Leveraging spatial relationships and visualization to improve public transit performance analysis

Abstract

Public transit agencies collect a tremendous amount of data in order to measure bus performance, including vehicle positions and passenger counts. These data are typically organized in the same way that the network is organized: split up by route, with each route further divided into stop-to-stop segments or timepoint-to-timepoint segments. This type of structure suffers from three drawbacks. First, it does not capture the spatial relationships between different routes. Second, route and stop identifiers are arbitrary and change over time, making it difficult to compare performance across time periods. Finally, even the stop-to-stop resolution is insufficient for certain applications, such as planning for transit priority infrastructure at the block or intersection level.

This thesis addresses those three issues by developing new practical methods that incorporate the geography of a transit network into bus performance measurement and analysis. These tools can be used to automate transit planning tasks that have typically involved specialized knowledge and considerable manual effort. Furthermore, each of the methods is intentionally designed to support visualization of performance data as an alternative to tabular representation in order to facilitate the identification of spatial patterns. A total of eight case studies, developed in concert with transit agency staff, are included to demonstrate how spatial analysis and visualization can address real transit planning challenges.

First, a map matching algorithm is described that facilitates the identification and classification of corridors served by multiple bus routes. A framework is then established for systematically aggregating different types of performance metrics across parallel routes, even if those performance metrics are only available at the stop-to-stop segment level. Case studies show how corridor identification and performance aggregation can be used to improve transit priority infrastructure planning, schedule coordination for parallel routes, and balancing service in local-express corridors.

Next, a method for increasing the resolution of performance data to the block-to-block level is proposed. Stop-to-stop segments are split at intersections and bus stops to create a unit of analysis that experiences uniform transit service across its 3 length. Performance measures are then assigned to the block-level segments, eliminating the dependence on arbitrary and fungible identifiers. This geography-based representation enables longitudinal comparison of performance that automatically captures changes in transit service as well as route and stop numbers. Two case studies demonstrate how these methods can be used to map the evolution of bus networks and ridership over many years.

Finally, a process is developed for extending the previous methods to include origin-destination (OD) estimates, enabling the spatial analysis and visualization of passenger journeys throughout the transit network. It also allows OD-based performance metrics that are not available from traditional sources to be assigned to blockto-block segments for longitudinal comparison. Case studies illustrate the strength of these methods in mapping the journeys of passengers whose trips involve a transfer, and for exploring travel pattern changes before and after route modifications.

Future research in this area includes the development of new bus performance metrics that incorporate spatial relationships between bus routes. Richer data sources, such as vehicle positions with a high sampling rate, could be leveraged to visualize travel speeds at the block-to-block level. Another research area is facilitated by the longitudinal analysis methods in this thesis: the ability to test theories related to bus network evolution over time. Finally, these methods lay the foundation for future research into new transit planning tools for strategic restoration of bus service after the COVID-19 pandemic.