Time-varying and non-linear associations between metro ridership and the built environment

The metro is the backbone of the transport system in many cities. Analyzing the built-environment correlates of metro ridership is crucial for transit-oriented development (TOD) planning and practice. Although numerous studies went along this line, they have rarely considered the non-linearity and temporal heterogeneity in the association of metro ridership with the built environment. After collecting transit smart card data, geo-data, and mobile phone signal data, this study adopts the random forest model to reveal the complex association of hourly metro ridership in November 2019 in Chengdu (China) with the built environment in three times of day (i.e., morning peak, noon off-peak, and evening peak hours). Notably, the contribution of several variables, such as the number of station entrances/overpasses and parking density, has rarely been considered in the literature. The results confirm the presence of non-linearity and temporal heterogeneity in the aforementioned association. Access to the city center and population density are strong predictors of metro ridership in the morning peak hour, whereas employment density, enterprise density, and road density are strong predictors in the evening peak hour. There are great differences in the correlates of metro ridership in different periods. Critical TOD planning parameters are also identified from the partial dependence plots obtained from random forest modeling.

Automated summary

The metro plays a crucial role in urban transportation systems. Understanding the relationship between metro ridership and the built environment is important for transit-oriented development planning. This study, using data from Chengdu (China), explores the complex association of metro ridership with the built environment during different times of day. The results show non-linearity and temporal heterogeneity in this association, and identify key predictors for metro ridership.