Work

Do Station-Level Ridership Models Travel Through Time? Conditional Transferability and New-Station Limits in Kaohsiung MRT

Year: 2026

Type: article

Abstract: Station-level direct-demand models relate observed station use to network role, location, service, and catchment attributes. They are used to benchmark station demand and, when an external systemwide forecast is available, to allocate that total across stations. This study examines the temporal stability and conditional transferability of a compact role-and-location model across twelve annual cross-sections of Kaohsiung MRT. Interchange coefficients remain positive and distance coefficients negative throughout 2013–2024, even though the annual estimates are not literally constant. In one-year-ahead rolling tests, transporting the preceding year’s slopes and calibrating them to a supplied target system total raises RMSE by an average of 2.9% relative to an in-sample target-year refit; carrying the prior intercept without target information raises RMSE by an average of 5.1% and by as much as 21.1%. A separate boundary test for a genuinely new station shows that 2013 demand at the newly opened South Gangshan was 2.27 times a system-total-calibrated prediction based on strictly pre-opening 2011 data; the original near-opening 2012 specification produces a similar result. The model, therefore, demonstrates useful conditional transferability among established stations, especially when an external systemwide total is available for calibration. That transferability has a clear limit, however: the same underprediction indicates that forecasting newly opened stations requires information that the established-station model does not provide. Supplementary 2024 results show that adding the tested population, street connectivity, and feeder-route measures did not improve the compact role-and-location specification for that cross-section. Together, the results establish where compact direct-demand models are temporally stable, where calibration enhances transferability, and where extrapolation requires more than a transferred equation.

Source: Future Transportation

DOI: 10.3390/futuretransp6050191

Authors: Wil Martens, David Emanuel Andersson, Hsin-Kun Su

Institutions: National Sun Yat-sen University, Hungkuang University

Language: English

Topic: Transportation Planning and Optimization

Subfield: Transportation

Field: Social Sciences

Domain: Social Sciences

SDGs: Sustainable cities and communities, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $1,207