Climate and Circulation Changes in Southern Africa (already finished)

Start date: 01.08.2000
End date: 31.03.2003
Funded by: DFG (Deutsche Forschungsgemeinschaft)
Local project leader: Prof. Dr. J. Jacobeit


Still many economies in southern Africa are dominated by agriculture, which strongly depends on rainfall. Climate variations in semiarid regions have a strong socioeconomic impact, because of large populations and limited natural resources. Understanding of past climate variations has a high value for the prediction of future climate, seasonal forecast and the crop and water management. Meanwhile climatic data are already used to predict cropyield and to reduce the vulnerability by droughts.

The present study intends to contribute to the understanding of climate trends, changes and variability in southern Africa during the past 100 years. Analyses of atmospheric circulation changes and couplings with sea-surface temperatures are one main part of this Ph.D Thesis, which is embedded in the DFG-post-graduate program "Joint Geoscientific Research in Africa".




The Global Mean Sea Level Pressure (GMSLP2.1) data set was developed in collaboration of the Hadley Centre, the Meteorological Office (U.K), CSIRO (Australia) and NIWA (New Zealand). It is a monthly gridded data set covering the period 1871-1994 with a 5° latitude by 5° longitude resolution. SSTs are taken from the Global Sea Ice and Sea Surface Temperature (GISST2.3b) data developed by the UK Met. Office. The data start in 1856 having a spatial resolution of 1° x 1°. The Kaplan SST Data are anomalies with respect to the 1961-1990 period. They cover the period 1856-1991 with a resolution of 5° x 5°. Precipitation and air temperature data are taken from the 0.5° x 0.5° Climate Research Unit (Norwich) Data covering the 1901-1998 period.




The major modes of monthly SST, SLP, temperature and precipitation variability have been derived by a principal component analysis (PCA) with Varimax rotation. S-mode PCAs were applied to describe the major centres of variability of the corresponding climate elements.
The resulting time coefficients of the PCs have further been analysed by canonical correlation analysis (CCA). The aim of this multivariate statistical method is to describe the common variation of pairs of different variables. The resulting canonical scores and loadings reveal the connections both on temporal and spatial scales.
In order to study possible remote forcings of the precipitation and temperature fields of southern Africa, CCAs between monthly SST and precipitation, as well as between SST and air temperature, SLP and precipitation and temperature were carried out. The particular influence of the Atlantic, Pacific and Indian Oceans on southern African precipitation was investigated by different analyses for each ocean and by an analysis treating them simultaneously.
Composites of months with positive and negative time coefficents (beyond +1 and -1, respectively in standard deviation terms) were calculated as means from monthly values. Weighted with the absolute value of the corresponding time coefficent representing real climate conditions in comparision to the statistically derived CCA results.





PCA is a useful method to describe the distribution and variation of temperature within the study period. The increasing air temperature during the last two decades, which is obvious for all seasons, is clearly revealed for the region of Botswana with the first t-mode PC of the CRU temperature data for January. The loadings show an increasing trend in two major steps since the mid of the last century. Decadal anomaly plots confirm the PCA results, obviously at the third and forth PCs. The third PC describes the warm period in the region of west Zimbabwe during the 1920s. The forth PC describes a cold period for Namibia during the 1930s (Figure 1).


Fig. 1:
First, third and forth principal components for January CRU temperature data. The explained variance is 33,1 %, 17,5 % and 16,2 % of the total variance, respectively. The component loadings are presented on the left diagrams, showing a marked trend pattern on the first, and pronounced warm periods on the third and forth PCs.
Decadal anomaly plots for selected periods are presented to the right of the spatial patterns.

In order to detect forcing factors for temperature changes, CCAs were performed with both SLP and SST. The results demonstrate a clear connection of the Mascarene High and temperature at the eastern part of southern Africa. Composites of 14 positive and 16 negative cases, respectively, show a distinct coincidence of high (low) subtropical pressure with high (low) air temperatures at the Atlantic and Indian Ocean coastal regions, including Madagascar. The Cape region and central southern Africa show an inverse coupling to SLP.
The pairs of canonical correlation patterns of SST (GISST data) and air temperature indicate a similar picture for the summer months with positive couplings to the south east Atlantic SSTs.




Fig. 2:
First pair of canonical correlation for sea-level pressure and temperature. The panels on the bottom present weighted composites plots for positive (upper section) and negative (lower section) time coefficients.


A PCA on southern African rainfall shows widespread variability, but no distinct trend. The regionalization of precipitation by means of s-mode PCAs gives 20 consistent regions from the 3530 CRU gridboxes, which have further been analysed by CCA.

The contribution of moist air from the Indian Ocean to southern African summer rainfall can be seen on Figure 3. There is a strong correlation between sea-level pressure northeast of Madagascar and precipitation south of 15°S. This characteristic pattern can be derived for nearly all austral summer months and doesn´t show any clear trend in the time coefficients.


Fig. 3:
First pair of canonical patterns (February) for gridded SLP and precipitation data. The canonical correlation r is 0,85. The explained variances are 5,02 % and 4,89 %.

The influence of SST on rainfall is a thoroughly studied objective. The focus of this study is to pay attention to changes in the coupling of SST and precipitation through time. In order to separate the influence of each southern ocean, CCAs were applied specifically to each southern ocean as well as to all of them together. The best coupling could be found for both each the Atlantic and the Indian Oceans.

The trend mode of this analysis presents on the first canonical pair a decrease of negative coupling and and increase of positive couplings of both Atlantic and Indian Ocean SSTs, that means, that precipitation deficits together with low SSTs are of decreasing importance within the study period. The region of Angola is out of phase with the rest of southern Africa.



Fig. 4:
First pair of canonical patterns for December with Atlantic SST (Kaplan-SST data) and precipitation. The canonical correlation r is 0.92. The explained variances are 16,02 % and 22,39 %. The canonical scores are shown for just one variable.




Fig. 5:
First pair of canonical patterns for December with Indian Oceans SST (Kaplan-SST data) and precipitation. The canonical correlation r is 0.82. The explained variances are 24,00 % and 17.30 %. The canonical scores are shown for just one variable.

SST Variability :



The dominant SST pattern for all months represents El Nino. The t-mode PCA, using the Kaplan SST anomalies, shows this on the first PC with the well-known time coefficients increasing since the mid 1970s.





Fig. 6:
First principal component for January SST. The explained variance is 14,5%  of the total variance. The time coefficients are shown on the bottom panel.

Other features as tropical dipole patterns in the Atlantic and Indian Oceans are less pronounced and seem to be more regional rather than global phenomena.
The tropical Indian Ocean Dipole could only be reproduced by performing an Indian Ocean PCA for October. Figure 7 presents the influence of SST on precipitation of East Africa on the 5th canonical pair with GISST data. The canonical scores clearly show the extreme precipitation event in East Africa in 1961, which can be traced back to anomalously high SST in the western Indian Ocean and cold SST in the eastern part, which means a reversal of normal conditions.





Fig. 7:
Fifth pair of canonical patterns for October with GISST and CRU Precipitation data. The canonical correlation is 0.63, the explained variances are 3.8% and 5.0%. The canonical scores are shown for the variable “SST”.

A CCA was also applied opposite to the common use, which investigates different climate elements (e.g. precipitation and SST), to compare the coupling of different regions by taking the PC scores of them. The aim is to investigate the coupling of tropical SSTs between the Atlantic, Indian and Pacific Oceans. For this purpose, the three oceans are assumed to have distinctly separate characteristics.

The correlation between the Atlantic and Indian Oceans is stronger than with the Pacific Ocean. The positive correlation between equatorial Atlantic SST and the west Indian Ocean shows a marked trend mode. The canonical scores of the pattern with cold SST in the Atlantic and cold SST in the western Indian Ocean are decreasing since the mid 1930s whereas the positive coupling is increasing. These results are consistent for all months.
The correlations of Pacific SST to Atlantic and Indian Oceans SST are weaker. The strongest couplings are between the El Nino region in the eastern Pacific and the Indian Ocean except for the equatorial region. This region is coupled out of phase with both Atlantic and Pacific Ocean SSTs.
In most cases the highest correlations can be found in the tropical Atlantic and the western Pacific Oceans.





Fig. 8:
Different pairs of canonical patterns. The upper panels show the first pair for February with the PCs of the Atlantic and the Indian Ocean SSTs. The canonical correlation r is 0.86. The explained variances are 10.13 % and 14.53 %. The central panels show the same for the Pacific and the Atlantic Oceans; r is 0.86. The explained variances are 11,08 % and 13.69 %. The lower panels show the canonical patterns for the Pacific and the Indian Ocean SST, r is 0.81, the explained variances are 6.58 % and 18.0 %. The particular time coefficients, the canonical scores are shown at the bottom.