A network-distance-based geographically weighted regression model to examine spatiotemporal effects of station-level built environments on metro ridership.

Understanding the relationships between built environment features and metro ridership is crucial to transit-oriented development (TOD). This study proposes a geographically weighted regression model based on metro network distance (ND-GWR) to capture the local patterns of relationships between the built environment (BE) and station-level metro ridership. The ND-GWR calculates network distances rather than traditional distance metrics (e.g., Euclidean, Manhattan, or Minkowski) to calibrate the spatial weight matrix, which can more truly characterize the associations among stations in the real world. Five types of independent variables, including demographic, land use, network, transfer, and station characteristics, are calculated by an exponential distance-decay function. This study also compares the modeling results across holidays, weekdays, and weekends to analyze temporal variations of BE's effects. Using data collected in 2019 in Shenzhen, China, the modeling results show that (1) the integration of ND-GWR and the decay-distance function can estimate the relationships between BE variables and metro ridership with higher accuracy; (2) bike-metro access-integrated use has positive impacts on metro ridership, and it presents more significant in the city periphery and on weekdays; (3) senior citizens (aged over 60) are unwilling to use the metro service due to the defect of their physical functions. The negative effects exhibit smaller in the city center and on holidays; (4) whether the effects of BE variables vary over space is highly affected by when the event happened. The findings are expected to provide insights for calibrating spatial models and increasing the attractiveness of metro systems. [ABSTRACT FROM AUTHOR]