Exploring the nonlinear and asymmetric influences of built environment on CO2 emission of ride-hailing trips

Emerging in many cities, ride-hailing is recognized as an approach to reduce car dependence and to lower CO2 emissions. Despite a number of studies on the impact of the built environment on travel behavior, the impacts of ride-hailing on carbon emissions is largely overlooked. Using gradient boosting decision trees (GBDT) method to deal with the data from Chengdu, China, this study examines the nonlinear influence of built environment on carbon emissions of ride-hailing trips at a disaggregated level. Meanwhile, the asymmetry of this influence is explored at origin and destination in several spatiotemporal contexts. The results show that, among the built environment variables, population density is the most crucial factor in predicting carbon emission, however there is a threshold effect. The distance to the subway station at origin and destination may have an opposite effect on emissions when asymmetric effects are taken into consideration. There is a 'U' type relationship between land use diversity and CO2 emission at the morning peak hour while pattern is different at evening peak. The impact of road density on CO2 emission does not show a consistent trend however display the asymmetry at origin and destination. These findings provide useful inputs to policymaking for ride-hailing management and sustainable urban development.