A Bayesian hierarchical regional flood model

Recent studies have shown that flood data from eastern Australian catchments may demonstrate variability in flood risk over multidecadal time scales, characterised by crossings of the Interdecadal Pacific Oscillation (IPO) climate index. This nonhomogeneity of flood risk may lead to a significant prospect of biased long-run flood risk from at-site flood data with insufficient coverage of both IPO epochs. This paper develops a Bayesian hierarchical regional model, implemented using the Gibbs sampler, to overcome this possible bias in flood risk. The hierarchical model proposes that the parameters of the flood frequency distribution at any site are random samples from a regional probability model, allowing for intersite variability, while also permitting spatial correlation between concurrent floods. An outcome is that the predictive uncertainty at an ungauged or gauged site may be quantified.