Model and data limitations: the sources and implications of epistemic uncertainty

Abstract

Chapter 2 focused entirely on aleatory uncertainty. This is the uncertainty that arises out of the randomness of the hazard itself, and also, possibly, out of the responses to the hazard outcome. That chapter chased this uncertainty through the footprint function and a loss operator to arrive at an exceedance probability (EP) curve. Such a structured approach (e.g. as opposed to a purely statistical approach) was motivated by the need to evaluate different interventions for choosing between different actions; and by the possibility of non-stationarity in the boundary conditions on policy-relevant timescales measured in decades. Different risk managers will have different loss operators, and hence different EP curves. Likewise, the same risk manager will have different EP curves for different actions. A very simple summary statistic of an EP curve is the area underneath it, which corresponds to the expected loss (‘expectation’ taken in the mathematical sense), which is defined to be the risk.