Sensitivity analysis of model output: variance-based methods make the difference

There are many different ways to perform sensitivity analyses in answering these questions but they may not yield identical results. In this paper we will be concentrating on item (a) above. Many authors, when referring to the degree to which an input parameter affects the model output, use the terms ‘sensitive’, ‘important’, ‘most influential’, etc. The methods that will be discussed and used to perform SA in this paper are called ‘Variancebased methods’, in that the variability, or uncertainty, associated with an important input parameter is propagated through the model resulting in a large contribution to the overall output variability. Methods such as “importance measure” (Iman and Hora 1990, Saltelli et al. 1993, Homma and Saltelli 1996), or “correlation ratio” (Krzykacz 1990, McKay 1996), are capable of estimating the “main effect” contribution of each parameter to the output variance. However, whether a parameter is influential or not depends also on the interactions and influences of all the parameters. Derived from quite a ABSTRACT