Assessing Congestion Spillover Effects of Freeway Crash Induced All Lane Closure Incidents: A Bayesian Copula-Based Approach

Abstract

Freeway Crash Induced All Lanes Closure (FCIALC) events are among the most severe non-recurrent incidents, often triggering substantial disruption across both the freeway and arterial network. This study develops a two-stage Bayesian modeling framework to quantify the interdependence between freeway lane closures and arterial congestion using real-world data. First, a hierarchical Bayesian Network (BN) is used to estimate the freeway all lanes closure duration (ALCD) levels based on pre-crash traffic conditions, crash severity, vehicle type, time of day, etc. By employing Bayesian Inference the model identifies critical scenarios involving fatal crashes and low pre-crash freeway speed as significant predictors of longer closure durations. In the second stage, a bivariate copula-based Bayesian regression model is used to jointly model freeway ALCD and arterial congestion duration. Results show that arterial congestion is significantly influenced by freeway ALCD, particularly when pre-crash arterial speeds are low. The Clayton copula outperforms other structures, indicating a positive lower-tail dependency, where short durations of freeway closure are often associated with short arterial congestion. The joint model substantially improves the predictive performance for arterial congestion duration over independent models, supporting the need to consider these dependencies in Integrated Corridor Management (ICM). Findings highlight the importance of optimizing diversion strategies based on real-time arterial capacity, broadcasting early warnings through multiple upstream Dynamic Message Signs (DMS), and adjusting signal timing of the surrounding arterial network to alleviate congestion. Given the sharp reduction in freeway capacity during FCIALC events, countermeasures such as temporary shoulder-use lanes or flex lanes managed by Lane Control Signs (LCS) can help restore freeway capacity. Moreover, the presence of lower-tail dependence in the joint model emphasizes that faster incident management is essential for faster network-wide recovery.