Analysing the dynamic performances of a bicycle network with a temporal analysis of GPS traces

The main goal of this paper is to describe temporal analysis methods of cyclists' GPS traces and apply them to a dataset recorded in the city of Bologna, Italy. The dataset consists of approximately 27,000 GPS bike traces registered in May 2016. The novelty of the developed method consists in a particular pre-processing step to reconstruction of the cyclist's speed profile, despite the disturbed location information of GPS traces. Thanks to a detailed representation of the road network, important indicators, such as speed in motion, average waiting times at intersections and even waiting times for specific turn manoeuvres inside junctions can be extracted from the speed profile. These are relevant quantities for the network analyses and the calibration of route choice models. The results reveal that waiting times represent 15% of the total trip duration in average and they differ significantly between cyclists and turn-typology, just as average speeds vary strongly among cyclists. These findings suggest that link-cost functions for bicycle routing should include both waiting times that could be modelled as a function of specific attributes of the network, as well as components that consider the different dynamic behavior of cyclists' types (careful, correct, reckless).