The first part of this study addresses the predictability of seasonal rainfall totals in the Sahel region of West Africa. We show that, when driven with observed sea surface temperature (SST), atmospheric General Circulation Models (GCMs) do usually contain information about the observed interannual Sahel rainfall variability, through their predictions of the low-level tropical Atlantic / West Africa wind field. Most current GCMs are unable to directly represent the SST impact on the Sahel's rainfall variability. Therefore, this new result is significant for GCM prediction of Sahel rainfall at seasonal timescales, and potentially other timescales as well. For seasonal predictions, lead-time beyond a month is difficult due to SST developments. The second major result found here is the potential for coupled models to break through the lead-time barrier, with skill superior to that previously reported using empirical or GCM approaches. For the CFS model, skill levels only slightly less than those achieved by the best approaches using observed SST, have been achieved at lead-times of up to six months, utilizing the same low-level regional wind field.

The onset of the rains is critical for agriculture. Two homogeneous onset regions over Senegal are defined (southern region and northern region). Both regions experience frequent false onsets. We find relationships between onset date and a mix of large scale boundary layer and regional atmospheric synoptic features. Variables containing information include low level wind, SST, precipitable water and moist static energy. The southern region onset is influenced by the SST gradient over the tropical Atlantic. The northern region is more related to SST in the global tropics. At the synoptic time scale, we find clear propagation of signals from the eastern Sahel: relatively slow for the southern region onset (about 2.5 m/s), with faster propagation typical for the northern region (about 8.5 m/s). The risk of a false onset is increased if a negative anomaly of PWAT is located in the vicinity of Senegal and nearby tropical Atlantic, with a strong advection of cold (convection inhibiting) air from the north, a further danger sign. The CFS coupled model is found to also have some skill for onset prediction for the southern region when initialized in early May. It is concluded that combining boundary forcing and synoptic propagation can provide useful advance information about onset.