Methods for Deriving Pesticide Aquatic Life Criteria - Reviews of Environmental Contamination and Toxicology - page 83

As discussed earlier, one challenge in the use of SSDs is to fit the data to an appro-
priate distribution before extrapolation. One way to achieve a better fit is to break data
into groups rather than pooling them in one SSD. Data may be grouped according to
toxicant mode of action, habitat (e.g., freshwater vs saltwater), reproductive strategy,
or life cycle (Solomon and Takacs 2002). Newman et al. (2000) found that cumulative
frequency models that did not fit log-normal or log-logistic models had distinct shifts
in slope corresponding to transitions among taxa in the ranked data set. When data
are grouped according to taxa or toxic mode of action, more data sets fit the log-
normal distribution (ECOFRAM 1999; Newman et al. 2002). Traas et al. (2002) also
support the idea of separating data into subgroups by taxa, or according to toxic mode
of action, before constructing SSDs. In constructing SSDs for pesticides, Maltby et al.
(2005) found that composition of taxonomic assemblages affected the hazard assess-
ment, but groupings by habitat and geographic distribution had no effect.
Only the USEPA criteria methodology (1985) explicitly separates data into groups
for constructing SSDs; moreover, the SSD is constructed using animal data only.
Plants are indirectly included in criteria derivation. If a plant proves to be the most
sensitive among tested species, then the FPV becomes the FCV. All other methodolo-
gies combine all aquatic data. The Netherlands methodology even includes NOECs
derived from secondary poisoning analysis for birds and mammals (RIVM 2001).
However, according to some of the guidelines, if statistical analysis shows that the
data do not fit the assumed SSD distribution, or if data show a bimodal distribution,
then data may be grouped to achieve a fit, with the most sensitive group used to derive
the criterion (RIVM 2001; ECB 2003). In deriving target values using the Australia/
New Zealand methodology (ANZECC and ARMCANZ 2000), which involves fitting
data to one of several possible distributions, it was possible to use all data sets in their
entirety (i.e., with all taxa combined).
Data have been grouped and/or excluded in other studies. For example, in con-
structing an SSD for an ecological risk assessment of chlorpyrifos, Giesy et al.
(1999) excluded data from rotifers, mollusks, and other insensitive organisms,
although such exclusions were not based on statistical analyses. Similarly, in a risk
assessment of diazinon in the Sacramento and San Joaquin River basins, Novartis
Crop Protection (1997) considered 10th percentile values for a combined fish and
arthropod data set, as well as for separate fish and arthropod sets. The 10th percen-
tile was derived from combined sets (3,710 ng/L), fish alone (79,900 ng/L) and for
arthropods (483 ng/L). Using these numbers, the risk to arthropods would be under-
estimated if fish and arthropod data were combined, indicating that data for the two
groups should be analyzed separately.
When the goal of a water quality criterion is to protect all species in an ecosystem,
it is important to include all species in the derivation procedure. However, it is
reasonable, especially in construction of SSDs, to separate species into groups if a
multimodal distribution is evident. If there is no statistically significant difference
between apparent groups (e.g., saltwater and freshwater, or plants and animals),
then the data should be pooled for criteria derivation.
Details of the currently utilized SSD procedures are presented in the following
paragraphs. Several countries use EPA methods and will not be individually