Abstract:
Public transport (PT) is not famed for its dependability, yet current PT accessibility models presume that the schedule is an adequate reflection of reality. Consequently, it is implicit that PT vehicles always operate according to their schedule, and that all passengers arrive at random rather than plan their arrival. In reality, such assumptions do not always hold. The sensitivity of model results to such assumptions is largely untested.
This thesis seeks to determine to what extent PT accessibility models based on the schedule are robust to substitution by automatic vehicle location (AVL) data on the real arrival times of PT vehicles. The inclusion of this high resolution data on punctuality enables the relaxation of assumptions about the random arrival of passengers, and the enumeration of expected waiting time in the presence of uncertain PT arrival time.
Using an open standard for the publication of public transit information (the General Transit Feed Specification (GTFS)), this thesis develops a number of models ofWellington’s transit system, assuming that passengers arrive at random, and vehicles arrive punctually. Equivalent models are then built with AVL data from the Greater Wellington Regional Council’s real-time information (RTI) system. This information is used to determine actual service headways, headway variability, probability of service arrival over an interval, and realistic vehicle travel times.
Results of pairwise comparisons of accessibility model outcomes indicate that for every definition of accessibility model considered, GTFS and AVL sources of information do not predict similar travel times. The magnitude and the direction of differences varied between model types, but in all cases were highly statistically significant.
The results of the research are of use to transportation and accessibility modellers, who previously could not estimate travel time error associated with making assumptions about PT reliability and passenger arrival behaviour (implied by the use of PT schedules in models).
