Review of Sensitivity Analysis Methods

This document is a review of sensitivity analysis methods. It aims to contribute to the general understanding of sensitivity analysis. Sensitivity analysis quantifies how the uncertainty in model parameters or factors (e.g. precipitation, in the case of a rainfall-runoff model) affects the uncertainty in the model output (e.g. discharge). The document centres around a decision tree to suggest a sensitivity method according to a set of questions as it is impossible to suggest a universal sensitivity analysis methodology, which is optimal for each situation. Questions include: how long does the model run; are the model equations simple and access to the model code available; or how many parameters and factors need to be investigated; or what is the aim of the sensitivity analysis (model simplification, prioritising research, mapping, calibration). This is followed by an introduction into different aspects and methodologies of sensitivity analysis (more mathematical details are given in an appendix). The chapter begins with a section on Experimental Design which describes ways parameters and factors can be varied within a sensitivity analysis framework. Fundamentally, sensitivity analysis can be grouped in four major subdivisions: Graphical methods include different ways to plot results of model outcome and parameters (e.g. scatter plots) or of other sensitivity analysis¿s themselves (e.g. in the form of histograms); Screening methods are often preliminary numerical experiments whose purpose is to isolate the most important factors from amongst a large number that may affect a particular model response; Local methods investigate how do small changes of parameters affect model output; and Global methods evaluate the effect on the model output by varying all parameters at the same time, in a multivariate fashion, usually based on a Monte Carlo approach. An important approach to sensitivity falls within the emulation context. The basic idea is to represent in a direct way the relationship between model output and model parameters or factors without the need to run the model multiple times. Another way to make a decision on the type of methodology, which could be used is through case studies. For example, a global sensitivity analysis of a sub-surface model highlights the complexities inherent in the sensitivities of the model factors, but demonstrated that the upslope pressure boundary condition and the Brooks Corey soil algorithm ¿air entry¿ value were particularly important. Understanding the controls on the numerical simulation of hydrological processes allow informed decision making, for example, where next to deploy field resources or on pollution transport and groundwater mixing problems. Other examples, in this document include flood inundation models, a weather prediction system and comparisons between different river water quality modelling concepts. The final chapter of the document contains a glossary, explaining many technical terms, which are used in the context of sensitivity analysis. In summary, this document should help practitioners, who maybe unfamiliar with the benefits and want to evaluate the added value of sensitivity analysis. It also acts as a quick reference guide for any reader who is confronted with sensitivity analysis and need to understand the advantages and disadvantages of a methodology.JRC.DG.G.9-Econometrics and applied statistic