Assessing the impacts of transit systems and urban street features on bike-sharing ridership: A graph-based spatiotemporal analysis and prediction model

Accurate analysis and forecasting of bike-sharing ridership, particularly accounting for the effects of urban street features and public transit systems, is vital for optimizing system design, improving operational efficiency, and promoting multimodal integration in urban transport. However, existing models focus more on spatiotemporal pattern analysis and prediction accuracy improvement, often overlooking the role of transit effects and street characteristics. This gap limits our understanding of their interplay and forces a trade-off between accuracy and interpretability. This study presented a graph-based modeling framework that incorporated spatiotemporal bike-sharing data with transit networks and schedules, street view imagery, demographics, built environment metrics, points of interest, and weather conditions to both analyze and predict ridership patterns and their underlying causes. This framework leveraged the predictive power of machine learning, the interpretability of manually extracted features, and the availability of data for these factors, particularly integrating transit networks and schedules to represent transit-related effects. We first employed Spatial Vector Autoregressive Lasso and graph-based models to identify key temporal variables, capture spatial dependencies, and extract spatiotemporal graph attributes. These were combined with other contextual variables and fed into an eXtreme Gradient Boosting (XGBoost) model to elucidate factor-ridership relationships and predict bike-sharing ridership. Using 2019 Capital Bikeshare trip data from Washington D.C., our results showed that incorporating transit and street features greatly improved ridership prediction performance, especially during rush hours and in high-demand areas. This implied strong connections among bike-sharing usage, public transit systems, and street forms. Notably, bike stations within 100 m of bus stops and 50 m of metro stops often showed higher ridership. Bike stations located near major transit hubs, busy streets, traffic intersections, and open urban areas with fewer buildings also experienced greater shared bike use. These findings emphasize the need to integrate transit accessibility and urban street form data into micromobility planning and operation, offering actionable insights for optimizing station placement, rebalancing strategies, and system integration with public transport to advance more efficient and sustainable urban mobility.