An origin-destination based train station choice model for new public transport connections to train stations

Abstract

A change in the access and egress system, like a new metro line, will automatically influence the station choices of train travellers. Given the fact that several new large public transport connections will be opened in the near future, there is the need to precisely forecast how the distribution of train travellers among stations will change, as it may give cause to (logistical) changes in the operation of the railway system. The most important factors for choosing at which station people will access the train are their origin and destination (ATOC, 2009; Hoogendoorn-Lanser, 2005; Debrezion, et al., 2007). Nevertheless, the current prognosis model of NS, De Kast, is however not able to realistically forecast the effect of new or significantly improved public transport connections to train stations on the distribution of train travellers among these stations as it does not take into account the train travellers' origins and destinations and different access/egress modes. This research therefore develops and estimates an origin-destination based station choice model that takes into account the access/egress modes walking, bicycle, car, and public transport. For an origin-destination based approach, detailed information about train travellers' origins and final destinations is required (ATOC, 2009; Wardman, et al., 2007). Nevertheless, filling a complete origin-destination matrix for a detailed geographical scale for all train travellers in the whole Netherlands would require massive amount of data, which simply is not available. Therefore, in this study, the city of Amsterdam was treated as an internal area with origins and destinations at the lowest geographical scale, hence the four-digit postal codes. Then it has been examined which destination stations in the rest of the Netherlands had the same distributions among the origin stations in the internal zone, hence the Amsterdam region. The results indicate that from the first intercity station onwards, where one would always travel to by an Intercity train instead of by a local train, the station choice probabilities are equal. This implies that all origins and destinations from the specific intercity station onward can be grouped into the same external zone. This implies that travellers that travel from the internal area to any destination station in the specific external zone would choose the same origin (intercity) station in the internal area to get there. On the short distance, this is however not justified as local trains mostly call at more stations in the internal area. Another external zone is therefore added between the long-distance external zone and the internal area, to account for the short distance trips. Then, using the survey data, per external zone the percentage of train travellers that travels to (inbound) or from (outbound) each internal zone is defined. This results in a relative distribution of train travellers between the considered in- and external zones for both in- and outbound trips. The shares are then multiplied by the number of travellers that travel from or to the external zones. It results in the number of train travellers between the in- and external zones, for in- and outbound trips. The danger of the revealed preference method is that it is unknown what the actual choice set was. To create choice sets that are logical and feasible to the respondent, conditions are applied to create origin-destination specific choice sets for each access/egress mode. Maximum travel times for walking and cycling are used to exclude unfeasible alternatives. Also, by conducting quantitative research based on the same survey data, the maximum extra time that people travel longer than the fastest possibility has been calculated for each access/egress mode both absolutely and relatively. Separately for each access/egress mode, utility functions are applied to the alternatives in the choice set to predict the station choice probabilities. As there appeared to be significant correlations among alternatives with the same access/egress mode, a nested logit model is used with the modes as nests. The included attributes that have been estimated using the Klimaat V survey that are significant, are the access/egress travel time by each mode, generalized travel time by train, intercity frequency at the station, number of metro lines at the station, number of bus/tram lines at the station, context constants for bicycle and car for inbound trips, and for public transport when travelling to an airport. The only significant differences between inbound and outbound trips are the mode specific constants of car and bicycle, caused by the vehicle availability at stations compared to at home. As for the rest, the choice behaviour is equal. The trip purpose of the train traveller appeared not significant at all. Finally, the predicted station choice probabilities per OD-pair are multiplied by the number of train travellers per OD-pair, which results in the number of train travellers per OD-pair that travels via a certain station with a certain access/egress mode. This is done for both the null situation and the new situation and the ratios between them are growth factors per station that form the input of De Kast. The model has been validated with the Ringlijn metro line case in Amsterdam. The model outcomes are comparable to the revealed results. It has finally been applied to forecast the effects of the Noordzuidlijn metro line in Amsterdam, with logical results: a steady situation at Amsterdam Centraal, 18% growth at Amsterdam Zuid, and a loss of passengers at most surrounding stations. When also including the future PHS timetable of NS in the model, the growth at Amsterdam Zuid rises to even 34%. The origin-destination based approach has appeared to be a strong tool for station choice predictions. By using revealed preference survey data to define the aggregation, calculate the OD-matrix, define the choice sets and estimate the model, it is possible to design an origin-destination based station choice model that gives logical forecasts and has a good model fit. Recommendations for further research and model improvement are to include vehicle availability and international travellers more extensively in the next Klimaat survey, and conduct a research about the behaviour of international and leisure travellers. Also, a limitation is that this model is currently limited to the current train travelling population. It should therefore be researched how the new train travelling population can be calculated for after the opening of new public transport connections to stations, or how the proposed model can cooperate best with the exogenous module that currently calculates that. Finally, a second validation is needed with a case in another city to see whether the parameters from the Amsterdam case are generically applicable. Other applications for the model might be forecasting the effects of new train stations and timetable changes.