Regional non-stationary frequency analysis based on extreme hydrological regionalization and hierarchical Bayesian framework

Key Points

• This research aims to develop a robust method for non-stationary hydrological frequency analysis by incorporating spatial correlations among site sequences.
• Utilized PAM clustering algorithm based on the F-madogram to classify hydrological regions.
• Constructed hierarchical Bayesian models including non-pooled, partially pooled, and fully pooled approaches for frequency analysis.
• Conducted a case study using annual maximum 24-hour extreme rainfall data from the Xiangjiang River Basin.
• Divided 36 rain gauge sites in the Xiangjiang River Basin into 3 hydrological regions using spatial clustering.
• Significant positive correlation noted between climate driving factors and rainfall, particularly in Hydrological Region I with FMA Nino12.
• Partially pooled model showed optimal performance, reducing uncertainty of regional parameters by 10% to 35% compared to stationary models.