Development of rainfall intensity-duration-frequency curves under nonstationary conditions

Abstract

Intensity-duration-frequency (IDF) curves, developed conventionally with the assumption of stationarity, for maximum rainfall time series have been used in the design of hydraulic structures for many years. However, recent studies have shown that climate change may violate the stationarity assumption on hydrometeorological time series, thus hindering the reliable use of probability distribution functions and their parameters in practice. Therefore, nonstationary models considering changes in extremes due to changes in climate have gained importance. This study focuses on the frequency analysis of 14 standard duration annual maximum rainfall time series recorded at 20 meteorological stations across the Black Sea Region in northern T & uuml;rkiye. Intensity-duration-frequency (IDF) curves were developed under both stationary and nonstationary conditions. Generalized Extreme Value (GEV) models were constructed for all stations assuming stationarity, while nonstationary models were developed only for stations with a trend component. Time and climate oscillations, specifically the Arctic Oscillation (AO) and the North Atlantic Oscillation (NAO), were chosen as covariates in nonstationary models based on their significant correlation with annual maximum rainfall. Additionally, this study investigated the 1-, 2-, and 3-month delayed effects of remote teleconnection patterns, which serve as covariates to derive nonstationary IDF curves for stations with temporal trends. The Deviance Information Criterion (DIC) was employed to determine the best-fit GEV model, and parameters were estimated using the Bayesian Markov Chain Monte Carlo (MCMC) simulation technique. The results indicated that stationary models were the best-fit for 10 stations, while nonstationary models were optimal for 10 stations, with the latter often incorporating time or climatic oscillations such as the NAO or AO. Notably, a significant finding was the increase in return levels with the nonstationary model incorporating a time covariate at the Rize station, ranging from 26.4 to 21.5% for durations of 5 min and 24 h, respectively. Furthermore, some stations showed a significant correlation between the maximum precipitation and the AO/NAO indices, either in the current month or with delays. The study emphasizes the importance of considering nonstationarity when developing IDF curves for annual maxima of daily and subdaily rainfall series.