Seasonal to interannual rainfall probabilistic forecasts for improved water supply management: Part 3 — A nonparametric probabilistic forecast model

This paper is the last in a series of three in the current issue that present a framework for long-term rainfall probabilistic forecasts. A nonparametric probabilistic forecast model is presented. The approach is based on accurate estimation of the conditional probability distribution of rainfall through the use of nonparametric kernel density estimation techniques. The kernel approach is data driven and avoids prior assumptions as to the form of dependence (e.g. linear or nonlinear) or of the probability density function (e.g. Gaussian). Quarterly rainfall at the Warragamba dam near Sydney, Australia, is used to illustrate the applicability of the nonparametric probabilistic forecast approach. Optimal predictor sets identified as discussed in the earlier two papers of this three-paper series are used to formulate the probabilistic forecast model. Two separate predictor sets, one based on commonly used El Niño Southern Oscillation indices, and the other based on spatially distributed sea surface temperature anomalies, are considered. Forecasts are made for lead times up to two years for selected years of the Warragamba dam rainfall record. Results indicate that use of the spatially distributed sea surface temperature anomalies leads to a reduction in the variability associated with the seasonal to interannual rainfall forecasts.