Nonstationary spatiotemporal evolution of extreme flood and low flow affected by climate change and human activities in the Yellow River basin

The Yellow River Basin, China. Extreme weather events occur frequently under global change, and the assumption of stationary for hydrological series may no longer be valid. Therefore, we proposed a new framework based on a nonstationary statistical model that incorporates machine learning for detecting spatiotemporal variations of extreme events. The series of these events at most stations show nonstationary characteristics during both the base period and the change period. By optimizing and evaluating different types of nonstationary models based on the Generalized Additive Model for Location, Scale and Shape (GAMLSS), the model with the climate index ( CI ) and the human-induced index ( HI) as covariates demonstrates superior applicability compared to the model using the CI and the reservoir index ( RI ). Furthermore, the higher probability of extreme flood and low flow were observed at Tangnaihai, while the lower probability of extreme low flow was identified at Huaxian. Extreme flood in the YRB show weak inter-station correlations with high spatial heterogeneity, especially between Tangnaihai and Huayuankou, while extreme low flow is generally well correlated except between Lanzhou and its downstream stations (Toudaoguai and Longmen) due to water withdrawals from irrigation districts. The results provide scientific basis for reservoir flood control, river ecological health and the safety and stability of power systems. • A new framework is proposed to detect spatiotemporal extreme events in the YRB. • The human-induced index outperforms the reservoir index as nonstationary covariate. • Climate change and anthropogenic factors impact extreme flood. • Irrigation strongly affects extreme low flow.