Systematic bikeability evaluation is essential to guide urban strategies and assess cycling infrastructure efficacy, fostering a sustainable, cyclist-friendly urban environment. While many studies use site visits or geospatial techniques to evaluate bikeability with neighborhood-level indicators, few address detailed street-level features due to data scarcity. This limitation poses significant challenges in understanding how various built environment (BE) features collectively influence bikeability. Additionally, although protected cycle lanes (segregated facilities) are widely recognized as effective for enhancing the cycling environment, their spatially heterogeneous effects on bikeability remain underexplored. This study presents a comprehensive framework using deep learning techniques and street view images (SVIs) to evaluate bikeability, validated using data from a docked bike-sharing system. Factor analysis is applied to integrate BE features into a bikeability evaluation framework, followed by a Multiscale Geographically Weighted Regression (MGWR) model revealing the hierarchical structure of bikeability: Accessibility & Feasibility, Safety, and Comfort & Pleasurability. The spatial results highlight the need for tailored measures in different functional areas to promote cycling behavior. These insights challenge the prevailing assumption that protected cycle lanes always ensure better bikeability and emphasize the importance of considering both street-level and neighborhood-level features to evaluate the biking environment heterogeneously. This study provides valuable policy implications for enhancing bike-friendly environments, especially in historical areas with distinct spatial characteristics.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)