ARTICLE IN PRESS

Journal of Environmental Management 82 (2007) 173–188

www.elsevier.com/locate/jenvman

An integrated fuzzy-stochastic modeling approach for risk assessment of

groundwater contamination
Jianbing Lia,, Gordon H. Huangb,c, Guangming Zengc, Imran Maqsoodd, Yuefei Huange
a
Environmental Engineering Program, University of Northern British Columbia, Prince George, British Columbia, Canada V2N 4Z9
b
Faculty of Engineering, University of Regina, Regina, Saskatchewan, Canada S4S 0A2
c
China-Canada Center of Energy and Environment Research, Hunan University, Changsha, 410082, China
d
Saskatchewan Environment, Regina, Saskatchewan, Canada S4S 0B1
e
Department of Hydraulic Engineering, Tsinghua University, Beijing 100084, China
Received 13 April 2005; received in revised form 14 December 2005; accepted 14 December 2005
Available online 30 March 2006

Abstract

An integrated fuzzy-stochastic risk assessment (IFSRA) approach was developed in this study to systematically quantify both
probabilistic and fuzzy uncertainties associated with site conditions, environmental guidelines, and health impact criteria. The
contaminant concentrations in groundwater predicted from a numerical model were associated with probabilistic uncertainties due to the
randomness in modeling input parameters, while the consequences of contaminant concentrations violating relevant environmental
quality guidelines and health evaluation criteria were linked with fuzzy uncertainties. The contaminant of interest in this study was
xylene. The environmental quality guideline was divided into three different strictness categories: ‘‘loose’’, ‘‘medium’’ and ‘‘strict’’. The
environmental-guideline-based risk (ER) and health risk (HR) due to xylene ingestion were systematically examined to obtain the general
risk levels through a fuzzy rule base. The ER and HR risk levels were divided into five categories of ‘‘low’’, ‘‘low-to-medium’’,
‘‘medium’’, ‘‘medium-to-high’’ and ‘‘high’’, respectively. The general risk levels included six categories ranging from ‘‘low’’ to ‘‘very
high’’. The fuzzy membership functions of the related fuzzy events and the fuzzy rule base were established based on a questionnaire
survey. Thus the IFSRA integrated fuzzy logic, expert involvement, and stochastic simulation within a general framework. The
robustness of the modeling processes was enhanced through the effective reflection of the two types of uncertainties as compared with the
conventional risk assessment approaches. The developed IFSRA was applied to a petroleum-contaminated groundwater system in
western Canada. Three scenarios with different environmental quality guidelines were analyzed, and reasonable results were obtained.
The risk assessment approach developed in this study offers a unique tool for systematically quantifying various uncertainties in
contaminated site management, and it also provides more realistic support for remediation-related decisions.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Contaminated site; Fuzzy set; Risk assessment; Simulation; Stochastic modeling; Uncertainty

1. Introduction ranking the severity of site contamination (CCME, 1996).

There are thousands of contaminated industrial sites that
pose significant threats to human health and the natural
environments in Canada. Risk assessment is a critical
procedure in the course of related decision-making regard-
ing effective remediation and management of these
polluted sites. It offers sound bases for assessing and
Corresponding author. Tel.: +1 250 9606397; fax: +1 250 9605845.
E-mail address: li@unbc.ca (J. Li).
However, the insight about risks is limited by the
randomness inherent in nature and the lack of sufficient
information related to the chances of risk occurrence
and the potential consequences of such occurrence. As a
result, risk assessment is inherently linked with uncertainty
(Wagner et al., 1992; Carrington and Bolger, 1998).
Negligence of uncertainty in the assessment procedures
may result in adverse consequences. For instance, over-
design of remediation systems may lead to waste of money
and resources; while underestimation of risks may result in

0301-4797/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jenvman.2005.12.018
 ARTICLE IN PRESS
174 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188
no actions or limited actions towards the management of
sites, and in fact will severely threaten human health and
the natural environments.
Previously, a large volume of literature has been
published on methods for implementing risk assessment
at contaminated sites under various source and/or aquifer
conditions. For example, Lee et al. (1994) proposed a
fuzzy-set-based approach to estimate human-health risk
from groundwater contamination and evaluate possible
regulatory actions; Goodrich and McCord (1995) applied
Monte Carlo methods to account for parameter uncertain-
ties in groundwater flow and solute transport processes,
and the modeling outputs were then used for exposure
assessment; Mills et al. (1996) developed a risk-based
method to evaluate the allowable levels of soil gasoline-
containing residues in order to protect human health;
Hamed and Bedient (1997) applied the first- and second-
order reliability methods in a risk assessment framework to
account for uncertainty; Batchelor et al. (1998) developed a
stochastic risk assessment model for a site by representing
relevant parameters as probability distribution functions;
Bennett et al. (1998) developed an integrated modeling
system for risk assessment by using Monte-Carlo-based
contaminant transport simulation results; Maxwell et al.
(1998, 1999) also developed an integrated system of linked
groundwater transport modeling and human exposure
assessment; Lee et al. (2002) applied Monte Carlo
simulation to estimate health risk in a groundwater-
contaminated community after on-site remediation. More
recently, a hybrid method has been proposed to combine
probabilistic and fuzzy-set approaches to represent model-
ing parameter uncertainties involved in the risk assessment
process (Li et al., 2003; Liu et al., 2004). For instance,
Guyonnet et al. (2003) integrated the Monte Carlo random
sampling of probability distribution functions with
fuzzy calculus to represent different uncertainties for
estimating the human exposure to soil cadmium at an
industrial site; Kentel and Aral (2004) conducted a health
risk analysis of multi-pathway exposure to contaminated
water by generating the fuzzy membership functions of
risks and the probability distributions of risks for various
alpha-cut levels of the membership function, where
the pollutant concentration and cancer potency factors
were treated as fuzzy variables while the remaining
modeling parameters were treated using probability
density functions (PDFs). Some other related studies
can be found in Chen et al. (2003) and Kentel and Aral
(2005).
It was found from the literature review that stochastic
and fuzzy-set techniques were commonly used to accom-
modate uncertainties associated with risk modeling inputs
and outputs. However, most of the previous works have
only dealt with parameter uncertainties in modeling of
contaminant transport, while the uncertainties in environ-
mental quality guidelines and health risk evaluation criteria
have received less attention (Minsker and Shoemaker,
1998; Chen et al., 2003). In addition, few previous works
were reported to effectively link different types of
uncertainties in a risk assessment framework. In fact, such
negligence may result in missed information and thus
impractical decision support. It is imperative that efforts
be made in developing advanced methodologies that
can effectively handle a variety of uncertainties. As an
extension of the previous efforts, the objective of this study
is to develop an integrated fuzzy-stochastic risk assessment
(IFSRA) approach that can quantify both probabilistic
and fuzzy uncertainties associated with site conditions,
environmental quality guidelines, and health impact
criteria. An attempt will be made to link both types of
uncertainties through the concepts of fuzzy logic and
stochastic analysis, and the developed IFSRA will then be
applied to a petroleum-contaminated groundwater system
in western Canada.
2. Probabilistic versus fuzzy reflection of uncertainties in
environmental systems
Uncertainties can be classified into two broad categories:
probabilistic and fuzzy ones (Destouni, 1992; Blair et al.,
2001). In the probabilistic (e.g. stochastic) approach,
probability distributions are used to describe random
variability in parameters. These distributions are then
propagated to the output variables through mathematical
models using some statistical sampling algorithms. In the
fuzzy set approach, on the other hand, membership
functions are used to characterize vagueness in human
thoughts. This approach has the power to handle the
concept of ‘‘partial truth’’ to quantify uncertainties
associated with linguistic variables (Chen and Pham,
2001). The rationales behind these two uncertainty-
manipulation approaches are different (Chen, 2000). For
example, Fig. 1a represents a PDF of soil porosity which is
an important input parameter for groundwater modeling,
while Fig. 1b shows a membership function of ‘‘porous
soil’’ in terms of soil porosity. The area under the porosity
PDF curve in an interval is equal to the probability of
porosity assuming any value in that interval, but the
probability of assuming a specific value is zero [i.e.
Pðf ¼ 0:35Þ ¼ 0] (Chen, 2000). The total area under the
PDF curve is equal to 1, meaning that the sum of all the
probabilities in the sample space is 1. On the other hand,
the area under the membership function curve has no
meaning, while the total area can be less or greater than
one. The membership function of porosity can be any value
between zero and one [i.e. mðf ¼ 0:35Þ ¼ 0:7, meaning that
the soil with porosity of 0.35 belongs to ‘‘porous soil’’ with
a possibility of 0.7].
The probabilistic approach is widely used when sufficient
information is available for estimating the probability
distributions of uncertain parameters, while the fuzzy set
method is well suited to deal with uncertainties when
little information is known (i.e. imprecise knowledge
from descriptions of human language). There have been
a number of environmental applications using either
 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188
Area under curve
<1.0 (i.e. =1
f (φ)
0.05)
0.35
(a) P (φ=0.35) = 0
Fig. 1. Comparison of probabilistic (a) and fuzzy (b) uncertainties.
probabilistic or fuzzy set approaches individually during
the past years (Dahab et al., 1994; James and Oldenburg,
1997; Batchelor et al., 1998; Maxwell et al., 1998;
Mohamed and Cote, 1999; Foussereau et al., 2000).
However, uncertainty-manipulation through individual
approaches may not be feasible in real-world situations
when different information qualities exist for various
parameters. For example, uncertainties in soil parameters
associated with a contaminant transport model may be
characterized by probability distributions, while the
uncertainties in environmental guidelines and health
impact criteria may be fuzzy in nature. Thus, if the
soil parameter uncertainties are represented through fuzzy
membership functions when they can be adequately
described through probabilistic distributions, then some
important information may be lost and not captured.
On the other hand, if the probabilistic distributions
are used to represent uncertainties that can only be
described by linguistic variables such as risk evalua-
tion criteria, this over-manipulation of input information
may lead to significant input errors at the very beginning
of modeling efforts. It is thus a challenging problem
to effectively deal with a combination of different
information quality levels for the uncertain inputs in
modeling procedures. An individual fuzzy or stochastic
method can hardly characterize such complexities (Blair
et al., 2001).
Risk assessment is a very important component of
environmental decision making, with various uncertainties
being associated with it. Since risk can be defined as an
exposure to undesirable consequences (Piver et al., 1998), it
involves two aspects: exposure and the nature of the
undesirable consequences. Therefore, the modeling of
uncertainty in a risk assessment can be the explicit
quantification of probabilities, possibilities, and potential
undesirable consequences based on all the information
available about the risks under consideration (Andricevic
and Cvetkovic, 1996). In this study, the uncertainties in
parameters associated with a subsurface contaminant
transport simulation model are assumed to be probabilistic
in nature, while the uncertainties in evaluation criteria
associated with risk assessment are assumed to be fuzzy in
nature. Thus, an integrated fuzzy-stochastic approach will
be developed for addressing such complexities and to offer
175
Meaningless for area
under curve
μ (φ)
1.0
0.7
φ 0.35 φ
(b) μ (φ=0.35) = 0.7 ≠ 0
a more realistic evaluation of the related risks due to site
contamination.
3. Risk assessment of groundwater contamination
3.1. Stochastic simulation of contaminant transport
Modeling of subsurface pollutant transport requires
inputs of various physical, chemical, and biological
parameters. However, some fundamental parameters, such
as soil permeability and porosity, are generally difficult to
acquire with accurate and deterministic values (Labieniec
et al., 1997). Among various stochastic techniques to
quantify uncertainties associated with such parameters, the
most popular approach has been Monte Carlo simulation
which uses repeated executions of a numerical model
(James and Oldenburg, 1997). Each execution produces a
sample output, and the output samples can then be
examined statistically to determine the related probabilistic
distributions. In this study, a Monte Carlo simulation
algorithm has been developed and incorporated within a
multiphase multi-component numerical simulator named
UTCHEM which can predict the spatial and temporal
distributions of the concentrations of multiple contami-
nants (UTA, 2000; Li et al., 2003). Thus, a stochastic
modeling system was then developed with the following
procedures: (a) generate random numbers for each
stochastic input parameter; (b) transform the random
numbers to the corresponding random variates based
on a specified statistical distribution for each parameter;
(c) store the generated random variates in an array for each
parameter; (d) obtain a value from the array for each
parameter and use it as a deterministic input parameter
in the multiphase multi-component numerical model;
(e) compute contaminant concentrations through the
numerical model for each Monte Carlo run; (f) store the
resulting outputs of contaminant concentrations from each
Monte Carlo run for further statistical analysis; (g) repeat
steps (a)–(f) for a specified number of Monte Carlo runs;
(h) stop computation when all the runs have been done,
and (i) analyze the contaminant concentrations and
compute the statistic moments (i.e. mean and standard
deviation of contaminant concentrations over the number
of Monte Carlo runs for each temporal and spatial unit).
 ARTICLE IN PRESS
176 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188
3.2. General approaches for risk assessment
The related risk characterization associated with a
contaminated site can usually be conducted through
environmental-guideline-based risk assessment (ERA) and
health risk assessment (HRA) (Carrington and Bolger,
1998). In this study, the environmental-guideline-based risk
(ER) is defined as the risk due to violation of environ-
mental guidelines or regulations, and the health risk (HR)
as the risk of health impacts due to chronic intake of the
contaminant. For the purpose of groundwater contamina-
tion risk assessment, the ERA approach is to compare
the contaminant concentration with the corresponding
groundwater quality standard. Through Monte Carlo
simulation, the probability (PF) under which the contami-
nant concentration exceeds the quality standard can be
described as follows:
PF ¼ PðC4C s Þ ¼ 1  F ðC s Þ, (1)
where C is the contaminant concentration, Cs is the
groundwater quality standard, and F ðC s Þ is the cumulative
distribution function (CDF) of contaminant concentration
which can be obtained from Monte Carlo simulation
results.
In cases when the concentrations are above guidelines
but there is few exposure, the cleanup may then be
perceived as being not a top priority issue at the site. This
situation cannot be characterized by ERA alone. In order
to better manage such situations, further HRA is needed.
To quantify human health risks, pollutants are character-
ized as carcinogens and non-carcinogens. HRA includes an
excess lifetime cancer risk (ELCR) model for assessing
carcinogenic risk, and a hazard index (HI) model for non-
carcinogenic risk. The degree of exposure to a chemical is a
function of many variables described as follows (USEPA,
1989, 1992):
CDI ¼ CW  IR  EF  ED=ðAT  BWÞ, (2)
where CDI is the chronic daily intake (mg/kg day), CW is
pollutant concentration in groundwater (mg/L), IR is
human ingestion rate (L/day), EF is exposure frequency
(days/year), ED is average exposure duration (year), BW is
average body weight (kg), and AT is averaging time
(AT ¼ 365  ED days). The ELCR and HI can then be
calculated by the following equations:
ELCR ¼ CDI  SF; (3)
HI ¼ CDI=RfD; (4)
where SF is a carcinogen slope factor (kg day/mg), and
RfD is a reference dose (mg/kg day). The calculated
carcinogen risk value (i.e. ELCR) implies the probability
of excess lifetime cancer from exposure to a specific
pollutant, and it means the number of people who will
get cancer per million of population because of exposure to
the contaminants. The USEPA defines acceptable risk
levels for carcinogens as being within the range of
104–106. On the other hand, the acceptable level of
non-carcinogen risk denoted by the HI is defined as lower
than 1.0.
4. Integrated fuzzy-stochastic risk assessment
Fig. 2 illustrates the integrated risk assessment frame-
work proposed in this study, and the detailed explanations
are described below.
4.1. Necessity of linkage between fuzzy and stochastic
uncertainties
The direct application of environmental risk analysis via
Monte Carlo simulation methods may lead to two
potential shortcomings. Firstly, risk analysis using Monte
Carlo outputs requires that distributions of input para-
meters should be precisely specified; secondly, researchers
mostly assume that input parameters are independent of
one another even though they are obviously not. Although
methods to simulate correlations among parameters exist,
they are not detailed enough for further risk quantification,
especially when the dependencies are not well known. As a
result, uncertainties still exist in the Monte Carlo simula-
tion outputs.
The event of a contaminant concentration exceeding the
corresponding guideline can be regarded as a stochastic
one due to randomness in the input parameters. The
occurrence of such an event can be predicted through
Monte Carlo simulation with the help of the probability
concept. However, its consequence (i.e. risk) can hardly be
characterized by using a probability due to uncertainties in
Monte Carlo simulation outputs and environmental guide-
lines as well as the fact that different people may have
different insights regarding risks. Such a consequence is of
a fuzzy nature which can be quantified by a degree of belief
(e.g. membership function). In fact, risk is related to two
aspects: event-occurrence (i.e. exposure) probability and
the consequences or outcomes associated with such
occurrence (Finley and Paustenbach, 1997). Systematic
consideration of any linkage between these two aspects is
critical for practical and realistic risk assessment.
In this study, the environmental-guideline-based risk is
treated as a fuzzy event by categorizing it into ‘‘low’’, ‘‘low-
to-medium’’, ‘‘medium’’, ‘‘medium-to-high’’, and ‘‘high’’.
Each category is associated with different magnitudes of
guideline-violation probability that is obtained from
Monte Carlo simulation outputs. In addition, the guide-
lines are categorized into three different-strictness-degree
sets, and are also taken into account within the modeling
system for quantifying risk levels. The construction of
membership functions for these fuzzy events will rely on a
questionnaire survey of experts in industry, government,
education and research organizations. A total of 48
questions were designed, including 3 questions for inves-
tigating the strictness of groundwater quality guidelines,
15 questions for the environmental risk levels under the
three different-strictness groundwater quality guidelines,
 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188 177

Fuzzy uncertainty
Modeling Environmental quality
parameters guideline

Medium guideline
Probabilistic

Loose guideline
Strict guideline
uncertainty Environmental-guideline-
based risk (ER) assessment
Modeling of
contaminant
transport

Medium-to-high ER risk
Low-to-medium ER risk

Medium ER risk

High ER risk
Low ER risk
Probabilistic
distribution of Guideline-violation probability
contaminant
concentration

Medium-to-high HR risk
Low-to-medium HR risk

Medium HR risk

High HR risk
Low HR risk

Exposure analysis
Fuzzy rule
Fuzzy base
uncertainty

General risk levels (GRL)

Health risk
evaluation criteria
Health risk (HR) assessment

Medium-to-high risk
Low-to-medium risk
Site management

Very high risk

Fuzzy membership functions

Medium risk
actions

High risk
Low risk
of GRL

Site risk score

Fig. 2. Modeling framework of the integrated fuzzy-stochastic risk assessment.

5 questions for the health risk levels, and 25 questions for Many guidelines are considered over-conservative and
the general risk levels (or generating the fuzzy rule base). impractical for environmental risk analysis (Chen et al.,
One example of such questions is ‘‘For a medium ground- 2003). As a result, the risk indicators such as the degrees of
water quality guideline, which of the following xylene guideline violation are not compatible among different
concentrations would you like to choose (please choose regions, leading to uncertainties in guideline applicability.
one): (a) approximately 0.02 mg/L; (b) approximately Such uncertainties can hardly be quantified by probabilistic
0.2 mg/L; (c) approximately 1.0 mg/L; (d) approximately distributions, but they can be described using linguistic
2 mg/L; (e) approximately 4 mg/L; (f) approximately 8 mg/ variables that can be quantified through a fuzzy logic
L; (g) approximately 10 mg/L’’. Thus, this study integrates approach.
fuzzy logic, expert involvement, and stochastic simulation To facilitate a guideline-based environmental risk
within a general framework, leading to the linkage between analysis, the guidelines were categorized into three fuzzy
probabilistic and fuzzy uncertainties using the fuzzy logic sets in this study, namely ‘‘strict’’, ‘‘medium’’ and ‘‘loose’’.
concept. A questionnaire survey was conducted to collect data for
establishing membership functions, and the fuzzy logic
4.2. Fuzzy environmental quality guidelines approach was further used for processing the subjective
opinions. The option with the highest survey response
The environmental-guideline-based risk assessment in- frequency was used for determination of corresponding
volves a comparison of contaminant concentration with its fuzzy sets. The conversion scale figures proposed by Chen
corresponding environmental guideline. In this study, et al. (1992) and Cheng (2000) were used to systematically
xylene is selected as the contaminant of interest. It has transform linguistic terms to their corresponding fuzzy sets.
significantly varying groundwater quality standards among These scale figures cover all the linguistic expressions of
countries, states and provinces. As indicated in Table 1, the ‘‘HIGH’’ vs ‘‘LOW’’. The corresponding linguistic-term
guidelines for this contaminant are 0.02, 0.04, 0.3, 0.4, 0.53, conversion procedure involves selecting a figure that
1.8, and 10 mg/L in Sweden, New Jersey, Canada, Japan, contains all the verbal terms given by the decision-maker,
North Carolina, California, and Illinois, respectively. and then using the membership function set for that figure
 ARTICLE IN PRESS
178 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188

Table 1
Environmental guidelines and the corresponding chronic daily intake and hazard index of xylene

Region Guideline for xylene (mg/L) Corresponding chronic Hazard index (HI)
daily intake (CDI)
(mg/kg day)
Corresponding to old RfD Corresponding to new RfD
(2.0 mg/kg day) (0.2 mg/kg day)

Environment Canada 0.3 0.0082 0.0041 0.0411

Saskatchewan 0.3 0.0082 0.0041 0.0411
Alberta 0.3 0.0082 0.0041 0.0411
USEPA 10 0.2740 0.1370 1.3699
Illinois 10 0.2740 0.1370 1.3699
California 1.8 0.0493 0.0247 0.2466
North Carolina 0.53 0.0145 0.0073 0.0726
New Jersey 0.04 0.0011 0.0005 0.0055
Thailand 10 0.2740 0.1370 1.3699
Sweden 0.02 0.0005 0.0003 0.0027
Japan 0.4 0.0110 0.0055 0.0548
World Health Organization 0.5 0.0137 0.0068 0.0685

should be approximately 50% or less’’ was of ‘‘low risk’’;

Strict Medium Loose 57.9% of the respondents selected ‘‘the probability of
1.0
guideline violation should be approximately 60%’’ as the
0.8 ‘‘low-to-medium risk’’; 57.9% of them chose ‘‘the prob-
ability of guideline violation should be approximately
0.6 70%’’ as a ‘‘medium risk’’; 57.9% of the respondents
chose ‘‘the probability of guideline violation should be
μg

0.4 approximately 80%’’ as a ‘‘medium-to-high risk’’; and

0.2
0.0
1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
guideline (mg/L)
Fig. 3. Membership functions of groundwater quality guidelines.
to represent the meaning of these verbal terms. In this
study, it was found that 52.6% of the surveyed respondents
indicated that ‘‘xylene concentration should be approxi-
mately 1.0 mg/L or less’’ was a ‘‘strict’’ guideline; 50.0% of
the respondents selected ‘‘xylene concentration should be
approximately 4.0 mg/L’’ as a ‘‘medium’’ one, and 47.4%
of them selected ‘‘xylene concentration should be approxi-
mately 8.0 mg/L or greater’’ as a ‘‘loose’’ one. As a result,
the membership functions of these three fuzzy sets can be
constructed as shown in Fig. 3. For example, if the
groundwater quality guideline is 1.8 mg/L, then it can be
categorized as partly ‘‘medium’’ (with a membership grade
of 0.27) and partly ‘‘strict’’ (with a membership grade of
0.60).
4.3. Fuzzy environmental-guideline-based risk assessment
Based on the questionnaire survey results in terms of
environmental-guideline-based risks under strict guidelines,
it was found that 55.3% of the surveyed respondents
indicated that ‘‘the probability of guideline violation
76.3% of the respondents selected ‘‘the probability of
guideline violation should be approximately 90% or
greater’’ as the ‘‘high risk’’. Again, according to Chen et
al. (1992) and Cheng (2000), the membership functions of
these fuzzy sets can be constructed based on these
investigation results (Fig. 4a). In this figure, ‘‘L’’, ‘‘L-M’’,
‘‘M’’, ‘‘M-H’’ and ‘‘H’’ represent ‘‘low’’, ‘‘low-to-med-
ium’’, ‘‘medium’’, ‘‘medium-to-high’’ and ‘‘high’’, respec-
tively. For example, if the probability of strict-guideline
violation is 75% from Monte Carlo simulation results, the
related environmental-guideline-based risk can be categor-
ized as partly ‘‘medium’’ (with a membership grade of 0.50)
and partly ‘‘medium-to-high’’ (with a membership grade
of 0.50). Similarly, the membership functions of fuzzy
environmental-guideline-based risks under the medium
and loose guidelines can be established according to the
questionnaire survey results as shown in Figs. 4b and c,
respectively.
4.4. Fuzzy health risk assessment
HRAs often adopt USEPA-published reference doses
(RfD) and slope factors (SF) that usually come from
laboratory studies on animals. As more and more
toxicologic studies are completed, many updated data on
RfD and cancer potency SF will be added to the USEPA
database. For example, USEPA recently decreased the
RfD value by 10 times (from 2.0 to 0.2 mg/kg d) for xylene
(USEPA, 2003). This leads to significant differences in
the relevant risk analysis. The HI of xylene under the old
 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188 179

L L-M M M-H H
1.0

μER 0.8

0.6

0.4

0.2

0.0
0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.0
(a) P

L L-M M M-H H
1.0

0.8
μER

0.6

0.4

0.2

0.0
0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.0
(b) P

L L-M M M-H H
1.0

0.8
μER

0.6

0.4

0.2

0.0
0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60
(c) P

Fig. 4. Membership function of fuzzy environmental risks associated with probability (P) of violating groundwater quality guideline with (a) for strict
standard, (b) for medium standard, and (c) for loose standard.
 ARTICLE IN PRESS
180 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188

L L-M M M-H H
1.0

0.8
μHR

0.6

0.4

0.2

0.0
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
(0.04) (0.06) (0.10) (0.16) (0.25) (0.40) (0.63) (1.0) (1.6)
Log (10HI) (HI)

Fig. 5. Membership function of health risks associated with hazard index (HI).

and new RfD formulations and different environmental A B A B A

guidelines is listed in Table 1, with the modeling parameter
values of IR ¼ 2 L/day, EF ¼ 350 days/year, ED ¼ 30
years, and BW ¼ 70 kg. For example, the HI is 0.137 under
the USEPA guideline when the old RfD is applied, and
1.37 when the new one is used. As a result, the different risk
values may result in different perceptions of health risks,
leading to different decision actions. Thus these uncertain-
ties due to incomplete toxicity information should be
incorporated within the process of risk analysis. However,
the quality of information available is not good enough
for using probabilistic distributions to describe the
applicability or suitability of the guidelines, resulting in
consequent difficulties in quantifying associated health
risks. Since such uncertainties usually show subjective
features, a fuzzy logic approach is proposed in this study to
account for related complexities.
In this study, non-carcinogenic health risk introduced by
xylene will be investigated. The health risk levels will be
categorized into fuzzy sets of ‘‘low’’, ‘‘low-to-medium’’,
‘‘medium’’, ‘‘medium-to-high’’, and ‘‘high’’ by associating
them with different magnitudes of the HI. A questionnaire
survey was conducted to obtain the associated membership
functions. The investigated HI was set to vary from 0.04
(corresponding to the Canadian groundwater quality
guideline for xylene) to 1.6 (approximately corresponding
to the USEPA guideline for xylene). Similar to the
identification of environmental-guideline-based risk levels,
the membership functions of related fuzzy health risk levels
can be established. Since the investigated HI levels vary
from 0.04 to 1.6, they were scaled down by using logarithm
values such as Log(10HI). Membership functions of these
fuzzy events are shown in Fig. 5. For example, if the
calculated HI is 0.40, then the related health risk can be
categorized as partly ‘‘medium’’ (with a membership grade
A and B
Not A
A or B
AND OR NOT
Min [μA(x), μB(x)] Max [μA(x), μB(x)] 1-μA(x)
Fig. 6. Basic operations in fuzzy logic.
of 0.5) and partly ‘‘medium-to-high’’ (with a membership
grade of 0.5).
4.5. General risk levels and rule base for risk management
The general risk level is derived from an integrated
consideration of environmental-guideline-based and health
risks. Since there are no mathematical models to relate
both types of risks, quantification of the general risk level
can only be based on subjective opinions rather than
through probabilistic analyses. A fuzzy inference process
will then be introduced for facilitating this kind of risk
quantification, by using fuzzy membership functions and
fuzzy rules. The rules usually contain a conditional part
(e.g. antecedent) and a conclusion part (e.g. consequence).
An antecedent may be a simple clause or may be a
combination of several clauses connected via fuzzy logical
operators AND, OR, and NOT (Fig. 6). For example, ‘‘if
environmental-guideline-based risk is MEDIUM and
health risk is HIGH, then the general risk is MEDIUM-
to-HIGH’’, where the ‘‘MEDIUM’’ is a fuzzy set of
 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188 181

Table 2
Survey results on fuzzy rules

Antecedent Survey result with highest response frequency

If environmental-guideline-based risk If health risk (HR) is Then the general risk level (GRL) is Frequency (%)
(ER) is

Low Low Low 65.8

Low Low-to-medium Low-to-medium 76.3
Low Medium Medium 78.9
Low Medium-to-high Medium-to-high 81.6
Low High High 73.7
Low-to-medium Low Low-to-medium 71.1
Low-to-medium Low-to-medium Low-to-medium 100
Low-to-medium Medium Medium 76.3
Low-to-medium Medium-to-high Medium-to-high 78.9
Low-to-medium High High 76.3
Medium Low Medium 76.3
Medium Low-to-medium Medium 78.9
Medium Medium Medium 100
Medium Medium-to-high Medium-to-high 81.6
Medium High High 73.7
Medium-to-high Low Medium-to-high 84.2
Medium-to-high Low-to-medium Medium-to-high 81.6
Medium-to-high Medium Medium-to-high 76.3
Medium-to-high Medium-to-high Medium-to-high 100
Medium-to-high High High 81.6
High Low High 71.1
High Low-to-medium High 71.1
High Medium High 68.4
High Medium-to-high High 73.7
High High Very High 63.2

‘‘environmental-guideline-based risk’’, the ‘‘HIGH’’ is a

fuzzy set of ‘‘health risk’’, and the ‘‘MEDIUM-to-HIGH’’ L L-M M M-H H V-H
is a fuzzy set of ‘‘general risk’’. This is a form of fuzzy rules, 1.0
where the ‘‘environmental-guidelined-based risk’’ and 0.8
‘‘health risk’’ are input variables, the ‘‘general risk’’ is an
μGR

output variable. Such a set of rules is known as the rule 0.6

base (Mohamed and Cote, 1999).
In this study, determination of the general risk will be
based on a series of fuzzy rules as acquired through a
questionnaire survey of experts and stakeholders. The risk
levels are set to include six categories of fuzzy sets: ‘‘low’’,
‘‘low-to-medium’’, ‘‘medium’’, ‘‘medium-high’’, ‘‘high’’,
and ‘‘very-high’’. The fuzzy logic operator ‘‘AND’’ is used
to join factors in the antecedent of the rules. Since both
types of risks include five categories of fuzzy events, there
will be a total of 150 rules (5  5  6). Assume that, if a rule
obtains the highest frequency in the survey, it will be kept in
the rule base for determination of the general risk level. As
a result, 25 fuzzy rules were obtained as shown in Table 2.
Since the general risk level can be categorized into
‘‘low’’, ‘‘low-to-medium’’, ‘‘medium’’, ‘‘medium-to-high’’,
‘‘high’’ and ‘‘very-high’’, the corresponding membership
functions of these fuzzy events can be established according
to Chen et al. (1992) and Mohamed and Cote (1999) as
shown in Fig. 7. The range of the general risk levels (i.e.
GRL ¼ [0, 100]) is subjectively given to the fuzzy sets in
order for them to have single numerical site risk scores
0.4
0.2
0.0
0 10 20 30 40 50 60 70 80 90 100
GRL
Fig. 7. Membership functions of fuzzy general risk levels.
(Mohamed and Cote, 1999). These numerical values have
no direct relationship with the values of input risk factors
(e.g. environmental-guideline-based risk and health risk).
However, after establishing the fuzzy sets of general
risk level, a numerical site risk score can be obtained
from Fig. 7 through the fuzzy ‘‘AND’’ or fuzzy ‘‘OR’’
operations based on the environmental guideline, the
probability of guideline violation, and the corresponding
HI. This will be illustrated in the following case study. The
site management decisions can then be made based on the
calculated site scores that describe the general risk level.
Table 3 lists the relationship between site scores and
 ARTICLE IN PRESS
182 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188

suggested management actions (Mohamed and Cote, proposed approach in this study. The site has been
1999). operated as a natural gas processing plant to remove
naphtha condensate from the gas stream prior to transport
to a regional transmission line. Fig. 8 shows an overview
5. Case study of the study site. Throughout the site history, naphtha
condensate, a waste liquid removed from natural gas by a
5.1. Overview of the study site series of scrubbers, was disposed of in a previous under-
ground storage tank (UST). Due to leakage of this UST in
A petroleum-contaminated site in western Canada was the past years, the site has been contaminated. During the
selected as a case to illustrate the applicability of the past several years, a number of field investigations have
been conducted to examine the site hydrogeologic condi-
Table 3 tions and related pollution problems. The groundwater
Recommended risk management actions flow was towards the northwest, which was determined
Calculated site score Risk management action through multiple site investigation results. The previous
UST that was at the up-gradient location of the site leaked
90–100 The site should be immediately approximately 55 m3 of petroleum products before it was
cleaned up
removed. Various forms of pollutants were found in the
70–90 Take full actions to treat the site
50–70 Contain the site and restrict down-gradient directions; they include free product, vapor-
groundwater use phase hydrocarbons, dissolved hydrocarbons in ground-
30–50 Take interim control measures and water, and trapped hydrocarbons in soil.
limit access to the site The constituent of interest is xylene in this study, which
10–30 The site should be monitored
is a toxic chemical and can result in impairment of lung and
0–10 No actions are required
respiratory functions, impairment of memory, damage of

BH206
BH205 BH207 14
15
N
BH104 BH109
BH108
11
BH204 BH203 BH107
BH111 Meter Building
Pump/Scrubber
16 Building BH101

BH304 BH401 BH209

BH202 BH106
13
A/G Tank 12
BH303 7
BH201 BH103 2
BH115 BH402
BH105 6
BH301
10 7 Compound
17
BH102
18 8 Existing
BH110 5 UST
BH208
BH112 Gate
3
BH113
Pole Mounted
Transformers Ditch
Ditch
BH302

Highway

Legend: Monitoring Well Approximate Extent of Residual Hydrocarbons

Borehole
Approximate Extent of LPH Plume
BH111 Well Name
Former UST Approximate Scale
Groundwater Flow
Direction Property boundary 0 10 m

Fig. 8. Overview of the study site.

 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188 183

Table 4
140 Groundwater Flow Part of the modeling parameters
NW 1.10
120 1.00
Parameter Value Unit
0.90
100 0.80 Mean of permeability of sandy soil in x, y, 2900 MD

0
0.
0.70 and z direction
Y (m)

0
80 0.60 Mean of permeability of clay till in x, y, 195 MD

0.
0. and z direction
9
0. 0.50
6
60 0.
3 0.40 Mean of permeability of silty clay in x, y, 380 MD
0. and z direction
0 0.30
40 Standard deviation of the logarithm of 0.60 —
0.20
sandy soil permeability
0.10 Standard deviation of the logarithm of clay 0.60 —
20
0.00 till permeability
0 Standard deviation of the logarithm of silty 0.60 —
0 20 40 60 80 100 120 140 160 180 clay permeability
X (m) Mean of porosity of sandy soil 0.35 —
Mean of porosity of clay till 0.30 —
Fig. 9. Observed xylene concentrations in groundwater (mg/L). Mean of porosity of silty clay 0.53 —
Standard deviation of sandy soil porosity 0.03 —
Standard deviation of silty clay porosity 0.03 —
Standard deviation of sandy soil porosity 0.03 —
X NAPL density 0.713 g/cm3
Longitudinal dispersivity of sandy soil 5 m
Y Longitudinal dispersivity of clay till 5 m
Z Layer 1 Longitudinal dispersivity of silty clay 5 m
Clay/silty clay Transverse dispersivity of sandy soil 0.5 m
Transverse dispersivity of clay till 0.5 m
Transverse dispersivity of silty clay 0.5 m
Hydraulic gradient 0.006 m/m
Clay/silty clay Till Layer 2 Benzene solubility 1750 mg/L
Ethylbenzene solubility 152 mg/L
Toluene solubility 535 mg/L
Flow Flow Xylene solubility 175 mg/L
Clay/ Till Sand Time step at t ¼ 0 0.101 Day
silty clay Layer 3 Maximum time step size 10 Day
Tolerance for concentration change 0.001 —

Till Sand Layer 4

Fig. 10. Soil properties in the simulation domain.
kidney and liver, lack of muscle coordination, dizziness,
and possibly reproductive effects. In extreme cases, it may
result in death (USEPA, 2003). Fig. 9 shows the observed
xylene concentration in groundwater at the site. The
highest xylene concentration of 1.12 mg/L was observed
at a monitoring well located to the west of the previous
UST. This concentration was higher than the regulated
local environmental guideline. Thus, soil and groundwater
contamination at this site may pose environmental risks
and health impacts on the surrounding communities.
5.2. Results of stochastic simulation
The developed stochastic modeling system was used for
simulating contaminant transport in the groundwater. The
site was considered as a three dimensional domain with an
area of 180  150 m2 and a depth of 20 m. The simulation
domain was vertically discretized into four gridblocks
corresponding to four layers; each layer was located at the
middle of each vertical grid block, since the model applied
a block-centered finite difference scheme. The vertical
gridblock size was 5 m. Each layer with an area of
180  150 m2 was discretized into 30  25 grid blocks. Each
grid had dimensions of 6, 6, and 5 m in the x, y, and z
directions, respectively. The total number of grids in the
3-D computational grid system was 3000 (30  25  4).
Layers 3 and 4 were located in the saturated zone, while
layers 1 and 2 were in the unsaturated zone. The site
possesses three soil types including sandy soil, clay till and
silty clay. Distribution of the three soil types in the
simulation domain is described in Fig. 10. Table 4 lists part
of the modeling inputs. The input data were obtained from
a number of sources after careful analysis and verification.
Among them, reports of the previous site investigation,
laboratory analysis, and a modeling study were the major
sources. The values of soil permeability (Kxx) and porosity
(f) were considered to be log-normally and normally
distributed, respectively. The mean values and standard
deviations of Kxx and f for three on-site soil types (sand,
clay till, and silty clay) were used for generating the Monte
Carlo realizations of Kxx and f based on assumed
probabilistic distributions. As a result, the contaminant-
concentration-distribution patterns can be obtained by
 ARTICLE IN PRESS
184 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188

nants (i.e. benzene, toluene, ethyl-benzene, xylene, etc.),

140
only the simulation results of xylene concentrations are
120 1300 presented in this paper. The mean and standard devia-
1200 tion of xylene concentration 10 years later from the
100
Monte Carlo simulations are shown in Figs. 11 and 12,
1000
respectively. With a conservative consideration, the peak
Y (m)

80

80
800 contaminant concentration at the site 10 years later for
0.0
0
40 0.00
60 600 each Monte Carlo run is identified for further fuzzy-
0.0
5
0

400 stochastic risk assessment. The peak values range from

40
200 0.931 to 1.951 mg/L with a mean of 1.35 mg/L and
20 standard deviation of 0.316 mg/L). Fig. 13 shows a
50
cumulative relative frequency distribution of the peak
0
xylene concentrations.
0 20 40 60 80 100 120 140 160 180
X (m)
5.3. Risk assessment results
Fig. 11. Mean of the xylene concentrations (mg/L) at 10 years later from
Monte Carlo simulations. Three scenarios were examined based on the drinking
water quality guidelines formulated by the Province of
Saskatchewan, the State of California and the USEPA. The
140 degrees of guideline strictness were analyzed, and the
probabilities of guideline violation were obtained to
120 320 analyze the associated environmental-guideline-based
risks. The mean of the peak xylene concentrations from
100 280
Monte Carlo simulation was used for analyzing the HI
240 through oral ingestion pathways. The associated health
Y (m)

80
180
risk could then be quantified. Thus, the general risk level
180 00

60 due to the environmental-guideline-based and health risks

50.
.00

140 was obtained through fuzzy ‘‘AND’’ and ‘‘OR’’ opera-

40
100 tions.
20 The first scenario was to apply the Saskatchewan
50
groundwater quality guideline of 0.3 mg/L for xylene. It
0 is indicated from Fig. 3 that this guideline is ‘‘strict’’ (with a
0 20 40 60 80 100 120 140 160 180 membership grade of 1.0). It is also found from Fig. 13 that
X (m) PðCo0:3Þ ¼ F ð0:3Þ ¼ 0:0; thus the probability of guideline
Fig. 12. Standard deviation of the xylene concentrations (mg/L) at 10 violation is PF ¼ 1  F ð0:3Þ ¼ 1:0 based on Eq. (1). As a
years later from Monte Carlo simulations. result, the environmental-guideline-based risk (ER) would
be ‘‘high’’ with a membership grade of 1.0 according to
Fig. 4a when the probability was 1.0 and the guideline was
1
‘‘strict’’, and it is illustrated in Figs. 14a and d. Since the
0.9
mean of the peak xylene concentrations was 1.35 mg/L, the
Cumulative relative frequency

0.8
corresponding chronic daily intake (CDI) can be calculated
0.7
as 0.037 mg/kg d according to Eq. (2). The associated HI
0.6 can be calculated as 0.185 according to Eq. (4) when a
0.5 reference dose (RfD) of 0.2 mg/kg d was used. When HI is
0.4 equal to 0.185, it could then be found from Fig. 5 that the
0.3 corresponding health risk (HR) would be partly ‘‘low-to-
0.2 medium’’ (with a membership grade of 0.33) and partly
0.1 ‘‘medium’’ (with a membership grade of 0.67), as shown in
0 Figs. 14b and c, respectively. Therefore, two combinations
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 of antecedents exist (i.e. two rules have to be analyzed)
Peak concentration of xylene (mg/L) including: (a) if ER is ‘‘high’’ and HR is ‘‘low-to-medium’’,
Fig. 13. Cumulative relative frequency distribution of peak xylene and (b) if ER is ‘‘high’’ and HR is ‘‘medium’’.
concentration. The related fuzzy inference process is shown in Fig. 14.
The fuzzy ‘‘AND’’ operation was applied to the rule’s
running the stochastic modeling system. Although the antecedent to determine its consequence (e.g. mGR ¼ Min
modeling system predicted the spatial and temporal (mER, mHR)) according to the rule base as shown in Table 2.
distributions of the concentrations of multiple contami- In other words, the minimum degree of membership grade
 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188
μER ER μHR HR
H L-M
0.33
(a) (b)
μER ER HR
H μHR
0.67
(d) (e)
Fuzzy “AND” operation
Note: fuzzy “AND” operation of (a) and (b) to obtain (c);
fuzzy “AND” operation of (d) and (e) to obtain (f); fuzzy
“OR” operation of (c) and (f) to obtain (g)
Fig. 14. Fuzzy inference process under scenario 1, (a) and (d): environmental risks; (b) and (e): health risks; (c), (f) and (g): general risk levels.
of the two input factors (ER and HR) was given to the
output factor (GRL). According to Table 2, the general
risk level would be ‘‘High’’ for the antecedent ‘‘if ER is
‘high’ and HR is ‘low-to-medium’’’. Based on Fig. 7, the
‘‘High’’ general risk level was represented by a triangular
membership function with the site score ranging from
60 to 100 as shown in Fig. 14c, and the corresponding
membership grade of this risk level was mGR ¼ Min (1.0,
0.33) ¼ 0.33. Similarly, the output of the second antecedent
‘‘if ER is ‘high’ and HR is ‘medium’’’ is a ‘‘High’’ general
risk level with a membership grade of mGR ¼ Min (1.0,
0.67) ¼ 0.67 as shown in Fig. 14f. As a result, the outputs
from the inference procedure which provide inputs for the
fuzzy composition process were then two scaled-down
fuzzy general risk levels (GRLs) (Mohamed and Cote,
1999). The fuzzy ‘‘OR’’ operation was then applied to the
two fuzzy GRLs. In other words, the two fuzzy events
represented by Figs. 14c and f were superimposed to obtain
the final fuzzy GRL. As shown in Fig. 14g, the final GRL
would be ‘‘High’’ with the site score ranging from 60 to 100
and with a membership grade of mGR ¼ Max (0.33,
0.67) ¼ 0.67 under this scenario. The crisp final GRL
value was then obtained by calculating the centroid of the
fuzzy GRL value as 80. As a result, the suggested risk
management action would be ‘‘take full actions to treat the
site’’ according to Table 3.
The second scenario was to apply the California ground-
water quality guideline of 1.8 mg/L for xylene. This guide-
line is looser than the corresponding Saskatchewan
environmental quality guideline. It is indicated from
Fig. 3 that this guideline is partly ‘‘strict’’ (with a member-
ship grade of 0.60) and partly ‘‘medium’’ (with a member-
ship grade of 0.27). It is also found from Fig. 13 that the
probability of guideline violation is PF ¼ 1  F ð1:8Þ ¼ 0:14.
185
μGR GRL
H
(c) 60 100
GRL
M μGR H
(f) 60 100
Fuzzy “OR” operation
μGR GRL
H
0.67
(g) 60 100
As a result, the environmental-guideline-based risk (ER)
would be ‘‘low’’ with a membership grade of 0.60 ( ¼ Min
(1.0, 0.60)) according to Fig. 4a when the probability was
0.14 and the guideline was ‘‘strict’’ by applying the fuzzy
‘‘AND’’ operation; ER would also be ‘‘low’’ with a
membership grade of 0.27 ( ¼ Min (0.80, 0.27)) according
to Fig. 4b when the probability was 0.14 and the guideline
was ‘‘medium’’ by applying the fuzzy ‘‘AND’’ operation;
ER could also be ‘‘low-to-medium’’ with a membership
grade of 0.2 ( ¼ Min (0.20, 0.27)) when the probability was
0.14 and the guideline was ‘‘medium’’. Since ER could be
‘‘low’’ with a membership grade of 0.60 or with a
membership grade of 0.27, then the fuzzy ‘‘OR’’ operation
would be applied to obtain the risk level of ER. At last, the
ER in this scenario would be partly ‘‘low’’ (with a
membership grade of 0.60 shown in Figs. 15a and d) and
partly ‘‘low-to-medium’’ (with a membership grade of 0.20
shown in Figs. 15g and j).
Since the calculation of health risk was based on the
health impact criteria rather than environmental quality
criteria, the health risk (HR) under the second scenario
would be the same as that under scenario 1. It would still
be partly ‘‘low-to-medium’’ and partly ‘‘medium’’. There-
fore, four combinations of the antecedents exist under
scenario 2: (a) if ER is ‘‘low’’ and HR is ‘‘low-to-medium’’;
(b) if ER is ‘‘low’’ and HR is ‘‘medium’’; (c) if ER is ‘‘low-
to-medium’’ and HR is ‘‘low-to-medium’’, and (d) if ER is
‘‘low-to-medium’’ and HR is ‘‘medium’’. Similar to the
analysis of scenario 1, the outputs from the inference
procedure were four scaled-down fuzzy GRLs as shown in
Figs. 15c, f, i and l, respectively. The final GRL would
therefore be partly ‘‘low-to-medium’’ with a membership
grade of 0.33 and partly ‘‘medium’’ with a membership
grade of 0.60 with site scores ranging from 0 to 60 as shown
 ARTICLE IN PRESS
186 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188

ER HR GRL
μER μHR μGR L-M
L-M
L
0.60
0.33

(a) (b) (c) 0 40

ER HR GRL
μER μHR μGR
M M
L 0.67
0.60

(d) (e) (f) 20 60

ER HR GRL
μER μHR μGR
L-M L-M L-M

0.33
0.20
(g) (h) (i) 0 40

ER HR GRL
μER μHR μGR
L-M M M

0.67

0.20
(j) (k) (l) 20 60
Fuzzy “OR” operation

Fuzzy “AND” operation

μGR GRL

Note: fuzzy “AND” operation of (a) and (b) to obtain (c);

M
fuzzy “AND” operation of (d) and (e) to obtain (f); fuzzy 0.60
“AND” operation of (g) and (h) to obtain (i); fuzzy L-M
“AND” operation of (j) and (k) to obtain (l); fuzzy “OR” 0.33
operation of (c), (f), (i) and (l) to obtain (m)
(m) 0 20 40 60

Fig. 15. Fuzzy inference process under scenario 2, (a), (d), (g) and (j): environmental risks; (b), (e), (h) and (k): health risks; (c), (f), (i), (l) and (m): general
risk levels.

in Fig. 15m. The crisp final GRL value was then obtained
by calculating the centroid of the fuzzy GRL value as 36,
which was lower than that under scenario 1. As a result, the
suggested risk management action would be ‘‘take interim
control measures and limit access to the site’’ according to
Table 3, and it is observed that the application of looser
environmental quality criteria will lead to less expensive
risk management actions, as compared to scenario 1.
The 3rd scenario was to apply the USEPA groundwater
quality guideline of 10 mg/L for xylene, which is much
looser than those under scenarios 1 and 2. Similar to the
analyses of scenarios 1 and 2, the final GRL would be
partly ‘‘low-to-medium’’ and partly ‘‘medium’’ with a crisp
final GRL value of 36 as shown in Fig. 16. This value is the
same as that under scenario 2, but is lower than that under
scenario 1. As a result, the suggested risk management
action would be ‘‘take interim control measures and limit
access to the site’’ according to Table 3, although a much
looser environmental quality criteria was applied, as
compared with scenarios 1 and 2. This is due to the fact
that the calculated health risk level (i.e. the same levels
under these three scenarios) would significantly affect the
general risk level (GRL) when the environmental-guide-
line-based risk level is low, according to the fuzzy rule base
as shown in Table 2.
6. Discussion
Uncertainties due to randomness and imprecision in a
risk assessment procedure have been addressed through the
integrated fuzzy-stochastic approach called IFSRA. The
contaminant concentrations in groundwater predicted
from the numerical model were associated with probabil-
istic uncertainties due to the randomness in modeling
input parameters, while the consequences of contaminant
concentrations violating relevant environmental quality
 ARTICLE IN PRESS
J. Li et al. / Journal of Environmental Management 82 (2007) 173–188
ER HR
μER μHR
L L-M
0.33
(a) (b)
ER HR
μER μHR
L M
0.67
(d) (e)
Fuzzy “AND” operation
Note: fuzzy “AND” operation of (a) and (b) to obtain (c);
fuzzy “AND” operation of (d) and (e) to obtain (f); fuzzy
“OR” operation of (c) and (f) to obtain (g)
Fig. 16. Fuzzy inference process under scenario 3, (a) and (d): environmental risks; (b) and (e): health risks; (c), (f) and (g): general risk levels.
guidelines and health evaluation criteria were linked with
fuzzy uncertainties. As compared to conventional risk
assessment methods, the proposed IFSRA took into
account the fuzzy uncertainties in environmental quality
guidelines and health impact criteria. The uncertainties of
risk assessment due to the uncertainties in Monte Carlo
simulation outputs were also addressed (see Section 4.1) by
establishing the linkage between fuzzy and stochastic
uncertainties (Fig. 4). As a result, the proposed approach
would provide a more realistic perception of risk for
decision makers.
The membership functions of relevant fuzzy events and
the fuzzy rule base for generating the general risk levels
were established based on questionnaire surveys, and they
represent the perceptions of the investigated stakeholders.
The bounds and shapes of the membership functions were
also subjective and dependent on the particular problem
investigated. Such uncertainties in the fuzzy membership
functions and rule base themselves were not incorporated
into the approach presented in this paper, and need further
investigation. In addition, this study considered uncertain-
ties in soil properties, environmental guidelines, and health
impacts. It only examined the non-carcinogenic risk from
one contaminant (i.e. xylene) although the multiphase
multi-component simulator can predict the concentrations
of multiple pollutants. However, the comprehensive con-
sideration of carcinogenic and non-carcinogenic risks from
multiple contaminants may be needed for site management
decisions, and the methodology developed in this study
needs to be further expanded to accommodate this kind of
consideration. Moreover, the uncertainties in many other
parameters used in the contaminant transport simulator
and risk assessment model, such as soil dispersivity, daily
187
GRL
μGR
L-M
(c) 0 40
GRL
μGR
M
(f) 20 60
Fuzzy “OR” operation
GRL
μGR
M
0.67
L-M
0.33
(g) 0 20 40 60
ingestion rate, body weight and exposure time, may also
affect the predicted contaminant concentration and the
resulting risk levels, and should be examined in further
studies.
7. Conclusions
An integrated fuzzy-stochastic risk assessment (IFSRA)
approach was developed to systematically quantify both
probabilistic and fuzzy uncertainties associated with site
conditions, environmental guidelines, and health impact
criteria. The linkage between both types of uncertainties
was also effectively established. This development was
based on (a) Monte Carlo simulation for the fate of
contaminants in the subsurface through a 3-D multiphase
multi-component numerical model to account for stochas-
tic uncertainties, (b) examination of simulation results
that were expressed as cumulative distribution functions,
(c) quantification of environmental guidelines and health
impacts using fuzzy membership functions acquired from a
questionnaire survey, (d) quantification of environmental
and health risks based on fuzzy and stochastic inputs, and
(e) assessment of general risk levels based on a fuzzy logic
approach. As a result, the IFSRA integrated fuzzy logic,
expert involvement, and stochastic simulation within a
general framework. The robustness of the modeling
processes was enhanced through the effective reflection of
the two types of uncertainties as compared with the
conventional risk assessment approaches. The developed
IFSRA was applied to a petroleum-contaminated ground-
water system in western Canada. Three scenarios with
application of different environmental quality guidelines
were analyzed, and reasonable results were obtained. The
 ARTICLE IN PRESS
188 J. Li et al. / Journal of Environmental Management 82 (2007) 173–188
developed risk assessment approach offers a unique tool
for systematically quantifying various types of uncertain-
ties associated with management of petroleum contami-
nated sites.
Acknowledgments
This research has been supported by the Natural
Sciences and Engineering Research Council of Canada
and The National Natural Science Foundation of China
(No.50225926, No.50425927). The authors thank the
anonymous reviewers for their comments and suggestions
that helped in improving the manuscript.
References
Andricevic, R., Cvetkovic, V., 1996. Evaluation of risk from contaminants
migrating by groundwater. Water Resoururces Research 32, 11–622.
Batchelor, B., Valdés, J., Araganth, V., 1998. Stochastic risk assessment of
sites contaminated by hazardous wastes. Journal of Environmental
Engineering 124, 380–388.
Bennett, D.H., James, A.L., McKone, T.E., Oldenburg, C.M., 1998. On
uncertainty in remediation analysis: variance propagation from
subsurface transport to exposure modeling. Reliability Engineering
and System Safety 62, 117–129.
Blair, A.N., Ayyub, B.M., Bender, W.J., 2001. Fuzzy stochastic risk-based
decision analysis with the mobile offshore base as a case study. Marine
Structures 14, 69–88.
Carrington, C.D., Bolger, P.M., 1998. Uncertainty and risk assessment.
Human and Ecological Risk Assessment 4, 253–257.
CCME (Canadian Council of Ministers of the Environment), 1996. A
Framework for Ecological Risk Assessment: General Guidance.
Canadian Council of Ministers of the Environment, National
Contaminated Sites Remediation Program, Ottawa, ON, Canada.
Chen, G., Pham, T.T., 2001. Introduction to Fuzzy Sets, Fuzzy Logic and
Fuzzy Control Systems. CRC Press, Boca Raton, FL.
Chen, S.J., Hwang, C.L., Beckmann, M.J., Krelle, W., 1992. Fuzzy
Multiple Attribute Decision Making: Methods and Applications.
Springer, New York.
Chen, S.Q., 2000. Comparing Probabilistic and Fuzzy Set Approaches for
Design in the Presence of Uncertainty. Ph.D. Dissertation, Virginia
Polytechnic Institute and State University, Blacksburg, Virginia.
Chen, Z., Huang, G.H., Chakma, A., 2003. Hybrid fuzzy-stochastic
modeling approach for assessing environmental risks at contaminated
groundwater systems. Journal of Environmental Engineering 129,
79–88.
Cheng, S., 2000. Development of a Fuzzy Multi-Criteria Decision Support
System for Municipal Solid Waste Management. M.Sc. thesis,
University of Regina, Regina, Canada.
Dahab, M.F., Lee, Y.W., Bogardi, I., 1994. A rule-based fuzzy-set
approach to risk analysis of nitrate-contaminated groundwater. Water
Science and Technology 30, 45–52.
Destouni, G., 1992. Predicting uncertainty in solute flux through
heterogeneous soil. Water Resources Research 28, 793–802.
Finley, B.L., Paustenbach, D.J., 1997. Using applied research to reduce
uncertainty in health risk assessment: five case studies involving human
exposure to chromium in soil and groundwater. Journal of Soil
Contamination 6, 649–705.
Foussereau, X., Graham, W., Aakpoji, A., Destouni, G., Rao, P.S.C.,
2000. Stochastic analysis of transport in unsaturated heterogeneous
soils under transient flow regimes. Water Resources Research 36,
911–921.
Goodrich, M.T., McCord, J.T., 1995. Quantification of uncertainty in
exposure assessments at hazardous waste sites. Ground Water 33 (5),
727–732.
Guyonnet, D., Bourgine, B., Dubois, D., Fargier, H., Come, B., Chiles,
J.P., 2003. Hybrid approach for addressing uncertainty in risk
assessments. Journal of Environmental Engineering 129, 68–78.
Hamed, M.M., Bedient, P.B., 1997. On the performance of computational
methods for the assessment of risk from ground-water contamination.
Ground Water 35 (4), 638–646.
James, A.L., Oldenburg, C.M., 1997. Linear and Monte Carlo uncertainty
analysis for subsurface contaminant transport simulation. Water
Resources Research 33, 2495–2508.
Kentel, E., Aral, M.M., 2004. Probabilistic-fuzzy health risk modeling.
Stochastic Environmental Research and Risk Assessment 18, 324–338.
Kentel, E., Aral, M.M., 2005. 2D Monte Carlo versus 2D fuzzy Monte
Carlo health risk assessment. Stochastic Environmental Research and
Risk Assessment 19, 86–96.
Labieniec, P.A., Dzombak, D.A., Siegrist, R.L., 1997. Evaluation of
uncertainty in a site-specific risk assessment. Journal of Environmental
Engineering 123, 234–243.
Lee, L.J.H., Chan, C.C., Chung, C.W., Ma, Y.C., Wang, G.S., Wang,
J.D., 2002. Health risk assessment on residents exposed to chlorinated
hydrocarbons contaminated in groundwater of a hazardous waste site.
Journal of Toxicology and Environmental Health-Part A 65, 219–235.
Lee, Y.W., Dahab, M.F., Borgardi, I., 1994. Fuzzy decision making in
ground water nitrate risk management. Water Resources Research 30,
135–148.
Li, J.B., Huang, G.H., Chakma, A., Zeng, G.M., Liu, L., 2003. Integrated
fuzzy-stochastic modeling of petroleum contamination in subsurface.
Energy Sources 25, 547–563.
Liu, L., Cheng, S.Y., Guo, H.C., 2004. A simulation-assessment modeling
approach for analyzing environmental risks of groundwater contam-
ination at waste landfill sites. Human and Ecological Risk Assessment
10, 373–388.
Maxwell, R.M., Pelmulder, S.D., Tompson, A.F.B., Kastenberg, W.E.,
1998. On the development of a new methodology for groundwater-
driven health risk assessment. Water Resources Research 34, 833–847.
Maxwell, R.M., Kastenberg, W.E., Rubin, Y., 1999. A methodology to
integrate site characterization information into groundwater-driven
health risk assessment. Water Resources Research 35, 2841–2856.
Mills, W.B., Johnson, K.M., Liu, S., Loh, J.Y., Lew, C.S., 1996.
Multimedia risk-based soil cleanup at a gasoline-contaminated site
using vapour extraction. Ground Water Monitoring and Remediation
16, 168–178.
Minsker, B.S., Shoemaker, C.A., 1998. Quantifying the effects of
uncertainty on optimal groundwater bioremediation policies. Water
Resources Research 34, 2626–3615.
Mohamed, A.M.O., Cote, K., 1999. Decision analysis of polluted sites—a
fuzzy set approach. Waste Management 19, 519–533.
Piver, W.T., Duval, L.A., Schreifer, J.A., 1998. Evaluating health risks
from ground-water contaminants. Journal of Environmental Engi-
neering 124, 475–478.
UTA (University of Texas at Austin), 2000. Technical Documentation for
UTCHEM-9.0: A Three-Dimensional Chemical Flood Simulator.
Reservoir Engineering Research Program, Center for Petroleum and
Geosystems Engineering, University of Texas at Austin, TX.
USEPA (US Environmental Protection Agency), 1989. Risk Assessment
Guidance for Superfund, vol. I: Human Health Evaluation Manual
(Part A). US Environmental Protection Agency, Office of Emergency
and Remedial Response, EPA 540/1-89/002, Washington, DC.
USEPA (US Environmental Protection Agency), 1992. Guidelines for
Exposure Assessment. USEPA 600Z-92/001, 29 May 1992, US
Environmental Protection Agency, Risk Assessment Forum, Washing-
ton, DC.
USEPA (US Environmental Protection Agency), 2003. Integrated
Risk Information System: Xylene (CASRN 1330-20-7). http://www.
epa.gov/iris/subst/0270.htm, accessed on May 10, 2003.
Wagner, J.M., Shamir, U., Nemati, H.R., 1992. Groundwater quality
management under uncertainty: stochastic programming approaches
and the value of information. Water Resources Research 28,
1233–1246.