Future Directions for Managing Uncertainty in Stochastic Traffic Models

A rapidly expanding range of traffic and transportation applications call for accurate dynamic modelling of traffic flow due to their potential impact on community and environmental decision making. The complexity of these applications dictates that detailed stochastic traffic simulation models are increasingly being used for such purposes. These models, having either stochastic inputs or stochastic model components or both, yield stochastic outputs, which is relevant and a required feature, being reality inherently uncertain. In order to consider a model valid, therefore, the analyst has to verify that the uncertainty in the model outputs be close enough to the uncertainty in the real world. Though apparently obvious the requirement has non-trivial implications: it calls for the ‘management of modelling uncertainty’, intended as the process of identification, quantification and reduction of model uncertainty. When applying mathematical models in support of policy decision making, this can be considered as a step of a broader process also referred as ‘sensitivity auditing’: “a practice of organised scepticism toward the inference provided by mathematical models” (Saltelli et al. 2013). It is claimed here that a shift towards the adoption of techniques to manage the uncertainty in traffic simulation is highly necessary.