Decoding the Spatial Heterogeneity of Bike-Sharing Impacts: Machine Learning Model of Meteorology, Epidemic, and Urban Factors

Previous studies on the factors affecting bike-sharing travel (BST) have not considered spatial differences, leading to insufficient understanding of the complex impacts of variables in different geographical locations. This study aims to reveal the differential spatial impacts of meteorological conditions, epidemics, and urban spatial variables on BST. Firstly, New York was selected as the study area, and the period from 2020 to 2021 was chosen for the study. Secondly, a high-precision urban information data set, including meteorological, epidemic, and urban spatial variables, was constructed using weighted Thiessen polygons as the segmentation method. Finally, machine learning was conducted, and the XGBoost ensemble learning algorithm, which yielded the best training results, was chosen for interpretable analysis. This examined the nonlinear correlations and spatial benefits of each variable with BST. The results show that (1) the impact of average temperature on shared bicycle travel is most significant among all factors, accounting for 26.15% of the total impact; (2) there is significant spatial heterogeneity in the influence of factors, and office closeness is negatively correlated with BST, contributing positively in the west and negatively in the east; (3) the southern part of Manhattan is significantly affected by meteorological (divided by SHAP value divided by = 484.18) and urban spatial sector (divided by SHAP value divided by = 122.65), while the central part of Manhattan is most significantly influenced by epidemic variables (divided by SHAP value divided by = 469.27). In summary, this study takes New York as an example to analyze the nonlinear effects and spatial benefits of meteorology, epidemics, and urban space on shared bicycle travel. Based on this, more targeted and effective urban traffic intervention strategies are provided for different regions of the city.