Stochastic Models for Daily Rainfall

Long sequences of daily rainfall are increasingly required, not only for classical hydrological purposes, but also to provide the inputs for models of landfills, tailings dams, land disposal of liquid waste, and other environmentally sensitive issues. This paper compares the performance of a range of models which have been proposed for the stochastic generation of daily rainfall data using historical data from a set of rainfall sites sampling most aspects of Australian rainfall regimes. Maximum likelihood estimators are used throughout for parameter estimation, and the Akaike information criterion has been used as a guide to parsimony in the number of parameters required. Monthly and annual models have been used for rainfall occurrence, and monthly models only for rainfall amounts. Particular aspects studied are effects of the length of the historical record on model select, and the need to group rain days according to the number of adjoining wet days. The results show that it is difficult to make substantial reductions in the number of parameters used in an existing model developed for Australian conditions.