Link-focused prediction of bike share trip volume using GPS data: A GIS based approach

Abstract

Modern bike share systems (BSSs) allow users to rent from a fleet of bicycles at hubs across the designated service area. With clear evidence of cycling being a health-positive form of active transport, furthering our understanding of the underlying processes that affect BSS ridership is essential to continue further adoption. Using 286,587 global positioning system (GPS) trajectories over a 12-month period between January 1st, 2018 and December 31st, 2018 from a BSS called SoBi (Social Bicycles) Hamilton, the number of trips on every traveled link in the service area are predicted. A GIS-based map-matching toolkit is used to generate cyclists’ routes along the cycling network of Hamilton, Ontario to determine the number of observed unique trips on every road segment (link) in the study area. To predict trips, several variables were created at the individual link level including accessibility measures, distances to important locations in the city, proximity to active travel infrastructure (SoBi hubs, bus stops), and bike infrastructure. Linear regression models were used to estimate trips. Eigenvector spatial filtering (ESF) was used to explicitly model spatial autocorrelation. The results suggest the largest positive predictors of cycling traffic in terms of cycling infrastructure are those that are physically separated from the automobile network (e.g., designated bike lanes). Additionally, hub-trip distance accessibility, a novel measure, was found to be the most significant variable in predicting trips. A demonstration of how the model can be used for strategic planning of road network upgrades is also presented.