Bayesian models with spatial autocorrelation for bike sharing ridership variability based on revealed preference GPS trajectory data

Abstract

Increasing bike sharing ridership to reduce traffic congestion and air pollution has become the major goal of policy makers and transportation planners. In this study, aggregated macro-level models are developed to investigate and quantify correlations of bike sharing ridership variability in the city of Seoul. The bike sharing kilometres travelled (BSKT) is introduced as the ridership measurement, and a revealed preference method for its estimation is proposed using more than five million trip trajectories of global positioning system (GPS) collected from the bike sharing system of Seoul city. Six different models grouped into three categories are estimated, including linear regression, Bayesian model without spatial effect and Bayesian models with spatial autocorrelation. Five categories of variables are used to explain the variability of BSKT in Seoul. The contribution of these variables in intra-zonal BSKT and inter-zonal BSKT are also modelled and compared. The findings suggest a strong correlation among BSKT, bike sharing stations and bicycle network variables. The model comparison results indicate that the performances of the models with spatial autocorrelation are considerably improved.