ABSTRACT
Global sensitivity analysis allows investigating the relationship between uncertainty in the inputs of a computational model and the uncertainty in the output. So-called variance-based techniques are based on a decomposition of the variance in the model output into components each depending on just one input variable, components each depending on two variables and so forth. Correspondingly, the output variance can be decomposed into contributions each coming from only one input variable (“first order effects’’), from just two variables (“second order’’), etc. A major drawback of the available algorithms (FAST(Saltelli et al. 1999), RBD(Tarantola et al. 2006), IHS(Saltelli 2002), Sobol’(Sobol’ et al. 2007), see also (Saltelli et al. 2000)) for the calculation of this variance decomposition is the requirement of special sampling schemes or additional model evaluations so that available data from previous model runs cannot be reused.
