This article discusses a combined model to perform climate forecast in a seasonal scale. In it, forecasts of specific stochastic models are aggregated to obtain the best forecasts in time. Stochastic models are used in the auto regressive integrated moving average, exponential smoothing and the analysis of forecasts by canonical correlation. The quality control of the forecast is based on the residual analysis and the evaluation of the percentage of reduction of the unexplained variance of the combined model with respect to the individual ones. Examples of application of those concepts to models developed at the Brazilian National Institute of Meteorology (INMET) show good results and illustrate that the forecast of the combined model exceeds in most cases each component model, when compared to observed data.
Seasonal forecasts; stochastic models; canonical correlations; cluster model
