Near “real-time” estimation of excess commuting from open-source data: Evidence from China's megacities

Urban commuting has continuously fascinated scholars and decision-makers. As few people live and work in the same place, there is always excess commuting (i.e., the non-optimal or surplus work travel occurring in cities because people do not minimize their journeys to work for most residents). Traditional commuting data sources (e.g., questionnaires and census surveys) are challenged by small samples, high cost, and low spatiotemporal resolution. In contrast, the big social-sensing data (e.g., smart card and mobile phone data) only consider one or two traffic mode of a route, which is not consistent with the real-life condition. This article proposes a framework for modeling excess commuting based on open-source data of the ten most populous megacities in China. We downloaded residential points of interest (POIs) from Lianjia Real Estate website and obtained workplace POIs from China's AMAP, which is widespread used as Google map. The stratified sampling approach was employed to derive commuting pairs. Both commuting distance and time were obtained by the shortest path under public transportation from AMAP. Then, the linear programming method was employed to calculate the theoretical minimum commuting time and distance of each city. We analyzed the statistical property and spatial distributions of excess commuting and found that (1) commuting distances and time (ranging from 9.1 to18.1 km and from 44.8 to 74.3 minutes) of all ten megacities follow a left-skewed normal distribution; (2) in terms of commute cost, all cities show universal core-periphery patterns where the spatial heterogeneity of the commuting time is more significant than that of distance; (3) for each city, the excess commuting measured by time (i.e. from 0.61 to 0.79) is lower than that measured by distance (i.e. 0.68 to 0.89); and (4) the role of mixing land use, waterbody distribution, and centripetal urbanization on urban commuting distance and time is significant.