Data-driven regional traffic simulation from heterogeneous mobility sources

Abstract

Regional traffic simulators increasingly combine mobile-phone, road-sensor and GPS data, but what each source contributes is rarely clear. We present an agent-based simulator of daily traffic in Emilia-Romagna, Italy: phone data give trips and their timing, half of the road sensors set the traffic level, all sensors correct it locally, and GPS traces guide route choice. We measure what each source adds against predictors without simulation, on sensors kept out of parameter calibration in winter 2020 and partly shared with it in summer 2021; as these sensors informed some modelling choices, the evaluation is not fully independent. Total traffic is within 1% of observations in winter and 10% in summer, and the simulator ranks roads better than a road-type predictor (Spearman 0.67-0.68 against 0.41-0.48, winter). GPS improves this ranking, sensors improve local estimates, and timing comes mostly from the demand. Single-road errors remain similar to those of simple predictors, and the test of specific origin-destination pairings was underpowered and inconclusive. These findings provide a practical framework for assessing the contribution and limitations of heterogeneous data sources in regional traffic simulation.