Predictions of African rainfall on the seasonal timescale

Real-time predictions of the rainfall in July-September 1997 and January-Ma rch 1998, respectively, for the Sahel and southern Africa at I month lead a re presented. The forecast technique is based on canonical correlation anal ysis (CCA), a multivariate linear statistical model that relates patterns i n the predict and held to patterns in the predictor field. In this CCA desi gn, four consecutive 3 month predictor periods are followed by a lead time and a single 3 month predictand period. Cross validation is used to estimat e the skill in the forecasts. The forecasts are expressed in terms of depar tures from climatological probabilities of three equiprobable categories fo r above-, near-, and below-normal rainfall. Expressing the forecasts in ter ms of probability anomalies communicates the level of confidence in the for ecasts. Forecast verifications have been performed using the Heidke skill ( S) score. For the Sahel the July-September 1997 rainfall prediction at 1 mo nth lead showed reasonable skill. Furthermore, it is shown that the Sahel r ainfall in July-September is influenced at least partly by Fl Nine-Southern Oscillation (ENSO). In addition, Sahel rainfall is affected by the sea sur face temperature pattern in the Atlantic Ocean and perhaps the Indian Ocean . For southern Africa, the S score was +30, suggesting a moderately good fo recast for the region as a whole. The results also indicate that ENSO plays an important role in modulating rainfall variations in southern Africa.