A crash prediction model for weaving sections in the Netherlands

Weaving is defined as the crossing of two streams travelling in the same direction along a significant section of the road without the assistance of traffic control devices. Merging and diverging vehicles need to make one or more lane-changes in a limited space and time, determined by the weaving section length. This situation creates intensive lane-change maneuvers, combined with heavy traffic volumes and variability in the speeds of the weaving and non-weaving vehicles. This often results in safety and operational problems. In the literature few studies developed crash prediction models for weaving sections. Therefore, the main objective of this study is to investigate how different geometric and traffic related variables affect the safety performance of motorways’ weaving sections, and develop a quantitative model for crash prediction. A sample of 110 weaving sections distributed all over the motorway network in The Netherlands was included. A database composed of the traffic and geometric characteristics of the weaving sections, and of their crash records was prepared. A Negative Binomial regression model was developed and the factors that mostly influence the crash frequency at weaving sections were identified. The results show that crash frequencies of weaving sections are significantly affected by the length of the weaving section, the average annual daily traffic (AADT), the percentage of weaving cars, the number of lanes on the main motorway and the location of the weaving section relative to the interchange (if inside or outside the interchange).