(GRS 412) Investigation of Modern Methods of Probalistic Sensitivity Analysis of Final Repository Performance Assessment Models (MOSEL)
Sensitivity analysis is a mathematical means for analysing the sensitivities of a computational model to variations of its input parameters. Thus, it is a tool for managing parameter uncertainties. It is often performed probabilistically as global sensitivity analysis, running the model a large number of times with different parameter value combinations. Going along with the increase of computer capabilities, global sensitivity analysis has been a field of mathematical research for some decades.
In the field of final repository modelling, probabilistic analysis is regarded a key element of a modern safety case. An appropriate uncertainty and sensitivity analysis can help identify parameters that need further dedicated research to reduce the overall uncertainty, generally leads to better system understanding and can thus contribute to building confidence in the models.
The purpose of the project described here was to systematically investigate different numerical and graphical techniques of sensitivity analysis with typical repository models, which produce a distinctly right-skewed and tailed output distribution and can exhibit a highly nonlinear, non-monotonic or even non-continuous behaviour.