Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries

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

A toxicity risk assessment method for spill incidents involving volatile

liquid hydrocarbons and aqueous solutions in enclosed areas
Nikolaos Kazantzis a, *, Vasiliki Kazantzi b
a
b

Department of Chemical Engineering, Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609-2280, USA
Department of Project Management, Technological Educational Institute (TEI) of Larissa, Larissa 41110, Greece

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 9 February 2010
Received in revised form
19 May 2010
Accepted 4 June 2010

A new method for the assessment of toxicity risk due to spill incidents involving volatile liquid hydrocarbons and aqueous solutions in enclosed areas is proposed. First, mass transfer models coupled with
time-varying evaporation and emission rates (source models) are used to estimate the dynamic concentration proles of potentially toxic gas/vapor pollutants resulting from spills of volatile liquid hydrocarbons
as well as aqueous solutions in an enclosed area or indoor environment. Recognizing that toxicity risk
depends nonlinearly on exposure duration and concentration, while the latter varies dynamically with
time at any receptor position, the use of the aforementioned models to reliably compute toxic loads in the
presence of time-varying concentration proles is pursued explicitly in the present study. In this manner,
one effectively overcomes the limitations of more traditional approaches to toxicity risk assessment where
toxic loads are estimated under constant (steady state, average or mean) concentration levels of the toxic
pollutant. Furthermore, instead of resorting to complex physiologically inspired pharmacokinetic models
and the associated formidable multi-parameter estimation problems requiring the availability of large sets
of good and reliable data, the proposed method incorporates also the idea of using a simple dynamic
description that provides the requisite degree of differentiation between the exposure concentration
obtained through the above models and the effective concentration (often associated with dose) that
reaches the receptor site as determined by the uptake rate of a toxic vapor/gas. On the basis of the timevarying effective concentration or dose, an effective toxic load that takes into account potential recovery
processes is then computed and integrated into a probit methodological framework where the proper
quantication of a population response to toxic exposure (effective toxic load) provides the means to assess
and characterize toxicity risk. Finally, the proposed method is evaluated through simulation studies in
a case study involving a spill episode of ammonia solution in an enclosed area.

2010 Elsevier Ltd. All rights reserved.

Keywords:
Chemical risk assessment
Spill incidents
Volatile liquid hydrocarbons
Aqueous solutions
Time-varying emission rates
Toxicity assessment

1. Introduction
The quantication of risk assessment associated with incidents
involving the natural, accidental or intentional release of potentially
hazardous chemicals into the environment has been recognized as
a rather useful methodological paradigm and well-justied
research endeavor by the scientic community, provided that its
inherent limitations and extent of scope are carefully acknowledged
in a complex world where uncertainty reigns (Holland & Sielken,
1993; van Leeuwen & Hermens, 1995; Louvar & Louvar, 1997;
Ramaswami, Milford, & Small, 2005; Scheringer, 2002; Trapp &
Matthies, 1998). In the presence of complexity, quantitative risk
models with varying degrees of sophistication should aim at
* Corresponding author. Tel.: 1 508 8315666; fax: 1 508 8315853.
E-mail addresses: nikolas@wpi.edu (N. Kazantzis), kazantzi@teilar.gr
(V. Kazantzi).
0950-4230/$ e see front matter
2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jlp.2010.06.005

complementing and skillfully guiding intuition, which should be

also exercised as models aspire to capture even value-laden facets of
complex situations (Holland & Sielken, 1993; van Leeuwen &
Hermens, 1995; Louvar & Louvar, 1997; Scheringer et al., 2001;
Slovic, Finucane, Peters, & MacGregor, 2004). Given:
1) the considerable uncertainties in modeling emission/release
conditions,
2) the complexity of the combined effect of inherent physicochemical properties and environmental processes on the fate of
a chemical,
3) the physiological responses of individuals to different levels of
exposure (notwithstanding the variability within a certain
population), and
4) the paucity of reliable data at all the above stages, risk models
should not be expected to provide the right answer and generate
predictions endowed with absolute mathematical precision, but

720

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

instead become useful tools enabling scientists to ask the right

questions without resorting to costly experimentation (Holland &
Sielken, 1993; Louvar & Louvar, 1997; Ramaswami et al., 2005;
Scheringer, 2002; Scheringer et al., 2001; Slovic et al., 2004; van
Leeuwen & Hermens, 1995). Indeed, they could be used to:
i) identify the sources of uncertainty with the highest impact on
risk assessment, and/or
ii) identify conditions under which signicant risk might occur,
and/or
iii) form the basis of a reliable risk-informed screening method
Therefore, the above quantitative risk assessment methods
should allow for (the nite) available resources to be concentrated on
and emphasis to be placed on cases where furthermore detailed
assessment is needed in order to reduce uncertainty and mitigate/
manage risk (van Leeuwen & Hermens, 1995; Scheringer, 2002;
Scheringer et al., 2001). Under this spirit, the present study aims at
developing a quantitative framework for the assessment of toxicity
risk posed by spill incidents involving volatile liquid hydrocarbons
and aqueous solutions in an enclosed area or indoor environment.
Following the principles of the conceptual framework presented in
(van Leeuwen & Hermens, 1995; Scheringer et al., 2001), one is rst
reminded that hazard identication is related to the inherent physicochemical properties of the chemical under consideration as well as
the occurrence conditions of the spill incident itself. Furthermore, the
overall risk is quantitatively represented as the product of the probability of occurrence of the above incident multiplied by the conditional probability of observing a fatality (the health end-point in the
present study) due to exposure to toxic vapor given the occurrence of
the spill incident. Notice that the second probabilistic term in the
overall risk is the term traditionally associated with chemical risk
assessment which also captures the effects (or consequences) on
public health and/or ecosystem functions (van Leeuwen & Hermens,
1995; Scheringer, 2002; Scheringer et al., 2001). Chemical risk
depends explicitly on a measure of exposure or dose which can be in
principle estimated through a chemodynamics model that explicitly
takes into account the combined effect of inherent physicochemical
properties and underlying environmental processes on the fate of the
chemical, often in conjunction with a physiologically inspired pharmacokinetic model in order to determine the physiologically
important effective dose (Andersen, 2003; Holland & Sielken, 1993;
van Leeuwen & Hermens, 1995; Ramaswami et al., 2005;
Scheringer, 2002; Trapp & Matthies, 1998). Typically, a quantitative
sense of the size or signicance of risk is obtained through the notion
of risk quotient which is based on a direct comparison of the aforementioned exposure measure or dose to an effect level or threshold
concentration such as the Permissible Exposure Limit (PEL) or the
Immediately Dangerous to Life and Health Limit (IDLH) (van Leeuwen
& Hermens, 1995; Louvar & Louvar, 1997; Ramaswami et al., 2005;
Scheringer, 2002; Scheringer et al., 2001). Finally, the extent or
consequences of the above risk are quantied through a doseresponse type of relationship such as the one associated with a probit
model, which, within the context of the present study, could provide
a probabilistic measure of an expected number (or percentage) of
fatalities due to exposure to toxic vapor of a certain population
(Finney, 1971; Lee, 2002; van Leeuwen & Hermens, 1995; Louvar &
Louvar, 1997; Ramaswami et al., 2005; Scheringer et al., 2001).
Overcoming some of the limitations associated with existing
approaches, the proposed method:
i) allows a more realistic time-varying rather than constant
evaporation/emission rate to be incorporated into a spill
model as in (Guo, 2002; Guo, Sparks, & Roache, 2008)
ii) uses explicitly a more realistic description of toxic load
resulting from a spill incident as a nonlinear time-varying

dynamic quantity rather than one based on a constant (steady

state, average or mean) concentration prole (Lee, 2002), and
incorporates it into an appropriate probit model. Indeed, it
has been convincingly demonstrated that nonlinear toxic
load models generalizing the traditional Habers law in toxicology are more suitable to reliably quantify the combined
health effect of exposure concentration and exposure time/
duration (ten Berge, Zwart, & Appelman, 1986; Hilderman,
Hrudey, & Wilson, 1999; Ride, 1984). In particular, they
overcome the limitations of Habers law associated with
overestimating health responses to low concentrations and
large exposure times, as well as underestimating health
responses at high concentrations and short exposure times
(as in acute toxicity episodes) (Hilderman et al., 1999; Holland
& Sielken, 1993; Ride, 1984; ten Berge et al., 1986). Furthermore, the above nonlinear generalizations of Habers law
profoundly change the effect of time-varying dynamic
concentration proles during the exposure window on health
responses compared to constant concentration values (steady
state, average or mean) (ten Berge et al., 1986; Hilderman
et al., 1999; Ride, 1984), and this should be necessarily
taken into account by any reliable chemical risk assessment
model for spill incidents.
iii motivated by the ideas introduced in the excellent work
(Hilderman et al., 1999), bypasses the complexity and challenges that often accompany the development of a comprehensive physiologically based pharmacokinetic model to
quantify the physiologically relevant effective concentration
or dose available at the receptor site (Andersen, 2003;
Holland & Sielken, 1993; Ramaswami et al., 2005). This is
achieved by introducing: 1) a simple linear rst-order
dynamic description with an empirical parameter called the
uptake time-constant relating the toxic vapor concentration
in the enclosed area calculated through the spill model to the
effective concentration (dose) responsible for health effects,
and 2) a second linear rst-order dynamic description with an
empirical parameter called the recovery time-constant that
captures underlying physiological recovery processes (metabolic reduction of chemical toxicity in the human body,
excretion mechanisms, tissue repair mechanisms, etc). On the
basis of the above models, an effective toxic load is then
dened and integrated into a probit methodological framework (Finney, 1971; Holland & Sielken, 1993) where the
proper quantication of a population response to toxic
exposure (effective toxic load) provides the means to assess
and characterize toxicity risk.
The present paper is organized as follows: Section 2 encompasses the requisite preliminaries associated with spill models
involving volatile liquid hydrocarbons and aqueous solutions in an
indoor environment, as well as the traditional structure of a probit
model for toxicity risk assessment due to exposure to toxic vapor
produced by the above spill incidents. The papers main results and
proposed method are presented in Section 3, followed by its evaluation in a case study discussed in Section 4 focusing on a spill
incident of ammonia solution in an enclosed area. Finally, a few
concluding remarks are provided in Section 5.
2. Spill models for volatile liquid hydrocarbons
and aqueous solutions in enclosed areas
As mentioned in the introductory Section 1, the proposed risk
assessment method involving spill incidents of volatile liquid
hydrocarbons and aqueous solutions relies on the integrated use of
an appropriate spill model coupled with a probit model and

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

a model for the effective toxic load that explicitly take into account
the dynamic time-varying nature of both the emission/evaporation
rate and toxic vapor concentration proles, a physiologically
important effective concentration prole at the appropriate
receptor site, as well as the nonlinearity of the health response of
a population to a particular exposure pattern. At this point, the
reader is reminded that modeling within the context of the
proposed quantitative toxicity risk assessment methodology is
viewed not as a means to provide a level of absolute accuracy and
reality capturing capacity in its predictions for an inherently
complex phenomenon such as the one under consideration, but
rather as a tool to acquire a comparative sense of the toxic risk
involved under different conditions and circumstances, thus identifying cases deserving further thorough assessments and possibly
the commitment of additional resources. Let us now begin with the
presentation of two quite popular modeling frameworks for spill
incidents involving volatile liquid hydrocarbons and aqueous
solutions in enclosed areas whose main structural features are
delineated in (Lee, 2002) and (Guo et al., 2008) respectively. The
rst modeling framework is structurally simpler and based on
a steady/constant evaporation rate for volatile liquid hydrocarbons
in a spill incident. It introduces a rst-order linear dynamic process
model mathematically represented by a single mass balance
equation applied to the enclosure volume V under the assumption
of ideal mixing (Lee, 2002):

dC
E  QC
dt

(1)

where C is the hydrocarbon vapor concentration, E the evaporation

rate and Q the ventilation ow rate (which is of course adjustable).
In this simple model the evaporation rate E is considered constant
and given by the following expression (Lee, 2002):

Mw KAP 0
RT

(2)

where Mw is the molecular weight of the hydrocarbon, A the spill

area (constant), P0 the vapor pressure of the hydrocarbon, R the
universal gas constant, T the liquid temperature and K the mass
transfer coefcient usually calculated through the appropriate
empirical expression (Lee, 2002). From equations (1) and (2) one
obtains the steady-state vapor concentration:

Cs

E
Mw KAP 0

Q
RTQ

(3)

which is used as a measure of exposure concentration (also used in

a risk quotients (van Leeuwen & Hermens, 1995; Scheringer et al.,
2001) in the risk assessment methodology presented in (Lee,
2002)
(using ppm units, expression (3) attains the form:
Cs ppm KAP 0 =QP  106 , where P is the partial pressure of the
hydrocarbon in air). Furthermore, one can easily integrate the
differential equation (1) and obtain the dynamic vapor concentration prole C(t) recognizing that C(t 0) 0:

Ct

E
Q
1  exp  t
Q
V


(4)

Notice that the quantity s V=Q is the process time-constant,

and as intuitively expected, is inversely proportional to the ventilation ow rate Q and proportional to the enclosure volume V. Quite
frequently, in an exposure scenario, the notion of a time-average
concentration value C is used (such as the 8-h average one for
workers (Scheringer et al., 2001)) as a measure of exposure
concentration instead of the steady-state value Cs, and dened as
follows (Scheringer et al., 2001):

1
C
T

721

ZT
Ctdt

(5)

where T is a measure of the exposure time or interval. It should be

pointed out that C captures the full dynamic history of the vapor
concentration prole, whereas Cs is based solely on its asymptotic
steady-state characteristics. Given expression (4), the integral in (5)
can be analytically calculated leading to:






E
V
Q
1
exp  T  1
Q
QT
V

(6)

Please notice that when the exposure time T is considerably

larger than the process time-constant: T >> s V=Q , then the
above equation (6) yields:

/N

(

)
expTs  1
E
E
T 
1 lim
Cs

Q
Q
Ts /N
s

(7)

and therefore the two measures of exposure concentration would

lead to similar risk estimates. However, if T < < 4s, i.e. the exposure
time is much shorter than the time practically required to reach
steady state in the above spill model, one obtains due to the
monotonicity of the function f x expx  1=x:


 

exp Ts  1
T 
s
(

)
expTs  1
E
T 
1
C
Q
s
C
Cs

<

exp4  1
0
4

E
< 0:8 0
Q

(8)

< 0:8

Therefore, within the above modeling framework and under the

above conditions, the use of the steady-state value Cs (as in (Lee,
2002)) instead of the time-average concentration C in a risk
assessment framework would unnecessarily overestimate the
associated risk, since the dynamic (transient) features of the vapor
concentration prole (which obviously become more important in
cases with short exposure times) would be overlooked.
The main ideas guiding the development of a more sophisticated,
comprehensive and experimentally validated model for spill incidents
involving aqueous solutions in an indoor environment that allows the
more realistic integration of time-varying evaporation rates (particularly monotonically decreasing evaporation rates due to the decreasing
concentration prole of the solute in the spilled liquid (Guo, 2002; Guo
et al., 2008)) into its structure have been introduced in the excellent
work (Guo et al., 2008). The proposed model is a fourth-order linear
dynamic process model comprised of four mass balance equations for
the mass of the aqueous solution spilled W, the water vapor concentration in the air m, the solute concentration in the spilled liquid CL and
the solute concentration in the air C (Guo et al., 2008):

dW
dt
dm
V
dt
dC
VL L
dt
dC
V
dt

Rw  Rs
Rw Q mout  m
Rs
Rs  QC

where V is the volume of the enclosed area, VL the volume of the

aqueous solution remaining on the oor, Rw Akgw(msat  m) is the
rate of water evaporation for the spilled liquid with A being the spill
area (time-varying), kgw the gas-phase mass transfer coefcient for
the solute, msat the saturated water vapor concentration in indoor
air, Rs AKOL CL  C=H is the rate of solute emission from the

722

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

spilled liquid (obviously time-varying) with KOL being the overall

liquid-phase mass transfer coefcient, H Henrys constant, and
nally mout the water vapor concentration in ambient air and Q the
air change ow rate. The set of initial conditions that accompanies
the above set of dynamic model equations is: (W(t 0), m(t 0), CL
(t 0), C(t 0)) (W0, mout, CL0, 0), where W0 is the mass of
aqueous solution spilled and CL0 the initial solute concentration in
the spilled liquid. Under the assumptions that (Guo et al., 2008): i)
the water evaporation rate is considerably larger than that of the
solute, ii) the moisture content of the indoor environment, i.e. m is
approximately constant (supported by experimental evidence (Guo
et al., 2008)), iii) the concentration of the solute in indoor air is
much lower than its equilibrium concentration (CL > > C/H), the
above dynamic equations can be analytically integrated leading to
the following time-varying dynamic concentration prole of the
solute in the air (Guo et al., 2008):

Ct

A0 KOL CL0
exp  Nt  exp  qt
Vq  N

(10)

where
N

Q/V
is
the
air
change
rate,
and
q A0 kgw msat 1  r=W0 KOL =q, with q VL/A being the
liquid lm thickness (considered constant; for a justication please
see (Guo et al., 2008)), A0 being the initial spill area and r the
relative humidity in indoor air (m/msat). Within the above modeling
framework, please notice that the evaporation rate is a monotonically decreasing quantity (Guo et al., 2008):



K t
Rs AKOL CL0 exp  OL

(11)

and consequently, the concentration C(t) asymptotically

approaches the zero value in agreement with the intuitively
expected behavior under realistic conditions. Furthermore, one can
show that the above concentration prole attains a maximum value
given by the following expression:

Cmax

A K C
0 OL L0
Vq  N

(  N   q )
N qN
N qN

q
q

K CT

(14)

where C is the exposure concentration, T the exposure time and

K a constant related to a specic level of fatalities in a population.
Habers law suggests that fatal health responses by members of
a population remain essentially invariant for various exposure
scenarios as long as the product CT stays constant: reducing the
exposure time by half while doubling the exposure concentration
value would lead to the same level of toxicity risk and similar health
responses. Quite often the product CT is known as the toxic load
and denoted by: L CT (this symbol will be retained throughout the
present paper to denote toxic load), or the innitesimal dL C(t)
dt (toxic load in differential form) whenever the concentration
prole is time-varying. Indeed, Habers law was also used in cases
where concentration proles were not necessarily constant, with C
in the original form (14) being now replaced by the time-average
concentration C (Hilderman et al., 1999). The rationale behind such
a use, is that the total toxic load under Habers law in the case of
time-varying concentrations (and hence the associated health
effects) remains invariant:

Z
L

Z
dL

Ctdt CT

(15)

However, new experimental studies, toxicological observations

and empirical ndings elucidated the inherent limitations of Habers law (ten Berge et al., 1986; Hilderman et al., 1999; Ride, 1984):

(12)

For this particular spill model, a time-average exposure

concentration value C can be analytically computed as well:

Z
1 T
Ctdt
T 0


A K C
1
1
0 OL L0
expqT  1  expNT  1
Vq  NT q
N

to exposure to toxic chemicals have early moorings in the

tumultuous scientic career and tragic personal life of Fritz
Haber who was the architect of the development of the chemical
weapons program of the German army in early 20th century.
On the basis of preliminary evidence accumulated through the use
of poison gases for military purposes, Haber postulated that fatal
toxicity can be described through the following simple law
(Andersen, 2003; ten Berge et al., 1986; Hilderman et al., 1999;
Holland & Sielken, 1993; Ride, 1984):

(13)

which can be considered as a measure of exposure concentration in

risk assessment studies.
3. The proposed toxicity risk assessment method
Traditionally, toxicological experiments have been conducted
in such a way that laboratory animals are exposed to constant
concentration levels of the toxic chemical under consideration for
a xed period of time and the corresponding number of fatalities
is observed and recorded (Andersen, 2003; Holland & Sielken,
1993; van Leeuwen & Hermens, 1995). From a practical point
of view, the selection of the number of fatalities as a health
end-point in toxicological studies of this nature is justied since
less severe effects on health, even if observed, can not be easily
quantied and are inherently associated with signicant variability within a certain population (Andersen, 2003; Hilderman
et al., 1999; Holland & Sielken, 1993; van Leeuwen & Hermens,
1995). Historically, acute toxicity studies and mortality rates due

i) Regardless of how low the exposure concentration levels are,

fatalities should be observed as long as the exposure time
becomes quite large. This prediction of Habers law leads to
overestimating the health effects at low concentration levels,
a prediction empirically refuted and theoretically weak since
it overlooks the underlying physiological mechanisms associated with recovery processes.
ii) The health effects due to high exposure concentration levels
can be modulated by shortening the exposure time. This
prediction of Habers law leads to underestimating the
severity of health effects often observed in acute toxicity
episodes due to the underlying complexity of the physiological mechanisms activated in order to cope with an overwhelming short exposure to high toxic concentration levels.
In light of i) and ii), toxicologists introduced nonlinear generalizations of Habers law which attain the following form (ten Berge
et al., 1986; Hilderman et al., 1999; Ride, 1984):

L CnT

(16)

where the exponent n is associated with the particular chemical

and customarily determined through experiments (with typical
values n  1). Notice that the above nonlinear generalization of
Habers toxic load is in agreement with the empirical evidence in
cases i) and ii), i.e. the behavior of toxic load realistically expected
under low and high exposure concentration levels. As before, in the
case of time-varying concentration proles one could use the

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

time-average exposure concentration value C in the above

nonlinear generalization:
n

Lm C T

(17)

and thus introduce the notion of a mean toxic load for the
exposure duration. Notice however that in the nonlinear case n > 1
(as opposed to the linear one where n 1) the dynamic changes in
C C(t) do matter, and the overall or total toxic load which is
properly dened as follows:

Z
Lt

Z
dL

C n tdt

(18)

(
Qn C1 ; .; Cm

m
1X
Cn
m i1 i

Z
Lt

Z
dL

C n tdt  C T Lm

(19)

thus representing a quite useful upper bound on toxic loads when

evaluating toxicity risk and associated health effects such as fatalities. At this point, a mathematical justication is provided of the
above result. Consider the standard Riemann representation of the
integral in (18):

Lt lim

m/N

m
X
i1

Cin Dt

(20)

under a partition of the exposure interval with mesh: Dt T=m

and Ci C(ti) with ti T=mi; T=mi 1. Similarly, under the same
Riemann partition, one obtains:
n

Lm C T

1
T n1

(Z

)n

Ctdt
0

1
T n1

(
lim

m/N

m
X

)n
Ci Dt

i1

(21)
Dene now the quantity:

)1
n

(22)

with C1, ., Cm > 0. Notice that:

Q1 C1 ; .; Cm

m
1X
C
m i1 i

(23)

Using the generalized mean inequality for n > 1 (Hardy,

Littlewood, & Plya, 1988), the following inequality can be established:

)1
(
n
m
m
P
P
1
1
Qn C1 ; .; Cm iQ1 C1 ; .; Cm 0 m
Cin > m
Ci 0
i1
i1
(
)n
(
)n
m
m
m
m
X
X
1X
1 X
Dtn
Cin > n
Ci 0Dt
Cin >
C
0
i
m i1
m
mDtn1 i 1
i )1
i1
(
(
)n
n
m
m
m
X
X
X
Pm
1
1
n
n
Ci Dt 0 lim
Ci Dt  n1 lim
Ci Dt 0Lt  Lm
i 1 Ci Dt > n1
m/N
m/N
T
T
i1
i1
i1

is not equal to (17) any more: Lt s Lm. In the special case of

a constant concentration prole where C C(t) M constant,
then, even in the nonlinear case (n > 1), C M and Lt MnT Lm.
The reader is reminded that in a typical toxicological experiment
the exposure concentration prole C is constant by design, and
therefore, no particular concern is caused about the choice and
use of the proper toxic load model. In a spill incident however, the
exposure concentration C is realistically expected to vary with
time (within the exposure time-window), and the selection of
total toxic load (18) is regarded as more useful since it captures
the simultaneous effect of nonlinear and time-varying exposure
concentrations on toxicity (ten Berge et al., 1986; Hilderman et al.,
1999; Ride, 1984). From a risk assessment, as well as regulatory
perspective, it is also important to show that in the nonlinear case
where n > 1, the total toxic load Lt is larger than the mean toxic
load Lm:

723

(24)

Motivated by the ideas introduced in (Hilderman et al., 1999),

one should acknowledge that the notion of overall/total toxic load
(18), despite its advantages, still lacks a degree of realism when used
for the assessment of toxicity risks in a real exposure scenario
following a spill incident such the ones under consideration. Indeed,
the recognition that it does not capture the undeniable facts that:
A) the uptake of any exposure concentration can not be instantaneous but rather one needs to consider an effective exposure
concentration or dose available at the receptor site which
potentially inuences health effects estimates (the number of
fatalities in our case), and
B) the underlying biological recovery processes do contribute to
the health response of a population to toxic exposure, and that
toxic loads can not increase indenitely as time progresses and
effective concentration levels become quite low. Apparently,
a scientically thorough and detailed description of cases A)
and B) would require the development of a comprehensive and
quite often complex physiologically based pharmacokinetic
model that by itself could pose considerable challenges
(considering all possible exposure routes, metabolic pathways,
the fate of the chemical inside the body, etc. which inevitably
introduce a large number of parameters that need to be reliably
estimated) (Andersen, 2003; Hilderman et al., 1999; Holland &
Sielken, 1993; Ramaswami et al., 2005). Instead, as the authors
show in (Hilderman et al., 1999), a simple rst-order dynamic
description of an effective exposure concentration Ceff available
at the receptor site could be introduced, and its relationship to
the concentration C of the toxic vapor (obtained from a spill
model) represented by:

dCeff
1
C  Ceff

sup
dt

(25)

where sup is an empirical parameter called the uptake timeconstant. Notice that small uptake time-constants sup /0 lead
almost instantaneously to: Ceff /C, whereas very large ones

724

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

sup /N to Ceff /0 as intuitively expected. For any nite value

0 < sup < N, one also concludes that asymptotically

limt/N Ceff C. In this case it is of course presupposed that the

exposure period T is quite large: T > > 4sup, otherwise, if T < 4sup,
then Ceff has not yet converged to C. This behavior is certainly
expected, since the above simple dynamic model nicely captures
the fact that the uptake process is not instantaneous and the short
exposure duration does not allow enough time for the effective
concentration level Ceff at the receptor site to reach C the concentration of the toxic vapor. Finally, it should be pointed out that
a more detailed uptake model based on rst-principles would
have to integrate all possible exposure and absorption routes
(inhalation path involving nose, throat and lungs in which case Ceff
is the concentration in the airways, metabolic paths involving the
bloodstream, absorption and transfer mechanisms between
the lungs and blood in which case Ceff is the concentration in the
bloodstream, etc.) (Andersen, 2003; Hilderman et al., 1999; Holland
& Sielken, 1993). On the basis of the notion of effective exposure
concentration Ceff, the notion of overall/total toxic load can be
modied as follows:

Z
Leff

T
0

Z
dL

T
0

n
Ceff
tdt

(26)

by introducing an effective toxic load Leff through which the

assumption of instantaneous uptake is relaxed.
With respect to case B), one should also be cognizant of the role
of complex underlying biological mechanisms (metabolic degradation of the chemicals toxicity, excretion mechanisms, physiological repair of damaged tissue) responsible for recovery processes
as evidenced by numerous studies (Andersen, 2003; Hilderman
et al., 1999; Ramaswami et al., 2005). These of course tend to
naturally reduce the effective toxic load. In such a case, a simple
rst-order recovery model for the effective toxic load can be
proposed (Hilderman et al., 1999):

dLeff
1
n
 Leff
Ceff
sr
dt

(27)

where sr is the recovery time-constant. Notice that (27) further

generalizes the effective toxic load model of (26):

dLeff
n
Ceff
dt

(28)

by capturing also the effect of recovery processes. When sr /N (no

recovery), the two models for Leff coincide. If sr /0 (immediate
recovery), then Leff /0 as expected. In cases where 0 < sr < N,
equation (27) describes the expected gradual decline of the effective toxic load values due to the latent recovery processes.
Summarizing, in the case of a noninstantaneous uptake the
following toxic load model is proposed:

dCeff
1
C  Ceff

sup
dt
dLeff
n
Ceff
dt

(29)

with Ceff(t 0) 0, Leff(t 0) 0 and C(t) representing the toxic

vapor concentration prole as given by the spill model equations
presented in Section 2. In cases where recovery processes are taken
explicitly into account, the proposed toxic load model takes the
following form:

dCeff
1
C  Ceff

sup
dt
dLeff
1
n
 Leff
Ceff
sr
dt

(30)

with Ceff(t 0) 0, Leff(t 0) 0 and C(t) representing the toxic

vapor concentration prole obtained through the spill model
equations presented in Section 2.
Given the availability of a working spill model and a model for
the effective toxic load, the last remaining component of a quantitative toxicity risk assessment framework is the development of
a risk model (a type of doseeresponse relationship) that relates the
effective toxic load to the number of fatalities (or a percentage) in
a population due to exposure to a toxic vapor resulting from a spill
incident. At this point, the reader is reminded of the two dimensions of a quantitative chemical risk assessment methodological
framework: the size of risk as realized through the notion of risk
quotient (the ratio of Ceff to a threshold value such as PEL or IDLH),
and the consequences of risk as realized through the appropriate
doseeresponse relationship and a specic health end-point
(fatalities for example in toxicity studies). In the context of the
present study, a probit doseeresponse model is proposed (Finney,
1971; Hilderman et al., 1999; Holland & Sielken, 1993). According
to this methodology the variability of individual tolerances to toxic
chemicals is recognized (Holland & Sielken, 1993). In particular, it is
assumed that a certain threshold dose or toxic load Leffth exists for
each individual, but such a threshold toxic load value varies
amongst the members of a population. Furthermore, it is assumed
that the individual threshold toxic load Leffth is log normally
distributed with a certain logarithmic mean m and standard deviation s (an assumption owing its justication to experimental
ndings and empirical data (Hilderman et al., 1999; Holland &
Sielken, 1993)). Therefore, the expected percentage of fatalities P
due to exposure to a certain toxic load Leff would be determined by
a threshold load that satises the inequality: Leffth < Leff, and thus
given by the cumulative distribution function shown below:

1
P p
2ps

lnLeff

N

exp

 )

1 xm 2
dx

s
2

(31)

The above expression (31) can be rewritten through a standard

change of variables as follows (Finney,1971; Holland & Sielken,1993):



Z ablnLeff

1
1
P p
exp  y2 dy F a b ln Leff
2
2p N

(32)

where F is the standard normal integral (Finney, 1971). One now

denes the probit variable Pr:

Prha bln Leff

(33)

and therefore:

P FPr

(34)

Notice that a, b coefcients in the above change of variables in

integral (32) are related to (m, s) of the above health response
distribution, depend on the specic chemical and are determined in
practice through experiments and on the basis of empirical data. For
a specic value of the effective toxic load Leff, equation (33) gives the
associated value of the probit variable Pr, and nally the percentage
of fatalities is obtained by using equation (34) (specically, tabulated values of the standard normal integral corresponding to
various values of its argument). Conversely, for a certain percentage
of fatalities, one inverts the P F(Pr) relationship computing the
associated value of the probit variable Pr, and then, using equation
(33) the corresponding value of the effective toxic load Leff that
induces the aforementioned percentage of fatalities is estimated. In
the next Section, the proposed toxicity risk assessment method is
evaluated through simulation studies in a case study involving
a spill episode of ammonia solution in an enclosed area.

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

4. Illustrative example: toxicity risk assessment

for an ammonia spill incident

Table 2
Spill incident II e process parameter values.

Pr a b ln Leff

(35)

with a 15.8, b 1, n 2, since it appears to be in closer

agreement with experimental results and methodologically
consistent with toxicological requirements involving more realistic
health responses of humans to toxic vapors (Schubach, 1995). The
1% fatalities level corresponding to a probit variable value of:
Pr z 2.6 is considered that is often related to a toxic load or
concentration threshold value close to the Immediately Dangerous
to Life and Health Limit (IDLH). Indeed, using equation (33), one can
explicitly calculate the corresponding level of effective toxic load:
7
2
3
L[1]
eff 9.7  10 [(mg) (min)/m )] that is probabilistically expected
to induce 1% fatalities. In the rst case, an acute exposure scenario
is considered where immediately after the spill and for a relatively
short period of time, the exposure concentration uctuates drastically attaining a maximum value of: Cmax z 5980 [mg/m3] which
is perilously close to levels characterized as severe by various
regulatory agencies (Schubach, 1995; U.S. Department of Health
and Human Services, 2004), rapidly decreasing afterwards to
much lower levels so that the 8-h average concentration value
Cz35 [mg/m3] stays close to the safe PEL level. In Fig. 1, please
notice that the total toxic load model (18) (in this case we consider
the simpler model with instantaneous uptake and no recovery
denoted by L1) crosses the L[1]
eff level that induces 1% fatalities after

Table 1
Spill Incident I e Process parameter values.
Process parameters

Values

kgw
KOL
msat
r
A0
W0
V
N Q/V

12.5 mh1
4E-3 mh1
18 g3
0.69
20.7 m2
15.4 kg
30 m3
0.4 min1
9.6E-6 m
1E6 g3

Process parameters

Values

kgw
KOL
msat
r
A0
W0
V
N Q/V

12.5 mh1
4E-3 mh1
18 g3
0.69
0.207 m2
154.3 g
30 m3
0.004 min1
9.6E-4 m
1E4 g3

CLO

approximately 5 min, capturing nicely the severity of the underlying acute exposure scenario. This behavior is contrasted with
a toxic load model based on the 8-h average (mean) concentration
value Cz35 [mg/m3], which, as shown in Fig. 1, grossly underestimates the associated toxicity risk. Therefore, this particular case
illustrates that it is certainly more advantageous to use even the
simple total toxic load model (18) in spill incidents, when
compared to more traditional models that use time-average (mean)
or other constant exposure concentration values.
In the second case, a different spill incident is considered under
conditions close to the experimental ones reported in (Guo et al.,
2008). In this case, the exposure concentration uctuates mildly
within a larger exposure window reaching a maximum value of
Cmax z 60 [mg/m3]. This is certainly a case of negligible toxicity risk
since all concentration values remain close to the PEL value, and the
standard regulatory characterization of ammonia toxicity under
these conditions is practically risk-free. The proposed effective toxic
load model (30) was used in our simulation studies with an uptake
time-constant sup 1[s] (a reasonable value under the assumption
that the uptake rate is determined by the inhalation rate in similar
cases as justied in (Hilderman et al., 1999; U.S. Department of
Health and Human Services, 2004)) and a recovery time-constant
sr 20[min] (a reasonable value justied in light of empirical
ndings and results from experimental studies on underlying
physiological, metabolic toxicity reduction and excretion mechanisms involving ammonia as discussed in (U.S. Department of
Health and Human Services, 2004)). Based on the simulation
results graphically depicted in Fig. 2, notice that the proposed

10

10
Load (mg/m3)2 min

An aqueous ammonia solution conforming to the specications

of the one used in the experimental study (Guo et al., 2008) is
considered (1.3% (w/w) concentration) and spill episodes are
simulated with the aid of the second model presented in Section 2.
In all the ensuing simulation runs all mass transfer coefcient
values as well as inherent physicochemical properties of the
ammonia solution are the same as the ones considered in the
experimental investigations in (Guo et al., 2008) that were conducted in order to validate the associated spill model. The rest of
the model input variables such as the enclosure volume, ventilation
rate, initial concentration of the aqueous solution, amount of
solution spilled, initial spill area could vary giving rise to different
scenarios. For the particular cases or spill incidents considered in
the present study the above sets of model input values are tabulated and presented in Tables 1 and 2. It should be also pointed out
that in both the above scenarios which physically correspond to
drastically different spill episodes, the same 8-h average (mean)
concentration value Cz35 [mg/m3] is considered, a value close to
the Permissible Exposure Limit (PEL) value as set by OSHA
(U.S. Department of Health and Human Services, 2004). Furthermore, the probit doseeresponse model equation used for the
ammonia solution was the one developed by the Netherlands
Organization of Applied Scientic Research (TNO):

CLO

725

10

10

L1 (instantaneous)
L(1% fatalities)
Load (mean)

10

10

10

10

10

t[min]
Fig. 1. Spill incident I: Toxic load versus time curves.

10

726

N. Kazantzis, V. Kazantzi / Journal of Loss Prevention in the Process Industries 23 (2010) 719e726

provides the requisite degree of differentiation between the

exposure concentration obtained through the aforementioned
models and the effective concentration that reaches the receptor
site as determined by a nite uptake rate of the toxic vapor/gas.
Furthermore, on the basis of the time-varying effective concentration, an effective toxic load that takes into account latent
recovery processes was integrated into a probit methodological
framework within which the quantication of a population
response to toxic exposure was made possible. Finally, the
proposed method was evaluated in a case study involving a spill
episode of ammonia solution in an enclosed area.

x 10

L1(effective)
L2(instantaneous)
L3(mean)

4.5
4

Load (mg/m3)2 min

3.5
3
2.5
2
1.5

Acknowledgements

1
0.5
0

50

100

150

200

250

300

350

t[min]
Fig. 2. Spill incident II: Toxic load versus time curves.

The authors would like to thank Professor S. M. Mannan and his

Associates at the Mary Kay OConnor Process Safety Center,
Department of Chemical Engineering, Texas A&M University for
their kind invitation to participate in the 2009 International Process
Safety Symposium. They are also indebted to the anonymous
reviewers for their thoughtful suggestions and helpful remarks.
References

effective toxic load (denoted by L1) has the capacity to realistically

capture the very low risk level under the set of conditions associated with the particular spill incident considered in this case.
However, in such a case, both the toxic load with instantaneous
uptake and no recovery (denoted by L2), as well as the toxic
load based on the 8-h average (mean) concentration value
Cz35 [mg/m3] predict unreasonably high risk levels, i.e. after
a certain exposure period they both cross the L[1]
eff level of 1%
fatalities, thus contradicting the established safety threshold values
and limits (such as PELs) which are deliberately set by regulatory
authorities in a conservative precautionary manner, and certainly
orders of magnitude below any fatalities-producing exposure
concentration levels. Therefore, as shown in this second case, the
proposed quantitative toxicity risk assessment method appears to
carry considerable merit when comparatively evaluated against
more traditional ones in pertinent studies.
5. Concluding remarks
A new method for the assessment of toxicity risk due to spill
episodes involving volatile liquid hydrocarbons and aqueous solutions in enclosed areas was proposed. Source models with timevarying evaporation and emission rates were used in conjunction
with appropriate mass transfer models to estimate dynamic
concentration proles of potentially toxic pollutants following
spills of volatile liquid hydrocarbons as well as aqueous solutions in
an indoor environment. Relaxing the assumption found in more
traditional approaches to toxicity risk assessment where toxic loads
are estimated under constant concentration levels of the toxic
pollutant, it was shown that the proposed approach takes explicitly
into account the dynamic features of the concentration prole
during the time of exposure. Furthermore, the proposed method
incorporated the idea of using a simple dynamic description that

Andersen, M. E. (2003). Toxicokinetic modeling and its applications in chemical risk

assessment. Toxicology Letters, 138, 9e27.
ten Berge, W. F., Zwart, A., & Appelman, L. M. (1986). Concentration-time mortality
response relationship of irritant and systematically acting vapours and gases.
Journal of Hazardous Materials, 13, 301e309.
Finney, D. J. (1971). Probit analysis. Cambridge, UK: Cambridge University Press.
Guo, Z. (2002). Review of indoor emission source models. Part 1. Overview. Environmental Pollution, 120, 533e549.
Guo, Z., Sparks, L. E., & Roache, N. F. (2008). Modeling small-scale spills of aqueous
solutions in the indoor environment. Journal of Hazardous Materials, 153,
444e453.
Hardy, G. H., Littlewood, J. E., & Plya, G. (1988). Inequalities. Cambridge, UK:
Cambridge University Press.
Hilderman, T. L., Hrudey, S. E., & Wilson, D. J. (1999). A model for effective toxic load
from uctuating gas concentrations. Journal of Hazardous Material, 64, 115e134.
Holland, C. D., & Sielken, R. L. (1993). Quantitative cancer modeling and risk assessment. New York: Prentice Hall.
Lee, K. W. (2002). A methodology for assessing risk from released hydrocarbon in an
enclosed area. Journal of Loss Prevention in the Process Industries, 15, 11e17.
van Leeuwen, C. J., & Hermens, J. L. M. (1995). Risk assessment of chemicals: An
introduction. Dordrecht, The Netherlands: Kluwer.
Louvar, J. F., & Louvar, B. D. (1997). Health and environmental risk analysis. New York:
Prentice Hall.
Ramaswami, A., Milford, J. B., & Small, M. J. (2005). Integrated environmental
modeling. Hoboken, NJ: Wiley.
Ride, D. J. (1984). An assessment of the effects of uctuations on the severity of
poisoning by toxic vapours. Journal of Hazardous Materials, 9, 235e240.
Scheringer, M. (2002). Persistence and spatial range of environmental chemicals: New
ethical and scientic concepts for risk assessment. Darmstadt, Germany: Wiley.
Scheringer, M., Vgl, T., von Grote, J., Capaul, B., Schubert, R., & Hungerbhler, K.
(2001). Scenario-based risk assessment of multi-use chemicals: application to
solvents. Risk Analysis, 21, 481e497.
Schubach, S. (1995). Comparison of probit expressions for the prediction of lethality
due to toxic exposure. Journal of Loss Prevention in the Process Industries, 8,
197e204.
Slovic, P., Finucane, M. L., Peters, E., & MacGregor, D. G. (2004). Risk as analysis and
risk as feelings: some thoughts about affect, reason, risk and rationality. Risk
Analysis, 24, 311e322.
Trapp, S., & Matthies, M. (1998). Chemodynamics and environmental modeling: An
introduction. Heidelberg, Germany: Springer.
U.S. Department of Health and Human Services. (2004). Toxicological prole of
ammonia. Washington D.C.: Agency for Toxic Substances and Disease Registry.