Ecological Risk Assessment of Diffuse and Local Soil Contamination Using Species Sensitivity Distributions - Dealing with Contaminated Sites: From Theory Towards Practical Application - page 654

important to use a defensible and consistent method to select a model. For
dose-response modeling, the U.S. EPA for example routinely uses Akaike's
information criterion (AIC) to select the best dose-effect model from a family
of optional mathematical dose-response models ( http://www.epa.gov/ncea/bmds/
bmds_training/methodology/intro.htm ). The AIC is an approach that compares sev-
eral competing statistical models when fitted to the same data set, and determines
the best model in terms of the information obtained from the data.

14.9.4 Options to Handle Small Sets of Input Data

For most new synthetic contaminants, no soil test values are available and
for most others the numbers of data are not sufficient to estimate an SSD.
The European inventory EINECS (European Inventory of Existing Commercial
Chemical Substances) alone contains more than 100,000 contaminants ( http://ecb.
jrc.ec.europa.eu/esis/index.php?PGM

ein ), while the RIVM e-toxBase contains
data for approximately 5000 contaminants.
The real or expected lack of input data for SSD-modeling implies a necessity to
consider methods for estimating test results. In such cases, existing ecotoxicity data
are used to predict the ecotoxicity of compounds for non-tested species (or species
groups) of interest, and to predict chronic input data based on available acute ecotox-
icity data. De Zwart ( 2002 ) for example has derived “rules of thumb” for deriving
the shape and position of SSDs of untested contaminants from test data for other
chemicals. The procedures are based on patterns in shape or position of SSDs, which
in turn relate to grouping of contaminants by Toxic Mode of Action. Aldenberg
and Luttik ( 2002 ) have proposed specific methods to handle small toxicity data
sets, again by using toxicity data from “similar” tested contaminants. Finally, the
U.S. EPA provides inter and intra species extrapolation models for contaminants for
which data are lacking. The ACE model addresses Acute-to-Chronic Extrapolation,
and makes use of the fact that for many contaminants there is a wealth of data
on acute effects, relative to chronic exposure test data (Ellersieck et al. 2003). It is
possible to extrapolate from acute to chronic data due to existing data patterns, yield-
ing the possibility to derive chronic SSDs when only acute data are available. The
ICE model is concerned with Interspecies Correlation Extrapolation, and makes use
of known correlations of toxicity values (e.g., EC50) between a surrogate species
(frequently tested) and an untested species or taxon of interest (Mayer et al. 2004).

=

14.9.5 Handling the Possible Causes of Misfit

SSD models may adequately fit the data, or not. Various statistical tests and diag-
nostics are implemented in SSD-software to investigate model fits. In the case of
a sufficient fit of the model to the data, the assessor may decide to report this, and
use the model. The statistically sufficient fit, however, may not be interpreted as any