Ecological Risk Assessment of Diffuse and Local Soil Contamination Using Species Sensitivity Distributions - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 638

Analogous to this is the issue of laboratory to field extrapolation in Ecological
Risk Assessment. This is a relevant concept because we usually do not (never?)
have sensitivity data for species that occur at particular field sites, and field sites
likely differ from each other and from the conditions at which the test data set was
established. This holds both for the exposure conditions (and Exposure Assessment)
and the local species (and their sensitivities). SSDs describe the distribution of
the sensitivities of the test species, in their test conditions, but not necessarily the
distribution of species sensitivities and the exposure situation at a contaminated
site .
When test data distributions must be extrapolated, it is crucial to investigate
how far a new situation could resemble the “test data set”, that is the set of data
to describe the distribution, and their exposure situations. For example, when the
“test data set” of the workshop participants' heights are adults, it is easy to under-
stand that a “blind” statistical use and interpretation of the model derived earlier
would overestimate “risks” for children, but (far) less so for workshop attendants in
general (since those tend to be populations of adults too). In a practical assessment,
one could easily see what the output means (and what it does not) when considering
the extra available data, and thus manage the case based on those extra data. For
any Ecological Risk Assessment, this means that one should scrutinize the resem-
blance of the contamination case to the “training data set”, with respect to both the
sensitivities of species and the exposure conditions.
The issue of extrapolation has caused considerable debate (e.g., Hopkin ( 1993 )
versus Van Straalen ( 1993 )). We resolve this issue at least semantically by recog-
nizing a difference between the fraction of workshop participants or species that is
probably affected and those at another workshop or in a natural species assemblage
that are potentially affected. When the distribution of the training data set is used,
the output of the distribution is a Probability that is directly valid for the test set .
But after extrapolating to another situation , the extra uncertainty introduced by the
extrapolation is frequently “captured” in a new term on the Y-axis, i.e., potential ,in
the phrase Potentially Affected Fraction (PAF).
If differences between the laboratory data set and the field are known, correc-
tion for those differences should be considered. There are various ways to correct
exposure data for differences in conditions (see e.g., De Zwart et al. 2008b ). These
corrections often address bioavailability, which usually implies that estimated PAF-
values are lower than when total contaminant concentrations are used. For effect
data, various authors have discussed the extrapolation between life stages, test dura-
tions, and levels of organization (Solomon et al. 2008). However, laboratory to field
extrapolations are, in general, highly uncertain.
As a useful alternative, it is becoming common practice to state at least qualita-
tively whether the field situation is likely to be more or less affected than modeled
(as is possible in the adults-to-children issue in the example). Thus, a qualitative
type of assessment output has been recently promoted as a common-sense solution
to the extrapolation problem in various contexts; see e.g. EUFRAM ( 2006 ), regard-
ing the evaluation of risk of plant protection products, and Risbey and Kandlikar
( 2007 ), on a formal scheme to handle Risk Assessment uncertainty.