Stochastic implications for long-range rainfall predictions

Rainfall prediction for a year in advance would be immensely valuable for numerous activities, if it were achievable. It is shown that in any one year, the chances of making a correct prediction is about 50%, but there is no way a priori of determining the correctness of such a prediction. This results primarily because annual mean time series of rainfall over most of the globe consists of white noise, i.e. they are random/stochastic. This outcome is shown to exist for both observations and output from a coupled global climatic model, based on autoregressive analysis. The major forcing mechanism for rainfall anomalies over much of the global is the El Nino/Southern Oscillation, but it explains only a modest part of the variance in the rainfall. Much of the remaining variance is attributed to internal climatic variability, and it is shown that this imposes a major limitation on rainfall predictability.