Development of Integrated Demand and Station Choice Models for Local Railway Stations and Services
Marcus Young, University of Southampton
This paper describes the development of models to forecast demand for new local rail stations that incorporate probabilistic catchments derived from station choice models applied to small-scale origin zones.
There has been substantial growth in rail passenger journeys in the UK and other European countries over recent years, and this is forecast to continue. This renaissance in rail travel has seen more than 50 new stations open in the UK during the last decade. However, the demand models that are typically used in the UK to assess the viability of proposed new station schemes have not always performed well, with many examples of substantial under- and over-prediction. While some improvements have been made in recent years, for example through the development of transferable trip end models calibrated using spatial regression techniques, these demand models still have a major flaw. Their representation of the areas served by stations is based on the definition of discrete and deterministic catchments that do not represent the complex interaction of real-world station catchments and do not allow for competition between stations. This paper describes the development of models that can overcome these problems and therefore provide a more accurate representation of station choice behaviour. These models are capable of forecasting the demand for a new station at any location in Britain, and incorporate probabilistic catchments derived from station choice models applied to small-scale origin zones.
In order to develop these integrated models, multinomial logit station choice models were first calibrated using some 15,000 validated trips obtained from on-train passenger surveys carried out in Scotland and Wales, incorporating a range of explanatory variables relating to the access journey, station facilities and service levels. These models were found to predict station choice substantially better than a base model that assumes the nearest station is always chosen. Multiple linear regression models were then developed to forecast passenger demand for all local stations in Great Britain, with the station catchments defined using either an established deterministic method or a new methodology based on the probability of each candidate station being chosen at each origin zone. Both model variants were used to predict demand at several recently opened stations, and the forecasts compared with actual usage figures.
While most prior station choice research has focussed on explaining choice behaviour in a local area, this study has developed transferable station choice models that are suitable for defining catchments for both trip end and direct-demand (flow) models in a range of contexts. This research has the potential to improve the models used to assess proposals for new railway stations and to enable better forecasting of the effects of changing service patterns, for example following the opening of new high speed lines or the introduction of new open access services. It also provides a methodology suitable for incorporating into standard demand forecasting guidance such as the British Passenger Demand Forecasting Handbook. While the modelling carried out in this study focused on the British context, the methods developed should be equally relevant to other contexts in Europe and beyond, particularly where major revisions or enhancements are planned for local rail services.
Association for European Transport