Illustrating sensitivity and uncertainty in environmental fate models using partitioning maps

Variations of model predictions of the environmental fate
of organic contaminants are usually analyzed for only one
or at most a few selected chemicals, even though
parameter sensitivity and contribution to uncertainty are
widely different for different chemicals. A graphical method
is introduced that allows for the comprehensive investigation of model sensitivity and uncertainty for all persistent organic nonelectrolytes at the same time. This is achieved by defining a two-dimensional hypothetical “chemical space” as a function of the equilibrium partition coefficients between air, water, and octanol (KOW, KAW, KOA), and plotting sensitivity and/or uncertainty of a specific model result to each input parameter as a function of this chemical space. The approach is illustrated for the bulk phase
concentrations in air, water, soil, and sediment calculated
by a level III model. Colored contour maps facilitate the
identification of those input parameters that cause a high
output variation of hypothetical and real chemicals.
They also allow for the easy categorization of chemicals
in terms of common parameter sensitivities, and thus
comparable environmental behavior. Sensitivity varies with
the mode of emission and the degradability of the
chemicals, making it necessary to develop multiple sets
of contour maps. Comparison of these sets of maps in turn
allows the investigation of how parameter sensitivities
change as a result of changes in mode of emission and
persistence. The presented method can be used for
investigating the sensitivity of any prediction obtained
with any linear fate model that characterizes the partitioning behavior of organic chemicals with KAW, KOW, and KOA.
Once the sensitivity maps have been constructed for a given
environmental scenario, it is possible to perform a
sensitivity analysis for a specific chemical by simple
placement of the substances’ partitioning combinations
within the chemical space. The maps can further contribute
to the mechanistic understanding of a model’s behavior,
can aid in explaining observations of divergent environmental behavior of related substances, and can provide a rationale for grouping chemicals with similar model behavior, or for selecting representative example chemicals for a model investigation. They can also help in deciding when accurate and precise knowledge of physical
chemical property data is crucial and when approximate
numbers suffice to conduct a model investigation.