PARENT SESSION

5A - Uncertainties and Variabilities
Hall 7
1:45 PM - 3:30 PM, Monday, 28 April 2003
Chair: Scheringer, M.¹, ¹
Co-chair: Matthies, D.M.², Green, J.³, ^{2 3}

(MO7/12) Predicting persistence and long-range transport potential with multimedia fate models: Robustness and sensitivity of results.

Fenner, Kathrin^{1, 2}, Scheringer, Martin², Hungerbühler, Konrad², Schwarzenbach, René^{1, 2}, ¹ EAWAG, Swiss Federal Institute for Environmental Science and Technology, Dübendorf, Switzerland² ETH, Swiss Federal Institute of Technology, Zürich, Switzerland

ABSTRACT- The outcomes of multimedia models often exhibit considerable variance due to large methodological uncertainties and variability in substance properties and environmental parameters. Model intercomparison studies show strongly differing absolute results for a selection of chemicals, but at the same time suggest that the results evaluated relative to a benchmark or in term of rankings are more similar between the models. However, most of these studies have not examined how well each separate model performs in distinguishing the behavior of different chemicals given the uncertainties and variability within the model input parameters (e.g. Do the model results, despite their variance, allow differentiating between typical POP and non-POP behavior?). To answer this question we ran a probabilistic uncertainty analysis of ChemRange, a multimedia box model set up for calculating P and LRT. Substance properties and environmental parameters were varied extensively to simulate substance fate for different environmental conditions. It is investigated how these variations affect the absolute as well as relative values of P, LRT and concentration predictions. The results indicate that the relative results show similar uncertainties as the absolute results. This finding is interpreted as being mainly caused by a weak understanding and lack of integration into the models of how substance properties depend on environmental properties. To understand whether it is possible to deduce such dependencies from existing substance data collections used for modeling, we analyzed a compilation of atrazine degradation rates in soil and water by means of multivariate analysis. The results of these studies are combined to draw conclusions with respect to what is, for a given data availability situation, an optimal level of model complexity and an appropriate interpretation of model results for P and LRT assessment.

Key words: Benchmark chemicals, Uncertainty analysis, Ranking, Chemical assessment