Road Surface temperature forecast. The statistical insight

Thermal mapping has been implemented since the late eighties to determine road ice susceptibility. It consists in measuring road pavement temperature along with some other atmospheric parameters to build a winter risk. Measurements are done using a vehicle embedded into the traffic in given road weather conditions. If the dew point temperature is lower than road surface temperature there is a risk of ice occurrence, and therefore a loss of grip for circulating vehicles. Road surface temperature is obtained with an infrared radiometer on board of a dedicated vehicle. Thermal mapping is usually done before dawn during wintertime in specific weather conditions, when the energy accumulated by the road during daytime is mainly dissipated and before the road structure starts a new thermal cycle. Measurements are anyhow time-consuming. A whole road network can hardly be analysed at once. So far, thermal fingerprints were either used directly to identify the susceptibility to ice occurrence of routes, or as an input to forecast road surface temperature with physical numerical models. These models are based on an energy balance description, and needs an accurate knowledge of both meteorological and road parameters. The objective of this work is to present how an approach based on multivariate data analysis was implemented to thermal mapping. The objective was to gather a given number of thermal fingerprints, with road surface temperature and air temperature at least, and in weather conditions representative of winter. Then data was processed through partial least-squared regression to establish a statistical forecast model for road surface temperature. Results obtained on a whole route provided a forecast with an error similar to the one of physical numerical model. Its application to specific zones could ease the forecast, reduce computation time, and provide an increasing role to thermal mapping.