Traffic safety in relation to multidimensional street network and land use features: A nonlinear analysis with population heterogeneity

Road traffic crashes remain a critical concern for public health and sustainable urban development. In recent years, there has been an emerging interest in applying nonlinear approaches to examine the relationship between the built environment and crash occurrence. Extending this line of inquiry, this research examines how diverse dimensions of street network configuration (geometry, hierarchy, topology) and land use (type, intensity, diversity) influence traffic crash density (i.e., crashes per unit area), and investigates how such effects vary by zonal population composition within a nonlinear framework. Using data from Wuhan, China, and gradient boosting decision trees, we find that street topology and land use density are the most influential correlates in explaining the variation in crash density. Nonlinear associations of street network and land use characteristics with crash density are prevalent, with most variation occurring within a specific threshold range. Moreover, the effects of these built environment characteristics vary significantly across zones with differing age and income structures. Zonal elderly population density amplifies the effects of most street network and land use characteristics on crash density. Low-income zones demonstrate a greater sensitivity to changes in certain built environment features, such as street density, residential land ratio, and land use diversity, resulting in more pronounced increases in crash density. Our findings provide a more comprehensive and nuanced understanding of the links between the built environment and traffic safety, and call for both recognitional and distributive considerations of spatial justice to be incorporated into traffic safety interventions.