How the built environment promotes public transportation in Wuhan: A multiscale geographically weighted regression analysis

Abstract

During rapid urbanization, the optimization of the built environment in metropolises to promote the use of public transportation considerably eases the road traffic pressure. Researchers usually adopt census or survey data and global models, which inadequately represent huge population and reflect the spatial non-stationarity. This study collected the location-based service data to identify commuters and their commuting modes and aggregated the share of transit commuting to traffic analysis zones (N = 586). While controlling three socio-economic attributes (age, gender & income), the 5Ds (Density, Diversity, Design, Distance to transit, and Destination accessibility) was formed to describe the built environment (BE) in Wuhan. Then we mainly compared the regression results of the geographically weighted regression (GWR) and multiscale geographically weighted regression (MGWR). Although the GWR model could reveal the spatial non-stationarity by local fitting, its regression coefficients with fixed bandwidth (BW = 153) tended to show high variability with abnormal sign reversals due to overfitting. The application of MGWR model with adaptative bandwidths could further consider the scale effect and provide robust estimation results. According to the various bandwidths provided by MGWR, the policies to improve the BE had been discussed in three scales. We emphasize that policies should be formulated based on the spatial distribution of the coefficients, that is, “Adapt to local conditions”.