Constructing a Routable Transit Network from a Real-time Vehicle Location Feed

Abstract

GIScience 2016 Short Paper Proceedings Constructing a Routable Transit Network from a Real-time Vehicle Location Feed Nate Wessel 1 , Jeff Allen 1 , Steven Farber 2 Department of Geography and Planning, University of Toronto, Sidney Smith Hall, 100 St. George Street, Toronto, ON M5S 3G3, Canada Email: {nate.wessel; jeff.allen}@mail.utoronto.ca Department of Human Geography, University of Toronto Scarborough, 1265 Military Trail, Toronto, ON M1C 1A4, Canada Email: steven.farber@utoronto.ca Abstract Many transit agencies publish open access feeds of real-time fleet locations for use by application developers to deliver mobile navigation services. However, by collecting a transit data feed over time, we demonstrate how to recreate retrospective, routable transit networks that are useful in answering network performance and accessibility research questions. In this research we develop an end-to-end GIS toolchain for 1) downloading and storing a transit data feed, 2) coercing the collected spatiotemporal database into a retrospective transit schedule adhering to the General Transit Feed Specification (GTFS) data standard, and 3) creating a routable transit network with time-dependent travel times in OpenTripPlanner. We further demonstrate how this toolchain can be used to identify discrepancies between scheduled and actual travel times on the network and motivate the usefulness of this approach through an accessibility analysis. 1. Introduction The emergence of the General Transit Feed Specification (GTFS) data standard, and the publication of GTFS packages by thousands of agencies worldwide has unleashed a flurry of tools development for researching transit networks (Hadas and Ranjitkar 2012), transit travel times (Farber et al. 2014), and time-dependent accessibility metrics (Fransen et al. 2015; Owen and Levinson 2015; Farber et al. 2016). One problem with travel time research based on GTFS schedules is that this format implicitly ignores inaccuracies in travel times due to, among other causes, operational delays, service interruptions or unrealistic schedules. Researchers requiring more accurate measures of transit travel times have relied on field measurements and large-scale simulation models of multimodal network assignments. But these sources of data are expensive to collect, may take years to implement, may not be very accurate, and are difficult to implement on a large network, in continuous time, and in perpetuity. To account for these shortcomings, we put forward a new methodology that capitalizes on freely available vehicle location data feeds to produce a routable retrospective transit network using open source tools. This network contains actual travel times that are true to observed transit network performance, not the expected performance contained in the official transit schedules. In this paper we will describe the toolchain we developed and demonstrate its use in detecting service that differs from the schedule, as well as the implications of these differences for end-to-end travel times and time-dependent accessibility scores. 2. Data and Methods This research uses two primary datasets, a current GTFS package for the Toronto Transit Commission (TTC) and a 48-hour extract of all vehicle locations reported by the TTC-NextBus