This paper deals with the relationship between the interannual variability of sea surface temperature (SST) and its associated atmospheric circulation and rainfall variability over southeastern South America (SSA), namely the subtropical region east of the Andes between 20° and 40°S, during the austral spring. Rainfall in SSA and SST interannual variability is studied using canonical correlation analysis. The first two modes show the well-known warm-wet and cold-dry pattern between the equatorial SST and rainfall over most of this part of the world. However, SST in the equatorial regions does not modulate rainfall variability among El Niño (EN) years or among La Niña (LN) years. On the other hand, it does modulate this variability between EN and LN cases as a whole and among neutral cases indicating that the SSA rainfall response to equatorial Pacific SST is not linear over the observed SST range, having no dependence on the extremes of this range. In contrast, among EN events, SST in the subtropical south-central Pacific (SSCP) modulates the seasonal rainfall over most of SSA. Also, when all the years are considered, this SST has a correlation with precipitation of magnitude similar to those corresponding to the SST in EN regions. Consistent with this, the circulation field has enhanced cyclonic (anticyclonic) advection over subtropical SSA when SST in the SSCP is cold (warm).
SSTs in EN-3 or EN-3.4 regions and in the SSCP are negatively correlated, but their correlation is practically zero when only EN cases or only neutral cases are considered, and very small in LN cases. This allows a stratification analysis composing cases according to different SSTs in EN-3 and EN-3.4 regions with almost constant SST in the SSCP and similarly according to different SST in the SSCP with approximately constant SST in EN regions. The composite difference fields with constant equatorial SST and with constant SST in the SSCP have a wave train in the Pacific similar to the ENSO-like pattern. Besides most of the ENSO circulation signal at 200 hPa over mid- and high latitudes is associated with SST variability at the SSCP rather than with SST variability in EN regions.
This paper deals with some aspects of El Niño (EN) and La Niña (LN) events and their relation with rainfall over southern South America. Concerning precipitation, the focus is on southeastern South America (SSA), namely the subtropical zone east of the Andes between 20° and 40°S, where there are significant signals of the El Niño–Southern Oscillation (ENSO) in the year-to-year rainfall variability. In fact, Aceituno (1988) found significant negative correlations between precipitation and the Southern Oscillation index (SOI) during October–November and Ropelewski and Halpert (1987, 1989) identified a large area comprising northeastern Argentina, southeastern Brazil, and Uruguay where there are positive rainfall anomalies from November of the starting year of EN [year (0)] to February of the following year [year (+)], and negative anomalies from July to December of the starting LN year. Also, Kiladis and Diaz (1989) have shown that in the same area but including also Paraguay there is a significant difference in the seasonal precipitation [September (0)–November (0)] between EN and LN years.
There are also detailed regional studies on the interannual response of precipitation to EN and LN events. Monthly rainfall correlates significantly with the SOI during the austral spring in the south of Brazil (Rao and Hada 1990). For the same area, Grimm et al. (1998) found that the highest rainfall anomalies takes place in November (0) of EN because of the intensification of the mesoscale convective activity, while the opposite happens in LN years. They also found that in part of this area there are similar signals during the winter (+). In Uruguay, Pisciotano et al. (1994) reported a significant tendency toward higher than normal precipitation during EN years from November (0) to January (+) and from March (+) to July (+) in the north of the country. They found a tendency toward lower than normal precipitation all over Uruguay during October (0)–November (0) and from March (+) to July (+) during the LN phase.
The greatest precipitation anomalies during EN and LN events are related to atmospheric circulation anomalies over the South American sector (Grimm et al. 2000). In the high troposphere during EN (LN), there is an anomalous circulation with a double dipole structure over the contiguous Pacific Ocean and over the Atlantic Ocean. Over the Pacific Ocean, there is a cyclonic (anticyclonic) circulation at the subtropics and an anticyclonic (cyclonic) center at midlatitudes. Over the Atlantic Ocean, there is a tendency toward an inverse pattern in the anomaly field. This dipole structure is stronger and more stable over the Pacific than over the Atlantic Ocean along both phases of ENSO. Although this pattern varies along each phase, with changes in the position and intensity of its centers, the general effect is an enhancement (decrease) of the subtropical upper-level westerly circulation and of the cyclonic advection over eastern SSA. This is particularly the case during the austral spring (0), a season when all studies coincide in that there are strong signals in the precipitation field. Garreaud and Battisti (1999, hereafter GB) suggested that this pattern in the eastern Pacific is consistent with more blocking events southwest of South America and with a shift of the storm tracks to the north during winter as was shown by Sinclair et al. (1997). The regional pattern is part of the hemispheric pattern associated with EN and LN conditions in the equatorial Pacific that is well documented (van Loon and Shea 1987; Karoly 1989; and GB). Karoly (1989) described the troposheric structure of this pattern for EN winter anomalies using the composite of three events. He found an equivalent barotropic wave train extending from the western subtropical Pacific to the Billingshausen Sea and turning to the northeast into South America and the Atlantic Ocean.
Although in some regions of SSA, the rainfall response to EN (LN) events is significant from the statistical point of view, rainfall response from one event to another is sometimes greater than the mean difference between EN and LN cases as it will be shown in sections 3 and 4. This can be also conjectured from the variability of the SSA river discharges between different EN (LN) events shown by Mechoso and Pérez Iribarren (1992). This behavior may be related to variability in both intensity and evolution EN (LN) events but it might also be caused by rainfall modulation by other regional or remote forcing. Whatever the cause of rainfall variability among EN (LN) cases, it is important to further explore this subject since most of the present ability for climate prediction in SSA is based on its response to ENSO. This issue is explored in this paper. The study is limited to the austral spring, which is a season with robust signals in the precipitation field during EN and LN events.
Data are described in section 2, and the climatic features of SSA rainfall are briefly described in section 3, including the variability of the regional response to EN and LN events during spring. The relationship between SST and rainfall over SSA during the spring season is discussed in section 4 showing that SST in other regions besides the equatorial Pacific contribute to modulate the rainfall interannual variability. Because SST in the subtropical south-central Pacific (SSCP) appears to be very important in this matter, the atmospheric circulation patterns associated with SST variability in this region are discussed in section 5. Section 6 assesses the relative contribution of the SST variations in the SSCP and in the equatorial Pacific to the ENSO signal in the atmospheric circulation field over the western Southern Hemisphere. The main conclusions are summarized in section 7.
Although the authors had access to more than 250 monthly precipitation series, most of them had large gaps or covered short periods of time. Also, some records present discontinuities due to operational problems or changes in the location of the stations. Therefore, only precipitation series from 91 stations from Argentina, Brazil, Paraguay, and Uruguay were selected for this study (Fig. 1). In this dataset, there was less than 2% data absent and none of the series have more than 7% data absent. To assure the quality of the data, most of the selected series came from synoptic stations, which were generally subject to a more careful control than other stations. The Argentine National Meteorological Service and the National Direction of Meteorology of Paraguay provided series from Argentina and Paraguay, respectively. For Brazil and Uruguay, precipitation records were taken from the National Center for Atmospheric Research (NCAR), Monthly Climate Data for the World, and also from the Institute of Agricultural Research of Rio Grande do Sul in the case of Brazil. NCAR's data were originally provided by the Institute of Meteorology of Brazil and the National Direction of Meteorology of Uruguay. Each of the institutions that provided the precipitation data carried out quality control procedures, checking their consistency with other meteorological observations. In the case of Argentina, extreme values were also checked with synoptic situations. The missing values were filled with data from the nearest station only in Uruguay. This technique proved to be as reliable as other more sophisticated interpolative schemes due to the great density of the precipitation network in this country, (Dirección Nacional de Meteorología and Universidad de la República 1998). In order to avoid distortions induced by the uneven distribution of the stations and smooth out singularities specific of single rain gauge measurements, precipitation data were averaged on a 5° latitude by 5° longitude array. This low resolution was determined by the sparse reliable data available to the authors, and constrains the value of the analyses to the large scale. The averaging was performed on monthly anomalies with respect to 1971–80, the period with the smallest proportion of missing data. Consequently, it was not necessary to fill in missing data, except for less than 1% of data from the 1971–80 records, which were completed with the mean monthly values of this period.
Global SST data with a 1° latitude by 1° longitude resolution were taken from the Meteorological Office Historical Sea Surface Temperature (version 5; Rayner et al. 1996). Sea level pressure (SLP), 200-hPa streamfunction and vertical velocity (ω) at 300-hPa composites were calculated from the National Centers for Environmental Prediction (NCEP)–NCAR reanalyses (Kalnay et al. 1996). These reanalyses use a frozen prediction model and an observational database that includes conventional surface and radiosonde observations, and aircraft and satellite data, being one of the most complete and physically consistent datasets. Although they have been extended back in time to 1948, their quality declines in years when data sources such as satellites and radiosondes were not available. Indeed, in most of the countries of the Southern Hemisphere, radiosonde observations started after 1958 and, of course, satellite data were available at a later data. Hence while for precipitation and SST analyses we have chosen the 1952–90 period because of the joint availability of these data, composite SLP and 200-hPa streamfunction analyses were restricted to the 1958–90 period. There was another problem with the NCEP–NCAR reanalysis arising from the incorrect assimilation of Australian SLP estimates at the ocean (PAOBS) between 1976 and 1992, when these data were erroneously shifted 180° in longitude. The assessment by NCEP–NCAR indicates that the impact of this mistake on monthly and longer-term timescales is fairly small (see online at http://westley.wwb.noaa.gov/paobs/). Since we used seasonal averages, the impact of this mistake is small, although as pointed out by Trenberth and Caron (2000), it may add some level of noise to the analyses.
EN and LN phases were taken as those defined by the Climate Prediction Center (available online from http://podaac.jpl.nasa.gov). Accordingly, the springs (see below) of EN events are 1957, 1958, 1963, 1965, 1968, 1969, 1972, 1976, 1977, 1982, 1986, 1987, and 1990; while 1954, 1955, 1956, 1964, 1970, 1971, 1973, 1974, 1975, 1983, 1984, and 1988 are the springs of LN events. In each case, this list includes the year if the event was present during the austral spring—October–November–December (OND). Therefore, a few differences should be expected with other characterizations that are based on the main or the starting years of the events, as in Quinn et al. (1987) for EN or Kiladis and Díaz (1989) for both EN and LN.
3. Climatic features of the SSA rainfall
Although positive trends have been important since 1960 in SSA (Barros et al. 2000a), the main features of the annual and monthly mean precipitation are still identified in the atlas based on data previous to 1960 (Hoffmann 1975). Annual precipitation regimes were described by Prohaska (1976) and, recently, by Gonzalez and Barros (1996). In general terms, annual precipitation in SSA decreases from east to west, although in northwestern Argentina, at around 65°W, there is a meridional band between 20° and 28°S with maxima of more than 800 mm. These relative maxima are believed to be caused by orographic precipitation associated with the sub-Andean mountains. The greatest annual precipitation—more than 1600 mm—is registered in the east between 25°–30°S and 51°–56°W, while to the west of 66°W, annual rainfall is less than 200 mm. There are considerable differences in the rainfall regime at the regional level. In northwestern Argentina, most of the annual precipitation falls during the austral summer, and there is a very pronounced dry season in winter. To the east, the dry season tends to disappear, which results in a decrease in the annual amplitude and even in a change to a winter maximum in the east of Uruguay and in the coastal area of southern Brazil. In eastern Argentina, rainfall south of 25°S presents a bimodal structure with on maximum in October–November and the other during March–April, and an absolute minimum in winter. These features can be seen in Fig. 1 of Grimm et al. (2000), and they are described in a more detailed way in Montecinos et al. (2000).
In spring, the interannual standard deviation of the rainfall is about one-third of its mean, Table 1. Regions included in Table 1 are shown in Fig. 1. They were selected according to a canonical correlation analysis (CCA) between SSA precipitation and SST, as will be explained in section 4. Part of the interannual variability is related to EN (LN) (Table 2a) that enhances (reduces) the physical processes related to rainfall conditions (Grimm et al. 2000). However, as it has been commented earlier, there is considerable rainfall variability among EN events, particularly in the northeast, and to a lesser extent, among LN cases (Table 1). Figure 2 illustrates this aspect, showing the spring rainfall of the northern region (N). There are EN cases with lower than normal rainfall, and also LN cases with higher than normal precipitation. A similar behavior is observed in the NE and SE regions (not shown).
4. SST–SSA precipitation relationship
A CCA (Barnston 1994; Grahan et al. 1987; Preisendorfer 1988) between seasonal SST and precipitation over SSA was carried out to study possible remote forcing on the precipitation field. The leading modes are more significantly correlated with precipitation over SSA during the austral spring. It is also during the austral spring that the strongest signals of EN and LN events are observed. Hence, this paper will focus on this season.
Figures 3 and 4 show the correlation between precipitation and the first and second canonical modes for the austral spring, the respective SST correlation fields with these canonical modes, and their time series. Each figure shows only one of the two time series, given that they are practically identical and their correlation being 0.99 in the first mode and 0.98 in the second. Correlation significance was calculated with the Student's t-test (Panofsky and Brier 1965). In the first mode, the strongest and more significant correlations are over southern Brazil, northeastern Argentina, and eastern Paraguay. Opposite and significant correlations also appear in southwestern Argentina. The rest of the region has little or no significant correlation. Regarding SST, there are some areas of significant correlation in the Pacific, in the South Atlantic and even in the Indian and the North Atlantic Oceans. The center in the EN-3 region has the same sign as the northeastern nucleus of the correlation with precipitation, while the two areas in the subtropical regions of each hemisphere have opposite signs. This means that cooler temperatures in the subtropical south-central Pacific Ocean between 120° and 170°W (SSCP) tends to be associated with enhanced rainfall in the northeastern region.
The SST correlation pattern of the first mode of the CCA, with weak negative values at the western equatorial Pacific Ocean and positive values at the subtropical latidues of both hemispheres, resembles the dominant feature of the interdecadal Pacific oscillation (Power et al. 1999). This feature can also be partially seen in the North Pacific decadal mode described by Zhang et al. (1997) who analyzed the SST field north of 30°S, and in the interdecadal mode of GB. This suggests that SST in the SSCP has considerable low-frequency variability, as can be confirmed from the visual inspection of Fig. 5. This is an indication that part of the rainfall variability over eastern SSA related to the subtropical western Pacific SST may belong to this scale. In fact, the time series of the interdecadal mode of GB, their Fig. 1, has a change in its mean value in 1976. This shift in the interdecadal mode was part of a well-documented global climate shift (Nitta and Yamada 1989; Trenberth 1990; and others). The time series of our CCA first mode has a similar but less abrupt change only a few years earlier, that is, around 1972 (Fig. 3). As it might be expected, it was accompanied by a sudden increase in precipitation in the northern part of SSA (Barros et al. 2000). This change was simultaneous and consistent with the growth of the Paraná River discharge documented by García and Vargas (1998).
The second mode has significant correlations with precipitation over most of the region with a maximum over Uruguay and part of eastern Argentina. Though the CCA analysis does not specifically separate timescale variability, the second mode seems to carry less interdecadal variability than the first (see time series in Figs. 3 and 4). As in the first mode, in the SST field there are various areas with significant correlation, most of them almost coinciding with those of the first mode. However, correlations over the Pacific Ocean are stronger than in the first mode, and the equatorial maximum is displaced to the EN-3.4 region (shown in Fig. 7). The fields in the Pacific and Atlantic Oceans are quite similar to the projection of the first mode of the principal component analysis (PCA; Preisendorfer 1988) on the global SST that carries most of the SST variability associated with ENSO (Fig. 6). This first PCA mode, that explains 20% of the variance of the SST field, has the strongest correlation in the Pacific Ocean. In this ocean, there are two centers with opposite correlation signs with respect to the equatorial region located in substropical and midlatitudes of both hemispheres. This feature has been shown in many studies, and it is part of both the interannual (ENSO) and the interdecadal (ENSO-like) variability (GB), denoting that variations of SST in SSCP tend to be associated with opposite sign variations of the equatorial SST. The spatial structure of the first PCA mode suggests that SSTs in EN regions and in the SSCP, as well as in its counterpart in the North Pacific (NP), are all reciprocally correlated. These ocean regions are depicted in Fig. 7. In fact, the correlation between the SST in the SSCP and SST in EN-3 and EN-3.4 is −0.58 and −0.63, respectively; while between subtropical SSTs, SSCP and NP, is 0.41. Also, the correlation between the SST in the NP and SSTs in EN-3 and EN-3.4 is −0.54 and −0.52, respectively. These correlations are all significant at the 95% level of confidence. However, when the series of SST in the SSCP and in the NP regions are regressed upon the SST of the EN-3.4 region and the residual time series are calculated, the correlation between these residuals is only 0.13, and it is not significant. This may indicate that, except for the common association with the equatorial SST variability, the SSTs of these two regions are not linked. Yet, this is not a definite conclusion, since both variables may be related not in a simultaneous manner but with certain lags or they may have some low-frequency link.
Diaz et al. (1998) made a canonical correlation analysis between global SST and seasonal precipitation. Their analysis was restricted to Uruguay and a part of southern Brazil, and their results were consistent with those presented herein for the region.
The CCA makes it possible to regionalize the precipitation field according to its correlation with the leading modes. Therefore, precipitation regions were selected according to their correlation with the first two CCA modes (Figs. 3 and 4) grouping the initial 5° × 5° boxes into larger areas. This regional classification was adopted only for the purpose of discussing the relationship of SST with the spring precipitation in SSA, and other classifications may certainly be more convenient for other purposes. Over SSA, precipitation has the strongest response to SST in the east, where it is correlated with the first canonical mode in the northern region, and with the second mode in the southern one. Thus, three regions were selected in the eastern side of SSA: N in the north, SE in the south, and NE, which shares the influence of both modes, in the middle (Fig. 1). The rest of the subtropical area east of the Andes was divided into another three regions. The northwest (NW) responds to the second mode and has a small interannual variability, especially among EN and LN cases (Table 1). In the west (W) there is significant correlation with the first mode and small correlation with mode 2, while in the extreme southeast (S) there is a relatively weak response to mode 2.
Table 2 presents the correlation between these regions and the SST of the ocean areas that appear more correlated with the first two modes of the CCA. Table 2 includes not only the correlation for all of the 39 spring cases (1952–90) but also the correlation among EN, LN, and the neutral cases. When all of the years are considered, precipitation in eastern SSA is correlated positively and almost everywhere significantly with SST in EN-3 regions. However, among EN cases, this is only true in the northeastern region N. Moreover, in most of subtropical Argentina, regions SE, NW, and S, this correlation is negative, though not significant (Table 2b). Similar results are found with the EN-3.4 region, except for smaller correlations with region N. In other words, considering only EN events, the SST in the EN-3 or EN-3.4 regions does not correlate significantly with the precipitation over most of SSA.
The same conclusion can be reached for LN cases, except in regions SE and NW, where precipitation is in fact strongly and positively correlated with SST in both EN-3 and EN-3.4 regions (Table 2d). On the other hand, except for the southernmost regions, correlation among neutral years between SST in both EN regions and regional precipitation are always positive, and in certain cases, even significant (Table 2c).
These results indicate that except in region N, SST in EN regions does not modulate the dispersion in rainfall among EN years or among LN years over SSA. However, it does modulate the rainfall between EN, LN, and neutral cases, indicating that the SSA rainfall response to SST in EN regions is not linear over the observed SST range, becoming approximately constant at both extremes of this range. To illustrate this point, scatter diagrams between seasonal rainfall at region NE and SST in these ocean regions are presented in Fig. 8. The SSCP and the tropical North Atlantic (TNA) are the only oceanic region whose SST modulates rainfall among EN cases, adding valuable information for prediction at the seasonal timescale (Table 2b and Fig. 8c). In the case of SST in the SSCP, this is possible because it is not dependent on the equatorial SST among EN cases, as can be seen in the scatter diagram between the SSCP and EN-3.4 spring SST (Fig. 9). In this figure, WC stands for the group of cases with extreme warm (W) equatorial SST and extreme cold (C) SST in the SSCP region with a similar convention CW, WW, and CC used to define groups. In the neutral cases there is no dependence between these regional SSTs, while in the case of LN there is some dependence for EN-3.4 SST below 25.5°C. Similar results are found when EN-3 instead of the EN-3.4 region is considered (not shown). However, when all the cases are considered together, there is a clear dependence between both SST (Fig. 9). Hence, since CCA was calculated taking into consideration all the years, it is not surprising that it does not separate the subtropical South Pacific SST and the equatorial Pacific SST variability into different leading modes.
As for LN, neither the equatorial Pacific nor the SSCP SSTs modulate the rainfall over most of SSA. The modulaiton is done by the SSTs of the nearby subtropical South Atlantic (SSA) and of the TNA as shown in Table 2d. In regions SE and NW, however, both EN-3 and EN-3.4 modulate the spring rainfall with a positive correlation.
In the NE region, rainfall has higher absolute value correlation with SST in the SSCP than with SST in EN regions among EN cases, as discussed before. In addition, when all of the years are considered, this SST has correlations similar in magnitude to those of the SST in EN regions (Tables 2a and 2b). In this later case, the correlations between SST in the SSCP and precipitation in regions N and NE, although not higher than in the case of SST in the EN regions, are still significant. This raises the question of which mechanisms cause the interannual response in spring rainfall in SSA in connection with SST variability in the equatorial Pacific and in the SSCP. In the next section, this issue is addressed through a discussion of the circulation patterns associated to these SSTs. Another question that has direct practical implications for seasonal prediction is in what measure would the diagnostic relationship between SSA rainfall and SST improve if SSCP SST anomalies were used in conjunction with SST anomalies in EN regions. Using a multiple regression technique, some improvement is found when only EN events are considered. This could be anticipated by inspection of the correlation coefficients shown in Table 2. The improvement is considerable for region N, less important for regions NE, SE, and W, and null in the others (Table 3). This simple exercise does not exhaust this issue, but it gives a hint about the progress that can be achieved in rainfall prediction adding SST in SSCP as predictor in EN cases.
Table 2 also shows the correlation of SSTs in other regions with regional precipitation in SSA. We excluded SST in the NP region because, according to the previous description, its correlation with SSA precipitation can be caused by its correlation with EN regions. They do not reach the 95% significance level, except with two exceptions in the cases of the Indian Ocean zone and of the TNA (Tables 2a and 2d). It is beyond the scope of this paper to discuss the possible physical links between spring precipitation and SST in each of these areas with significant correlation. Some of the mechanisms that have a relationship with SST and can affect precipitation over SSA were addressed by Mechoso and Robertson (1998) in the case of the tropical Atlantic Ocean, and by Robertson and Mechoso (2000) and Barros et al. (2000b) in the case of the subtropical South Atlantic Ocean. Therefore, since SST in the SSCP together with SST in EN regions have the strongest signals on SSA precipitation, the rest of the paper will be devoted to discuss the atmospheric patterns related to these SSTs.
5. Atmospheric circulation patterns associated with SST variability at the SSCP
According to the results of the previous section, it will be convenient to explore how the SSA circulation anomalies are associated to the SSCP interannual variability and also how they are related to the ENSO signal. Therefore, it is opportune to briefly review this signal. Grimm et al. (2000) discussed the anomaly circulation along each phase of ENSO over South America and the neighboring oceans, finding that the greater anomalies tend to be approximately at the same location during EN and LN, but with opposite signs. Therefore, the composite difference between EN and LN cases provides an insight for both signals in a synthetic and at the same time enhanced form.
Figure 10a shows the difference between the composite mean of the 200-hPa streamfunction field for the austral spring of EN and LN years. There is a wave train over the Pacific Ocean that starts with two anticyclonic circulations at low latitudes of both hemispheres centered east of the data line. It extends to the southeast down to the west of the Antarctic Peninsula, where it turns to the northeast toward the east of South America. At subtropical latitudes, this implies an enhanced subtropical jet over most of the Pacific Ocean. This wave train has been documented by Karoly (1989) for the winter 200-hPa geopotential, and attributed to barotropic Rossby wave propagation. Also, the structure of Fig. 10a is basically similar to one of the leading modes of the interannual variability of the 300-hPa streamfunction related to ENSO described by Kidson (1999), but in our case with decreasing amplitude with latitude. Near South America, this pattern appears as a strong dipole on the eastern Pacific with negative differences in subtropical latitudes and positive differences in mid- and high latitudes. There is another dipole with inverse sign, which is less intense over the east of South America and the South Atlantic, at approximately the same latitudes. As a result, there is an intensification of the cyclonic vorticity advection over SSA in EN years with respect to LN cases. At surface level, there is a similar pattern at mid- and high latitudes, indicating that this signal is basically equivalent barotropic at these latitudes as discussed by Karoly (1989) and others (Fig. 10b). The main features of this SLP difference field, which is positive over Australia and the Indian Ocean, negative over the eastern Pacific, and positive to the southwest of South America, was documented by many authors (e.g., van Loon and Shea 1987). Over eastern SSA, the difference field shows that during EN years, the advection of warm and humid air from the humid tropical continent is enhanced with respect to LN years. This is the most important source of atmospheric water vapor for SSA (Nogues-Paegle and Mo 1997). Furthermore, the anomalies over the South American region could be linked to the mechanism explained by GB in the case of EN winter that is summarized as follows. The enhancement of the anticyclonic circulation to the southwest of South America reflects the increase of blocking episodes, as reported by Rutlant and Fuenzalida (1991), and Renwick (1998). This feature, together with the weakening of the subtropical high, reveals that there is a northward shift of the cyclone track, as confirmed by the higher frequency of cyclones over subtropical South America during EN events (Sinclair et al. 1997). The same mechanism could be acting in spring for EN cases (Fig. 10b). Therefore, both the upper- and the lower-tropospheric circulation contribute to enhance the rainfall processes during EN years with respect to LN years, as it is in fact observed in all of the SSA subregions (Table 1).
As mentioned earlier, the scatter diagram between SST in EN-3.4 and in the SSCP regions shows a dependence of these two SSTs. However, when only EN cases are considered, there is no dependence between these two SSTs (Fig. 9). Therefore, a stratified difference between EN years with extreme SST in the SSCP isolates these SST effect on the circulation from those of the EN-3.4 SST. In fact, by choosing the four lowest and highest SSCP cases, the respective EN-3.4 or EN-3 mean SSTs are practically the same (Table 4). In this selection, we excluded the 1957 EN because of the reasons discussed in section 2. The composite differences of these cases for the 200-hPa streamfunction and SLP are presented in Fig. 11. The wave train pattern over the western Pacific Ocean, and even over the South Atlantic Ocean, is remarkably similar to—although with less amplitude than—the one associated with EN minus LN cases over the tropical and subtropical latitudes of the Pacific Ocean. The anticyclonic center to the west of the Antarctic Peninsula is indeed equally strong. Another dissimilarity is that the wave train is practically meridional oriented over the Pacific Ocean, and except at high latitudes, is farther to the west with respect to the EN minus LN pattern. It can be concluded that the coldest anomalies in the SSCP during EN enhance the typical high-tropospheric circulation in the western part of the Southern Hemisphere associated to these events. Therefore, the enhancement of cyclonic advection of the mean circulation over South America is consistent with the precipitation response in eastern South America to SST in the SSCP during EN events (Fig. 8). Furthermore, the mechanism discussed by GB and described in the previous paragraph is probably more active in EN events with colder SST in the SSCP, because the anticyclonic anomaly to the southwest of South America is enhanced in these cases (Fig. 11).
The difference at 200 hPa between cases with the coldest and warmest SSTs in the SSCP of EN years (Fig. 11) does not only have a similar structure to the EN minus LN pattern in most of the western Southern Hemisphere, but it also has a near-equivalent-barotropic structure at mid- and high latitudes. However, there is a dissimilitude in the SLP pattern over South America that implies a reversal of the low-level anomaly circulation with respect to EN minus LN due to the shift in longitude (Figs. 10b and 11b). Therefore, the coldest SSTs in the SSCP are associated with inhibited mean advection of warm and humid air over most of SSA. Nevertheless, rainfall in regions N and NE is negatively correlated with SST in the SSCP (Table 2b) indicating that another mechanism prevails over the diminished water vapor advection, presumably the one described by GB, which we have extended to the spring in previous paragraphs. On the other hand, consistent with the low-level circulation, this does not occur in the western regions (Table 2b), where according to the annual cycle, water vapor advection seems to be the limiting factor for rainfall (Doyle and Barros 2000).
When a similar composite difference between the coldest and the warmest SST in the SSCP is calculated using only neutral cases, a weaker but otherwise similar wave train pattern to those shown in Fig. 11 is insinuated (not shown). This reinforces the idea that the variability of SST in SSCP is a contributing forcing to the EN minus LN pattern.
In the PCA, the first mode shows that cold (warm) anomalies at the SSCP are associated with warm (cold) SST in the equatorial Pacific (Fig. 6). The preceding stratified analysis permits one to attribute to the SST variability in the SSCP part of the anomalous circulation over the western Southern Hemisphere associated with EN events (Figs. 10 and 11). Therefore, it is appropriated to determine the contribution of the equatorial SST variability to the EN minus LN circulation signal. According to Fig. 9 and Table 4, the mean of the warmest SST in the SSCP during EN cases is approximately the same as the mean of the coldest SST in the SSCP during LN cases; namely, WW and CC cases. This allows stratification according to EN and LN cases independent of SST in the SSCP. Again, the EN and LN events before 1958 were not included in this grouping. Figure 12 shows the composite differences of these cases. At 200 hPa, the spatial structure over tropical and subtropical latitudes is similar to the one corresponding to the complete EN minus LN case (Figs. 10a and 12a). In the case of the tropical anticyclonic circulation of the Northern Hemisphere, both cases are practically identical near the date line indicating that this signal is mostly induced by the equatorial SST variability, and not by the SST in the SSCP (Figs. 10a and 12a). However, this circulation is less intense than in the EN minus LN case over the eastern Pacific Ocean, indicating a possible forcing from the SSCP. On the other hand, the amplitude of the wave pattern over the Pacific Ocean decreases toward the south, and almost vanishes to the west of the Antarctic Peninsula, where in the EN minus LN case there is a strong anticyclonic circulation. This distinction is less conspicuous at surface level (Figs. 10b and 12b), but the anticyclonic center is considerably weaker than in the EN minus LN case. The wave pattern is shifted to the east by almost 30° in longitude with respect to the case with extreme SST differences in the SSCP of EN cases. This is related to the different position of the thermal forcing, and will be discussed in the next section.
If instead of using EN-3.4 SST, as it was done in this section, EN-3 SST were used, the preceding results would be identical, because the years with extreme SST are the same in both cases. Therefore, it can be concluded that the circulation anomaly field over the western part of the Southern Hemisphere associated to EN or to EN minus LN cases responds not only to the equatorial SST variability, but also to that of the SSCP. This conclusion is also valid for the respective regional anomalies over southern South America and consequently, it is consistent with the correlation between precipitation in eastern South America and SST in the SSCP.
6. Relative importance of SSCP and equatorial Pacific SSTs in the atmospheric difference pattern of the Southern Hemisphere associated with EN and LN events
Over the western Southern Hemisphere, many of the features of the circulation field associated with EN minus LN events (ENSO-like pattern) are basically reproduced when stratified differences are taken either at quasi-constant equatorial SST (cold minus warm SST in SSCP in EN cases) or at quasi-constant SST in the SSCP (warm minus cold equatorial SST cases). Therefore, it is possible that the difference between the more extreme cases in both SSTs will produce a more amplified ENSO-like pattern. Figure 9 indicates that in the case of LN there is some dependence between EN-3.4 and SSCP SST. However, if the 1973 case is not considered, the coldest and warmest LN in the SSCP, namely, CC and CW cases, have almost the same mean SST in the EN-3.4 region (Table 4). Therefore, it is possible to separate the signal between the extreme cases, WC and CW, into two parts: one with quasi-constant SST in the SSCP, and another with quasi-constant SST in the EN-3.4 region (Fig. 9). In fact, since the composition and difference of fields are linear operations
where the meaning of WC, CW, WW, and CC was already explained in section 4, when Fig. 9 was first introduced. Hence, the difference between extreme cases, the right-hand side of Eq. (1), can be decomposed in the difference between cases with an almost constant SST in the SSCP region, the first term in parentheses in the right-hand side of Eq. (1), and in differences between cases with an almost constant SST in the EN-3.4 region, the term in brackets in the right-hand side of Eq. (1).
Figure 13 shows the difference fields of the 200-hPa streamfunction between the extreme cases described by the left-hand side of Eq. (1). The streamfunction difference pattern of the Western Hemisphere at 200 hPa is very similar to the EN minus LN case. As expected, the amplitude of the wave train extending from the central tropical Pacific to the west of the Antarctic Peninsula is enhanced with respect to the EN minus LN case (Figs. 10a and 13a). This enhancement is stronger in the anticyclonic circulation to the west of the Antarctic Peninsula and in the wave train over the South Atlantic–South American sector. Since SST differences in the EN-3.4 region were approximately similar in both cases, this enhancement on the signal can be attributed to the greater SST difference at the SSCP. This is confirmed when the respective fields associated with the first and second terms on the right-hand side of Eq. (1) are compared (Figs. 12 and 14).
The difference field associated to the first term on the right-hand side of Eq. (1) was already been discussed (Fig. 12) whereas the one corresponding to the second term is shown in Fig. 14. From now on, they will be referred to as the equatorial and the subtropical forced fields. Although at 200 hPa, both present an ENSO-like wave train in the Western Hemisphere, they are shifted in longitude by approximately 30°. This is precisely the difference in latitude of the anomalous thermal forcing in between them, as revealed by the vertical motion (Fig. 15). In the case of the subtropical forcing, the subtropical circulation center at 200 hPa coincides approximately with the thermal forcing. Despite of the forcing being at different latitudes, both cases may have the vorticity source at about the same latitude. This is because in the case of the equatorial forcing, the vorticity source is displaced away from the divergence anomaly to subtropical latitudes (Sardeshmukh and Hoskins 1988). In fact, in the equatorial forced field, the anticyclonic circulation at 200 hPa is centered at 20°S and coincides approximately in longitude with its thermal source. The anticyclonic circulation on the tropical Northern Hemisphere west of the date line, that appears in the difference of extreme cases, left-hand term of Eq. (1), should be attributed mostly to the equatorial forcing. At these longitudes, it is considerably smaller in the case of the subtropical forcing (Figs. 12a and 14a). However, the subtropical forcing contribution outweighs the signal associated with the equatorial forcing over the eastern Pacific of the Northern Hemisphere. This feature is consistent with the Rossby wave meridional propagation theory, since at high levels there is westerly flow over the equatorial eastern Pacific, and therefore there is not an evanescent effect over the subtropical forced wave train. Based on the same argument, the Southern Hemisphere patterns resulting from the composite analysis could have some influence from the Northern Hemisphere. However, this is unlikely because the only SST that may force a meridional Rossby wave propagation through the eastern Pacific, namely, SST in the NP, is only related to SSCP through equatorial SSTs, and these remain practically constant when we consider the subtropical forcing.
If the anticyclonic circulation west of the Antarctic Peninsula in the subtropical forced case is not considered, in both cases there is an almost meridional propagation from the source region up to 60°S where they diminish their amplitude and are refracted first to the east, and then turn to the northeast. This is consistent with what should be expected in the Southern Hemisphere (James 1994) from linear ray tracing theory (Hoskins and Karoly 1981) for a local zonal wavenumber near 2 (Figs. 12 and 14).
The anticyclonic circulation to the west of the Antarctic Peninsula is almost entirely associated with the subtropical forcing at the SSCP (Figs. 12a, 13a, and 14a). In the subtropical forced field, it is almost 30° to the west of the longitude of the cyclonic center at midlatitude, a feature not consistent with linear Rossby wave meridional propagation theory.
For the rest of the wave train at 200 hPa over the Pacific Ocean, both terms of the right-hand side of Eq. (1) contribute approximately the same to the wave amplitude. However, the zonal shift between them, together with the different rate of decrease with latitude makes their sum, left-hand side of Eq. (1), to have a wave train with southeast propagation. Therefore, this feature of the ENSO-like pattern in the western Southern Hemisphere is not the result of the equatorial forcing alone but is actually caused by the forcing of both, the equatorial and the subtropical SST forcing.
Despite the equatorial forced circulation being initially shifted westward with respect to the subtropical forced case, the subtropical jet enhancement extends farther east, reaching South America. A possible explanation for this is that variations in the equatorial SST modify the moist content of the troposphere more than variations in the subtropical SST, where subsidence predominates. The higher water vapor content leads to enhanced and earlier cyclonic development on one side, and prevents the decoupling of the low and high circulation on the other, contributing to sustain an eastward trajectory of cyclones (Orlanski 1998, 2000).
As it has been explained before, there is a zonal shift between the equatorial and the subtropical forced fields over the southern South America and western Atlantic Ocean sector (Figs. 12a and 14a). However, their sum produces a considerable anticyclonic circulation centered off the coast of southern Brazil (Fig. 13a), which implies an enhanced cyclonic vorticity advection in EN cases with cold SST in the SSCP with respect to LN cases with warm SST in the SSCP.
At SLP, the mid- and high-latitude SLP anomalies on the Western Hemisphere associated to the subtropical forcing are also considerably amplified with respect to EN minus LN case (Figs. 10b and 14b). Although the anticyclonic signal west of the Antarctic Peninsula appears in the pattern associated to the equatorial forcing, it is considerably weaker than the respective anticyclonic signal associated to the subtropical forcing (Figs. 12b and 14b).
Finally, it should be mentioned that if SST in the EN-3 region were used instead of SST in the EN-3.4 region, the preceding analysis would not change as years with extreme SST in EN and LN cases remain the same.
During OND, the SST in EN regions does not modulate the rainfall variability among EN years or among LN years over most of SSA. However, it does modulate rainfall between EN, LN, and neutral cases and, also among neutral years. This indicates that in the austral spring, the response of rainfall in SSA to SST in El Niño regions is not linear, becoming independent of it at both extremes of its registered range.
On the other hand, SST in the SSCP modulates rainfall over most of SSA among EN events during OND, even though among EN cases this temperature is independent of the SST in EN regions. This occurs because the cold minus warm SST cases in the SSCP during EN events are associated with an atmospheric circulation pattern over the western Southern Hemisphere, that is similar, in many aspects, to the ENSO-like pattern. In the upper troposphere, this pattern favors the cyclonic advection over SSA, but at low levels inhibits the advection of humid air from the north. As a result, the enhancement of the dynamic factors only leads to more rainfall in the east of the SSA where there are other sources of water vapor. In addition, the strong enhancement of the anticyclonic circulation to the west of the Antarctic Peninsula during EN with respect to LN is mostly associated to the forcing in the SSCP and very little to the equatorial forcing. Therefore, this may favor a northward shift of the cyclone track during periods with cold SST in the SSCP, increasing the precipitation in the eastern regions.
The ENSO circulation signal in the Southern Hemisphere is forced not only by the equatorial SST variability in EN regions but also by SST variability in the SSCP. This can be seen when the circulation field associated to the difference between EN and LN events with extreme SSTs at the SSCP is decomposed into one part with almost constant SST in EN regions, and another with the same characteristic in the SSCP. In the case of the SST forcing at the equator, the streamfunction response in the tropical latitudes corresponds with theory, as the anticyclonic circulations are displaced to higher latitudes of the upper divergence (Sardeshmukh and Hoskins 1988). For the subtropical SST forcing, the vertical motion response is centered at 20°S and 140°W coinciding with the anticyclonic circulation. In both cases, there is a wave train that propagates almost from north to south over the Pacific Ocean, turning to the east at high latitudes, and then to the northeast either, over South America or over the Atlantic Ocean. The respective wave trains are shifted one from the other around 30° in longitude, responding to similar shift in their respective thermal forcing.
Finally, the southeastern propagation of the wave train in the ENSO-like pattern of the western Southern Hemisphere does not result from the equatorial forcing alone, which actually has a meridional propagation. Rather, it is caused by the forcing of both, the equatorial and the subtropical SST.
This work was funded by the University of Buenos Aires under Grant UBACYT TW075, and by the ANPCyT under Grant PICT 00561.
Corresponding author address: Dr. Vicente R. Barros, Department of Atmospheric Sciences, University of Buenos Aires, CONICET, Cuidad Universitaria, Pabellon II, 2° Piso, Buenos Aires 1428, Argentina. Email: firstname.lastname@example.org