Stochastic Generation of Annual Rainfall Data.

Cooperative Research Centre for Catchment Hydrology, Technical Report 02/6.

This report describes the generation of annual rainfall data with and without incorporating parameter uncertainty. A first order autoregressive [AR(1)] model is generally used to generate annual rainfall data and recently the use of this model has been criticised because of its inability to model explicitly the wet and dry rainfall years observed in the observed data. Thyer and Kuczera (2000) developed a hidden state Markov (HSM) model which explicitly models the wet and dry rainfall years. The HSM model parameters were estimated using the Markov Chain Monte Carlo method and this results in posterior probability distributions for the model parameters. Both the AR(1) and HSM models were applied to 44 rainfall stations located in various parts of Australia and their performance evaluated. When the models were applied without parameter uncertainty, only the HSM model performed satisfactorily for all the stations. However, when parameter uncertainty was incorporated, both models performed equally well and it was difficult to separate one from the other in terms of model performance.