The road surface conditions are important factors in the driving strategy of road users. During the winter, these conditions depend on the weather, the road characteristics, the measures taken to maintain the flow of traffic and of course on the traffic itself. Road authorities have at their disposal forecast tools and decision making softwares. The forecast tools and methods of protective intervention could be improved. They need to be analysed such that their performances can be improved. For this purpose, a research program on the thermal behaviour of road surface during winter, called Winter Road Maintenance, has been undertaken by the Laboratoire central des Ponts et Chaussees - France in co-operation with Meteo-France. In this paper, we design a model to predict the short term road surface temperature (3 hours) taking into account the thermal and physical characteristics of the road as well as the weather conditions. This model, COGEL, combines the program GEL1D, a model for soil and road behaviour, and the program COBEL, a model for the atmosphere. The first program, GEL1D, allows the study of the evolution of the temperature field in a multi-layered, one-dimensional structure. The second program, COBEL, predicts fog formation and computes the distribution of atmospherical conditions : temperature, humidity, in the nocturnal boundary layer between the road surface and an altitude of approximately 1500 metres. The combined method in COGEL has the advantage of entirely preserving the structures of the programs GEL1D and COBEL. The interactions between the programs result from energy exchanges. In order to validate COGEL, in situ measurements have been recorded on two roads: RN 225 at Steenvorde (with traffic) and RN 455 at La Sentinelle (without traffic). Comparisons between the field measurements and the predictive model COGEL are discussed in the paper: they show a good agreement between simulations and actual evolutions.