Optimal Sensitivity Analysis under Constraints: Application to fisheries

摘要

We propose a new optimal (in some sense) methodology to conduct a global sensitivity analysis (GSA) of numerical model outputs relative to model inputs when three specific constraints exist. The three constraints considered here are: (i) A computation time that is too long to perform all simulations required when using the usual methods of GSA, typically those based on LHS, Sobol sequences, etc., and typically when only hundreds of simulations are possible; (ii) The inputs are not independent because some structural correlations or functional relationships exist between them (or part of them), or bounded combinations of inputs exist; (iii) Qualitative inputs are present in addition to quantitative inputs.