Sensitivity analysis of spatial and/or temporal phenomena

Abstract

This section presents several sensitivity analysis methods to deal with spatial and/or temporal models. Focusing on the variance-based approach, solutions are proposed to perform global sensitivity analysis with functional inputs and outputs. Some of these solutions are illustrated on two industrial case studies: An environmental model for flood risk assessment and an atmospheric dispersion model for radionuclide release. These test cases are fully described at the beginning of the paper. Then a section is dedicated to spatiotemporal inputs and proposes several sensitivity analysis methods. The use of metamodels is also addressed. Pros and cons of the various methods are then discussed. In a subsequent section, solutions to deal with spatiotemporal outputs are proposed: Aggregated, site, and block indices are described. The use of functional metamodels for sensitivity analysis purpose is also discussed.