Detection of change point in annual maximum flood series over eastern Australia using a Bayesian approach

Inference about the existence and characteristics of changes in the mean and variance of streamflow sequences that may violate the stationary assumption is an important step in many water resources planning and management tasks. These abrupt changes in the flood peaks can be caused by climatic as well as other anthropogenic effects such as changes in land use ad land cover and construction of dams. This study seeks to investigate the existence of change point in flood peaks time series data in eastern Australia using a Bayesian approach, based on a single shifting model to investigate a change in the mean level for the catchments of a very long record length. Two univariate normal models are considered: the no change hypothesis and a single change in the mean level. A Markov Chain Monte Carlo algorithm with Gibbs sampler is used to undertake inference analysis of the posterior probabilities for each of the hypotheses. This uses annual maximum flood data from fifteen catchments in NSW and Queensland states with record lengths of over 75 years to specify the informative prior distributions. The results show change points in the late forties and mid seventies in the annual maximum flood time series data for the selected catchments.