Multivariate models to investigate the relationship between collision risk and reliability outcomes on horizontal curves

Abstract

This paper proposes a reliability-based framework to address the risk associated with limitations in the Available Sight Distance (ASD) on curved highway segments considering a three-dimensional (3D) sight distance computation approach. To facilitate this assessment, the ASD on horizontal curves was evaluated and an accurate inventory of curve attribute information was generated using LiDAR (Light Detection and Ranging) data in an automated and efficient manner. These datasets were then used to estimate the risk (probability of noncompliance, P_nc) associated with sight distance insufficiencies. Full Bayes multivariate Poisson log-normal safety performance functions were developed to relate the P_nc to the expected number of collisions. The results show that there was a statistically significant relationship between P_nc and collision frequency. There was also a significant correlation of 0.444 to 0.452 across collision severity levels indicating that curves with high Property-Damage-Only (PDO) collisions could be associated with higher injury and fatal (I + F) collisions. It was also found that P_nc had a greater impact on increasing PDO collisions than I + F collisions, suggesting that collisions associated with insufficient sight distance are likely to be less severe. The results of this analysis are expected to improve our understanding of the risks associated with deviations from design guidelines and quantitatively assess the safety margins due to these variations. The framework presented in this paper can be used to compare different design alternatives and investigate the influence of design deficiencies on collision occurrence across various severity levels.

Keywords: 3D sight distance; Full bayes; Horizontal curves; Multivariate models; Reliability analysis; Safety performance functions.