Regionalisation of Generalised Extreme Value Distribution in Regional Flood Frequency Analysis: A Case Study for South-East Australia

This study presents regionalisation of the Generalised Extreme Value (GEV) distribution by adopting L moments and Bayesian Generalised Least Squares (BGLS) regression. This uses data from 88 gauged catchments in New South Wales, Australia. The regional GEV distribution is compared using the Parameter Regression Technique (PRT) and Quantile Regression Technique (QRT) under fixed-region and Region-of-Influence (ROI) approaches. Results indicate improved regional flood estimation, capturing spatial variability and reducing uncertainty in flood quantile estimation through the ROI approach compared to fixed regions. The mean flood statistic shows a higher spatial heterogeneity in the fixed region approach, while the ROI approach more effectively reduces uncertainty and enhances predictive performance. For fixed regions, PRT achieves a median relative error (Rer%) of 34–40% and a relative root mean square error (RRMSE%) of 59–65%, compared with QRT’s Rer% of 40–63% and RRMSE of 59–94%. In the ROI framework, both techniques yield similar Rer% (31–40%), though QRT-ROI exhibits slightly reduced RRMSE.