Discovering the spatio-temporal impacts of built environment on metro ridership using smart card data

Abstract

The primary objective of this study is to identify the spatio-temporal relationship between metro ridership and built environment by integrating smart card data with online point of interest (POI) data. This study proposes the geographically weighted regression (GWR) model to address the spatial autocorrelation and nonstationarity of metro ridership at the station level. Different from the Euclidean distance (ED) commonly used in the traditional GWR, the Minkowski distance (MD) metric is introduced into the study to measure geographical distance and improve weighting matrix calibrations. This study provides useful insights in adaptive model settings based on selections of spatial range and distance metric parameters. A case study is conducted in Nanjing, China, by utilizing large-scale metro smart card data and POI data obtained from the Gaode map by a web crawler written in Python. The comparative analysis indicates that MD-GWR achieves better goodness-of-fit than the global ordinary least squares (OLS) and traditional GWR in the case of modeling metro ridership. Finally, this study quantifies the impacts of built environment at both spatial and temporal scales, under the proposed modeling structure. We examine the existence of job-housing separation and corresponding areas in spatio-temporal ridership analysis. Impacts of non-commuting activities and intermodal connection on boarding and alighting ridership differ in spatial and temporal ranges. The proposed modeling structure promotes modeling precision and thus enhances the understanding of the relationship between station-level ridership and surrounding built environment, which can provide useful guidance for planning departments and transit agencies to implement targeted policies and create accessible, livable and vibrant communities.