Modelling Under Risk and Uncertainty: An Introduction to Statistical, Phenomenological and Computational Methods

2.4 Modelling Under Uncertainty – The General Case

2.4.1 Phenomenological Models Under Uncertainty and Residual Model Error

Models are always an imperfect representation of reality: beyond the basics given in Section 3.2, a more thorough definition of a system model should end with ‘... approximately , accurately enough for decision-making’, leaving room for residual model uncertainty or inaccuracy. The whole rationale of system modelling is that information brought by the phenomenological statement defining the system model itself (physical laws, accident sequence analysis, etc.) plus the information available on the extent of uncertainty in its inputs x will result in a less uncertain inference than a prediction of z based on direct observations or expertise. However, any effort to establish sound modelling in the context of risk and uncertainty should account more explicitly for such residual inaccuracies in the system model.

Indeed, a large class of modelling practice does involve an explicit representation of it to some extent. The previous section illustrated the situation where the amount of physics-based or logical knowledge resulted in a solidly-established system model M, prior to the consideration of the available statistical information. Differently, a ‘phenomenological’ system model, in the environmental or sanitary fields, for instance, may be regarded from the outset ...