Abstract: Based on hourly 2m air temperature product of SWC-WINGS model with a resolution of 0.01° × 0.01°, a new 1km grid air temperature forecast with hourly rolling update was obtained by using time-lagged ensemble and real-time bias correction. The hourly air temperature forecast from July to August 2022 was verified by prediction accuracy, average error, and average absolute error. The results showed the accuracy of time-lagged ensemble forecast was higher than that of the up-to-date forecast, and the real-time bias correction leaded to obvious improvement with 17.3% in 1~6 h forecasts. The SWC-WINGS model had obvious systematic bias in the prediction of high and low temperature in Sichuan, and the time-lagged ensemble reduced the deviation weakly. The real-time bias correction could control the average absolute error of low and high temperature forecast in most areas of Sichuan within 1 ℃ and 2 ℃, respectively. For the high temperature weather in Sichuan on 13 August 2022, the integration forecast of time-lagged ensemble and real-time bias correction had a high positive skill compared with SWC-WINGS model.