A COMPREHENSIVE COMPARISON OF TOTAL-ORDER ESTIMATORS FOR GLOBAL SENSITIVITY ANALYSIS

Sensitivity analysis helps identify which model inputs convey the most uncertainty to output. One of authoritative measures in global sensitivity is Sobol' total-order index, can be computed with several different estimators. Although previous comparisons exist, it hard know estimator performs best since results are contingent on benchmark setting defined by analyst (the sampling method, distribution inputs, number runs, test function or and its dimensionality, weight higher order effects performance measure selected). Here we compare estimators an eight-dimension hypercube where these parameters treated as random parameters. This arrangement significantly relaxes dependency design. We observe that accurate Razavi Gupta's, Jansen's Janon/Monod's for factor prioritization, Jansen's, Azzini Rosati's approaching "true" indices. The rest lag considerably behind. Our work analysts navigate myriad formulae reducing selection appropriate estimator.