You are here:
UNCERTAINTY IN MODEL PREDICTIONS-PLAUSIBLE OUTCOMES FROM ESTIMATES OF INPUT RANGES
Citation:
Weaver, J W., C L. TebesStevens, AND K Wolfe. UNCERTAINTY IN MODEL PREDICTIONS-PLAUSIBLE OUTCOMES FROM ESTIMATES OF INPUT RANGES. Presented at Brownfields 2002, Charlotte, NC, November 13-15, 2002.
Impact/Purpose:
To assist decision-making by improving approaches to subsurface contaminant transport modeling based on evaluation of field observations and subsequent development of appropriate modeling approaches and tools.
Description:
Models are commonly used to predict the future extent of contamination given estimates of hydraulic conductivity, porosity, hydraulic gradient, biodegradation rate, and other parameters. Often best estimates or averages of these are used as inputs to models, which then transform them into output concentrations. Despite this evident certainty, all properties of the subsurface are both uncertain, because of imperfect measurement methods, and subject to point-to-point variability, because of geologic heterogeneity. Where used as purely predictive tools (i.e., in the absence of model calibration to field data), uncertainty and variability lead to the need for assessment of the plausible range of model outcomes. For Brownfields sites, the need to provide rapid assessment of contamination may not allow extensive field and modeling studies, where calibration data sets would be obtained. In those and other cases there is a need for evaluation of model uncertainty given input variation. Our approach is to use all combinations of input parameters to determine the earliest and latest first arrivals, the lowest and highest peak concentration, the shortest and longest duration of contamination, and the lowest and highest risk scenarios. Results of simulations show that even moderate ranges of input variation generate significant differences in model predictions. These differences are greater than obtained from simple one parameter at a time uncertainty analyses, because of combined influences of multiple parameters. For example, hydraulic conductivity, porosity and gradient together determine the seepage velocity, and variation of each of parameter needs to be considered in order to determine the extremes of velocity. The extreme parameter sets were found to be different for some of the four predicted model outputs (first arrival, maximum concentration, duration, risk). This result shows that selection of worst case parameter set depends on the desired output of the model. The simulations showed that the best and worst case parameter sets for first arrival time, maximum concentration and duration were consistent across all simulations and could thus be selected a priori. Those for risk , however, could be determined only by performing an uncertainty analysis for each input parameter set.