Exploring the multiscale relationship between the built environment and metro station ridership

• The multiscale geographically weighted regression (MGWR) is employed to estimate Bangkok metro ridership. • The MGWR model was found to reveal spatial heterogeneity and scale effects with better goodness of fit than the other models. • By K-means clustering, the Bangkok metro stations were classified into five groups based on their built environment factors. • The estimated coefficients of each station provide local insights to determine Bangkok’s functional land use policy. Accurately estimating passenger volume is essential for metro station planning. The number of station-level ridership is a crucial variable to capture relationships with land use factors surrounding metro stations. Understanding different metro station characteristics with their significant built environment factors from a local perspective is vital for transportation agencies. Previous studies assumed that the relationships between passenger volume and independent variables are described by global or local models separately, ignoring spatial scale. In this study, the multiscale geographically weighted regression (MGWR) is employed to address spatial autocorrelation issues found in ordinary least squares (OLS) regression and previous versions of geographically weighted regression (GWR) by adjusting the bandwidth for each variable (spatial scale). The Bangkok metro ridership data were collected from automated fare collection (AFC) in June 2019. Employed explanatory factors include land-use-related points of interest (POIs), intermodal transportation attributes, and network structure, including centrality analysis. The OLS, GWR and MGWR models were estimated and compared. We found that the MGWR model, which reveals spatial heterogeneity and scale effects, provides better goodness of fit based on the residual sum of squares, R-squared, and AIC compared than the other models. The eight variables that have significant positive correlations with the ridership demand are commercial POI, attraction POI, residential POI, industrial POI, degree centrality, betweenness centrality, terminal station, and feeder transport variable. By K-means clustering, the Bangkok metro stations were classified into five groups based on their influenced built environment factors. The estimated coefficients of each station provide local insights to determine Bangkok’s functional land use policy promoting the usage of metro service.