Derivation of nonstationary rainfall intensity-duration-frequency curves considering the impacts of climate change and urbanization

Urban infrastructure traditionally relies on stationary rainfall intensity-duration-frequency (IDF) curves. However, this assumption is challenged by climate change and urbanization. Many studies tried to update IDF using time covariate which lacks physical significance. More importantly, the stationary (ST) design method is not applicable for nonstationary (NS) design where the distributions of extreme precipitation change over time. For the annual maximum precipitation (AMP) in Beijing, we utilized local factors (urbanization and temperature) and global factors (ENSO and EASM etc.) to develop NS models, with the average annual reliability method first employed to update the IDF curves. Short-duration (shorter than 6-h) AMP of most stations show upward trends, whereas the AMP with longer durations exhibits downward trends. The NS modeling reveals that the 18-h AMPs is mainly affected by global processes (ENSO and EASM). The predictive accuracy of the optimal NS model outperforms ST model by a remarkable 219% during the validation period. In addition, the ST design rainfall tends to overestimate rainfall for durations longer than 12-h. Interestingly, the gap between NS and ST design uncertainties diminishes as duration/return period expands. The above findings provide new insights about impacts of local and global physical processes on the variation of extreme rainfall. •Best combination of distribution and covariate for updating IDF curve is analyzed.•Urbanization/construction area is used to build and predict nonstationary model.•Split-sample test is used to validate the superiority of nonstationary model.•Gap between NS and ST design uncertainty shrinks as duration/return period expands.•Average annual reliability is first used to derive the nonstationary rainfall IDF.