxicity (0.1, 0.2, 0.3)
Ecotoxicity (0.1, 0.2, 0.3) Human toxicity (0.5, 0.6, 0.7)
Cancer (0.0, 0.1, 0.2)
Ecosystem Climate change (0.7, 0.8, 0.9) Loss of biodiversity (0.0, 0.1, 0.2)
Ionising radiation (0.2, 0.3, 0.4) Loss of biodiversity (0.5, 0.6, 0.7)
Acidification (0.3, 0.4, 0.5) Loss of biodiversity (0.2, 0.3, 0.4)
Eutrophication (0.4, 0.5, 0.6) Loss of biodiversity (0.2, 0.3, 0.4)
Ecotoxicity (0.1, 0.2, 0.3) Loss of biodiversity (0.5, 0.6, 0.7)
Crops and woods Climate change (0.7, 0.8, 0.9) Loss of productivity of crops and woods (0.2, 0.3, 0.4)
Ozone depletion (0.4, 0.5, 0.6) Loss of productivity of crops and woods (0.3, 0.4, 0.5)
Increased tropospheric ozone concentration (0.2, 0.3, 0.4) Loss of productivity of crops and woods (0.2, 0.3, 0.4)
Acidification (0.2, 0.3, 0.4) Loss of productivity of crops and woods (0.4, 0.5, 0.6)
Wildlife Ozone depletion (0.2, 0.3, 0.4) Disappearance of species (0.0, 0.1, 0.2)
Increased tropospheric ozone concentration (0.2, 0.3, 0.4) Disappearance of species (0.1, 0.2, 0.3)
Acidification (0.1, 0.2, 0.3) Disappearance of species (0.1, 0.2, 0.3)
Eutrophication (0.1, 0.2, 0.3) Disappearance of species (0.3, 0.4, 0.5)
Fish production Ozone depletion (0.2, 0.3, 0.4) Loss of fish catch (0.1, 0.2, 0.3)
Acidification (0.0, 0.1, 0.2) Loss of fish catch (0.3, 0.4, 0.5)
Eutrophication (0.3, 0.4, 0.5) Loss of fish catch (0.5, 0.6, 0.7)
860 Stoch Environ Res Risk Assess (2013) 27:849–866
123
Ta
ble
4F
uzz
yri
sks
of
dam
age
bef
ore
and
afte
rth
eex
pan
sio
nan
dth
eir
def
uzz
ifica
tio
ns
Po
llu
tan
tD
amag
eB
efo
reth
eex
pan
sio
nA
fter
the
exp
ansi
on
Incr
ease
ind
Fu
zzy
risk
RC
entr
oid
of
R0
Fu
zzy
risk
RC
entr
oid
of
R0
lm
rx
yd
lm
rx
yd
SO
xH
um
anto
xic
ity
0.1
89
0.7
58
1.7
05
0.9
57
0.2
95
1.0
01
0.1
97
0.7
89
1.7
75
0.9
96
0.2
95
1.0
39
0.0
38
3.9
%
Lo
sso
fb
iod
iver
sity
1.1
37
2.2
74
3.7
89
2.5
38
0.2
88
2.5
54
1.1
84
2.3
67
3.9
46
2.6
42
0.2
88
2.6
58
0.1
04
Dis
app
eara
nce
of
spec
ies
0.1
89
0.7
58
1.7
05
0.9
57
0.2
95
1.0
01
0.1
97
0.7
89
1.7
75
0.9
96
0.2
95
1.0
39
0.0
38
Lo
sso
fp
rod
uct
ivit
yo
fcr
op
san
dw
oo
ds
1.5
16
2.8
42
4.5
47
3.1
27
0.2
87
3.1
40
1.5
78
2.9
59
4.7
35
3.2
56
0.2
87
3.2
69
0.1
28
Lo
sso
ffi
shca
tch
0.0
00
0.7
58
1.8
95
0.9
79
0.2
92
1.0
21
0.0
00
0.7
89
1.9
73
1.0
19
0.2
92
1.0
60
0.0
39
NO
xR
esp
irat
ory
dis
ease
s0
.00
00
.81
32
.16
81
.09
80
.29
51
.13
70
.00
00
.90
32
.40
91
.22
00
.29
51
.25
50
.11
81
0.8
%
Hu
man
tox
icit
y0
.27
11
.08
42
.43
91
.36
90
.29
51
.40
00
.30
11
.20
42
.71
01
.52
10
.29
51
.54
90
.14
9
Lo
sso
fb
iod
iver
sity
0.0
00
2.1
68
4.8
77
2.6
06
0.2
86
2.6
22
0.0
00
2.4
09
5.4
20
2.8
96
0.2
86
2.9
10
0.2
88
Dis
app
eara
nce
of
spec
ies
0.5
42
1.6
26
3.2
52
1.9
42
0.2
92
1.9
64
0.6
02
1.8
07
3.6
13
2.1
58
0.2
92
2.1
78
0.2
14
Lo
sso
fp
rod
uct
ivit
yo
fcr
op
san
dw
oo
ds
2.1
68
4.0
64
6.5
03
4.4
73
0.2
87
4.4
82
2.4
09
4.5
17
7.2
27
4.9
70
0.2
87
4.9
79
0.4
97
VO
Cs
Res
pir
ato
ryd
isea
ses
0.0
00
1.5
87
4.2
31
2.1
43
0.2
95
2.1
63
0.0
00
2.0
52
5.4
72
2.7
71
0.2
95
2.7
87
0.6
24
29
.1%
Dis
app
eara
nce
of
spec
ies
1.0
58
3.1
73
6.3
46
3.7
90
0.2
92
3.8
01
1.3
68
4.1
04
8.2
09
4.9
02
0.2
92
4.9
11
1.1
10
Lo
sso
fp
rod
uct
ivit
yo
fcr
op
san
dw
oo
ds
2.1
15
4.7
60
8.4
62
5.4
37
0.2
89
5.4
45
2.7
36
6.1
56
10
.94
57
.03
20
.28
97
.03
81
.59
4
CO
Res
pir
ato
ryd
isea
ses
0.0
00
0.8
66
2.3
11
1.1
70
0.2
95
1.2
07
0.0
00
0.8
66
2.3
11
1.1
70
0.2
95
1.2
07
0.0
00
0.0
%
Dis
app
eara
nce
of
spec
ies
0.5
78
1.7
33
3.4
66
2.0
70
0.2
92
2.0
90
0.5
78
1.7
33
3.4
66
2.0
70
0.2
92
2.0
90
0.0
00
Lo
sso
fp
rod
uct
ivit
yo
fcr
op
san
dw
oo
ds
1.1
55
2.5
99
4.6
21
2.9
69
0.2
89
2.9
83
1.1
55
2.5
99
4.6
21
2.9
69
0.2
89
2.9
83
0.0
00
TS
PC
ard
iov
ascu
lar
dis
ease
0.0
00
2.3
04
5.5
29
2.8
94
0.2
89
2.9
08
0.0
00
2.8
23
6.7
75
3.5
46
0.2
89
3.5
58
0.6
49
22
.4%
Res
pir
ato
ryd
isea
ses
9.2
15
13
.82
31
9.3
52
14
.67
20
.28
41
4.6
75
11
.29
21
6.9
38
23
.71
31
7.9
78
0.2
84
17
.98
03
.30
6
No
ise
Psy
chas
then
ia2
.48
16
.61
61
2.4
04
7.6
75
0.2
89
7.6
80
2.4
84
6.6
25
12
.42
17
.68
50
.28
97
.69
00
.01
00
.1%
Sle
epd
iso
rder
s4
.13
59
.92
31
7.3
66
11
.16
90
.28
71
1.1
73
4.1
40
9.9
37
17
.38
91
1.1
84
0.2
87
11
.18
80
.01
5
BO
DL
oss
of
bio
div
ersi
ty0
.00
07
.54
41
6.9
75
9.0
71
0.2
86
9.0
76
0.0
00
7.8
39
17
.63
89
.42
60
.28
69
.43
00
.35
43
.9%
Dis
app
eara
nce
of
spec
ies
2.8
29
7.5
44
14
.14
68
.75
20
.28
98
.75
72
.94
07
.83
91
4.6
99
9.0
94
0.2
89
9.0
99
0.3
42
Lo
sso
ffi
shca
tch
14
.14
62
2.6
33
33
.00
72
4.2
66
0.2
85
24
.26
81
4.6
99
23
.51
83
4.2
97
25
.21
40
.28
52
5.2
16
0.9
48
PO
43-
Lo
sso
fb
iod
iver
sity
0.0
00
7.5
44
16
.97
59
.07
10
.28
69
.07
60
.00
07
.83
91
7.6
38
9.4
26
0.2
86
9.4
30
0.3
54
3.9
%
Dis
app
eara
nce
of
spec
ies
2.8
29
7.5
44
14
.14
68
.75
20
.28
98
.75
72
.94
07
.83
91
4.6
99
9.0
94
0.2
89
9.0
99
0.3
42
Lo
sso
ffi
shca
tch
14
.14
62
2.6
33
33
.00
72
4.2
66
0.2
85
24
.26
81
4.6
99
23
.51
83
4.2
97
25
.21
40
.28
52
5.2
16
0.9
48
Stoch Environ Res Risk Assess (2013) 27:849–866 861
123
5 Conclusions
This study proposes the integration of a LCIA and a FRA
by using a LCIA to identify the causal linkage for hazard–
pathway–receptor–damage, using fuzzy logic for release
assessment, using ST for comparison with standard values,
estimating the frequency of hazard occurrence, estimating
the probability that the receptors will be exposed to mid-
point effects, evaluating the probability of receptors being
exposed to standardized hazards, using the vertex method
to compute risk of damage and using distance method to
defuzzify the risk. The tool can be used to assess EIA
reports, because a LCIA can clearly identify the causal
linkage for hazard–pathway–receptor–damage and then
better explain the significance of the impact; furthermore,
FRA copes with fuzzy and probabilistic situations in the
assessment of pollution severity and the estimation of the
probability of exposure.
The integrated tool was demonstrated with a practical
case study. The release assessment shows that the STs of
BOD and PO43- are very high before and after expansion,
due to their high concentrations. However, the greatest
relative increases in ST are VOCs (by 29.3 %) and TSP (by
22.6 %). The risk characterization shows that the high STs
for BOD and PO43- also lead to a severe reduction on the
fish catch of more than 20.0. Meanwhile, the risks resulting
from VOCs and TSP have the greatest relative increases
(29.1 and 22.4 %) after expansion. However, the greatest
absolute increase in risks is 3.306 for the respiratory dis-
eases caused by TSP. The joint risk of respiratory diseases
exceeds 20.0 and has the greatest absolute increase (3.750)
which implies an increase in the expected value of the
percentage of humans that are the subject of respiratory
diseases. Assuming that the risk of respiratory diseases
resulting from TSP can be reduced from 17.980 to 16.135,
which means that the concentration of TSP must be reduced
from 37.02 to 32.73 mg/m3 and that its emission must be
reduced from 341.0 to 301.4 tons, by installing more dust
collectors, the increase of 3.750 will be reduced to 2.000.
Uncertainty should be considered not only for the RA
part of the integration, but also for the LCIA. Uncertainty
in the LCA may be due to data, choices and relationships
(Finnveden et al. 2009). Data can show variability, be
badly-specified, erroneous, incomplete or imprecise and
Table 5 Joint risks of damage due to various pollutants before and after the expansion
Damage Before the expansion After the expansion
Extent
(km)
Joint risk
(fuzzy)
Extent
(km)
Joint risk
(fuzzy)
Increase
(fuzzy)
Joint risk
(crisp)
Cardiovascular disease 7.3 2.908 8.3 3.558 0.649 2.837
Psychasthenia 1.406 7.680 1.407 7.690 0.010 6.630
Sleep disorders 1.406 11.173 1.407 11.188 0.015 9.941
Respiratory diseases 1.8 18.461 1.8 22.211 3.750 20.159
2.8 17.469 3.2 21.267 19.429
7.1 16.520 8.3 20.266 18.660
7.3 14.675 8.6 17.980 16.940
Human toxicity 2.8 2.388 3.0 2.572 0.185 2.062
3.2 1.549 1.236
Loss of biodiversity 2.8 5.109 3.0 5.491 0.382 4.751
3.2 2.910 2.425
(In the sea) 2.01 17.328 2.23 17.971 0.643 15.073
Disappearance of species 1.8 8.585 1.8 9.870 1.285 8.291
2.8 6.635 3.0 7.948 6.654
7.1 3.801 3.2 6.982 5.866
8.6 4.911 4.114
(In the sea) 2.01 16.747 2.23 17.370 0.622 15.073
Loss of productivity of crops and woods 1.8 15.122 1.8 17.093 1.971 15.337
2.8 12.519 3.0 14.554 13.072
7.1 5.445 3.2 11.667 10.409
8.6 7.038 6.163
Loss of fish catch 2.01 43.232 2.23 44.666 1.434 41.996
(In the sea) 2.8 1.021 3.0 1.060 0.837
862 Stoch Environ Res Risk Assess (2013) 27:849–866
123
(d)(c)
(a) (b)
(h)(g)
(e) (f)
Fig. 8 Sensitivity analysis of the fuzzy logic system with different
operators. a Our selection (E&M vs. S), b our selection (D&M vs. S),
c And operator: min, d Implication operator: min, e Defuzzification:
Bisector, f Defuzzification: Mom, lom, som, g Membership function:
trapezoid, h Membership function: Gaussian
Stoch Environ Res Risk Assess (2013) 27:849–866 863
123
choices may be inconsistent in system boundaries, allocation
principles, and time horizon; relationships may be wrongly
assumed to have a linear dependence between outcomes and
inputs. The types of uncertainty in LCA can be approxi-
mately categorized as variable and fuzzy. Variability results
from spatial or temporal differences, or stochastic nature and
fuzziness occurs because of linguistic ambiguity or impre-
cise measurement. For example, Ardente et al. (2004) used
fuzzy methods to deal with inexact and incomplete input
data and again to evaluate the quality of life cycle inventory.
The authors are working on the fuzziness of a LCIA; how-
ever, because of limitations of space, this important topic
will be the subject of a future paper.
Acknowledgments The authors would like to thank the National
Science Council of the Republic of China (Taiwan) for financially
supporting this research under Contract NSC 99-2221-E-131-010-
MY2. The author also appreciates the editorial assistance provided by
Dr. Michael McGarrigle.
Appendix 1: Fuzzy Logic
Fuzzy logic (Zadeh 1996) has the ability to compute with
words, to model qualitative human thought processes in the
analysis of complex systems and decisions. Fuzzy logic
represents qualitative perception-based reasoning by ‘IF-
THEN’ fuzzy rules. An example of fuzzy logic, in which a
new fuzzy value is derived on the basis of a fuzzy rule (i.e.,
the ith rule in a fuzzy-rule base) with three antecedents and
three fuzzy facts, is represented as follows:
If X1 is Fi1 AND X2 is Fi2 AND X3 is Fi3 THEN Y is Gi
X1 is F01 AND X2 is F02 AND X3 is F02 AND X3 is F03Y is G01
ð7Þ
where Xj and Y are linguistic variables, Fij and F0j are fuzzy
sets of Uj and Gij and G0j are fuzzy sets of V. In the
framework of the compositional rule of inference (Zadeh
1975), G0j is computed by
G0i ¼ F01 ^ F02 ^ F03� �
� Fi1 ^ Fi2 ^ Fi3ð Þ ! Gið Þ ð8Þ
where ^ denotes a t-norm operator, � is a composition
operator and ? indicates an implication operator.
The selection of operators is important for the calcula-
tion of G0. If ‘sup-min’ is chosen as the composition
operator (Zadeh 1975), the membership function of G0 is
computed by:
lG0iðvÞ ¼ max
u1;u2;u3min lF01^F02^F03
u1; u2; u3ð Þ; lFi1^Fi2^Fi3!G
h
u1; u2; u3; vð Þi
ð9Þ
Furthermore, if ‘min’ is the t-norm operator (i.e.,
a ^ b = min (a, b)) and the Mamdani’s implication
operator (i.e., a ? b = min(a, b)), Eq. 9 becomes the
well-known ‘Mamdani’s fuzzy reasoning’, which can be
expressed as
lG0iðvÞ ¼ max
u1;u2;u3min lF0
1ðu1Þ; lF0
2ðu2Þ; lF0
3ðu3Þ; lFi1
ðu1Þ;h
lFi2ðu2Þ; lFi3
ðu3Þ; lGðvÞi
ð10Þ
Equation 10 can be further depicted in another form:
lG0iðvÞ ¼ min max
u1lF0
1^Fi1ðu1Þ;max
u2lF0
2^Fi2ðu2Þ;
�
maxu3
lF03^Fi3ðu3Þ; lGðvÞ
� ð11Þ
where F0j ^ Fij denotes the intersection of fuzzy sets F0j and
Fij; maxuj
lF0j^FijðujÞ is the highest degree of membership of
the intersection and can be interpreted as the compatibility
Cij between F0j and Fij; min maxu1
lF01^Fi1
�ðu1Þ;max
u2lF0
2^Fi2
ðu2Þ;maxu3
lF03^Fi3ðu3Þ� can be viewed as the overall
compatibility Ci between the facts and the rule; and Ci is
used to truncate Gi to obtain G0i. Moreover, if F0j is a precise
value (i.e., say uj), Eq. (11) becomes:
lG0iðvÞ ¼ min lFi1
u1ð Þ; lFi2u2ð Þ; lFi3
u3ð Þ; lGðvÞ� �
ð12Þ
where min lFi1u1ð Þ; lFi2
u2ð Þ; lFi3u3ð Þ
� �can be viewed as the
overall compatibility Ci between the facts and the rule; Ci
is used to truncate Gi to obtain G0i.
Appendix 2: Analysis of the sensitivity of operators
in fuzzy logic
The sensitivity analysis of this study’ fuzzy logic system
with different operators is expressed by three-dimensional
surfaces, which represent the dependency of the output
(severity) on any two of the three inputs (magnitude, spatial
extent and temporal duration), as shown in Fig. 8. When
any horizontal plane exists it implies that both of the inputs
are not sensitive to the output; in other words, any change in
the inputs within the plane does not alter the output. The
selection of operators (‘product’ for the ‘and operator’ and
the ‘implication operator’; ‘centroid’ for the ‘defuzzifica-
tion operator’) is acceptable in the sensitivity analysis, as
shown in Fig. 8a, b. Even if either the ‘and operator’ or the
‘implication operator’ uses ‘min’, the sensitivity analysis is
still acceptable, as shown in Fig. 8c, d. However, it is
unacceptable, due to the existence of horizontal planes, if
the ‘defuzzification operator’ uses other settings (‘bisector’,
‘mom’, ‘lom’, or ‘som’), or if the membership functions are
changed from ‘triangular’ into ‘trapezoidal’ or ‘Gaussian’.
864 Stoch Environ Res Risk Assess (2013) 27:849–866
123
References
Ardente F, Beccali M, Cellura M (2004) F.A.L.C.A.D.E.: a fuzzy
software for the energy and environmental balances of products.
Ecol Model 176(3–4):359–379
Bare JC (2006) Risk assessment and life-cycle impact assessment
(LCIA) for human health cancerous and noncancerous emis-
sions: integrated and complementary with consistency within the
USEPA. Hum Ecol Risk Assess 12:493–509
Benetto E, Tiruta-Barna L, Perrodin Y (2007) Combining lifecycle and
risk assessments of mineral waste reuse scenarios for decision
making support. Environ Impact Assess Rev 27(3):266–285
Carpenter AC, Gardner KH, Fopiano J, Benson CH, Edil TB (2007)
Life cycle based risk assessment of recycled materials in roadway
construction. Waste Manage (Oxford) 27(10):1458–1464
Chen Z, Huang GH, Chakma A (2003) Hybrid fuzzy-stochastic
modeling approach for assessing environmental risks at contam-
inated groundwater systems. J Environ Eng 129(1):79–88
Cheng CH (1998) A new approach for ranking fuzzy numbers by
distance method. Fuzzy Sets Syst 95:307–317
Chiu SY (2011) Life cycle impact assessment as a tool in
environmental impact assessment and its uncertainty analysis.
Master thesis, Ming Chi University of Technology, Taiwan
Chu TC, Tsao CT (2002) Ranking fuzzy numbers with an area
between the centroid point and original point. Comput Math
Appl 43:111–117
Darbra RM, Demichela M, Mure S (2008) Preliminary risk assessment
of ecotoxic substances accidental releases in major risk installa-
tions through fuzzy logic. Process Saf Environ Prot 86(2):103–111
DEFRA (2011) Guidelines for environmental risk assessment and
management. http://www.defra.gov.uk/publications/files/pb136
70-green-leaves-iii-1111071.pdf
Dong W, Shah HC (1987) Vertex method for computing functions of
fuzzy variables. Fuzzy Sets Syst 24(1):65–78
Finnveden G, Hauschild MZ, Ekvall T, Guinee J, Heijungs R,
Hellweg S, Koehler A, Pennington D, Suh S (2009) Recent
developments in life cycle assessment. J Environ Manage
91(1):1–21
Guyonnet D, Bourgine B, Dubois D, Fargier H, Come B, Chiles JP
(2003) Hybrid approach for addressing uncertainty in risk
assessments. J Environ Eng 129(1):68–78
Hao F, Chen G (2010) A fuzzy multi-criteria group decision-making
model based on weighted borda scoring method for watershed
ecological risk management: a case study of three gorges
reservoir area of China. Water Resour Manage 24(10):2139–2165
Hsiao HY (2009) Air pollution and health risks of communities
surrounding the No. 6 Naphtha Cracking Complex. Master
thesis, National Taiwan University, Taiwan
ISO 14040 (2006) Environmental management—life cycle assess-
ment—principles and framework (ISO 14040:2006). Interna-
tional Standard. ISO, Geneva
Karimi I, Hullermeier E (2007) Risk assessment system of natural
hazards: a new approach based on fuzzy probability. Fuzzy Sets
Syst 158(9):987–999
Kaya I, Kahraman C (2009) Fuzzy robust process capability indices
for risk assessment of air pollution. Stoch Environ Res Risk
Assess 23(4):529–541
Kentel E, Aral MM (2004) Probabilistic-fuzzy health risk modeling.
Stoch Environ Res Risk Assess 18(5):324–338
Kentel E, Aral MM (2005) 2D Monte Carlo versus 2D Fuzzy Monte
Carlo health risk assessment. Stoch Environ Res Risk Assess
19(1):86–96
Khan FI, Sadiq R, Husain T (2002) GreenPro-I: a risk-based life cycle
assessment and decision-making methodology for process plant
design. Environ Model Softw 17(8):669–692
Li JB, Chakma A, Zeng GM, Liu L (2003) Integrated fuzzy-stochastic
modeling of petroleum contamination in subsurface. Energy
Sources 25(6):547–563
Li J, Huang GH, Zeng G, Maqsood I, Huang Y (2007) An integrated
fuzzy-stochastic modeling approach for risk assessment of
groundwater contamination. J Environ Manage 82(2):173–188
Li HL, Huang GH, Zou Y (2008) An integrated fuzzy-stochastic
modeling approach for assessing health-impact risk from air
pollution. Stoch Environ Res Risk Assess 22(6):789–803
Liu J, Chen Q, Li Y, Yang Z (2011) Fuzzy synthetic model for risk
assessment on Haihe River basin. Ecotoxicology 20(5):1131–1140
Mofarrah A, Husain T (2011) Fuzzy based health risk assessment of
heavy metals introduced into the marine environment. Water
Qual Expo Health 3(1):25–36
Nishioka Y, Levy JI, Norris GA, Wilson A, Hofstetter P, Spengler JD
(2002) Integrating risk assessment and life cycle assessment: a
case study of insulation. Risk Anal 22(5):1003–1017
Petts J (1999) Handbook of environmental impact assessment volume
2: environmental impact assessment in practice: impact and
limitations. Blackwell Science Ltd., Oxford
Qin XS (2011) Assessing environmental risks through fuzzy param-
eterized probabilistic analysis. Stoch Environ Res Risk Assess
26:43–58
Rehana S, Mujumdar PP (2009) An imprecise fuzzy risk approach for
water quality management of a river system. J Environ Manage
90(11):3653–3664
RSC, Environment, Health and Safety Committee of Royal Society of
Chemistry (2008) Note on environmental risk assessment.
http://www.rsc.org/ScienceAndTechnology/Policy/EHSC/
EnvironmentalRiskAssessment.asp
Sadiq R, Husain T (2005) A fuzzy-based methodology for an
aggregative environmental risk assessment: a case study of
drilling waste. Environ Model Softw 20(1):33–46
Sadiq R, Rodriguezb MJ (2004) Fuzzy synthetic evaluation of
disinfection by-products-a risk-based indexing system. J Environ
Manage 73(1):1–13
Sadiq R, Husain T, Veitch B, Bose N (2004) Risk-based decision-
making for drilling waste discharges using a fuzzy synthetic
evaluation technique. Ocean Eng 31(16):1929–1953
Sadiq R, Rodriguez MJ, Imran SA, Najjaran H (2007) Communicat-
ing human health risks associated with disinfection by-products
in drinking water supplies: a fuzzy-based approach. Stoch
Environ Res Risk Assess 21(4):341–353
Shakhawat C, Tahir H, Neil B (2006) Fuzzy rule-based modelling for
human health risk from naturally occurring radioactive materialsin produced water. J Environ Radioact 89(1):1–17
Socolof ML, Geibig JR (2006) Evaluating human and ecological
impacts of a product life cycle: the complementary roles of life-
cycle assessment and risk assessment. Hum Ecol Risk Assess
12:510–527
Sonnemann G, Castells F, Schuhmacher M (2003) Integrated life-
cycle and risk assessment for industrial processes. CRC Press
LLC, Boca Raton
Tesfamariam S, Sadiq R (2006) Risk-based environmental decision-
making using fuzzy analytic hierarchy process (F-AHP). Stoch
Environ Res Risk Assess 21(1):35–50
Udo de Haes HA, Jolliet O, Finnveden G, Hauschild M, Krewitt W,
Muller-Wenk R (1999) Best available practice regarding impact
categories and category indicators in life cycle impact assess-
ment. Int J Life Cycle Assess 4(3):167–174
Udo de Haes HA, Sleeswijk AW, Heijungs R (2006) Similarities,
differences and synergisms between HERA and LCA—an
analysis at three levels. Hum Ecol Risk Assess 12:431–449
US EPA (United States Environmental Protection Agency) (2012)
Website information. http://epa.gov/riskassessment
Stoch Environ Res Risk Assess (2013) 27:849–866 865
123
Zadeh LA (1975) The concept of a linguistic variable and its
application to approximate reasoning. Inf Sci 8(3):199–249
Zadeh LA (1978) Fuzzy sets as the basis for a theory of possibility.
Fuzzy Sets Syst 1(1):3–28
Zadeh LA (1996) Fuzzy logic computing with words. IEEE Trans
Fuzzy Syst 4(1):103–111
Zhang K, Kluck C, Achari G (2009) A comparative approach for
ranking contaminated sites based on the risk assessment
paradigm using fuzzy PROMETHEE. Environ Manage 44(5):
952–967
866 Stoch Environ Res Risk Assess (2013) 27:849–866
123