Calibration of Traffic Simulation Models for Non-Circular Roundabouts with Emerging Data

Abstract

Non-circular roundabouts, including elliptical, peanut-shaped, egg-shaped, and turbo roundabouts, are increasingly being adopted across the United States to improve operational efficiency and enhance traffic safety for various intersections. However, existing microsimulation state-level guidance has been developed primarily for conventional circular roundabouts. This leaves a critical research gap regarding how non-circular roundabouts should be accurately modeled and calibrated in PTV VISSIM. This limitation is highly significant because varying entry curvatures, non-circular central islands, and non-uniform circulating paths produce localized driving behaviors that are simply not captured by default simulation parameters. As part of an Alabama Department of Transportation (ALDOT) project, this dissertation develops a robust, data-driven calibration framework for PTV VISSIM roundabout models utilizing emerging high-resolution vehicle trajectory data from drone video. The study evaluated eighteen roundabouts in Alabama, comprising thirteen circular and five non-circular sites. Drone video data were processed using DataFromSky to extract critical microscopic traffic characteristics, including vehicle trajectories, traffic volumes, turning movements, travel speeds, and detailed gap-acceptance behaviors. To support a systematic calibration process, this research implemented an automated randomized grid-search procedure using Python and the VISSIM COM interface. This framework seamlessly automated parameter adjustment, simulation execution, and performance evaluation. Calibration specifically targeted segment-specific speed distributions, approach-specific priority-rule parameters, and the Wiedemann 74 car-following model. Results revealed substantial speed variations both across and within the sites, proving that non-circular roundabouts necessitate multiple localized speed zones. Furthermore, while the default minimum gap time of 3.0 seconds remained effective, minimum clearance values required approach-specific calibration. These calibrated clearance values were significantly higher than VISSIM’s defaults, indicating standard parameters underestimate necessary yielding distances at complex entries. Under the observed traffic conditions, car-following adjustments had a limited impact. Validation demonstrated that the calibrated models achieved improved agreement with field-observed conditions, with all GEH (Geoffrey E. Havers Statistic) values falling below 5.0 and calibrated RMSNE values outperforming default models. Ultimately, this dissertation successfully bridges a vital gap in roundabout microsimulation. By establishing an observation-based calibration methodology, it delivers practical guidelines for ALDOT and provides a highly transferable framework for evaluating the performance of complex, emerging intersection alternatives including non-circular roundabout.