Night icing prediction using vehicle-mounted pavement temperature sensor and atmospheric data

Abstract

Faced with high accident rates on ice-covered roads at night, notifying road managers of the most dangerous locations would allow them the opportunity to treat the icy roads by applying de-icing chemicals. In addition, drivers can be better prepared for the impending danger. In this regard, this study developed a web-based night icing potential information system and a deep learning-based night icing forecasting model. Input data included pavement temperature, collected by patrol cars operating between 23:00 and 07:00 on a daily basis in the winter season, and atmospheric data provided by the Korea Meteorological Administration. For evaluating the forecasting model, baseline data were generated based on the physical principle that ice is formed when pavement temperature is negative and lower than the dew point temperature. The forecasting model was assessed using pavement temperature and atmospheric data obtained on a 40 km stretch of a rural highway, passing through a mountainous region in South Korea. As a result, the model showed favorable performances, with 94% and 90% accuracies for bridge and roadway segments, respectively. Considering the increasing emphasis on preventive maintenance, the developed forecasting model can be applied to a preventive anti-icing measure, in turn, increasing traffic safety on winter roads.

Keywords

• Black ice
• Prediction
• Pavement temperature
• Deep learning