ABSTRACT
The study aimed to model the effect of spatial relationship between adjacent curves on the severity of curve-based crashes along with driver and crash causal characteristics, reflecting driver's short-term expectancy. The crash and other associated data was retrieved from the web-based Road Accident Data Management System available in Himachal Pradesh, India, and curvature attributes were extracted using GIS. Overall, the study included 1113 curve based crashes. The driver's perception of the sharpness of a curve was quantified as a single representative categorical factor based simultaneously on its length and radius using K-Medoid Clustering. Separate crash severity models catering to the two possible approach directions of the subject curve were developed reflecting its independent interaction with its corresponding adjacent curve in each direction. Partial Proportional Odds models were developed to overcome the predictive limitations of Ordinal and Multinomial logit models. Indicators of spatial relationship and the intensity of sharpness of the subject curve were found to be statistically significant. A sharp approach curve (radius:40-60 m) increased the risk of fatality by 2.16 times with a similar increase (2.5 times) observed for a short (length:30-60 m) adjacent curve. Adjacent curves turning in the same direction were 2.34 times more prone to fatalities. A very sharp subject curve with radius ≤ 40 m increased the risk of fatal crashes by 2.5 times, as did the short subject curves (30-60 m) (at least 3 times). Subject curves characterized by a short length and a very sharp curvature contributed relatively 3-4 times more to fatal crashes. The identified risk factors and their impact can help the relevant stakeholders to take appropriate actions and can further assist them in identifying high risk scenarios.