Based on the surface energy balance method and the soil temperature calculation module of the Community Land Model (CLM), a numerical forecast model for road surface temperature on expressway is developed. The hourly observation data of 8 conventional meteorological stations and 6 traffic automatic weather monitoring stations from 1 June 2014 to 31 May 2015 are used to evaluate and test the model forecast products. The results show that the prediction skill increases with the decrease of forecast lead time. When the forecast lead time is 1 and 6 hours, the average absolute errors of the conventional stations (traffic stations) are 1.64 ℃ (1.82 ℃) and 3.27 ℃ (3.69 ℃), respectively. The sensitivity analysis of the model to the meteorological elements indicates that the model is most sensitive to 2m temperature, followed by relative humidity and total cloud amount, and the sensitivity increases with the increase of forecast lead time. Combined with the forecasting products of the Zhejiang WRFADAS Rapid Refresh System, the Zhejiang Expressway Surface Temperature Forecasting System is established to provide the refined forecasting products of hourly highway surface temperature in the future 12 hours with an update rate of 12 hours for the whole province. The evaluation results show that the average absolute errors of conventional stations (traffic stations) are 2.81 ℃ (3.23 ℃) and 2.50 ℃ (2.93 ℃), respectively, when the forecast lead time are 1 and 11 hours. The system has high forecasting skills for extreme high (low) temperature forecasting.