Using regional wind fields to improve general circulation model forecasts of July–September Sahel rainfall

Abstract This study develops and applies a model output statistics (MOS) approach for correcting poor general circulation model (GCM) seasonal rainfall predictions over the Sahel region of West Africa. It illustrates a methodology for approaching the MOS prediction of regional rainfall, drawing on knowledge of the regional circulation system. The ECHAM4.5 GCM has very little skill in predicting July–September Sahel rainfall. However, the GCM is much more capable of reproducing the regional wind circulation at 925 hPa, especially over the tropical Atlantic. This is capitalized upon using a MOS approach that applies empirical orthogonal functions (EOFs) of the model's regional (tropical Atlantic and West Africa) 925 hPa wind as predictors in a regression with observed Sahel rainfall as the predictand. Over 1968–2002, the MOS system requires only the first wind EOF and improves the correlation skill of July–September Sahel rainfall from 0.07 (raw GCM) to 0.57. The MOS system is applied to GCM experiments using persisted sea‐surface temperature (SST) anomalies from the months of June, May and April respectively, to estimate the potential of the system to make forecasts with lead times between 0 and 2 months ahead of the July–September season. Almost identical skill is achieved from the June SST, but May and April SSTs show a substantial decline in skill. This is mainly associated with a tendency in tropical Pacific SST anomalies from May to June, highlighting development of El Niño (La Niña) in Sahel dry (wet) years, though statistically significant tendencies are also found in the tropical Atlantic and Indian Oceans. The MOS approach is robust in simulation experiments over the 1950–2002 period, when one additional EOF in the MOS system is found to best capture the tropical Atlantic influence. For increasing forecast lead time, targeted prediction of SST from April to June is motivated by these findings. Copyright © 2008 Royal Meteorological Society