Spatial and temporal variability of rainfall related to a third-order Markov model

Many types of analysis require rainfall data, which must often be generated to overcome limitations in the historical record. The ability to model outlying rainfall years satisfactorily is particularly important in risk studies. We describe a rainfall generator with this ability, based on a third-order Markov process, and we show how the fitted parameters of the model vary for an illustrative sample of 18 sites with highly contrasting climates. Some of the parameters of the model show patterns that are characteristic of the climate type. The model should thus be suitable for interpolating rainfall data where they do not exist. We also fitted the model to subsets of some long data records to illustrate the marked changes in rainfall characteristics that have occurred in some locations over the last 70 years, and to help understand the nature of these changes. (C) 1997 Elsevier Science B.V.