Identifying the determinants of metro ridership is essential for metro planning and passenger flow management. However, few studies to date have empirically examined how accessibility affects metro ridership and even fewer have emphasized the non-linear impacts from a spatiotemporal perspective. This study demarcates station areas via the network-distance method and precisely quantifies the accessibility of metro stations both internally and externally. This is combined with a gradient boosting regression trees (GBRT) model and a Shapley additive explanations (SHAP) model to understand the non-linear impacts of accessibility on metro ridership from a spatiotemporal perspective. The results show that accessibility indicators collectively contribute more than 60% of the predictive power for metro ridership at different times and the external accessibility has a greater impact on metro ridership than internal accessibility. Some indicators, such as the shortest path and population density show threshold effects on metro ridership. More importantly, the results demonstrate significant spatial heterogeneity in the effects of accessibility indicators on metro ridership and geographic trends generally from urban to suburban areas. The findings are expected to help planning departments and transit agencies improve the coordinated development of metro systems.

确定地铁乘客量的决定因素对于地铁规划和客流管理至关重要。然而，迄今为止，很少有研究从经验上检验可达性如何影响地铁乘客，从时空角度强调非线性影响的研究更少。本研究通过网络距离方法划分车站区域，并精确量化地铁站内部和外部的可达性。这与梯度提升回归树 (GBRT) 模型和 Shapley 加性解释 (SHAP) 模型相结合，以从时空角度理解可达性对地铁乘客量的非线性影响。结果表明，可达性指标在不同时间共同贡献了超过 60% 的地铁乘客预测能力，外部可达性对地铁乘客的影响比内部可达性更大。一些指标，例如最短路径和人口密度，显示出对地铁乘客量的阈值影响。更重要的是，结果表明，可达性指标对地铁乘客量和从城市到郊区的地理趋势的影响存在显着的空间异质性。研究结果有望帮助规划部门和交通机构改善地铁系统的协调发展。例如最短路径和人口密度对地铁乘客量显示出阈值效应。更重要的是，结果表明，可达性指标对地铁乘客量和从城市到郊区的地理趋势的影响存在显着的空间异质性。研究结果有望帮助规划部门和交通机构改善地铁系统的协调发展。例如最短路径和人口密度对地铁乘客量显示出阈值效应。更重要的是，结果表明，可达性指标对地铁乘客量和从城市到郊区的地理趋势的影响存在显着的空间异质性。研究结果有望帮助规划部门和交通机构改善地铁系统的协调发展。