Non-parametric methods for global sensitivity analysis of model output with dependent inputs

Quelques méthodes non-paramétriques pour l'analyse de sensibilité de modèles à entrées dépendantes

Thierry A. Mara (1, 2) , Stefano Tarantola (3) , Paola Annoni (4)

1 GdR MASCOT-NUM - Méthodes d'Analyse Stochastique des Codes et Traitements Numériques
2 PIMENT - Physique et Ingénierie Mathématique pour l'Énergie, l'environnemeNt et le bâtimenT
3 Institute for the Protection and Security of the Citizen
4 Joint Research Center Institute for the Protection and Security of the Citizen

This paper addresses the issue of performing global sensitivity analysis of model output with dependent inputs. First, we define variance-based sensi- tivity indices that allow for distinguishing the independent contributions of the inputs to the response variance from their mutual dependent contribu- tions. Then, two sampling strategies are proposed for their non-parametric, numerical estimation. This approach allows us to estimate the sensitivity indices not only for individual inputs but also for groups of inputs. After testing the accuracy of the non-parametric method on some analytical test functions, the approach is employed to assess the importance of dependent inputs on a computer model for the migration of radioactive substances in the geosphere.

Mots clés: variance-based sensitivity indices; dependent contributions; Iman & Conover sampling procedure; radionuclide migration; independent contributions; dependent inputs; Rosenblatt transformation

Domaines: Applications [stat.AP]

Thierry A. Mara, Stefano Tarantola, Paola Annoni. Non-parametric methods for global sensitivity analysis of model output with dependent inputs. Environmental Modelling and Software, 2015, 72, pp.173-183. ⟨10.1016/j.envsoft.2015.07.010⟩. ⟨hal-01182302⟩