Abstract Manager

Trains are an important part of transit services that provide superior travel time and travel time reliability compared to bus services. Most of the previous research focused on train ridership associations with link-level, demographics, and built–environment variables. A few studies presented transit ridership prediction methods. In practice, modelers and planners rely on disaggregate travel demand models such as the four-step mode or the activity-based model when available. This is because these models are sensitive to variables such as population, employment, transit fare, and transportation network. On the other hand, a full run of a travel demand model takes hours. This study presents a novel route-based accessibility approach to develop a stand-alone train ridership prediction model with seconds of runtime for predictions. The proposed approach mainly accounts for 1) the total number of trips between OD that may use the train, 2) the share of trips that use trains instead of other modes (e.g., auto, bus, walk) based on the travel time and travel time perception, and 3) train trips distribution between stations. The proposed approach is applied using Metro Vancouver’s Skytrain ridership data which consists of 53 stations distributed over three lines. The model employs travel time between 1700 TAZs in Metro Vancouver during the AM and PM peaks extracted from The Greater Vancouver Regional Travel Model (RTM). Also, the model employs population and employment variables at the TAZ level. The developed train ridership model was then validated against several runs of the RTM to test the model's ability to predict the train ridership change due to changes in population and employment especially around the train station’s location. The validation using the RTM shows a mean absolute percentage error below 10%.