Using the forecast data of regional model in South China (CMA-GD) and the 2m temperature observation data of 97 national stations in Hunan from August 2019 to July 2020, the hourly 2m temperature forecast of CMA-GD was evaluated and corrected in Hunan by using the stepwise regression method. The results show that the forecast bias of the 2m temperature of CMA-GD has obvious diurnal variation, because the forecast accuracy decreases in the day and increases in the night. The deviation increases with the extension of the forecast time. The forecast deviation is larger in summer. The forecast in the western Hunan is better than in the eastern Hunan being affected by the terrain and the model deviation. It`s better the 2m temperature forecast from 00:00 than that of from 12:00(UTC). On the basis of the 2m temperature forecast of CMA-GD, a stepwise regression correction model is established by time, regardless of time period, and season. The correction forecast has obvious positive skills compared with the model forecast, because of being reduced of the error and significantly improved of the accuracy.