Relations between ENSO and the South Atlantic SST modes and their effects on the South American rainfall

Authors


M. T. Kayano, Instituto Nacional de Pesquisas Espaciais, Centro de Previsão de Tempo e Estudos Climáticos, Avenida dos Astronautas, 1758, 12227-010 São José dos Campos, SP, Brazil. E-mail: mary.kayano@cptec.inpe.br

Abstract

This paper analyses the relations between the El Niño/Southern Oscillation (ENSO) mode in the tropical Pacific and the sea surface temperature (SST) modes in the South Atlantic for the 1900–2006 period. In the South Atlantic, two modes are analysed in more detail: the South Atlantic dipole (SAD) mode, with centres at (15°S, Greenwich longitude) and at (37.5°S, 25°W) and the southwestern South Atlantic (SWSA) mode, with centres at (30°S, 40°W) and at the southern midlatitudes. The ENSO and SAD modes are lagged or lead connected depending on the period of analysis. An El Niño (a La Niña) precedes by up to 6 months the establishment of a positive (negative) SAD mode during the 1920–1930 and 1940–1980 periods. Otherwise, a positive (negative) SAD mode precedes by up to 1 year the establishment of a La Niña (El Niño) during the 1975–2000 period. On the other hand, the SWSA is strongly driven by the ENSO. The effects of both the SAD and SWSA modes on the South American rainfall are also discussed on a seasonal basis. In general, the SAD (ENSO) mode has a weak influence on the ENSO-related (SAD-related) rainfall anomalies over South America. On the other hand, in general, the SWSA and ENSO modes have a combined effect on rainfall of the southern and southeastern regions of South America. A particularly important result of the present analysis for climate monitoring and forecasting purposes seems to be the changing relations between the SAD and ENSO modes. Copyright © 2012 Royal Meteorological Society

1. Introduction

It is well known that the El Niño/Southern Oscillation (ENSO) teleconnections established through Rossby-wave trains or through east–west type circulations associated with anomalous tropical heating (Nogués-Paegle et al., 2002) and, also through the ENSO-related changes in the Hadley circulation (Zhou and Lau, 2001) are responsible for interannual climate variations in large portions of the tropics. Over South America, the ENSO-related climate variations have mostly been documented for the precipitation (Ropelewski and Halpert, 1987; Aceituno, 1988; Kayano et al., 1988; Kiladis and Diaz, 1989; Kousky and Ropelewski, 1989; Ropelewski and Halpert, 1989; Rao and Hada, 1990; Pisciottano et al., 1994; Grimm et al., 1998; Giannini et al., 2000; Grimm et al., 2000; Montecinos et al., 2000; Souza et al., 2000; Giannini et al., 2001; Zhou and Lau, 2001; Cazes-Boezio et al., 2003; Andreoli and Kayano, 2005; Kayano and Andreoli, 2006). In fact, drier (wetter) than normal conditions over the northern and northeastern sectors and the opposite conditions over the western equatorial coast and southern/southeastern sectors of this continent have been attributed to the occurrence of El Niño (La Niña) episodes.

However, the interannual sea surface temperature (SST) variability modes in the South Atlantic Ocean may also contribute to the South American climate (Andreoli and Kayano, 2006). Indeed, severe droughts in northeastern Brazil (NEB) have been associated with the presence of negative SST anomalies in the tropical South Atlantic (Markham and McLain, 1977). Diaz et al. (1998) obtained positive simultaneous correlations between the SST anomalies in the southwestern subtropical Atlantic and rainfall anomalies over Uruguay and southern Brazil during austral summer. They proposed that both the SST anomalous patterns in the southwestern Atlantic and the rainfall anomalies over southern South America might be local manifestations of the ENSO teleconnections. Differently, Barros and Silvestri (2002) showed that the rainfall over southeastern South America is modulated by the SST in the subtropical central South Pacific during El Niño years and, by the SST in the subtropical South Atlantic during La Niña years.

Notwithstanding the important implications of the SST variability modes in the South Atlantic for the South American climate, they only received attention since the end of the 1990s. Venegas et al. (1997), using the SST and sea level pressure (SLP) data for the 1953–1992 period, and the single value decomposition (SVD) technique determined three-coupled modes of the atmosphere–ocean variability in the South Atlantic, all of them related to variations in the intensity or position of the South Atlantic Subtropical high (SASH). According to these authors, the first SVD mode describes the SASH intensity variations at a 14–16 year time scale and features an SST dipole with one centre located in the equatorial South Atlantic [centre approximately at (17°S, 12°W)], and a secondary opposite sign centre in the extratropical South Atlantic [centre at (40°S, 25°W)]. Bombardi and Carvalho (2011) referred to this mode as the South Atlantic dipole (SAD). A positive (negative) SAD mode features positive (negative) SST anomalies in the northern centre and negative (positive) SST anomalies in the extratropical South Atlantic, and relates to weakened (intensified) SASH (Venegas et al., 1997).

The second SVD mode obtained by Venegas et al. (1997) shows an SST monopole with its centre at (10°S, 0°W), and describes the east–west shifts of the SASH at a 6 to 7 year time scale. Finally, their third SVD mode shows more structure in space with the strongest SST anomalies in a large latitudinal area bounded at 15°S, 35°S, 40°W and 10°E, and relates to north–south shifts of the SASH at a 4 year time scale. They also showed that this mode is strongly linked to the ENSO, with the SST anomalies in the South Atlantic preceding the Pacific warming by 1–2 months.

Sterl and Hazeleger (2003) studied the coupled atmosphere–ocean variability in the South Atlantic for the 1949–2000 period using the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis. They showed that the SST anomalies are mostly induced by the atmospheric circulation anomalies through latent heat flux and mixed layer deepening. They also found that the variability in the South Atlantic Ocean is largely independent of the variability in the other oceans. They found only a weak relation with the ENSO. Consistently, Bombardi and Carvalho (2011), using ten global coupled climate models for the 20th century climate scenario from 1971 to 2000, found that the SAD variability is independent of the ENSO. Using observational and reanalysis data sets from 1950 to 2008, Nnamchi et al. (2011) proved the existence of the SAD mode in the South Atlantic as a distinct mode from the Atlantic equatorial mode (AEM), and independent of the direct influence of the ENSO in the tropical Pacific. The AEM is an interannual SST mode with its main anomalous pattern in the eastern side of the equatorial Atlantic (Zebiak, 1993).

Colberg et al. (2004), using simulations with an ocean general circulation model (ORCA2) forced with the NCEP/NCAR reanalysis data for the 1948–1999 period, found significant influence of the ENSO, through the wind stress anomalies, on the upper ocean temperatures in the South Atlantic. They suggested that under El Niño onset, the negative SLP anomalous centre in the South Atlantic causes a weakening of the southerly trades and a strengthening of the midlatitude westerlies. In consequence, the reduction of the southward Ekman heat transport in the tropics and the enhancement of the northward Ekman heat transport in the midlatitudes cause a warming between the equator and 25°S and a cooling in the midlatitudes during the season from October to December (OND). They noted similar, but not exactly reversed patterns, during OND of the La Niña onset years. The composite patterns obtained by Colberg et al. (2004) and the third SVD mode presented by Venegas et al. (1997) show differences which are indicative that they might not be the same mode. In fact, the ENSO-related SST composites obtained by Colberg et al. (2004), for some stages, present close resemblances with the first SVD mode shown by Venegas et al. (1997).

So, most papers searching for the connections between the ENSO and the South Atlantic SST variability modes considered them simultaneously or when the Pacific mode precedes the South Atlantic modes. From another point of view, several authors showed that the AEM is strongly linked to the ENSO, with the SST anomalies in the Atlantic preceding the opposite sign anomalies in the eastern equatorial Pacific. Indeed, Losada et al. (2010) showed that a warm AEM in May–June, through an anomalous Atlantic Walker circulation, favours the establishment of La Niña conditions in the tropical Pacific during the July–August–September period. Under warm AEM conditions, the enhanced surface divergence in the central and eastern Pacific due to the descending branch of the anomalous Atlantic Walker circulation shallows the equatorial thermocline, and triggers the La Niña (Rodríguez-Fonseca et al., 2009). More recently, Kayano et al. (2011) noted that the Atlantic SST anomalous conditions associated with the AEM are persistent and might be noted 8–9 months before the establishment of the ENSO conditions in the tropical Pacific. They also documented the evolving patterns of the SST and SLP leading to the ENSO extremes forced by or independent of the AEM, as well as their relations to the rainfall over South America.

To summarize, some authors found close relations between the ENSO and some of the South Atlantic SST modes, while others did not. In addition, the temporal precedence of the ENSO and South Atlantic modes, particularly the SAD mode, was not clearly established yet. Thus, in this paper we will further examine the connections of the ENSO with the dominant South Atlantic SST variability modes as well as their relations with the rainfall variations over South America. The following sections are organized as follows: a brief description of data and methods in Section '2. Data and methods'; analyses of the South Atlantic and tropical Pacific SST modes, and their relations and influences on the South American rainfall, in Section '3. Results'. Conclusions are given in Section '4. Concluding remarks'.

2. Data and methods

The dataset used in this article is the version 3B of the reconstructed monthly SST series obtained by Smith et al. (2008) at 2° by 2° latitude–longitude resolution grid. According to these authors, this version was obtained using in situ SST data and improved statistical tools that allowed stable reconstruction using sparse data. They also stated that in this version, the improved low-frequency tuning and historical merged analysis of SST eliminated most of the low-frequency damping error. They also used improved high-frequency tuning. According to these authors, as the local and short-term variations were smoothed in the version 3B SST data, this dataset is appropriate for long-term global and basin-wide studies. With this information in mind, the SST data is used here for large-scale and long-term analyses. The SST time series are selected in two areas, one in the South Atlantic bounded at equator, 60°S, 70°W and 15°E and in the tropical Pacific limited at 30°N, 30°S, 120°E and 70°W for the 1900–2006 period.

Monthly gridded precipitation time series at 1° resolution derived from the gauge-based reconstructions of the Global Precipitation Climatology Center (GPCC) (Beck et al., 2005; Rudolf and Rubel, 2005) are also used here. The GPCC full data reanalysis version 5 is used. Precipitation time series are selected for the 1901–2006 period in the area bounded at 15°N, 50°S, 90°W and 30°W, which encompasses most of the South American region. The gridded precipitation data obtained from the GPCC were generated by an operational analysis system which includes integration of data from different sources, quality-control and calculation of area-averaged precipitation on the grid cells (Rudolf and Schneider, 2005). Although errors might exist in this dataset due to deficiencies in the measuring process or to regionally sparse data, it contains high-quality gridded precipitation data. Error analysis in this dataset is out of the scope of the present article. So, taking into account that one of the possible sources of errors is sparse observational data, we will focus our analysis in the South American regions with dense rain gauge network.

The general aspects of the data processing are described here. Prior to any calculation, the linear trend is removed from the SST time series at each grid point. The SST and precipitation climatologies are calculated for each calendar month at each grid point by averaging the data over the 1900–2006, and 1901–2006 periods, respectively. Monthly anomalies are determined as departures from the mean annual cycle of each variable. The monthly precipitation anomaly series at each grid point is standardized by the corresponding monthly standard deviation.

The SST variability modes are obtained from the empirical orthogonal function (EOF) analyses of the SST anomalies in selected areas. In these analyses, the covariance matrix is used and the method proposed by North et al. (1982) is adopted to examine the separation of the modes. The eigenvectors are presented as correlation patterns. To examine the correlation statistical significance, the number of degrees of freedom is estimated as the time interval for two independent realizations, the lag needed to obtain autocorrelation coefficients of the time series close to zero. The principal component (PC) is standardized and gives the temporal variations of the corresponding mode. Spectral features of the PC time series are obtained through the Morlet wavelet transform whose calculation is given by Torrence and Compo (1998). The Morlet wavelet is a complex exponential modulated by a Gaussian, math image, with η = t/s, where t is the time, s is the wavelet scale and ω0 is a non-dimensional frequency (Torrence and Compo, 1998).

The partial correlation analysis is also used here. The partial correlation between two variables, X1 and X3, while excluding the effects of a third independent variable, X2, is defined as (Panofsky and Brier, 1968):

equation image

where, r13, r12 and r23 are the linear correlations between X1 and X3, between X1 and X2 and between X2 and X3, respectively. The linear correlation between X1 and X2 is referred to as ‘X1 × X2’, and the partial correlation between X1 × X3 without the effects of X2, ‘X1 × X3X2’.

3. Results

3.1. EOF modes for monthly SST data

The EOF analyses of the monthly SST anomalies in the South Atlantic and tropical Pacific are done separately for the two oceanic sectors. The mean number of degrees of freedom for the tropical Pacific is 50 and for the South Atlantic is 57. The Student's t-test for 50 and 57 degrees of freedom give respectively the thresholds of 0.35 and 0.33, for the correlations to be significant at the 99% confidence level.

The leading mode in the tropical Pacific explains 40.0% of the total variance for the SST in this oceanic sector (Figure 1(a)), and is well separated from the higher modes (North et al., 1982). The largest significant positive loadings are centred in the central tropical Pacific, and extend meridionally in the eastern side of the basin. As one might expect, this mode reflects mainly the ENSO-related SST variations in the tropical Pacific. Indeed, the largest positive (negative) values of the corresponding PC time series (PPC1) occur mostly during El Niño (La Niña) events (Figure 1). The Morlet wavelet spectral analysis of PPC1 shows the largest variances dominantly at the interannual time scale which yield a significant peak approximately at 4 year (figure not shown). So, the PPC1 time series may be used to describe the ENSO cycles.

Figure 1.

First EOF mode pattern of the SST monthly anomalies in the tropical Pacific for the 1900–2006 period. Contour interval is 0.20; the continuous (dashed) line is positive (negative), and the zero line, omitted. Shaded areas encompass significant correlations at the 99% confidence level. The lower panel displays the principal component of this mode (PPC1). Explained variance is on top of the panel in percentage

The patterns of the first three EOF modes of the SST anomalies in the South Atlantic and the corresponding PC time series (APC1, APC2 and APC3) are displayed in Figure 2. These modes explain respectively 19.7, 13.3 and 9.3% of the total variance for the SST in the South Atlantic. According to the North et al. (1982) method, they are well separated from each other.

Figure 2.

The EOF patterns and the corresponding APC time series of the first three modes of the SST monthly anomalies in the South Atlantic for the 1900–2006 period. Display is the same as in Figure 1

The first mode features a negative SAD mode with centres at (15°S, Greenwich longitude) and at (37.5°S, 25°W). These centres show significant negative and positive loadings with nearly balanced magnitudes. The SAD mode obtained here holds resemblances with the first SVD mode in the South Atlantic found by Venegas et al. (1997). The APC1 time series shows interannual to decadal time scale fluctuations. The Morlet wavelet analysis of this time series (figure not shown) shows significant interannual time scale variances scattered during the period of analysis, which lead to a significant peak approximately at 1.5 year.

The equatorial sector of the SAD mode holds similarities with the AEM mode previously documented by Zebiak (1993) and retrieved by Kayano et al. (2011) through EOF analysis of the interannual SST anomalies in the tropical Atlantic. To examine this aspect the interannual component of the APC1 and the PC time series of the AEM obtained by Kayano et al. (2011) are compared. The APC1 time series is then filtered with a band-pass 1–6 year Morlet wavelet filter which is the same one used by Kayano et al. (2011). The linear correlation between these two interannual time scale series is 0.9 which, using the Student t-test, is highly significant at a confidence level above 99.9%. Since the analysis by Kayano et al. (2011) is limited to the tropics, some cases of the AEM identified in their study might be indeed the equatorial side of the SAD mode. Furthermore, Nnamchi et al.'s (2011) criterion to identify an AEM is based on a weak or inexistent centre in the southwestern South Atlantic (SWSA). With this criterion, some weak SAD modes may be identified as a pure AEM. All these indicate that some cases of AEM identified in previous studies might be the equatorial side of an SAD mode.

The second mode in the South Atlantic features only one main area with significant loadings located between 35°S and 50°S (Figure 2). The largest loadings of this mode extend over an area with sparse data measurements in the past, most of which was avoided in the analysis done by Venegas et al. (1997). Consistently, this mode has no correspondence to any of the South Atlantic SVD modes described by Venegas et al. (1997). The APC2 shows interannual to multidecadal variations before 1965, and considerably reduced amplitudes from 1965 to 1990 (Figure 2). The Morlet wavelet analysis of APC2 (figure not shown) shows significant variances from seasonal to interannual and from decadal to multidecadal time scales before 1960, and no significant variances afterwards. Since the high- and low-frequency signals are stronger prior 1960 when observational data were fewer than afterwards, these signals in the southeastern South Atlantic might not represent a physical mode. So, this mode will not be further analysed.

The third SST mode in the South Atlantic shows a dipolar structure with a strong negative centre at (30°S, 40°W) in the SWSA that extends between 15°S and 45°S, and a less extensive positive centre in the southern midlatitudes (Figure 2). Since the negative centre is more extensive and stronger than the positive one, the negative centre will be used to refer to this mode. So, this mode will be referred to as the SWSA mode. It is dominated by interannual fluctuations, as shown in its PC time series (APC3) (Figure 2) and confirmed by the Morlet wavelet analysis of APC3 that shows the largest variances at the interannual time scale, with a significant peak at 4 year (figure not shown). The SWSA mode holds some similarities with the third SVD mode obtained by Venegas et al. (1997), which they related to the ENSO, and with the SST anomaly composite for El Niño during January–February–March (JFM) season shown by Colberg et al. (2004). It is apparent in the analysis here that the APC3 are mostly negative (positive) during El Niño (La Niña) years. So, the surface waters off southern Brazil and Uruguay warm up (cool down) during El Niño (La Niña) years. This result confirms the El Niño relation to warmed surface waters off southern Brazil and Uruguay noted in previous papers (Diaz et al., 1998; Barreiro, 2010). Moreover, the results here evidence that the El Niño-related warmed (La Niña-related cooled) SST pattern in the SWSA is part of a dipole pattern of the SWSA mode that extends from tropical latitudes to midlatitudes.

3.2. Relations between the Atlantic modes and the ENSO

The relationships of the SAD and SWSA modes in the South Atlantic to the ENSO, represented here by the first SST mode in the tropical Pacific, are searched for using correlation analysis. Since the PPC1 shows large interannual variability, and the APC1 and APC3 feature higher frequency variability superimposed to decadal time scale variability, their corresponding seasonal time series are obtained. Each seasonal time series contains 484 seasonally averaged values spanning from December–January–February 1900–1901 to September-October-November 2006. This procedure eliminates the difference in the high-frequency variability between the PPC1 and the Atlantic PC time series. Lag and simultaneous 41-season running correlations between the ENSO and the Atlantic PC time series are calculated. The statistical significance of the correlations is examined using the random-phase test suggested by Ebisuzaki (1997). So, 1000 random-phase time series with the same power spectrum as the PPC1 time series are generated. The distribution of the correlations between the random-phase time series and the Atlantic PC time series is then used to define the threshold for significant correlations. The 95% percentile is used. The resulting running correlations are limited to the 1905–2001 period and displayed in time versus lag plots. Negative (positive) lags indicate that PPC1 time series leads (lags) the Atlantic PC time series.

Figure 3 shows the running correlations between PPC1 and APC1 displayed in time versus lag plot. It is interesting to note that the correlations between PPC1 and APC1 for negative (positive) lags are predominantly negative (positive). Furthermore, the PPC1 and APC1 show very weak simultaneous correlations for some periods, such as during 1905–1935 and 1975–2000. This result concords with those obtained by Bombardi and Carvalho (2011) who found no significant correlation for ENSO simultaneous to SAD or for ENSO leading SAD for the 1971–2000 period.

Figure 3.

Time versus lag plot of the 41-season running correlations between PPC1 and APC1. Lag interval in the abscissa is one season. Shaded areas encompass significant correlations at the 95% confidence level

The negative correlations are noticeable from lags − 2 seasons to + 1 season particularly during the 1920–1930 and 1940–1970 periods. For the negative lag, the interpretation is that an El Niño (a La Niña) precedes by up to two seasons the establishment of a positive (negative) SAD in the South Atlantic. For the positive lag, the interpretation is that a positive (negative) SAD precedes by up to one season the establishment of an El Niño (La Niña). Simultaneous occurrences of an El Niño (a La Niña) and a positive (negative) SAD were previously documented by Kayano and Andreoli (2006). They showed that this relation occurs for an ENSO event with its main centre located in the central Pacific and without SST ENSO signature in the eastern equatorial Pacific off the South American coast. In agreement, Rodrigues et al. (2011), analysing the warm ENSO phase, found that a positive SAD develops in boreal winter only during weak and short El Niño events located in the central Pacific.

The PPC1 and APC1 time series are significantly and positively correlated from lag + 1 to lag + 4 seasons during the 1975–2000 period (Figure 3). So, a negative (positive) SAD in the South Atlantic precedes by one season to 1 year the establishment of an El Niño (a La Niña) event. Once the negative (positive) SAD persists, the persistent descending (ascending) motion in the equatorial Atlantic related to the cooling (warming) there gradually drives an anomalous east–west circulation in the equatorial–vertical plane with an ascending (descending) branch in the eastern equatorial Pacific. This east–west circulation contributes to weaken (enhance) the surface divergence in the eastern equatorial Pacific, leading to the sinking (shallowing) the equatorial termocline. Under this persistent condition, an El Niño (a La Niña) is established one season to 1 year later. This mechanism is the same discussed previously concerning the inter-Pacific-Atlantic east–west SST gradient which through a feedback process relates the ENSO and the AEM, with the AEM preceding by some months the establishment of the ENSO mode (Wang, 2006; Losada et al., 2010; Rodríguez-Fonseca et al., 2009; Kayano et al., 2011).

The results show that the ENSO and SAD modes are closely lagged or lead connected depending on the period. Another interesting aspect is the modulation of the correlations by the decadal and multidecadal variability. Indeed, the significant positive correlations for positive lags during the 1975–2000 period show maxima at a decadal time scale. Furthermore, the multidecadal modulation of the positive correlations for positive lags as well as the decadal modulation of the largest negative correlations for negative lags during the 1940–1970 period are also apparent. These modulations result from the decadal component of the SAD mode.

Figure 4 shows the running correlations between PPC1 and APC3 displayed in time versus lag plot. These time series are negatively correlated from lag − 4 to lag + 2 seasons during most of the 1905–2001 period. However, significant negative correlations are conspicuous in three periods: 1905–1920, 1945–1962 and 1970–1980 for the lag interval from − 4 seasons to 0 season. The interpretation is that an El Niño (a La Niña) event precedes by up to 1 year the establishment of positive (negative) SST anomalies in the SWSA between 15°S and 45°S. The mechanism relating the ENSO and the SST anomalies in the southwestern Atlantic Ocean might be that suggested by Colberg et al. (2004). This mechanism involves variations in the low-level winds which cause alteration in the Ekman heat transport. They suggested that under El Niño onset, the weakening of the southerly trades and the strengthening of the midlatitudes westerlies yield a reduction of the southward Ekman heat transport in the tropics and an enhancement of the northward Ekman heat transport in the midlatitudes. In consequence, a warming in the band from the equator to 25°S and a cooling in the midlatitudes are noted one season later. They noted similar, but not exactly reversed patterns, during the La Niña onset years.

Figure 4.

Time versus lag plot of the 41 month running correlations between PPC1 and APC3. Display is the same as in Figure 3

Significant negative correlations are also noted during the 1925–1945 and 1980–1990 periods for positive lags up to two seasons. In this case, the interpretation is that positive (negative) SST anomalies in the SWSA between 15°S and 45°S precedes by up to two seasons the establishment of an El Niño (a La Niña) event.

3.3. Atlantic seasonal SST modes

To examine the seasonal differences of the dominant SST variability modes in the South Atlantic, separated EOF analyses are done for seasonally averaged SST anomalies in the South Atlantic. The seasons are referred to with the initials of the months such as DJF for summer, MAM for autumn, JJA for winter and SON for spring. The mean number of degrees of freedom in the South Atlantic for each season is 20. The Student's t-test for 20 degrees of freedom gives the threshold of 0.42, for the correlations to be significant at the 95% confidence level. The patterns of the first three EOF modes of the seasonal SST anomalies in the South Atlantic for DJF, MAM, JJA and, SON are displayed in Figures 5(a)–(d), respectively. As in the monthly analysis, the first three modes of each season are well separated from each other (North et al., 1982).

Figure 5.

The EOF patterns of the first three modes of the seasonal SST anomalies in the South Atlantic for the 1900–2006 period: (a) DJF, (b) MAM, (c) JJA, (d) SON. Contour interval is 0.20; the continuous (dashed) line is positive (negative), and the zero line, omitted. Shaded areas encompass significant correlations at the 95% confidence level. Explained variance of each mode is on top of the corresponding panel in percentage

The seasonal analyses reproduce the major features of the first three EOF modes of the monthly analysis. For sake of comparison, the monthly APC time series are seasonally averaged. Table 1 lists the correlations between the APC time series of the seasonal analyses and the seasonally averaged APC time series of the monthly analysis. For the significance of these correlations, the estimated number of degrees of freedom is 20 for each season. These correlations are all highly significant at confidence levels above 99.9%, except the correlation of 0.45 between the APC3 of the seasonal analysis and the seasonal average APC2 for JJA which is significant at the 95% confidence level. These highly significant correlations indicate that the first three EOF modes of the monthly analysis are features of the low-frequency SST variability in the South Atlantic, as shown above.

Table 1. Correlation between seasonally averaged APC (indicated by ‘M’) and seasonal APC (indicated by ‘S’) for each season
APC number seasonM1 × S1M2 × S2M3 × S3M2 × S3M3 × S2
DJF0.990.910.68
MAM0.970.950.95
JJA0.970.450.81
SON0.990.960.93

The SAD mode with a northeast–southwest dipolar structure is reproduced in the first EOF seasonal modes (left column in Figure 5). Its northeastern and southwestern centres show nearly balanced magnitudes for SON and DJF seasons when the SAD mode is well defined. On the other hand, the weaker southwestern centre than the northeastern centre during MAM and JJA indicates a weak SAD mode during these seasons. The seasonal dependence noted here on the SAD mode intensity is in agreement with previous results. Indeed, Morioka et al. (2011) found that the SAD mode growth phase occurs in late spring, its mature phase, in summer, and its decay phase, in early autumn. Haarsma et al. (2003) found the strongest atmospheric response to the SAD mode during austral summer.

Since Nnamchi et al. (2011) also used the EOF analyses to get seasonal SST variability modes in the South Atlantic, comparisons between the modes obtained here and their modes (their Figure 2) are done. Contrasting with the results here, Nnamchi et al. (2011) found that the SAD mode is stronger during winter. The possible sources of the discrepancies are the differences in the data processing procedures, period of analysis and data source. Here, the linear trends have been removed from the SST time series. This justifies why they obtained the SAD mode in the second EOF, while here it is the first EOF mode. Whereas they standardized the data in each grid point prior to the EOF analyses, here the data are not standardized and the covariance matrix is used in the EOF calculations. We chose to avoid standardization because this procedure gives equal importance to regions of low and high temporal variance and a more uniform spatial structure in the EOF modes. We suspect that this is the main cause of discrepancies between our and Nnamchi et al.'s (2011) results.

The second SST variability mode for DJF, MAM, SON and the third SST variability mode for JJA (Figure 5) are similar to the second SST variability mode of the monthly analysis (Figure 2). Because this mode might not be a physical mode, as stated above, the corresponding seasonal modes are not further examined.

The third SST variability mode for DJF, MAM, SON and the second variability mode for JJA (Figure 5) are similar to the SWSA mode of the monthly analysis (Figure 2). The SWSA mode is well reproduced for MAM, JJA and SON. The dipolar structure of this mode with a strong centre in the SWSA and an opposite sign centre in the southern midlatitudes is less defined in DJF. This explains the lowest correlation of 0.68 between the APC3 of the seasonal and monthly analyses for DJF compared to the relatively high correlations between the corresponding APC time series for the other seasons (see Table 1). However, in general the SWSA mode is reproduced in the seasonal analyses.

3.4. Influences of seasonal SST modes on the South American rainfall

Although the lagged correlations between the PPC1 and APC1 show changes during the period of analysis, the corresponding ENSO and SAD modes might separately or together affect the South American rainfall. In addition, the ENSO and SWSA modes may also separately or together affect the South American rainfall. So, the simultaneous effects of the ENSO and SAD modes as well as of the ENSO and SWSA modes on the South American rainfall are examined here for the 1901–2006 period. These analyses are done on seasonal basis using correlation and partial correlation calculations.

To carry out seasonal analyses, the monthly precipitation anomalies, PPC1, APC1 and APC3 from the monthly analysis are seasonally averaged. The use of seasonally averaged PPC1, APC1 and APC3 instead of the PC time series from the seasonal analyses simplifies the calculations and is justified because the ENSO, SAD and SWSA modes are well reproduced in the seasonal analyses (Figures for ENSO are not shown here). For 106 degrees of freedom (number of years), the Student t-test gives the threshold values of 0.2 for correlations to be significant at the 95% confidence level.

The correlation maps between PPC1 and the precipitation (PPC1 × Prp), as well as the partial correlation maps between PPC1 and precipitation while removing the effect of the APC1 (PPC1 × Prp − APC1) and while removing the effect of the APC3 (PPC1 × Prp − APC3) are presented in Figure 6. In general, the El Niño- (La Niña-) related dipolar structure of the correlations with significant negative (positive) values in the northern portion of South America and opposite sign significant values in the southeastern/southern portion of this continent is reproduced in all seasons. However, the centres of this dipole show seasonal differences in their locations and intensities. In fact, significant negative correlations extend from NEB westward to the central Amazon during MAM and occupy a region in the northern and northwestern South America during the other seasons. Most months of MAM overlaps the northeastern and central Amazon rainy season. So, the main seasonal system modulating the rainy season in this area, the Atlantic intertropical convergence zone (ITCZ) (Ratisbona, 1976; Hastenrath and Heller, 1977), is affected by the ENSO. Another interesting seasonal aspect is the well-defined centre with significant positive correlations over Uruguay, southern and eastern Brazil during austral spring. This aspect reflects the strongest impacts of the El Niño and La Niña events noted over Uruguay (Pisciottano et al., 1994) and southern Brazil (Rao and Hada, 1990; Grimm et al., 1998) during SON.

Figure 6.

Seasonal correlation maps between PPC1 and precipitation (Prp) for: (a) DJF, (b) MAM, (c) JJA and (d) SON. Seasonal partial correlation maps between PPC1 and Prp while removing the effect of APC1 for: (e) DJF, (f) MAM, (g) JJA and (h) SON. Seasonal partial correlation maps between PPC1 and precipitation while removing the effect of APC3 for: (i) DJF, (j) MAM, (k) JJA, and (l) SON. Contour interval is 0.20; the continuous (dashed) line is positive (negative), and the zero line, omitted. Shaded areas encompass significant correlations at the 95% confidence level. The period of analyses is 1901–2006

The correlation and partial correlation maps show quite similar patterns for all seasons. This is particularly true for PPC1 × Prp and PPC1 × Prp − APC1. So, the SAD mode has a weak influence on the ENSO-related rainfall anomalies over South America.

Otherwise, the differences between PPC1 × Prp and PPC1 × Prp − APC3 are quite evident in the southern and southeastern sectors of the continent, where significant positive partial correlations are weakened and/or occupy relatively smaller areas than the significant positive correlations. The interpretation is that the wet (dry) conditions in the southeastern and southern South America associated with the El Niño (La Niña) are reinforced by the anomalously warm (cold) surface waters in the SWSA. Thus, the SWSA mode and ENSO have combined effects in increasing the precipitation anomalies in the southern and southeastern South America.

The correlation maps between APC1 and the precipitation (APC1 × Prp) show interesting seasonal differences (Figure 7). A dipolar structure with significant positive correlations over northern South America and the opposite sign correlations over northeastern South America is noted during DJF and MAM. A less organized structure with significant negative correlations over northern South America and the positive ones scattered over central and eastern South America is noted during JJA and SON. To interpret these results it is important to take into account that the main system modulating the seasonal rainfall over northern and northeastern South America, the Atlantic ITCZ, performs a seasonal meridional migration with its extreme southerly position in March–April and its extreme northerly position in September–October (Ratisbona, 1976). A negative (positive) SAD mode prevailing during the austral summer and autumn seasons impedes (facilitates) the seasonal southward migration of the ITCZ. In consequence, above (below) normal rainfall occurs in areas over northern South America and below (above) normal rainfall, over NEB. Otherwise, as the ITCZ migrates northward from the austral winter to spring, during these seasons the rainfall over most areas of South America might be under a more direct influence of the SAD mode. In fact, the negative (positive) SAD mode during JJA and SON might have a local influence causing below (above) normal rainfall over northern South America (eastern Venezuela, Guyana, French Guiana and surrounding northern Brazil areas for JJA and northern Brazil for SON), and above (below) normal rainfall in small areas scattered over central and eastern South America.

Figure 7.

Seasonal correlation maps between APC1 and precipitation for: (a) DJF, (b) MAM, (c) JJA, and (d) SON. Seasonal partial correlation maps between APC1 and precipitation while removing the effect of PPC1 for: (e) DJF, (f) MAM, (g) JJA, and (h) SON. Display is the same as in Figure 6

The ENSO influence on the relation between the SAD mode and the rainfall over South America is mostly weak, except during DJF. The partial correlation map (APC1 × Prp − PPC1) for this season shows that most of the significant positive correlations noted over northwestern South America for the correlation map are weakened or absent (Figure 7).

The seasonal maps of correlations between APC3 and the precipitation (APC3 × Prp) and of the partial correlations between APC3 and the precipitation while removing the effect of the PPC1 (APC3 × Prp − PPC1) are displayed in Figure 8. The correlation maps show remarkable seasonal differences. Significant negative correlations are noted over northern Peru only during DJF and over southeastern South America in all seasons. The significant negative correlations are found over southeastern South America in JJA, and over Uruguay, southern and southeastern Brazil during SON. Significant positive correlations are found over NEB in MAM, and over part of northern South America in JJA and SON. However, most of these significant correlations seem to be due to the ENSO action. Indeed, most of the significant correlations noted in the seasonal correlation maps are weakened or absent in the corresponding partial correlation maps. This is the case of the significant positive correlations noted over NEB, during MAM, and over northern South America during JJA and SON, and of the negative ones, over Uruguay, and southern and southeastern Brazil during SON. Thus, in general, the significant correlations between APC3 and precipitation represent mostly the ENSO effect on the South American rainfall through the SWSA mode. Therefore, the SWSA, third EOF mode for the SST in the South Atlantic, is closely related to the ENSO, as shown above with the running correlation analysis of the APC3 and PPC1. Nevertheless, the significant negative correlations noted over Peru during DJF, and over part of southeastern South America during DJF, MAM and JJA remain quite strong for the corresponding partial correlation maps (APC3 × Prp − PPC1). In these cases, the interpretation is that the dry (wet) conditions over these regions and seasons are ENSO independent and in part related to cooled (warmed) surface waters off southern Brazil and Uruguay.

Figure 8.

Seasonal correlation maps between APC3 and precipitation for: (a) DJF, (b) MAM, (c) JJA, and (d) SON. Seasonal partial correlation maps between APC3 and precipitation while removing the effect of PPC1 for: (e) DJF, (f) MAM, (g) JJA, and (h) SON. Display is the same as in Figure 6

3.5. Relations between the seasonal SST modes and the South American rainfall for three sub-periods

As illustrated above, the relations between the ENSO and Atlantic SST modes show a temporal dependence (Figures 3 and 4). In the previous sub-section, the influences of these modes on the South American rainfall are analysed without considering the temporal dependence of the relations between the Pacific and Atlantic SST modes. This aspect is examined by repeating the correlation and partial correlation analyses for three non-overlapping sub-periods: 1911–1940, 1941–1970, and 1971–2000. To facilitate comparisons among the sub-periods, they are selected with a same length but their choice is based on Figure 3. For 30 degrees of freedom (number of years), the Student t-test gives the threshold values of 0.3 for correlations to be significant at the 90% confidence level.

Differences among the sub-periods are noted in the correlation maps between the SST PC time series (PPC1, APC1 and APC3) and the rainfall anomalies. However, the correlation and partial correlation maps for each sub-period hold similar relations as those for the whole period. These relations refer to increase, decrease or maintenance of the correlation values. Therefore, the partial correlation maps for the sub-periods are not shown here.

The seasonal correlation maps between the PPC1 and the rainfall anomalies are shown in Figure 9. The correlation map differences among the sub-periods are mostly in the spatial extension of the significant values. For all seasons, the correlation maps of the 1911–1940 sub-period show smooth patterns with significant negative values over northern South America; and the positive ones to the south extend over southeastern South America in DJF, central South America in MAM, southern South America and part of eastern central Brazil in JJA, and central and eastern South America in SON. The 1941–1970 sub-period correlation maps show significant values in relatively large and well-defined areas in DJF and SON, and in small areas scattered in southern and southeastern South America during the other seasons. The correlation maps of the 1971–2000 sub-period are very similar to those of the 1901–2006 period.

Figure 9.

Seasonal correlation maps between PPC1 and precipitation for: (a) DJF, (b) MAM, (c) JJA, and (d) SON for the 1911–1940 sub-period; (e) DJF, (f) MAM, (g) JJA, (h) SON for the 1941–1970 sub-period; (i) DJF, (j) MAM, (k) JJA, (l) SON for the 1971–2000 sub-period. Contour interval is 0.30; the continuous (dashed) line is positive (negative), and the zero line, omitted. Shaded areas encompass significant correlations at the 90% confidence level

The seasonal correlation maps between the APC1 and rainfall anomalies are depicted in Figure 10. Again, the correlation map differences among the sub-periods are in the spatial extension of the significant values. The smoothest patterns are also noted in the 1911–1940 sub-period, in particular during DJF, MAM and JJA. In this case, for each season the correlation map of the 1901–2006 period seems to be a combination of the corresponding sub-period seasonal maps.

Figure 10.

Seasonal correlation maps between APC1 and precipitation for: (a) DJF, (b) MAM, (c) JJA, and (d) SON for the 1911–1940 sub-period; (e) DJF, (f) MAM, (g) JJA, (h) SON for the 1941–1970 sub-period; (i) DJF, (j) MAM, (k) JJA, (l) SON for the 1971–2000 sub-period. Display is the same as in Figure 9

The seasonal correlation maps between APC3 and the South American rainfall anomalies are illustrated in Figure 11. For each season, they show differences among the sub-periods in the extension of the significant correlation areas. Again, the smoothest correlation patterns are noted for the 1911–1940 sub-period. Furthermore, most correlation maps for the 1941–1970 and 1971–2000 sub-periods show only very small areas with significant values. The exception is the JJA correlation map for 1971–2000. This map features significant positive correlations over northern South America and the negative ones in a large area extending from central southeastern South America northward to southern Amazon and then eastward to NEB. This map resembles the corresponding map for the 1911–1940 sub-period.

Figure 11.

Seasonal correlation maps between APC3 and precipitation for: (a) DJF, (b) MAM, (c) JJA, and (d) SON for the 1911–1940 sub-period; (e) DJF, (f) MAM, (g) JJA, (h) SON for the 1941–1970 sub-period; (i) DJF, (j) MAM, (k) JJA, (l) SON for the 1971–2000 sub-period. Display is the same as in Figure 9

The correlation maps between SST PC time series and the South American rainfall anomalies show differences among the three 30 year sub-periods. These differences among sub-periods are mostly in the spatial extension of the significant correlations. However, as the correlation and partial correlation maps for each sub-period hold relations similar to those for the whole period, the discussion above for the 1901–2006 can also be extended to the sub-periods.

4. Concluding remarks

This work examines the relations between the ENSO in the tropical Pacific and the dominant SST modes in the South Atlantic as well as their influences on the South American rainfall variability. The SST modes are obtained from independent EOF analyses of the SST anomalies in these two oceanic areas. Of the first three SST modes in the South Atlantic, the SAD and the SWSA modes are analysed in more detail.

The existence of the SAD mode as the dominant SST variability mode in the South Atlantic has been demonstrated since the end of the 1990s (Venegas et al., 1997; Bombardi and Carvalho, 2011; Nnamchi et al., 2011). It has been associated with the SASH intensity variations at a decadal time scale (Venegas et al., 1997). It is worth mentioning that the methods adopted here to get and to examine the SAD mode do not allow us to make a distinction between the SAD and the AEM, as done by Nnamchi et al. (2011). This distinction is out of the scope of this analysis. Rather, a time series giving the temporal variations of the SAD mode, necessary to perform correlation, partial correlation and wavelet analyses, is aimed here.

An important result here is that the ENSO and SAD modes are closely lagged or lead related depending on the period of analysis. This relation is modulated by the decadal component of the SAD mode. The negative correlations for negative lags between PPC1 and APC1 during the 1920–1930 and 1940–1970 periods indicate that an El Niño (a La Niña) precedes by up to two seasons the establishment of a positive (negative) SAD mode. The simultaneous occurrence of an El Niño (a La Niña) and a positive (negative) SAD mode during austral autumn was previously documented to occur in association with weak ENSO events located in the central Pacific (Kayano and Andreoli, 2006; Rodrigues et al., 2011. Although most El Niño and La Niña events classified as this type occurred during the 1920–1930 and 1940–1970 periods, other types of El Niño and La Niña events occurred in these periods (Kayano and Andreoli, 2006).

On the other hand, the positive (negative) SAD mode during the 1975–2000 period precedes by up to 1 year the establishment of a La Niña (El Niño). In this case, previous results indicate that the relations between the Pacific and Atlantic modes might occur through the east–west circulation in the equatorial latitudes (Nogués-Paegle et al., 2002; Wang, 2006; Rodrigues-Fonseca et al., 2009; Kayano et al., 2011).

The SWSA mode is indeed the South Atlantic SST mode driven by the ENSO as previously found in some papers (Diaz et al., 1998; Colberg et al., 2004). The mechanism relating the ENSO and the SST anomalies in the southwestern Atlantic Ocean might involve variations in the low-level winds which cause alteration in the Ekman heat transport as suggested by Colberg et al. (2004). They suggested that under El Niño onset, the weakening of the southerly trades and the strengthening of the midlatitude westerlies yield a reduction of the southward Ekman heat transport in the tropics and an enhancement of the northward Ekman heat transport in the midlatitudes. In consequence, a warming is noted between equator and 25°S and a cooling in the midlatitudes.

The effects of the SST modes on the South American rainfall are analysed on a seasonal basis. In general, the SAD and ENSO modes might act mutually independently on the South American rainfall. Indeed, the SAD (ENSO) mode has a weak influence on the ENSO-related (SAD-related) rainfall anomalies over South America. Exception is the ENSO influence on the SAD-related rainfall anomalies over northern South America during DJF. The strongest influence of the SAD mode on the South American rainfall is noted over northern and northeastern parts of this continent. Since the main system modulating the seasonal rainfall over South America is the Atlantic ITCZ, and it performs a seasonal meridional migration (Ratisbona, 1976), the SAD mode affects the seasonal South American rainfall through the ITCZ. A negative (positive) SAD mode during the austral summer and autumn seasons impedes (facilitates) the seasonal southward migration of the ITCZ, thus causing above (below) normal rainfall over northern South American areas and below (above) normal rainfall, over NEB. As the ITCZ migrates northward from the austral winter to spring, the rainfall over South America might be under a more direct influence of the SAD mode. In these seasons, the negative (positive) SAD contributes to below (above) normal rainfall over northern South America and above (below) normal rainfall in small areas scattered over eastern and southeastern South America.

The analyses of the relations of the ENSO and SWSA modes and their effects on the South American rainfall showed that most of the significant correlations between APC3 and precipitation represent basically the ENSO effect on the South American rainfall through the SWSA mode. Exceptions are the significant negative correlations noted over Peru during DJF and in part of southeastern South America during DJF, MAM and JJA. In these cases, the dry (wet) conditions over most of Peru in DJF and in part of southeastern South America during DJF, MAM and JJA are ENSO independent and related to the cooled (warmed) surface waters off southern Brazil and Uruguay.

The most important result of the analysis seems to be the changing relations between the SAD and ENSO modes. Our knowledge of what determine the temporal precedence of the ENSO and SAD modes is important for climate monitoring or prediction purposes. So, more detailed analyses should be done in the future to search this aspect.

In addition, the correlation between the South American rainfall anomalies and the SST PC time series show differences among the three 30 year sub-periods. These differences are mostly in the spatial extension of the significant correlations. Investigation on the causes of these differences is out of the scope of this article. However, it is known that this type of difference might be due to very low-frequency variability modulation, which creates a climate background for the interannual relations between the South American rainfall and the SST modes. In this context, Kayano et al. (2009) discussed the differences between the 1948–1976 and 1977–2002 periods in the interannual relations between some tropical SST indices and the South American rainfall. They suggested that one of these low-frequency phenomena is the Pacific decadal oscillation. However, the analyses here strongly suggest that other low-frequency phenomena should also be considered and further examined. This issue will be the subject of a future study by the authors.

Acknowledgements

The authors thank the two anonymous reviewers for their helpful comments and suggestions. The authors were partially supported by the Conselho Nacional de Desenvolvimento Científico e Tecnológico of Brazil. The authors thank the Conselho Nacional de Desenvolvimento Científico e Tecnológico and Financiadora de Estudos e Projetos of Brazil for the research support (CNPq, 569749/2008-5 and FINEP/REMCLAM). Wavelet software was kindly provided by C. Torrence and G. Compo and is available at URL: http://paos.colorado.edu/research/wavelets.

Ancillary