Improved models of roundabout lane capacity

Accurate roundabout capacity models are essential for optimising roundabout designs and minimising queues and delays, but there exists significant variation in the capacities predicted by different state-of-the-art models. This paper presents the results of an empirical study into roundabout lane entry capacity, based on geometric and at-capacity flow data from 35 roundabout entry lanes. A large number of model forms and explanatory variable sets were tested due to the general lack of consensus on the variables which affect roundabout capacity and the form of their impacts in existing research. Two regression models and an artificial neural network were developed. It was found that, compared to linear-in-Qc models similar to those used in the U.K., a nonlinear exponential relationship could provide better lane capacity predictions particularly at high and low circulating flows, and better reflected the overall trends in the aggregated capacity data. Comparison with neural network analyses on the same dataset showed that the regression models performed relatively well and provided better information on the impacts of the variables. The new empirical models also performed better than existing international capacity models, and provide evidence that entry-exit separation and exiting flows on the same arm have significant effects which were potentially more than those of other commonly used variables such as entry angle and entry radius. The findings from this study could thus contribute to an improved understanding of the factors and variables which affect entry lane capacity, and therefore towards the development of better roundabout capacity models