Ecological Risk Assessment - Part 5 pptx

Part V
Risk Characterization
Statements about single events can’t be decided by a calculator; they have to be hashed out by
weighing the evidence, evaluating the persuasiveness of arguments, recasting the statements to
make them easier to evaluate, and all the other fallible processes by which mortal beings make
inductive guesses about an unknowable future.
Pinker (1997)
Risk characterization is the phase of ecological risk assessment that integrates the exposure
and the exposure–response profiles to evaluate the likelihood of adverse ecological effects and
uses those results to synthesize a useful conclusion. In other words, it is the process of
estimating and interpreting the risks and associated uncertainties. There are two fundamen-
tally different types of risk characterizations. Screening assessments are intended to quickly
and easily divide risks into those that need more attention and those that can be ignored
(Chap ter 31). Definit ive a ssessment s are intende d to infor m a decision -making pr ocess by
providi ng risk estimat es for all asses sment end points (Ch apter 32).
Riskcharacterizations may bealgorithmic in thattheymay use a standard procedure basedon a
standard set of input information using standard assumptions, scenarios, and models. Algorith-
mic approaches are used primarily in ecological risk assessments of pesticides and industrial
chemicals (Luttik and van Raaij 2003; EPPO 2004). They are desirable in that context, because
they are efficient and fair to all of the competitive products that come before a chemical regulator.
They are popular with regulated parties, because the data requirements are clear, and the outcome
of a regulatory assessment can be predicted. Algorithmic approaches are disadvantageous when
chemicals have properties that are not considered in the algorithm. The obvious example is
endocrine disruptors that are not addressed by standard test batteries or effects models.
Alternatively, risk characterization may be performed ad hoc. The advantage of ad hoc
approaches is that they can be designed to provide the best estimate of risk and uncertainty
given the types of information that are available and the particular circumstances of the
assessment. Ad hoc approaches have been used for contaminated sites, because the condi-
tions and information sets are highly variable. Ad hoc approaches are also employed when
assessments are highly contentious or when unusual issues such as developmental deformities
are involved.

ß 2006 by Taylor & Francis Group, LLC.
Infer ence in risk charact erization takes diff erent form s de pending on the type of assessment
and the types of infor mation that are avail able. They differ in how they use the avail able lines
of evidence to reach a con clusion. In risk ch aracteriza tion, a line of evidence is an estimat e of
expo sure an d a corres pondi ng exposure–r esponse relationshi p.
Single line of eviden ce : The classic form of infer ence uses one line of ev idence, whi ch is either
the only availab le evidence or the best eviden ce. For ch emicals, the most c ommon line of
evidence is an expo sure estimat e from a mathemati cal model and a num erical en dpoint from a
toxic ity test.
Weight of eviden ce : If mult iple lines of evidence are avail able, they may be joint ly con-
sider ed. The multiple lines may be from a single type of evidence (e.g., exp osure–r esponse
relation ships from different tests) or from multiple types (e.g., chemical toxic ity tests, tests of
con taminate d media , and biologi cal surveys ).
Risk ch aracteriza tions may also be diff erentiated by the form of the infer ence.
Rule -based inference : Risk asses sors may be provided with an inferen tial rule to de termine
whet her a risk is accepta ble. The sim plest an d most common is: if the e xposure estimat e
exceed s the benchmark effec ts level (i.e., HQ > 1; Secti on 31.1), the risk is unacce ptable.
A more complex rule is: if the 90 th percent ile of the e xposure dist ribution exceeds the 10th
percen tile of the effec ts dist ribution, the risk is una cceptable (Section 30.5). Rule -based
inference is most common in algorithmic assessments of new chemicals. How ever, an infer-
ential rule may be developed for an individual assessment during the problem formulation
(Ch apter 18). Rule-based inference may be applie d to screeni ng or defini tive asses sment s. It is
usuall y limit ed to a single line of evidence but , in its original form , the sedim ent qualit y triad
is a rule- based inferenti al method for three lines of evidence (Chap ter 32).
Ad hoc judgment: In many cases, risk characterizations include judgments concerning
acceptability of a risk without a priori rules or guidance. This approach provides the greatest
flexibility and influence to the assessors, but lacks transparency and diminishes the role of
stakeholders and decision makers.
Structured judgment: Many risk characterizations are too complex and the evidence too
ambiguous to allow rule-based inference, but ad hoc judgment gives too much latitude to

assessors. In such cases, the assessor’s judgment can be guided by an inferential structure
including organization of the input data by type of evidence, the use of standard consider-
ations to evaluate the evidence, and scoring systems. Examples of structures for judgment for
causal analys is and risk charact eriza tion are present ed in Chapt er 4 an d Chapter 32,
respectively.
Risk estimation: One may estimate risks and uncertainties and report them to a risk
manager who interprets the estimates and makes a decision. Risk estimation is used in
definitive assessments and may be based on any number of lines of evidence. Risk estimates
are essential if the results of risk characterization are to be used in an economic analysis,
decision analysis, or other quantitative decision-support tool.
Comparison of alternatives: Rather than characteriz ing risks from an agent or activity to
determine its acceptability, one may compare alternatives to determine which is preferable
(Ch apter 33). Exa mples include alterna tive chemi cals with the same us e, alte rnative remedial
actions for a contaminated or disturbed site, and alternative management plans for a forest.
These approaches to inference are not mutually exclusive. For example, it is often appro-
priate to use structured judgment to determine whether significant effects are likely and then,
if the results are positive, use risk estimation to inform a decision.
ß 2006 by Taylor & Francis Group, LLC.
29
Criteria and Benchmarks
For various reasons , it is somet imes desir able to red uce the c omplex ities of expo sure–re sponse
relationshi ps for v arious taxa, pro cesses, an d other eco logical pro perties to a single num ber
that is presu med to be a suff iciently protect ive level. Those that are us ed to separat e
accepta ble from unaccep table concen trations for regula tory purp oses are termed crite ria or
standar ds (hencef orth, sim ply crit eria). Thos e that are used for screenin g or priori tization are
termed screeni ng bench marks or screeni ng values .
29.1 CRITERIA
Criter ia are con centrations of contam inants in water or other media that are intende d to
consti tute the bounds of regula tory accep tability given prescr ibed co nditions (Section 2.2).
The only national ecological criteria in the United States are the acute and chronic National

Ambient Water Quality Criteria (NAWQC). Criteria were prop osed for sediments by the
Environ mental Pro tection Agency (EP A) but wer e conve rted to screeni ng gu idelines (Sect ion
29.2.) The acu te NAW QC are calculated by the EPA as half the final acute value, whi ch is the
5th percentile of the distribution of 48 to 96 h LC
50
values or equivalent median effective
concentration (EC
50
) values for each criterion chemical (Stephan et al. 1985). The acute
NAWQC are intended to correspond to concentrations that would cause less than 50%
mortality in 5% of exposed species in a relative ly brief exposure. Because the criterion is
not a no-effect level, the criterion is lowered if an impor tant species is among the most
sensitive 5% (Figure 29.1) . The chronic NAWQC are final acute values divided by the final
acute=chronic ratio, which is the geometric mean of quotients of at least three LC
50
=CV ratios
from tests of organisms belonging to different families of aquatic organisms (Stephan et al.
1985). Chronic NAWQC are intended to prevent significant toxic effects in most chronic
exposures. Some, termed final residue values, are based on protection of humans or other
piscivorous organisms rather than protection of aquatic organisms.
Because criteria are applied to an entire state or nation, they should be derived in a way that
accounts for variance among sites and uncertainty. Site-specific standards may incorporate
site properties to reduce either variance or uncertainty. For example, the NAWQC for many
metals are functions of hardness, so that important sources of variance can be eliminated in
site-specific applications (Spehar and Carlson 1984; Stephan et al. 1985). Similarly, results
from testing of local species may be used to modify national criteria in deriving site-specific
standards. More broadly, standards may be derived for different classes of ecosystems (e.g.,
freshwater and saltwater standards in the United States), different uses (e.g., agricultural,
residential, commercial, and industrial land uses in Canadian soil guidelines), or different
levels of protection (e.g., the designation of National Parks as Class I under the US Clean

Air Act).
435
ß 2006 by Taylor & Francis Group, LLC.
NAWQC are applicable regulatory criteria and are generally adequatel y protective, but
they are often not good risk estimators for particular sites. If they are applied to a site,
assessors should co nsider deriving site-specific criteria using the water effect ratio. This is a
factor for adjusting criteria to site water that may be derived using an EPA procedure (EPA
1983; Office of Science and Technology 1994). It requires performing toxicity tests with the
chemical in site waters, and, optionally, with site species (Figure 29.2). The time and expense
0
0.1
1
10
10
2
10
3
10
4
10
5
0.2 0.4
Freshwater final acute value* = 2.1 µg/L dissolved cadium @ 50 mg/L total hardness
Criteria maximum concentration = 1.0 µg/L dissolved cadium @ 50 mg/L total hardness
Freshwater
% Rank GMAVs
Ranked summary of cadmium GMAVs
Cadmium effect concentration (µg/L)
0.6 0.8 1
Freshwater invertebrates

Freshwater fish
Freshwater amphibians
(lowered to protect rainbow trout)
*
FIGURE 29.1 Acute and chronic ambient freshwater quality criteria for cadmium at 50 mg=L hardness
(horizontal lines), and the acute species sensitivity distribution. The acute values (LC
50
s and EC
50
s) for
species and genera are geometrically averaged so the points are genus mean acute values (GMAVs).
(From EPA (U.S. Environmental Protection Agency), 2001 Update of Ambient Water Quality Criteria
for Cadmium, EPA-822-R-01-001, Office of Water, Washington, DC, 2006a. With permission.)
Toxicity in site
water = 0.4 mg/L
Toxicity in
laboratory water
= 0.1 mg/L
Water effect ratio =
0.4:0.1 = 4
Site-specific criterion
= 4 ϫ 0.06 =
0.24 mg/L
Water quality
criterion =
0.06 mg/L
FIGURE 29.2 An illustration of the derivation and use of water effect ratios.
ß 2006 by Taylor & Francis Group, LLC.
requir ed to calculate site-spe cific criteri a co uld be worthw hile if the water ch emistry at a site
differs significan tly from conven tional laborato ry test waters. Othe rwise, the effor t is be tter

expend ed on tests of ambie nt waters (Sect ion 24.5).
Currently, in the United States, the methodology for deriving ambient water quality criteria
is being reexamined, and the risk assessment framework is being applied. In particular,
derivation of new criteria will begin with a problem formulation to determine the appropriate
endpoints for the chemical, important exposure pathways, and the availability and utility of
unconventional effects data. The more flexible approach is reflected in recent criteria and
proposed criteria that use field data or novel modeling approaches (EPA 2000a, 2003a, 2004a,
2006a). For suspended and bedded sediments, a framework for deriving regional or water-
shed-specific values by multiple methods and weighing the results has been developed (EPA
2006b).
Many nations have criteria for water and other media, and comments about the utility of
the US criteria may not apply to them. The utility of these criteria in risk assessments should
be considered where they are potentially applicable. It is often appropriate to estimate the risk
of exceeding a criterion in addition to estimating risks to ecological endpo ints.
29.2 SCREENING BENCHMARKS
Screening benchmarks are concentrations of chemicals that are believed to constitute thresholds
for potential toxic effects on some category of receptors exposed to the chemical in some medium.
Since they are used for screening chemicals, they should be somewhat conservative so that
chemicals that do in fact cause effects at a particular site are not screened out of the assessment
(Chap ter 31). It is mo re impor tant to ensure that hazardo us che micals are retained than to avoid
retention of chemicals that are not hazardous. However, excessive conservatism decreases the
value of screening assessments, because effort is wasted on nonhazardous chemicals that might
better be expended on the truly hazardous ones. Because of this deliberate conservatism, it is
important to avoid adoption of screening benchmarks as remedial goals or other thresholds for
action without some additional assessment to determine that they are appropriate.
There is little consensus about the best methods for deriving screening benchmarks. The
following alternatives are based on US practices. Screening benchmarks used in Australia,
Europe, and North America are reviewed by Barron and Wharton (2005).
29.2.1 CRITERIA AS SCREENING BENCHMARKS
Criteria are commonly used as screeni ng benchmarks because exceedence of one of these values

constitutes cause for concern. The US NAWQC have been recommended for screening at
contaminated sites by the EPA (Office of Emergency and Remedial Respon se 1996). However,
it is not clear that they are sufficiently conservative, since they are assumed to be sufficiently
close to the true threshold of effects to justify regulatory action and because of other concerns
(Suter 1996c). These concerns are supported by the finding that nickel concentrations in a
waste-contaminated stream on the Oak Ridge Reservation that were below chronic NAWQC
were nonetheless toxic to daphnids (Kszos et al. 1992). When used for regulation of effluents—
their intended purpose—these criteria achieve additional conservatism by being applied to
relatively short exposure durations. That conservatism does not app ly to contaminated sites.
29.2.2 TIER II VALUES
If NAWQC are not available for a chemical, the Tier II method described in the EPA
Proposed Water Quality Guidance for the Great Lakes System or a slight variation used at
ß 2006 by Taylor & Francis Group, LLC.
OR NL may be applied (EPA 1993e; Suter and Ts ao 1996). Tier II va lues wer e de veloped so
that aquati c life crit eria could be conserva tively estimat ed with fewer da ta than are requir ed
for the NAWQC . Tier II values are conc entrations that woul d be expecte d to be higher
than NAW QC in no more than 20% of ca ses, if suff icient test data wer e obtaine d to calculate
NAW QC. For exampl e, if there is only one acu te value (LC
50
or EC
50
) for a ch emical, that
value is divided by 20.5 if it is a daphnid and 242 if it is not. Equi valent factors are available
for other numb ers of acu te values in Appen dix B of Suter and Tsao (1996) . The sources of
data for the Tier II values , and the pro cedure an d fact ors used to calculate the SAVs and
SCVs, are presen ted by EPA (1993e ) and Suter a nd Tsao (1996).
29.2.3 B ENCHMARKS B ASED ON EXPOSURE –RESPONSE MODELS
Screen ing bench marks might be based on low percent iles of exposure–r espo nse relation ships.
In particu lar, one can calcul ate an LC
0

or EC
0
for chemi cals with apparent effects thresho lds.
Alternat ivel y, the practice in human healt h risk asses sment of using the lower 95 % confide nce
limit on a be nchmark dos e (the EC
10
) can be applied to nonhum an organis ms (Linder et al.
2004). Thi s value is consider ed by the US EPA to app roximatel y co rrespond to a no observed
adverse effect level (NOAEL ) for human healt h effe cts, but is more consistent.
29.2.4 T HRESHOLDS FOR S TATISTICAL SIGNIFICANCE
Test endpoints based on statistica l signi ficance are commonl y used as screening bench marks.
The e ndpoint used varies among media and recept ors.
Lowes t chroni c values : Chronic values (CVs ) are geomet ric means of no observed effect
con centrations (NO ECs) and lowest observed effe ct concentra tions (LOEC s). They are used
to calculate the chronic NAW QC , and may be present ed in place of ch ronic criteri a by the
EPA when chro nic criteri a canno t be calcul ated (EPA 1985). CVs are not con servative
ben chmark values.
Wild life NO AELs : Screening ben chmarks for wildlife are conventi onally based on
NOAE Ls from chronic or subch ronic toxicity tests with mamm als or birds. The major
varia bles in deriva tion of wildli fe benchmarks are the test en dpoints used an d whet her
allom etric scali ng or safety fact ors are used . Wildlif e bench marks use reprod uctive or other
effe cts as end points, allometr ic equ ations for inter species extrap olations, and factors to allow
for shortco mings in the test design (Sampl e et al. 1996c; Office of Solid Waste and Emergency
Respons e 2005). The resul ting screeni ng dose, terme d the wi ldlife toxicity reference values
(TRVs) must be co nverted to a concen tration in soil or other medium to screen those media
(Efroy mson et a l. 1997; Office of Solid Waste and Emergency Respons e 2005). That requ ires
an expo sure mod el (Chap ter 22).
29.2.5 T EST ENDPOINTS WITH S AFETY F ACTORS
Some states and EPA regions ba se screeni ng benchmarks on test e ndpoints divide d by safety
fact ors. These fact ors do not have the scientif ic basis of the fact ors used to derive the Tier II

values (above) or the fact ors propo sed by Cal abrese and Bald win (Tab le 26.3) . However, the
use of factors of 10, 100, or 1 000 have a long hist ory in the EPA (Dour son and Stara 1983;
Nabho lz et al. 1 997) (Table 26.1), an d such factors ca n be easily app lied to any test endp oint.
29.2.6 DISTRIBUTIONS OF EFFECTS LEVELS
Sets of screening benchmarks for sediments and soils have been derived from distributions of
effects or no-effects levels. An estimate of the threshold effects concentration for a particular
ß 2006 by Taylor & Francis Group, LLC.
chemi cal is derive d from a pe rcentile of the distribut ion of reported effects or no-effect s
concen trations. Thes e concen trations vary due to varian ce in the phy sical and chemi cal
propert ies of soil s or sedimen ts, varian ce among the measur ed responses , and varia nce in
the sensi tivities of the species or commun ities. Therefor e, the benchmarks de rived in this way
may be tho ught to protect some propo rtion of combination s of specie s, responses , and media .
The foll owing are examples of this approach .
Effect s range- low and e ffects range-med ian for sediment s: The Nation al Oceani c and At-
mospheri c Admi nistratio n (NOAA ) uses three method s: (1) equ ilibrium partiti oning;
(2) spiked sedim ent toxic ity test s; an d (3) field su rveys to develop exposure–r esponse rela-
tions hips (Lon g et al. 1995). Chem ical concentra tions obs erved or estimat ed to be associ ated
with biologi cal effe cts are ranked , and the low er 10th percent ile (effect s ran ge-low, ERL ) an d
the med ian (effect s range- media n, ER M) conce ntrations are identifi ed. A variant of this
approac h is Florida ’s Thresh old Effect s Lev els (MacD ona ld et al. 1996).
Screenin g level concent rations : Thes e bench marks are derive d from synop tic data on
sedim ent ch emical concen trations and benthic inverteb rate dist ributions. They are estimat es
of the highest co ncentra tion that can be tolerated by a specified percent age of benthic species.
Example s include the Ontario Minis try of the Environme nt Lowest and Severe Effect Levels
(Pesaud et a l. 1993).
Oak Ridge Nat ional Laborat ory benchm arks for soil : Bench marks for toxic ity to plants, soil
inverteb rates, an d micr obial pro cesses have been developed from the 10th percen tile dist ri-
butions of toxic ity test data (Efroy mson e t al. 1997a, b).
29.2.7 EQUILIBRIUM PARTITIONING BENCHMARKS
Equilibrium partitioning benchmarks are bulk sediment concentrations derive d from aqueous

criteria or benchmark concentrations based on the tendency of nonionic organic chemicals to
partition between the sediment pore water and sediment organic carbon and for metals to be
bound to sulfides (Sect ion 22.3). The fundame ntal a ssumptions are that pore water is the
principal exposure route for most benthic organisms and that the sensitivities of benthic species
is similar to that of the species tested to derive the aqueous benchmarks, predominantly the
water column species. Examples include the US EPA’s equilibrium partitioning sedim ent
guidelines (EPA 2000b, 2002c–f) and consensus sediment guidelines for PAHs (Swartz 1999).
29.2.8 AVERAGED VALUES AS BENCHMARKS
Sometimes the most sensitive response is thought to be too conservative, criteria for identi-
fying the best value are not apparent, and there is no agreement concerning how to extrapo-
late to a safe level. In such cases, benchmarks may be derived by simply averaging test
endpoints that are deemed to be relevant and of sufficient quality. This approach was used
in the US EPA’s soil screening values for plants and soil invertebrates (Office of Solid Waste
and Emergency Response 2005).
29.2.9 ECOEPIDEMIOLOGICAL BENCHMARKS
When effects are observed in the field and the cause has been de termined (Chapter 4), the
effective exposure levels determined in those studies can be used as benchmarks at other sites.
For example, tund ra swans and other waterfowl were foun d dead or suffering toxicosis in the
Coeur d’Alene Basin, Idaho, an area of lead mining. Field and laboratory studies were used to
relate sediment lead to dietary lead to lead body burdens and effects. The result was an
estimated toxic threshold of 530 mg lead per gram sediment dry weight and a lethal level of
1800 mg=g (Beyer et al. 2000; Henny 2003).
ß 2006 by Taylor & Francis Group, LLC.
29.2.10 SUMMARY OF SCREENING B ENCHMARKS
Curr ently the de velopm ent of screening benchmarks is inconsi stent across media . The large
and relative ly consis tent body of data for aq uatic animals has led to the de velopm ent of more
than a doz en alternati ve types of ben chmarks . Simi larly there are severa l alternati ve bench-
marks for sedimen ts, but they have been developed for fewer chemic als. Wildl ife benchmarks
are nearly alw ays ba sed on NOEC values , so usually only one type of be nchmark is availa ble.
How ever, there is consider able varian ce in what effects are included and in the exposure

models used to extra polate back to soil concen trations. Finally, bench marks for plan ts,
invert ebrate s, an d micro bes in soil are inconsi stent and are ava ilable only for few chemi cals.
Give n the lack of valida tion or even a common definiti on of validity , no singl e type of
ben chmark can be demonst rated to be consis tently reliable. When there are multiple bench-
marks for a chemi cal an d none are clearly superi or, ‘‘cons ensus’ ’ ben chmark values may be
sim ply derive d by av eraging. Swar tz (1999) derived a thres hold effe cts concentra tion for total
PAHs (0.3 mg =g OC) as the arithmet ic mean of five divers e bench marks. He found that it
was a reasonab le thres hold value for PAH effec ts in inde pendent da ta sets from PAH-
con taminate d sites. Alternat ively, the unc ertainty co ncerning the most app ropriate bench-
mark may be treat ed by cho osing the low est be nchmark for each chemi cal.
Bec ause the degree of conserva tism of benchmarks is uncerta in, concerns that truly toxic
chemi cals may be screened out may be relieved by using unc ertainty factors. An exampl e of
the use of unc ertainty fact ors for this purpo se is the eco logical risk assessment for the Rocky
Mo untain Arsenal, in which factors were a pplied to account for intrataxon variability,
inter taxon variab ility, uncerta inty of critical effect, exp osure duratio n, en dpoint extrapo la-
tion, and resid ual unc ertainty (Banton et al. 1996). For each of these six issue s, a fact or of 1,
2, or 3 was applie d signi fying low , medium , or high uncerta inty, respect ivel y. Clearl y, the
magni tudes of these factors are not related to estimat es of actual varian ce or unc ertainty
associ ated with each issue, and the multiplicat ion of fact ors bears no relat ionship to any
estimat e of the total uncerta inty in the ben chmarks . Ho wever, uncerta inty factors pro vide an
assura nce of conserva tism withou t appeari ng complet ely arbitrary . An alternati ve is to
derive unc ertainty factors based on estimat es of actual varian ce or uncerta inty. An exampl e
is the pred iction inter vals on the inter taxon extrap olations and the unc ertainty factors on the
predict ion inter vals (PIs) for a given taxonom ic level present ed in Table 26.2 through
Table 2 6.5.
ß 2006 by Taylor & Francis Group, LLC.
30
Integrating Exposure and
Exposure–Response
The primary task of risk characterization is to integrate the exposure estimates from the

analysis of exposure with the exposure–response relationships from the analysis of effects to
estimate the nature and magnitude of risks. In effect, response is esti mated by solving the
exposure–response function for the exposure estimate. In most assessments, this task has
been performed by simple methods that require little thought. However, as more attention
is paid to varia bility and uncerta inty (Chapter 5), probabil istic methods are be coming more
common.
30.1 QUOTIENT METHODS
If the analysis of exposure has generated a point estimate of exposure (e.g., the maximum
measured concentration) and the analysis of effects has reduced the exposure–response
relationship to a point (e.g., an LC
50
), integration of the two reduces to the quotient method.
The hazard quotient (HQ) is the quotient of an exposure concentration (C
e
) divided by a
toxicological benchmark concentration (C
b
):
HQ ¼ C
e
=C
b
(30:1)
Because this is a widely used assessment method, the terms have many representations. In
Europe, C
e
is usually termed the predicted environmental concentration (PEC) and C
b
is
termed the predicted no effect concentration (PNEC). If exposure and effect are expressed as

doses, the HQ is equivalent [D
e
=D
b
]. The same simple model may be applied to a variety of
agents such as temperature, percent fines, and radiation. Because of its simplicity, the
quotient method is nearly always used in screening assessments, but it is also the most
common method of risk characterization in definitive assessments.
Although some assessors have used Monte Carlo analysis (Chapter 5) to perform prob-
abilist ic analys es of HQs (as in the Hong Kon g exampl e, Sectio n 30.7.3 , an d Zolezz i et al.
2005), they may be performed analytically (IAEA 1989; Hammonds et al. 1994). The quotient
model can be expressed as:
ln HQ ¼ ln C
e
À ln C
b
(30:2)
HQ will be approximately log-normal even if the distributions assigned to C
e
and C
b
are not
(IAEA 1989; Hammonds et al. 1994). Hence, the geometric mean of HQ is the antilog of the
difference of the means of the logs of C
e
and C
b
, and the geometric variance is the antilog of
the sum of the variances of the logs of C
e

and C
b
.
ß 2006 by Taylor & Francis Group, LLC.
If the number of exposure values and effects values are finite, one may simply determine the
distribution of all possible values of HQ. For example, in an assessment of risks to pond
communities from pyrethroid pesticides, Maund et al. (2001) determined the distribution of
quotients for 90th percentile concentrations in each of 72 pond categories with each acute and
chronic toxicity datum (Figure 30.1).
While the HQ expresses how bad things are, a related concept, the margin of safety,
expresses how good they are. The relative margin of safety is simply the inverse of the HQ.
A relative margin of safety of 100 suggests that the exposure concentration must be increased
by a factor of 100 to reach a toxic level. The absolute margin of safety is the difference
between a toxic level and the exposure level. An absolute margin of safety of 100 mg=L
suggests that the exposure concentration must be raised by that amount to reach a toxic level.
An example of the use of margins of safety in ecological risk assessment is presented by
Newsted et al. (2002).
30.2 EXPOSURE IS DISTRIBUTED AND RESPONSE IS FIXED
Frequently, the exposure–response relationship is reduced to a point, such as a criterion
value, but the exposure estimate is distributed. The exposure distribution may come from the
distribution of measured concentrations in the environment, from Monte Carlo analysis of
a transport and fate model or from expert judgment. In such cases, the probability of
exceeding the benchmark value (C
b
) is the integral of the probability density function above
C
b
(i.e., 1—the cumulative probability at C
b
). An example of this approach is the analyses of

risks to herons and egret s in Hong Kong with determ inate effects thresh olds (Sect ion 30.6).
10
−2
10
−1
Risk quotient
10
0
10
1
10
−3
0.1
Exceedance probability (%)
0.5
1
2
5
10
20
30
40
50
Water column instantaneous EEC: invertebrate acute toxicity
Water column 96 h EEC: invertebrate acute toxicity
Water column 21 d EEC: invertebrate chronic toxicity
FIGURE 30.1 Distributions of acute and chronic quotients for invertebrates from an assessment of risks
to pond communities from pyrethroid pesticides. EEC ¼ estimated exposure concentration. (From
Maund, S.J., Travis, K.R., Hendley, P., Giddings, J.M., and Solomon, K.R., Environ. Toxicol. Chem.,
20, 687, 2001. With permission.)

ß 2006 by Taylor & Francis Group, LLC.
30.3 BOTH EXPOSURE AND RESPONSE ARE DISTRIBUTED
Given distributions of exposure and response with respect to a common variable (e.g.,
concentration), one may calculate risk as the probability that a random draw from the
exposure distribution exceeds a random draw from the response distribution (Suter et al.
1983). This concept of risk as the joint probability of exposure and effects distributions was
applied to effects expressed as species sensitivity distributions (SSDs) by Van Straalen (1990)
and Parkhurst et al. (1996a,b). Risk is the integral of the product of the probability density of
the exposure concentration C
e
and the cumulative distribution of the benchmark concentra-
tion C
b
(Figure 30.2c). The derivation of this form ula and alternatives, including a discrete
approximation, are clearly presented by Van Straalen (2002b). This is c onceptually equivalent
to the prob abilist ic HQs (Sect ion 30.1) but is both clear er and more elegant .
A variant of this approach is proposed by the ECOFRAM Aquatic Workgroup (1999) and
applied to pesticide risk assessments (Giddings et al. 2005) as well as contaminated site
assessments (Moore et al. 1999). From an exposure distribution (proportion of locations,
times, or episodes, with respect to concen tration) and an effects distribution (SSDs or other
exposure–response distributions) one can derive a plot of exposure proportions vs. effects
levels that is called a risk curve (Figur e 30.3 an d Figure 30.4) . Since both proportio ns of
exposures and responses are distributed with respect to concentrations, there are correspond-
ing values of each. The area under the curve is called the mean risk. It is equivalent to risk
estimated as a joint probability, discussed earlier.
(a) (c)
p(c)
N(c)
(d)
p(c)

N(c)
Probability
Probability density
Probability density
Probability density
Probability density
Probability
Probability
Concentration Concentration
Concentration Concentration
1-P(c)
(b)
1-P(c)
n(c)
1
1
1
0
Probability
1
0
δ
n(c)
δ
δ
δ
FIGURE 30.2 Graphical representations of the estimation of ecological risks (d) defined as the prob-
ability that exposure concentrations are greater than no-effect concentrations (NECs). The probability
density of exposure concentrations is denoted as p(c), the distribution of NECs is denoted as n(c). P(C)
and N(C ) are the corresponding cumulative distributions. In a and c, both variables are distributed. In b

and d, the exposure concentration is assumed to be constant. (From Van Straalen, N.M., in
L. Posthuma, G.W. Suter II, and T. Traas, eds., Species Sensitivity Distributions in Ecotoxicology,
Lewis Publishers, Boca Raton, FL, 2002. With permission.)
ß 2006 by Taylor & Francis Group, LLC.
When exposure and effects distributions are used as part of a logical weighing of evidence
(Ch apter 32), it may be appropri ate to logically inter pret them rather than calculati ng joint
probabilities. For example, the following interpretation occurs in the risk assessment for fish
community of the Poplar Cree k embayment of the Clinch River (Suter et al. 1999).
Copper. The dist ributions of ambient copper concentrations and aqueous test endpoints are
shown in Figu re 30.5. The ambie nt concen trations were dissolved phase conc entrations in the
subreaches (3.04 and 4.01) with potentially hazardous levels of Cu. The toxic concentrations
were those from tests performed in waters with hardness approximately equal to the site
Percent of species affected
0
Probability of exceedence
0
20
40
60
80
100
10 20 30 40 50 60 70 80 90 100
Concentration
42% probability,
>5% of species affected
4% probability,
>25% of species affected
10
Probability of exceedence
0

20
40
60
80
100
Percent of species affected
30
10
1
50
70
90
99
100 1,000 10,000 100,000
Exposure
distribution
Toxicity
distribution
42% exceedence
25% of species affected
5% of species affected
4% exceedence
A
B
Risk curve
A
B
FIGURE 30.3 A demonstration of the derivation of a risk curve from distributions of exposure
(probability of exceedence) and response (percent of species affected) with respect to concentration.
(From Giddings, J.M., Anderson, T.A., Hall, L.W. Jr., et al., Atrazine in North American Surface

Waters: A Probabilistic Risk Assessment, SETAC Press, Pensacola, FL, 2005. With permission.)
ß 2006 by Taylor & Francis Group, LLC.
water. The ambient concentrations fall into two phases. Concentrations below 0.01 mg=L
display a fairly smooth increase suggestive of a log-normal distribution. The upper end of this
phase of the distribution (above the 75th percentile of 4.01 and the 80th percentile of 3.04)
exceed the lowest chronic value (CV) (a bluntnose minnow CV for reproductive effects).
However, the distributions above the 90th percentile are not continuous with the other points.
The break in the curve suggests that some episodic phenomenon causes exceptionally high
concentrations. The two points in 4.01 and one in 3.04 that lie above this break ex ceed
approximately 90% of the CVs, approximately 30% of the acute values, and both the acute
and chronic Nati onal Ambient Water Quality Criteria (NAWQC). These results are suggest-
ive of a small risk of chronic toxicity from routine exposures, but a high risk of short-term
toxic effects of Cu during episodic exposures in lower Poplar Creek embayment and the
Clinch River.
This sort of interpretation is a mixture of quantitative and qualitative analysis that can, as
in this case, provide more information than a purely quantitative analysis. Had the Aquatic
Risk Assessment and Mitigation Dialog Group criterion been applie d or a joint probability
been calculated, the results would have been less ad hoc but would have provided less basis
for inference concerning the cause of observed effects and toxicity.
30.4 INTEGRATED SIMULATION MODELS
When a mathematical simulation model, such as a chemical transport and fate model
(Chap ter 21), is used to estimate exposure, one may simply add the exposure–r espon se
function to the exposure model so that the model output is an effects level. Similarly, when
a population or ecosystem model is used to estimate responses to exposures, an exposure level
0
Exceedence frequency (%)
0
20
40
60

80
100
Tier 2, Region 2, corn
10 20 30 40
Ma
g
nitude of effect (%)
50 60 70 80 90 100
Rainbow trout Daphnia magna (chronic)
Lemna gibba
FIGURE 30.4 Risk curve for atrazine applied to corn based on estimated annual maximum instantaneous
concentrations in pond water in a defined region, for three endpoints: duck weed (Lemna gibba) growth
inhibition, cladoceran (Daphnia magna) reproduction inhibition, and rainbow trout mortality. Because of
the relative insensitivity of acute lethality to trout, the corresponding line appears to be vertical in this plot.
(From Giddings, J.M., Anderson, T.A., Hall, L.W. Jr., et al., Atrazine in North American Surface Waters:
A Probabilistic Risk Assessment, SETAC Press, Pensacola, FL, 2005. With permission.)
ß 2006 by Taylor & Francis Group, LLC.
or an exposure model my be linked to produce an integrated model that estimates effects from
loading rates or a mbient level s (O’Nei ll et al. 1982). Monte Carlo ana lysis (Chapter 5) is used
with such models to estimate risks as probabilities of effects.
30.5 INTEGRATION OF SENSE AND NONSENSE
When integrating exposure with exposure–response relationships, it is essential to ensure that
they can be combined in a way that makes sense, i.e., they must be concordant. This requires
first that the common units be consistent. This is not simply a matter of assuring that, for
example, the exposure concentration and the concentration in the exposure–response rela-
tionship are both mg=L of copper. If the response concentration is a 96 h LC
50
for dissolved
copper, and the exposure concentration is an annual average of measured total copper
concentrations, they are not concordant. Other measures of exposure or response must be

used, or one of the measures must be adjusted to achieve concordance. For example, a metal
speciation model may be used to estimate dissolved copper concentrations in the field and the
peak 96 h concentration might be estimated from the time series of measurements.
Concordance becomes more complex when parameters are expressed as distributions and
results are expressed as probabilities. For every dist ribution, it is essential to ask what is
distributed and with respect to what it is distributed. Is a distribution of dose to mink
(mg=kg=d) the distribution of the average dos e across a mink population, the dose to the
0.001
0
20
40
Percentile
60
80
100
0.01 0.1
Concentration (m
g
/L)
1
Invertebrate: acute
Fish: acute
CVs
Acute NAWQC
Chronic NAWQC
Reach 3.04
Reach 4.01
10
FIGURE 30.5 Empirical distribution functions (species sensitivity distributions, SSDs) for acute toxicity
(LC

50
and EC
50
values) and chronic toxicity (chronic values) of copper to fish and aquatic invertebrates,
and for individual measurements of copper in surface water from two stream reaches. Vertical lines are
acute and chronic National Ambient Water Quality Criteria (NAWQC).
ß 2006 by Taylor & Francis Group, LLC.
media n mink, or the dos e to an indivi dual mink occup ying a pa rticular locat ion? Is it
distribut ed with respect to space (e.g., from sampl ing points on a site), to time (e.g ., from
year- to-year variation in diet), to individua ls (e.g., from varian ce in size and dieta ry pr efer-
ence), or to degree of belief (e.g., an expression of an asses sor’s unc ertainty concerning the
dose esti mate)? If it was gen erated by Monte Car lo analysis of an exp osure model, the dose
distribut ion might be a hodgepo dge of variance of con sumption rates across individu als,
varia nce in drinkin g wat er contam ination over time, varia nce in contam inant level s
across individ ual fish in a pond, variance in contam inant levels in mice across space, an d
varia nce in dieta ry composi tion across different studi es. The best that could be said of such a
dose dist ribution is that the probabil ities express the asses sor’s uncerta inly con cerning dose as
degrees of be lief.
A response distribut ion (e.g ., one derive d from a reproduct ive test) may also take different
form s. In the sim plest case, a dose–res ponse distribut ion may be derive d for the propo rtional
reducti on in the num ber of live births per fema le. That dist ribution might be us ed to estimat e
the average proporti onal reductio n at a given dos e, or the varia nce in the parame ters of the
fitted mod el might be used to estimat e the dist ribution with respect to ind ividual fema les of
the dose causing a given prop ortional redu ction (e.g., an ED
10
). If the test was performed with
rats rather than mink, an unc ertainty fact or may be app lied resulting in a distribut ion of the
ED
10
with respect to degree of belie f (Bo x 30.1) .

Note that any pa ired exposure dist ribution and effe cts distribut ion from the previous tw o
paragra phs superf icially app ear to be con cordant, becau se they are all proba bilities as
functio ns of dose to mink. How ever, the probabil ities ex press very different qualities .
The appro priate integ ratio n of ex posure an d response depend s on the asses sment endpo int,
the prefer ences of the risk manage r, and the available infor mation. In the sim plest case of the
mink exampl e, one might use an HQ. The poin t esti mate of the an nual average daily dose for
the site might be divide d by the ED
10
and the risk could be dec lared signi ficant if HQ > 1. To
include uncerta inty, one might sub jectivel y estimat e a low er confide nce bound of HQ= 100 for
use in screeni ng, or to be precau tionary. One might estimat e the distribut ions of both
exposure and respon se dos es and esti mate the distribut ion analytical ly. Howe ver, to a ctually
estimat e the most likel y effe ct or the risks of prescribed levels of effect, one would need to
settle on a con cordant set of distribut ions of exp osure and response parame ters . For exampl e,
to e stimate the probabil ity that a female mink on a con taminate d site experie nces a repro-
ductiv e de crement, one co uld use the dist ribution of the ED
10
(i.e., the pr oportio n of females
with fecun dity at least 10% below control s) an d a distribut ion of dos e with respect to
individ ual female mink esti mated as describ ed in Secti on 22.11. Since both the exposu re
dose and the effective dose woul d be distribut ed with respect to indivi dual female mink, the
joint probab ility would be the prop ortion of fema les in the site popul ation estimat ed to
experi ence a 10% or great er redu ction in fecundi ty.
The problem of combini ng exposure an d response dist ributions may be simplified by
devisin g rules like the one provided by the Aquat ic Ris k Assessm ent and M itigation Dialo gue
Group (1994) (see also Solomon et al. 1996). They reduced risk characterization for exposure
and effects distributions to a dichotomous criterion; the risk is significant if the 90th percentile
of the distribution of aqueous concentrations exceeds the 10th percentile of the SSD. Al-
though this method has been recommended by a distinguished group of scientists, the
criterion is not supported by legal or policy considerations. In addition, it does not interpret

the distribut ions in terms of either variability or uncertainty. It is simply an easy an d
consistent rule, which may be adopted or adapted to an assessment if it makes sense.
The possible combinations of distributions to characterize ecological risks are effectively
infinite, so this section can give only a sample of the possible problems in achieving concordance.
It is incumbent on those who perform probabilistic risk assessments to carefully consider what is
ß 2006 by Taylor & Francis Group, LLC.
distributed, with respect to what it is distributed, and in what sense do probabilities from the
distribution constitute risks to the assessment endpoint. If the nature of the probabilities is
unclear in the assessor’s mind, it will not be clear to the users and reviewers of the assessment,
and there is a good chance that it is wrong. In such cases, assessors should consider seeking help or
using a less complex analysis.
30.6 INTEGRATION IN SPACE
In regional assessments or other assessments at large spatial scales, the assessment must
integrate risks over space. The most common approach is to divide the area into reasonably
uniform units, estimate risks for each unit, and then generate a summary such as an area-
weighted average effect or a distribution of effects across units to estimate risks on a site or in
a region. The spatial units might be habitat types on a site, watersheds in a region, areas with
distinct types of disturbance or contamination, or other relevant divisions of the area being
BOX 30.1
Variability, Uncertainty, and Distributions of Effects
Conventionally, the relationship between an exposure variable (dose, concentration, duration,
etc.) and a response variable (growth, death, etc.) is quantified using a distribution function
(normal, logistic, etc.), the exposure–response function (Chapter 23). In ecological risk assess-
ment, these functions are most commonly based on responses of exposed individual organisms
(dose–response or concentration–response) or individual species (species sensitivity distributions,
SSDs). These distribution functions are usually described as probability distributions without
careful thought as to what mechanism generated the distribution.
Error: All individuals may be effectively the same or all species in a community may be
effectively the same and the distribution is due to random effects in the tests (these random effects
are termed error, but need not imply actual errors in conducting the tests). Hence, the output of

the model is the probability of the prescribed response given the uncertainty due to experimental
error (randomness).
Variability: There are actual differences among the individuals or species that are measured
without effective error. Hence, the distribution describes that variability and the output of the
model is the deterministic proportion of individuals or species responding at a given exposure
level. This is the assumption underlying the estimation of risks to human populations, and the
most common assumption in the case of species distributions, leading to the calculation of the
potentially affected fraction (PAF) of species.
Identity: There are actual differences among the individuals or species that are measured
without effective error. However, we are interested in the risk to an untested individual or species,
rather than the proportion of individuals in a population or species in a community. This
interpretation leads to an estimate of individual risks (probability of effects on an exposed
organism other than a member of the test population) or species risks (probability of effects on
an untested species). This is the assumption underlying the estimation of risks to human individ-
uals. In the case of species distributions, this assumption implies that the endpoint species is a
random draw from the same population as the set of test species used to define the distribution.
Extrapolation: The variability and uncertainty inherent in the test and the model fitted to the
test data are negligible relative to the uncertainty associated with extrapolation to the endpoint
species or community. In such cases, we may ignore the variance among organisms or other test
units, estimate a midpoint of the observed responses (e.g., LC
50
or EC
50
) and assign a standard
deviation, range, or other distribution parameter to express extrapolation uncertainty.
ß 2006 by Taylor & Francis Group, LLC.
asses sed. In most cases, the exp osure esti mates vary amon g units , but the expo sure–re sponse
relationshi ps may vary as well due to diff erences in the biotic communi ties.
A sop histicate d elabora tion of this ap proach is found in the asses sment of aq uatic eco-
logic al risks from cotton pyrethr oids in Yazoo Count y, Mississipp i (Hendl ey e t al. 2001;

Travis and Hendley 2 001). The units wer e ponds and their associ ated watershe ds. A geo-
graphic infor matio n syst em (GIS) was used with trans port an d fate models to estimate
90th percen tile concen trations in 597 ponds in the county an d compare them to SSDs for
acute an d chron ic effects of the pesticides . A furt her step woul d include spatial v ariance or
even spatial dynami cs of the en dpoint organ isms, popul ations, or co mmuni ties alon g with the
spatial varia nce in expo sure.
Anothe r ap proach estimate s risks at points ; usuall y, points at whi ch soil or sedimen t has
been sampl ed an d a nalyzed. Kriging, Thiesse n pol ygons, or some other geospat ial statistica l
method is then used to de fine areas within whi ch risks fall in de fined ranges. Thi s a pproach is
approp riate for organisms with littl e mobil ity such as plants and be nthic inverteb rates. If
toxic ity tests are perfor med on soil or sediment sampl es, and if the tests are measur es of effects
on an asses sment endpoint, this approach c an be applie d to tho se resul ts as well (Figur e 20.3).
Figure 30.6 illustr ates a simple techn ique that is appropri ate for organisms with terr itories
or hom e ranges occurrin g on relative ly sim ple contam inate d sites. A contam inant has be en
dumped or spilled at a point, and the average soil concen tration dro ps off app roximatel y
expon entially from that point as the area average d increa ses. Equi valent curves could be
plotted for other pa tterns of con taminatio n around a point. Hori zontal dashed lines indica te
soil concentrations that are estimated to be thresholds for effects on small mammals and
birds. The vertical dashed lines indicate the average home range or territory size for the
endpoint species (shrew and woodcock in this case). If the average concentration falls below
the effects concentrations before intersecting the home range size, not even one individual is
01 6
Area (acre)
0
200
400
600
800
1,000
Concentration in soil (mg/kg)

Baseline
conditions
Remedial action for
soils containing at
least 150 mg/kg
Acute lethal effect level for song birds
Chronic lethal effect level for small mammals
Reproductive effect level for song birds
Reproductive effect level for small mammals
Shrew
Woodcock
2345
FIGURE 30.6 Mean concentration of a contaminant as a function of averaging area centered on the
point of maximum concentration. Vertical lines indicate the home range area for a shrew and a
woodcock. Horizontal dashed lines indicate soil concentrations estimated to produce toxic doses
associated with specific effects. (Graphic redrawn from an unpublished graphic provided by C. Menzie).
ß 2006 by Taylor & Francis Group, LLC.
expecte d to be affected. In Figu re 30.6, no effects on woodcock are exp ected, but a shrew is
estimated to experience reproductive effects and has a marginal risk of death.
To estimate risks to multiple organisms with territories or home ranges on a contaminated
site or other defined area, a GIS can be used to cover the area with polygons having the area
(on the map scale) of the territory or home range. Exposure levels in those areas can then be
averaged to estimate risks to those individuals or reproducing pairs.
30.7 EXAMPLES
The following examples provide a taste of the range of approaches to integrating exposure
and exposure–response information in ecological risk assessments.
30.7.1 SHREWS ON A MERCURY-CONTAMINATED SITE
Talmage and Walton (1993) collected shrews on the mercury-contaminated floodplain of East
Fork Poplar Creek, Tennessee. They analyzed the mercury concentration in kidneys, the
target organ, and compared them to the 20 mg=g threshold for mercury toxicity in rodents.

They found that 75% of shrews exceeded that threshold.
30.7.2 EGRETS AND EAGLES IN SOUTH FLORIDA
A large stormwater treatment pond in South Florida was found to have high methyl mercury
concentrations. To obtain a discharge permi t, the State agreed to assess risks to great egrets
and bald eagles foraging on the pond (Rumbold 2005). The exposure–response relationships
were lowest observed adverse effect levels (LOAELs) divided by a factor of 3. Exposure was
estimated from concentrations in fish collected from the pond and from dietary uptake
models for pre-nesting females and nestlings. The nestling exposure model also included
maternal mercury deposited in the eggs. Monte Carlo simulation of the uptake models was
used to estimate the distributions of exposure, based on variance in the mercury concentra-
tions in appropriate fish for each avian species. The assessment found that risks of exceeding
the effects thresholds were low and similar to other areas in the region.
30.7.3 EGRETS AND HERONS IN HONG KONG
Connell et al. (2003) assessed risks from organochlorine compounds to the reproduction of
black-crowned night herons and little egrets in the New Territories of Hong Kong. They
estimated exposure by analyzing contaminants in eggs and established that DDE posed the
greatest hazard. They developed a concentration–response relationship using published stud-
ies of the relationship between survival of young ardeids and DDE concentrations in eggs
(Figur e 30.7). They judge d that 1000 ng=g DDE was a thres hold for significa nt redu ction in
survival. By applying that value to the probability density functions for egg concentrations,
they estimated that 12.4% of night herons and 40.9% of egrets were exposed at levels
exceeding the threshold. Finally, they used Monte Carlo simulation to estimate the probabil-
ity of exceeding the threshold given the uncertainty in the threshold. However, they consid-
ered only the possibility that the threshold was underestimated.
In a companion study, Connell et al. (2002) related the distributions of metal concentra-
tions in heron and egret feathers to effects thresholds from the literature. In this case, the
Monte Carlo analysis used the observed distribution of concentrations in feathers and a
uniform distribution of the effects threshold between the highest level reported to have no
effects (3 mg=g mercury) and the lowest level reported to reduce reproductive success in
ardeids (5 mg=g).

ß 2006 by Taylor & Francis Group, LLC.
30.7.4 BIOACCUMULATIVE CONTAMINANTS IN A STREAM
After completion of the remedial investigation for East Fork Poplar Creek in Oak Ridge,
Tennessee, the US Department of Energy commissioned a new ecological risk assessment to
test new probabilistic techniques (Moore et al. 1999). Exposure of belted kingfishers and mink
to mercury and polychlorinated biphenyls (PCBs) was estimated using Monte Carlo simula-
tion of a multiroute model including inhalation, drinking, and feeding. The exposure–
response relationships were estimated using generalized linear modeling of published test
data. A single best study was used for kingfishers, but for mink data from multiple tests were
combined to generate the dose–response functions. The exposure an d exposure–response
functio ns wer e co mbined to generat e risk curves (Figur e 30.8). Thes e an alyses showe d
significant risks to mink and kingfishers from mercury and to mink from PCBs. These results
differed from the assessments for the remedial investigation, which found little risk to
piscivorous wildlife. However, that assessment did not use measured co ncentrations in fish,
but rather modeled uptake based on contaminants in the sediment, which was the medium
that would be remediated under Superfund (Burns et al. 1997). The results by Moore et al.
(1999) gave more similar resul ts to a reservation-wide wildlife risk asses sment that also
used measured fish concentrations (Sample et al. 1996b). The exception was PCBs in
mink, and the difference was that Sample et al. (1996b) used the LOAEL as a threshold.
Consideration of the dose–response relationship showed that PCBs caused large reproductive
effects at the LOAEL.
30.7.5 SECONDARY POISONING IN HAWAII
An ecological risk assessment of possible secondary poisoning in Hawaii provides an example
of Monte Carlo simulation of an integrate d exposure and effects model (Johnston et al. 2005).
Broadcast baits contai ning rodenticides are consumed by invertebrates, which are in turn
10 100 1,000 10,000 100,000
0
20
40
60

80
100
Night herons
Little egrets
Findholt and Trost (1985)
Henny et al. (1984)
Findholt (1984)
0
20
40
60
80
100
DDE Concentration in e
gg
s (n
g
/
g
; lo
g
scale)
Cumulative probability (%)
Reduction in survival of young (%)
FIGURE 30.7 Cumulative distribution of concentrations of DDE in eggs for night herons and little
egrets and a model with confidence interval fitted to data on percent reduction in the survival of young
birds as a function of DDE in eggs. The vertical line is the estimated threshold effective concentration of
1000 ng=g. (From Connell, D.W., Fung, C.N., Minh, T.B., Tanabe, S., Lam, P.K.S., Wong, B.S.F.,
Lam, M.H.W., Wong, L.C., Wu, R.S.S., and Richardson, B.J., Water Res., 37, 459, 2003. With
permission.)

ß 2006 by Taylor & Francis Group, LLC.
con sumed by birds. As illustr ated in Figure 30.9, the acute dose to Po’ouli (a honeycreeper )
was esti mated from the distribut ion of co ncentra tions in snails and slugs from treat ed areas,
and from estimat ed dist ributions of sn ail con sumption rates derived from caloric req uire-
ments , fraction moll uscs in diets , and en ergy content of moll uscs. The exp osure–r esponse
model was de rived from distribut ions assign ed to the avian LD
50
, the slope of the dose–
respon se relationshi p, and the sensitiv ity of honeycreeper s relat ive to avian test sp ecies.
Mo dels were also develop ed for 5 and 14 d ex posures. Median probab ilities of acute mort ality
wer e 0.03% for adu lts and 0.57% for juveniles (Figur e 30.10) .
30.7.6 ATRAZINE
The ecologi cal risk asses sment for atraz ine (Sect ion 32 .4.4) illustr ates the use of risk curves
to integ rate exp osure and expo sure–r esponse distribut ions (Sect ion 30.3). The exposure–
response distributions are SSDs. The exposure distributions include both distributions of
measured concentrations and distributions of concentrations from Monte Carlo analyses of
models of pond scenarios.
30.7.7 WARMING SUBALPINE FORESTS
Bolliger et al. (2000) estimated risks from climatic warming to the abundance and distribution
of tree species in the Alps. The exposure–response relationships were logit regression models
0
2,000
4,000
6,000
8,000
10,000
0 100 200 300 400
0 200 400 600
Exposure (μg/kg bw/day)
Cumulative frequency

0
1
2
3
4
5
6
0 5 10 15
Concentration (μg/g)
#Kits/Female
0
20
40
60
80
100
Dose (μg/kg bw/day)
Decline in fecundity (%)
0
20
40
60
80
100
0 20406080100
Decline in fecundity (%)
Exceedence probability
(a)
(c)
(b)

(d)
FIGURE 30.8 (a) Cumulative distribution function for female mink exposure to polychlorinated
biphenyls (PCBs). (b) Dietary concentration–response curve for effects of PCBs on mink fecundity.
(c) Dose–response distribution estimated from (b) and food intake rates. (d) Risk curve for fecundity of
female mink exposed to PCBs. (From Moore, D.R.J., Sample, B.E., Suter, G.W., Parkhurst, B.R., and
Teed, R.S., Environ. Toxicol. Chem., 18, 2941, 1999. With permission.)
ß 2006 by Taylor & Francis Group, LLC.
of the presence or absence of each major tree species at locations as a function of five
biophysical habitat properties. This model was derived by fitting to forest inventory data,
using a map of degree-days, radiation in July, summer frost frequency, July water budget, an d
slope. Exposure was expressed as current conditions and three scenarios: warming of 100,
200, and 400 degree-days, converted into the biophysical variables. The results showed little
change in overall abundance, but distributions shifted, and species did not move together. In
particular, spruce and beech, which currently occur together became segregated, leaving the
subalpine belt dominated by spruce.
30.8 SUMMARY
The critical step in the characterization of ecological risks is the integration of exposure
estimates with exposure–response relationships to estimate effects or probabilities of effects.
Diphacinone
concentration
in snails (µg/g)
Weight of
daily diet
(g food/g bw)
Daily energy
requirements
(kJ/g bw)
Fraction of
insects in
diet

Fraction of
snails in
diet
Energy
content of
insects
(kJ/g)
Energy
content
of diet
(kJ/g)
Energy
content of
snails
(kJ/g)
Weight of
snails
(g/g bw)
0.88
1.90
2.92
1.05 1.50
1
1.95
0.31 0.61 0.90 0.31 0.61 0.90
2.10 6.20 10.30
0.35 1.77 3.18
Probability of
mortality
vs.

diphacinone
dose
Slope
LD
50
Interspecies
extrapolation
Probability of mortality
0.40 0.54 0.68
8.45 9.18 9.91
0

04
125
0
6
12
0
0.4 0.8
1.2
Log dose (mg/kg)
Probit
Weight of
insects
(g)
Dose
(mg/kg bw)
FIGURE 30.9 Diagram of a probabilistic ecological risk assessment for a single-day acute exposure of a
Hawaiian honeycreeper to diphacinone. Dose is estimated from measured concentrations in snails and
the consumption of snails estimated from energy requirements, dietary habits, and energy contents of

dietary items. The exposure–response relationship is estimated from a reported avian LD
50
and slope
values and a distribution of safety factors (1–25) for interspecies extrapolation. (From Johnston, J.J.,
Pitt, W.C., Sugihara, R.T., Eisemann, J.D., Primus, T.M., Holmes, M.J., Crocker, J., and Hart, A.,
Environ. Toxicol. Chem., 24, 1557, 2005. With permission.)
ß 2006 by Taylor & Francis Group, LLC.
It may be deterministic or probabilistic based on varia bility or uncertainty in either compon-
ent. The critical considerations are the relevance of any of the sources of variability and
uncertainty to the decision and the concordance of the units of the two components and of the
distributions.
Percent
0.000
0.250
0.500
0.750
1.000
0
10,000
0.00
Probability
Frequency
Frequency
Probability
1.38 2.76 4.14 5.52
Adult Po’ouli
0.000
0.250
0.500
0.750

1.000
0
10,000
0.00 1.10 2.21 3.31
4.41
Juvenile Po’ouli
Percent
Mean mortality = 0.03%
Mean mortality = 0.57%
FIGURE 30.10 Results of the assessment methodology shown in Figure 30.9. Probabilities of mortality
and equivalent frequencies are presented as functions of percent occurrence. (From Johnston, J.J., Pitt,
W.C., Sugihara, R.T., Eisemann, J.D., Primus, T.M., Holmes, M.J., Crocker, J., and Hart, A., Environ.
Toxicol. Chem., 24, 1557, 2005. With permission.)
ß 2006 by Taylor & Francis Group, LLC.
31
Screening Characterization
Risk characteriz ation in a screeni ng eco logical risk assessment con sists of using exposure an d
effects informat ion to screen the risks into categor ies. The most general categorizat ion is:
.
De minim is—ris ks that are clear ly insi gnificant and can be ignored in subsequent assess-
ments or decisio ns
.
Indeter minate—ri sks that are not clear ly signi fican t or insi gnificant and must be resol ved
by furth er assessment or con siderati on of other issue s such as co sts, benefi ts, engineer ing
feasibil ity, or public concerns
.
De manifes tis —ris ks that are clear ly signifi cant and need not be assessed further but
should be referred to the risk manager for remedi ation or control
These Latinisms are derived from legal terminology. In particular, some risks or other legal
issues are so small that they are considered trivial (de minimis) and therefore not worthy of the

law’s attention (Travis et al. 1987; Whipple 1987). Commonly, this three-part logic is reduced to
two parts by replacing de manifestis and indeterminate risks with a single category, nontrivial
risks. In that case, the nontrivial risks are carried forward to subsequent assessments.
Screening ecologi cal risk assessment s have a num ber of potenti al uses.
To prompt acti on : In some cases, a screening assessment will reveal that risks are manif estly
signifi cant, and remedi al or pr eventative action sho uld be taken wi thout furt her data c ollec-
tion or asses sment .
To deter mine the need for furt her asses sment : A screeni ng assessment may reveal that
signifi cant risks are highly unlikel y or that the risks are of low priority relative to other
risks or relative to the likely costs of asses sment and managem ent.
To d efine the scope of a definit ive asses smen t: A screenin g assessment may screen out certa in
contam inants , recept ors, media , routes of exposure, or porti ons of a site by demo nstrating
that they are associated with negligible risks.
To guide data collecti on : Data colle ction for subsequen t tiers of asses sment may be focused
on the hazards that have not been screened out as clearly insignificant or to focus testing or
modeling on routes of exposure or mechanisms of action that are likely to be associated with
an agent.
In some assessment schemes, particularly rule-based assessments of industrial chemicals,
there is no defini tive assessment in the sense that risks are neve r estimat ed (Figur e 3.8).
Rather, data are generated and screening approaches are applied until the chemical falls into
the acceptable or unacceptable category.
31.1 SCREENING CHEMICALS AND OTHER AGENTS
When characterizing risks in a screening assessment it is not necessary to estimate the nature,
magnitude, or probability of effects, but it is necessary to assure that hazards fall into the
ß 2006 by Taylor & Francis Group, LLC.
correct categor y. It is especi ally important to avoid allowi ng a hazard to pass throug h the
screens sim ply becau se few or no data are avail able or be cause the haz ard is poorly
charact erize d.
31.1.1 QUOTIENTS
The standar d model for risk charact erization in screeni ng asses sment s is the ha zard quotient

(HQ) (Section 30.1) . The be nchmark and exposure levels may be de rived in a varie ty of ways.
The bench mark concentra tion or dose may be a regula tory standar d or a value derive d
specif ically for screeni ng assessment s or a standar d test en dpoint (Chapter 29). The sim plest
ben chmark is the thres hold for toxico logical concern (T TC), ‘‘a level of exp osure to chemic als
below whi ch no significan t risk is expe cted to ex ist’’ (Kroes et al. 2000). For exampl e, 0.01
m g=L has been proposed as a TTC for aquati c toxicity of organic chemi cals (De Wolf et al.
2005). The exposure conce ntration for asses sment s of new pesticides or ind ustrial ch emicals is
derive d by employ ing standar d use, relea se, or disposa l scenari os to standar d en vironmen tal
models such as in the Europ ean Union syst em for the evaluation of su bstance s (EUSE S). For
existing contam ination, the de rivation of a n exposu re concentra tion is complex , and is
discus sed with respect to contam inated sit e screeni ng (Sect ion 31.2) .
Screen ing of mixtures of co ntaminan ts may emp loy tests of the mixt ure or models based on
the toxic ity of the con stituent s (C hapter 8). The only model that is rou tinely used for
ecologi cal screen ing asses sments of mixtures is the hazard index. The assessor calculates :
HI ¼
X
( C
ei
=C
bi
); (31 :1)
where HI is the hazard index, C
ei
, the exposure concentration of chemical i,andC
bi
,the
corresponding benchmark concentration. If the sum is greater than 1, the mixture is potentially
hazardous and must be retained for further assessment. If the individual constituents must be
screened, one might retain those chemicals that contribute more than 10%, 1%, or some other
percentile of the HI. Alternatively, one may include chemicals with individual quotients greater

than some value. Parkhurst et al. (1996a) recommend a minimum quotient of 0.3.
If multiple ch emicals occur at potenti ally toxic co ncentra tions but the assum ption of
add itive toxic ity is not reasonabl e, it is us eful to calcul ate the sum of toxic units ( S TU) as
an index of total toxicity (Sect ion 8.1.2). This pe rmits the asses sor an d reviewer s to compare
the relative contri butions of ch emicals to toxicity without necessa rily assum ing that the
con taminants are concen tration- additive. TUs are quot ients of the conc entration of a che m-
ical in a medium divide d by the standar d test endpoint concentra tion for that ch emical. A TU
is sim ilar to an HQ and a STU is simila r to an HI except that, beca use TUs are used for
compa rative purposes rather than to dra w conclusi ons, a common test end point is use d rather
than conserva tive be nchmarks or most relev ant test endpo ints. The express ion of con centra-
tion and the test endpoint vary among media ; for wat er they are typic ally the mean or upper
95% con fidence lim it exposu re co ncentra tion an d the 48 h EC
50
for Daphnia sp. (the most
common aquati c test endp oint). The c hemicals that constitut e a potenti ally signi ficant com-
ponent of toxicity (i.e., TU > 0.01) should be plotted for each reach or area for water,
sedim ent, soil , a nd wildlife intake (e.g., Figure 20.2). The choice of a cutoff for inclusion is
based on the fact that acute values are used in calculating the TUs, and chronic effects can
occur at concentrations as much as two orders of magnitude below acute values. Other values
may be used if specific circumstances warrant. The height of the plot at each subreach is the
STU for that medium and subreach (Figure 20.2). This value can be conservatively inter-
preted as the total toxicity-normalized concentration and therefore as a relative indication of
the toxicity of the medium in that subreach.
ß 2006 by Taylor & Francis Group, LLC.
31.1.2 SCORING S YSTEMS
When data are unavailable or there are no simple models of the hazard being assessed, expert
judgment based on experience with the issues being assessed or similar issues may be used for
screening (UK Department of the Environment 2000). These may be qualitative (e.g., high,
medium, low) or semiquantitative. For example, for each hazard, scores (e.g., 1–5) may be
applied to the source, transport pathways, receptor exposure, and response, and then summed.

Because risk assessments should be transparent and subject to replication, it is important to
clearly characterize the bases for judgments. This may be accomplished by developing a formal
scoring system. Scoring systems have been used for decades to rank the risks from chemicals or
from more diverse sets of agents (Harwell et al. 1992; Swanson and Socha 1997). However, to
serve as screening tools, these systems should be calibrated to actual risks so that the total score
is at least roughly linearly related to risk and cutoff scores can be defined for the screening
categories. If scoring systems are subjective (i.e., not calibrated), it is important to avoid giving
an impression of scientific accuracy to the numeric results.
31.1.3 SCREENING FOR P ROPERTIES
Chem icals and other agen ts may be sub ject to particular asses sment standar ds if they possess
certa in propert ies such as pe rsistence, or particular mod es of ac tion such as mutag enici ty or
teratoge nicity. For ex ample, the US Food Qual ity Pr otection Act requ ires screeni ng an d
subsequ ent testing of pesticides for endocrine- disrupting prop erties. This screeni ng may be
performed using in vitro tests (ER-CALUX test for estrogenicity), simple rapid whole
organism tests (e.g., use of Japanese medaka embryos to screen waters for teratogenicity),
or quantitative structure–activity relationships (QSARs) for particular modes of action
(Sect ion 26.1). If the chemic al is not posit ive in the test or doe s not fit the model of the
mode of action, it is screened out with respect to those standards.
31.1.4 LOGICAL CRITERIA
In some cases, particularly for agents other than chemicals, neither quantitative nor semi-
quantitative methods are feasible, but simple logical criteria may be applied to determine
whether a hazard exists. For example, to screen nonnative plant species to determine whether
they should be assessed, Morse et al. (2004) asked: (1) Is the species established outside
cultivation in the region of interest? (2) Is the species established in conservation areas or
other native species habitats in that region? If the answer to either question is no, the plant
species is screened out from further assessment.
31.2 SCREENING SITES
1
While the screening of chemicals, materials, and other agents is largely constrained to compar-
isons of exposure and effects metrics, assessments of contaminated sites are more complex.

The primary purpose of screening is to narrow the scope of subsequent assessment
activities by focusing on those aspects of the site that constitute credible potential risks.
Screening is performed by a process of elimination. Beginning with a site description and the
full list of chemicals that are suspected to constitute site contaminants, one can potentially
eliminate:
1
This section is based on Chapter 5 of Suter et al. (2000).
ß 2006 by Taylor & Francis Group, LLC.