Based on the real-time hourly temperature data of global surface meteorological observation stations and non-meteorological observation stations of the National Meteorological Information Center, a real-time updating quality detection process is designed by applying the monthly threshold, block threshold, zonal threshold and spatial consistency comprehensive detection methods. The global hourly surface temperature data set from 2016 to 2020 is established to provide reliable data support for global forecasting services and scientific researches. The detection results of data in the recent five years show that suspicious and wrong data accounted for 0.85% of the total global data. More suspicious and wrong data in Asia accounted for 1.67% of the data samples in Asia, about 72% of the global suspicious error data, and more data are detected in summer and autumn. About 79% of the suspicious data of nonmeteorological observation stations are detected by the block threshold and spatial consistency method. More than 9% of the suspicious data of observation stations are detected by the station extreme value and spatial consistency method.