Degree of heterogeneity versus prediction error in Regional Flood Frequency Analysis : a case study for Victoria, Australia

![CDATA[In flood management and hydraulic infrastructure design, flood risk assessment is needed. To estimate flood quantiles accurately at an ungauged catchment Regional Flood Frequency Analysis (RFFA) is widely adopted. In RFFA, the homogeneity of a region refers to the state of similar flood responses, which is mostly the reflection of similar flood and catchment characteristics. This study examines the homogeneity of 113 gauged catchments in Victoria, Australia. The selected catchments are divided into two groups by drainage division and then subdivided each of them into two sub-regions. Hosking and Wallis (HW) test statistics (H) are applied, and few sites are detected as discordant. H1-statistics are relatively low (ranging from 3.6 to 20.2) in the sub-groups but highest (26.6) in Victoria as a single region, which indicates that these regions were highly heterogeneous. A log-log model is used to develop prediction equations using ordinary least squares regression (OLS). To check the relative accuracy of the developed RFFA models a leave-one-out (LOO) is adopted. It is found that the degree of heterogeneity does not have any direct effect on the accuracy of design flood estimates. More investigation is needed to better understand the association between the degree of regional heterogeneity and model accuracy in RFFA.]]