Impacts of downscaled inputs on the predicted performance of taxi fleets in agent-based scenarios including Mobility-as-a-Service

Abstract

The impacts of using downscaled inputs in mobility simulation involving taxi fleets have not been well studied. Downscaled inputs, that is, inputs with a fraction of the full-scale population, are often used for large-scale scenarios in order to keep run times short and to avoid using high-performance computing resources. In this paper, a large-scale multi-modal scenario of the Munich metropolitan region, with a taxi fleet that serves the first/last mile legs of public transit passengers, is used to systematically quantify the impacts of downscaling. The results show that the full-scale population is required to obtain accurate spatio-temporal estimations of the fleet’s performance. Sample sizes of at least 30% of the full-scale population can thus be used if only an approximate knowledge of spatially aggregated performance metrics (such as average waiting time, fleet utilization and empty mileage) is desired. However, with downscaled inputs, different characteristics during peak hours, such as a shift in the peak utilization, can result. Moreover, we find evidence that downscaling can affect the spatial distributions of trip densities of taxi fleets.