This paper quantifies the impacts of incident, railroad, environment. and train/car characteristics on the probability of hazardous material (hazmat) release in a hazmat-carrying train incident and provides a prediction tool for hazmat release. Two sets of models utilized the Federal Railroad Administration 2012–2016 rail equipment incident dataset. The units of analyses for these two sets were trains and hazmat cars. Binary logit and binary mixed logit models were investigated to account for hazmat release and potential single-level and two-level grouping in the data (owing to possible hazmat release interdependence among hazmat cars belonging to a train and trains belonging to an incident). Development of receiver operating characteristics curves improved the prediction performance of the models by defining an appropriate cutoff point. Results showed that derailment increased hazmat release probability more than other incident types. Incidents resulting from signal and communication causes were most likely to result in hazmat release. Higher proportion of damaged/derailed hazmat cars and proportion of hazmat cars in a train, track classes 2 and 3, higher train speed, and train gross tonnage were the other important factors. Results of mixed models showed hazmat release from cars belonging to a train were interdependent and hazmat release from trains belonging to an incident were independent. Although models at both levels led to useful results, car-level models had better prediction performance.
ASJC Scopus subject areas
- Civil and Structural Engineering
- Mechanical Engineering