Probabilistic forecasting of time-dependent origin-destination matrices by a complex activity-based model system: effects of model uncertainty

No previous studies seem to have examined uncertainty in forecasts of origin destination matrix (OD) tables, predicted by advanced activity-based models of travel demand. This paper documents the design and results of a study on the effects of model uncertainty of the Albatross model on predicted time-dependent OD matrices, for the Rotterdam area, the Netherlands, as a case study. The study involves 1000 runs of model system for a synthetic population of 41,668 individuals. Results indicate that the average uncertainty in the predicted OD matrices due to model uncertainty is 45%, and.13% for destination totals based on these simulation runs. In general, uncertainty is lower for the destinations with higher traffic volumes. Uncertainty in predicted traffic volumes, represented by the cells of the OD matrix, tends to be higher. Finally, for both types of indicators, there is evidence of spatial variability in coefficients of variation, capturing uncertainty in destination totals and traffic volumes. Generally, uncertainty is a non-linear function of the number of samples.