Abstract

The existence of a significant simultaneous correlation between bimonthly mean precipitation anomalies over southeastern South America (SESA) and either the first or the second (depending on season) leading mode of interannual variability of upper-level wind over South America (SA) is demonstrated during all seasons except winter. The pattern associated with these modes of variability is similar during all seasons and consists of a continental-scale vortex centered over the eastern coast of subtropical SA. The vortex has a quasi-barotropic structure during all seasons, and its variability modifies moisture transport from the South American low-level jet and the western tropical Atlantic to SESA thus creating precipitation anomalies in this region. During spring (October–November) and summer (January–February) the circulation creates a second center of precipitation anomalies over the South Atlantic convergence zone that are of opposite sign to those over SESA, while during fall (April–May) precipitation anomalies are primarily confined to SESA. On the basis of the correlation between upper-level winds and precipitation, an empirical method to produce long-range forecasts of bimonthly mean precipitation over SESA is developed. Method tests in hindcast mode for the period 1959–2001 show a potential for reliable predictions during the southern spring, summer, and fall. The method is further tested in an experimental mode by using Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) wind hindcasts. Forecasts obtained in this way are skillful during spring only, with highest skill during El Niño–Southern Oscillation years. During summer and fall, the DEMETER forecasts of wind anomalies limit the method’s ability to make reliable real predictions.

1. Introduction

Reliable climate forecasts help decision makers in the planning of effective responses to departures from the mean seasonal cycle (seasonal anomalies) in anticipation of their occurrence. Precipitation anomalies can have strong impacts on human activities, especially in regions strongly dependent upon agriculture and hydroelectricity production such as the La Plata River basin in southeastern South America (SESA). In SESA, climatological precipitation has small seasonal variations (e.g., Montecinos et al. 2000). Summer and spring precipitation is part of the South American monsoon system, whereas fall and winter precipitation originates from synoptic weather systems. Seasonal precipitation anomalies over SESA have a dominant pattern of uniform sign over the region and strong interannual variations with values ranging from 50% to 150% of the mean (Fig. 1). The SESA anomalies form the southern component of an approximately north–south dipole of precipitation anomalies over South America (SA) during spring and summer (e.g., Casarin and Kousky 1986; Paegle and Mo 1997; Paegle et al. 2002; Robertson and Mechoso 2000; Díaz and Aceituno 2003). The dipole variability has been linked to anomalies in other outstanding climate features such as the intensity of the South Atlantic convergence zone (SACZ), South American low-level jet (SALLJ), Pacific–South American (PSA) modes, and El Niño–Southern Oscillation (ENSO). Several papers extensively document such links (e.g., Robertson and Mechoso 2000; Díaz and Aceituno 2003; Mo and Paegle 2001; Paegle and Mo 2002; Liebmann et al. 2004; Ferreira et al. 2003; Grimm 2003; Grimm et al. 2007; Silva et al. 2009).

Fig. 1.

Precipitation (shaded, mm day−1) and vertically integrated moisture flux [vectors, hg (m s)−1] climatology over SA for (a) ON, (b) JF, and (c) AM and interannual variability (solid line) of the mean precipitation over SESA (mm day−1) and climatology (dotted line) for (d) ON and (e) AM. The box in panel (a) represents SESA.

Fig. 1.

Precipitation (shaded, mm day−1) and vertically integrated moisture flux [vectors, hg (m s)−1] climatology over SA for (a) ON, (b) JF, and (c) AM and interannual variability (solid line) of the mean precipitation over SESA (mm day−1) and climatology (dotted line) for (d) ON and (e) AM. The box in panel (a) represents SESA.

During the austral spring ENSO is significantly and positively correlated with rainfall variability in subtropical SA (Montecinos et al. 2000); this relationship serves as the foundation for interannual predictions of rainfall in that region. The different phases of ENSO (El Niño and La Niña) tend to generate precipitation anomalies with opposite sign, which indicates a large degree of linearity in the teleconnection with the tropical Pacific Ocean (Grimm et al. 2000).

ENSO impacts are found during summer as well, though with weaker intensity. It has been suggested that the effects of the South American monsoon system predominate during summer over those associated with remote forcings (e.g., Cazes-Boezio et al. 2003; Grimm 2003). Paegle and Mo (2002) described ENSO’s impacts on SA during summer by means of EOF analysis of rainfall and found anomalies of opposite sign over northeastern Brazil and SESA. Cazes-Boezio et al. (2003) outlined a dynamical picture of ENSO’s impact on the South American climate by EOF analysis of upper-level winds and regression maps of the leading principal components (PCs) over different fields. Their results show the leading mode of upper-level variability over SA during spring to be part of a wave pattern that is highly positively correlated with sea surface temperatures (SSTs) in the tropical Pacific (up to 0.6 in the El Niño-3.4 region). The pattern is also significantly and positively correlated with anomalies in the vertical velocity over northeastern SA and negatively correlated with those over SESA. (A consistent result is obtained, by the same authors, for the mean value of rainfall anomalies obtained by 13 rainfall gauge stations over Uruguay.) Cazes-Boezio et al. (2003) proposed that the interannual variability of upper-level wind can be interpreted as changes in the frequency of occurrence of intraseasonal circulation regimes over the South Pacific, which would be triggered by ENSO during spring and would originate from nonlinear processes during fall–winter. A different perspective is that Rossby wave propagation from the central and western equatorial Pacific contributes to the development of the leading PSA (PSA1) during all seasons (Mo 2000; Mo and Paegle 2001). Precipitation anomalies over SA have been interpreted as downstream effects of the PSA1 and PSA2 modes (Mo and Paegle 2001) during summer and spring, respectively; the same authors did not investigate similar relationship for precipitation over SA during the fall season.

Doyle and Barros (2002) suggested tropical and southwestern Atlantic SST anomalies as source of predictability during summer. Cazes-Boezio et al. (2003) found significant correlations between the second mode of upper-level wind variability over SA and rainfall over SESA and a north–south dipole in the SSTs field over the South Atlantic.

Diaz et al. (1998) identified a connection between Atlantic SST anomalies under the oceanic extension of the SACZ and precipitation over southern Brazil and Uruguay during fall–early winter. In their canonical correlation analysis warm (cold) SST anomalies correspond to wet (dry) rainfall anomalies. Cazes-Boezio et al. (2003) reported the same type of relationship by correlation analysis for the fall–winter season.

This short review demonstrates that various processes have been recognized as modulators of seasonal rainfall over SESA on interannual time scales. Current operational models, however, seem unable to exploit the associated potential for predictability of seasonal precipitation anomalies in the region. For example, the International Research Institute for Climate and Society (IRI) provides, since 1997, probabilistic forecasts for below-, near-, and above-normal for precipitation and temperature, one month in advance for the upcoming three months. The forecasts are obtained by combining products of dynamical models and simple empirical prediction methods based on the phases of ENSO. The skill of IRI seasonal forecast over subtropical SA, as evaluated by ranked probabilistic skill score (RPSS) with respect to climatology (Wilks 1995), is between −0.1 and 0.1 for all seasons except for the austral spring, when the skill can increase up to 0.3 (Goddard et al. 2003). Starting December 1997, the Climate Outlook Forum (COF) has been producing seasonal precipitation forecasts for SA east of the Andes and between 20° and 40°S. The forecasts are obtained by combining outputs of several numerical and statistical models, results of diagnostic analysis, and statistics on the regional climate variability and are released in the same format as those from IRI. Berri et al. (2005) found that the average RPSS of COF forecasts for all seasons is negative over most of SESA. The European Centre for Medium-Range Weather Forecasts (ECMWF), together with other research and operational centers in Europe, created the Development of a European Multimodel Ensemble System for Seasonal-to-Interannual Prediction (DEMETER) project in order to assess the capability of a multiple GCM ensemble forecast system to produce seasonal to interannual forecasts (Palmer et al. 2004). The GCMs participating in the project produced hindcasts from 1959 to 2001. The multimodel ensemble mean hindcast by ECMWF, the United Kingdom Met Office (UKMO) and Météo-France (METF) captures the main pattern and intensity of precipitation climatology over SA during all seasons. Nevertheless, precipitation anomaly forecasts over SESA present small departures from the climatology, and most of the observed values fall outside of the 95% prediction interval as defined by the variance of the ensemble mean. Hindcasts have marginal skill when GCMs are run the same month the prediction targets, as during November–December (ND) and May–June (MJ). Recently, Coelho et al. (2006) have developed an integrated seasonal forecast system that combines an empirical model, which uses Pacific and Atlantic SSTs as predictor of precipitation over SA, and a multimodel coupled system. For the southern summer the method generally produces marginally skillful predictions over SESA (see their Fig. 9), while better predictions result in the occurrence of ENSO events (see their Fig. 10).

In this study we are interested in identifying other predictors of seasonal precipitation over SESA during all seasons and in assessing their potential for improved predictions. We start by finding a relationship between interannual variability of precipitation over SESA and of winds at upper levels. Next we provide a physical interpretation of the documented relationship. Finally, we explore how the relationship found can contribute to produce long-range reliable seasonal precipitation forecasts over SESA.

The datasets we use are described in section 2. Section 3 presents the physical relationship between upper-level wind and precipitation variability. Section 4 describes our proposed methodology for long-range precipitation forecasts and discusses the expected forecast skill. Section 5 includes a summary and concluding remarks.

2. Data

Our main focus is on the region 24°–38°S, 60°–45°W, which represents SESA (Fig. 1a). For the selected region, we represent seasonal mean precipitation by the domain average over the central 2-month periods during each season. This procedure captures the coherent signal throughout each season and filters high-frequency variability. The bimonthly periods, therefore, are October–November (ON), January–February (JF), April–May (AM), and July–August (JA). We investigate the period 1959–2001.

The precipitation dataset was compiled by the Climate Research Unit at the University of East Anglia (http://www.cru.uea.ac.uk/cru/data/hrg.htm). In this dataset, values of monthly precipitation were reconstructed from gauge measurements. The fields are available at 0.5° × 0.5° horizontal resolution over land during the period 1901–2002 (New et al. 1999, 2000).

The wind fields at 1000, 925, 850, 700, 600, 500, 400, 300, and 200 hPa; latent heat flux; and specific humidity correspond to the global reanalysis at the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR). This dataset is available at 2.5° × 2.5° horizontal resolution from 1948 to the present (Kalnay et al. 1996).

For SSTs we use the National Oceanic and Atmospheric Administration’s (NOAA) extended reconstructed SST dataset (Smith and Reynolds 2004). In this dataset, monthly mean values are available for the global ocean at 2° × 2° horizontal resolution, from 1855 to the present.

For the tests of our method in a realistic forecast framework, we use the predictions of 200-hPa winds produced by the DEMETER GCMs. The initial conditions for the atmospheric and land states in these predictions were provided by the 40-yr ECMWF Re-Analysis (ERA-40) four times each year on the first day of February, May, August, and November. Nine ensemble members of 6-month-long predictions were generated by each GCM by changing the ocean initial conditions (Palmer et al. 2004). Since predictions by different models cover different periods, we chose the ECMWF, UKMO, and METF products, which have the longest overlap and consequently define our period of investigation. The DEMETER data have 2–3 months lag with respect to the initial conditions for the bimonthly periods that we considered.

For simplicity, we consider the multimodel ensemble mean over all the ensemble members of the three GCMs, and we will refer to it as “EMU” in the remainder of the text. The advantage of the multimodel approach is to enlarge the variability simulated by each model’s physics since the ensemble covers a larger portion of the phase space.

3. Upper-level wind and rainfall variability

a. Interannual variability of upper-level winds over South America

As discussed in the introduction, Rossby wave trains on the southern Pacific Ocean generated by ENSO heating anomalies have been linked with rainfall over subtropical SA during spring (Montecinos et al. 2000; Mo and Paegle 2001; Paegle et al. 2002). A similar link has been found during fall between upper-level wind circulation and precipitation, but this particular one seems not related to ENSO (Cazes-Boezio et al. 2003). To investigate the relationship between precipitation over SESA and upper-level wind anomalies, we described the winds’ interannual variability by the mean of an EOF analysis.

The EOF analysis was performed on winds over SA (10°N, 60°S; 90°, 30°W) at the 200-hPa level, for the period 1948–2002, for ON, JF, AM, and JA individually. As a first step, we computed seasonal anomalous meridional and zonal wind fields with respect to the long-term mean of each season at each grid box. Next, we merged the two components in one vector for each time step and obtained eigenvectors and eigenvalues of the covariance matrix.

Depending on season, the variability of upper-level winds over SA is dominated by 2–3 distinct modes, which together account for 40%–55% of the total variance; for all four seasons the eigenvalues spectrum levels off after those 2–3 leading modes, and higher-order modes are indistinguishable. The patterns associated with the first leading mode during ON and AM, and with the second leading mode during JF and JA, have very similar structure and account for a large fraction of the total variance during each season: 25% during ON, 19% during JF, 28% during AM, and 14% during JA. They consist of a continental-scale vortex, centered on the Atlantic coast of subtropical SA near 25°S, 50°W (see, e.g., Fig. 2). For all four seasons results are robust: performing the EOF analysis by considering the period 1959–2001 does not affect the outcomes. Regression maps of corresponding PC on upper-level winds over the South Pacific Ocean resemble the PSA1 pattern during spring and fall, while during summer the pattern is barely identifiable (not shown). The vortex has a quasi-barotropic structure during spring, summer, and fall, as shown for example during AM, by the composite of wind at 200, 500, and 850 hPa over events in which the normalized PC of upper-level winds variability is greater than 1 (see Fig. 3). Similar patterns are found for the other seasons, and composites over values of the PC smaller than −1 (not shown) depict the same circulation except for the opposite sign.

Fig. 2.

Leading EOF pattern of 200-hPa interannual variability of wind for ON.

Fig. 2.

Leading EOF pattern of 200-hPa interannual variability of wind for ON.

Fig. 3.

Composite of horizontal wind (m s−1) for AM at (left to right) 200, 500, and 850 hPa over events in which the PC of 200-hPa winds is greater than one standard deviation.

Fig. 3.

Composite of horizontal wind (m s−1) for AM at (left to right) 200, 500, and 850 hPa over events in which the PC of 200-hPa winds is greater than one standard deviation.

We find a significant positive simultaneous correlation, approximately 0.6, between the corresponding PCs of these modes and bimonthly mean precipitation anomaly over SESA during spring, summer, and fall. During winter the relationship does not hold and we no longer discuss this season. For simplicity, we consider the same region (SESA) for all seasons, even though slight differences in the location of larger precipitation anomalies appear during ON, while the location is largely comparable during JF and AM (Figs. 4, 5, and 6). None of the other PCs is significantly correlated with precipitation during any season.

Fig. 4.

Water budget terms. The composites of precipitation anomalies (shaded, mm day−1) and vertically integrated moisture flux anomalies [vectors, hg (m s)−1] over events in which the PC of 200-hPa winds is (a) greater than one standard deviation and (b) smaller than minus one standard deviation for ON. The regression of the PC of winds onto (c) evaporation (mm day−1) and (d) divergence of the vertically integrated moisture flux (mm day−1). Note the scale of the evaporation represents anomalies one order of magnitude smaller compared to the other fields.

Fig. 4.

Water budget terms. The composites of precipitation anomalies (shaded, mm day−1) and vertically integrated moisture flux anomalies [vectors, hg (m s)−1] over events in which the PC of 200-hPa winds is (a) greater than one standard deviation and (b) smaller than minus one standard deviation for ON. The regression of the PC of winds onto (c) evaporation (mm day−1) and (d) divergence of the vertically integrated moisture flux (mm day−1). Note the scale of the evaporation represents anomalies one order of magnitude smaller compared to the other fields.

Fig. 5.

As in Fig. 4, but for JF and (c) is the composite of the total vertically integrated moisture flux [hg (m s)−1] over events in which the PC of 200-hPa winds is greater than one standard deviation.

Fig. 5.

As in Fig. 4, but for JF and (c) is the composite of the total vertically integrated moisture flux [hg (m s)−1] over events in which the PC of 200-hPa winds is greater than one standard deviation.

Fig. 6.

As in Fig. 4, but for AM.

Fig. 6.

As in Fig. 4, but for AM.

We have repeated the EOF analysis with the ERA-40 dataset and obtained patterns that match very closely with those obtained using the NCEP–NCAR reanalysis. There are small differences, on the order of 1%–2%, in the amount of variance accounted by each mode. The correlation between corresponding PCs is 0.95 during ON, 0.75 during JF, and 0.93 during AM. Finally, the correlations between ERA-40 PCs and mean precipitation over SESA are basically the same during ON and AM, while during JF the correlation is 0.45 instead of the 0.67 that we reported using the NCEP–NCAR data. For this reason, the results regarding the JF season are less reliable.

b. The relationship between rainfall and winds

During all seasons, intense trade winds entering northern SA are deflected southward over the western tropical regions forming the SALLJ (Fig. 1). Besides this common behavior, the circulation has a significant seasonal variability. During ON moisture flux is intense down to 25°S where it weakens and becomes southeastward over SESA. During JF the overall circulation is at its maximum strength, and over eastern Brazil (near 20°S, 50°W) it merges with the moisture flux coming from the Atlantic subtropical high. During AM the SALLJ circulation is weakest, but nevertheless it reaches SESA at 65°–70°W where it is oriented southeastward; its main source is the easterlies between 5° and 15°S. A comparison of the precipitation patterns in Fig. 1 reveals an interesting feature over SESA: precipitation amounts are fairly constant throughout the seasons (Diaz et al. 1998). This is particularly evident during the fall season when the dominant pattern of precipitation has retreated to the north and precipitation amounts over SESA contrast with those of the surrounding areas. Different phenomena are responsible for precipitation during different seasons while the SALLJ is a common element in providing moisture to the region. Indeed, larger precipitation amounts over SESA closely correspond to the SALLJ pattern. To explore the mechanisms underlying the relation between upper-level wind variability and precipitation anomalies over SESA, we examine in detail the composites of rainfall and vertically integrated moisture flux anomalies over events in which the normalized PC is greater (smaller) than 1(−1). All composites are, at least over portions of the patterns, 90% significant according to the Monte Carlo test. We also evaluate other water budget terms by simple regression of the PC onto the latent heat flux (evaporation) and the divergence of the vertically integrated moisture fluxes. Positive (negative) events occurred during ON of 1972, 1982, 1984, 1994, and 1997 (1962, 1964, 1970, 1971, 1974, 1976, 1983, 1989, 1998, and 1999); during JF of 1971, 1973, and 1984 (1962, 1979, 1980, 1982, and 1991); and during AM of 1959, 1973, 1980, and 1984 (1960, 1962, 1968, 1978, 1979, and 1982).

During ON, in the composite of positive events, the anomalous moisture flux is characterized by a southward extended SALLJ; precipitation is increased over SESA where the moisture flux associated with the SALLJ is diverted to the east (Fig. 4a). The circulation at higher levels (not shown) conveys moisture to the southwest from the tropical Atlantic Ocean near 12°S, which results as a secondary source of moisture for SESA. Negative precipitation anomalies are observed over the SACZ, creating the dipole in precipitation previously reported by several authors (Fig. 4a). In the composite of negative events (Fig. 4b) the dipole in precipitation reverses sign, even though there are slight differences in the intensity and structure of the two centers and, particularly over SESA, precipitation anomalies extend further south (near 38°S). The moisture flux distribution is characterized by a well-defined cyclone centered over the ocean at 27°S, 35°W, which represents a marked asymmetry with respect to the positive phase of the circulation. The associated moisture flux anomalies weaken and confine the SALLJ to the north consistently with reduced precipitation over SESA. Southwesterly anomalies over eastern Brazil confine the moisture flux from the tropical Atlantic over the SACZ and contribute in increasing precipitation there. The circulation found in the negative phase contrasts with the findings by Díaz and Aceituno (2003) who interpreted the positive precipitation anomalies over the SACZ as the result of a diverted SALLJ into that region, while here we suggest that increased precipitation over the SACZ is the result of the confined moisture flux from tropical western Atlantic Ocean over the area. We notice over Chile the presence of precipitation anomalies of the same sign and intensity as those over SESA for both phases of the circulation.

The intensity and location of precipitation anomalies are in qualitative agreement with the divergence of vertically integrated moisture flux pattern (Fig. 4c): the regression map shows convergence over SESA and divergence over the SACZ corresponding to the anticyclonic phase of the vortex (positive PC values) and vice versa. In particular precipitation anomalies during the cyclonic phase (Fig. 4b) are consistent with convergence over the region receiving direct moisture from the ocean (SACZ) and divergence over SESA. Evaporation anomalies are not an important contributor to the water budget since their intensity is approximately one order of magnitude smaller than precipitation anomalies (Fig. 4c). We disregard the large anomalies over the Andes between 80° and 70°W and 0° and 30°S since we believe them to be artifacts of the reanalysis data. Asymmetries among positive and negative cases and the limited accuracy of reanalysis data to capture small differences in the moisture flux do not allow for drawing firm quantitative conclusions. Nevertheless, the results presented support the hypothesis that precipitation anomalies are created by the circulation anomalies we documented.

During JF a well-defined vortex centered on the eastern coast of SA characterizes both composites. During the positive phase, corresponding to the southwestern–SESA (northeastern–SACZ) branch of the vortex, positive (negative) precipitation anomalies are observed (Fig. 5a). Anomalies of the opposite sign are observed during the negative phase of the circulation. The upper-level circulation has been described as a stationary Rossby wave during January–March (JFM) by Robertson and Mechoso (2000), who also reported a dipole in the vertical velocity consistent with precipitation anomalies.

In the anticyclonic (positive) circulation (Fig. 5a), the main source of moisture for SESA is the SALLJ. To better identify this feature we also show the composite of the total vertically integrated moisture flux for the positive phase of the circulation (Fig. 5c). The anomalous vortex depicted in Fig. 5a is consistent with a strengthened and zonally broader SALLJ. This results in a large anomalous moisture flux that reaches more meridional (approximately 35°S) and western (approximately 65°W) regions. As for ON, a secondary source of moisture originates in the tropical western Atlantic; it merges with the eastern flank of the SALLJ over central Brazil (Fig. 5c) and reaches SESA. The pole of intense precipitation over Ecuador, Peru, and Colombia corresponds to a locally reduced intensity of the flux from the trade winds (Fig. 5a). During the opposite phase (Fig. 5b) the SALLJ is diverted eastward and, together with the northerly moisture flux from equatorial Atlantic at 50°W, it creates positive precipitation anomalies over the SACZ and central Brazil and negative anomalies over SESA. As for ON, precipitation anomalies of the same sign as in SESA are observed over Chile; however, compared to those during ON, they have a smaller extension.

The regression map of the divergence is in qualitative agreement with precipitation anomalies: it shows a narrow area of negative anomalies over the SACZ and a positive area to the south of SESA. Clearly the correspondence of the divergence pattern with those of precipitation is not strict. An exact correspondence between the anomalous moisture flux and precipitation patterns is not expected in view of the asymmetries between the positive and negative cases, together with the already discussed lack of accuracy of the reanalysis data. As for ON, during JF evaporation anomalies are not an important contributor to the water budget (not shown).

During AM in its anticyclonic phase, the vortex is centered over the ocean but the circulation extends over the central-eastern part of the continent combining the SALLJ and moisture from the western tropical Atlantic. Consistently, positive precipitation anomalies are observed over SESA (Fig. 6a). On the contrary, the circulation confines the SALLJ to the north during the cyclonic phase of the circulation and negative precipitation anomalies are found in SESA (Fig. 6b). In both cases, a dipole of precipitation is observed over the north-northeast part of the continent possibly because of the combined effects of the eddy and the trade winds from the equatorial Atlantic. As for the previous seasons, precipitation anomalies are observed over Chile and have, during this season, the largest spatial extension. The regression over the divergence field presents a broad area of positive anomalies and it is in qualitative agreement with precipitation anomalies (Fig. 6d). Similarly to the other seasons, evaporation anomalies contribute little to the water budget (Fig. 6c).

Summarizing, we reported asymmetries between the cyclonic and anticyclonic phase of the circulation that affect precipitation anomalies over SESA and the SACZ. Precipitation anomalies are not only the result of a deflected SALLJ over the region presenting positive precipitation anomalies (Díaz and Aceituno 2003; Liebmann et al. 2004), but they are the result of the entire trade winds moisture flux transport. We emphasize that the source of moisture impacting precipitation during different seasons can be different, which is a result not previously reported in the literature. We also indicate other centers of precipitation anomalies over SA modulated by interannual wind variability, which are particularly consistent over Chile.

c. On the relationship between the Atlantic Ocean and precipitation over SESA during the fall season

During the fall season, some authors (Diaz et al. 1998; Cazes-Boezio et al. 2003) have suggested an impact of subtropical Atlantic SSTs in the region close to the South American coast on precipitation over SESA. Instead, our analysis suggests that the wind anomalies induce SST anomalies, which confirms the relationships reported but give them a different interpretation. In Fig. 7 we show that, corresponding to the anticyclonic phase of the circulation, reduced evaporation (negative latent heat flux anomalies) and consistent increased SSTs are observed near 25°–30°S on the South American coast. If the atmosphere were driven by the ocean, low-level cyclonic and baroclinic circulation would correspond to warmer SSTs. Similar anomalies are found in the opposite case, although negative SST anomalies are displaced to the northeast and are possibly due to nonnegligible ocean dynamics. In this context, our results extend for the fall season the concept of the atmosphere forcing the ocean under the SACZ during summer as suggested by Robertson and Mechoso (2000) and supported by Chaves and Nobre (2004).

Fig. 7.

Composites over events in which the PC of 200-hPa winds is greater than one standard deviation for AM of (a) latent heat flux anomalies (shaded, W m−2) and vertically integrated moisture flux anomalies [vectors, hg (m s)−1], and (b) SST anomalies (shaded, °C) and vertically integrated moisture flux anomalies [vectors, hg (m s)−1].

Fig. 7.

Composites over events in which the PC of 200-hPa winds is greater than one standard deviation for AM of (a) latent heat flux anomalies (shaded, W m−2) and vertically integrated moisture flux anomalies [vectors, hg (m s)−1], and (b) SST anomalies (shaded, °C) and vertically integrated moisture flux anomalies [vectors, hg (m s)−1].

4. Long-range forecasts

a. Empirical model

Motivated by the relationship between upper-level wind variability and precipitation anomalies reported above, we developed a methodology to predict precipitation over SESA (24°–38°S, 60°–45°W). Our methodology is the same during all seasons, except that the mode used as predictor depends on the season. This dependence is discussed in section 3a.

To associate an uncertainty to the prediction, the forecast is provided in the form of a distribution each year. We assume a normal distribution since the observed mean precipitation over SESA is approximately Gaussian during spring, summer, and fall during 1959–2001. In the initial development of the methodology we have been following the idea Coelho et al. (2004) developed for the prior distribution of Niño-3.4 SST.

The linear regression between the predictor and predictand is determined in cross validation, leaving out the data of the year being forecasted. The computation of the linear regression is therefore repeated for each year. This technique allows the maximum use of the dataset to develop the model and verify it without including any piece of information already considered.

For a particular year, the mean of the empirical distribution is obtained from the equation of the regression line and the value of the predictor for that year; the standard deviation is the root-mean-square of residuals of the regression line σo increased by two terms that account for the finite size of the dataset and the distance of the predictor from its time mean (Wilks 1995). In compact notation

 
formula

where P(θt|ψt) stands for probability distribution of the variable θt given ψt, and the subscript t represents the year to which the forecast refers. In our case, θt is precipitation and ψt is the PC selected as predictor. Here, N(μot, σot) stands for Gaussian distribution with mean μot and standard deviation σot; σo is the root-mean-squared error of the regression line; n is the number of years used to compute the linear regression; ψt is the time mean of the predictor over all years except the tth; and β0 and β1 are the coefficients of the regression line.

Since the highest correlation between precipitation and PC is found in the case with no time lag between corresponding time series, we examine the potential skill of the method by using PCs computed from the reanalysis. To test the method in a more realistic forecast framework the PC must be predicted as well. The PC forecast is defined by the projection of wind anomalies of the DEMETER hindcasts at 200 hPa onto the EOF patterns computed with NCEP–NCAR reanalysis. An alternative method is to perform the EOF analysis directly on the GCMs fields. We chose the first method since we consider EOF modes as known features of the system, available each time a forecast is released, and we can reasonably expect the variability of the forecasted wind fields to be smaller than in the observation. For a simple evaluation of our results we compute the skill score as the mean-squared error with respect to climatology (Wilks 1995); this procedure is commonly used in forecast verification. Through the values of the standard deviation of the distribution we provide a confidence estimate on the results.

b. Potential predictability

The empirical method provides useful predictions of precipitation anomalies when PCs based on reanalysis data are used as predictors. The skill score is approximately 0.4 during spring, summer, and fall (Table 1). The value is higher than the one of the bias-corrected EMU as well as the one reported by IRI (Goddard et al. 2003) and also by COF (Berri et al. 2005). Although the variability reproduced by the method is still smaller than the observed one (Fig. 8), the empirical model can forecast a large range of values. Corresponding to large anomalies, there is, in most cases, an 80% probability of correctly forecasted values below or above the climatology (see, e.g., 1982 and 1997 during ON, 1984 during JF, 1982 during AM in Fig. 8). Further, the method is reliable in general and only few values fall outside of the 80% confidence interval. As reference for the reader, the mean1 standard deviations of the forecasts are 0.94, 0.77, and 0.96 mm day−1 for ON, JF, and AM, respectively.

Table 1.

Skill of the empirical model for the predictability and experimental forecasts (see text for details). Skill is measured by the mean-squared error skill score.

Skill of the empirical model for the predictability and experimental forecasts (see text for details). Skill is measured by the mean-squared error skill score.
Skill of the empirical model for the predictability and experimental forecasts (see text for details). Skill is measured by the mean-squared error skill score.
Fig. 8.

Potential predictability of precipitation (mm day−1) over SESA. Mean of the distribution of the empirical model (black), 80% confidence interval (light gray), climatology (dotted), and observations (gray) for (a) ON, (b) JF, and (c) AM.

Fig. 8.

Potential predictability of precipitation (mm day−1) over SESA. Mean of the distribution of the empirical model (black), 80% confidence interval (light gray), climatology (dotted), and observations (gray) for (a) ON, (b) JF, and (c) AM.

c. Experimental forecasts

Results based on using PCs obtained by reanalysis data indicate the potential of the 200-hPa wind field as predictor and encouraged us to test our method in a more realistic forecast framework. As expected, the model’ skill is lower when forecasted PCs are used as predictor. Nevertheless, skillful predictions are still obtained during spring (Table 1 and Fig. 9). The skill score during summer and fall drops to zero from 0.4 of the potential predictability case. During fall and summer the method is not reliable since at the 80% confidence level it would provide the same forecast every year and more than 20% of the forecasts fall outside this interval. We note that variability of the predictions is very small relative to that of the observations (e.g., see Fig. 9).

Fig. 9.

As in Fig. 8, but for experimental forecasts of precipitation (mm day−1).

Fig. 9.

As in Fig. 8, but for experimental forecasts of precipitation (mm day−1).

We next explore why real predictions are skillful during spring only and the reasons for the unrealized predictability during fall and summer. It is apparent that the only cause for the decrease in skill is the use of forecasted wind fields. In this regard, the correlation between forecasted and PCs based on the reanalysis is 0.6 during spring, but it is not significant during summer and fall. The wind field climatology is well simulated by EMU during all seasons (not shown), but the anomalous circulation is only reasonably captured during spring. Composites of NCEP–NCAR reanalysis fields over events in which the normalized PC1 is greater (smaller) than 1(−1) closely resembles the EOF pattern during spring, summer, and fall: there is a vast anticyclonic (cyclonic) circulation centered over the eastern coast of subtropical SA reaching the northwestern coast of the continent. A pattern resembling the eastern part of the PSA1 mode appears to the southwest of the domain in the composite of ON and AM. The corresponding composites for spring from EMU do not reproduce the anomalies in the tropics. Nevertheless, a signature of the PSA pattern over the southwest and the vortex are captured, although its center is slightly displaced to the east (Fig. 10). During summer the eddy is not found in either of the composites (not shown). During fall, there is a weak signature of the PSA pattern and the eddy is captured but its center is displaced approximately 20° in the composite of positive events, while in the composite of negative events the whole pattern is missed (Fig. 11).

Fig. 10.

Composite of 200-hPa winds (m s−1) over events in which the PC of 200-hPa winds is (top) greater than one standard deviation and (bottom) smaller than minus one standard deviation for the (left) NCEP–NCAR reanalysis and (right) EMU for ON.

Fig. 10.

Composite of 200-hPa winds (m s−1) over events in which the PC of 200-hPa winds is (top) greater than one standard deviation and (bottom) smaller than minus one standard deviation for the (left) NCEP–NCAR reanalysis and (right) EMU for ON.

Fig. 11.

Same as Fig. 10 but for AM.

Fig. 11.

Same as Fig. 10 but for AM.

When the EOF pattern is projected onto an anomalous field not well captured in the forecasts, the resulting value of the “forecasted PC” is small. Thus the relationship between precipitation and PC found by using observational data cannot be reproduced.

These results are consistent with ENSO triggering a PSA1-like regime through Rossby wave propagation during spring only (Cazes-Boezio et al. 2003). Robertson and Mechoso (2003) proposed a different interpretation of the PSA-like regimes as geographically fixed circulations during fall, winter, and, to a lesser extent, during spring. Nevertheless, they also found a strong correlation between ENSO and the frequency of occurrence of the PSA-like regimes during spring. We can reasonably summarize that during southern spring ENSO, one of the strongest forcing of the climate system in the interannual time scale, is associated with, and even possibly generates, the anomalous vortex over SA. This is consistent with a higher capability of current GCMs in reproducing the mechanism and pattern in the wind field. We have conducted a preliminary investigation on the relationship between the vortex over SA, the PSA, and ENSO for the three seasons, and we found that ENSO may trigger the leading PSA, which in turn is related to the vortex over SA. This chain of elements, however, presents differences during different seasons and does not fully account for the occurrence of the vortex over SA (Zamboni 2009). At present, for the seasons other than spring, the mechanisms responsible for the occurrence of the vortex over SA are unknown, and further analysis is required to understand why the atmosphere over SA presents the described preferred configuration and thus to properly represent it in the GCMs.

d. Predictability associated with ENSO

Corresponding to stronger ENSO events, we can expect predictability to increase because of a larger the “signal-to-noise ratio.” In our study, this is of particular interest for the spring season since we may expect to find the circulation anomalies we have discussed at a more frequent rate and the GCMs to better capture them. To investigate whether the predictability of the system and the skill of the forecast are accordingly higher, we developed and tested the empirical model considering ENSO events only for the spring season. For this test, we adopted the definition of ENSO events by the Japan Meteorological Agency (JMA), which classifies years as events if the 5-month running mean of SSTs anomalies over 4°S–4°N, 150°–90°W is equal to or higher (lower) than 0.5°C (−0.5°C) for 6 consecutive months. According to this definition there are 10 El Niños and 10 La Niñas from 1959 to 2001.

The skill score of precipitation forecasts during spring is doubled (0.46) using forecasted PC of wind as predictor corresponding to the occurrence of ENSO events. To evaluate the effects on the skill of the strong events 1982/83 and 1987/88, we applied the method after having removed those cases. The skill score results 0.5.

To use our method in a real forecast framework a prediction for the occurrence of an ENSO event is required. We do not believe this requirement makes a significant difference since we expect state-of-the-art GCMs to be reasonably sensitive to ENSO events. For example, the correlation between the observations and EMU over 4°S–4°N, 150°–90°W is 0.94. The increase of the skill when the empirical model is applied to ENSO events only reveals a further potential of the method besides the assumed linearity of the response of the system to El Niño and La Niña and the relationship between precipitation and PC of upper-level wind over SA.

5. Conclusions

We have shown evidence of a simultaneous linear relationship at interannual time scales between precipitation anomalies over SESA (24°–38°S, 60°–45°W), a region that comprises parts of the La Plata River basin in SA, and the interannual variability of upper-level wind over SA. The relationship holds during spring, summer, and fall. The anomalous wind pattern related to precipitation anomalies is very similar during all seasons: it consists of a continental-scale vortex, centered on the Atlantic coast of subtropical SA at 25°S, 50°W (Fig. 2). This configuration is the first leading EOF mode during ON and AM and the second leading EOF mode during JF; they account for 25%, 19%, and 28% of the total variance, respectively. In addition, regression maps suggest that the pattern could be related to the PSA1 during spring and fall, while their connection is less evident during summer. The vortex has a quasi-barotropic structure during spring, summer, and fall (see, e.g., Fig. 3).

The circulation associated with the anticyclonic phase of the upper-level vortex combines at lower levels moisture fluxes from the SALLJ and the western tropical Atlantic contributing to increased precipitation over SESA during spring, summer, and fall and diminished precipitation over the SACZ during spring–summer, thus creating the north–south precipitation dipole previously reported by several authors. During the cyclonic phase of such vortex, when increased precipitation is observed over the SACZ (during spring–summer) and decreased precipitation is observed over SESA (during spring, summer, and fall), seasonal differences arise. During spring precipitation anomalies are associated with a confined moisture flux from the tropical Atlantic over the SACZ and with a weakened SALLJ. During summer they are associated with an eastward diverted SALLJ toward the SACZ, while during fall the SALLJ is weakened and confined to the north.

We also indicate other centers of precipitation anomalies over SA modulated by interannual wind variability; particularly over Chile precipitation anomalies are consistent throughout the seasons and of the same sign of those over SESA.

Even though asymmetries among positive and negative cases and the limited accuracy of reanalysis data to detect small differences in the moisture flux do not allow for drawing firm quantitative conclusions, our results do support the hypothesis that precipitation anomalies are associated with the quasi-barotropic circulation we identified. In fact, the intensity and distribution of precipitation anomalies are in qualitative agreement with those of vertically integrated moisture flux divergence, while evaporation anomalies are one order of magnitude smaller (see Figs. 4, 5, and 6).

During the fall season, some authors (Diaz et al. 1998; Cazes-Boezio et al. 2003) have suggested that subtropical Atlantic SST anomalies near the South American coast are linked to precipitation over SESA. We suggest that wind circulation anomalies induce the one in the SSTs (see Fig. 7). Since oceanic anomalies are forced by atmospheric anomalies, the former are associated with, but cannot be used as predictors of, the latter.

We have shown that long-range seasonal precipitation forecasts over SESA from operational centers have low skills (Goddard et al. 2003; Berri et al. 2005; Coelho et al. 2006). In view of this, we proposed an empirical model as alternative method to provide rainfall predictions during spring, summer, and fall based on the relationship we documented between upper-level wind variability and precipitation.

To provide an estimate of the precipitation predictability associated with upper-level winds in a real forecast, we carried out the analysis using predictors obtained by the reanalysis data first and forecasted predictor values next. For the latter investigation, we made use of the DEMETER forecasted 200-hPa wind fields.

PCs can potentially be used as predictor of the mean precipitation over SESA during spring, summer, and fall: the skill score is approximately 0.4 in all cases. The empirical model can forecast a large range of values, and it is reliable with just few values falling outside the 80% confidence interval (see Fig. 8).

During summer and fall the DEMETER GCMs forecasts of wind anomalies limit the method’s potential: the forecasted wind anomalies are not well captured and consequently the relationship between precipitation and PC found in the observations is not well reproduced either (see Figs. 10 and 11). Wind anomalies are reasonably well captured during spring, and skillful precipitation predictions are obtained during this season, the skill score being 0.26.

It is furthermore shown that predictions for spring are improved when the method is applied in the occurrence of ENSO events; in this case the skill score is 0.46. The result is consistent with the hypothesis that ENSO is able to trigger a pattern associated with the PSA1 during spring (Cazes-Boezio et al. 2003; Robertson and Mechoso 2003). For the seasons other than spring, the mechanisms responsible for the occurrence of the vortex over SA are unknown and further analysis is required to understand why the atmosphere over SA presents the described preferred configuration and thus to properly represent it in the GCMs.

Acknowledgments

This research was supported by the Dipartimento di Matematica, Universitá degli Studi di Trieste, Italy, the Abdus Salam International Centre for Theoretical Physics, and by NOAA under the Grant NA05OAR4310009. The authors wish to thank Dr. G. Cazes-Boezio and Dr. A.W. Robertson for the stimulating discussions and helpful suggestions.

REFERENCES

REFERENCES
Berri
,
G. J.
,
P. L.
Antico
, and
L.
Goddard
,
2005
:
Evaluation of the climate outlook forum’s seasonal precipitation forecasts of southeast South America during 1998-2002.
Int. J. Climatol.
,
25
,
365
377
.
Casarin
,
D. P.
, and
V. E.
Kousky
,
1986
:
Anomalias de precipitação no sul do Brasil e variações na circulação atmosférica.
Rev. Bras. Med.
,
1
,
83
90
.
Cazes-Boezio
,
G.
,
A. W.
Robertson
, and
C. R.
Mechoso
,
2003
:
Seasonal dependence of ENSO teleconnections over South America and relationships with precipitation over Uruguay.
J. Climate
,
16
,
1159
1176
.
Chaves
,
R. R.
, and
P.
Nobre
,
2004
:
Interactions between sea surface temperature over the South Atlantic Ocean and the South Atlantic Convergence Zone.
Geophys. Res. Lett.
,
31
,
L03204
.
doi:10.1029/2003GL018647
.
Coelho
,
C. A. S.
,
S.
Pezzulli
,
M.
Balmaseda
,
F. J.
Doblas-Reyes
, and
D. B.
Stephenson
,
2004
:
Forecast calibration and combination: A simple Bayesian approach for ENSO.
J. Climate
,
17
,
1504
1516
.
Coelho
,
C. A. S.
,
D. B.
Stephenson
,
M.
Balmaseda
,
F. J.
Doblas-Reyes
, and
G. J.
van Oldenborgh
,
2006
:
Toward an integrated seasonal forecasting system for South America.
J. Climate
,
19
,
3704
3721
.
Díaz
,
A.
, and
P.
Aceituno
,
2003
:
Atmospheric circulation anomalies during episodes of enhanced and reduced convective cloudiness over Uruguay.
J. Climate
,
16
,
3171
3185
.
Díaz
,
A.
,
C. D.
Studzinski
, and
C. R.
Mechoso
,
1998
:
Relationships between precipitation anomalies in Uruguay and southeastern Brazil and sea surface temperature in the Pacific and Atlantic Oceans.
J. Climate
,
11
,
251
271
.
Doyle
,
M.
, and
V.
Barros
,
2002
:
Midsummer low-level circulation and precipitation in subtropical South America and related sea surface temperature anomalies in the South Atlantic.
J. Climate
,
15
,
3394
3410
.
Ferreira
,
R. N.
,
T. M.
Rickenbach
,
D. L.
Herdies
, and
L. M. V.
Carvalho
,
2003
:
Variability of South American convective cloud systems and tropospheric circulation during January–March 1998 and 1999.
Mon. Wea. Rev.
,
131
,
961
973
.
Goddard
,
L.
,
A. D.
Barnston
, and
S. J.
Mason
,
2003
:
Evaluation of the IRI’s “net assessment” seasonal climate forecasts 1997–2001.
Bull. Amer. Meteor. Soc.
,
84
,
1761
1781
.
Grimm
,
A. M.
,
2003
:
The El Niño impact on the summer monsoon in Brazil: Regional processes versus remote influences.
J. Climate
,
16
,
263
280
.
Grimm
,
A. M.
,
V. R.
Barros
, and
M. E.
Doyle
,
2000
:
Climate variability in southern South America associated with El Niño and La Niña events.
J. Climate
,
13
,
35
58
.
Grimm
,
A. M.
,
J. S.
Pal
, and
F.
Giorgi
,
2007
:
Connection between spring conditions and peak summer monsoon rainfall in South America: Role of soil moisture, surface temperature, and topography in eastern Brazil.
J. Climate
,
20
,
5929
5945
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Liebmann
,
B.
,
G. N.
Kiladis
,
C. S.
Vera
,
A. C.
Saulo
, and
L. M. V.
Carvalho
,
2004
:
Subseasonal variations of rainfall in South America in the vicinity of the low-level jet east of the Andes and comparison to those in the South Atlantic convergence zone.
J. Climate
,
17
,
3829
3842
.
Mo
,
K. C.
,
2000
:
Relationships between low-frequency variability in the Southern Hemisphere and sea surface temperature anomalies.
J. Climate
,
13
,
3599
3610
.
Mo
,
K. C.
, and
J. N.
Paegle
,
2001
:
The Pacific–South American modes and their downstream effects.
Int. J. Climatol.
,
21
,
1211
1229
.
Montecinos
,
A.
,
A.
Diaz
, and
P.
Aceituno
,
2000
:
Seasonal diagnostic and predictability of rainfall in subtropical South America based on tropical Pacific SST.
J. Climate
,
13
,
746
758
.
New
,
M.
,
M.
Hulme
, and
P.
Jones
,
1999
:
Representing twentieth-century space–time climate variability. Part I: Development of a 1961–90 mean monthly terrestrial climatology.
J. Climate
,
12
,
829
856
.
New
,
M.
,
M.
Hulme
, and
P.
Jones
,
2000
:
Representing twentieth-century space–time climate variability. Part II: Development of monthly grids of terrestrial surface climate.
J. Climate
,
13
,
2217
2238
.
Paegle
,
J. N.
, and
K. C.
Mo
,
1997
:
Alternating wet and dry conditions over South America during summer.
Mon. Wea. Rev.
,
125
,
279
291
.
Paegle
,
J. N.
, and
K. C.
Mo
,
2002
:
Linkages between summer rainfall variability over South America and sea surface temperature anomalies.
J. Climate
,
15
,
1389
1407
.
Paegle
,
J. N.
, and
Coauthors
,
2002
:
Progress in pan-American CLIVART research: Understanding the South American monsoon.
Meteorologica
,
27
,
3
32
.
Palmer
,
T. N.
, and
Coauthors
,
2004
:
The economic value of ensemble forecasts as a tool for risk assessment: From days to decades.
Quart. J. Roy. Meteor. Soc.
,
128
,
747
774
.
Robertson
,
A. W.
, and
C. R.
Mechoso
,
2000
:
Interannual and interdecadal variability of the South Atlantic convergence zone.
Mon. Wea. Rev.
,
128
,
2947
2957
.
Robertson
,
A. W.
, and
C. R.
Mechoso
,
2003
:
Circulation regimes and low-frequency oscillations in the South Pacific sector.
Mon. Wea. Rev.
,
131
,
1566
1576
.
Silva
,
G. A. M.
,
T.
Ambrizzi
, and
J. A.
Marengo
,
2009
:
Observational evidences on the modulation of the South American low level jet east of the Andes according the ENSO variability.
Ann. Geophys.
,
27
,
645
657
.
Smith
,
T. M.
, and
R. W.
Reynolds
,
2004
:
Improved extended reconstruction of SST (1854–1997).
J. Climate
,
17
,
2466
2477
.
Wilks
,
D. S.
,
1995
:
Statistical Methods in the Atmospheric Sciences.
Academic Press, 467 pp
.
Zamboni
,
L.
,
2009
:
Relationships between upper level circulation over South America and rainfall over southeastern South America: A physical base for seasonal predictions. Ph.D. thesis, Universitá degli Studi di Trieste, 103 pp
.

Footnotes

Corresponding author address: Laura Zamboni, Centro Euro-Mediterraneo per i Cambiamenti Climatici (CMCC), Viale Aldo Moro, 44 I-40127 Bologna, Italy. Email: laura.zamboni@cmcc.it

1

There is a value for the standard deviation for each year being the forecast provided in the form of a distribution.