Nonlinear effects of built environment on metro-feeder bike-sharing as a dynamic green exposure pathway: a dual perspective of connection frequency and modal share

Background Bike-sharing integrated with metro transit can be conceptualized as a potential dynamic green exposure pathway linking blue-green infrastructure, active mobility, and public health. The built environment is a key determinant, but existing research focuses on connection frequency and neglects modal share, which better reflects feeder efficiency and competitiveness. Objective This study compares built environment effects on metro-feeder bike-sharing connection frequency and modal share and identifies nonlinear thresholds. Methods Using Tianjin, China as a case study, we integrated multi-source travel data with geospatial data. The Green View Index (GVI), derived from Baidu Street View via deep learning, served as a station-catchment-level proxy for potential street-level green exposure. Actual cycling trajectories and health outcomes were not measured. A LightGBM-SHAP model quantified variable contributions and uncovered nonlinear associations. Results Three findings emerge. First, connection frequency concentrates in the city center, whereas modal share clusters in suburban ring zones, suggesting spatial unevenness in access to bike-sharing as a green feeder mode. Second, drivers differ: connection frequency is mainly explained by governmental (39.8%), commercial (11.5%), and cultural-educational POI density (9.4%), reflecting commuting-oriented mechanisms; modal share is dominated by healthcare POI density (27.0%), distance to city center (15.0%), and bus stop density (10.0%), highlighting accessibility and inter-modal competition. Third, nonlinear thresholds are pervasive: governmental POI density turns positive beyond 30 POIs/km 2 ; GVI above 90 is associated with positive feeder cycling; bus stop density beyond 4 stops/km 2 shifts from complementarity to competition; and floor area ratio has an inverted U-shaped effect on modal share, positive at 0.5–2.1. Conclusion This study is the first to quantitatively disentangle the differentiated nonlinear effects of the built environment on bike-sharing connection frequency and modal share from both “volume enhancement” and “efficiency optimization” perspectives. The identified thresholds provide evidence-based planning guidance for optimizing metro-feeder bike-sharing systems and supporting climate-adaptive and more spatially equitable green feeder access. Because GVI is a station-catchment proxy and health outcomes were not measured, the health-equity implications are interpretive rather than causal.