Moment independent and variance-based sensitivity analysis with correlations: an application to the stability of a chemical reactor

Recent works have attracted interest toward sensitivity measures that use the entire model output distribution, without dependence on any of its particular moments (e.g., variance). However, the computation of moment-independent importance measures in the presence of dependencies among model inputs has not been dealt with yet. This work has two purposes. On the one hand, to introduce moment independent techniques in the analysis of chemical reaction models. On the other hand, to allow their computation in the presence of correlations. To do so, a new approach based on Gibbs sampling is presented that allows the joint estimation of variance-based and moment independent sensitivity measures in the presence of correlations. The application to the stability of a chemical reactor is then discussed, allowing full consideration of historical data that included a correlation coefficient of 0.7 between two of the model parameters.