How to improve public environmental health by facilitating metro usage on weekend: exploring the non-linear and threshold impacts of the built environment

Abstract

Introduction: The accelerated motorization has brought a series of environmental concerns and damaged public environmental health by causing severe air and noise pollution. The advocate of urban rail transit system such as metro is effective to reduce the private car dependence and alleviate associated environmental outcomes. Meanwhile, the increased metro usage can also benefit public and individual health by facilitating physical activities such as walking or cycling to the metro station. Therefore, promoting metro usage by discovering the nonlinear associations between the built environment and metro ridership is critical for the government to benefit public health, while most studies ignored the non-linear and threshold effects of built environment on weekend metro usage.

Method: Using multi-source datasets in Shanghai, this study applies Gradient Boosting Decision Trees (GBDT), a nonlinear machine learning approach to estimate the non-linear and threshold effects of the built environment on weekend metro ridership.

Results: Results show that land use mixture, distance to CBD, number of bus line, employment density and rooftop density are top five most important variables by both relative importance analysis and Shapley additive explanations (SHAP) values. Employment density and distance to city center are top five important variables by feature importance. According to the Partial Dependence Plots (PDPs), every built environment variable shows non-linear impacts on weekend metro ridership, while most of them have certain effective ranges to facilitate the metro usage. Maximum weekend ridership occurs when land use mixture entropy index is less than 0.7, number of bus lines reaches 35, rooftop density reaches 0.25, and number of bus stops reaches 10.

Implication: Research findings can not only help government the non-linear and threshold effects of the built environment in planning practice, but also benefit public health by providing practical guidance for policymakers to increase weekend metro usage with station-level built environment optimization.

Keywords: built environment; machine learning; metro ridership; nonlinearity; public environmental health.