Abstract:

This presentation shows the advantages and drawbacks of the Sobol method.  Sobolmethod is attractive because: 1) it measures parameter sensitivities across the entire input space (i.e. global sensitivity analysis); 2) It can deal with nonlinear responses effectively; 3) It can measure the effect of interactions in non-additive systems; 4) The sensitivity of measures are directly quantifiable in percentages; 5) The analysis of results is intuitive and easy to understand. The main drawback, as with other variance-based methods, is the cost of the analysis, which in the case of computationally intensive models can become prohibitive even when using the approach described that reduces the total number of model executions.  This problem is addressed by: The use of advanced proxy models  and another viable alternative for computationally intensive models is the Morris method (presented in symposium 2014).