INTERNATIONAL JOURNAL OF CLIMATOLOGY, VOL. 17, 235–265 (1997) INTERANNUAL VARIABILITY OF SOUTH-EASTERN AFRICAN SUMMER RAINFALL. PART 1: RELATIONSHIPS WITH AIR–SEA INTERACTION PROCESSES ALFREDO ROCHA Departamento de Fı́sica, Universidade de Aveiro, 3800 Aveiro, Portugal email:firstname.lastname@example.org AND IAN SIMMONDS School of Earth Sciences, University of Melbourne, Parkville, Victoria, 3052, Australia email: email@example.com Received 29 September 1995 Revised 28 June 1996 Accepted 4 July 1996 ABSTRACT This paper investigates the role that air–sea interaction processes may play in interannual variability of south-eastern African summer rainfall. The principal spatial modes of south-eastern African summer rainfall are first identified using principal component analysis. Four modes are retained. The most important mode of variability is found to represent rainfall variability over most of the domain, particularly in the regions to the south. The influence of ENSO (as measured by the SOI) on summer rainfall is investigated in detail for different SOI leads. The relationship is such that during the summer following the onset of an ENSO event, south-eastern Africa tends to experience dry conditions. Strongest relationships are found with the SOI leading rainfall by about 3 to 6 months. A second index, the Brandon–Marion Index (BMI) which is indicative of changes in the pressure field over the Indian Ocean correlates with rainfall better than the SOI. Strongest correlations are found when this index leads rainfall by about 1 to 3 months. More importantly, a partial correlation analysis reveals that the BMI influences rainfall independently of ENSO. Both the SOI and the BMI are potential predictors of summer rainfall. An investigation of rainfall associations with global SST anomalies reveals areas in the tropical Indian and Pacific Oceans that are linked with rainfall changes over the subcontinent. The relationship is such that warm anomalies tend to be followed by dry conditions over much of south-eastern Africa. Strongest relationships are found when SSTs lead the rainfall season by about 1 to 3 months. Well-defined atmospheric anomalies are identified during dry south-eastern African summers. These include, amongst others, anomalously warm tropospheric temperatures and marked low-level cyclonic circulation anomalies over the central Indian Ocean, which generate abnormally weak easterly winds along much of the south-eastern coast of Africa. These perturbations to the low-level flow divert moisture from the continent and result in precipitation decreases. An important and related finding is the fact that the SST–rainfall link over the Indian Ocean remains strong after the ENSO effects have been removed, suggesting that the atmospheric circulation anomalies observed over south-eastern Africa during dry summers, are linked mainly to SST anomalies over the Indian Ocean. This hypothesis will be tested in a companion paper through a series of GCM simulations. south-eastern Africa; principal component analysis; correlation; anomaly fields; air–sea interaction; summer rainfall; atmospheric circulation; Brandon–Marion Index. KEY WORDS: 1. INTRODUCTION Extremes of interannual climate variability can often create stresses in many aspects of human life. Drought has been recognized as a common occurrence in many parts of south-eastern Africa, and is characterized by a recurrent distribution in time and spatial coherence over large areas (Nicholson, 1986a,b). These can be exemplified by the severe droughts of 1982–83 and 1991–92, which spread famine and distress amongst most # CCC 0899-8418/97/030235-31 $17.50 1997 by the Royal Meteorological Society 236 A. ROCHA AND I. SIMMONDS south-eastern African countries. A large proportion of the subcontinent experiences a single rainy season centred on January, and its failure leaves the local population deprived of water for the rest of the year. Most of the research on the year-to-year variability of summer rainfall in south-eastern Africa has been carried out for limited areas, particularly South Africa. Some of those studies have concentrated on describing the temporal behaviour of drought and the associated pressure anomalies that develop over the region and surrounding oceans (e.g. Rubin, 1956; Miron and Tyson, 1984; Tyson, 1984). During the last decade or so a considerable amount of research has been undertaken seeking non-local causes of south-eastern African drought. Particular attention has been given to the influence of the El Niño–Southern Oscillation (ENSO) phenomenon on summer rainfall (e.g. Dyer, 1979; Ropelewski and Halpert, 1987, 1989; Lindesay, 1988a; Kiladis and Diaz, 1989; Matarira, 1990). However, these studies have often been performed for restricted areas or using coarse rainfall networks. Also, lag relationships between ENSO and summer rainfall have rarely been investigated in detail. Evidence of an association between ENSO and rainfall in the region has been documented in many studies (Pittock, 1983; Nicholson and Entekhabi, 1986; Schulze, 1986; Lindesay, 1988a; van Heerden et al., 1988; Matarira, 1990). The relationship, which explains about 20 per cent of rainfall interannual variability (Lindesay, 1988b), is such that dry conditions tend to occur during ENSO episodes. Lindesay and Vogel (1990) analysed documentary rainfall data since 1820 and show this relationship to be stable. Using annual rainfall data for stations in the tropics, Stoeckenius (1981) has also identified an area in central South Africa that correlates with an annual SO index. In a global-scale analysis of monthly rainfall variability, Ropelewski and Halpert (1987) and Kiladis and Diaz (1989) have reported a tendency for drier than normal conditions to occur over south-eastern Africa during the summer following an ENSO event. Janowiak (1988) confirms the ENSO–rainfall link for a region immediately to the north of the summer rainfall area, in equatorial eastern Africa, which exhibits concurrent rainfall anomalies of opposite sign, that is, wet conditions during ENSO events. This dipole pattern between south-eastern and eastern equatorial Africa is shown in Figure 1 by the two Figure 1. The two core regions represent ENSO related interannual rainfall anomalies as obtained by Repelewski and Halpert (1987). The signs represent those of rainfall anomalies during ENSO events. Stars represent rainfall stations used in this study. The crosses represent the locations of St Brandon and Marion Island, used to derive an index in section 2 SOUTH-EAST AFRICAN RAIN: PART 1 237 core areas of ENSO related interannual rainfall anomalies (reproduced from Ropelewski and Halpert, 1987) and is also present in the study performed by Nicholson and Entekhabi (1986). In this study the ENSO–rainfall link is investigated using a relatively fine rainfall network covering most of the summer rainfall region of south-eastern Africa. The most important modes of interannual rainfall variability are identified over the subcontinent using principal component analysis (PCA). Their association with an index of ENSO is analysed and contrasted with the spatial signature of ENSO over the region. Lag relationships up to 12 months are detailed in an attempt to evaluate the potential of ENSO as a predictor of south-eastern African summer rainfall. We will also examine the link between rainfall projected on to these modes of variability and a measure of the regional circulation, in an attempt to determine factors that are associated with south-eastern African rainfall, but which are independent of ENSO. Because ENSO is a consequence of air–sea interaction processes, a physical understanding of the ENSO– rainfall link might be best achieved if changes at the air–sea interface are investigated. Large-scale sea-surface temperature (SST) anomalies have the potential to generate imbalance in the heat-flux field, which in turn can set up anomalous atmospheric circulation and rainfall patterns (Horel, 1982; Rasmusson and Carpenter, 1982). They may persist for several months over large areas, and they appear as the most easily monitored potential perturbers of the climatic system. However, it is still unclear to what extent the SST anomalies are a causal factor as opposed to simply being a response to large-scale circulation changes that produce both SST and rainfall anomalies. Here we are concerned mainly with the identification of SST anomalies (not necessarily related to ENSO) that are associated with interannual variations in climate, in particular large-scale rainfall, in the summer rainfall regions of south-eastern Africa. We have carried out an observational study of the possible links between summer rainfall and world-wide SST anomalies, and related large-scale atmospheric features. Verification of the hypothesis suggested by this paper will be achieved in a subsequent paper through a modelling approach. Meanwhile, one should mention that this study is not concerned directly with the synoptic systems whereby the large-scale circulation changes generate rainfall anomalies. Rather, attention is focused upon the large-scale circulation adjustment to SST forcing and consequent rainfall changes. The main motivation behind the research to be carried out here is the hope that knowledge emerging from it will be useful for future forecasting techniques of summer rainfall in the region. Emphasis will thus be placed on lead–lag relationships. 2. DATA AND ANALYSIS TECHNIQUES 2.1. Rainfall data Monthly rainfall data for the summer (November to March) for the period 1951–1989 were obtained from a number of different sources (Lindesay, pers. comm.; Mulenga, pers. comm.; Instituto Nacional de Meteorologia (Mozambique); ESSA, 1967; NCDC, 1971 and later years). Owing to the highly variable nature in time and space of monthly rainfall totals, it proved to be difficult to check for irregularities in the data. However, a few simple quality controls were applied. The final set comprised 85 stations covering an approximate area of 2 800 000 km2, representing an average density of about 10 stations per 330 000 km2. The location of these stations is displayed in Figure 1. For each station, summer seasonal totals were subsequently computed. The data were tested for normality by computing the skewness and kurtosis quantities. Both the monthly and seasonal data showed distributions far from normal and, therefore, several of the most common transformations were applied to these data, namely square-root, cube-root, logarithmic, and natural logarithmic transformations. After a careful examination of the transformed data sets we decided to adopt the square-root transformation, the rainfall distribution of which assumed a near-normal form. This transformation has been applied elsewhere to monthly rainfall amounts (e.g. Wright, 1974; Whetton, 1988). Next, the seasonal cycle was removed from data by subtracting each monthly (and seasonal) value from its respective monthly (and seasonal) average calculated for the period 1951–1989. These anomaly data were then standardized by the respective monthly (and seasonal) standard deviation, resulting in time series of average zero and unit standard deviation. This procedure facilitates the comparison of stations with different rainfall variability. 238 A. ROCHA AND I. SIMMONDS Table I. ENSO and anti-ENSO events from 1950 to 1989 (Events up to 1985 were obtained from van Loon and Shea (1985) and those from 1986 to 1989 from Kiladis and Diaz (1989)) Warm events (ENSO) Cold events (anti-ENSO) 1951–1952 1953–1954 1954–1955 1957–1958 1963–1964 1964–1965 1965–1966 1966–1967 1969–1970 1970–1971 1972–1973 1973–1974 1976–1977 1978–1979 1982–1983 1986–1987 1988–1989 2.2. The Southern Oscillation Index As an index of ENSO we use in this study the Southern Oscillation Index (SOI) as derived by the Climate Analysis Center (1986). This SOI is the normalized difference between the normalized mean monthly mean sealevel pressures at Tahiti and Darwin. Monthly values of the SOI from 1950 to 1989 were obtained from the monthly issues of the Climate Analysis Center (1983 and later years). Next, all data were filtered using a 3-month moving average. As shown by Trenberth (1984), such a filter can increase the SOI signal-to-noise ratio (mostly caused by small-scale disturbances of the circulation) from 1144, for monthly data, to 1197 without losing the variance on the ENSO time-scales. However, one is aware that such a procedure reduces the effective number of degrees of freedom as explained by Trenberth (1984). Table I displays ENSO and anti-ENSO events that have occurred since 1950, as classified by van Loon and Shea (1985) (up to 1985) and Kiladis and Diaz (1989) (from 1986 to 1989). 2.3. A circulation index over the Indian Ocean A number of studies have related dry and wet conditions in southern Africa with the surface pressure anomaly field over the region and neighbouring oceans (e.g. Tyson, 1981; Miron and Tyson, 1984; Matarira, 1990). Most of these studies have obtained results that are consistent, in that, at interannual time-scales, during wet years lower than normal pressure occurs over southern Africa, with the reverse in dry years. However, the structure of the wet and dry composite pressure anomalies over the adjacent oceans is not so clear. Tyson (1981) and Miron and Tyson (1984) have reported that during wet summers pressure is abnormally high to the south-west over the South Atlantic, and to a lesser extent over the Indian Ocean to the south-east. Using summer 1000 hPa pressure anomaly data, Matarira (1990) has shown that during dry south-eastern African years positive pressure anomalies occur over the land, whereas the surrounding oceans experience lower than normal pressures, with the exception of an area centred approximately at Marion Island (46 53 S, 37 52 E) where anomalies are strongly positive. This seems to indicate that the mid-latitude trough associated with the standing waves 1 and 3 (Tyson, 1981), which is normally located at about 30 E, shifts eastwards. The trough and its coupling with tropical easterly waves are 0 0 SOUTH-EAST AFRICAN RAIN: PART 1 239 related to major cloud bands, which usually form over the subcontinent in summer (Kuhnel, 1989). These cloud bands are one of the most important rain-bringing systems in southern Africa (Harrison, 1986). To gain an insight into the importance of pressure changes over the adjacent oceans and how they relate to south-eastern African rainfall, a circulation index was constructed. Monthly mean sea-level pressure data for two stations in the Indian Ocean were obtained for the period 1955–1988. These stations, Marion Island and St Brandon (16 27 S, 59 37 E) (represented by the crosses in Figure 1), are located in two ocean areas of Matarira’s pressure composite map corresponding to dry summers (his Figure 4(c)), which show high positive and negative pressure anomalies, respectively. Monthly normalized pressure difference time series between St Brandon and Marion Island were constructed in the same way as the SOI was using data for Tahiti and Darwin (monthly values of the BMI are presented in the Appendix). This index, which we name the Brandon–Marion Index (BMI), can be seen as a measure of changes in the position of the trough. The BMI will assume negative values when the trough shifts eastwards. Note that Harrison (1983) has shown that during dry South African summers the cloud bands associated with the trough are located more to the east towards the Indian Ocean. As for the SOI, this time series was filtered using a 3-month moving average in order to eliminate high frequency oscillations. 0 0 2.4. Ocean surface data Monthly averages of SST, mean sea-level pressure (MSLP), surface specific humidity and wind were obtained from the Comprehensive Ocean–Atmosphere Data Set (COADS) (Slutz et al., 1985; Woodruff et al., 1987). The data were obtained on a 2 2 grid over the world oceans for the period 1946–1987. Here we use the ‘trimmed’ subset of COADS and no extra quality control has been performed on these data. To improve their spatial coverage the data were subject to the same preprocessing described by Simmonds and Rocha (1991). Sea-surface temperature trends were found in the data, particularly over the north and equatorial eastern Pacific Ocean, with typical values up to ÿ015 C and 015 C per century respectively, and, because this paper is concerned with interannual time-scales, the SST anomaly data set was detrended. In addition, a seasonal subset was created by averaging, for every grid-point, the monthly values into four groups; December, January, and February (DJF), March, April, and May (MAM), June, July, and August (JJA), and September, October, and Novemer (SON). 6 2.5. The Southern Hemisphere atmospheric data set In this study the Southern Hemisphere data analyses prepared by the World Meteorological Centre (WMC) in Melbourne, Australia, are also used. These data consist of daily analyses at 0000 UTC produced since 1972 and are derived from station, buoy, ship and satellite data. In this paper we use MSLP, wind temperature, and water vapour mixing ratio at various pressure levels. The data base and analysis scheme used are described by Le Marshall et al. (1985) and Guymer (1986). Temporal and spatial inhomogneities of the data set have been mentioned by Trenberth (1979) and Guymer and Le Marshall (1980), amongst others. A comparison between the monthly climatologies of the WMC analyses and of a similar daily chart series developed by the South African Weather Bureau (the NOTOS chart series) for the period 1951–1962 (Taljaard et al., 1969) is presented by Le Marshall et al., (1985). The reliability of the monthly mean sea-level pressure of the WMC data set has been assessed by Jones (1991). The daily analyses from June 1972 to December 1984 were used to compute monthly means for every month and year. Owing to its relatively short coverage (12 years) it proved difficult to inspect the data for trends. 2.6. Correlation analysis In this study we use the Pearson’s correlation coefficient (r) as defined by Spiegel (1988), with significane assessed using Student’s t-test. In all cases a two-tailed, 5 per cent significance level is adopted unless indicated otherwise. It should be mentioned that this significance test assumes normally distributed populations for both variables. As described above, the frequencey distributions of all data sets were checked for normality. Partial 240 A. ROCHA AND I. SIMMONDS correlation is also applied to some variables in order to evaluate how a relationship between two of them is independent of a third (Spiegel, 1988). The reduction of the number of degrees of freedom due to autocorrelation can be a problem in correlation analysis. How this is determined is not entirely clear, particularly with relatively short data records. Greenhut (1979) presents a method (following Davis, 1976) to calculate the effective number of degrees of freedom. It should be emphasized that only when both time series display an autoregressive structure, is there a reduction of the degrees of freedom (see equation (2) of Greenhut). Rainfall is a variable in all correlations computed in this paper. We have computed the power spectrum of summer rainfall in the regions. All peaks were not significantly different from white noise processes. So, even if other variables possess autoregressive structure, and it is known that the SOI does (Trenberth and Shea, 1987), the persistence of rainfall anomalies are sufficiently weak to not be of great concern in our analysis. We emphasize that in this paper, data of all time series used to calculate correlations are separated by 12 months, so memory in sequential months does not affect the degrees of freedom. 2.7. Principal component analysis Principal component analysis is a statistical technique that has the objective of identifying variability patterns in a data set. It has been applied in meteorology to fluctuations in mean sea-level pressure (e.g. Kutzbach, 1967; Trenberth and Paolino, 1981), geopotential height (e.g. Craddock and Flood, 1969; Horel, 1981), temperature (Diaz and Fulbright, 1981), and precipitation (Dyer, 1975), amongst other variables. In most studies, as here, the principal component (PC) patterns are spatial fields and their coefficient series represent their amplitude in time. This usually is referred to as an S-mode PCA (Richman, 1986). Principal component analysis was performed using the correlation matrix of the standardized rainfall data. In order to obtain the PCs that best represent the rainfall data, Varimax rotation was applied and the resulting PCs compared with those before the rotation. We also performed Oblimin rotation but its spatial patterns were similar to those of Varimax. 3. PRINCIPAL COMPONENTS OF SUMMER RAINFALL In order to investigate relationships between large-scale rainfall changes and other atmospheric and oceanic variables it is important that the most important modes of rainfall variability be identified. To achieve this we carried out PCA on the monthly summer rainfall data. Gutman’s cut-off criterion of the eigenvalue series was adopted. By using this method, all eigenvalues smaller than unity were discarded. The same rule has been applied in numerous meteorological studies (e.g. Horel, 1981; ). By applying this truncation rule, the first 21 PCs are retained. These PCs were then rotated using the Varimax and Oblimin methods. The latter solution was not considered further because it gave results similar to the former. Table II displays the percentage of the total variance associated with eigenvalues of each of the 21 unrotated and rotated PCs. Figure 2 shows the sampling error for each of the first 15 eigenvalues of the unrotated solution as calculated using the method of North et al., (1982). The first four eigenvalues are well separated from each other. Eigenvalue pair four and five and pair five and six, although passing this cut-off criterion, have spacings comparable to their sampling errors. All the remaining eigenvalues form degenerate multiplets and can be discarded according to this criterion. One should note the differences between cut-off rules. Whereas Gutman’s truncation criterion discards eigenvalues beyond 21, the North et al., rule retains only the first four eigenvalues. We also performed rotation only on the first four PCs, but the resulting spatial patterns (not shown) were to a great extent present on the 21 PC rotation solution. 3.1. Principal component spatial patterns The spatial loadings of the first four rotated PCs are displayed in Figure 3. Each of these PCs have their highest loadings located in one particular region. Principal component 1 represents a large-scale pattern of rainfall anomalies located just south of the mean ITCZ summer position. It coincides broadly with the first PC of the PCA performed by Janowiak (1988) on African rainfall for the DJFM season, and with the south-eastern African rainfall anomaly type 4 reported by Nicholson (1986a). Principal component 2 lies mostly on the northern parts of 241 SOUTH-EAST AFRICAN RAIN: PART 1 Table II. The percentage of total variance associated with eigenvalues of each of the first 21 unrotated and rotated PCs PC 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Sum Unrotated Rotated 2417 919 511 319 313 217 214 212 211 119 118 117 117 116 115 114 114 114 114 112 112 7416 1616 612 510 418 416 312 312 219 219 219 218 215 213 212 210 119 118 118 118 116 116 7416 the Drakensberg Ranges and displays a similar pattern to that of Janowiak’s PC 3, south of 15 S. Principal component 3 is located in regions with elevations between 500 and 1000 m. Principal component 4 occupies the gap between PC 1 and PC 2, slightly overlapping the latter. In order to verify how well the PCs represent the underlying interstation correlation fields, the method suggested by Wigley et al. (1984) and used by Richman and Lamb (1985) was adopted. For each PC, the station with the highest loading is identified. Point-correlations between rainfall time series (same data as used to compute the correlation matrix) at this and all the remaining stations were performed. The resultant spatial correlation field was mapped and compared with the loadings of the PC. This procedure was carried out for all unrotated and rotated PC pairs. A quantitative measure of the match between these two types of maps, the loadings and the point-correlations, is obtained by calculating the congruence coefficient (Richman and Lamb, 1985). The difference between the correlation and the congruence coefficients is that the latter does not remove the means of the two variables. Therefore, it is not only a measure of pattern similarity, such as the correlation coefficient, but also of magnitude similarity. In sharp contrast with the unrotated PCs, almost all rotated PCs yield high congruence coefficients with a mean value of 0179, against 0109 for the unrotated modes over all 21 PCs. It is believed that the spatial orthogonality constraint of the unrotated solution does not allow PCs to represent the clusters present in the data, partly occurring due the convective nature of rainfall in the region. On the other hand, rotated PCs are able to isolate these clusters successfully. Hereafter only rotated PCs will be considered. The main purpose here is to associate a region with each of the four PCs, and construct a respective time series. Therefore, we shall not use the time series of amplitudes (i.e. scores) to represent rainfall changes in these regions for the following reasons. The PC scores are a measure of the strength of PC spatial patterns (i.e. loadings) in time and represent an average (over the whole area) of standardized station rainfall weighted by the respective station loading (this latter normalized by the respective eigenvalue). Although they are dominated by stations with high loadings, in a particular year, if these stations have low rainfall and low-loading stations have high rainfall, the respective PC score is not as dominated by the stations with high loadings as it otherwise would be. For this reason, regional rainfall time series were constructed based solely on the stations with the highest loadings of 242 A. ROCHA AND I. SIMMONDS Figure 2. The percentage of the total variance associated with each of the first 15 eigenvalues (unrotated solution) of south-eastern African summer rainfall. Bars denote the eigenvalue sampling error calculated using the method of North et al., (1982) each PC. These time series were obtained by considering only the area inside the 015 isopleth of every PC. The 015 isopleth level was chosen for two reasons. Firstly, loadings are correlations and can, therefore, be subject to statistical significance testing usually performed on the correlation coefficient. The critical correlation coefficient significantly different from zero at the 1 per cent significance level is 0118, according to Student’s t-test. The 015 loading isopleth is, thus, highly significant. Secondly, by using this isopleth level little overlapping occurs amongst the PCs, which means that regionalization can be accomplished without ambiguity. This method has been applied by many authors (e.g. Walsh et al., 1982; Ehrendorfer, 1987; White et al., 1991). Examples of alternative methods would be to assign each station to the PC to which they best relate (e.g. Karl and Koscielny, 1982) or to consider a given isopleth level of the point-correlation maps (e.g. Mallants and Feyen, 1990). Next, rainfall time series were obtained for each of the four regions by averaging the data from all stations within the region. The PC spatial pattern within each region was taken into account by weighting each station data by the respective PC loading. The interregion correlations of summer rainfall were computed and are displayed in Table III. Region 1 correlates positively with the other three regions but most highly with region 4. This may, despite the convective nature of rainfall in the area, indicate the importance of the large-scale in determining summer rainfall. Regions 2 and 4 correlate negatively with region 3. This north–south dipole is consistent with the south-eastern African anomaly types 5 and 6 reported by Nicholson (1986a) and with PC 1 and PC 2 of Janowiak (1988). 243 SOUTH-EAST AFRICAN RAIN: PART 1 Figure 3. Spatial loadings of the first four rotated PCs of south-eastern African summer rainfall. Isopleth levels are ÿ017, ÿ015, ÿ012, 010, 012, 017. Areas with loadings greater than 015 are stippled Table III. Cross-correlations of summer rainfall between the four most important rainfall regions. Italic numbers represent correlations different from zero at the 1 per cent significance level Region 2 Region 1 Region 2 Region 3 0144 Region 3 0139 0 34 ÿ 1 Region 4 0177 0168 ÿ0103 4. RELATIONSHIPS BETWEEN ATMOSPHERIC INDICES AND RAINFALL In this section we examine the associations between the SOI and the BMI, and rainfall. We consider the cases where the indices lead rainfall by up to 12 months and, for convenience, leads and lags are referred relative to January rainfall (the rainfall season’s middle month). 244 A. ROCHA AND I. SIMMONDS 4.1. SOI–rainfall associations Figure 4 displays the standardized rainfall anomaly time series for regions 1 to 4. Year refers to the calendar year at the end of the rainy season (e.g. 1983 means the rainy season starting in November 1982 and ending in March 1983). The ENSO and anti-ENSO years are denoted by ‘W’ and ‘C’, respectively. Inspection of these time series reflects what has been reported previously by some studies (e.g. Lindesay, 1988b; van Heerden et al., 1988; Matarira, 1990) in that most dry summers tend to coincide with ENSO and wet summers with anti-ENSO events, particularly for regions 1, 2, and 4. However, it is also evident that some dry summers did not occur during ENSO years. Indeed, the driest summer in regions 1 and 4 since 1950, namely 1968, was not associated with ENSO. The same applies for wet summers and anti-ENSO years. The general impression from Figure 4 is that, although rainfall and ENSO are related, the link is not strong. In order to quantify the relationship, summer rainfall in the four regions was correlated with the SOI, the index leading rainfall by up to 12 months. Figure 5 displays these lag correlations. Almost all correlations are positive, meaning that during ENSO years (negative SOI) the region tends to experience below normal rainfall. A feature of note is that correlations are strongest for region 2. For this region the correlation steadily increases with SOI lead decreasing, peaks at 4 months lead (r 0150 for September SOI), and decays thereafter. Correlations are significant (1 per cent for SOI leads from 3 to 6 months. There are no significant correlations for regions 1, 3, and 4. In order to investigate the temporal structure of the SOI signal, the same lag correlations stratified by rainfall month were computed. Once more, highest correlations are observed for region 2 (Figure 6). For this region, December rainfall correlates more highly with the SOI than rainfall in the other months. For this rainfall month correlation peaks for August and September SOI as it does for summer rainfall (see Figure 5). Despite the small, non-significant (1 per cent) correlations in regions 1, 3, and 4 (not shown), some features are worth mentioning. In general, the SOI correlates best with November, December, and March rainfall (the exception is for region 3, where January and February rainfall appear to dominate the relationship with the SOI). For region 3 (not shown) much of the correlations are negative. This might mean a change of signal, from positive to negative, in the rainfall–SOI relationship from the south to the north. 4.2. BMI–rainfall associations In this subsection, lag correlations between the BMI and rainfall are computed in the same way as for the SOI. Figure 7 presents the lag correlations between the BMI and summer rainfall, the index leading rainfall by up to 12 months. The BMI–rainfall relationship is strongest in regions 1 and 4 and weakest in region 3. The correlations steadily increase with decreasing BMI lead, peak when the index leads rainfall by 2 to 4 months (September to November BMI) and decay at 1 month lead. Region 2 displays a similar correlation curve when BMI leads, but values are lower than in regions 1 and 4, and only marginally significant (1 per cent) for BMI leads from 2 to 4 months. Summer rainfall in region 3 appears to have no relationship with the BMI at any of the considered leads and correlations are, therefore, not shown. A comparison with similar correlations with the SOI (see Figure 5) clearly highlights the greater rainfall forecast potential of the BMI for regions 1 to 4. As for the SOI, we computed lag correlations between monthly stratified rainfall and the BMI, for BMI leads up to 12 months. These are strongest for regions 1 (Figure 8), 2, and 4, but unlike with the SOI, highest values occur during the mid-summer months of December, January, and February and for BMI leads of 1 to 4 months. For region 3 correlations are weak and not significant. To unveil the spatial signature of the BMI, we correlated station summer rainfall with the BMI, the index leading rainfall by up to 12 months. One of the strongest of these correlation patterns, that between October BMI and rainfall, is displayed in Figure 9. Most of the area with correlations greater than 014 (significant at the 1 per cent significance level) fall within regions 1 and 4, in close agreement with the correlations between the BMI and seasonal rainfall in each of these regions. As defined, the BMI is negative when the pressure is abnormally low over the Indian Ocean, just east of Madagascar, and above normal at Marion Island where the mid-latitude trough is preferentially located in summer. Positive BMI–rainfall correlations indicate reduced precipitation over the subcontinent 2 to 4 months after the negative index anomalies take place or increased precipitation for reversed SOUTH-EAST AFRICAN RAIN: PART 1 245 Figure 4. Time series of summer rainfal index for regions 1, 2, 3, and 4. The SOI leads rainfall by up to 12 months. (ENSO and anti-ENSO events are denoted by ‘W’ and ‘C’, respectively) 246 A. ROCHA AND I. SIMMONDS Figure 5. Lag correlations between the SOI and summer rainfall for regions 1, 2, 3, and 4. The SOI leads rainfall by up to 12 months. Correlations greater than 0.4 are different from zero at the 1 per cent significance level pressure anomalies (i.e. abnormally strong trough). The first scenario seems to be consistent with an eastward shift of the preferred cloud band mean position towards the Indian Ocean during dry summers, as reported by Harrison (1986). Next, the BMI–rainfall associations were investigated with the effects of ENSO removed. A partial correlation analysis was performed between summer rainfall and the BMI with the effects of the SOI removed (the SOI simultaneous with the BMI). Figure 10 displays these partial correlations for regions 1 to 4 (the index leading rainfall by up to 12 months). Figure 10 is comparable with Figure 7. Monthly stratified partial correlations for region 1 are shown in Figure 11 which can also be contrasted with Figure 8. It is evident that Figures 7 and 8 are almost identical to Figures 10 and 11, respectively. One can therefore conclude that the BMI–rainfall association is independent of the SOI. We have computed lag correlations between monthly SOI and BMI, Figure 6. Lag correlations between monthly summer rainfall and monthly SOI for region 2. SOI months refer to the year in which the rainfall season starts. (The SOI leads rainfall except for November rainfall, and November and December SOI (zero lag and SOI lags by one month, respectively), and for December rainfall and December SOI (zero lag)). Correlations greater than 014 are different from zero at the 1 per cent significance level SOUTH-EAST AFRICAN RAIN: PART 1 247 Figure 7. Lag correlations between the BMI and summer rainfall for regions 1, 2, 3, and 4. The BMI leads rainfall by up to 12 months. Correlations greater than 014 are different from zero at the 1 per cent significance level Figure 8. As in Figure 6 but for the BMI and for region 1 for lags from ÿ72 months (SOI leading) to 72 months (SOI lagging). Correlations were all below 0115 for lags between ÿ12 and 12 months. Strongest correlations (0125) were found for the SOI leading the BMI by 50 months, but even these were barely significant if the reduction of the degrees of freedom due to memory in both time series is taken into account. 5. RELATIONSHIPS BETWEEN SEA–SURFACE TEMPERATURES AND RAINFALL The main objective here is to identify large-scale ocean areas that are related with summer rainfall in southeastern Africa. As with the atmospheric indices in section 4, our interest is twofold. Firstly, we hope to find SST anomalies and rainfall lag associations, with the SSTs leading rainfall, which have some forecast potential. Secondly, these observed relationships, together with observed atmospheric circulation changes characteristic of dry years, will be used to suggest a SST–rainfall physical link. 248 A. ROCHA AND I. SIMMONDS Figure 9. Lag correlations between station summer rainfall and the BMI for October. The BMI leads rainfall. Isopleth interval is 012. Areas with correlations greater than 014 are stippled Figure 10. Lag partial correlations between the BMI and summer rainfall for regions 1, 2, 3, and 4, with the effect of the SOI (simultaneous with the BMI) removed. The BMI leads rainfall by up to 12 months. Correlations greater than 014 are different from zero at the 1 per cent significance level 5.1. Global correlation analysis 6 The SST–rainfall link is investigated by correlating summer rainfall in regions 1 to 4 with world-wide SST anomalies in every 2 2 grid box. The seasonal (DJF, MAM, JJA, and SON) SST anomaly data sets are used here. Leads and lags are labelled, for convenience, with respect to the middle month of the rainfall season, January. We have correlated summer rainfall with SST anomalies leading by 12 months (DJF SSTs), 9 months (MAM SSTs), 6 months (JJA SSTs) 3 months (SON SSTs), at zero lag (DJF SSTs) and lagging by 3 months (MAM SSTs). Only a selection of the most important spatial correlation patterns obtained are presented here. Correlations were computed only if the SST grid-point in question had at least 10 common values (years) with SOUTH-EAST AFRICAN RAIN: PART 1 249 Figure 11. Lag partial correlations between monthly summer rainfall and monthly BMI for Region 1, with the effect of the SOI (simultaneous with the BMI) removed. BMI months refer to the year in which the rainfall season starts. (The BMI leads rainfall except for November rainfall and November and December BMI (zero lag and the BMI lags by 1 month, respectively), and for December rainfall and December BMI (zero lag)). Correlations greater than 014 are different from zero at the 1 per cent significance level rainfall. For clarity in the subsequent analysis, all spatial correlation patterns may occasionally be discussed only in terms of what they represent with respect to dry conditions. 5.1.1. Region 1. The spatial patterns of correlations between summer rainfall in region 1 and grid-point SST anomalies indicate that during JJA preceding dry summers (not shown), ocean waters tend to be abnormally warm in most of the tropical Indian and Pacific Oceans. The SSTs are above normal in the Atlantic Ocean and a core area of significant correlations is located in its south-eastern parts. This global correlation pattern is very similar to the SST anomalies during ENSO. Correlations are, however, not significantly different from zero over large areas of the ocean. Three months later, in SON (Figure 12), the correlation patterns are better defined and a large area of negative significant correlations (positive SST anomalies in dry years), with magnitudes in excess of 015, are observed in the central Indian Ocean. To the south and south-east of Madagascar, the surface of the ocean Figure 12. Lag correlations between summer rainfall in region 1 and SST anomalies in SON (three months lead). The isopleth interval is 012. Stippling indicates correlations significant at the 5 per cent significance level 250 A. ROCHA AND I. SIMMONDS tends to be cooler than normal during dry years (positive correlations). In the Pacific scattered pools of significant associations are also evident. The SSTs in the Atlantic are weakly correlated with rainfall. In DJF (the peak of the rainy season) (not shown), significant correlations cover a smaller area than in SON, particularly over the central Indian Ocean but the general patterns present in SON remains unchanged. An area of relatively high positive correlations (anomalous cold waters in dry summers) appears in the Indian Ocean just off the South African east coast. 5.1.2. Region 2. For region 2 these spatial lag correlation patterns are similar but the values and significance of the correlations are higher. During JJA (not shown) only in the Pacific Ocean are correlations significant over considerable areas. Three months later, during SON (Figure 13), there is a strong association between rainfall and SST anomalies in most of the equatorial Indian and Pacific Oceans, as evident from the large areas covered by stippling. Correlations in some of those areas reach 017. Warm oceans tend to be observed in the southern parts of the Indian Ocean, as in the correlations with region 1. The Atlantic SSTs show at this time no relation with summer rainfall. During DJF (not shown) the SST–rainfall link weakens in the Indian and Pacific but a zone of significant correlations appears in the eastern Atlantic. Also, the band of positive correlations in the south Indian Ocean during SON expands eastwards, even becoming significant in its core. Walker (1989) has investigated the association between Indian Ocean SSTs and rainfall in the south African summer region (its northern parts coincide with our region 2). As in this study, Walker found that, over the Indian Ocean, dry summers are related with warm SSTs north of about 20 S whereas to the south ocean waters tend to be cooler. 5.1.3. Regions 3 and 4. For region 3 the rainfall–SST link is much weaker than for regions 1 and 2. The shape of all patterns resembles that of an ENSO SST anomaly composite but correlations are in general weak and not significant. Rainfall in region 3 does not seem to be related to SST anomalies in any coherent way and are, therefore not shown. Results for region 4 are not shown also because they fall between those of regions 1 and 2. For all four rainfall regions, the SST–rainfall link is weaker for SST leads greater than 6 months (i.e. JJA SSTs) as well as for SSTs lagging rainfall by 3 months (MAM SSTs after the rainy season). Confirmation of the correlations obtained above was sought by constructing composite SST fields for the six driest (1967–1968, 1972–1973, 1982–1983, 1963–1964, 1969–1970, and 1959–1960) minus the six wettest (1981–1980, 1962–1963, 1952–1953, 1977–1978, 1954–1955 and 1973–1974) summers in regions 1 to 4. In general, these are consistent with the correlation analysis and are not shown here. Figure 13. Lag correlations between summer rainfall in region 2 and SST anomalies in SON (three months lead). The isopleth interval is 012. Stippling indicates correlations significant at the 5 per cent significance level SOUTH-EAST AFRICAN RAIN: PART 1 251 5.2 Partial correlation analysis: sea-surface temperature and rainfall associations in the absence of ENSO effects It has been shown above that summer rainfall in regions 1 and 2 correlates significantly with SST anomalies in vast areas of the tropical Indian and Pacific Oceans. Strongest associations were found with the SSTs leading rainfall by approximately 3 months (SON SSTs). One may question how much of that relationship is ENSOrelated because it is well known that ENSO encompasses most of the tropical oceans. A partial correlation analysis was performed between summer rainfall and grid-point SST anomalies after removing the effect of the SOI (taken synchronously with SST). Only regions 1 and 2 were considered here because rainfall in region 3 does not seem to be related in any coherent way with SSTs. The correlation spatial patterns obtained when SSTs lead rainfall by 3 months (SON SSTs) are shown in Figure 14 (a) for region 1 and Figure 14 (b) for region 2. For both regions, the area of significant correlations over the Indian Ocean was reduced when compared with that of the total correlation fields (see Figures 12 and 13 for regions 1 and 2, respectively). Differences are greatest for region 2. However, much of the SST–rainfall link is still present after the ENSO effects are removed. These findings are consistent with the weak associations between the SOI and summer rainfall in region 2. Over the Pacific, the influence of ENSO on the SST–rainfall total correlation fields is, as expected, considerably stronger. The strong negative associations present in the total correlations (see Figures 12 and 13) virtually disappear with the removal of ENSO effects. Similar changes occur in both oceans if rainfall is correlated with SSTs at zero lag (DJF SSTs) (not shown). The relative independence of the SST– rainfall association from ENSO over the Indian Ocean is also evident in the work of Walker (1989). In that study, after the removal of ENSO effects, South African summer rainfall still showed significant correlations with SSTs over vast coherent areas of the Indian Ocean. 6. RELATIONSHIPS BETWEEN ATMOSPHERIC VARIABLES AND RAINFALL In order to investigate the nature of the SST–rainfall link, atmospheric and circulation conditions typical of dry summers will now be identified. Some studies have reported on the atmospheric changes taking place over the subcontinent and nearby oceans that occur during dry south-eastern African summers. It has been shown in the previous sections that the SOI, BMI, and SST associations with rainfall in region 1 are representative of those in regions 2 and 4. Therefore, only atmospheric anomalies typical of dry conditions in region 1 will be identified. The COADS contains only data for the ocean surface and, therefore, the Australian Southern Hemisphere data set described above is used mainly in the following correlation analysis. However, due to the shorter period of the Southern Hemisphere data set compared with that of the COADS, compositing is based on the latter. 6.1. Mean sea-level pressure Correlations were performed between summer rainfall in region 1 and mean sea-level pressure at each gridpoint, with pressure leading rainfall by about 3 months (SON) and at zero lag (DJF). These two correlation fields are displayed in Figure 15. Three months before the rainy season, pressure changes over the land and the adjacent oceans in a coherent manner over large areas. Correlations indicate that before dry summers a band of positive pressure anomalies extends from the central and eastern south-eastern Africa to the south and south-western Indian Ocean. At the same time, negative pressure anomalies occur in the central Indian ocean, east of Madagascar. Most of the South Atlantic is dominated by abnormally low pressures. Correlations are barely significant at 3 months lead. At zero lag (DJF), the pressure anomaly pattern of SON becomes stronger. The spatial structure of the correlations stays the same only over region 1 and to the east and south of Madagascar. In other regions the pattern changes and the sign of correlations even reverse. In Figure 15, Marion Island and St Brandon, the stations used to derived the BMI index in section 2, are denoted by crosses. The correlations shown here reflect the strong relationship obtained between rainfall in region 1 and the BMI (with Marion Island in phase and St Brandon out of phase with pressure changes over the subcontinent), thereby justifying the usage of the BMI as an indicator of rainfall changes over region 1. This 252 A. ROCHA AND I. SIMMONDS Figure 14. Partial correlations between summer rainfall and SST anomalies in SON with the SOI (simultaneous with the SSTs) being kept constant, for (a) region 1 and (b) region 2. The isopleth interval is 012. Stippling indicates correlations significant at a 5 per cent significance level pressure anomaly dipole (with centres near Madagascar and to its south) may also be related with fluctuations of the preferred position of the ridge associated with the standing waves 1 and 3 which are normally located over the South Atlantic and south-western Indian Oceans, respectively (Streten, 1973). The southern African cloud band (one of the most important rain-bringing systems in southern Africa during summer) forms ahead of the midlatitude trough associated with wave number 3 (Streten, 1973). The spatial correlation patterns of Figure 15 agree, when interpreted in terms of what they represent during day conditions, with the pressure composites constructed by Matarira (1990) for dry minus normal south-eastern African years (his figure 4(c). When superimposed on the climatological MSLP field, these anomalies reflect a weakening of the South Atlantic and Indian Oceans high-pressure cells (not so clear during SON) and, consequently, a reduction of the surface pressure gradient directed from the oceans to the subcontinent. Clearly, large-scale pressure changes take place 3 months before abnormally dry rainy seasons and intensify at zero lag with rainfall. Only during DJF are these anomalies significantly correlated with rainfall. Global analysis of atmospheric changes during ENSO years have shown that over southern Africa pressures tend to be abnormally high (e.g. van Loon and Madden, 1982; van Loon and Shea, 1987). An important feature of the low SOUTH-EAST AFRICAN RAIN: PART 1 253 Figure 15. Correlations between summer rainfall in region 1 and MSLP in (a) SON and (b) DJF. The isopleth interval is 011. Stippling indicates correlations significant at a 5 per cent significance level. The crosses represent the locations of St Brandon and Marion Island pressure anomaly east of Madagascar is that it is located just south of the positive SST anomalies indicated by the negative correlations of Figure 12. The surface pressure composite analysis performed by Cadet (1985) for the Indian Ocean shows that during SON of El Niño years, an area of positive anomalies is observed north-east of Madagascar, broadly coincident with the zone of positive correlations (negative anomalies in dry summers) of Figure 15. This may indicate an ENSO signal in the correlation fields shown here. As for the SSTs, MSLP composites were constructed. They agree well with the correlation maps but are not shown here. 254 A. ROCHA AND I. SIMMONDS 6.2. Winds As with pressure, correlations were computed between summer rainfall in region 1 and the zonal wind component at 850 hPa and 200 hPa (U850, U200) and the meridional wind component at 850 hPa (V850). Figures 16 and 17 display the spatial correlation fields for U850 and V850, respectively, during DJF. Maps for SON are similar but correlations are weaker (not displayed). They indicate that 3 months before dry summers, low tropospheric wind anomalies are south-easterly over much of south-eastern Africa south of about 20 S and north-easterly over south-western Africa. At zero lag (DJF), the spatial pattern of the meridional wind anomalies changes little but the zonal winds are now anomalous westerly in the southern parts. To the north, a zone of high positive significant correlations (easterly anomalies during dry summers) is located over central south-eastern Africa. When superimposed on the mean vector field, these correlations represent, particularly for DJF, an anomalous anticyclonic circulation which acts to weaken the thermal low pressure cell characteristic of the region during summer. At 200 hPa, the zonal wind correlates negatively with rainfall in SON (westerly anomalies in dry years) over most of south-eastern Africa (not shown). Strongest significant associations exist at about 15 S over northern Zambia, Mozambique, and Madagascar. At zero lag (DJF), correlations indicate that during dry years westerly anomalies persist over most of the domain south of 15–20 S. Westerly wind anomalies have been reported to occur over southern Africa during ENSO years (Arkin, 1982; Lindesay, 1988b) but the easterly wind anomalies implied here by the positive correlations at low latitudes, particularly over the Atlantic during DJF, are not present in the analyses of Arkin and Lindesay. Weaker easterlies at 200 hPa have been related to a lower frequency of easterly wave and cloud band formation over the subcontinent, resulting in below average rainfall over the summer rainfall region of South Africa (Harrison, 1983, 1986), the northern parts of which fall within our region 2. Westerly surface wind anomalies across much of the eastern coast of southern Africa have been reported by Pan and Oort (1983) to occur when SSTs are abnormally warm in the central equatorial Pacific (i.e. ENSO events). At the same time, westerly wind anomalies are observed at 200 hPa over most of the tropical Atlantic, Indian, and eastern Pacific Oceans. The composite analysis of Cadet (1985) also depicts surface westerly wind anomalies over the western Indian Ocean during ENSO. Figure 16. As in Figure 15 but for DJF zonal wind at 850 hPa SOUTH-EAST AFRICAN RAIN: PART 1 255 Figure 17. As in Figure 15 but for DJF meridional wind at 850 hPa 6.3. Temperature Figure 18 displays the correlation fields between rainfall in region 1 and DJF temperature at (a) 850 (T850) and (b) 500 hPa (T500). They indicate that during dry summers the low and middle troposphere tends to be abnormally warm over much of the tropics of our domain. Similar but weaker patterns were obtained for SON T850 and T500 (not shown). As with the parameters analysed before, the rainfall–temperature association is stronger during DJF, when significant correlations cover a larger area. This is particular evident for T850 during DJF, when significant correlations occur over most of south-eastern Africa. Warm low-level temperatures during dry summers may be a result of reduced cloud cover and a consequent increase in insolation. Lindesay (1988b) has also reported widespread low and middle tropospheric warming over southern Africa during ENSO years. Her results are consistent with the global studies performed by van Loon and Madden (1981) and Kiladis and Diaz (1989). Indeed, the studies of Horel and Wallace (1981) and Pan and Oort (1983) have shown that most of the troposphere is anomalously warm during ENSO, particularly during DJF. 6.4. Relative humidity A monthly relative humidity subset was calculated using monthly mixing ratio and temperature data for the 850 and 500 hPa levels. Correlation fields between DJF relative humidity and summer rainfall are shown in Figure 19 for (a) 850 hPa and (b) 500 hPa. They indicate that during dry years, relative humidity at 500 hPa is lower than normal (positive correlations) over region 1. At zero lag (DJF), correlations are strongly positive and significant over region 1 and strongly negative and significant over south-western Africa. Over the Indian Ocean, east of Madagascar, the relationship is negative (increased relative humidity during dry summers). There seems to be a spatially coherent relationship between the pressure and 500 hPa relative humidity anomalies over the continent and Indian Ocean during dry region 1 summers. Areas of positive MSLP anomalies (negative correlations in Figure 15(b) tend to coincide broadly with areas of reduced relative humidity (positive correlations). This is particularly evident during DJF when an alternating three pole pattern of positive and negative anomalies can be observed over the eastern Atlantic and south-eastern Africa (positive correlations) and western Indian Ocean east of Madagascar (negative correlations). 256 A. ROCHA AND I. SIMMONDS Figure 18. As in Figure 15 but for DJF temperature at (a) 850 hPa and (b) 500 hPa The following scenario is suggested to explain the observed relative humidity anomalies. During dry summers in region 1, pressure increases over the land weakening the thermal low. Weaker vertical motion associated with these pressure anomalies would not favour convection, resulting in decreased moisture condensation in the middle troposphere and, hence, reduced rainfall. Reversed conditions occur over south-western Africa and the western Indian Ocean. This scenario is consistent with the eastward shift of the southern African cloud band from SOUTH-EAST AFRICAN RAIN: PART 1 257 Figure 19. As in Figure 18 but for relative humidity the subcontinent to the western Indian Ocean, described by Harrison (1983) as occurring during dry South African summers. Anomalous features at 850 hPa are very similar to those at 500 hPa. Significance is only achieved in the mid-latitudes, particularly over the south-east Atlantic Ocean during DJF where correlations represent reduced relative humidity during region 1 dry summers. Inspection of low-level horizontal moisture flux changes during region 1 summers using the Southern Hemisphere data (not shown) indicates that maximum anomalous equatorward moisture divergence takes place around 40 S parallel near the Greenwich meridian 258 A. ROCHA AND I. SIMMONDS (where correlations imply a maximum reduction in the relative humidity, as shown in Figure 19(a)). The positive anomalies in relative humidity over south-west Africa are at least partly due to anomalous moisture convergence from the equatorial regions there. 6.5. Surface horizontal moisture flux To investigate the anomalous transport of moisture from the oceans to south-eastern Africa, the stationary component of the surface moisture flux was computed. These were calculated using monthly values of the surface winds and moisture mixing ratio obtained from the COADS. Although COADS provides a poor spatial coverage south of 40 S, the major moisture sources of the summer rainfall in the region originate in the tropical oceans, in particular over the Indian Ocean. Here we present the structure of the horizontal surface moisture fluxes calculated from the seasonal mean surface wind and moisture. As calculated, this represents only the ‘stationary’ part of the moisture flux. Moisture fluxes affected by the transient components may be important, particularly for southern coastal areas where frontal systems in the mid-latitude westerly circulation contribute to summer precipitation totals (Tyson, 1986). One believes, however, that the steady component is appropriate to represent the moisture changes associated with the large-scale circulation controls in the tropics (Chen, 1985). It is assumed here also that surface moisture transport is an appropriate representation of the moisture flux in the lower troposphere where most moisture is concentrated (Chen, 1985). To analyse how this moisture flux is related to rainfall, composites were constructed for the six driest minus the six wettest summers of region 1. As a reference for the subsequent analysis, the climatological stationary moisture flux is presented in Figure 20 for SON (a) and DJF (b). Figure 21 displays the moisture flux anomaly composites for SON (a) and DJF (b). Three months before the rainy season (SON) when SSTs are anomalously warm over most of the tropical Indian Ocean, a strong easterly moisture flux anomaly is observed emanating from the western flank of the Indian Ocean high-pressure cell towards the warm waters. Along the African eastern coast, a reduction of moisture transport inland is generally evident by the westerly anomalies (although weak). During DJF the basic flow over the Indian Ocean is quite different from that in SON (compare Figures 20(a) and 20(b)). Note that in DJF the ITCZ is well defined by the confluence zone of the north-east monsoonal winds and the south-east Trades at about 10 S. During spring (SON), the north-east monsoon is not yet established and the circulation near the east African coast is south-easterly. One should also notice that over the Indian Ocean the ITCZ encompasses a large area where the horizontal advection is rather weak, and therefore likely to allow heating anomalies associated with warm SSTs to propagate relatively fast in the vertical, as suggested by the theory of Webster (1981). The composite for DJF shows that a large and well organized cyclonic anomaly in the moisture flux field develops at the surface east of Madagascar, resulting in a reduction of moisture advection into the continent across the eastern coast. Along the Mozambique Channel, wind anomalies have a southerly component, which agrees with the correlation analysis performed earlier (see Figure 17) and with the wind anomalies typical of ENSO years reported by Lindesay (1988b). 8. DISCUSSION It has been shown above that summer rainfall in the central and south-eastern parts of southern Africa is modulated by ENSO (no relationships were found for the northern regions of our domain). Strongest relationships with the SOI were found with region 2, which encompasses most of northern South Africa. This confirms the link between rainfall in the summer rainfall region of South Africa (which coincides broadly with region 2 here) and the SOI reported in a number of studies (e.g. Schulze, 1986; Nicholson and Entekhabi, 1886; Ismail, 1987; Lindesay, 1988b; Matarira, 1990) and the moderate ENSO signal on rainfall in an area extending from southern Mozambique into Zimbabwe, noted by Matarira (1990). When monthly instead of seasonal rainfall is considered, the association is practically non-existent for January rainfall (at any SOI lead) whereas December rainfall yields the highest correlations with the SOI. A number of factors can be suggested for the instability of these correlations within the rainy season. January is the peak summer month and, by this time, the heat low and ITCZ are well established over the subcontinent and 259 SOUTH-EAST AFRICAN RAIN: PART 1 Figue 20. Climatological stationary surface moisture flux for (a) SON and (b) DJF. The longest vector corresponds to 0114 (Kg Kg 1)m s ÿ ÿ 1 embedded in a highly barotropic atmosphere (Lindesay, 1988b). During January, rainfall is predominantly of a convective nature and rainfall-bringing baroclinic systems play a less important role than at the beginning and end of summer. It is therefore possible that the higher spatial and temporal variability of rainfall during January, when compared with the other summer months (at most stations of region 2, highest rainfall interannual variability is found for January) may contribute to weaken the correlations with the SOI. It may also happen that 260 A. ROCHA AND I. SIMMONDS Figure 21. Composite of the stationary surface moisture for the six driest minus the six wettest region 1 summers for (a) SON and (b) DJF. The longest vector corresponds to 018 (Kg Kg 1)m71 in (a) and 0106 (Kg Kg 1)m s 1 in (b) ÿ ÿ ÿ ENSO modulates summer rainfall through a predominance of certain types of atmospheric systems embedded in the tropical(Lindesay, 1988b) or extratropical (van Heerden et al., 1988) regimes. The BMI correlates strongly with rainfall over the central regions of the summer rainfall area (regions 1 and 4 of this study). The relationship is such that steeper than normal MSLP gradients between the western and the 261 SOUTH-EAST AFRICAN RAIN: PART 1 southwestern Indian Ocean tend to be followed by positive rainfall anomalies, or less precipitation for weaker gradients. The spatial signature of the association is broadly coincident with region 1. It is orientated in the northwest–south-east direction which happens also to coincide with the preferred orientation of the southern African cloud band (Kuhnel, 1989). This cloud band is one of the most important rain-bringing systems in southern Africa (Harrison 1984a,b; Smith, 1985). The BMI could, thus, be considered in any long-range forecast scheme of summer rainfall over the central areas of the summer rainfall region. It has been shown by the partial correlation between rainfall and the BMI with the effects of the SOI removed that both indices are, to a great extent, independent. Because both indices correlate well with summer rainfall in regions 2 and 4, the potential predictability of rainfall is increased if both are used as predictors. The low correlations found between the BMI and the SOI for different lags reinforces this potential. Interannual rainfall variability over the summer rainfall region of south-eastern Africa south of about 16 S has been shown to be related to global SST patterns. Strongest associations are observed for SST leads of about 3 months and decay at zero lag with rainfall. The spatial pattern and the temporal evolution of these anomalies are reminiscent of a typical ENSO episode. However, a partial correlation analysis has revealed that, although the anomalies in the Pacific are ENSO related, over the central Indian Ocean warming is to a great extent independent of ENSO. In fact, a careful inspection of monthly global SST anomalies and region 1 rainfall interannual variability since 1950 shows that four (1959–1960, 1967–68, 1981–82 and 1983–84) of the eight driest summers not associated with ENSO events coincided with warming over the central Indian Ocean. The relative independence of the SST–rainfall association from ENSO over the Indian Ocean is also evident in the work of Walker (1989). In that study, after the removal of ENSO effects, South African summer rainfall still showed significant correlations with SSTs over vast coherent areas of the Indian Ocean. Similar positive SST anomalies were reported by Reverdin et al., (1986) as having occurred in the Indian Ocean during non-ENSO years. During dry region 1 summers, the atmosphere over the subcontinent and adjacent oceans undergoes major changes. A broad zone of positive MSLP anomalies are observed over the central and south-eastern parts of southern Africa extending south-eastwards into the Indian Ocean. At the same time, over much of the tropical south-eastern Atlantic and north-east of Madagascar, pressure is anomalously low. In the lower troposphere, meridional wind anomalies are predominantly southerly along the eastern coast but zonal wind changes display a more complex spatial structure, with westerly anomalies located south of 25 S and easterly anomalies to the north (see Figure 16). Over the Indian Ocean anomalies of the opposite sign are observed. The net effect of these changes over the land (an anticyclonic circulation anomaly), which represents a weakening of the low pressure cell, is consistent with the readjustment of the MSLP field. A spatial correlation analysis performed by Lindesay (1988) between simultaneous SOI and winds at 700 hPA during OND and JFM displays some of the features shown in Figures 16 and 17. This may indicate that part of the wind anomalies reported here are related to ENSO, despite the fact that rainfall data and season classification on which correlations were based are different in both studies. During dry years, temperatures throughout the lower and middle troposphere are abnormally high over the subcontinent and neighbouring oceans, a situation usually linked to ENSO, as a result of the westward expansion of the warm SSTs in the tropical Pacific and Indian Oceans. In the relative humidity field at 500 hPa, a stationary synoptic-scale wave is observed over south-eastern Africa such that an area of moisture deficit is located above region 1 and positive anomalies are observed to the west, over south-western Africa, and to the east over Madagascar and the western Indian Ocean. These anomalies are consistent with reduced condensation at 500 hPA over south-eastern Africa, and can be associated with a shift in the preferred cloud band position from the continent towards the eastern Indian Ocean, as reported by Harrison (1983). Coincident with the negative MSLP anomalies east of Madagascar, a stationary cyclonic circulation anomaly is observed (Figure 21). As a result, part of the moisture carried by the south-east Trades and north-east monsoon is diverted from its usual trajectory, and instead of entering the continent through the eastern coast, recurves towards the anomalous low pressure area. Over the Atlantic, minor changes take place in the low level circulation despite the relatively large SST anomalies observed during DJF. Circulation changes during two very dry region 1 summers (those of 1967–1968 and 1982–1983) were also investigated. Three months before and during the summers of both years, SST anomalies were positive over the Indian Ocean, but in the equatorial central and eastern Pacific Ocean SSTs were very different. In 1967–1968, 262 A. ROCHA AND I. SIMMONDS surface waters were abnormally cool, whereas during 1982–1983 positive SST anomalies occupied most of the region. During both summers, enhanced surface level moisture convergence took place over the central Indian Ocean, in close agreement with the moisture flux composites for dry minus wet summers. However, it was suggested by Harrison (1986b) and Lindesay (1988) that westerly wind anomalies at 200 hPa over southern Africa, as a result of changes in the Walker Circulation during ENSO events, contributes to summer rainfall deficits in the region. Indeed, during January 1983 westerly wind anomalies at 200hPa of the order of 10 m s 1 were observed over south-eastern Africa. ÿ 9. CONCLUDING REMARKS No work to date has considered interannual variability of summer rainfall over south-eastern Africa in detail, and for such a large area. Therefore, identification of its principal modes of variability has been accomplished in order to investigate the spatial signature of precipitation variability. This has been achieved through PCA, and a careful selection of the modes (unrotated versus Varimax rotated) that best represented the underlying variability of the data. It has been shown that the most important mode of rainfall variability (region 1) covers much of central and eastern parts of the summer rainfall region of southern Africa. The influence of ENSO (as measured by the SOI) on summer rainfall has been found to be important only over the southern parts of our domain, but even there the relationship is not strong. The association is such that during ENSO (negative SOI) dry conditions tend to prevail. To the north, the SOI influence weakens considerably, with the correlations changing sign over the northernmost areas. Moreover, strong associations are found when the SOI leads rainfall by about 4–5 months. An index of the Indian Ocean atmospheric circulation, the BMI, has been devised in this study which is independent of the SOI. The association of rainfall with the BMI has been found to be much stronger than with the SOI, and is such that weak mean sea-level pressure gradients between St Brandon (east of Madagascar) and Marion Island (south-east of Africa) tend to be followed by dry conditions over south-eastern Africa. Unlike the SOI, the BMI correlates with summer rainfall over a large area of the dominion, but strongest values are found for the central eastern regions. The relationship peaks when the BMI leads rainfall by approximately 2 to 4 months, and decays for shorter leads. The BMI–rainfall association remains unaltered when the SOI effects are removed, suggesting that over the western and south-western Indian Ocean, the atmospheric circulation is at least partly independent of ENSO. The SSTs over large areas of the tropical Indian and Pacific Oceans correlate significantly with rainfall in regions 1, 2, and 4, such that anomalously warm waters in those oceanic areas tend to coincide with dry conditions. Correlations for region 3 are weak and spatially ill-defined, as are, in general, those over the Atlantic Ocean. Strongest correlations have been found for regions 2 and 4 but their spatial structures are very similar to that for region 1, and are strongly reminiscent of a typical ENSO SST anomaly pattern. The association peaks when SSTs lead rainfall by about 3 months and decays thereafter. A partial correlation analysis where the ENSO effects have been removed has revealed that, whereas much of the relationship disappears over the Pacific, strong significant correlations are still present in the Indian Ocean. This further supports the finding of the BMI–rainfall, that air–sea processes over the Indian Ocean are, in fact, partly independent of ENSO. We have identified here anomalous atmospheric features that tend to occur during dry region 1 summers (in fact during most dry south-eastern African summers). These are detailed below for the peak of summer (DJF): (i) decreased MSLP takes place east of Madagascar whereas over south-east Africa positive MSLP anomalies are observed; (ii) north-westerly low-level wind anomalies prevail along much of the east coast of Africa; (iii) at 200 hPa the wind is anomalously westerly over south-eastern Africa; (iv) anomalously warm tropospheric temperatures occur over the subcontinent and adjacent tropical oceans; (v) relative humidity at 850 and 500 hPa decreases over south-east Africa but positive anomalies are present over south-west Africa and the South Atlantic Ocean; (vi) low-level moisture flux convergence weakens over south-eastern Africa but intensifies over the tropical Indian Ocean, particularly to the east of Madagascar. 263 SOUTH-EAST AFRICAN RAIN: PART 1 The hypothesis we have formed is that SST anomaly patterns identified as typical of dry south-eastern Africa summers generate dry conditions through the above-mentioned atmospheric anomalous features. This hypothesis will be tested through a series of modelling experiments in which the model’s atmosphere has been forced with spatial- and time-evolving SST anomalies characteristic of dry summers over south-eastern Africa. The results of these experiments are reported in Rocha and Simmonds (1997). APPENDIX Monthly values of the BMI from 1955 to 1988. Missing values are denoted by Year January February March 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 03 17 ÿ016 ÿ012 ÿ019 ÿ111 013 ÿ113 ÿ114 011 ÿ014 117 018 010 211 011 ÿ018 ÿ9919 ÿ011 ÿ9919 118 016 110 016 019 ÿ112 115 ÿ012 015 ÿ116 ÿ011 ÿ019 013 012 ÿ 1 ÿ 1 14 015 110 018 ÿ011 ÿ018 111 014 ÿ016 ÿ019 010 211 110 014 119 ÿ110 ÿ119 ÿ018 ÿ111 ÿ9919 014 016 012 114 ÿ016 ÿ011 ÿ014 ÿ011 ÿ016 ÿ9919 ÿ212 012 016 ÿ011 ÿ 1 13 010 ÿ012 ÿ011 ÿ110 ÿ115 212 ÿ018 ÿ113 ÿ012 016 ÿ014 012 ÿ019 016 ÿ112 ÿ015 014 ÿ011 ÿ9919 ÿ014 116 213 017 010 ÿ116 114 ÿ012 016 011 111 ÿ9919 ÿ011 ÿ9919 ÿ 1 April May June 011 115 ÿ011 ÿ016 ÿ114 017 018 ÿ9919 ÿ013 ÿ114 ÿ016 ÿ018 ÿ117 019 114 112 117 119 ÿ111 ÿ9919 015 013 ÿ016 017 ÿ011 111 ÿ111 ÿ111 ÿ012 ÿ014 ÿ014 017 ÿ115 010 02 210 ÿ015 019 ÿ117 ÿ016 ÿ014 ÿ012 ÿ014 ÿ019 017 112 ÿ115 110 ÿ011 ÿ011 013 014 011 ÿ9919 012 ÿ012 014 211 016 013 ÿ310 ÿ017 ÿ016 016 ÿ018 019 ÿ011 012 03 113 111 114 116 112 ÿ014 ÿ017 ÿ016 012 ÿ018 ÿ9919 ÿ012 014 ÿ012 ÿ016 012 ÿ310 ÿ014 ÿ9919 ÿ013 011 113 111 018 018 ÿ112 011 ÿ111 016 ÿ013 ÿ117 ÿ013 010 ÿ 1 ÿ 1 ÿ 9919 July August September October November December 05 012 110 ÿ214 014 ÿ011 ÿ110 015 011 ÿ113 ÿ013 014 ÿ110 212 015 117 ÿ112 114 ÿ115 ÿ015 013 112 010 ÿ018 019 019 ÿ116 ÿ018 ÿ012 014 012 014 010 014 ÿ 1 16 111 017 ÿ112 ÿ014 210 ÿ018 013 ÿ012 019 ÿ014 ÿ015 ÿ017 114 ÿ019 012 ÿ012 ÿ118 ÿ013 018 112 110 011 119 ÿ113 015 ÿ015 ÿ016 ÿ9919 016 ÿ011 ÿ116 016 ÿ9919 ÿ 1 10 02 110 ÿ016 ÿ212 010 015 017 010 012 016 013 ÿ015 113 ÿ112 ÿ012 016 014 017 012 ÿ110 ÿ111 111 019 212 112 ÿ014 ÿ211 012 ÿ110 2 018 ÿ9919 ÿ013 ÿ112 ÿ 1 ÿ 1 03 012 115 010 ÿ016 112 019 012 ÿ115 ÿ012 ÿ016 010 ÿ016 114 ÿ014 ÿ218 ÿ012 016 113 114 ÿ012 013 014 ÿ9919 015 018 ÿ011 ÿ112 ÿ110 ÿ016 117 ÿ016 ÿ116 011 ÿ 1 02 9919 116 012 ÿ115 ÿ018 ÿ016 ÿ014 ÿ018 210 013 ÿ018 ÿ014 ÿ013 010 016 ÿ015 ÿ116 112 118 110 012 111 210 011 019 ÿ110 ÿ111 ÿ016 ÿ015 ÿ9919 ÿ015 ÿ015 ÿ112 ÿ 1 ÿ 010 015 ÿ011 013 011 017 ÿ210 013 ÿ016 015 111 ÿ112 ÿ116 113 019 ÿ011 112 ÿ012 ÿ012 114 013 110 017 014 013 014 ÿ016 ÿ113 ÿ212 018 ÿ214 011 013 ÿ9919 REFERENCES Arkin, P. A. 1982. ‘The relationship between interannual variability in the 200 mb tropical wind field and the Southern Oscillation’, Mon. Wea. Rev., 110, 1393–1404. Cadet, D. L. 1985. ‘The Southern Oscillation over the Indian Ocean’, J. Climatol., 5, 189–212. Chen, T.-C. 1985. ‘Global water vapor flux and maintenance during FGGE’, Mon. Wea. Rev., 113, 1801–1819. Climate Analysis Center 1983. Climate Diagnostics Bulletin, NMC, National Weather Service, NOAA, Washington, DC. Climate Analysis Center 1986. Climate Diagnostics Bulletin No. 86/4, NMC, National Weather Service, NOAA, Washington, DC, 12 pp. Craddock, J. M. and Flood, C. R. 1969. ‘Eigenvectors for representing the 500 mb geopotential surface over the Northern Hemisphere’, Q. J. Meteorol. Soc., 95, 576–593. 264 A. ROCHA AND I. SIMMONDS Davis, R. e. 1976. ‘Predictability of sea surface temperature and sea level pressure anomalies over the North Pacific’, J. Phys. Oceanogr., 6, 249–266. Diaz, H. F. and Fulbright, D. C. 1981. ‘Eigenvector analysis of seasonal temperature, precipitation and synoptic-scale frequency over the contiguous United States. Part I: winter’, Mon. Wea. Rev., 109, 1267–1284. Dyer, T. G. J. 1975. ‘The assignment of rainfall stations into homogeneous groups: an application of principal component analysis’, Q. J. Meteorol. Soc., 101, 1005–1013. Dyer, T. G. J. 1979. ‘Rainfall along the east coast of southern Africa, the Southern Oscillation, and the latitude of the subtropical high pressure belt’, Q. J. Meteorol. Soc., 105, 445–451. Ehrendorfer, M. 1987. ‘A regionalization of Austria’s precipitation climate using principal component analysis’, J. Climatol., 7, 71–89. ESSA 1967. World Weather Records 1951–60. Africa, 5, Environmental Science Services Administration, Washington, DC, 545 pp. Greenhut, G. K. 1979. ‘Reply to Comments on ‘‘Correlations between rainfall and sea surface temperature during GATE’’ ’, J. Phys. Oceanogr., 9, 1298–1299. Guymer, L. B. 1986. ‘Procedures and concepts used in Southern Hemisphere analyses at WMC Melbourne’, Second International Conference on Southern Hemisphere Meteorology, American Meteorological Society, pp. 10–16. Guymer, L. B. and Le Marshall, J. F. 1980. ‘Impact of FGGE buoy data on Southern Hemisphere analysis’, Aust. Meteorol. Mag., 28, 19–42. Harrison, M. S. J. 1983. ‘The Southern Oscillation, zonal equatorial circulation cells and South African rainfall’, First International Conference on Southern Hemisphere Meteorology, American Meteorological Society, pp. 302–305. Harrison, M. S. J. 1984a. ‘A generalized classification of South African summer rain-bearing synoptic systems’, J. Climatol., 4, 547–560. Harrison, M. S. J. 1984b. ‘Comparison of rainfall time series over South Africa generated from real data and through principal component analysis’, J. Climatol., 4, 561–564. Harrison, M. S. J. 1986. A synoptic climatology of South African rainfall variations, Unpublished PhD thesis, University of the Witwatersrand, 341 pp. Horel, J. D. 1981. ‘A rotated principal component analysis of the interannual variability of the Northern Hemisphere 500 mb height field’, Mon. Wea. Rev., 109, 2080–2092. Horel, J. D. 1982. ‘On the annual cycle of the tropical Pacific atmosphere and ocean’, Mon. Wea. Rev., 110, 1863–1878. Horel, J. D. and Wallace, J. M. 1981. ‘Planetary-scale atmospheric phenomena associated with the Southern Oscillation’, Mon. Wea. Rev., 109, 813–829. Ismail, S. A. 1987. ‘Long-range seasonal rainfall forecast for Zimbabwe and its relation with El Niño/Southern Oscillation (ENSO)’, Theor. Appl. Climatol., 38, 93–102. Janowiak, J. E. 1988. ‘An investigation of interannual rainfall variability in Africa’, J. Climate, 1, 240–255. Jones, P. D. 1991. ‘Southern Hemisphere sea-level pressure data: an analysis and reconstruction back to 1951 and 1911’, Int. J. Climatol., 11, 585–607. Karl, T. R. and Koscielny, A. J. 1982. ‘Drought in the United States: 1895–1981’, J. Climatol., 2, 313–329. Kiladis, G. N. and Diaz, H. F. 1989. ‘Global climatic anomalies associated with extremes in the Southern Oscillation’, J. Climate, 2, 1069– 1090. Kuhnel, I. 1989. ‘Tropical-extratropical cloudband climatology’, Int. J. Climatol., 9, 441–463. Kutzbach, J. E. 1967. ‘Empirical eigenvectors of sea level pressure, surface temperature and precipitation complexes over North America’, J. Appl. Meteorol., 6, 791–802. Le Marshall, J. F., Kelly, G. A. M. and Karoly, D. 1985. ‘An atmospheric climatology of the Southern Hemisphere based on ten years of daily numerical analyses (1972–82): I. Overview’, Aust. Meteorol. Mag., 33, 65–85. Lindesay, J. A. 1988a. ‘South African rainfall, the Southern Oscillation and a Southern Hemisphere semi-annual cycle’, J. Climatol., 8, 17–30. Lindesay, J. A. 1988b. The Southern Oscillation and atmospheric circulation changes over southern Africa, Unpublished PhD thesis, University of the Witwatersrand, 284 pp. Lindesay, J. A. and Vogel, C. H. 1990. ‘Historical evidence for Southern Oscillation-southern African rainfall relationships’, Int. J. Climatol., 10, 679–689. Mallants, D. and Feyen, J. 1990. ‘Defining homogeneous precipitatoin regions by means of principal component analysis’, J. Appl. Meteorol., 29, 892–901. Matarira, C. H. 1990. ‘Drought over Zimbabwe in a regional and global context’, Int. J. Climatol., 10, 609–625. Miron, O. and Tyson, P. D. 1984. ‘Wet and dry conditions and pressure anomaly fields over South Africa and the adjacent oceans 1963–1979’, Mon. Wea. Rev., 112, 2127–2132. NCDC 1971. Monthly Climatic Data for the World, National Climatic Data Center, Asheville, USA. Nicholson, S. E. 1986a. ‘The nature of rainfall variability in Africa south of the Equator’, J. Climatol., 6, 515–530. Nicholson, S. E. 1986b. ‘The spatial coherence of African rainfall anomalies: interhemispheric teleconnections’, J. Clim. Appl. Meteorol., 25, 1365–1381. Nicholson, S. E. and Entekhabi, D. 1986. ‘The quasi-periodic behavior of rainfall variability in Africa and its relationship to the Southern Oscillation’, Arch. Meteorol. Geophys. Bioklimatol., Ser. A, 34, 311–348. North, G. R., Bell, T. L., Cahalan, R. F. and Moeng, F. J. 1982. ‘Sampling errors in the estimation of empirical orthogonal functions’, Mon. Wea. Rev., 110, 699–706. Pan, Y. H. and Oort, A. H. 1983. ‘Global climate variations connected with sea surface temperature anomalies in the eastern equatorial Pacific Ocean for the 1958–73 period’, Mon. Wea. Rev., 111, 1244–1258. Pittock, A. B. 1983. ‘Climatic teleconnections in the Southern Hemisphere’, First International Conference on Southern Hemispherre Meteorology, American Meteorological Society, pp. 283–286. Rasmusson, E. M. and Carpenter, T. H. 1982. ‘Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño’, Mon. Wea. Rev., 110, 354–384. Reverdin, G., Cadet, D. L. and Gutzler, D. 1986. ‘Interannual displacements of convection and surface circulation over the equatorial Indian Ocean’, Q. J. R. Meteorol. Soc., 112, 43–67. Richman, M. B. 1986. ‘Rotation of principal components’, J. Climatol., 6, 293–335. 265 SOUTH-EAST AFRICAN RAIN: PART 1 Richman, M. B. and Lamb, P. J. 1985. Climatic pattern analysis of three- and seven-day summer rainfall in the central United States: some methodological considerations and a regionalization’, J. Clim. Appl. Meteorol., 24, 1325–1343. Rocha, A. and Simmonds, I. 1997. ‘Interannual variability of south-eastern African summer rainfall. Part II: modelling the impact of seasurface temperatures on rainfall and circulation’, Int. J. Climatol., 17, 267–290. Ropelewski, C. F. and Halpert, M. S. 1987. ‘Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation’, Mon. Wea. Rev., 115, 1606–1626. Ropelewski, C. F. and Halpert, M. S. 1989. ‘Precipitation patterns associated with the high phase of the Southern Oscillation’, J. Climate, 2, 268–284. Rubin, M. J. 1956. ‘The associated precipitation and circulation patterns over southern Africa’, Notos, 5, 53–59. Schulze, G. C. 1986. ‘South African rainfall related to warm and cold events’, Second International Conference on Southern Hemisphere Meteorology, American Meteorological Society, pp. 465–467. Simmonds, I. and Rocha, A. 1991. ‘The association of Australian winter climate with ocean temperatures to the west’, J. Climate, 4, 1147– 1161. Slutz, R. J., Lubker, S. J., Hiscox, J. D., Woodruff, S. D., Jene, R. J., Joseph, D. H., Steurer, P. M. and Elms, J. D. 1985. Comprehensive Ocean-Atmosphere Data Set — Release 1, Climate Research Program, ERL, Boulder, Colorado, April 1985. Smith, A. V. 1985. ‘Studies of the effects of cold fronts during the rainy season in Zimbabwe’, Weather, 40, 198–203. Spiegel, M. R. 1988. Theory and Problems of Statistics, Schaum’s Outline Series, McGraw-Hill, 5th Printing, 359 pp. Stoeckenius, T. 1981. ‘Interannual variations of tropical precipitation patterns’, Mon. Wea. Rev., 109, 1233–1247. Streten, N. A. 1973. ‘Some characteristics of satellite-observed bands of persistent cloudiness over the Southern Hemisphere’, Mon. Wea. Rev., 101, 486–495. Taljaard, J. J., van Loon, H., Crutcher, H. L. and Jenne, R. L. 1969. Climate of the Upper Air: Southern Hemisphere, Vol. 1, NAVAIR 50-IC55, Chief of Naval Operations, Washington, DC, 135 pp. Trenberth, K. E. 1979. ‘Interannual variability of the 500 mb zonal mean flow in the Southern Hemisphere’, Mon. Wea. Rev., 107, 1515–1524. Trenberth, K. E. 1984. ‘Signal versus noise in the Southern Oscillation’, Mon. Wea. Rev., 112, 326–332. Trenberth, K. E. and Paolino, D. A. 1981. ‘Characteristic patterns of variability of sea level pressure in the Northern Hemisphere’, Mon. Wea. Rev., 109, 1169–1189. Trenberth, K. E. and Paolino, D. J. 1987. ‘On the evolution of the Southern Oscillation’, Mon. Wea. Rev., 115, 3078–3096. Tyson, P. D, 1981. ‘Atmospheric circulation variations and the occurrence of extended wet and dry spells over southern Africa’, J. Climatol., 1, 115–130. Tyson, P. D. 1984. ‘The atmospheric modulation of extended wet and dry spells over South Africa, 1958–78’, J. Climatol., 4, 621–635. Tyson, P. D. 1986. Climatic Change and Variability in Southern Africa, Oxford University Press, Cape town, 220 pp. Van Heerden, J., Terblanche, D. E. and Schulze, G. C. 1988. ‘The Southern Oscillation and South African summer rainfall’, J. Climatol., 8, 577–597. Van Loon, H. and Madden, R. A. 1981. ‘The Southern Oscillation. Part I: global associations with pressure and temperatures in northern winter’, Mon. Wea. Rev., 109, 1150–1162. Van Loon, H. and Shea, D. J. 1985. ‘The Southern Oscillation. Part IV: the precursors south of 15 to the extremes of the oscillation’, Mon. Wea. Rev., 113, 2063–2074. Van Loon, H. and Shea, D. J. 1987. ‘The Southern Oscillation. Part VI: anomalies of sea level pressure of the Southern Hemisphere and of Pacific sea surface temperature during the development of a warm event’, Mon. Wea. Rev., 115, 370–379. Walker, N. D. 1989. ‘Sea surface temperature–rainfall relationships and associated ocean–atmosphere coupling mechanisms in the southern Africa region’, CSIR Research Report 683, Stellenbosch, South Africa, 173 pp. Walsh, J. E., Richman, M, B. and Allen, D. W. 1982. ‘Spatial coherence of monthly precipitation in the United States’, Mon. Wea. Rev., 110, 272–286. Webster, P. J. 1981. ‘Mechanisms determining the atmospheric response to sea surface temperature anomalies’, J. Atmos. Sci., 38, 554–571. Whetton, P. H. 1988. ‘A synoptic climatological analysis of rainfall variability in south-eastern Australia’, J. Climatol., 8, 155–177. White, D., Richman, M. and Yarnal, B. 1991. ‘Climate regionalization and rotation of principal components’, Int. J. Climatol., 11, 1–25. Wigley, T. M., Lough, J. M. and Jones, P. D. 1984. ‘Spatial patterns of precipitation in England and Wales and a revised, homogeneous England and Wales precipitation series’, J. Climatol., 4, 1–25. Woodruff, S. D., Slutz, R. J., Jenne, R. L. and Steurer, P. M. 1987. ‘A comprehensive ocean–atmosphere data set’, Bull. Am. Meteorol. Soc., 68, 1239–1250. Wright, P. B. 1974. ‘Seasonal rainfall in southwestern Australia and the general circulation’, Mon. Wea. Rev., 102, 219–232.