R2 PM Evaluating Short- and Long- Term Effectiveness In Sediment Remediation
Thursday, 17 November 2005: 1:50 PM - 5:30 PM in Ballroom 2

(DEK-1117-717903) Forecasting Short- and Long-term Remedy Effectiveness within an Uncertainty Context: How Quantified Uncertainty Impacts Decisions on Remedy Effectiveness.

Dekker, T¹, Helfand, J¹, Redder, T¹, Lautenbach, D¹, ¹ Limno-Tech, Inc., Ann Arbor, MI, USA

ABSTRACT- Decisions about remedial alternatives at contaminated sediment sites are almost always made in the context of considerable uncertainty about future system behavior. To date, models have often been used to forecast the long-term behavior of a site, simulating changing concentrations in various system compartments, including biota, sediment, and water. While it is often acknowledged that such models are uncertain and that uncertainty increases with longer forecast times, rigorous tools for evaluating forecast uncertainty are rarely employed. In this paper, a novel method for assessing model uncertainty within the constraints of a calibrated modeling framework is presented. The generalized parameter estimation tool of Doherty (Dohery, 2004) is used to provide a diagnostic evaluation of model behavior under a range of conditions that provide different but equally acceptable correspondence to calibration targets. The model calibration and sensitivity analysis process generates a Jacobian matrix describing and quantifying model sensitivity to critical input parameters. This process, termed calibration-constrained uncertainty analysis, is then used to run a series of illustrative simulations, illustrating how: 1) A model's value can be expressed in terms of its ability to discern between the relative merits of different remedies, in terms of some measurable future benefit 2) Models can be used both to quantify uncertainty and to provide guidance as to how uncertainty could be reduced. 3) Models are useful for resolving management questions when uncertainty is sufficiently small to allow different outcomes of a given management question to be discerned. 4) Models which allow for characterization of both the expected value and the distribution of a given output allow for design of improved long-term monitoring plans, and more quantitative assessment of the achievement (or probability of achievement) of remedial targets.

Key words: sediment, uncertainty, model, decision