Chapter 10 - Global Sensitivity Analysis for Uncertain Parameters, Models, and Scenarios

Abstract

Listen

Translate article

Sensitivity analysis has been used widely for model development and improvement in Earth and environmental sciences. Local sensitivity analysis is to identify important model parameters for a given set of parameter values (nominal values), whereas global sensitivity analysis is to identify important parameters when parameter values are not specified but vary in parameter space. This chapter first briefly introduces two widely used methods of GSA, i.e., the Morris method (a screening method) and the Sobol' method (a sampling-based method). The Sobol' method is always implemented using Monte Carlo (MC) approaches that are computationally demanding. This chapter discusses two numerically efficient methods that were recently developed for implementing the Sobol' method without using MC approaches. In general, the Morris and Sobol' methods and other global sensitivity methods are designed for a fixed model and modeling scenario, assuming that there is no uncertainty in model structure and modeling scenario. However, Earth and environmental systems are open and complex, rending them prone to multiple model structures and modeling scenarios. The model and scenario uncertainties raises the following two immediate questions to GSA. (1) Are the parameters important to one model still important to other models? (2) Are the parameters important under one modeling scenario still important under other modeling scenarios? GSA without considering model and scenario uncertainties may result in biased selection of important parameters, i.e., the identified important parameters based on a single model and a single scenario may not be important to the system of interest. This chapter introduces a recently developed method of GSA that integrates the Sobol' method into a hierarchical framework of quantifying uncertainty in model parameters, structures, and scenarios. An example application of the new method of GSA is presented in the context of environmental modeling, followed by a discussion of existing and potential applications of GSA to satellite data. Several perspectives for future research on GSA are given at the end of this chapter.