Abstract:In this paper we made a statistical analysis of the road surface temperature based on observations of the selected five road stations (A1027, A1325, A1412, A1414, A1512) and the meteorological elements output from the Beijing Rapid Update Cycle (BJ RUC) numerical forecasting model with 3 km resolution from 1 November 2012 to 30 March 2013. We used the stepwise regression model methods to build three types of statistical models for hourly road surface temperature in 24 h in the winter half year for the different initial forecasting times (08:00,14:00, 05:00 BT) and the different months. Then the best type is used to forecast the road surface temperature from 1 November 2013 to 30 March 2014. The results are as follows. The road surface temperature is significantly correlated to T2 and the short wave radiation, but secondarily correlated to the long wave radiation and humidity output from RUC. Compared to the type of statistical model with the only one factor for the previous day, the type of regression model with meteorological elements of remarkable correlation inserted performs better in terms of the road surface temperature forecast accuracy by more than 25%, and the prediction error decreases by 1℃. For further enhancing the forecast accuracy rate, we selected the different initial times for verification so as to control error within ±3℃. The result of evaluation shows that the forecast value of the road surface temperature in the daytime is better than that over night, and sunny days are better than any other kinds of weather.