Regional flood estimation for very frequent floods based on peaks-over-threshold approach : a case study for south-east Australia

Regional flood frequency analysis (RFFA) is widely applied to estimate flood quantiles at ungauged locations. Multiple linear regression analysis is one the most applied techniques to develop regional prediction equations in RFFA, which establish a relationship between the dependent variable (flood quantile) and independent variables (climatic and catchment characteristics). Traditionally, RFFA uses at-site flood estimates from annual maximum (AM) flood series as the dependent variable. For estimating very frequent floods, AM modelling approach is not suitable. As an alternative, peaks-over-threshold (POT) approach can overcome some of the drawbacks associated with AM model in the very frequent ranges. However, the application of using POT for very frequent flood estimation (e.g., 12EY, 6EY, 4EY, 3EY, 2EY, 1EY, 0.5EY and 0.2EY) is limited in Australia. This study used 103 gauged catchments in the state of Victoria to develop a POT-based RFFA technique. Very frequent rainfall intensity with varied exceedances per year and baseflow index (peak and volume) data were used as additional independent variables. This study adopts a range of statistical measures and three different validation techniques to evaluate model performance, including k-folds, repeated cross and leave-one-out validation. It is found that accurate regional prediction equations can be developed based on the POT approach for very frequent floods.