Challenges and Future Outlook of Sensitivity Analysis

As Earth and Environmental System models have rapidly become more complex and computationally intensive, growing in parameter dimensionality as they reflect our growing understanding about the nature and functioning of the world, the need for robust, informative, and computationally efficient sensitivity analysis techniques and tools has become ever more pressing. To date, a variety of different approaches to sensitivity analysis have been proposed in the literature, each of which has its strengths and weaknesses. Furthermore, their application is typically inhibited by computational expense and their usefulness is limited by the inability to extract useful diagnostic information from the model response. This chapter reviews and contrasts some of the major strategies for sensitivity analysis that have been proposed and discusses several challenges that need critical attention. Perhaps the most important of these is to establish a clear definition of how to "compactly" characterize the sensitivity of a model response to perturbations in its causal factors in such a manner that the "diagnostic" information provided by the analysis is maximized.