A nonparametric model for stochastic generation of daily rainfall amounts

This paper presents a model for stochastic generation of rainfall amounts on wet days that is nonparametric, accommodates seasonality, and reproduces important distributional and dependence properties of observed rainfall. The model uses kernel density estimation techniques, which minimize the assumptions that are made about the underlying probability density, and representation of seasonal variations is achieved through the use of a moving window approach. Four different classes of rainfall amount are considered and categorized according to the number of adjacent wet days, and the model is conditioned on the rainfall amount on the previous day. The proposed model can emulate the day-to-day features that exist in the historical rainfall record, including the lag 1 correlation structure of rainfall amounts. We link this model with long sequences of rainfall occurrence generated by the model of Harrold et al. [2003], which is designed to reproduce the longer-term variability of the observed record. The approach is applied to daily rainfall from Sydney, Australia, and the performance of the approach is demonstrated by presentation of model results at daily, seasonal, annual, and interannual timescales.