Revisiting the Fundamental Basis of Global Sensitivity Analysis for Dynamical Environmental Models
Keywords
Global Sensitivity Analysis, Dynamical Systems, Variogram Analsysis
Start Date
27-6-2018 10:40 AM
End Date
27-6-2018 12:00 PM
Abstract
In this presentation, we raise fundamental issues with the conventional approaches to global sensitivity analysis (GSA) of dynamical environmental models. We argue that (a) the current approaches are actually designed to carry out parameter “identifiability analysis”, (b) the use of model-performance metrics as model response in GSA may distort the information extracted about relative parameter importance, and (c) it is a conceptual flaw to interpret the results of such an analysis as being consistent and accurate indications of the sensitivity of the model response to parameter perturbations. Further, because such approaches depend on availability of system state/output observational data, the analysis they provide is necessarily incomplete.
We reframe the GSA problem from first principles, using trajectories of the partial derivatives of model outputs with respect to controlling factors as the theoretical basis for sensitivity, and construct a global sensitivity matrix from which statistical indices of total-period time-aggregate parameter importance, and time-series of time-varying parameter importance, can be inferred. We apply the HBV-SASK conceptual hydrologic model set up for a Canadian basin to show the utility of the new, performance metric-free method and demonstrate how it disagrees with the Morris and Sobol’ methods regarding which parameters exert the strongest controls on model behavior. We further show that the new method is highly efficient, requiring limited number of model runs to obtain stable and robust parameter importance assessments.
Revisiting the Fundamental Basis of Global Sensitivity Analysis for Dynamical Environmental Models
In this presentation, we raise fundamental issues with the conventional approaches to global sensitivity analysis (GSA) of dynamical environmental models. We argue that (a) the current approaches are actually designed to carry out parameter “identifiability analysis”, (b) the use of model-performance metrics as model response in GSA may distort the information extracted about relative parameter importance, and (c) it is a conceptual flaw to interpret the results of such an analysis as being consistent and accurate indications of the sensitivity of the model response to parameter perturbations. Further, because such approaches depend on availability of system state/output observational data, the analysis they provide is necessarily incomplete.
We reframe the GSA problem from first principles, using trajectories of the partial derivatives of model outputs with respect to controlling factors as the theoretical basis for sensitivity, and construct a global sensitivity matrix from which statistical indices of total-period time-aggregate parameter importance, and time-series of time-varying parameter importance, can be inferred. We apply the HBV-SASK conceptual hydrologic model set up for a Canadian basin to show the utility of the new, performance metric-free method and demonstrate how it disagrees with the Morris and Sobol’ methods regarding which parameters exert the strongest controls on model behavior. We further show that the new method is highly efficient, requiring limited number of model runs to obtain stable and robust parameter importance assessments.
Stream and Session
E3: Complexity, Sensitivity, and Uncertainty Issues in Integrated Environmental Models