Population fraction, number of simulation runs and model uncertainty: Results for Albatross travel demand model system

The shift from four-step to activity-based models of travel demand reflects a transition from aggregate deterministic to disaggregate stochastic models. Multiple simulation runs are conducted to differentiate policy effects from model uncertainty. In principle, the models simulate the behavior of all individuals compromising a synthetic population. However, in practice, forecasts are commonly based on a fraction of the population to reduce computing times. Both the fraction size of the synthetic population and the number of runs will affect model uncertainty, but little is known about their relative contribution. This study addresses this topic by comparing the uncertainty in km travelled, travel time, duration of activities, and number of trips as predicted by Albatross for up to 1000 runs and 10, 30 and 50% fractions of a synthetic population of the city of Rotterdam. Results indicate fraction size has a bigger impact on model uncertainty that the number of simulation runs.