Abstract:Based on hourly rainfall observational data of meteorological stations during 1960 to 2014, a non-stationary frequency analysis of the Annual Maxima (AM) sub-daily rainfall series (1, 2, 3, 6, 12 and 24 h rainfall, using a moving window approach) for North China was conducted, then the difference between stationary and non-stationary Intensity-Duration-Frequency (IDF) curves using Bayesian inference was estimated. Results show that the trends of 1, 2, 3, 6, 12 and 24 h annual maxima rainfall in North China from 1960 to 2014 are complex. While the shorter the duration, the more stations that show an upward trend; and for over 6 h extreme precipitation, the more stations that show a downward trend. For a 20 to 100 years 1 h extreme precipitation, the difference between the non-stationarity and stationarity extreme precipitation is larger, and at the station with the upward trend that a stationary climate assumption may lead to underestimation of extreme precipitation about 30%-43%; the average difference of 6 h to 24 h is relatively small, and the difference at station with downward trend is about -43%- -32%. The difference between the non-stationarity and stationarity extreme precipitation decreases with the extension of the duration, and the uncertainty increases as the return period increases in all conditions. Shorter precipitation events have been intensified more in the past decades in North China, the shorter the duration the larger the differences between the nonstationary and stationary extremes. Therefore, this non-stationary extreme value analysis can potentially reduce the risk of extreme precipitation.