Identifying crash-prone traffic conditions under different weather on freeways

• We explore crash-prone traffic conditions under different weather conditions on freeways.
• We link crash risk with traffic flow variables by Bayesian random intercept logit model.
• Random intercept logit model improves goodness-of-fit of crash risk models.
• Impacts of traffic flow variables on crash risk are different across various weather conditions.
• Model predictive accuracy in clear weather is greater than that in adverse weather.

IntroductionUnderstanding the relationships between traffic flow characteristics and crash risk under adverse weather conditions will help highway agencies develop proactive safety management strategies to improve traffic safety in adverse weather conditions.MethodThe primary objective is to develop separate crash risk prediction models for different weather conditions. The crash data, weather data, and traffic data used in this study were collected on the I-880N freeway in California in 2008 and 2010. This study considered three different weather conditions: clear weather, rainy weather, and reduced visibility weather. The preliminary analysis showed that there was some heterogeneity in the risk estimates for traffic flow characteristics by weather conditions, and that the crash risk prediction model for all weather conditions cannot capture the impacts of the traffic flow variables on crash risk under adverse weather conditions. The Bayesian random intercept logistic regression models were applied to link the likelihood of crash occurrence with various traffic flow characteristics under different weather conditions. The crash risk prediction models were compared to their corresponding logistic regression model.ResultsIt was found that the random intercept model improved the goodness-of-fit of the crash risk prediction models. The model estimation results showed that the traffic flow characteristics contributing to crash risk were different across different weather conditions. The speed difference between upstream and downstream stations was found to be significant in each crash risk prediction model. Speed difference between upstream and downstream stations had the largest impact on crash risk in reduced visibility weather, followed by that in rainy weather. The ROC curves were further developed to evaluate the predictive performance of the crash risk prediction models under different weather conditions. The predictive performance of the crash risk model for clear weather was better than those of the crash risk models for adverse weather conditions.Impact on industryThe research results could promote a better understanding of the impacts of traffic flow characteristics on crash risk under adverse weather conditions, which will help transportation professionals to develop better crash prevention strategies in adverse weather.