Making sense of global sensitivity analyses

This study presents improved understanding of sensitivity analysis methods through a comparison of the local sensitivity and two global sensitivity analysis methods: the Morris and Sobol′/Saltelli methods. We re-interpret the variance-based sensitivity indices from the Sobol′/Saltelli method as difference-based measures. It suggests that the difference-based local and Morris methods provide the effect of each parameter including its interaction with others, similar to the total sensitivity index from the Sobol′/Saltelli method. We also develop an alternative approximation method to efficiently compute the Sobol′ index, using one-dimensional fitting of system responses from a Monte-Carlo simulation. For illustration, we conduct a sensitivity analysis of pressure propagation induced by fluid injection and leakage in a reservoir–aquitard–aquifer system. The results show that the three methods provide consistent parameter importance rankings in this system. Our study also reveals that the three methods can provide additional information to improve system understanding.