Abstract: Analyzing the variation pattern of temperature observation deviation in the radiation shield of regional automatic meteorological stations can help to correct the quality of temperature data, thereby improving the usability of temperature data. Therefore, based on the hourly temperature data observed by the radiation shield of regional automatic meteorological stations and its neighboring station with thermometer screen at the Hubei Province from June 2019 to May 2022, this study first analyzes the seasonal and daily variation characteristics of the hourly temperature deviation ( T_\textbs ) between the two types of stations, and explores the influence of meteorological factors such as precipitation, relative humidity, sunlight, and wind speed on T_\textbs . Then, multiple linear regression and random forest methods were used to establish two temperature correction models for radiation shield stations respectively, and the correction effects of the two models on the temperature observation deviation at the radiation shield stations were analyzed. The results are as follows. (1) Overall, the observed temperature at the radiation shield stations during the day is higher than that of its neighboring stations with thermometer screen . The average annual number of high temperature days at the radiation shield station is 20.0 days higher than that of its neighboring stations with thermometer screen. (2) There are obvious seasonal and daily variations in T_\textbs between the two types of stations, with an overall trend of high in summer and low in winter, high during the day, and low at night and early morning. The average T_\textbs is highest at 13:00 BT on a clear day, reaching over 1.0 ℃. (3) T_\textbs will vary with changes in meteorological conditions at the station, with a greater value in the absence of precipitation and approaching 0 ℃ in the presence of precipitation. The T_\textbs has a negative correlation with relative humidity, a positive correlation with sunshine hours, and a positive correlation with wind speed at first, followed by a negative correlation as the wind speed increases to the critical value. (4) Multiple linear regression and random forest models have good correction effects on temperature deviation at radiation shield stations, reducing the average T_\textbs from 0.72 ℃ to 0.17 ℃ and 0.16 ℃, respectively. The correction effect of the random forest model is slightly better than that of the multiple linear regression model, and the random forest model has better correction effect for high temperatures exceeding 35 ℃. After correction, the total number of high temperature days at the radiation shield station has decreased by more than 55%.