Probabilistic risk assessment phase 2. Guidance for quality assurance
Öberg, T., Sander, P., Bergbäck, B.
Report 5621
Swedish Environmental Protection Agency, 2006.
 

Abstract
Probabilistic methods are anticipated to find an increased use in risk assessment of contaminated land. The methodology is easy to integrate with current risk assessment models, but a need exists for quality assurance of the work process and reporting routines. The purpose of this report is to provide guidance and recommendations to ensure comparability and facilitate independent review.

A previous review of the literature showed a wide area of application for probabilistic risk assessment. This type of risk assessments can often provide distinct answers to questions regarding the presence of health- and environmental risks, suitable cleanup targets, and the classification of different areas of land. Probabilistic risk assessments should therefore be considered when a traditional point estimate cannot ensure a lack of risk and when the costs for cleanup are substantial. The applications are then mainly within exposure assessments.

The purpose of a probabilistic risk assessment is to handle uncertainty (lack of knowledge) and variability (natural variation) with a rational and scientific defensible approach. An important source of uncertainty is lack of knowledge in the estimation of parameters in the risk assessment model. The largest contribution to variability is the natural variation between individuals. These sources of uncertainty and variability should be described quantitatively.

Different methods are available to describe uncertainty and/or variability in an input variable and to carry it through the risk assessment model. Three methods are treated here: Interval estimates, Monte Carlo simulations, and probability bounds analysis (PBA). PBA is a relatively new method that can be viewed as a combination of the first two. The choice of method is to a large extent dependent upon the availability of data, where Monte Carlo methods are more demanding than the other two.

Three software have been evaluated within the frame of this project. Crystal Ball® and Analytica® are suitable tools for Monte Carlo simulations and Risk Calc™ can be used for estimating intervals and probability bounds analysis. All three software are equivalent when the probabilistic calculation handle uncertainty and variability similarly, or when only one of these is estimated.

The choice and specification of probability distributions and dependence between input variables are the separate components that have most influence upon the outcome of a probabilistic risk assessment. The choice of input distributions and treatment of dependencies should therefore be motivated and documented.

Sensitivity analysis denotes methods to quantitate the influence by input variables on the outcome of model calculations. A probabilistic risk assessment can be viewed as a general sensitivity analysis, but usually beneficial to estimate the influence of each input variable separately. There are several methods to choose among and therefore this choice should also be motivated.

An important part for the quality assurance is that all calculation results should be possible to verify from the information provided. We recommend that risk assessments of larger cleanup projects are peer reviewed by independent experts. To gain acceptance among decision makers and the public it is necessary that the results are reported in a manner that satisfies the need of different groups of stakeholders. Probabilistic risk assessments may then become an efficient tool to facilitate risk communication by disclosing uncertainties that are hidden by the traditional deterministic approach.

A Swedish version of this report is available for download as a PDF file (1830 kb)