You are on page 1of 14

Journal of Environmental Management 91 (2010) 11691182

Contents lists available at ScienceDirect

Journal of Environmental Management


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

Risk-based economic decision analysis of remediation options at


a PCE-contaminated site
Gitte Lemming a, *, Peter Friis-Hansen b, Poul L. Bjerg a
a
b

Department of Environmental Engineering, Technical University of Denmark, Miljoevej, building 113, DK-2800 Kgs. Lyngby, Denmark
Det Norske Veritas (DNV), Veritasveien 1, 1322 Hvik, Norway

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 21 November 2008
Received in revised form
18 December 2009
Accepted 10 January 2010
Available online 1 February 2010

Remediation methods for contaminated sites cover a wide range of technical solutions with different
remedial efciencies and costs. Additionally, they may vary in their secondary impacts on the environment i.e. the potential impacts generated due to emissions and resource use caused by the remediation activities. More attention is increasingly being given to these secondary environmental impacts
when evaluating remediation options. This paper presents a methodology for an integrated economic
decision analysis which combines assessments of remediation costs, health risk costs and potential
environmental costs. The health risks costs are associated with the residual contamination left at the site
and its migration to groundwater used for drinking water. A probabilistic exposure model using rst- and
second-order reliability methods (FORM/SORM) is used to estimate the contaminant concentrations at
a downstream groundwater well. Potential environmental impacts on the local, regional and global
scales due to the site remediation activities are evaluated using life cycle assessments (LCA). The
potential impacts on health and environment are converted to monetary units using a simplied cost
model.
A case study based upon the developed methodology is presented in which the following remediation
scenarios are analyzed and compared: (a) no action, (b) excavation and off-site treatment of soil, (c) soil
vapor extraction and (d) thermally enhanced soil vapor extraction by electrical heating of the soil.
Ultimately, the developed methodology facilitates societal cost estimations of remediation scenarios
which can be used for internal ranking of the analyzed options. Despite the inherent uncertainties of
placing a value on health and environmental impacts, the presented methodology is believed to be
valuable in supporting decisions on remedial interventions.
2010 Elsevier Ltd. All rights reserved.

Keywords:
Decision support
Remediation
Contaminated sites
Groundwater
Life cycle assessment
Chlorinated solvents
Health risk
Uncertainty modeling
First- and second-order reliability methods

1. Introduction
The management of contaminated sites as undertaken by
municipal, regional or national authorities is not solely a matter of
whether or not a site is contaminated and if a site should be
remediated but also how the site should be remediated. Decisionmakers are often faced with a broad range of different technical
approaches for site cleanup, including biological, chemical and
physical (thermal) technologies that can be implemented either ex
situ or in situ. Efciency and cleanup times may vary substantially
between remediation technologies as may associated costs and
environmental impact of each method. While remediation of
a contaminated site reduces a local contamination problem, other
environmental impacts are also created on global, regional and

* Corresponding author. Tel.: 45 4525 1595; fax: 45 4593 2850.


E-mail address: gil@env.dtu.dk (G. Lemming).
0301-4797/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jenvman.2010.01.011

local scales as a consequence of the energy and materials consumed


by the remediation process itself. Such impacts can be quantied
using life cycle assessment (LCA), which has been suggested as
a valuable decision support tool for comparing the environmental
impacts arising from remedial activities (Lemming et al., in press;
Cadotte et al., 2007; Bayer and Finkel, 2006; Toffoletto et al., 2005).
The risk-based cost-benet approach to aid decision-making
within the eld of hydrogeology was described by Freeze et al.
(1990) and further exemplied for subsurface remediation by
Massmann et al. (1991). The objective function (Eq. (1)) describes
the net present value, 4, of a decision alternative as a function of
annual benet, B, annual cost, C, and a risk term, R, covering the
expected cost of failure. T is the time span of the analysis and r is the
discount rate (Freeze et al., 1990):

T
X
t 0 1

1
t

h
i
Bt  Ct  Rt

(1)

1170

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

If the included benets, risks and costs reect the preferences of


the decision maker, then the optimal decision is one which maximizes the objective function. The risk term is dened as the probability of failure, Pf, (or generally the frequency of failure)
multiplied by the cost of failure Cf:

Rt Pf t$Cf t

(2)

The risk-based cost-benet framework has been used previously


to evaluate groundwater remediation strategies (e.g. Khadam and
Kaluarachchi, 2003; Rosen et al., 1998), however neglecting benet
evaluation and simplied to a risk-cost minimization model:

T
X
t 0 1

1
r

i
Ct Rt

(3)

Rosen et al. (1998) used this framework to evaluate two alternatives to reduce groundwater impacts from diffuse nitrate and
aluminium sources and monetized the groundwater contamination
risk based on the water price. Khadam and Kaluarachchi (2003)
used the risk-cost-benet approach to evaluate the optimal
pumping period for a pump-and-treat system, and expressed the
cost-effectiveness in terms of remediation costs per life saved.
Other decision support frameworks for contaminated site
management include multi-criteria utility theory, as used by Scholz
and Schnabel (2006), to calculate the overall utility of a decision
alternative. This is based on the sum of four partial utilities which
describe human health impacts, remediation costs, soil productivity and market value of land after remediation. The utilities were
normalized to absolute values between zero and one, with a single
utility score of 1 representing the most favorable outcome, i.e. no
health effects or a remediation cost of zero. In this way, the authors
overcome the difculties in monetizing the various impacts of the
remediation alternatives. Nevertheless the method implies an
internal weighting between the four utilities, and the nal utility
sum can be difcult to interpret. The decision support system by
Nasiri et al. (2007) aimed to assess the compatibility of groundwater remediation technologies based on characteristics of the
contaminated zone. Fuzzy sets theory was used to include uncertainty of linguistic data in their multiple attribute decision analysis.
In brief, none of the reviewed decision frameworks include
external environmental costs in their analyses, such as externalities
due to air emissions. These external societal costs should ideally be
zdemiroglu, 2005).
a part of a societal analysis (Hardisty and O
Studies which apply LCA to evaluations of environmental impacts
from remediation express these impacts in terms of an emission of
a reference substance (e.g. kg of CO2-equivalents for impacts on
climate change). The impacts may be normalized (e.g. expressed as
person equivalents, PE) and weighted to a single impact index but

are not monetized and cannot directly be used in an economic


decision analysis.
This article presents a holistic methodology for combining
assessments of health risk and environmental impacts associated
with remediation of a contaminated site. The method applies
a probabilistic risk assessment using rst- and second-order reliability methods (Ditlevsen and Madsen, 1996) to estimate the
human health risk and cost associated with the residual contamination and the potential ingestion of contaminated drinking water.
Environmental impacts caused by the remedial activities are evaluated using life cycle assessments (LCA) of each remediation
scenario. The human health risk cost is estimated using the Life
Quality Time Allocation Index (Ditlevsen and Friis-Hansen, 2007b)
and the environmental cost is estimated using a preliminary
environmental cost model. Together with the direct remediation
cost of each method, the health cost and the environmental costs
combines into a relative cost estimate of each remediation scenario
that can be used for an internal ranking of the assessed remediation
technologies. The methodology is illustrated in Fig. 1. Human health
costs estimated in the methodology are associated with the
potential ingestion of contaminated drinking water. Other exposure
routes such as exposure via indoor air or via soil are disregarded in
this analysis for simplicity purposes.
2. Materials and methods
2.1. Site description
The risk-based economic decision analysis presented in this
article is used to assess three alternatives for remediating a site in
Denmark which is heavily contaminated with chlorinated ethenes.
The contaminant source zone lies in an unsaturated and possibly
fractured clayey till which extends approximately 7.5 m below
ground surface. The top 1 m of soil is regarded as unpolluted. The
mass of chlorinated solvents is estimated at 10 tons and primarily
in the form of tetrachloroethene (PCE). A signicant part of this
contamination is present as an immobile separate phase trapped as
blobs within the soil pores (residual phase contamination). The
clayey till overlies a sand layer which is saturated from approximately 20 m below ground surface. A thin clayey till layer separates
the upper aquifer from the regional chalk aquifer from where
drinking water is abstracted. The protective clay layer diminishes in
the direction towards the water supply. The contaminated site is
situated within the groundwater catchment of the largest Danish
well eld located approximately 2000 m down gradient from the
site. Groundwater is the sole source of drinking water in this area
and therefore represents a scarce and valuable resource. A
conceptual model of the site is presented in Fig. 2.

Fig. 1. Overview of the framework for the integrated analysis of remediation costs, environmental costs and health costs of contaminated site remediation.

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

1171

1. Excavation and off-site treatment


2. Soil vapor extraction (SVE)
3. Thermally enhanced soil vapor extraction by electrical heating
of the soil

Fig. 2. Conceptual model of the contaminated site (not to scale).

2.2. Calculating the probability of an event: rst- and second-order


reliability methods
First- and second-order reliability methods (FORM/SORM) are
analytical models developed in the eld of structural engineering to
calculate the probability of failure of engineered structures. They
have recently been applied to address uncertainty within
contaminant transport modeling by e.g. Hamed and El-Beshry
(2006) and Baalousha and Kongeter (2006). Both methods involve
dening a limit state function GX that divides the set of function
values into a failed set GX  0 and a safe set GX > 0. The failure
probability as dened in Eq. (4) is (Ditlevsen and Madsen, 1996):

h
i
P GX  0

f xdx

(4)

GX0

where f x is the multivariate density function of the random


variable X of dimension n. Note that capital letters refer to random
variables and that lower case letters refer to a specic outcome of
the capital lettered random variable.
Standard numerical integration techniques are generally not
feasible to solve the high-dimensional integral in Eq. (4) (large n). In
general, either Monte Carlo simulation (MCS) or analytically based
rst- and second-order reliability methods (FORM/SORM) must
therefore be used. It is straightforward to apply MCS to obtain the
cumulative density function, FGX gx. However, it can be a drawback that MCS may require a signicant amount of function calls to
evaluate the possible complex function GX, especially for small
probability levels. FORM/SORM are analytical probability integration methods and thus, when they do apply, are quite fast. It should
be noted that FORM and SORM apply only to random variable reliability problems where the set of basic variables are continuous.
An important advantage of the FORM/SORM method compared
to MCS is the importance factors (often called the a-vector), which
are by-products of the calculation. These describe how much the
individual random variables in the X-vector contributes to the
uncertainty in the probability calculation. If the importance factor of
a variable is below 1014%, the variable may be changed to its mean
value alone without any signicant loss in accuracy. This is known as
sensitivity omission (Madsen et al., 1986). The four steps in a probability computation by FORM/SORM are presented in Appendix I.
3. Remediation methods and costs
Three main technologies are considered for remediating the
contaminated clayey till formation. These have been outlined in
a sketch project by the consulting company Kruger:

The rst option involves the active removal of contaminated soil


volume by excavation. The soil is then transported to a soil treatment facility for ex situ aeration. The extracted PCE-contaminated
air is collected and treated with activated carbon. The excavation pit
is backlled using gravel from a local gravel pit. In option 2, soil
vapor extraction (SVE), the contamination is left in the ground and
the aim is to prevent contaminant migration to the groundwater
aquifer by extracting soil vapor from the unsaturated sand below
the source and pumping of groundwater from the upper shallow
aquifer. An SVE system has already been established at the site as
a temporary solution until a permanent solution is chosen. The SVE
option provides a very slow removal of contaminant mass from the
source due to a slow release from the low-permeability clay till
layer. A minimum operating time of 30 years is anticipated in the
project outline. Finally in option 3, the extraction of soil vapor is
combined with electrical heating of the soil using heating elements
submersed into steel cased wells (in situ thermal desorption). By
heating the soil to above 100  C, a fast volatilization of chlorinated
solvents and thus a high removal rate is achieved. The in situ
treatment period is thereby anticipated to be reduced to approximately 10 months of heating. Extracted PCE vapors from the SVE
systems in options 2 and 3 will be cleaned on-site by adsorption to
activated carbon before being released to the atmosphere.
Table 1 summarizes the remediation costs for each of the
considered remediation methods. Excavation is estimated to be the
most expensive solution, and the budget for this method includes
the cost of expropriating two resident houses within the excavated
area. Soil vapor extraction is estimated as the cheapest option; the
remediation costs were estimated based on a 30-year operation
period.
4. Health cost model
4.1. Probabilistic contaminant exposure model
The applied limit state function represents the difference in
a certain target concentration level Ctarget and the 30-year average
contaminant concentration Cd,av in drinking water produced at the
down gradient well eld:

GX Ctarget  Cd;av Ctarget 

1
T

T
0

Cd tdt

T
0


Ctarget  Cd t fT tdt

(5)

Table 1
Remediation costs estimated by Kruger. All costs are in million Euro (MEuro, 2007
prices). No discounting of costs is applied. Prices were converted from DKK 2005
values by updating to 2007 prices using Denmark Consumer Price Index (OECD Stat
Extracts) and converted to Euro using Purchasing Power Parities (OECD Stat
Extracts).
Phase

Excavation

Soil vapor
extraction

Electrical
heating

Project design phase


Establishment phase
Operation and maintenance phase
Dismantling phase

0.04
3.2
0
0

0
0.06
1.8
0.01

0.16
1.2
1.1
0.08

Total remediation costs

3.2

1.9

2.5

1172

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

T 30 years is the considered time horizon, and fT t is the


density function for time, modeled as uniformly distributed in the
interval 0; T1. The failure probability PGX  0 thus represents
the probability of exceeding a certain target concentration and is
calculated for a range of target concentration levels. Cd(t) is the time
dependent contaminant concentration in the drinking water
calculated as described below.
A subsurface system consisting of a contaminant source located
in an unsaturated soil layer is considered. The ux of contaminant
mass, J(t) (g yr1), leaving the source zone is modeled as the
downward advective mass ux following the approach in Troldborg
et al. (2008):

Jt Cw t$I$A

(6)
2

1

where I (m yr ) is the inltration rate, A (m ) is the source zone


area perpendicular to the ow and Cw(t) (g m3) is the time
dependent solute contaminant concentration in the source zone.
Source mass is depleted with time due to dissolution in inltrating
water and the change in contaminant mass in the 3-phase system
(air, soil and water). This is expressed as (Troldborg et al., 2008):

dM
dCw
I$A$Cw t
Vqw Kd rb qa KH
dt
dt

(7)

where qw and qa describe the fraction of the pore volume occupied


by water and air respectively and Kd (L kg1) and KH (dimensionless) are the soil-water and the air-water partitioning coefcients
respectively. rb (kg L1) is the bulk density of the soil and V is the
total soil volume (m3).
By solving Eq. (7), an exponentially declining solute source
concentration is obtained:

Cw t Cw $explt
where l

(8)

I
Dqw Kd rb qa KH

Cw (g m3) is the initial contaminant concentration and D (m) is


the depth of the contaminated zone. The time dependent
contaminant ux then becomes:

Jt I$A$Cw $explt

(9)

For simplication, it is assumed that no degradation occurs in


the source zone as the contaminant in focus, PCE, is persistent in
the aerobic environment within this specic case (see Site
description, Section 2.1).
For areas assessed to contain contamination as a separate or
residual phase, the initial solute concentration is assumed to be
equal to the aqueous solubility of the contaminant, S (g m3). The
total mass ux is then calculated as the sum of the mass ux from
the hotspot area (J2) and the ux from the surrounding lighter
contaminated soil volume (J1):

Jtotal t J1 t J2 t I$A$explt$ 1  x$Cw1 x$Cw2

(10)
where x denotes the fraction of the total area occupied by residual
phase contamination. For the no action scenario, Cw is the initial
solute concentration, whereas in the remediation scenarios Cw is
reduced as a consequence of the remedial effort. The resulting
solute concentrations after remediation are estimated from the
expected efciencies (effi) of the remediation technologies to
remove the source or prevent leaching of contaminant mass to the
saturated zone:

1
This modeling adjustment allows averaging out to be directly within the FORM/
SORM analysis by extending it one additional random variable, fT(t).

Cw;after Cw;initial 1  effi

(11)

The remedial effect is assumed to occur instantaneously.


The worst case contaminant concentration in the down gradient
supply well Cp(t) (g m3) is found by dividing the contaminant mass
ux, J(t) (g yr1) by the pumping rate, Qp, (m3 yr1) at the well eld
(Einarson and Mackay, 2001). This concentration expresses the
theoretical maximum concentration in the abstracted water
assuming that the plume is fully captured by the well and that no
degradation occurs:

Cp t

Jt
Qp

(12)

Conventional treatment of groundwater for drinking water


supply in Denmark involves aeration and sand ltration. Assuming
that a cascade aerator is used, it can be expected that approximately 70% of the chlorinated solvents are stripped to the atmosphere (Arvin, 1992). Thus a stripping potential, Ps, of 0.7 is
included in the analysis and the resulting drinking water concentration Cd(t) is:

Cd t 1  Ps $Cp t

(13)

4.2. Input data and uncertainty modeling


The variability and uncertainty in input data is addressed in the
FORM/SORM analysis by representing variables as statistical
distributions instead of deterministic values. For this purpose
probability distributions of the beta type as well as the log-normal
type are used. The beta distribution is dened by its upper and
lower boundaries (a, b) and the two shape parameters (r, s). This is
advantageous since it makes it possible to specify minimum and
maximum limits (a, b) for the distribution, whereas the log-normal
distribution is dened from zero (or a possible offset) to innity.
The beta distribution was used for variables for which more information was available from the site characterization or for which
a physical upper and lower boundary exist e.g. fractions such as
porosity and water content.
The beta shape parameters were determined by tting the
distribution to observation data or based on expert judgment of
mean, standard deviation and upper/lower limits. For a beta
distribution with shape parameters (r, s) and minimum and
maximum limits (a, b) the corresponding mean, m, and standard
deviation, s, are (Johnson et al., 1995):

 r 
rs

m a b  a$

s b  a$

 r  r
s
$
rs
r$r s 1

(14)

(15)

Table 2 presents the site-specic input parameters used in


uncertainty modeling of contaminant concentrations in drinking
water. The initial pore water concentration in areas without
residual phase contamination is represented by a probability
distribution of the beta type with a minimum value of 0 and
a maximum of 227 g/m3 (equal to the aqueous solubility of PCE at
10  C). Mean and standard deviation of the distribution were
determined by tting observation data to the beta distribution
using the method of least squares (see Fig. 3).
Table 3 presents the beta distributions applied for the uncertainty modeling of the remedial efciencies of the three remediation techniques. Compilations of remedial efciency data in the
literature is however scarce. For the SVE method no compilation of
efciencies could be found in the literature. Instead the average

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182


Table 2
Uncertainty modeling of the site-specic variables. For the log-normal distribution
the mean and standard deviation are given in brackets [m,s]. For the beta distributions the values in parentheses represent the two shape parameters and the upper
and lower limit of the parameter respectively (r, s, a, b). The corresponding mean and
standard deviation of the beta distribution are given in brackets [m,s].
Variable

Symbol Unit

Inltration
Contaminated area
Fraction of area with
residual phase
contamination
Depth of contaminated
zone
Initial solute source
concentration of PCE
Porosity in source zone

I
A
x

Water saturation in
unsaturated zone
Bulk density of soil
Fraction of organic carbon
Pumping rate

Distribution Parameters

m yr1 Log-normal
m2
Log-normal
Decimal Beta
fraction

[0.15, 0.04]
[1150, 150]
(17.2, 40.1, 0, 1)
[0.3, 0.06]

(3, 2, 5, 7.5)
[6.5, 0.5]
(0.24, 1.4, 0, 227)
[33, 49]
(5.1, 5.1, 0.2, 0,5)
[0.4, 0,08]
(55.5, 18.5, 0, 1)
[0.75, 0.05]
[1.6, 0.1]
[0.0053, 0.0024]
6  106

Beta

Cw,initial g m3

Beta

Vol/vol

Beta

Sat

Decimal
fraction
kg L1
g kg1
m3 yr1

Beta

rb
foc
Qp

Log-normal
Log-normal
Fixed

1173

Table 3
Uncertainty modeling of remediation efciencies. For the beta distributions the
values in parentheses represent the two shape parameters and the upper and lower
limit of the parameter respectively (r, s, a, b). The corresponding mean and standard
deviation are given in brackets [m,s].
Variable

Symbol

Unit

Distribution

Parameters

Remediation efciency
excavation
Remediation efciency
soil vapor extraction
Remediation efciency
electrical heating

effexc

Decimal
fraction
Decimal
fraction
Decimal
fraction

Beta

(2.9, 0.03, 0, 1)
[0.98, 0.04]
(6.38, 0.61, 0, 1)
[0.91, 0.1]
(10.6, 0.33, 0, 1)
[0.97, 0.05]

effSVE
effheat

Beta
Beta

equilibrium concentration in the water phase, Cw, results in


concentrations above the aqueous solubility of PCE:

Cw

CT rb
qw Kd rb qa KH

(16)

4.3. Contaminant exposure results from the FORM/SORM analysis


efciencies for the similar air sparging method from Bass et al.
(2000) was assumed to represent a rough estimate for the mean
efciency of the SVE method (91%). The average remediation efciency of 97% found in McGuire et al. (2006) was used to represent
the electrical heating method. This efciency is however based on
thermal remediation using steam injection. The excavation method
mean efciency was assumed to be 98%. Compound-specic
physicalchemical parameters for PCE are presented in Table 4. Due
to the generally low variability in this type of data compared to
other parameters, these were modeled as xed values.
The extent of residual phase contamination, x, is estimated
based on the fraction of total samples (water or soil) that indicates
presence of residual phase contamination. For water samples, PCE
concentrations greater than 1% of the aqueous solubility are
assumed to indicate the presence of residual/separate phase
contamination (US EPA, 1992). For soil samples with a total
contaminant concentration of CT (mg/kg soil), residual phase
contamination is assumed to exist if the calculated 3-phase

The 30-year average contaminant concentration in drinking


water (equation (5)) was calculated for each scenario using FORM/
SORM. The resulting cumulative probability distributions of the
contaminant concentrations are seen in Fig. 4. The curves were
tted to a probability distribution of the gamma type using the
method of least squares, and the mean value and standard deviation of the distribution were determined. The mean contaminant
concentrations in drinking water range from 0.03 mg/L (excavation
scenario) to 18 mg/L (no action scenario), see Table 5. All remediation scenarios have mean values below the current Danish
Maximum Concentration Limit for PCE in drinking water of 1 mg/L.
The gamma distributions were used as input to the health cost
calculation, see Section 4.4.
As earlier described, the importance factors are a very important
by-product of the FORM/SORM analysis. They dene how much
each individual variable contributes to the total uncertainty of the
result. A low importance factor (approx. < 10%)2 implies that the
variable without signicant error may be considered to be deterministic. Importance factors of each variable are illustrated in Fig. 5
as a function of the resulting 30-year average concentration. In the
no action scenario and for small values of Cd,av, the inltration rate
followed by the fraction of residual phase contamination and the
size of the contaminated area are the variables with highest
importance for the result. With increasing concentration levels, the
initial contaminant concentration becomes the most important
parameter. In all three remediation scenarios, the remediation
efciency is by far the most dominating variable for the entire
range of values of the PCE concentration in drinking water. The
initial concentration has some inuence at higher concentration
levels. The remaining variables all have (summed) importance
factors below 10%, which implies that they all, without loss of
accuracy of the cumulated density function of Cd,av, may be replaced
by their deterministic mean value.
4.4. Probabilistic health cost model
4.4.1. Health risk model
The average daily dose, ADD, (mg kg1 d1) of PCE from intake of
drinking water with an average contaminant concentration of Cav

Fig. 3. Cumulative probability of observations of initial solute source concentrations,


Cw,initial, and of the beta distribution tted to the observations. The probability density
function of the beta t is plotted on the secondary axis.

2
Note that if several variables each have an importance factor of 10%, these of
course cannot all be removed.

1174

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

Table 4
Compound-specic parameters for PCE.
Variable

Symbol

Unit

Distribution

Parameter

Log octanol-water partioning


coefcient a
Henrys law constant (10  C)b

LogKow

Fixed

2.71

Fixed

0.32

Aqueous solubility (10  C)c


Stripping potential of PCE
during drinking water
aeration in waterworks

S
Ps

Dimension
less
Dimension
less
g m3
Decimal
fraction

Fixed
Fixed

227
0.70

KH

Table 5
Mean and standard deviation of the 30-year average PCE concentrations (Cd,av) as
found by tting the FORM/SORM results to a gamma distribution.

No action
Excavation
Soil vapor extraction
Electrical heating

Mean PCE concentration


(mg L1)

Standard deviation
(mg L1)

1.80
0.03
0.12
0.04

0.63
0.08
0.20
0.11

Average value from literature study in Schaerlaekens et al. (1999).


Calculated for 10  C using regression formula from Gosset (1987).
c
The vapor pressure, p (kPa), at 10  C is estimated with Antoines equation for
PCE: log p AB/(t C), where t is temperature ( C), A 6.1017, B 1386.9 and
C 217.52 (Riddick et al., 1986). The aqueous solubility (mol L1) is calculated from
the vapor pressure (atm) at 10  C and Henrys law constant as: S p/(KH R T), where
R is the gas constant (0.0821 L atm mol1 K1) and T is the temperature (K).
b

(mg L1) during the averaging time, AT (yr), is estimated as


(Maxwell et al., 1998):

ADD Cav $


IR ED$EF
BW AT

(17)

IR/BW (L kg1 d1) is the ingestion rate per body weight, ED is


the exposure duration (yr) and EF is the exposure frequency
(d yr1). The health-risk-specic parameters used for estimating
the average daily dose were adopted from Maxwell et al. (1998) and
can be seen in Table 6. The 30-year average exposure concentration
in each scenario was represented by gamma type distributions (cf.
Section 4.3).
The incremental excess lifetime cancer risk IELCR due to exposure of an organic contaminant can be described with the one-hit
equation (US EPA, 1989), expressing the probability of an individual
developing cancer:

IELCR 1  expADD$CPF

(18)

CPF is the chemical specic cancer potency factor (kg d mg1).


The model assumes that exposure to any amount of a carcinogen
will increase the risk of cancer, i.e. there is no safe or threshold
dosage. For small cancer probabilities (P < 0.01) the linear version
of the equation is considered valid (US EPA, 1989):

IELCR ADD$CPF

(19)

Fig. 4. Cumulative probability distributions of the 30-year average PCE concentration


in drinking water calculated using FORM/SORM.

Fig. 5. Importance factors of variables for determining the 30-year average contaminant concentration of drinking water of each scenario. The mean value and the
cumulative density function (CDF) of the contaminant concentration have been plotted
together with the importance factors.

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182


Table 6
Probability density functions of health-risk-specic variables. For the log-normal
distribution the mean and standard deviation are given in brackets [m,s].
Variable

Symbol

Unit

Distribution

Parameters

Ingestion rate per


body weight
Exposure duration
Averaging time
Exposure
frequency
Cancer potency
factora
Exposed population

IR/BW

L kg1 d1

Log-normal

ED
AT
EF

yr
d
d yr1

Fixed
Fixed
Fixed

[3.3  102,
1.3  102]
30
25 550 (70 yr)
350

CPF

kg d mg1

Fixed

0.051

Number of
people

Log-normal

[1.1  105,
1  104]

The provisional oral cancer slope factor calculated by US EPA (1988).

4.4.2. Health cost estimation


It is beyond rational reasoning to place a value on a human life in
good health. However, rational principles may be formulated to
guide this setting. Inspired by the so-called Life Quality Index,
Ditlevsen and Friis-Hansen (2007a,b) formulated the Life Quality
Time Allocation Index (LQTAI), which is a social indicator that
serves the purpose of allocating a reasonable part of a countrys
Gross Domestic Product (GDP) to life-saving initiatives. Human life
value is assessed by requiring LQTAI invariance, implying that any
activity that reduces the life in good health must be balanced by an
equivalent increase in societal productivity, such that life quality
does not decrease by the activity.
In risk and decision analysis the LQTAI can be used to obtain
loss values associated with fatalities and injuries. The empirical
model was tted to a time series of working hours and
productivity for Denmark and 18 other OECD countries (Ditlevsen and Friis-Hansen, 2007b), and model tted constants (c, pmin)
representing the country-specic equilibrium between working
time and leisure time were established based on these. The tted
coefcients show limited variation the 18 evaluated OECD
countries.
According to the LQTAI model, the maximum productivity time,
DTp, that a country can allocate to averting a fatality or an injury is
calculated using the empirical expressions in Eqs. (20) and (21)
respectively (Ditlevsen and Friis-Hansen, 2007b):

DTpfatality

DTpinjury


r0 1
1 VL2 E$pmin
c 2



1 1
r  c
ET$pmin
log 0
c r0
pmin c

(20)

(21)

r0 denotes the fraction of life that can be converted to working time


and is set to 2/3 assuming that 1/3 of a persons life is occupied by
sleep. c (0.085) and pmin (1.81) are model tted constants for
Denmark. E is the life expectancy, which is set to 80 years here with
a coefcient of variation VL of 0.2. E[T] denotes the expected
duration of an illness period i.e. the experienced lifetime loss in
good health for at a person that survives cancer. Such data could not
be located; however a 2-year-period was assumed as an average
value. Using these values, a productivity time allocation of 591
years for averting a fatality and 48 years for averting an injury is
obtained. The allocated time values can be converted to a countryspecic monetary values by multiplying with the annual work
based salary, S. S is equal to the GDP divided by the equilibrium
productivity per working time, pmin. Hereby the incurred costs of
averting a fatality (ICAF) and an injury (ICAI), respectively, are
found. With a GDP of 33 340 Euro per capita for Denmark the
following values are obtained:

1175

33 340 Euro=yr
ICAF 591 yr$
10:9 MEuro
1:81
ICAI 48 yr$

33 340 Euro=yr
0:9 MEuro
1:81

These values are used to estimate health costs associated with


increased cancer probability in the population exposed to
contaminated drinking water. The model takes into account that
a person diagnosed with cancer will either die of the disease or
recover after a certain illness period. The general probability of
surviving cancer once diagnosed, psurv, has been set to 0.39, which
corresponds to the number of cancer incidences per year divided by
the number of deaths due to cancer per year. These data are based
on Danish statistics for all cancer types except skin cancer for the
period 19962000 (Danish Cancer Society, 2006). The health cost
associated with fatalities and illness periods due to the increased
probability of cancer is calculated based on the following expression, where N denotes the size of the exposed population:

Health cost IELCR$NICAF$1  psurv ICAI$psurv

(22)

The European Commission (2001) recommends a best estimate


value of 1.0 MEuro (year 2000 prices) for preventing a fatality. For
death as a result of cancer, this value is increased to 1.5 MEuro when
including the period of ill health. This value is applicable to deaths
in a principally elderly population where the reduction in life
expectancy is likely to be short (less than a year). If a carcinogenic
pollution affects a population characterized by a more average age,
then the Commission recommends adjusting for this by multiplying with a factor of 1.43. Table 7 presents the best estimate value
as well as the upper and lower range values calculated for a mortal
cancer incidence based on these guidelines. The best estimate value
from the European Commission will be used in a sensitivity
scenario.
4.5. Health cost results from the FORM/SORM analysis
Fig. 6 presents the baseline results of the health cost modeling
using a LQTAI value of 10.9 MEuro. By tting the results to gamma
type distributions, mean values and standard deviations were
determined and summarized in Table 8. The mean value of the
health cost for the no action scenario is 0.94 MEuro, whereas the
health costs of the remediation scenarios span from 0.016 MEuro
(excavation) to 0.062 MEuro (soil vapor extraction). The resulting
mean health costs when applying the lower ICAF value of 2.7
MEuro are a factor of 3.5 lower.
The importance factor plots in Fig. 7 show that for the no action
scenario the uncertainty in the PCE exposure concentration and
ingestion rate per body weight contribute signicantly to the result,
whereas in the remediation scenarios only the PCE concentration in
drinking water has a high importance. For very high health costs
the importance factor for the ingestion rate per body weight in the

Table 7
Recommended values by the European Commission (2001) for preventing
a cancer-related fatality for an average age population.

Best estimate value


Lower range value
Upper range value

2000 pricesa (MEuro)

2007 pricesb (MEuro)

2.1
1.4
5.4

2.7
1.8
6.9

a
Based on the values 1.5, 1.0 and 3.8 multiplied with a factor of 1.43 to adjust for
an average age population.
b
The value is updated from 2000 to 2007 prices using the OECD-Europe
Excluding High Ination Consumer Price Index. (OECD Stat Extracts).

1176

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

Fig. 6. The cumulative probability distributions of the incurred health costs for each
scenario.

SVE scenario increases, but still the uncertainty in the ingestion rate
is of limited inuence.
5. Environmental cost model
5.1. Life cycle assessment modeling
The primary goal of the remediation efforts is to reduce a local
contamination problem, but secondary effects or externalities to the
environment are costs that should be included in the overall cost
evaluation. Seen in a life cycle perspective, the remediation activities consume energy and other non-renewable resources and
causes emissions on local, regional and global scales. In order to
quantify and compare these impacts in a systematic and consistent
way, a life cycle assessment (LCA) of each remediation method is
conducted. The LCA result describes the aggregated potential
environmental impacts over the entire life cycle of each remediation technology from extraction of raw materials to production of
components and electricity to end-of-life disposal, reuse or recycling of utilized materials.
The life cycle assessment was modeled using the GaBi 4 LCA
Software and the EDIP unit process database combined with
additional data collection. The EDIP97 Impact Assessment methodology was chosen as impact assessment method (Environmental
Design of Industrial Products, Wenzel et al., 1997). As the LCA itself
is not the primary scope for this paper, the analysis is only briey
presented.
The goal of the LCA is to compare the environmental impacts of
the three different approaches for site cleanup. The compared
service (the functional unit in the LCA) is dened as the treatment
of the 7500 m3 of contaminated soil within a 30-year time frame.
The functional unit does not dene a cleanup level for the remediation. This implies that different levels of residual contamination
will be left in the ground depending on the evaluated method. The

Table 8
Mean health cost values of each scenario. Standard deviations are given in
parentheses.

No action
Excavation
Soil vapor extraction
Electrical heating

Health costs (MEuro)


ICAF of 10.9 MEuro

Health costs (MEuro)


ICAF of 2.7 MEuro

0.94
0.016
0.062
0.022

0.27
0.005
0.018
0.006

(0.49)
(0.043)
(0.11)
(0.055)

(0.14)
(0.012)
(0.030)
(0.016)

Fig. 7. Importance factors of variables for determining the health costs of each
scenario. The mean and the cumulative density function (CDF) of the health cost have
been plotted together with the importance factors.

consequences of the residual contamination are dealt with specifically in the health cost part of this framework.
The main operation activities included in the different remediation scenarios are presented in Table 9. Raw material extraction
and production of materials (steel, activated carbon, plastic materials, concrete etc.) and energy sources (electricity, diesel) are not
mentioned specically in Table 9 but are also included. Average
Danish electricity is assumed for on-site operations and off-site soil
treatment. As a common limitation, the construction of vehicles
such as excavators, drill rigs, trucks and cars are excluded as it is
assumed that only a minor part of their service life is ascribed to

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

this remediation case. Laboratory analyses of soil, water and air


samples were furthermore excluded due to their expected negligible impact.
The EDIP97 impact assessment includes a number of impact
categories on the global scale (global warming and ozone depletion) as well as on regional and local scales (photochemical ozone
formation, acidication, nutrient enrichment, human and ecotoxic
impacts and waste generation). In addition to this, the consumption
of nite resources is accounted for. Emissions and resource
consumption are normalized to person equivalents (PE) by dividing
with the impacts/resource use from an average person. The
normalized impacts are afterwards weighted based on politically
set targets for the reference year 2004. Resource consumption is
weighted according to the inverse supply horizon of each resource
type, i.e. consumption of crude oil has a higher weighting factor
than coal since the supply horizon is shorter. The weighted impacts
are expressed in the unit targeted person equivalents (PET) and the
resource consumption is expressed in person reserves (PR).
Appendix II lists all the applied normalization references and
weighting factors.

1177

Table 10
Unit cost for environmental impacts and resource consumption. The Stern Review
value is included for comparison only and is not used in the valuation.
Emission unit cost
Stern (2006)
Danish EPA (2007)

Euro/ton CO2
73
18

Resource unit cost

Euro/kg oil

Crude oil price (2007 average)

0.434

Euro/PETc
570
140
Euro/PRd
10 700

All Currency conversion was done using Purchasing Power Parities from OECD Stat
Extracts.
a
The value in Euro/ton CO2 represents the alternative costs of reducing CO2
emissions as estimated at 180 Danish kroner per ton CO2 by Danish EPA (2007).
b
1 PET (Targeted Person Equivalent) is equal to the average emission of 8.7 tons
CO2-eq./cap/yr divided by the weighting factor for global warming of 1.12
(Stranddorf et al., 2005).
c
The value in Euro/kg of crude oil is based on the 2007 world average oil price of
69.2 USD/barrel (EIA, 2008).
d
1 PR (person reserve) is equal to the average consumption of 588 kg oil/cap/yr
divided by a weighting factor of 0.024 (LCA Center, 2005).

of 108 Euro/ton and 0.506 Euro/kg respectively can however be


compared to the applied unit costs in this study.

5.2. Environmental cost estimation


Ideally each unit of weighted emissions (PET) should represent
the same environmental cost since they are based on politically set
weightings. Therefore an environmental cost is calculated
assuming the same unit cost per weighted impact. The Stern
Review (Stern, 2006) estimated a damage cost of 85 US Dollar (73
Euro) per ton of CO2 emitted if no action is done to control
greenhouse gas emissions. In this analysis a unit cost of 18 Euro per
ton of CO2 emission is used to estimate the environmental costs.
This value represents the estimated average market price of CO2
offsets (i.e. alternative cost of reducing CO2 emissions) from 2005
onwards and is suggested by the Danish EPA to be used in costbenet evaluations (Danish EPA, 2007). The environmental cost of
resource consumption is likewise estimated assuming that each
person reserve represents the same economic value. The average
2007 crude oil price is used as a basis for placing a value on the
resource consumption. Table 10 summarizes the applied unit costs.
It should be mentioned that life cycle impact assessment methods
including a monetary valuation step do exist e.g. Environmental
Priority System (EPS) by Steen (1999). However, EPS is based on an
endpoint impact assessment method and cannot be combined with
the midpoint LCA result obtained with the EDIP method. The EPS
environmental load units for CO2 emission and crude oil depletion

5.3. Environmental cost results


The weighted results from the life cycle assessments are seen in
Fig. 8 (environmental impacts) and Fig. 9 (resource consumption).
As expected, the environmental impacts and resource use proles
are very similar for soil vapor extraction and electrical heating,
since these methods are technically similar and both very electricity consuming. Conversely, the excavation method is highly
diesel consuming for excavation and transportation, which results
in higher toxic impacts and consumption of more crude oil than the
other methods. Soil for backlling of the excavation pit is the main
resource used in the excavation scenario.
Fig. 10 presents the summed weighted impact potential and
resource use of the three remediation methods. Electrical heating
has the highest aggregated impact load (2000 PET), but excavation
is only slightly lower at 1930 PET. Soil vapor extraction has both the
lowest impact load (1740 PET) and resource use (64 PR), whereas
excavation has the highest aggregated resource consumption
equivalent to 70 PR.
Converted to a cost using the unit costs for emissions of 140
Euro/PET and for resources of 10 700 Euro/PR (cf. Table 10), the
aggregated environmental costs sum to 1 MEuro for the excavation

Table 9
Main activities included in the three life cycle assessments.
Excavation

Soil Vapor Extraction

Electrical heating

On site:
 Excavation and backlling
 Demolition of 2 houses

On site:
 Drilling of wells
 Materials for wells, pumps and activated carbon units
 Electricity use for 30 y of soil ventilation, water
extraction and house ventilation
 Activated carbon use for water and air treatment

On site:
 Drilling of wells
 Materials for wells
 Materials for ventilation layer, isolation
and capping.
 Electricity use for 8 months of heating
 Activated carbon use for treatment
of extracted air

Off site:
 Construction of 2 houses
 Electricity for 1 y of soil aeration
 Materials for construction of treatment facilitya
 Activated carbon use for treatment
of extracted air
 Extraction of soil for backlling
Off site:
 Transport of soil to treatment (80 km) and to
nal disposal (40 km)
 Transport of materials, equipment
and people
a

Transport:
 Transport of materials, equipment and people

Transport:
 Transport of materials, equipment
and people

Only a part of the materials are ascribed to this remediation project based on the assumed treatment capacity and lifetime of the facility.

1178

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

Fig. 8. Weighted environmental impact potentials quantied in the life cycle assessment (PET: Targeted Person Equivalent).

and electrical heating methods and to 0.9 MEuro for the SVE
method (see Fig. 11).
6. Ranking of remedial alternatives
As remediation benets are not included in the evaluation, the
riskcostbenet model is reduced to a riskcost minimization
model. The time frame of the analysis is set to 30 years. No discounting of costs and benets is carried out. However, it can be
shown that a decision model with a nite time frame, T, and a discounting rate, r, of zero is decision theoretically equivalent to
a decision model with an innite time frame and a discounting rate
of 1/T, see Friis-Hansen and Ditlevsen (2003). Consequently, a 30year time frame as used in the present analysis is essentially equal
to applying an innite time frame and a discount rate of 1/
30 3.3%, which is similar to currently recommended discount
rates by Danish EPA (2007). The total cost associated with a remediation initiative, 4, is thus calculated as the sum of the remediation
costs and the risks in terms of environmental and health costs
(equal to Eq. (3) without discounting):

T h
i
X
Ct Rt

(23)

t0

Fig. 12a illustrates the results of the total cost estimates made for
the no action scenario and each of the remediation scenarios. The
total societal costs of the three remediation scenarios range from 2.9
MEuro (soil vapor extraction) to 4.3 MEuro (excavation). All

remediation scenarios have higher summed total societal costs than


the no action scenario which has an expected health cost of 0.9
MEuro. The remediation cost is in this case the most signicant
difference between the scenarios, because the health costs are low
and the environmental costs are of similar magnitude (see Table 11).
Ranking the methods based on their total societal costs places
soil vapor extraction rst, followed by electrical heating, and excavation is placed last. In the sensitivity scenario, Fig. 12b, the incurred
cost of averting a fatality was reduced from 10.9 MEuro to 2.7 MEuro.
This however does not alter the internal ranking of the methods
since the health costs are insignicant for the total cost result.
7. Discussion
The presented methodology combines assessment of remediation costs, health costs and environmental costs into an overall
societal cost comparison of different technical solutions for site
cleanup. Results can be used to support decision-making regarding
the selection of remedial strategies for a contaminated site. By
including both impacts on health and environment in the overall
cost estimate, the methodology enables a more complete comparison of remedial actions compares to the reviewed studies that
focused only on one of these issues.
7.1. Model approach
The chosen rst order model used to describe contaminant mass
ux to the aquifer over time is a relatively simple model of

Fig. 9. Weighted resource consumption quantied in the life cycle assessment (PR: person reserve).

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

1179

Fig. 11. Environmental cost of each remediation scenario.


Fig. 10. Sum of weighted impact potentials and sum of weighted resource consumption for each scenario.

a complex contaminant transport system. For a source zone in a low


permeable and possibly fractured media such as clay till, more
complex models could be applied to give a better description of the
mass input to the aquifer. Attenuation in the unsaturated zone and
aquifer is disregarded, possibly overestimating concentrations in
drinking water. Another simplication is that the remedial effect is
assumed to occur instantly in the abstracted groundwater at the
well eld, when in fact, this effect is projected in time.
The conducted life cycle assessments constitute the backbone of
the evaluation of externalities to the environment incurred by the
remedial actions. Life cycle assessments are elaborate evaluations
that quantify a wide spectrum of potential environmental impacts
in a systematic and comparable way. However, they are also associated with uncertainty both stemming from the data input and the
underlying impact assessment models. The applied impact
assessment method, EDIP97, was chosen as it has been a widely
applied method. It should be noted, however, that more recent
methods with more advanced impact assessment models are
available especially in regard to the modeling of human and
ecotoxic impacts e.g. the USEtox model (Rosenbaum et al., 2008).
In the present analysis a time frame of 30 years was chosen. It
should be noted, however, that this time frame may favor the soil
vapor extraction method, since the slow release of contaminants
from the low-permeability layer indicates that the treatment could
exceed 30 years. This would then result in higher environmental
costs. A more likely scenario of 100 years of soil vapor extraction
causes a signicant increase in both remediation and environmental costs, as seen in Fig. 12c and d.
Finally, any inconvenience and disruption (e.g. noise, dust,
esthetic deterioration of the landscape etc.) experienced by residents at or near the contaminated site was not included in the
analysis, but may be a decisive factor in the regulators decisionmaking. Site excavation and in situ electrical heating will cause
a high level of disruption for a relatively short period. Conversely,
the level of disruption experienced due to soil vapor extraction is
lower but will prevail for decades.
7.2. Importance of variables in the contaminant exposure model
The FORM/SORM procedure used to calculate the exposure
concentrations and expected health costs provides a fast and

functional way of generating probabilistic model results. Uncertain


variables were represented as continuous probability distributions
either of the log-normal or the beta type. The log-normal distribution is dened from zero to innity, whereas the beta distribution is bound by lower and upper values. Although somewhat
complex, the modeling of bounded variables using the beta distribution is preferred in this study because of the distributions high
degree of exibility.
The importance factors, which are easily calculated in FORM/
SORM, can be used for an easy identication of which uncertain
variables contribute the most to the uncertainty of the nal result.
Thus the analysis identies which variables should be investigated
further in order to reduce the uncertainty of the overall outcome.
In the present analysis it is found that the most important variables depend on whether a no action scenario or a remediation
scenario is considered. For the no action scenario, important
variable uncertainty is associated especially with the characterization of the source area represented by the source zone
contaminant concentration, the presence of residual phase
contamination and the size of the contaminated area. In addition,
the inltration rate to the aquifer is of high importance. In the
three remediation scenarios, the remediation efciency is by far
the single most dominating cause of uncertainty. Thus a better
description of remediation efciencies would be benecial for
future use of this model.
7.3. Economic valuation
The results of the analysis showed that the environmental costs
due to externalities from remediation were indeed signicant and
added 3050% to the total cost estimates of the remediation
scenarios. The applied valuation method for environmental
impacts and resource consumption was rather simplied assuming
a xed cost for each weighted unit of environmental impact (1 PET)
and each weighted unit of resource use (1 PR) respectively. The
applied unit cost for environmental impacts was based on the
expected CO2 offset market price. In comparison, damage estimates
presented in the Stern Review are a factor of 4 higher, which
suggests that this cost may be underestimated. Regarding resource
use, the market price of oil was used as a basis for the valuation of 1
person reserve. This might underestimate the total socioeconomic
value of a resource but serves as a best estimate. The signicant
contribution of the environmental costs to the total costs estimates

1180

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

Fig. 12. Remediation cost, environmental cost and health cost of each remediation scenario. All costs are in million Euro (MEuro). (a) Health costs based on an ICAF value of 10.9
MEuro. (b) Health costs based on an ICAF value of 2.7 MEuro. (c) Health costs based on an ICAF value of 10.9 MEuro, 100 year time frame for soil vapor extraction. (d). Health costs
based on an ICAF value of 2.7 MEuro, 100 year time frame for soil vapor extraction. The error bars mark the 5th and 95th percentile of the health cost distribution.

indicates that these are in fact important to include. Sensitivity


studies may show how sensitive the identied optimal solution is
to the chosen values.
The incurred cost of averting a fatality (ICAF) represents the
societal cost which can be justied for life-saving initiatives. In
the baseline scenarios, a value of 10.9 MEuro is applied based on
the Life Quality Time Allocation Index principle by Ditlevsen and
Friis-Hansen (2007a,b). This value in fact can be viewed as an
upper boundary value since it is based on the maximal
production time that a country can justify to allocate for the
prevention of a fatality. The lower value of 2.7 MEuro, as recommended by the European Commission (2001), represents
a more typical value applied in cost-benet evaluations. The
health costs due to ingestion of contaminated groundwater were
negligible compared to the total cost estimates of all scenarios
except for the no action scenario. Using the lower ICAF value did
therefore not affect the internal ranking of the three remediation
scenarios.
The analysis showed that remediation gives a signicant
reduction in the expected health costs associated with the
contaminated site. However, the total societal costs accrued in the
remediation scenarios are all larger than that of the no action

scenario. Due to the limitations in the analysis, the results obtained


cannot be considered as complete cost-benet comparisons since
benets were not included. It should furthermore be noted that for
simplicity purposes the health costs are only related to the ingestion of drinking water. Health effects associated with the contaminated soil e.g. via evaporation to indoor air and/or direct soil
contact could affect the cost analysis. Also the 30-year time
boundary used in the analysis favors the no action scenario since
the groundwater contamination will most likely persist for
centuries.
In the future, benets such as increased land value and the value
of maintaining good quality groundwater should be included if the
aim is to provide a full cost-benet evaluation for site cleanup. The
total economic value of an environmental good such as groundwater can be considered to consist of both a use value and a nonuse value (NRC, 1997). Little research exists on the non-use values
of groundwater but nonetheless indicate that the willingness to pay
for non-use benets is signicant. The fact that groundwater is
clean resource that can be passed on to future generations is
valuable even if the resource is not used presently (Hardisty and
zdemiroglu, 2005).
O
8. Conclusion

Table 11
Costs (MEuro) of remediation options. Values in parentheses are responding to
health costs calculated using an ICAF value of 2.7 MEuro.
No action

Excavation

Soil vapor
extraction

Electrical
heating

Remediation cost
Environmental cost
Health cost

0
0
0.9 (0.3)

3.2
1.0
0.02 (0.005)

1.9
0.9
0.06 (0.018)

2.5
1.0
0.02 (0.006)

Total cost

0.9

4.3

2.9

3.5

The presented methodology includes an economic evaluation


encompassing three aspects of site remediation: remedial costs,
external costs to the environment and health costs associated with
residual contamination left after remediation. The method allows
for an evaluation of the total societal costs associated with
a cleanup strategy for a contaminated site within a drinking water
catchment which is also compared with a no action scenario. The
health costs were evaluated using a 2-step probabilistic FORM/
SORM analysis in order to include variable uncertainty. This

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

analysis showed that the source characterization variables


(contaminant concentrations, presence of residual phase contamination and size of contaminated area) were of high importance for
the contaminant concentration levels in the drinking water well in
the no action scenario, whereas the remedial efciencies were the
single most important variable in the remediation scenarios. The
health costs associated with the ingestion of contaminated
drinking water were, however, only minor compared to the direct
remediation costs and the environmental costs. In the analyzed
case the most favorable remediation technique was soil vapor
extraction when a time boundary of 30 years was used. This time
boundary, however, favors remediation options with long time
frames, since operation, health and environmental costs occurring
after this period are omitted. If a more realistic time frame is used,
the soil vapor extraction method becomes less favorable and the
electrical heating method is the preferred option. The methodology
brings insight to the overall societal costs of a remedial strategy and
can facilitate more holistic decision-making that does not solely
focus on the direct cost of remediation but also incorporates
external costs to health and environment.
Acknowledgements
The authors wish to acknowledge E. Pfeilschifter, M.S. Mller
and E. Sgaard for their contribution to the data collection for the
conducted life cycle assessments and S.I. Olesen, DTU Management,
for technical support with the LCA software. C.B. Jensen (Capital
Region of Denmark) and J. Elkjr (Capital Region of Denmark, now
at Kbenhavns Energi) provided data for the case study and DTU
funded the Ph.D scholarship.

1181

space to u-space is exact. The one-to-one transformation, T, of


x-space on u-space maps the failure surface fxjGX 0g into
a
corresponding
surface
fujgU 0g.
Note
that
gu GTx.
2. identication of the most likely failure point. This point is
dened as the point of maximum density on the failure surface
in u-space and is obtained by solving the optimization problem
fminjuj; s:t: gu 0g. The optimal solution is in the most
likely failure point, u*, which is also called the design point.
3. approximation of the failure surface GX 0 in the u-space.
Since the identied design point describes the point in the uspace of maximum contribution to the failure probability, it
is natural to approximate the limit state around this point.
When using FORM the failure surface is approximated by
a tangent hyper plane through the u* and when using SORM
the failure surface is approximated by a second-order hyper
sphere.
4. computation of the failure probability corresponding to the
approximating failure surfaces. When using FORM the failure
probability is simply calculated as pf Fb, where F is
the standard normal cumulative density function and b ju*j
is the distance from the origin in the u-space to the design
point. For SORM results the integration becomes more
involved, see Ditlevsen and Madsen (1996) p. 104 Eq. 6.49.
Since the optimization is performed directly in the u-space
where the search is for the shortest distance from the origin to the
failure surface, the calculation time becomes independent of the
sought probability level. In general FORM gives good results for low
probability estimations, whereas it is recommended to use SORM
for mid-range probabilities.

Appendix I
Appendix II
It is not the intention to give a comprehensive summary of
FORM/SORM here, but only to refer to the main principles. Reference is left to Madsen et al. (1986) or Ditlevsen and Madsen (1996)
for an in-depth treatment.
A probability computation by FORM/SORM consists of four main
steps,
1. transformation of the random variables X into a standard
normal vector U by solving Fui Fxi , in which F$ is the
standard normal density function. This holds only for independent variables. It is worth emphasizing that FORM/SORM
are full distributional methods and the transformation from x-

The normalization references and weighting factors applied in


the life cycle assessments are presented in Tables A1 and A2
together with the associated reference region and year. Normalization references and weighting factors for environmental impacts
updated EDIP values are from Stranddorf et al. (2005) except for the
waste categories which are from Wenzel et al. (1997). Resource use
was normalized and weighted according to global values from 2004
(LCA Center, 2005). Normalization and weighting of the soil
resource is not included in the EDIP impact assessment methodology. Therefore Danish normalization references and weighting
factors from ScanRail Consult et al. (2000) were used.

Table A1
Normalization references and weighting factors for environmental impacts.
Impact category

Global warming
Ozone depletion
Acidication
Eutrophication
Photochemical oxidant potential
Ecotoxicity water chronic
Ecotoxicity water acute
Ecotoxicity soil chronic
Human toxicity air
Human toxicity water
Human toxicity soil
Bulk waste
Hazardous waste
Radioactive waste
Slags and ashes

Normalization reference

8700
0.103
74
119
25
352000
29100
964000
6.09  1010
52200
127
1350
20.7
0.035
350

Weighting factor
Unit

Reference region (year)

kg CO2-eq./cap/yr
kg CFC-11-eq./cap/yr
kg SO2-eq./cap/yr
kg NO3-eq./cap/yr
kg C2H4-eq./cap/yr
m3 water/cap/yr
m3 water/cap/yr
m3 soil/cap/yr
m3 air/cap/yr
m3 water/cap/yr
m3 soil/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr

Global (1994)
Global (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
EU15 (1994)
DK (1991)
DK (1991)
S (1989)
DK (1991)

Reference region (year)


1.12
63
1.27
1.22
1.33
1.18
1.11
1
1.4
1.3
1.23
1.1
1.1
1.1
1.1

Global (2004)
Global (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
EU15 (2004)
DK (2000)
DK (2000)
DK (2000)
DK (2000)

1182

G. Lemming et al. / Journal of Environmental Management 91 (2010) 11691182

Table A2
Normalization references and weighting factors for nite resources.
Resource

Aluminium
Chromium
Copper
Crude oil
Hard coal
Iron
Lignite
Manganese
Natural gas
Nickel
Uranium
Zinc
Sand, gravel

Normalization reference

4.52
0.83
2.27
604
602
97.7
264
1.72
353
0.22
0.0056
1.42
3306

Unit

Reference region
(year)

kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr
kg/cap/yr

Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
DK (1995)

Weighting factor
Reference region
(year)
0.0068
0.021
0.031
0.024
0.0080
0.0078
0.0039
0.029
0.015
0.023
0.010
0.041
0.0040

Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
Global (2004)
DK (2000)

References
Arvin, E., 1992. Udluftning Af Chlororganiske Forbindelser-Konsekvenser for Vandets Kvalitet. (Stripping of Chloroorganic Compounds-Consequenses for Water
Quality). In: Course in Water Supply, vol. 41. University of Aarhus, Danish Water
Association (in Danish).
Baalousha, H., Kongeter, E., 2006. Stochastic modelling and risk analysis of
groundwater pollution using FORM coupled with automatic differentiation.
Advances in Water Resources 29, 18151832.
Bass, D.H., Hastings, N.A., Brown, R.A., 2000. Performance of air sparging systems:
a review of case studies. Journal of Hazardous Materials 72, 101119.
Bayer, P., Finkel, M., 2006. Life cycle assessment of active and passive groundwater
remediation technologies. Journal of Contaminant Hydrology 83, 171199.
Cadotte, M., Deschenes, L., Samson, R., 2007. Selection of a remediation scenario for
a diesel-contaminated site using LCA. The International Journal of Life Cycle
Assessment 12, 239251.
Danish Cancer Society, 2006. Krft i Danmark en opslagsbog. (Cancer in Denmark
a reference book). FADLs forlag (in Danish).
Danish EPA, 2007. Ngletal for Samfundskonomiske Beregninger P Milj- Og
Naturomrdet. Miljprojekt Nr. 1199 2007. Danish Environmental Protection
Agency, Danish Ministry of the Environment (in Danish).
Ditlevsen, O., Friis-Hansen, P., 2007a. Cost and Benet Including Value of Life and
Limb Measured in Time Units. Paper from Special Workshop on Risk acceptance
and Risk Communication. 2627 March, 2007, Stanford University.
Ditlevsen, O., Friis-Hansen, P., 2007b. Life quality index - an empirical or
a normative concept? International Journal of Risk Assessment & Management
7, 895921.
Ditlevsen, O., Madsen, H.O., 1996. Structural Reliability Methods. John Wiley & Sons,
Chichester, UK, Edition 1. Edition 2.2.5: Internet Edition July 2005. http://www.
web.mek.dtu.dk/staff/od/books.htm.
EIA, 2008. Energy Information Administration. Ofcial Energy Statistics from the
U.S. Government. Accessed at: http://tonto.eia.doe.gov/dnav/pet/pet_pri_wco_
k_w.htm.
Einarson, M.D., Mackay, D.M., 2001. Predicting impacts of groundwater contamination. Environmental Science and Technology 35, 66A73A.
European Commission, 2001. Recommended Interim Values for the Value of Preventing a Fatality in DG Environment Cost Benet Analysis. Accessed at: http://ec.
europa.eu/environment/enveco/others/pdf/recommended_interim_values.pdf.
Freeze, R.A., Massmann, J., Smith, L., Sperling, T., James, B., 1990. Hydrogeological
decision-analysis .1. A framework. Ground Water 28, 738766.
Friis-Hansen, P., Ditlevsen, O., 2003. Nature preservation acceptance model applied
to tanker oil spill simulations. Structural Safety 25, 134.
Gosset, J.M., 1987. Measurement of Henrys Law constants for C1 and C2 chlorinated
hydrocarbons. Environmental Science and Technology 21, 202208.
Hamed, M.M., El-Beshry, M.Z., 2006. Application of rst-order reliability method to
modelling the fate and transport of benzene in groundwater. International
Journal of Environment and Pollution 26, 327346.
zdemiroglu, E., 2005. The Economics of Groundwater Remediation
Hardisty, P.E., O
and Protection. CRC Press, New York.
Johnson, N.L., Kotz, S., Balakrishnan, N., 1995. Continuous Univariate Distributions,
second ed., vol. 2. John Wiley & Sons, New York.

Khadam, I., Kaluarachchi, J.J., 2003. Applicability of risk-based management and the
need for risk-based economic decision analysis at hazardous waste contaminated sites. Environment International 29, 503519.
Lemming, G., Hauschild, M.Z., Bjerg, P.L., 2010. Life cycle assessment of soil and
groundwater remediation technologies: literature review. International Journal
of Life Cycle Assessment 15, 115127.
LCA Center. 2005. List of EDIP factors downloaded from LCA Center Denmark 04-112008 at <http://www.lca-center.dk/cms/site.aspx?p=1595>. The updating was
done in 2005 by Henrik Fred Larsen from IPU Produktion, Denmark, according
to the principles of the EDIP methodology.
Madsen, H.O., Krenk, S., Lind, N.C., 1986. Methods of Structural Safety. Prentice Hall.
Massmann, J., Freeze, R.A., Smith, L., Sperling, T., James, B., 1991. Hydrogeological
decision-analysis .2. Applications to groundwater contamination. Ground Water
29, 536548.
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, 833847.
McGuire, T.M., McDade, J.M., Newell, C.J., 2006. Performance of DNAPL source
depletion technologies at 59 chlorinated solvent-impacted sites. Ground Water
Monitoring and Remediation 26, 7384.
Nasiri, F., Huang, G., Fuller, N., 2007. Prioritizing groundwater remediation policies:
a fuzzy compatibility analysis decision aid. Journal of Environmental Management 82 (1), 1323. Ref Type: Generic.
NRC, 1997. Valuing Ground Water-Economics, Concepts and Approaches. National
Research Council. National Academy Press, Washington.
Riddick, J.R., Bunger, W.B., Sakano, T.K., 1986. Organic Solvents, fourth ed. John
Wiley & Sons, New York.
Rosen, L., Wladis, D., Ramaekers, D., 1998. Risk and decision analysis of groundwater
protection alternatives on the European scale with emphasis on nitrate and
aluminium contamination from diffuse sources. Journal of Hazardous Materials
61, 329336.
Rosenbaum, R.K., Bachmann, T.M., Gold, L.S., Huijbregts, M.A.J., Jolliet, O., Juraske, R.,
Koehler, A., Larsen, H.F., MacLeod, M., Margni, M., McKone, T.E., Payet, J.,
Schuhmacher, M., van de Meent, D., Hauschild, M.Z., 2008. USEtox-the UNEPSETAC toxicity model: recommended characterisation factors for human
toxicity and freshwater ecotoxicity in life cycle impact assessment. International Journal of Life Cycle Assessment 13, 532546.
ScanRail Consult, HOH Water Technology A/S, NIRAS Consulting Engineers and
Planners A/S, and Revisorsamvirket/Pannell Kerr Forster. 2000. Environmental/
Economic Evaluation and Optimising of Contaminated Sites Remediation. EU
LIFE Project no. 96ENV/DK/0016. Copenhagen, Denmark.
Schaerlaekens, J., Mallants, D., Simunek, J., van Genuchten, M.T., Feyen, J., 1999.
Numerical simulation of transport and sequential biodegradation of chlorinated
aliphatic hydrocarbons using CHAIN_2D. Hydrological Processes 13,
28472859.
Scholz, R.W., Schnabel, U., 2006. Decision making under uncertainty in case of soil
remediation. Journal of Environmental Management 80, 132147.
Steen, B., 1999. A Systematic Approach to Environmental Priority Strategies in
Product Development (EPS). Version 2000 Models and Data of the Default
Method. CPM Report 1999, vol. 5. Chalmers University of Technology, Gothenburg, Sweden.
Stern, N., 2006. The Economics of Climate Change: The Stern Review. Accessed at:
http://www.hm-treasury.gov.uk/stern_review_nal_report.htm.
Stranddorf, H.K., Hoffmann, L., Schmidt, A., 2005. Pvirkningskategorier, Normalisering og Vgtning i LCA. Opdatering af Udvalgte UMIP97-Data. (Impact
Categories, Normalisation and Weighting in LCA. Updating of Selected EDIP97
Data). Miljnyt Nr. 77 2005. Danish Environmental Protection Agency, Ministry
of the Environment (in Danish).
Toffoletto, L., Deschenes, L., Samson, R., 2005. LCA of ex-situ bioremediation of
diesel-contaminated soil. International Journal of Life Cycle Assessment 10,
406416.
Troldborg, M., Lemming, G., Binning, P.J., Tuxen, N., Bjerg, P.L., 2008. Risk assessment
and prioritisation of contaminated sites on the catchment scale. Journal of
Contaminant Hydrology 101, 1428.
US EPA, 1988. Health Effects Assessment for Tetrachloroethylene. EPA/600/8-89096. Environmental Criteria and Assessment Ofce, Ofce of Health and Environmental Assessment, Ofce of Research and Development, Cincinnati, OH,
United States Environmental Protection Agency.
US EPA, 1989. Risk Assessment Guidance for Superfund. In: Human Health Evaluation Manual (Part A). EPA/540/189/002, vol. I. Ofce of Emergency and
Remedial Response, United States Environmental Protection Agency.
US EPA, 1992. Estimating Potential for Occurence of DNAPL at Superfund Sites.
9355.407FS. United States Environmental Protection Agency.
Wenzel, H., Hauschild, M., Alting, L., 1997. Environmental Assessment of Products.
1: Methodology, Tools, and Case Studies in Product Development. ISBN 0 412
80800 5. Kluwer Academic Publishers, Hingham, MA. USA. Chapman & Hall,
United Kingdom, 1997.

You might also like