Author(s): K. Haddad; A. Rahman; G. Kuczera; T. Micevski
Linked Author(s):
Keywords: Floods; Regional flood frequency analysis; Australian Rainfall and Runoff; Parameter regression; Quantile regression; Region-of-influence
Abstract: This paper compares the region of influence (ROI) and fixed region approaches for regional flood frequency analysis for both the quantile regression technique and parameter regression techniques using a Bayesian Generalised Least Squares (BGLS) regression. It uses data from 96 catchments in New South Wales. The BGLS is used to regionalise selected quantiles (QRT) and the first three moments of the log-Pearson type 3 distribution (PRT). The BGLS regression accounts for inter-station correlation and importantly distinguishes between sampling error and model error. To assess the performance of the models, one-at-a-time cross validation is used. It has been found that the ROI deals with regional heterogeneity in a better way than the fixed region approach. The developed models for both the fixed and ROI approaches satisfy the underlying model assumptions quite well, however the ROI approach provides more accurate flood quantile estimates with less uncertainty than the fixed region. It has also been found that the BGLS-QRT-ROI and BGLSPRT-ROI perform very similarly for the study data set in New South Wales.
Year: 2011