Statistical approach for forecasting road surface temperature

Snow and ice make road conditions and use difficult and represent a major challenge for the winter road maintenance service. Optimizing winter maintenance service and safety thus requires accurate short‐term forecasts of the meteorological state of the roads. The most common approach to forecasting road conditions is an energy balance model based on a one‐dimensional diffusion equation. Physical models can predict the road surface temperature, which is the most important parameter for determining the road surface condition (e.g. dry, wet, ice, snow). However, such models can show a large degree of error at sites where physical processes are too complex to be simulated correctly. To solve this problem, physical models are often combined with statistical approaches. This paper proposes a purely statistical method for forecasting road surface temperature based on stepwise linear regression analysis with appropriate selection of the input parameters and separate models for different time intervals. The method is tested on data from several Slovenian road weather stations. Its accuracy is comparable to or better than that of physical models.