ASSESSMENT OF CHILD EXPOSURE TO LEAD ON AN IRONWORKS BROWNFIELD : UNCERTAINTY ANALYSIS

Uncertainty regarding parameters involved in health risk assessment models is generally addressed within a purely probabilistic framework: it is assumed that knowledge regarding model parameters is of a purely random nature (expressing variability). Such knowledge is represented by single probability distributions (PDFs) and the uncertainty is typically propagated through the risk model using the Monte-Carlo technique. Yet in real cases of site investigations and risk assessments, information regarding model parameters is often vague or incomplete (due to time and financial constraints). Such information does not warrant the use of single PDFs that are truly representative of available information. For example, in their analysis of the uncertainty in the propagation of chlorinated organic pollutants in groundwater, McNab et al. (1999) define 12 probability distributions for their model parameters, 9 of which are qualified as "postulated". As shown in particular by Ferson & Ginzburg (1996) assuming random variability when faced with partial incomplete/imprecise information (partial ignorance) seriously biases the outcome of the analysis by reducing the range of possible outcomes. These authors suggest that distinct methods are needed to adequately represent and propagate random variability (often referred to as "objective uncertainty") and imprecision (often referred to as "subjective uncertainty").