Preliminary analysis of the Morris method for identifying influential parameters

Performance assessment of the geologic disposal of radioactive waste is an inherently complex problem involving significant uncertainties. To ensure that the disposal of the waste is safe over long time periods, performance assessments that involve uncertainty and sensitivity analysis are conducted using a mathematical model that stochastically samples uncertain parameters over their range of plausible values. Analysis of the sensitivity of the performance measure to these parameters is conducted with the goal of reducing the list of parameters to a manageable level so that more attention can be given to the most influential parameters driving the system performance. Various techniques, such as scatter plots, differential analysis, regression methods, and response surface techniques, are available to quantify the uncertainty and sensitivity in complex models. These techniques either are cumbersome (e.g., scatter plots) or require that the relationship between input and output variables be linear or quadratic (i.e., response surface techniques). The authors describe the implementation of an experimental design-based method, referred to as the Morris method, to conduct sensitivity analysis for a model with a large number of sampled parameters. Additionally, they present the sensitivity measures used in the study and associated parameter transformations for the best representation of the overall problem. The method employs a sequence of randomized one-factor-at-a-time numerical experiments such that changes in the output corresponding to these isolated parameter increments are unambiguously attributable to the specific change in the input. The number of model computations needed to conduct the sensitivity analysis is proportional to the number of parameters under investigation