How can global sensitivity analysis serve traffic and transportation modelling? background and examples

Quantitative sensitivity analysis (QSA) of models is becoming an essential element of model-based analyses. On the one hand, as far as mathematical models are increasingly concerned with policy decision making, QSA is being invoked for the corroboration, quality assurance and defensibility of model-based analysis. The crucial question, here, is the sensitivity of final results to changes in starting assumptions. On the other hand, QSA is increasingly considered as an essential ingredient for building and improving models, being able to unveil the hidden relationships between model inputs and outputs: quoting Saltelli, 2008, “QSA is the study of how the uncertainty in the output of a model can be apportioned to different sources of uncertainty in the model input”. Variance-based techniques are among the most advanced methods for QSA. They are based on the Sobol decomposition of variance and on the implementation of quasi-Monte Carlo algorithms (i.e. based on low-discrepancy sequences of numbers) for the variance computation. In this seminar, variance-based techniques for QSA will be first introduced. Some examples of applications of such techniques to traffic and rail simulation will follow.