Sensitivity analysis of car-following models: Methodology and applications

In this work variance-based techniques for model sensitivity analysis have been discussed and applied to the Intelligent Driver Model (IDM) and the Gipps’ model. Throughout the paper it is argued that the application of such methods is crucial for a true comprehension and the correct use of these models. In particular, concerning the subject of their parameters estimation or calibration. Important issues arising when setting up a sensitivity analysis have been investigated and commented in the specific application to car-following models. Among these issues, the importance of the characterization of the uncertainty in the inputs, also known as data assimilation, and the influence of the data richness (i.e. the coverage of a wide spectrum of traffic patterns) on the characterization of the parameters significance and on their successive calibration. The method allowed us to show that both in the case of the IDM and the Gipps’ model, some parameters, which are generally considered fixed in the field literature, account for a high share of the output uncertainty and thus require to be calibrated. Sensitivity indices let us also to evaluate the parsimony of models, intended as the ability to describe reality with a minimum of adjusting parameters.