Heterogeneous And Non-linear Effects Of The Built Environment On Bicycle Use And Bicycle Safety

This study aims to explore how to improve bicycle use and optimize a safe cycling environment by improving the built environment so that bicycle transport can play a better role in alleviating urban traffic congestion,reducing pollution and carbon emissions,and improving public health.From the perspective of micro individual behavior and macro spatiotemporal space,the thesis applies machine learning and utility theory to construct non-linear and heterogeneous analytical models,and takes Xi’an and Greater Melbourne as case cities to conduct empirical research on bicycle use and bicycle safety.The innovative research results:(1)We applied the Modal Choice of Generalize additive mixed model(MCGAMM)based on non-linear assumptions and the binomial logit model based on linear assumptions respectively and analyzed the heterogeneous association between the built environment and cyclist’s bicycle choice of shared bikes and private bikes.A case study of bicycle travel in Xi’an showed that the MCGAMM fitted better than the binomial logit model;the street pattern was linearly related to bicycle choice behavior: the average geodesic distance(an indicator of street pattern)and intersection density were both negative with regard to the likelihood of choosing a shared bike;an increase in the proportion of mixed bus/bicycle lanes(an indicator of bicycle lane facilities)non-linearly and significantly increased the likelihood of choosing a shared bike;an increase in the proportion of mixed transit/bike lanes(an indicator of bicycle lane facilities)non-linearly and significantly increased the likelihood of choosing to share a bicycle;the differences proved in the impact of various types of bicycle lane facilities(separated bicycle lanes by striped lines,green belt or barriers and mixed bus/bike lanes)on the likelihood of choosing to a shared bike among cyclists with higher and lower education,due to the fact that cyclists with different levels of education may have various needs and perceptions of different types of cycle lanes.(2)We combined e Xtreme Gradient Boosting(XGBoost)and SHapley Additive ex Planations(SHAP)to quantify and visualize the non-linear and interactive effects of the built environment on cycling distance,in terms of individual trips.The method identifies non-linear relationships between variables and uncovers the threshold effects of the built environment on cycling distance.A case study of bicycle travel in Xi’an showed that the street pattern indicators(average geodesic distance,intersection density,and network betweenness centrality)make the largest contribution to cycling distance(more than 50%);road networks with average geodesic distance with under 2.8 and intersection density with above 30 per square kilometer increase cycling distance.This suggests that cycling distance could be increased by street network form planning and design,such as connecting the cul-de-sac roads in areas with high intersection density to lower the average geodesic distance.(3)A generalized additive mixed model with Negative binomial distribution(GAMMNB)was constructed to explore the impact of the built environment on hourly bikeshare trips in traffic analysis zones(TAZs)as well as different districts.The spatial-temporal heterogeneity of the relationship between land use and bicycle trips is revealed by considering the effect of spatial-temporal correlation on the model estimation.A case study of bikeshare trips in Xi’an shows that the density of land use(e.g.,residential areas,leisure facilities,and car parking lots)is linearly related to the number of bicycle trips in some districts,and non-linearly in others.This suggests that land use patterns have spatial heterogeneous impacts on bikeshare use,and attention should be paid to planning sub-area land use patterns to increase the use of bicycles.(4)A linear model of Coregionalization(LMC)and a semiparametric geographically weighted poisson regression(s GWPR)to analyse the spatially heterogeneous relationship between the built environment and bike accidents.A case study of Greater Melbourne bike accidents shows that: the combined model has better explanatory power than the traditional s GWPR model based on AICc(Akaike Information Criterion)values;the effects of residential road density,residential land use and bike lane intersection density vary with spatial location;pavement density and small-scale commercial land use are associated with cycling accidents were positively correlated with accident occurrence.This study applies the LMC model to the field of accident analysis for the first time,providing a new perspective and attempting to study this heterogeneous relationship.The main contribution of the thesis is to propose suitable models to estimate and identify non-linear and heterogeneous impacts between the built environment and bicycle travel based on spatial and temporal dimensions and granularity(macro and micro)characteristics of data segmentation,extending the models and theories of studying the relationship between the built environment and bicycle travel.