Understanding and analysis of bicycle travel and safety

Abstract

Cycling has received more and more attention in urban and transport planning in recent years. As an active transport mode, cycling does not only relieve traffic congestion and reduce vehicle emissions, but also improves the well-being of society. Despite the benefits for health and environment, bicyclists are vulnerable to injuries and mortalities in road crashes. It is crucial to identify the influencing factors that affect the bicycle crash risk. Therefore, effective countermeasures can be implemented to improve overall bicycle safety. In this study, effects of policy interventions on bicycle travel and safety are examined, based on comprehensive traffic and crash data. For example, policy interventions including low emission zone, congestion charging scheme, and public bicycle rental scheme are considered. The propensity score matching method is applied to account for the effects of confounding factors like built environment and population socio-demographics. Results indicate that bicycle travel increases remarkably after the implementation of low emission zone, especially for short and intermediate bicycle trips. However, bicycle crash frequencies also increase after the introduction of congestion charging and public bicycle rental schemes. On the other hand, association between built environment, population socio-demographics, road network configuration, traffic characteristics, and bicycle crash frequency at zonal level is measured, with which the bicycle crash exposure is accounted. For example, bicycle usage data from the public bicycle rental system is used to estimate the bicycle crash exposure. In addition, a weighted shortest path approach is proposed to estimate the bicycle distance travelled, with which the configuration of cycle lane network and safety perception of bicyclists are considered. Results indicate that bicycle crash frequency model that incorporates bicycle distance travelled as exposure is superior to those using bicycle time travelled and bicycle trip frequency as exposure. Furthermore, factors including land use, bicycle infrastructure, population density, gender, age, median household income, and weather condition are found to affect bicycle crash frequency, after controlling for the effects of unobserved heterogeneity and spatial correlation. Last but not least, advanced statistical and deep learning models are developed to resolve the prevalent problems in safety analysis. For example, a multivariate Poisson-lognormal regression model is developed to account for the correlation between the frequencies of different bicycle crash types. Furthermore, imbalanced crash data and boundary crash problems are resolved using the deep learning approaches including augmented variational autoencoder and crash feature-based allocation methods. Results indicate that crash frequency models developed using the aforementioned approaches have better prediction performances. More importantly, more influencing factors can be identified. To sum up, findings of this study can enhance the understanding on the roles of environmental, physical, social, and political factors in bicycle travel and safety. This should shed light on the optimal urban planning, engineering design, and transport policy that can promote bicycle travel and improve bicycle safety in the long run.