| dc.description.abstract | Extreme rainfall events have devastating socio-economic impacts associated with
food shortages, insufficient hydropower, water, shelter, loss of life and property.
Prediction and timely early warning of such extreme events are critical in order to reduce
the vulnerability of the society. Most of the past rainfall predictability studies have
concentrated on short term to seasonal/interannual time scales, yet decadal time scales
of floods or droughts phases seem to be quite common. The objective of this study is to
investigate the space-time patterns of East Africa decadal rainfall variability modes and
their predictability potentials using Sea Surface Temperatures.
Station rainfall data from 1920-2004 for the three East African countries and
global sea surface temperatures (SSTs) for the period 1950-2004 were used. Simple
correlation; trend and cyclical analyses; Principal Component Analysis (PCA); and
Canonical Correlation Analyses (CCA) methods were employed in the study.
The results showed that too much / too little rainfall received in one or two years
could determine the general trend of the decadal mean rainfall. The decades of 1921-1930
and 1961-1970 were noted to be generally wet while 1931-1940, 1951-1960 and 1991-
2000 were dry for the March-May rainfall season. During the October-December season,
the decades of 1961-1970, 1981-1990 and 1991-2000 were generally wet while 1931-
1940, 1941-1950 and 1951-1960 were dry. The decades of 1931-1940 and 1961-1970
were regionally dry and wet respectively. The June-August season showed wet/dry
decades to be 1941-1950, 1951-1960, 1981-1990, 1991-2000 / 1921-1930, 1931-1940,
1961-1970, 1971-1980 over the western region. Over the coastal areas, the wet/dry
decades were 1921-1930, 1931-1940, 1961-1970, 1971-1980 / 1941-1950, 1951-1960,
1981-1990, 1991-2000. Some of these anomalies were associated with extreme dry / wet conditions brought about by strong La Nina / EI Nino Southern Oscillation (ENSO) and
Indian Ocean Dipole (rOD) that were observed during few specific years.
Results from spectral analysis of seasonal rainfall record indicated spectral peaks
of 2.0-3.3, 3.5-4.0, 5.0-7.5, and 9.0-11.0 years most of which were not statistically
significant at 95% confidence level.
Results from rainfall Principal Component Analysis indicated that about six, five
and four modes accounting for 80%, 81.3% and 65.1% of total variance were significant
for March - May, October-December and June-August seasons respectively.
Canonical correlation results showed existence of some significant SST modes
that had significant predictability signals for seasonal rainfall. For example significant
positive correlations were observed between the sea surface temperatures and the
canonical component time series over the central equatorial Indian Ocean. March - May
positive loadings were spread over the coastal and Lake Victoria regions and negative
loading over the rest of the region with significant canonical correlation skill of -0.79 and
0.96 for lags one and zero respectively. The maximum skill was 0.72 for JJA rainfall and
Atlantic SSTs with negative loadings centred on the tropical western Atlantic Ocean
associated with the wet / dry regimes over western / eastern sectors, while the canonical
correlation skill between October - December rainfall and the predictor SSTs were 0.88
and 0.97 for lags one and Zero.
The study has shown evidence of decadal variability in East African rainfall
which has teleconnection to modes of variabilities over parts of the global oceans. The
results would be useful in developing policies that could be used to address some
mitigation and / or adaptation options for addressing challenges that are associated with
decadal variability climate extremes of decadal time scale, and climate change impacts. | en |
