Abstract:

In the context of risk assessment, we focus on the prediction of an unknown quantity Z whose value is realised in the future, and for which experimental data are not available. We deal with the issue of the uncertainty associated to the difference between the output of the model used for the prediction of Z and the true unknown value of Z itself. Accepted principles and methods for handling this uncertainty in the specific conditions of risk assessment are still lacking. Through the paper we seek to contribute by:

1. making a clear distinction between model output uncertainty (epistemic uncertainty about the differences between the true values of the output quantities and the values predicted by the model) and sources of model output uncertainty, from incomplete/imprecise knowledge on the values of the parameters of the model to model assumptions, simplifications and approximations introduced in the model,
2. distinguishing between model output uncertainty, structural model uncertainty and parameter (model input quantities) uncertainty,
3. establishing explicit links between the different purposes of modelling and risk assessment, discussing how model output uncertainty should be treated in the different instances.

We argue that in risk assessment, quantification of model output uncertainty serves for the qualification and acceptance of the models used, whose outputs feed the following risk-informed decision making process.