Benchmarking the PAWN distribution-based method against the variance-based method in global sensitivity analysis: Empirical results. (December 2019)

• Record Type: Journal Article
• Authors: Mora, Esteve Borràs; Spelling, James; van der Weijde, Adriaan H.
• Is Part Of: Environmental modelling & software. Volume 122(2019)
• Publication Date: 2019-12
• Subjects: Global sensitivity analysis -- PAWN distribution-based method -- Variance-based sensitivity analysis; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2019.104556
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: The search for new and more efficient global sensitivity analysis methods has led to the development of the PAWN distribution-based method. This method has been proven to overcome one of the main limitation of variance-based methods – the moment independent property. In this regard, the distribution-based method has outperformed the variance-based method for some highly-skewed or multi-modal distributions. However, despite its increasing popularity, there is a lack of understanding about the performance and properties of the distribution-based method. The benchmark presented in this paper is an attempt to remedy this. We compare the distribution-based method against the variance-based method for a set of well-known test functions. We show that, whereas the distribution-based method can be used as a complementary approach to variance-based methods, which is especially useful when dealing with highly-skewed or multi-modal distributions, it fails to rank different inputs that have different orders of magnitude in their contribution of the response. Highlights: We demonstrate that the PAWN distribution-based method can be used as a complementary approach to variance-based methods. We demonstrate that the PAWN distribution-based method fails to rank different inputs when these have different orders of magnitude in their contribution. We benchmark the PAWN distribution-based method against the variance-based method for a set of well-known test functions. We show when the PAWN distribution-based method can outperform the variance-based method.