Abstract: In order to detect the deeply hidden precipitation quality anomaly data, the non-real-time precipitation quality anomaly events are defined, and a non-real-time quality control method for hourly precipitation data from automatic weather stations is proposed by combining Madsen-Allerupt and percentile method. The hourly precipitation data from 2841 regional automatic weather stations in Hubei Province during 2005 to 2022 are further quality-controlled by the proposed method and the hourly precipitation quality-controlled dataset is produced. The study results show that: (1) The designed method can effectively classify and detect the deeply hidden quality anomaly data. The continuous low precipitation events account for 87.9%, precipitation hours anomaly events account for 8.7% and continuous high precipitation events account for 3.4%. (2) The total amount of quality anomaly data tends to decrease year by year, and there is no obvious differences in the spatial distribution. The total amount of quality anomaly events from April to October is lower than that in other months with the highest proportion of 36.7% in winter and the lowest of 13.1% in summer. (3) The quality of precipitation data in Hubei Province is generally poor at the beginning of the construction of the station. Since 2008, it has been steadily improved. The average annual data availability ratio from 2008 to 2022 is 95.6%, and the average data availability ratio from April to October of the year is 97.8%, which is higher than 92.6% in other months.