A low-cost approach to identify hazard curvature for local road networks using open-source data

Vehicle crashes are a leading cause of death in the United States. Curvature in local roadways has been identified as one of the most significant factors that lead to fatal crashes. Given the large number of local roads and their relatively low traffic volume - compared with interstates or freeways - most local roads may not receive priorities in the first phase of highway upgrades, and critical locations, e.g., sharp curves (vertical and/or horizontal), in the network may be a deadly threat for both advanced autonomous vehicles and conventional vehicles. Furthermore, identifying local roadway curvatures presents various obstacles, such as high budgets and lack of survey data. To fill this gap, this study offers a low-cost approach to constructing three-dimensional geometric profiles for local roads in a relatively large study area using open-source data. Given these profiles, critical road segments, including extreme horizontal and vertical curves and their combinations, can be identified. This study re-classifies the local road segments into 20 sub-categories based on the calculated vertical grades and curve radii and incorporates those segments into a zero-inflated native binomial model for crash occurrence. Model results showed that grades or curves were associated with decreased crash frequency compared with straight and flat roads. However, segments with larger horizontal curve radii and low grades were found to be associated with increased crash frequency. Further implications are discussed in the paper