Abstract
We evaluate the skill of predictions of the East African short‐rains (October–December) season, of up to 5 months lead, based on a two‐stage dynamical/statistical hybrid approach. The statistical component is a well‐established, zero‐month lead, method based on observed ocean‐scale patterns of July–September sea surface temperature. A strong advantage of using this method is that its skill is high and also very robustly measured – a correlation of 0.77 between forecast and observed area‐mean rainfall is achieved over the 26 years of real‐time forecasts. Here, we extend the lead time by driving the statistical component with predicted July–September sea surface temperatures available from a number of dynamical model seasonal prediction systems. There is correlation skill near 0.5 for models initialized in May and positive skill from forecasts initialized as early as the previous October.