Extreme wet years over southern Africa: Role of Indian Ocean sea surface temperatures



[1] Southern Africa is a predominantly semiarid region with a high degree of interannual rainfall variability. Although much of the recent climate research has focused on the causes of drought events, the region has also experienced extremes of above average rainfall, the most recent examples being the major flooding episodes that devastated Mozambique during 2000 and 2001. This paper investigates extremely wet years over southern Africa during the twentieth century. Focusing on the two most extreme years, 1974 and 1976, we show that while ENSO serves as an important control on rainfall variability, a specific pattern of SSTs in the SW Indian Ocean, with warm anomalies in the subtropical SW Indian Ocean and cool anomalies in the northern SW Indian Ocean that is statistically independent of ENSO, plays a crucial role in generating extreme conditions. To do this, we use a series of multimodel experiments, to demonstrate first the importance of global sea surface temperatures. Through additional idealized experiments with HadAM3, we then isolate the role of SST anomalies in the Indian Ocean. The anomalies are based on the observed SSTs with the ENSO signal linearly removed. The critical influence is tied to cold SST anomalies in the Mascarene region which induce an anomalous anticyclonic circulation driving an anomalous low-level easterly moisture flux along 10–20°S into eastern southern Africa. This results in enhanced moist convective uplift, conducive to enhanced rainfall, over a large part of southern Africa. Near surface humidity and 500-hPa omega fields extend from eastern southern Africa into the Agulhas region in a tropical-temperate cloud band like structure. The similarity between the reanalysis fields for the extreme years and the model experiments is striking.

1. Introduction

[2] Southern Africa is a predominantly semiarid region with a high degree of interannual rainfall variability [Tyson, 1986; Mason and Jury, 1997]. A heavy reliance on subsistence agriculture exposes the population base to drought events [Vogel, 1994; Mason and Joubert, 1995; Washington and Downing, 1999; Dube and Jury, 2000; Jury, 2002], a feature of the climate system which has received significant recent emphasis [e.g., Cook, 2001; Jury, 2002; Jury and Mwafulirwa, 2002; Endfield and Nash, 2002; Nash and Endfield, 2002; Levinson, 2004; Thomson et al., 2003; Usman and Reason, 2004]. The increased likelihood of the greater frequency and intensity of sustained drought [Hoerling et al., 2006], land degradation and desertification [Thomas and Leason, 2005; Thomas et al., 2005] in future decades, set against the pressure of increasing population, further raises the vulnerability of this region.

[3] In addition to the hazard of drought, southern Africa has also experienced extremes of above average rainfall, the most recent examples being the major flooding episodes that devastated Mozambique during 2000 and 2001 during which hundreds of people were killed and nearly 200,000 people made homeless. Total cost of the damage resulting from the 2000 event was approximately 500 million USD [du Plessis, 2002]. In parts of southern Africa, an increased frequency of heavy rainfall events has been identified over the period 1979–2002 [Usman and Reason, 2004]. Overall there is somewhat less research on wet events over southern Africa (examples include Mason and Joubert [1997] and Reason and Keibel [2004]).

[4] A more complete understanding of climate mechanisms in southern Africa can only come about if both wet and dry years are considered. Accordingly, this paper investigates years of large rainfall excess over southern Africa. In particular, we are concerned with the sea surface temperature (SST) patterns and their associated atmospheric circulation anomalies that accompany wet extremes. The paper examines the hypothesis that Indian Ocean SST anomalies are crucial for explaining extremely wet years. Section 2 presents an analysis of observed rainfall and sea surface temperatures associated with years of extreme rainfall. Experiments with atmosphere climate models driven with historical SST and idealized SST patterns are examined in section 3. Section 4 provides a discussion and summary of the main findings.

2. Observed Rainfall and Sea Surface Temperatures

2.1. Background

[5] Southern Africa south of 15°S is a predominately semiarid region. Annual rainfall totals averaged over land south of 15°S are 316 mm.yr−1, with 40% of the total rainfall occurring between January and March, associated with the southward migration of tropical convection to the subtropical latitudes (approximately 20°S) and the formation of tropical-temperate cloud bands at this time of year [Mason and Jury, 1997; Washington and Todd, 1999].

2.2. Rainfall Variability

[6] We start with the time series of area average land-based rainfall anomalies for southern Africa south of 15°S (Figure 1) for 1900 to 1998. The area average for each JFM (January, February, March) season was calculated from the mean of rainfall anomalies (based on 1961–1990 period) using the Hulme et al. [1998] monthly data set (2.5° latitude × 3.75° longitude resolution). Three years in the record of area average rainfall exceed 2.5 standard deviations above the mean (1976, 1974 and 1909). The two most extreme years fall into the latter half of the twentieth century during which meteorological and oceanographic records are of sufficient quality to support detailed analysis. In this study, we therefore focus on the JFM season of 1976 and 1974, neither of which has been comprehensively studied before.

Figure 1.

Southern African January to March (JFM) rainfall south of 15°S from the Hulme et al. [1998] data set, 1900–1998. Vertical axis shows standardized anomalies relative to the 1961–1990 climatological mean (mean is 316 mm and standard deviation is 80 mm).

2.3. ENSO and Anomalously Wet Years

[7] One simple explanation for the extreme rains in 1974 and 1976 lies with the ENSO phenomenon. It has long been known that southern African summer rainfall is significantly influenced by ENSO [Lindesay, 1988; Van Heerden et al., 1988; Ropelewski and Halpert, 1989; Nicholson, 1997; Nicholson and Kim, 1997; Cook, 2000, 2001] with La Nina episodes generally bringing anomalously wet conditions (Figure 2). This association is also evident in at least 200 years of proxy and observed data [Lindesay and Vogel, 1990], although the strength of the relationship has varied over the past century. Table 1 confirms that for each of the months January through March 1974 and 1976, extreme La Nina conditions prevailed in the Pacific. During each of these months in 1974, the Southern Oscillation Index (SOI) was the highest value recorded throughout the period 1950 to 1998 inclusive. Likewise, 1976 was ranked the 5th highest SOI in January and February and the 3rd highest in March. The Nino3 SST rankings are similarly extreme.

Figure 2.

Correlations between JFM Southern Oscillation Index and JFM southern African rainfall for 1900–1998 (shaded regions significant at 5% level).

Table 1. ENSO Statistics for January, February and March 1974 and 1976
1974 SOI4.33.23.6
Rank 1974 SOI 1950–1998111
1976 SOI2.42.62.2
Rank 1976 SOI 1950–1998553
1974 Nino 3−1.64−1.20−0.70
Rank 1974 Nino 3 1950–1998146
1976 Nino 3−1.78−0.97−0.59
Rank 1976 Nino 3 1950–1998369

[8] However, a clear indication that Pacific sea surface temperature conditions are not a simple determinate of anomalously wet conditions over southern Africa, is that other years with anomalously cool Nino3 SSTs between 1950 and 1998, some of which rank higher than 1976 in widely used ENSO indices (including SOI and Nino3), were not anomalously wet. Examples include 1968, which was anomalously dry and 1970/1971, a near average year. Likewise, 1967/1968 featured an extreme La Nina. Conditions were predominately wet over southern Africa but at 0.55 standard deviations above the mean, not excessively so.

2.4. Non-ENSO Related Rainfall Variability

[9] A key question relating to the two extreme rainfall years of 1974 and 1976 is whether these years were simply a result of unusually strong La Nina events or whether other conditions, in addition to and perhaps independent of, the La Nina events were favorable for above normal rainfall. The existence of extreme La Nina years which were not excessively wet (and some of which were dry) over southern Africa (section 2.3) suggests that other influences may also control rainfall and there is indeed much in the literature to suggest that SSTs in surrounding ocean basins exert an independent influence on southern African rainfall [Walker, 1990; Rocha and Simmonds, 1997a, 1997b; Goddard and Graham, 1999; Behera and Yamagata, 2001; Reason, 2002]. It is also clear that ENSO controls much less than half of the interannual variance of JFM SSTs in the Indian Ocean (Figure 3).

Figure 3.

Correlation (r) between the JFM SOI and JFM SSTs (1959–1998). Shading indicates significance at the 5% level. Contour interval 0.1. Interannual variance in JFM SSTs explained by ENSO as measured by the SOI is determined by r2.

[10] To test the possible role of non-ENSO influences, we create an index of southern African rainfall with ENSO linearly removed by regression. Although there are a number of limitations to this technique, most notably those relating to linearity, it does provide a basic way of analyzing variability independent from ENSO and has been used in a number of previous studies for the same purpose [e.g., Jones, 1988, 1994; Christy and McNider, 1994; Zhang et al., 1996; Kruger, 1999; Barreiro et al., 2002]. Regression was favored over the use of bandpass filtering. In the latter technique, removal of the interannual component of variability (2–7 years), which characterizes ENSO, would have also resulted in the removal of non-ENSO modes of variability on similar timescales.

[11] The regression approach used to linearly remove ENSO may be formalized as follows:

equation image
equation image



= Seasonal (JFM) rainfall anomaly (with reference to 1961–1990)


= Seasonal average ENSO index value


= Regression coefficient


= Regression coefficient

equation imagei

= Estimated value for yi given xi


= Seasonal (JFM) rainfall value with ENSO signal removed

such that each rainfall grid box has a unique regression equation.

[12] With the ENSO signal linearly removed from each grid box of JFM rainfall over southern Africa, we construct the area average time series of the y'i anomalies using the same method by which data in Figure 1 were generated (section 2.2).

[13] We have investigated the sensitivity of the linear regression approach to the precise index of ENSO used in the regression, including OND and JFM averages of the Nino3.4, Nino3, SOI and a multivariate index [Wolter and Timlin, 1993] over the period 1950–1998. The rankings of the top 8 wet years are not changed as a function of the index used and we conclude that the results are largely insensitive to the precise ENSO index.

[14] Even with ENSO linearly removed, both 1974 and 1976 appear as extreme wet anomalies with 1974 as the third and 1976 the fourth wettest JFM anomalies in the in the yi time series from 1900 to 1998. While noting the assumptions of linearity in the regression based approach, we conclude that the extremely wet years of 1974 and 1976 may not have been solely a product of Pacific based ENSO forcing.

[15] An important caveat in this method and the conclusion we have reached above is that we have not considered the multidecadal variability of ENSO and the impact this may have on the Indian Ocean and southern Africa [e.g., Reason and Rouault, 2002; Allan et al., 2003] in this method. The method may also not deal adequately with the differences between protracted and shorter ENSO events [Allan et al., 2003] which may turn out to be important in the reasons for the wet conditions in the 1970s.

2.5. Extreme Rainfall and SSTs

[16] In order to establish non-ENSO related SST patterns associated with the extreme rainfall years of 1974 and 1976, we derive JFM composites of SST anomalies for ocean basins surrounding southern Africa. We use the Global Ice and Sea Surface Temperature data (GISST2.3b) [Rayner et al., 1996], but prior to computing the composites, we linearly regress the ENSO signal from each grid box (2.5° latitude × 3.75° longitude resolution) of the SST data, using the SOI as a measure of ENSO, following the method described for linearly removing the ENSO signal from rainfall in section 2.4.

[17] Anomalous SST gradients with positive (negative) anomalies approaching 1°C (−1°C) in the subtropical (tropical) southwest Indian Ocean are clearly evident in these resulting composites (Figure 4). There are also years, namely 1970 and 1964, where anomalous SST gradients of the opposite sign are clearly evident (Figure 5). These years, 1970 and 1964 are the second and sixth driest in the yi time series from 1900 to 1998. During 1970, Zimbabwe was seriously affected by intense drought [Jury et al., 1992]. What emerges from this simple statistical analysis, is a pattern of SST variability in the SW Indian Ocean with an enhanced poleward gradient associated with some extreme dry years and a relaxed poleward gradient associated with some extreme wet years.

Figure 4.

Composite JFM SST anomalies for 1974 and 1976 (contour interval 0.1°C). The JFM SOI signal has been linearly regressed from all data.

Figure 5.

Composite JFM SST anomalies for 1964 and 1970 (contour interval 0.1°C). The JFM SOI signal has been linearly regressed from all data.

3. Model Experiments

[18] In section 2 we have shown that the wet years of 1974 and 1976 were associated with anomalous SST patterns in the southwest Indian Ocean and also with strong La Nina conditions. To evaluate whether the association with the SST patterns was simply coincidental, and if not, to test whether the rains could have resulted from Indian Ocean SSTs separate from a Pacific based ENSO influence, we present the results of several atmosphere general circulation model (AGCM) experiments.

3.1. Historical SST Experiments

[19] In this section we assess the 1974 and 1976 JFM rainfall anomalies over a broad southern African domain from the ensemble mean of 5 atmospheric GCMs forced with historical SSTs (Figure 6). These years have been sampled from longer integrations of the models, typically running from 1950 onward, with each member of the ensemble integrated from differing initial conditions in the start years. The models, approximate resolution and number of members making up the ensemble mean are shown in Table 2.

Figure 6.

AGCM simulations for (left) JFM 1974 and (right) JFM 1976. First row is CCM3, second row is COLA T63, third row is ECHAM4.5, fourth row is NCEP MRF9, and fifth row is NSIPP-1. Units are rainfall anomalies in mm.day−1.

Table 2. Details of Atmospheric GCMs Used in the Historical SST Runs
ModelResolution: LatitudeResolution: LongitudeNumber of Ensemble Members
COLA C2.21.8751.86467810
ECHAM 4.52.81252.78932824
NCEP MRF92.81252.78932810

[20] Positive total rainfall anomalies over southern Africa are found in all 5 models in JFM 1974 and in 4 of the 5 models in 1976 (the exception being NSIPP). The anomalies are generally larger and more spatially coherent in 1974 than 1976, although the rainfall response within one year is not the same in any of the models, probably owing to differences in the basic state of the models. The COLA, CCM3 and NCEP (and NSIPP to a lesser extent) simulations all show rainfall anomalies extending from the tropics to subtropical-extratropical margins during 1974. The structure of this anomaly is reminiscent of the tropical-temperature cloud bands (TTCB) that are regarded as the leading southern African rainfall producing system [Harrison, 1984; Todd et al., 2004]. In all the simulations and for both 1974 and 1976, dry anomalies are found north, west or northwest of Madagascar in the southwest Indian Ocean. The conclusion we reach on the basis of these simulations is that the observed global sea surface temperature conditions during the austral summer of 1974 and 1976 were most likely to have been responsible for generating the anomalously wet conditions over southern Africa. What these experiments are unable to demonstrate is whether Indian Ocean SSTs could have induced the wet conditions on their own.

3.2. Idealized SST Experiments

[21] We now go on to use idealized sea surface temperature experiments with the atmosphere only model, HadAM3, to determine whether the pattern of sea surface temperatures evident in the southwest Indian Ocean during JFM 1974 and 1976 is capable of forcing the anomalously wet conditions over southern Africa on its own. The experiments are part of a large (>20) suite of experiments using HadAM3 aimed at an improved understanding southern Africa rainfall variability and SSTs.

3.2.1. Background to HadAM3

[22] A large proportion of previous southern African climate research, including many of the climate change studies in the mid-1990s, was undertaken using the Commonwealth Scientific and Industrial Research Organisation (CSIRO) model [Mason et al., 1994; Joubert et al., 1996; Mason, 1996; Mason and Joubert, 1997; Rautenbach and Smith, 2001]. More recently, there have been several sensitivity experiments featuring the Melbourne University GCM (MUGCM) [Rocha and Simmonds, 1997b; Reason and Mulenga, 1999; Reason, 2001, 2002] and the application of the European Centre/Hamburg Model (ECHAM) in the context of seasonal forecasting and model output statistics [Landman and Goddard, 2002, 2005]. HadAM3 has been studied in the context of intraseasonal [Tennant, 2003] and interannual variability [e.g., Reason and Jagadheesha, 2005], as well as for operational seasonal forecasting.

[23] A detailed description of the United Kingdom Meteorological Office (UKMO)'s model, HadAM3, is provided by Pope et al. [2000]. In these experiments, the hydrostatic model is run in standard climate resolution with a horizontal grid of 2.5° latitude by 3.75° longitude with 19 vertical levels (5 representing boundary layer processes) calculated in hybrid sigma/pressure coordinates [Simmons and Burridge, 1981].

[24] HadAM3 correctly simulates a unimodal annual cycle with maximum rainfall in the austral summer (October–March) and minimum rainfall in the austral winter (April–September). The model does overestimate summer rainfall. This is most pronounced in the early summer season (October to December) with the positive bias continuing into the late summer months (January to March). The annual rainfall total over the whole subcontinent in HadAM3 for JFM is 387 mm compared to the observed total of 316 mm. The overall simulation of African climate has been thoroughly investigated by Preston [2004].

3.2.2. Idealized Indian Ocean SST Experiments: Setup

[25] In experiment 1, the poleward gradient of SSTs in the SW Indian Ocean is decreased by imposing a warm SST anomaly of 1.5°C centered at 32°S, 55°E on the climatological SST values and a cold SST anomaly (with a minimum value of −1.0°C) centered at 12°S, 65°E (Figure 7). The location and magnitude of these anomalies is based on SSTs with ENSO linearly removed during 1974 and 1976 (see section 2.5), although the idealized SSTs are smoothed and symmetrical compared with the observed and SSTs. Also, outside the idealized dipole areas in this experiment and in experiments 2 and 3, SSTs over the remaining part of the Indian Ocean follow the climatological values. In experiment 2 and experiment 3, only the subtropical warm anomaly and the tropical cold anomaly respectively are included, so that the role of these separate components of experiment 1 can be considered in isolation. Climatological SSTs are specified in all other parts of the global oceans. Each of the experiments has a common design, consisting of 10 integrations, which are identical apart from the arbitrarily different initial starting conditions, thereby allowing for the averaging of internal variability [Mote, 2000; Wehner, 2000]. The SST anomalies are introduced in December and are persisted through the late summer season, although the base SST climatology evolves according to the seasonal cycle with 5-day updates. The integrations for these experiments all start in November and run for 5 months until the end of March.

Figure 7.

Idealized SST anomalies (°C) superimposed on the SST climatology to force HadAM3 in experiment 1. Experiment 2 was forced by the warm anomaly only, and experiment 3 was forced by the cold anomaly only. Climatological SSTs were prescribed in the rest of the global oceans.

[26] The three experiments are evaluated against a control experiment which consists of 10 years of integrations forced with climatological SSTs, starting in April, to give a long spin-up period during which the model is able to reach an equilibrium with the climatological SST forcing before analysis in the late austral summer (JFM) data of the following year. The significance (at the 0.05 level) of differences between the experiments and the control run is determined by means of a t-test calculated locally on each model grid box.

3.2.3. Idealized Indian Ocean SST Experiments: Results Experiment 1

[27] Experiment 1, which includes 2 SST anomaly loci in the SW Indian Ocean, induces large-scale significant atmospheric anomalies over southern Africa and the Indian Ocean. Positive sea level pressure (SLP) anomalies dominate the tropical SWIO, with the peak departures from the climatology (3 hPa) in the Mascarene region to the east of Madagascar (Figure 8) representing an equatorward expansion of the semipermanent anticyclone and a reduction in the tropical Indian Ocean SLP minimum (1009 hPa). Accompanying this adjustment are significant anticyclonic 850-hPa wind and 700-hPa moisture flux anomalies which are strongest at 10–20°S, along the northern edge of the SLP anomaly, all but replacing the westerly climatological flow in this sector. The effect is to allow the northward expansion of the southeasterly trade winds. The significant easterly 700-hPa moisture flux anomalies (which are representative of the integrated moisture flux) converge with weaker significant westerly anomalies over Malawi and Zimbabwe (Figure 9). This results in significant positive 700-hPa specific humidity anomalies (0.5 g.kg−1) and significant negative 500-hPa omega anomalies (−0.02 m.s−1) over eastern southern Africa which extend southeast into the Agulhas region as a TTCB form. While the model does not show a significant increase in rainfall, probably because of convective parameterization, the ingredients for enhanced rainfall are present.

Figure 8.

JFM SLP anomalies for experiment 1 minus control (contour interval 0.5 hPa). Shading indicates statistical significance at the 5% level according to a two-tailed Student's t-test.

Figure 9.

JFM 700 hPa moisture flux anomalies for experiment 1 minus control (scale vector = 30 g.kg−1.s−1). Shading indicates statistical significance at the 5% level according to a two-tailed Student's t-test. Experiment 2

[28] Are both the cold and warm components of the SST anomaly of experiment 1 necessary to bring about the enhanced moisture flux, specific humidity and convection over southern Africa? On the whole, the SST anomaly in experiment 2 (warm subtropical component of experiment 1 only) appears to have little effect on the large-scale atmospheric circulation over southern Africa and the surrounding ocean basins. There are no significant SLP changes in around southern Africa although there are slight decreases (−0.5 hPa) throughout the southern subtropical Indian Ocean and to the south of southern Africa (Figure 10) consistent with a limited weakening and contraction of the South Indian High. Associated with these SLP anomalies, there are nevertheless significant northwesterly 850-hPa wind anomalies and 700-hPa moisture flux anomalies in the southeast Indian Ocean which amount to a weakening of the southeasterly trade winds There are significant positive humidity anomalies (0.2 g.kg−1) over Lake Malawi and the Mozambique Channel with small and nonsignificant positive humidity and negative omega anomalies over the warm SST forcing anomaly.

Figure 10.

JFM SLP anomalies for experiment 2 minus control (contour interval 0.5 hPa). Shading indicates statistical significance at the 5% level according to a two-tailed Student's t-test. Experiment 3

[29] Experiment 3, consisting of the cold SST anomaly component of experiment 1 (in the Mascarene region), generates a response which is very similar to experiment 1, especially in the tropical SWIO. There are significant positive SLP anomalies of a similar magnitude to experiment 1 in the Mascarene region to the east of Madagascar (Figure 11). The 850-hPa wind anomalies and 700-hPa moisture flux anomalies (Figure 12) are also similar between the two experiments. However, there are slight shifts in the orientation of the anticyclonic feature and the easterly anomalies along 10°S are slightly weaker in experiment 3 compared with experiment 1.

Figure 11.

JFM SLP anomalies for experiment 3 minus control (contour interval 0.5 hPa). Shading indicates statistical significance at the 5% level according to a two-tailed Student's t-test.

Figure 12.

JFM 700 hPa moisture flux anomalies for experiment 3 minus control (scale vector = 30 g.kg−1.s−1). Shading indicates statistical significance at the 5% level according to a two-tailed Student's t-test.

[30] Associated with the relatively weaker moisture flux anomalies, the negative 700-hPa specific humidity anomalies in the Mascarene region are weaker in experiment 3 than in experiment 1 (−0.9 g.kg−1 compared to −1.5 g.kg−1). The positive 500-hPa omega anomalies are of a similar strength (0.05 m.s−1 compared to 0.06 m.s−1). Over eastern southern Africa, the humidity and omega anomalies are of a similar magnitude and location in both experiments.

3.2.4. Inverse Idealized Indian Ocean SST Experiments: Experiment 4

[31] From the 3 idealized SST experiments, we can conclude that when HadAM3 is forced with an anomalous SST gradient in the Indian Ocean reminiscent of the SST anomalies independent of ENSO during 1974 and 1976, circulation features favorable to enhanced rainfall over southern Africa result. The model atmosphere's response is particularly strong when both the cold and the warm SST forcing is present, although a similar response results from the cold anomaly alone. The role of SST anomalies in forcing the atmosphere can only reasonably be answered by model experiments, but by their nature, idealized experiments impose a substantial separation between the model and the observed climate system. To be precise, we can conclude that HadAM3 responds to certain SST anomalies when the rest of the global oceans are constrained to climatological SSTs. The idealized SST anomalies are at least based on statistical analysis of the observed data and are in this sense realistic. However, the anomalies cover only a small portion of the Indian Ocean while the rest of the atmosphere is subjected to forcing by climatological SSTs, which is something that never happens in the real world.

[32] It is possible to design an experiment that relaxes this constraint. Accordingly, in experiment 4, sea surface temperatures are permitted to follow their observed values over the global oceans for the period of integration (1950 to 1990), except in that part of the Indian Ocean overlying the region of the SST anomalies specified in experiment 3. In this region, the climatological SST values are specified, with a smoothing relaxing the SSTs to the observed (historical) values at the edges of the region. Experiment 4 thus has varying SSTs everywhere except for the region for which the model was shown to be sensitive in experiment 3.

[33] To evaluate the experiment, we are looking for a reduction, relative to a control experiment, in the variance of the interannual variability of the anomalous 700 hPa easterly moisture flux over the SW Indian Ocean between 10–15°S and 40–60°E. Moisture flux in this area has previously been found to control the strength and frequency of TTCBs over southern Africa [Todd and Washington, 1999]. We chose this criterion since it is a characteristic response of HadAM3 to the SSTs in experiments 1 and 3. Note that the control experiment in this case is necessarily different from that in experiments 1 through 3 and is an integration from 1950 to 1990 with historical SSTs over the entire global oceans.

[34] The results of experiment 4 show that the variance of 700 hPa moisture flux in the control run is 3.5 times the variance of the experiment over the SW Indian Ocean box. Over southern Africa itself south of 20°S, moisture flux is 4 times higher in the control run. Experiment 4 therefore confirms, in an inverse sense, the sensitivity of the model atmosphere to SST anomalies in the Mascarene region. However, it does this in a way that is less constrained that experiments 1 and 3.

4. Summary and Discussion

4.1. Summary of Observed and Model Results

[35] From the analysis of observed southern African summer rainfall and SSTs, a clear pattern of SW Indian Ocean SSTs linearly independent from ENSO characterized by warm anomalies in the subtropical SW Indian Ocean and cool anomalies in the northern SW Indian Ocean was associated with the two wettest years of the twentieth century. It is likely that the effects of strong La Nina events in 1974 and 1976 were additive to effects of the SW Indian Ocean SSTs. From a series of multimodel experiments, we have shown that global sea surface temperatures were capable of generating anomalously wet conditions over southern Africa in those years. Additional experiments with HadAM3 indicate that an idealized pattern of SSTs which isolate the role of anomalies in the Indian Ocean, based on the observed SSTs with the ENSO signal linearly removed, is capable of forcing a model response conducive to rainfall. ANOVA testing indicates that approximately 40% of the significant circulation response (700 hPa moisture flux over the SW Indian Ocean box and over eastern southern Africa) is due to the anomalous SST forcing (experiments 1 and 3), a figure that is in broad agreement with studies pointing to the importance of tropical SSTs in atmospheric variability [e.g., Rowell, 1998; Washington, 2000].

[36] The idealized model experiments show that significant circulation changes resulting from tropical SWIO cooling (experiment 3) are greater in magnitude and spatial extent than the more localized responses resulting from the subtropical SWIO warming (experiment 2). However, the largest response comes from the combination of cool and warm forcing (experiment 1). In both experiments 1 and 3, the model generates a very similar atmospheric response to reanalysis fields which have been statistically treated to emulate the idealized experiment (Figures 13 and 14) by removal of the ENSO signal through regression. Specifically, wet years for these fields are characterized by an anomalous anticyclonic circulation in the Mascarene region, which drives an anomalous low-level easterly moisture flux along 10–20°S into eastern southern Africa resulting in enhanced moist convective uplift, conducive to enhanced rainfall, over large part of southern Africa. Statistically significant 700-hPa humidity and 500-hPa omega in both the idealized experiments and the statistically treated reanalysis data (not shown) extend from eastern southern Africa into the Agulhas region in a TTCB-like structure. The similarity between the reanalysis field and the model experiments is striking, notwithstanding the key caveats of atmosphere-only model experiments in which ocean-atmosphere coupling, known to be crucial in this region [Spencer, 2002], is ruled out. This paper emphasizes the role that the Indian Ocean plays in climate and is part of the growing evidence of the widespread impact the ocean basin has, for example on tropical cyclones [Xie et al., 2002], Sahel rainfall [Giannini et al., 2003], west Pacific rainfall [Annamalai et al., 2005] and globally [e.g., Saji and Yamagata, 2003].

Figure 13.

Correlation between JFM southern African rainfall and JFM SLP 1959–1998. Shading indicates statistical significance at the 5% level. SLP data are from the NCEP/NCAR reanalysis data set. The JFM SOI signal has been linearly regressed from all data prior to the calculation of the correlations.

Figure 14.

Correlation between southern African rainfall and JFM 700-hPa moisture flux 1959–1998 calculated from the NCEP/NCAR reanalysis data set. Shading indicates statistical significance a t the 5% level. The JFM SOI signal has been linearly regressed from all data prior to the calculation of the correlations.

4.2. Interpretation of Model Response and Comparison With Analogous Experiments

[37] The circulation response evident in experiments 1 and 3 over the Mascarene region is largely baroclinic, with anticyclonic (cyclonic) circulation anomalies in the lower (upper) troposphere. This is associated with anomalous descent and a reduction in strength of the Indian Ocean tropical convection maximum. Taken together, the circulation changes over the Mascarene region and eastern southern Africa are consistent with an observed dipole in moist convective uplift over southern Africa and the SWIO. The importance of this dipole in relation to southern African rainfall variability has been highlighted in previous research [Jury et al., 1992, 1995, 1996; Mason and Jury, 1997].

[38] Similar results have emerged from other modeling studies. Goddard and Graham [1999], using ECHAM3 AGCM, find that decreased convective heating (due to the cooler SSTs) in the tropical SWIO during the austral summer of 1975/1976 induces the formation of an anomalous anticyclonic circulation to the east of Madagascar (centered at 20°S, 60°E). Consequent on this local reduction in convective rainfall, lower-tropospheric air remains moist and is transported by anomalously strong tropical easterlies over southern Africa. This results in anomalous moisture convergence and enhanced rainfall over eastern southern Africa. Using the Center for Ocean-Land-Atmosphere Studies (COLA) AGCM, Tennant [1996] imposed an idealized cold SST anomaly (with a minimum central value of −2.0°C) in the tropical SWIO (5°N–20°S, 50–90°E). In line with the responses to cooling in the SW Indian Ocean in HadAM3, Tennant [1996] found a local decrease in rainfall in the Mascarene region associated with increased SLP.

[39] On the basis of bandpass filtered observed data, Reason and Mulenga [1999] have shown that warmer SST in the southwest Indian Ocean tends to be associated with wetter conditions over east and central South Africa. Idealized experiments with the Melbourne University AGCM forced by warming in the southwest Indian Ocean produce wet conditions over eastern South Africa and neighboring regions. The circulation response is similar to, though much more vigorous than, the equivalent for experiment 2, suggesting that HadAM3 may not be as responsive to SSTs in this region. Some caution should therefore be exercised in the emphasis given to the cold anomaly by HadAM3 in this paper. Furthermore, Reason [2002] has assessed the sensitivity of the Melbourne AGCM to a SST dipole in the SW Indian Ocean. The dipole in this case is similar to Behera and Yamagata [2001] with the warm part of the dipole near southern Africa (as in our experiment 2) but with the cold anomaly west of Australia. The results of experiment 1 are similar to Reason [2002] except that the anticyclonic (cyclonic) anomaly is stronger (weaker). Since both the model used and the SST forcing is different, it is difficult to be precise about the cause of this reversal in emphasis of response but does suggest that there is merit in exploring the sensitivity of southern African climate to the Indian Ocean SSTs using a variety of climate models.

[40] Overall these circulation changes are characteristic of a reversed Gill-type atmospheric circulation response (rising motion in the region of the positive diabatic heating anomaly coincident with descent to the east, associated with an equatorially trapped Kelvin wave, and off equatorial descent to the west, associated with two stationary Rossby waves) to an off-equatorial SST forcing. Further work using simple linear baroclinic models [e.g., Webster, 1981] and aquaplanet models [e.g., Hoskins et al., 1999; Neale, 1999] has shown that positive diabatic heating and associated Gill-type circulation responses can result from positive SST forcing in equatorial regions. In experiments 1 and 3, the responses are largely the reverse of those proposed by Gill [1980]. Instead they relate to negative diabatic heating anomalies induced by the cold SST forcing resulting in a vorticity change in the subtropical anticyclone. Using convective rainfall rate as a proxy for diabatic heating, this is supported by significant negative rainfall anomalies (−5 mm.day−1) over the tropical SWIO in experiments 1 and 3. The reduction in moist convective uplift to the east of Madagascar for these experiments is probably due to tropical SWIO SSTs having been reduced below a critical convective threshold [e.g., Gadgil et al., 1984; Graham and Barnett, 1987; Waliser and Graham, 1993; Zhang, 1993; Lau et al., 1997]. The northward shift in the location of the 28°C isotherm to approximately 5°S for experiments 1 and 3 is in line with the observed SST changes in 1974 and 1976.

4.3. Potential Causes of Tropical SWIO SST Variability

[41] The idealized experiments 1 to 3 and the design of experiment 4 have taken the SST pattern as given. An important issue surrounds the mechanisms which could cause the SST variability in the SW Indian Ocean. ENSO forcing is known to be responsible for most of the interannual variability of tropical Indian Ocean SSTs [Cadet, 1985; Reverdin et al., 1986; Kiladis and Diaz, 1989; Hastenrath et al., 1993; Latif and Barnett, 1995; Nagai et al., 1995; Tourre and White, 1995, 1997; Chambers et al., 1999; Murtugudde and Busalacchi, 1999; Reason et al., 2000; Behera et al., 2000].

[42] The strong association between Indian Ocean SSTs and ENSO activity is exemplified by significant negative correlations with the SOI during the austral summer. These correlations are strongest when Indian Ocean SST anomalies lag changes in the eastern tropical Pacific by 3–4 months [Lanzante, 1996; Wallace et al., 1998; Klein et al., 1999; Venzke et al., 2000; Fauchereau et al., 2003]. Analyses of all strong events over the last century [Reason et al., 2000] suggests that La Nino initially responds more quickly than El Nino but by the mature phase, the El Nino anomalies tend to be larger.

[43] The statistical analysis of observed SSTs showed that removal of the ENSO signal still left spatially and temporally coherent SST anomalies in the SW Indian Ocean. It is therefore important to explore mechanisms independent from ENSO that could cause these SST patterns. The tropical SWIO northeast of Madagascar is a region of open-ocean upwelling [Hastenrath et al., 1993; Xie et al., 2002] where dynamics impose an important control on SST variability [Chambers et al., 1999; Murtugudde and Busalacchi, 1999; Xie et al., 2002]. This contrasts with other parts of the tropical Indian Ocean where SST anomalies are found to largely result from surface heat flux changes [Chambers et al., 1999; Klein et al., 1999]. Xie et al. [2002] suggests that approximately 50% of SST variability in the tropical SWIO is due to oceanic Rossby waves that propagate from the east. The importance of oceanic Rossby waves within Indian Ocean circulation variability has also been shown in a number of other studies [Perigaud and Delecluse, 1993; Masumoto and Meyers, 1998; White, 2000; Schouten et al., 2002]. This study has emphasized the importance of the tropical SST anomaly to the atmosphere and to the extreme climate of southern Africa. Importantly, the tropical SST anomalies originate from subsurface ocean dynamics and which then go on to induce a very robust response in the atmosphere.

[44] While the appearance of anomalous equatorial easterlies in the eastern tropical Indian Ocean is generally associated with remote Pacific El Niño forcing [Klein et al., 1999; Murtugudde and Busalacchi, 1999; Venzke et al., 2000], there has been less research on their causes during non-ENSO years. One possible explanation, put forward by Xie et al. [2002], is that the anomalous easterlies (and associated anomalous Rossby waves) may be generated by anomalous SST cooling off Sumatra.

[45] The leading mode of SST-southern Africa rainfall variability in the coupled climate model HadCM3 is revealed as an SST pattern remarkably similar to the idealized SSTs in experiment 1 and to the composite SST pattern for JFM 1974 and 1976 (not shown). This provides the opportunity for a detailed examination of SST evolution in the model. The availability of long integrations (1000 years) similarly provide the opportunity to assess whether these SST modes undergo multidecadal variability in the same way that climate change experiments with HadCM3 allow assessments of the stability of the SST mode under conditions of significant Indian Ocean warming and with a changing basic state in the atmosphere. This work is ongoing and will shed light on whether the wet years of 1974 and 1976 will have any counterparts in the twenty first century.


[46] We are grateful to the IRI for making the historical AGCM experiments available. Thanks to Mike Bithell for help in setting up HadAM3 on our linux network and to Simon Mason for suggesting the “inverse idealized experiment.” The work was part funded by NERC, UK and by the Tyndall Centre for Climate Change Research, UK, under the “CLOUD” project (grant T2:32).