A Bayesian hierarchical nonhomogeneous regional flood model for New South Wales, Australia

Long-term climate variability has been shown to affect flood frequency in New South Wales, Australia (Kiem et al., 2003). This study extends this work by developing a-Bayesian hierarchical model for predicting floods at ungauged locations. Flood records were stratified using a threshold value of the Interdecadal Pacific Oscillation (IPO), which characterises the low frequency component of sea surface temperature anomalies in the Southern Pacific Ocean. It is shown that the stratified flood distributions differ significantly with ratios of flood magnitudes being approximately 1.6 times greater during the IPO negative years. Moreover, the evidence suggests that the log-normal distribution fits the stratified flood records satisfactorily. The stratified flood records were then subjected to regional flood analysis using a Bayesian hierarchical approach. The regional model considers at-site floods to be spatially correlated via an intersite distance correlation function. The hierarchical model proposes that the parameters of the flood frequency distribution for any site are random samples from a regional probability model. This allows the inclusion of catchment characteristics, while also explicitly allowing for intersite variability. Model calibration was performed using Markov chain Monte Carlo methods. An important outcome is the quantification of predictive uncertainty at an ungauged catchment.