Seasonal forecasting of the Ethiopian summer rains

We present a new seasonal forecasting model for the June-September rains in Ethiopia. It has previously been found that the total June-September rainfall over the whole country is difficult to predict using statistical methods. A detailed study of all available data shows the rainfall seasonality varies greatly from one region to another, which would explain why the total June-September rainfall over all regions is a difficult property to forecast. In addition, the correlation between rainfall and the southern oscillation index varies spatially, with a strong teleconnection present only in some regions. This study accounts for the spatial variability in rainfall by grouping the rain gauge stations into four geographical clusters based on seasonality and cross-correlation of rainfall anomalies. Linear regression equations are then developed separately for each cluster. The variables we use for the regressions are sea-surface temperature anomalies in the preceding March, April and May of the tropical western Indian Ocean, the tropical eastern Indian Ocean, and Nino3.4. Formal skill testing of the equations shows that the new forecasting scheme is more effective in central western Ethiopia than either climatology or persistence - the methods currently used by the Ethiopian National Meteorological Services Agency. Copyright (C) 2004 Royal Meteorological Society.