Comparison of nonstationary models in analyzing bivariate flood frequency at the Three Gorges Dam

Abstract

Traditional flood frequency analysis is usually performed under the stationary assumption that flood elements, such as flood peak and flood volume, are independent and identically distributed (i.i.d). However, with global climate change and human activities across the landscape, stationary approaches cannot capture the nonstationary characteristics of flood events. This study aims to explore the key elements in establishing an effective bivariate nonstationary flood frequency model for flood peak and volume series at the Three Gorges Dam in China. A stationary reference model and six nonstationary models were established and rigorously compared to evaluate their benefits and limitations, including two time-informed models and four climate-informed models that consider time and climate indices as explanatory variables. The results suggest that nonstationary models are clearly superior to stationary models with respect to model performance based on the deviance information criterion (DIC). The time-informed nonstationary model provides a potential tool to project future trends in flood return periods with minimal data availability, yet the uncertainty in the estimation of design values of such models is limited. On the other hand, explicitly incorporating climate indices as explanatory variables of flood frequency distribution can significantly improve model performance and reduce uncertainty but at the cost of increased model complexity.