Decoding Street-Level Dockless Bike-Sharing Cycling Flow Patterns: Insights from Physical–Perceptual Environments and Explainable Machine Learning

Promoting active travel modes, such as dockless bike-sharing (DBS) usage, is essential for achieving urban carbon neutrality goals and advancing sustainable transportation. However, the nonlinear relationship between the multilevel built environment and DBS cycling behavior at the street level remains underexplored. Therefore, this study constructs an integrated measurement framework that focuses on street-level DBS cycling flow and incorporates multilevel built environment factors from both physical and perceptual dimensions. Furthermore, this study integrates the Gradient Boosting Decision Tree (GBDT) model and SHapley Additive exPlanations (SHAP) method to reveal the nonlinear effects of the built environment on DBS cycling flow. The results reveal that the following: (1) The physical neighborhood environment plays a dominant role in shaping DBS cycling flow, with residential density emerging as the most influential predictor. (2) Several environmental factors exhibit pronounced multi-stage variation patterns and temporal heterogeneity in their effects on street-level DBS flow. (3) Identifying dominant spatial patterns, localized clustering structures, and unit-level effect mechanisms of environmental influences enables the formulation of targeted and precision-oriented intervention strategies. This study advances street-scale analytical framework for examining the relationship between the built environment and DBS travel, and provides practical insights for cycling-friendly spatial renewal and the optimization of street-level cycling systems.