The methodology used in this study comprised correlation analysis of the mean annual runoff and three-month quarter warm ENSO indices during the concurrent El Niño years considered. The hydrological year or seasonal cycle defined as a 12-month period starting from the month of lowest average rainfall for each year was selected. For all the 1 degree grids covering southern Africa, the seasonal cycle of flow and the annual flow depths are determined. The type and magnitude of the correlation are also investigated. We have also attempted to explore the possible evidence of a link between the long-term ENSO patterns and trends in seasonal and annual runoff pattern across the region of Southern Africa.

The variance accounted for by El Niño in the natural variability of precipitation of selected locations in Southern Africa is established. In order to avoid possible lag correlation between seasonal precipitation and seasonal SLPs, a correlation analysis is conducted with the annual discharge, against eight quarters of SOI in terms of SLP anomalies. Eight quarters are formed from four ENSO/SOI quarters during El Niño years, and two quarters each before and after El Niño years.

We measured the association between ENSO and precipitation using the linear correlation coefficient, r (also called the product-moment correlation coefficient, or Pearson’s r). In order to assess whether a correlation is significant, we use the test statistic, t, to identify the confidence level of the correlations between ENSO and the terrestrial hydro-climatological variable of precipitation. The test statistic, t, is defined by [33] as:

t = r N − 2 1 − r 2 (1)

which is distributed in the null hypothesis case (of no correlation) like the student’s t-distribution denoted by A(t/v) with v = N − 2 degrees of freedom. 1 − A(t/v) is the significance level at which the hypothesis that the means are equal is disproved. The mathematical definition of the function is given by:

A ( t / v ) = 1 v 1 / 2 B ( 1 2 , v 2 ) ∫ − t t ( 1 + x 2 v ) − v + 1 2 d x (2)

with lower and upper limiting values, respectively given by:

A ( 0 / v ) = 0 and A ( 1 / ∞ ) = 1 (3)

A(t/v) is related to the incomplete beta function Ix(a; b) by:

A ( t / v ) = 1 − I v v + t 2 ( v 2 , 1 2 ) (4)

The confidence level, γ for the t-statistic t(r, v) is then computed as given in [34] and further noted of its application in [23] as follows:

γ ( t , v ) = ∫ − t ( r , v ) t ( r , v ) f ( t , v ) d t (5)

Sample size uncertainties and the Pearson correlation model capabilities are investigated to determine the contribution of internal ENSO variability on the rainfall trend and drought regime in the region [33] [34]. This is particularly true when the recent period ENSO events from 2005 to early 2020 only capture about 5 cycles that relate to lower rainfall depths obtained from the corresponding period of GPCC gridded rainfall data of the region of southern Africa.

The number of samples used to quantify the linkage between ENSO and precipitation in terms of the Pearson correlation model capabilities would affect degree of robustness of the model (Figure 3) [33] [34]. By combining Equations (1) and (5), the confidence interval, β = (1 − γ)100% is computed (see Figure 3 for the sample size of 5, which is the number of datasets attributed to the warm ENSO events that occurred during the period of analysis of date set from 2005 to February 2020). A confidence level γ indicates that (1 − γ)100% of the variability is attributable to random variability. FORTRAN routines provided in Numerical Recipes [34] are used to compute the confidence intervals. Before applying the test, we normalised the data and also demonstrated the possible loss of power of the t-test as a result of the small sample size considered as shown in Figure 3 for

different sample sizes ranging between 5 to 15 samples. It can be noted that for correlation coefficient of 0.5 the confidence interval in the estimate of r = 0.5 would increase from 60% to nearly 85% as sample size increase from 5 to 15 sample sizes.

A correlation coefficient between seasonal warm ENSO indices and annual of a typical station in western Botswana bordering SE Namibia and NW South Africa over 8 consecutive quarters is shown in Table 2. It shows that the mean annual precipitation is positively correlated with the four quarter ENSO periods observed during the El Niño years. Amongst the 8 quarters, a strong positive correlation (r > 0.52) seems to be apparent in the rainy season DJF and MAM quarters. The confidence level of the correlation coefficient for these quarters is above 95%. The positive correlation diminishes for the other seasonal quarters, and the correlation becomes negative in the quarters that precede and follow the El Niño years. For this reason, we studied the correlation between the DJF quarter warm ENSO indices and annual and concurrent seasonal precipitation of the region.

Table 2 and Table 3 show the correlation coefficients and confidence interval between the seasonal and annual rainfall at selected sites across the Southern African region and the three-month mean warm ENSO indices of December to February (DJF). The computations were made based on two datasets that represent the period before and after 2005 over a span of early 1960s to early 2000s.

The link between seasonal rainfall in the region of Southern Africa and ENSO of December to February (DJF) in terms of Pearson correlation coefficient and

Key: [E]—El Niño year; [−]—Quarters preceding El Niño year; [+]—Quarters following El Niño year.

the confidence interval is shown in Figure 4. The corresponding correlation coefficients and the confidence interval between annual rainfall in the region of Southern Africa and quarterly ENSO of December to February (DJF) is presented in Figure 5. There is a marked similarity between these two figures, which is a clear indication that DJF quarter is a rainy season whose rainfall represents substantial proportion of the annual rainfall amount in the region.

Comparative investigations have been conducted to further investigate the association between annual rainfall and annual ENSO activity over annual time cycle presented in Table 4. It can be seen that this figure is still very much closer to the previous two figures highlighting the strong correlation between rainfall and ENSO activity, with a more pronounced effect observed during ENSO quarterly season of DJF.

Regional summary of correlation coefficients and confidence intervals of Seasonal and annual rainfall with DJF ENSO indices is provided in Table 5, which summarizes the correlation coefficient for the region of southern Africa in terms of percentage area distribution of the Pearson coefficient as well as confidence intervals.

Of the entire 1 degree gridded area of southern Africa, 40.5%, 32.6% and 26.9% of the region has Pearson correlation coefficient with ranges of 0.25 to 1.0; −0.25 to 0.25 and −0.25 to −1.0 between the seasonal rainfall and ENSO during the quarters of December to February (DJF). The same corresponding correlation coefficient ranges in the region between annual rainfall and ENSO is found to be 41.2%, 33.0% and 26.0%, respectively.

Considering the entire southern African region, 40.5% of the surface area exhibit strong positive correlation in terms of the December to February quarter rainfall and ENSO indices, including 17.4% of the area attributed to positive but weak correlation.

The confidence interval of the correlation coefficients between seasonal rainfall ENSO anomalies during DJF quarters is 74% whereas, between annual precipitation and DJF quarter ENSO indices reaches 84%, signifying substantial variance of precipitation during ENSO years is accounted for by the warm ENSO activity.

Strong positive correlation with ENSO activity is found to cover southern Zimbabwe, area around south eastern Namibia/North Western South Africa and southern Botswana/northern South Africa as well as North Zambia/Southern DRC. Most of these areas in the region represent areas with lower rainfall regime in general with lower water security, with implications of water scarcity which were hard-hit with water shortages in 2019 and early 2020.

Occurrence of below-normal rainfall amounts from December to March during El Niño years and warm ENSO positive association with drier rainfall conditions have been noted in South Africa in several studies [2] [14] [21] [22].

In particular, the results of the regression analysis (Figure 4) confirm a weak yet statistically significant association between ENSO and the latitudinal position of the Angola low as noted in [35]. This also agrees with the conclusions in [35] [36], specifically for January to March season of 1998, which further supports the notion that internal atmospheric extratropical variability is partly responsible of the Angola low low-frequency variability.

Annual rainfall variation at a station located in an area centered on grid 25.50˚E/25.50˚S in southern Botswana and northern South Africa is shown in Figure 6. It clearly shows the 2019/20 rainfall signifies very low rainfall and drought that occurred in the region. Below-average production and drought impacts are felt in larger and some pockets of various countries in southern Africa, including but not limited to Angola, Botswana, Lesotho, Namibia, Zambia, Zimbabwe, South Africa and Mozambique. The areas identified in this study with below rainfall regimes associated with warm ENSO events are similar to recent observation and model simulation of North American Multi-Model Ensemble (NMME) probabilistic forecast for precipitation for January to March 2020 based on NOAA/CPC, which is widely used for forecasting outlook assessment of drought impacts by FEWSnet [37].

A number of climate indicators are noted to have changed in the year up to end of 1990s, especially around the Pacific Basin. As can be seen on Figure 2, before end of 1990s El Niño seems to have occurred with about equal frequency, each at intervals of about 3 to 7 years, and between 1960 and 1990 alone, there have been about 7 El Niños (almost one every 4 years). Figure 6 depicts standardized annual rainfall variation showing ENSO/La Nina years at a station located in an area centered on grid 25.50˚E/25.50˚S in southern Botswana and northern South Africa.

Below normal in the annual rainfall of 2019/20 in southern Africa, calculated from global gridded data base of the Global Precipitation Climatology Centre (GPCC) provided by Ziese et al. [28] is depicted in Figure 1. Below normal rainfall are recorded in southern Africa and higher rainfalls are recorded in equatorial parts of Southern Africa especially in the Congo River basin.

This analysis of the long-term trend of the El Niño phenomenon over the period 1950 to early 2020, depicted in Figure 2 and Figure 6, supports the findings of decline in precipitation in Southern Africa shown in Table 4 and Figures 4-6 and corroborated in other studies, e.g. [1] [19] [20] [38] [39].

The studies demonstrate the occurrence of low seasonal rainfall totals together with high seasonal temperatures during preceding seasons of El Nino/ENSO years are likely to be more frequent under future climate change as noted in Archer et al. [21] and [22]. Also the studies and the reported results also highlight the potential importance of the ENSO activity for southern African rainfall [36] [40] [41] [42]. ENSO events are also found to be an important phenomenon in modulating the skills of regional climate and CMIP5 models as illustrated by Dieppois et al. [1] and Taylor et al. [27].

Recent studies by Monerie et al. [18], Archer et al. [21] and Winsemius et al. [22] corroborate occurrence of low seasonal rainfall totals together with high seasonal temperatures during preceding seasons during El Nino/ENSO years are likely to be more frequent under future climate change.

Predicting climate for future seasons to several decades is useful for decision makers to adapt policies to near-term climate change [43]. The effect of low precipitations would be a potential revelation for decline of runoff of some rivers in southern Africa which would determine water resources availability for societal demands as noted by Alemaw and Chaoka [23].

The need to anticipate damages due to climate variability is a stressful problem, especially in developing countries, which are more vulnerable to climate hazards. Climate projections are mostly provided by simulations performed with Atmosphere-Ocean General Circulation Models (AOGCM) under the Climate Model Intercomparison Project, phase 5 (CMIP5) [1] [18] [27].

The ability of CMIP5 models to simulate realistic summer rainfall variability of rainfall over southern African and noted the summer teleconnections with global SSTs from extended reconstructed SST version 4 (ERSST.v4) dataset of the National Climatic Data Centre is examined by Dieppois et al. [1]. The modelling study through various models demonstrated the impact of SST variability in the South Atlantic and South Indian Oceans on rainfall variability in southern Africa. With various degrees of association between SST and SA rainfall is also reported and documented for different time scales and time horizons.

These precipitation forecast products have been applied for forecasting and operational applications including the following: climate modelling [1] [21] [41]; precipitation extremes [38]; and regional agriculture and vulnerability assessment [37] [44]. Other applications also include rainfall and river flow analysis [5] [23]; real-time precipitation monitoring [43] [45]; fisheries and aquaculture [17]; and study of greenhouse gas forcing [24].

All together these results also corroborate the potential importance of the ENSO activity for southern African rainfall, as suggested in earlier studies in the region including [36] Nicholson and Kim [36], and others [19] [20] [38] [39] [40] [41] [42]; and ENSO events in modulating the skills of CMIP5 models reported by Dieppois et al. [1] and Taylor et al. [27].