patterns are not necessarily those that explain interannual variability. Within this region, mean annual rainfall and its seasonality are highly heterogeneous (Figure 13), being associated with several features of the general atmospheric circulation as well as with regional-scale orographic and lacustrine effects. Nevertheless, the patterns of interannual variability are markedly similar throughout the region (Figure 14), so that there is a remarkable resemblance between the time series for the region as a whole, and for those four climatologically homogeneous subregions. The first eigenvector of annual rainfall shows loadings of the same sign throughout the region; the pattern explains 36 percent of the variability of annual rainfall and 52 percent of the variability during the "short rains" (Nyenzi, 1988).
In summary, there are several notable characteristics of rainfall variability that must be explained by mechanisms for producing this variability. These include, first of all, the continental-scale coherence and the seemingly contradictory contrast between the dominant low-frequency fluctuations in Sahelo-Saharan regions and the high-frequency fluctuations elsewhere, in regions that show strong teleconnections to the Sahel on time scales of 1, 10, and 100 years. The characteristic time scales of 2.3, 3.5, and 5 to 6 years must likewise be explained, as must be the seasonality of the changes and the coherent patterns of interannual variability in climatically heterogeneous regions.
Most studies of African rainfall have focused on interannual variability. Numerous authors have demonstrated relationships between SSTs and rainfall in various parts of Africa (e.g., Lamb, 1978; Lough, 1986; Semazzi et al., 1988; Wolter, 1989; Folland et al., 1991; Nicholson and Entekhabi, 1987). For parts of eastern and southern Africa linkages to ENSO have also been clearly established (by, e.g., Farmer, 1988; Ropelewski and Halpert, 1987; Ogallo, 1987; Harrison, 1983; Lindesay et al., 1986; Nicholson and Entekhabi, 1986, 1987). For Sahel rainfall, a number of other factors, such as changes in the upper-level winds and in the Hadley and Walker circulations, have also been implicated (Kanamitsu and Krishnamurti, 1978; Newell and Kidson, 1984).
Very little attention has been paid to the causes of the lower-frequency variability. The exception is the work by Folland and collaborators (e.g., Folland et al., 1991), which is limited to the Sahel. Virtually no studies have been carried out on the causes of the continental rainfall fluctuations on decadal time scales.
Rowell et al. (1992, 1995) suggest that the twentieth-century trends in Sahel rainfall can be attributed to interhemispheric differences in sea surface temperature. The interhemispheric contrast increased in mid-century and decreased sharply as the droughts began in the 1970s and 1980s. This mechanism does not, however, account for the extreme persistence of anomalies in the Sahel, or their lack of persistence in regions such as southern Africa, which otherwise appear to have teleconnections to the Sahel in individual years and on decadal-scales.
As an example of how this might work, consider a year during which anomalous large-scale circulation triggers drought in the Sahel and in other African regions showing strong teleconnections to the Sahel. When the large-scale forcing returns to normal, so do rainfall conditions in most areas. However, in the Sahel the drought conditions might be reinforced by local, land-surface feedback. In that case they will persist until a year in which a large-scale circula-