ABSTRACT
Abrupt changes in the environment, such as increasingly frequent and intense weather events due to climate change or the extreme disruption caused by the coronavirus pandemic, have triggered massive and precipitous human mobility changes. The ability to quickly predict traffic patterns in different scenarios has become more urgent to support short-term operations and long-term transportation planning, emergency management, and resource allocation. Urban traffic exhibits a high spatial correlation in which links adjacent to a congested link are likely to become congested due to spillback effects. The spillback behavior requires modeling the entire metropolitan area to recognize all of the upstream and downstream effects from intentional or unintentional perturbations to the network. However, there is a well-known trade-off between increasing the level of detail of a model and decreasing computational performance. This paper addresses these performance shortcomings by introducing a new platform MANTA for traffic microsimulation at the metropolitan-scale. MANTA employs a highly parallelized GPU implementation that is fast enough to run simulations on large-scale demand and networks within a few minutes. We test our platform to simulate the entire Bay Area metropolitan region over the course of the morning using half-second time steps. The runtime for the nine-county Bay Area simulation is just over four minutes, not including routing and initialization. This computational performance significantly improves state of the art in large-scale traffic microsimulation and offers new capacity for analyzing the detailed travel patterns and travel choices of individuals for infrastructure planning and emergency management.