Investigating the Safety Effectiveness of Wildlife–Vehicle Crash Countermeasures using a Bayesian Approach with a Comparison between Carcass Removal Data and Traditional Crash Data

Abstract

Wildlife–vehicle crashes (WVCs) pose a significant threat to not only wildlife populations but also highway safety. The most expensive WVC countermeasures include crossing structures with fencing, whereas the least expensive countermeasure is the wildlife warning signs. This study is aimed at estimating the crash modification factors (CMFs) for these two countermeasures using cross-sectional analysis. Two types of WVC data are used in this study: carcass removal data and traditional crash data. A random-intercept Bayesian approach was utilized to incorporate the contributing factors representing traffic volume, roadway geometry, weather conditions, and unobserved heterogeneity resulting from between-site variance. The No-U-Turn Hamiltonian Monte Carlo sampling technique was employed because of its high efficiency in handling complex models. The results suggest that the treatment of implementing wildlife warning signs on hotspots of high WVCs has been ineffective, perhaps because warning signs are passive devices, and the high uncertainty of wildlife interaction still remains. The crossing structures are found to be effective, with an estimated CMF of 0.66 and 0.55 using the carcass data and crash data, respectively. Recommendations could be made to implement more active information dissemination via wildlife-actuated warning signs, where crossing structures may not be feasible. The findings from this study indicate that the carcass removal data are more comprehensive than the crash data, despite the underreporting issue existing in both datasets. Therefore, a unique identifier should be added in both datasets to enable merging the data and obtain more complete results from the analyses.