Unraveling Latent Transfer Patterns Between Metro and Bus From Large-Scale Smart Card Data

Abstract

Intermodal transfer patterns could provide a better understanding of urban mobility in an integrated transit system. However, the existing literature is still limited related to in-depth data-driven investigations of transfer patterns, particularly between metro and bus. In this study, we propose a generic framework to unravel latent transfer patterns at both aggregated and disaggregated levels. K-means clustering method is first used to classify metro stations based on surrounding built environment. We then introduce the cube technique which allows hierarchical aggregations for each dimension and cross-tabulations of massive ridership data. Last, we redesign the generative mechanism of the structural topic model that is capable of capturing the variability and interdependence in transfer patterns with passenger-level attributes. An empirical study is conducted with large-scale metro and bus smart card data, as well as built environment data in Nanjing, China. The results indicate intermodal transfers exhibit significant heterogeneous patterns in different transfer types over space and time. Although transfers per station in central business district (CBD) areas rank the highest, urban sprawl helps increase the utilization rate of transfer services in inner suburbs with most types of transfer behaviors. Strong correlations are identified among commuting transfer patterns, while correlations remain weak for pattern pairs between commuting and non-commuting activities, reflecting high consistency among commuting activities. Adults play a dominant role in commuting patterns in CBD areas and inner suburbs. Yet, students and the elderly are found to influence the pattern prevalence more effectively in outer suburbs and periphery areas.