Rainfall Generator for Nonstationary Extreme Rainfall Condition

Stochastic weather generators are generally used to produce scenarios of climate variability on a daily timescale for hydrological modeling and water resource planning applications. Most of the available weather generators assume extreme rainfall series as stationary series. However, it is currently perceived that global climate change is increasing the intensity and frequency of extreme rainfall events and creating a nonstationary component in extreme rainfall time series. Consequently, the realistic modeling of rainfall extremes in a nonstationary context is indispensable. In this study, we propose a modified version of a k-nearest neighbor (KNN) weather generator that incorporates nonstationarity in the extreme rainfall series. The proposed algorithm first models the nonlinear trend in the extreme rainfall series that exceeds the defined threshold u and perturbs the original-KNN-simulated extreme rainfall using the knowledge available in the nonstationary model. The proposed algorithm is demonstrated with three case studies, and the performance of the proposed algorithm is validated using various extreme precipitation indices. The results of the three case studies indicate that extreme rainfall characteristics are consistently well simulated with the proposed algorithm. Particularly, based on the results of the three case studies, the proposed algorithm decreases the root-mean-square error (RMSE) in rainfall simulation with respect to the original KNN algorithm by at least 40%. (c) 2019 American Society of Civil Engineers.