Application and evaluation of computationally efficient techniques for uncertainty and variability characterization in exposure assessment
S. Isukapalli, P.G. Georgopoulos
Environmental and Occupational Health Sciences Institute, UMDNJ - R.W. Johnson Medical School and Rutgers University
Uncertainty and variability in exposure modeling need to be characterized separately. This is often addressed via two-dimensional Monte Carlo (MC) techniques, because one-dimensional MC methods require prohibitively large number of simulations to keep the separate propagation of uncertainty and variability tractable. However, even the two dimensional MC techniques require significantly large numbers of model runs, and do not provide estimates of contributions from uncertainties in individual inputs and parameters. New, computationally efficient techniques based on direct uncertainty propagation, such as the Stochastic Response Surface Method (SRSM), enable mathematically tractable propagation of uncertainty and variability, and thus can be used as alternatives to two-dimensional MC analyses. Uncertainty propagation in the SRSM is accomplished by transforming the inputs and parameters into functions of standard random variables (srv's), and expressing the outputs in terms of a polynomial chaos expansion of the srv's. The coefficients of output approximations are estimated using model outputs at heuristically selected sample points; the number of sample points is substantially lower than the number of model simulations needed for two-dimensional Monte Carlo analysis. The coefficients of the polynomial expansion, along with the input approximations, provide insight into the overall uncertainty in output metrics, and also provide information on contributions from individual components of inputs, thus allowing separate characterization of uncertainty and variability. Relevant applications of the SRSM involving various modules that focus on specific components of exposure and risk assessment are presented. Performance and accuracy of the SRSM are evaluated using corresponding two-dimensional MC analyses.
This work had been funded in part by the US Environmental Protection Agency under Cooperative Agreement # EPAR-827033 to the Environmental and Occupational Health Sciences Institute (EOHSI). The viewpoints expressed here are the responsibility of the authors and do not necessarily reflect the views of the USEPA or its contractors.