Risk based tunnel design by vehicle involved in road crashes. Models and tunnel length

Between 2018 and 2020, in the approximately 2,600 km of Italian road tunnels, 2,899 people were injured and 60 died in 1,885 road accidents. The accident frequency was lower than on open roads, while the injury/fatality rate was higher. Using the recursive partitioning and regression trees method (rpart), we developed two accident models useful for predicting the probability of involvement of "vehicle type" in short and long tunnels. Variables such as the type of accident, the circumstances, the type of road, the carriageway, the time of the accident, the journey purpose (whether work-related or not), and the length of the tunnel defined the nodes and paths of the regression tree associated with a vehicle type involved. The “road type” was the best predictors for short tunnels while the “journey purpose” was the best predictor for long tunnels. The most important result of the study refers to the similarity between the probability of an accident in short and long tunnels for a specific segment of road users: commuting and non-commuting car drivers and drivers of heavy goods vehicles on-duty. The study showed that this road user segment in short tunnels has an accident probability half that observed in long tunnels.