1. Introduction
The task of understanding climate variability in the tropical Atlantic region presents a considerable challenge to climate scientists. Unlike the tropical Pacific, seasonal-to-decadal climate variability in the tropical Atlantic is not dominated by any single mode such as the El Niño–Southern Oscillation (ENSO). Rather, this region is subject to multiple competing influences of comparable importance. Some of these influences originate in regions remote from the tropical Atlantic, while others arise from local processes; some are potentially predictable, whereas others are essentially stochastic. These influences interact in subtle ways to determine the evolution of the atmosphere–ocean system.
Fluctuations in the climate of the tropical Atlantic region have major effects on the human populations of the surrounding continents. Agriculture in northeast Brazil and in the Sahel, for example, is strongly affected by large year-to-year fluctuations of rainfall (Moura and Shukla 1981; Lough 1986; Folland et al. 1986; Hulme et al. 1992a,b; Rowell et al. 1995). The number and intensity of hurricanes, which develop in the subtropical Atlantic, also show considerable year-to-year variability (Landsea 1993; Goldenberg and Shapiro 1996). Research to date has already demonstrated the possibility of forecasting some of these fluctuations in climate (e.g., Ward et al. 1993; Gray et al. 1993; Hastenrath and Greischar 1993), and indeed experimental forecasts have been issued for a number of years. It is reasonable to expect that further progress in understanding the causes of climate variability in the tropical Atlantic region will lead to improvements in the skill of such forecasts.
Central to understanding the variability of the atmosphere–ocean system is understanding variability in sea surface temperature (SST). Previous work has established the following basic features of SST variability in the tropical Atlantic region. Spectra of SST variability in general show good agreement with the “red-noise” form that is predicted by the stochastic model of Frankignoul and Hasselmann (1977). There have, however, been indications from some studies of enhanced interannual variability on and near to the equator (Zebiak 1993; Latif and Gröñer 2000), and enhanced decadal variability off the equator, particularly to the south (Mehta 1998). The interannual variance of SST on and near to the equator peaks in boreal summer (Latif and Gröñer 2000), whereas the variance to the north of the equator peaks in boreal spring (Nobre and Shukla 1996). SST variability to south of the equator exhibits only weak correlations with that to the north (Houghton and Tourre 1992; Mehta 1998; Rajagopalan et al. 1998).
The atmospheric variability in the tropical Atlantic region (or indeed in any other region) may be partitioned between internal variability, which arises from instabilities within the atmosphere, and externally forced variability, which arises as a response to changing boundary conditions such as SST. The fraction of the total atmospheric variance that is a response to SST changes is relatively high in the Tropics and relatively low in the extratropics (e.g., Rowell 1998). Thus in the tropical Atlantic region we know that SST exerts a significant control on the atmosphere. Previous work suggests that more than one region of SST is important; three particular SST patterns and associated patterns of response are discussed in the literature. In no particular order these are the following. (a) A remote response to SST in the Pacific. A prominent feature of this response is that during El Niño events the northeast trade winds in the Atlantic basin weaken, particularly during boreal winter (Curtis and Hastenrath 1995; Enfield and Mayer 1997). (b) A local atmospheric response to fluctuations in the cross-equator SST gradient. In this case the atmospheric response involves anomalous cross-equator flow, particularly in boreal spring, directed toward the hemisphere in which SST is anomalously high (e.g., Moura and Shukla 1981). (c) Fluctuations in the equatorial easterlies in response to fluctuations in central/eastern equatorial Atlantic SST anomalies (Zebiak 1993).
In the instances where the atmospheric fluctuations arise in response to fluctuations in the local tropical Atlantic SST there may be feedbacks onto the ocean that could support a coupled ocean–atmosphere mode of variability. Zebiak (1993) suggested that, associated with the third of the signals mentioned above, there is a coupled mode of interannual variability in the equatorial Atlantic that is analogous to ENSO but has smaller amplitude and a shorter timescale. Chang et al. (1997) suggested there could be a decadal timescale coupled mode associated with the second of the signals (b) mentioned above. It has yet to be clearly established, however, whether the feedbacks postulated by Chang et al. (1997) arise in the real world and, if they do arise, whether they have sufficient strength to compete successfully with quasi-stochastic internal variability.
The aim of this study is to further investigate the characteristics of both internal and SST-forced atmospheric variability in the tropical Atlantic region. We use an ensemble of integrations with an atmospheric GCM to separate the internal variability from the SST-forced variability (a separation that is not possible with a single model integration or with observational data). By using a recently developed algorithm (Allen and Smith 1997; Venzke et al. 1999), we then estimate, in each of the four seasons, the space–time characteristics of the dominant modes of internal and SST-forced variability. We identify the patterns of SST associated with the latter and assess feedbacks onto the ocean. An important feature of our study is that we are able to obtain quantitative estimates of the comparative importance, in different regions and different seasons, of the three SST-forced signals described above, and of the internal variability.
The structure of the paper is as follows. In section 2 the model, ensemble integrations, and preliminary data analysis are described. In section 3 we use a local analysis of variance (e.g., Rowell 1998) to assess the comparative importance of internal and SST-forced variability in the tropical Atlantic region. We then examine the dominant modes of internal variability. In section 4 we examine the dominant modes of SST-forced variability. We use the optimal filter patterns, which are an additional output of the Venzke et al. (1999) algorithm, to assist our physical interpretation of these modes. We examine their seasonality and, in section 5, investigate feedbacks from the atmosphere back onto the ocean.
While most of the analyses in this paper seek to distinguish different components of climate variability in the tropical Atlantic from each other, these components are not linearly independent. In section 6 we consider how the various elements combine together to determine the total variability, and discuss the consequences for the predictability of climate fluctuations in the tropical Atlantic region. Conclusions are presented in section 7.
2. Atmospheric model integrations and data analysis
The GCM used for this study is a version of the U.K. Hadley Centre model known as HadAM1. This version was used in the Atmospheric Model Intercomparison Project (Gates 1992). It is a gridpoint model with a horizontal resolution of 2.5° lat × 3.75° long and 19 hybrid levels in the vertical. Physical parameterizations include: a gravity wave drag scheme, a radiation scheme that computes fluxes in four longwave bands and six shortwave bands and responds to prognostic cloud variables, a penetrative convection scheme with stability-dependent closure, and a land surface scheme. Further details about the model development and the physics and dynamics of the precise version used here are given by Cullen (1993) and Phillips (1994), respectively.
The simulations used here are the same six-member ensemble employed by, for example, Potts et al. (1996);Davies et al. (1997); Rowell (1998); Renshaw et al. (1998), and Venzke et al. (1999). Each member was integrated from 1 October 1948 to 1 December 1993 with lower boundary forcing supplied by version 1.1 of the Global Sea-Ice and Sea Surface Temperature dataset (Parker et al. 1995). The initial atmospheric state and surface temperatures were taken from U.K. Meteorological Office analyses for six different dates, arbitrarily chosen, but all close to 1 October. Soil moisture and snow depth were initialized from an adaption of the Willmott et al. (1985) climatology. Due to likely spinup effects, the first 2 months of each integration were discarded, leaving 6 × 45 = 270 yr of model data, with each member starting on 1 December 1948 and finishing on 30 November 1993.
The climatologies of model variables were computed for each month by averaging that month over all 270 model years. Monthly anomalies were obtained by subtracting the climatology from the monthly data. Standard seasonal means (December–February, DJF; March–May, MAM; June–August, JJA; September–November, SON) were constructed by averaging over 3-month periods, and each season was analyzed separately.
The principal variable we analyze is the surface wind stress. We make this choice because of our interest in how atmospheric variability (internal or forced) forces the ocean. For the same reason we mask out grid points over land. For some analyses we treat the zonal and meridional components of wind stress separately, but in most cases we combine them into a single field. To do this we first weight each component with a scale factor equal to the reciprocal of its spatially averaged standard deviation. The components are deweighted before plotting.
Figure 1 shows seasonal means of the surface wind stress climatology, superimposed on the seasonal mean SST. It can be seen that the ITCZ is generally collocated with the region of highest SST, is farthest south in MAM, and farthest north in JJA and SON. The model wind stress shows very good agreement with the observed mean wind stress computed from the da Silva et al. (1994) dataset (not shown).
We use a number of techniques to analyze the variability in our model integrations. Common to all these techniques, however, is an attempt to distinguish between internal atmospheric variability and variability that is forced by the time varying surface boundary conditions (SST and sea-ice extent). We define the “forced response” as the component of the evolution of the system that is determined by the external forcing, independent of the initial conditions, and would therefore be common to all members of a hypothetical infinite ensemble. If we had such an infinite ensemble available, the forced response would simply be the time-varying part of the ensemble mean.
For small ensembles such as ours, however, internal variability makes an important contribution to the variability in time of the ensemble mean. A consequence of this “contamination” is that any nonlinear function of the ensemble mean (e.g., variance, EOFs, etc.) is a biased estimate of corresponding function of the true forced response. By employing information about the internal variability derived from the ensemble, however, we are able to derive unbiased estimates. Rowell (1998) and Venzke et al. (1999) provide the details of how this is done for local variance and for spatial patterns, respectively. We note in passing that, unless steps are taken to account for the contribution of internal variability, the leading EOFs of the ensemble mean can give a very misleading impression of the dominant modes in the forced response (Venzke et al. 1999).
3. Internal and forced atmospheric variability in the tropical Atlantic region
a. Local analysis of variance
We present first a local analysis of the percentage of the total variance in wind stress that is due to SST and sea-ice1 forcing. This analysis provides an important baseline for the subsequent analyses since here we show (locally) the full percentage of forced variance, whereas later we separate out the contributions from particular elements of climate variability.
Figure 2 shows the results of ANOVA analyses (Rowell et al. 1995; Rowell 1998) of the zonal and meridional components of wind stress. The basic picture is clear: the ratio of externally forced to internal variance is high in the Tropics and low in the extratropics. In all seasons the percentages are highest, sometimes exceeding 80%, in the region within 5°–10° of the equator. The highest percentages are found in MAM and JJA. Especially for zonal wind stress the largest values arise on the western side of the equatorial Atlantic. Between 5° and 25° north and south values fall rapidly to 20% or less.
b. Internal atmospheric variability
The local ANOVA gives no information about the space–time characteristics of the internal or the forced variability, which are the major focus of interest in this paper. We consider first the internal variability, which we estimate by computing the deviations of the individual ensemble members from the ensemble mean. As we are considering interannual and longer timescales and the internal memory of the atmosphere rarely exceeds a month or so, we expect the internal variability to be uncorrelated in time and this is indeed the case (Rowell and Zweirs 1999). We do expect, however, preferred spatial patterns. To determine these patterns we perform an empirical orthogonal function (EOF) analysis.
Figure 3 shows results from the joint EOF analysis of the two components of surface wind stress in the region 30°S–30°N, 80°W–20°E. (Note that land points and points over the Pacific Ocean are masked out.) Table 1 shows the fraction of the total internal variance explained by each of the first three EOFs. In all four seasons the leading mode is associated with a pattern that has largest amplitude away from the equatorial belt. This result is in line with the ANOVA results shown in Fig. 2 and with our general expectations that internal variability is more important in the extratropics than in the Tropics.
In DJF and MAM the first EOF is well separated from the higher order EOFs in the eigenvalue spectrum (Table 1). In both these seasons the pattern is dominated by variability in the Northern Hemisphere, especially in the region poleward of 15°N. Comparison with Fig. 1 shows that the pattern describes modulation of the poleward flank of the trade winds. These fluctuations in the trades are correlated with fluctuations in the higher latitude westerlies (not shown) and may be viewed as a low-latitude manifestation of the North Atlantic oscillation (NAO). Note that the influence of the NAO extends as far south as 10°–15°N in DJF.
In JJA, and especially SON, the first EOF is less well separated from the higher-order EOFs in the eigen spectrum. In JJA the dominant variability is found in the Southern Hemisphere as one might expect during austral winter. In SON, however, variability in the Northern Hemisphere dominates again.
In summary, our results show that low-frequency internal variability in the tropical Atlantic region (as defined in this analysis) is dominated by the equatorward extension, especially from the Northern Hemisphere, of extratropical patterns (or “modes”). The NAO is the most important such mode.
4. Atmospheric variability forced by changing SST
a. Optimal detection of the dominant pattern of SST-forced variability
In this section we present the results of an analysis of the leading mode of SST-forced variability in each of the four seasons. We refer the reader to Venzke et al. (1999) for a full account of the algorithm we use. A key stage in its application involves projection of the ensemble mean onto the leading EOFs of internal variability (referred to in this context as “noise” and estimated using the deviations from the ensemble mean as in section 3b). A choice must be made about how many noise EOFs to retain. Tests showed that in the Tropics more EOFs must be retained than is typical for midlatitudes. This is because much of the SST-forced variability in the Tropics is close to the equator where the internal variability has low variance. Therefore to ensure that the noise EOFs span the space in which the SST-forced signal lies, a relatively large number must be retained. We found that it was necessary to retain 30 noise EOFs. The results were very similar when 40 noise EOFs were retained.
The algorithm yields a spatial pattern that is an unbiased estimate of the first EOF of a hypothetical infinite ensemble (or, in other words, the first EOF of the SST-forced component of variability), together with an associated time series (or principal component). In addition, the algorithm yields “optimal filter” patterns. By definition these are the patterns that optimally discriminate between the SST-forced variability and the internal variability; that is, they have large weights where the signal-to-noise ratio is high.2 Although the optimal filters are not themselves the SST-forced signals, they can, as we shall see, sometimes assist the physical interpretation of these signals. In particular, because we often expect the signal-to-noise ratio to be high at the locations where the forcing is operating, the optimal filter patterns may help us to identify these special locations.
Unlike canonical correlation analysis (CCA) and singular value decomposition analysis (Bretherton et al. 1992), the algorithm of Venzke et al. (1999) makes no direct use of the SST field to identify the forced signals. Consequently we identify the SST pattern that (linearly) induces the atmospheric response after performing the Venzke et al. (1999) analysis by regressing the SST onto the first principal component of SST-forced variability. The algorithm also yields an estimate of the signal-to-noise ratio associated with a particular pattern. Table 2 (columns 2 and 3) shows that in all seasons the leading mode is easily detectable at the 99% confidence level, not only in the ensemble mean but also in an individual ensemble member. Note that the numbers in the table measure the global significance of each pattern. We will consider local significance in section 4b.
Figure 4a shows the leading mode of SST-forced wind stress variability in DJF, together with the associated SST field. The SST pattern clearly shows that the wind fluctuations arise in response to ENSO events in the Pacific. The most prominent feature in the wind stress pattern is the modulation of the northeast (NE) trades, such that during Pacific warm events the NE trades are anomalously weak. This feature of the Atlantic response to ENSO has been noted in observational studies (Curtis and Hastenrath 1995; Enfield and Mayer 1997). Figure 4a suggests that, concurrent with the weakening of the NE trades, there is strengthening of the easterlies in the western equatorial Atlantic and weakening of the SE trades. These secondary features, however, were not seen in the observational studies cited above. The discrepancies may indicate an error in the model response to ENSO or weaknesses in the observational record.
The optimal filter pattern for DJF looks very similar to the pattern of SST-forced variability, except that it has lower weights at higher latitudes. This is as we expect since the local analysis of variance (section 3a) showed that the signal-to-noise ratio is highest close to the equator. An intriguing feature that the optimal filter pattern draws attention to is the divergence around 7.5°N. This latitude lies significantly to the north of the mean position of the ITCZ at this time of year (Fig. 1), and inspection of other fields (not shown) indicates that the divergence is not related to local anomalous convection. What, then, is the origin of this feature?
There are at least two mechanisms via which ENSO may influence the tropical Atlantic region (e.g., Saravanan and Chang 2000). One mechanism is via changes to the tropical Walker and Hadley circulations and the associated patterns of deep convection. A second mechanism is via a Rossby wave–like disturbance propagating through the extratropics of the winter (Northern) hemisphere (Nobre and Shukla 1996). We speculate that the divergence at ∼7.5°N may delineate the domains of these two competing influences. To the north the extratropical Rossby disturbance dominates, and the weakening of the NE trades may, perhaps, be viewed as the “tail-end” of the Pacific North American pattern. To the south, changes in the Walker and Hadley circulations dominate.
The picture in MAM (Fig. 4b) is rather different to that seen in DJF. The regression of SST on the optimized time series again picks out an ENSO-like pattern in the Pacific, but also a dipole pattern in the tropical Atlantic with anomalies of one sign between 10° and 20°N and anomalies of the opposite sign between 0° and 20°S. The latter are especially intensified near the coast of Africa. The wind stress pattern indicates that, as in DJF, the NE trades are weaker than usual when SST in the central Pacific is anomalously high. Unlike in DJF, however, there is a prominent enhancement of the SE trades in the latitude band 0°–10°S. There is anomalous flow across the climatological position of the ITCZ in this season (see Fig. 1) toward the positive lobe of the SST dipole and away from the negative lobe. This cross-Equator flow, which is related to a displacement of the ITCZ, has the qualitative form that is predicted by the simple boundary layer model of Lindzen and Nigam (1987), wherein the SST dipole would be expected to set up a dipole in sea level pressure. On the equator, where the Coriolis force vanishes, a frictionally balanced flow down the pressure gradient can be set up. Moving away from the equator in the direction of the flow, the Coriolis force induces a veering to the right in the Northern Hemisphere, and to the left in the Southern Hemisphere.
The optimal filter pattern for MAM highlights the cross-Equator flow at the expense of the weakening of the NE trades. This result suggests that the response to the local tropical Atlantic SST anomalies dominates the response to the remote tropical Pacific SST anomalies during this season. This same conclusion was reached by Chang et al. (2000) and Saravanan and Chang (2000), and is explicit in the empirical analysis of Pacific and Atlantic influences on northeast Brazil rainfall of Ward and Folland (1991).
Why should fluctuations in the Atlantic dipole SST pattern show any correlation with ENSO-related tropical Pacific SST? Past studies shed some light on this question. In a composite analysis of observational data Curtis and Hastenrath (1995) described how, following a Pacific warm event, positive SST anomalies develop north of the equator in March and April (and persist into boreal summer) and weaker negative SST anomalies develop south of the equator in May and June. They explained the former anomalies in terms of the ocean’s response to the weakening of the NE trades in boreal winter. They could find no explanation in their observational datasets, however, for the development of the latter anomalies. Our results may offer some explanation. Figure 4 suggests that, in the boreal spring following Pacific warm events, there is a strengthening of the southeast (SE) trades in the latitude band 0°–10°S. Associated with this strengthening we would expect enhanced evaporation and enhanced upwelling off the west coast of Africa. Both these processes will act to reduce SST. (See, however, the further analysis in section 5.)
In the above view, El Niño events do not force dipolelike SST anomalies directly; rather, there is a two stage process. First, El Niño forces reduced evaporation, leading to warming of the oceanic mixed layer, to the north of the equator during boreal winter. Second, the atmospheric response to this northern warming forces cooling of the southern oceanic mixed layer, which lags the warming to the north. This lag is clearly evident in the analysis of Curtis and Hastenrath (1995), but in our analysis it is somewhat obscured by the low time resolution inherent in studying seasonal means.
The fact that ENSO can force the Atlantic SST dipole does not mean that all fluctuations in the dipole are simply a passive response to ENSO. Many other processes can affect the SST dipole, as will be discussed in section 6. See also Saravanan and Chang (2000) for further discussion of these issues.
Figure 4c shows results for JJA. The SST pattern again picks out the tropical Atlantic dipole, and the optimized time series is highly correlated (coefficient 0.8) with that for MAM, presumably indicating memory in the oceanic mixed layer. The main wind stress fluctuations are in the western equatorial Atlantic where there is an intensification of the SE trades. (At this time of year the SE trades extend across the equator; see Fig. 1). As in MAM, the anomalous cross-equator flow is directed toward the hemisphere in which SST is anomalously high.
An interesting feature of the optimal filter pattern for JJA is that it draws attention to anomalous monsoonal flow onto North Africa. When the Atlantic SST to the south of the equator is anomalously low the monsoonal flow is anomalously strong, as one would expect. This result suggests potential predictability of North African summer rainfall (Folland et al. 1991; Ward et al. 1993;Carson 1998).
The SST pattern for SON (Fig. 4d) highlights ENSO-type anomalies in the Pacific, though the largest weights are farther east along the equator than was found for DJF. The wind stress pattern suggests a modest strengthening (note that the scale in Fig. 4d differs to that used in Figs. 4a–c) of the NE trades in response to warm events, which interestingly is the opposite sign to the response found in DJF. South of the equator the results suggest a modest weakening of the SE trades, which could be viewed as the counterpart to the weakening of the NE trades in MAM.
As in the other seasons, the optimal filter pattern for SON highlights wind stress fluctuations close to the equator. Our attention is particularly drawn to the fluctuations in the near-equatorial trades in the western part of the basin. Close by there is a notable feature in the midbasin equatorial SST. The relationship between these equatorial features is such that, when the midbasin SST is anomalously high, the trade winds to the west are anomalously weak. This is the same relationship that is found between the equatorial trades and equatorial SST in the Pacific, and lies at the heart of the Bjerknes (1969) instability mechanism for ENSO. It is also the relationship that Zebiak (1993) presented as evidence of an Atlantic ENSO-like mode of climate variability. This analysis suggests that boreal fall is likely to be a key season during which the ocean–atmosphere coupling takes the form needed to sustain such a mode (or at least sustain the instability mechanism) in the Atlantic. The same conclusion has been reached by Chang et al. (2000).
Figure 4d suggests that fluctuations in midbasin equatorial Atlantic SST are to some extent correlated with ENSO during SON. This suggestion is at odds with the view of Zebiak (1993) that the Pacific ENSO mode and the Atlantic ENSO-like mode are essentially independent. The possibility that there could, nevertheless, be interactions between these modes will be discussed further in section 6.
b. Regression on indices of SST variability
The results presented in the previous section establish that, in the model we have used, the dominant modes of SST-forced variability in the tropical Atlantic region are responses to (a) tropical Pacific SST anomalies associated with ENSO and (b) a tropical Atlantic dipole SST pattern. There is also evidence of a response to midbasin equatorial Atlantic SST anomalies. In this section we investigate further the seasonality of these three signals. Rather than further exploring the dominant modes of the atmospheric variability, the approach we take here is to regress the ensemble mean fields onto indices of SST. This approach enables us to focus on the atmospheric fluctuations associated with a particular feature of the SST field, regardless of whether it is the dominant influence. Furthermore we are able to examine the atmospheric fluctuations associated with different features of the SST field without our results being subject to unphysical orthogonality constraints that necessarily influence the structure of modes in an EOF or similar analysis. Last, we are able to assess the local significance of the various SST-forced signals. The previous analysis provided only global measures of significance (summarized in Table 2).
The first step in the regression analysis is to choose SST indices that are suitable for characterizing the three signals of interest. One possibility for the first two indices would be to use the time series derived from the analysis in the previous section. As we have seen, however, which signal dominates depends on season, so we would have to favor one season in our choice. Another possibility, which we select, is to define indices based on SST variability in prescribed spatial regions. There is necessarily a degree of arbitrariness in the choice of region, but in practice the results are not very sensitive to small changes in the boundaries of these regions. Figure 5 shows the regions we use to define our SST indices. The definition of each region is independent of season. The indices themselves vary with season but Fig. 6 shows that, by comparison with the common fluctuations, this variation is generally small. Note that the different indices are not uncorrelated. In particular, the dipole and ATL3 [“ATL3” is a region of the equatorial Atlantic defined by Zebiak (1993); see Fig. 5] indices are anticorrelated at a level of ∼0.5–0.6 throughout the year. This anticorrelation results principally from the fact that SST in ATL3 is correlated with SST in the southern lobe of the dipole. The ENSO and Atlantic dipole indices are best correlated (coefficient ∼0.4) in MAM, are marginally correlated (coefficient ∼0.2) in JJA, and exhibit no significant correlation in DJF and SON. The ENSO and ATL3 indices show no significant correlation in any season, except possibly DJF (coefficient ∼0.2). The physical significance of these correlations between the different SST indices will be discussed in section 6.
Figure 7 shows the results of regressing the ensemble mean surface wind stress on the ENSO SST index. In the regression procedure the residual is modeled as an AR(1) (red noise) process (Allen and Smith 1994). In DJF, MAM, and SON the patterns of wind stress agree well with the results of the EOF analysis (Fig. 4) as we would expect. In DJF and MAM the largest fraction of the ensemble mean variance (50%–70%) is explained in the region of the NE trades. In MAM there is a secondary maximum in explained variance (∼30%) associated with the cross-equatorial flow in the western Atlantic. The wind stress pattern for JJA differs from that of the dominant mode of variability (Fig. 4c), consistent with the influence of ENSO being relatively weak in this season. The most coherent feature in JJA is the weakening of the SE trades. This signal during austral winter could be viewed as the counterpart of the weakening of the NE trades during boreal winter. As the mean state in the tropical Atlantic is so asymmetric (e.g., the ITCZ is, except briefly in boreal spring, always found north of the equator) this symmetry in the response to ENSO is perhaps surprising. It lends further support to the view that the weakening of the trades derives from a disturbance propagating through the extratropics of the winter hemisphere, rather than from changes to the tropical Walker and Hadley circulations.
Figure 8 shows the results of regression on the dipole SST index. As we would expect from the analysis in section 4a this index accounts for most variance in MAM and JJA. The pattern of wind stress in MAM is very similar to that seen in the regression on the ENSO index (Fig. 7), but there are notable differences in the distribution of explained variance. The dipole index accounts for a substantially greater fraction of the cross-equatorial flow (∼50%–70% vs ∼30%) than did the ENSO index, while the ENSO index accounts for a greater fraction of variance in the NE trades (∼50% vs ∼30%–40%). These results support the view that the cross-equatorial flow is most strongly influenced by variability in the Atlantic dipole pattern, while the NE trade winds are most strongly influenced by ENSO variability in the tropical Pacific. As we discussed in the previous section, the dipole pattern is itself partly forced by ENSO.
Anomalous cross-equatorial flow in the western Atlantic, directed toward the positive lobe of the Dipole, is found in all four seasons. This result suggests that, inspite of the rather large seasonal changes to the background mean state (which imply that a given SST anomaly will be associated with quite different absolute values of SST; see Fig. 1), there are consistent elements in the response to the SST dipole anomaly pattern. There are also differences, however. In JJA the response is preferentially found in the Northern Hemisphere, while in DJF there is a stronger response in the Southern Hemisphere. This contrast suggests a bias in the response toward the hemisphere where summer conditions (and therefore warmer absolute SSTs) prevail. SON, like JJA, shows a bias toward the Northern Hemisphere.
Figure 9 shows results of regression on the ATL3 SST index. Zebiak (1993) presented evidence from observations of anomalous convergence toward the ATL3 region, especially from the west, when positive SST anomalies are located there. He did not present a seasonal decomposition of the wind stress anomalies. Figure 9 suggests a substantial seasonal dependence of the wind stress anomalies associated with fluctuations in ATL3 SST. JJA and SON show the clearest convergence, predominantly from the west, toward the ATL3 region. This result supports the suggestion in section 4a and Chang et al. (2000) that the Bjerknes (1969) instability mechanism could be effective in the tropical Atlantic in boreal fall, and suggests that in addition it could be effective in boreal summer.
The wind stress pattern in MAM is rather different to that in JJA and SON. Comparison with Fig. 8 shows that it is close to being the reverse of the pattern associated with the dipole index (or equivalently the pattern associated with a negative value of the dipole index). This is consistent with the fact that the ATL3 index is anticorrelated with the dipole index. The presence of westerly anomalies in the equatorial trades suggests there may be some scope for the Bjerknes mechanism to operate in this season too, but the results also suggest that interactions between variability associated with the dipole index and variability in equatorial Atlantic SST are particularly likely in this season. The possibility and implications of such interactions will be discussed further in section 6.
The wind stress pattern in DJF suggests that the Bjerknes instability mechanism is unlikely to be operative in the tropical Atlantic in this season. The fractions of variance explained by the ATL3 index are also lowest, suggesting that fluctuations in equatorial SST are not an important influence in this season.
5. The influence of atmospheric variability on the ocean
In this section we consider how the patterns of atmospheric variability identified in the previous analyses would force the ocean. We are especially interested in how the atmospheric variability could force changes in SST. Such changes may result from changes in the air– sea heat fluxes (Nobre and Shukla 1996; Carton et al. 1996) or from wind-forced changes in ocean dynamics (Zebiak 1993; Carton et al. 1996; Huang and Shukla 1997). The nonlocal nature of the latter processes, which are likely to be especially important close to the equator, means that integrations of an ocean model are needed to investigate them fully. Such integrations are beyond the scope of the present study; thus we focus here on changes in the air&ndash≃a heat fluxes. In particular we consider changes in the latent and shortwave surface heat fluxes, which make the largest contributions to variations in the total surface heat flux. Variations in the latent heat flux at these latitudes are mainly due to variations in the wind speed rather than variations in near-surface specific humidity (Cayan 1992).
We compute the heat flux anomalies associated with each of the three SST indices by regressing the ensemble mean heat flux onto each index in turn. Figure 10 shows the latent and shortwave heat fluxes obtained by regression onto the ENSO SST index. The major signals are in DJF and MAM and results are presented for these seasons only. The weakening of the NE trades in DJF is associated with a positive latent heat flux anomaly (i.e., a reduction in the mean cooling by evaporation). In the real world this effect leads to warmer than usual SST anomalies in the spring following an ENSO event (Curtis and Hastenrath 1995; Enfield and Mayer 1997). Our results for DJF suggest a second positive latent heat flux anomaly south of the equator, associated with weakening of the SE trades. In this region too we might therefore expect warmer than usual SST in the spring following an ENSO event. This expectation is at odds with the claim of Curtis and Hastenrath (1995) that ENSO events are followed by colder than usual boreal spring SST in the southern tropical Atlantic. This apparent discrepancy may indicate that the response to ENSO south of the equator is incorrect in the model. Alternatively, it may be that the SST tendency in this region is determined by influences other than the DJF latent heat flux anomalies.
Figure 10 shows that the latent heat flux anomaly associated with ENSO in MAM is similar to that in DJF, with positive anomalies to the north and south of the equator. The anomalies are, however, weaker than in DJF and in the latitude band 0°–10°S are not significantly different from zero. Figure 4b showed anomalous strengthening of the SE trades in this latitude hand, and it was postulated in section 4a that this strengthening would generate negative latent heat flux anomalies. The fact that any such anomalies that arise in the model do not achieve significance suggests that the cooling of SST by anomalous latent heat flux is a weak effect in this region and season. Cooling by anomalous upwelling along the coast of Africa could be a stronger effect. The fact that the largest magnitude SST anomalies south of the equator are found adjacent to the African coast in Fig. 4b supports this suggestion.
The anomalies in surface shortwave heat flux associated with ENSO are shown in Figs. 10c (DJF) and 10d (MAM). The major feature in Fig. 10c indicates suppression of convection over South America during El Niño events. This result supports the suggestion made in section 4a that changes in the tropical Walker circulation play an important role in the tropical Atlantic response to ENSO.
Figure 10d indicates a large positive heat flux anomaly into the ocean in the latitude band 0°–10°S in the boreal spring of El Niño years. This signal is presumably related to a weakening, or northward retreat, of convection in the ITCZ. North–South displacements of the ITCZ in association with anomalies in the cross-equator SST gradient (which, as we have seen, is correlated with ENSO in MAM) have been noted in several observational studies (Moura and Shukla 1981; Curtis and Hastenrath 1995; Enfield and Mayer 1997). The anomalies in the shortwave heat flux, if realistic, will tend to force warmer than usual SST south of the equator in the boreal summer following an ENSO event.
In the northwestern part of the domain considered we find negative shortwave heat flux anomalies in MAM. These may be associated with enhanced convection during El Niño events. A similar but weaker negative anomaly is seen in DJF.
We do not present the results for JJA and SON but the main features in the anomalous latent flux may be readily anticipated from Fig. 7 (in conjunction with Fig. 1). Where the wind anomalies reduce the mean wind speed, principally to the south of the equator, there are positive latent heat flux anomalies and where the wind anomalies increase the mean wind speed, for example, along the northwest coast of South America in JJA, there are negative latent heat flux anomalies. There are no major anomalies in shortwave heat flux.
Figure 11 shows the surface heat flux anomalies associated with variations in the Atlantic dipole SST index. We have seen that the influence of the Dipole is most important in MAM and JJA and latent heat flux anomalies are shown for these seasons only. In MAM negative anomalies in the western Atlantic at around 5°S are surrounded by a horseshoe-like pattern of positive anomalies. The former are clearly associated with the enhanced cross-equatorial flow that was seen in Fig. 8 and discussed in section 4b. The horseshoe-like pattern of the latter is less easy to anticipate from Fig. 8 but comparison with Fig. 1 shows that there is broad agreement between the regions where the anomalous winds are tending to reduce the mean wind speed and regions where the latent heat flux anomalies are tending to warm the ocean. In JJA the latent heat flux anomalies show a northwest-to-southeast-oriented dipole pattern. Comparison with Fig. 4c shows that the southeastern latent heat flux anomaly is roughly coincident with the southern lobe of the dipole SST anomaly, and would act to damp it. The northwestern latent heat flux anomaly lies equatorward of the northern lobe of the dipole SST anomaly and would therefore neither damp nor reinforce it. The pattern of latent heat flux anomalies for SON is very similar to that for JJA (not shown). With the possible exception of the small region along the South American coast around 10°N in DJF, in no season do we find any evidence of the positive latent heat flux feedback that Chang et al. (1997) suggested could reinforce the SST Dipole pattern.
The anomalies in the surface shortwave heat flux are shown in Fig. 11 for all seasons because they give interesting insights into the way in which the patterns of convection change in response to (or in association with) changes in the SST dipole. In DJF, positive values of the dipole index (i.e., warm to the north of the equator) are associated with increased surface heating by shortwave radiation, suggesting reduced convection, in the region of the ITCZ (close to the equator in this season— see Fig. 1) and South Atlantic convergence zone (SACZ). Since we would expect positive SST anomalies to boost convection, this result suggests that convection in these two regions is influenced more by the southern lobe of the SST dipole than the northern lobe (i.e., when SST in the southern lobe is anomalously cold, convection in the ITCZ and SACZ is suppressed). Between the SACZ and ITCZ, and to the north of the ITCZ, there is reduced surface shortwave, suggesting enhanced convection.
The surface shortwave signature in MAM is very similar to that associated with the ENSO SST index (Fig. 10) and was discussed above. As in DJF, anomalous convection in the ITCZ takes the same sign as the SST anomalies to the south of the equator. In JJA the mean ITCZ is located farther north, around 5°N (see Fig. 1). The pattern of anomalous surface shortwave in Fig. 11e exhibits a dipole suggesting meridional displacements of the ITCZ, toward the hemisphere in which SST is anomalously warm, in response to fluctuations in the SST Dipole (e.g., Lough 1986; Rowell et al. 1995). The pattern for SON (Fig. 11f) is similar, albeit somewhat less clear.
As with the latent heat flux, it is interesting to ask how the anomalies in shortwave heat flux could feedback onto the ocean to affect the SST dipole. Because the pattern of shortwave anomalies differs considerably from the pattern of SST anomalies, the feedback cannot be simply described as “positive” or “negative.” This said, careful comparison between Figs. 11 and 4c shows that when shortwave anomalies are found near the centers of action of the SST dipole, their sign generally suggests a negative feedback.
Figure 12 shows the surface heat flux anomalies associated with variations in the ATL3 SST index. Figure 9 showed that the most important response to ATL3 SST occurs in JJA and SON, so heat flux anomalies are shown for these two seasons. Both the latent and shortwave heat flux anomalies are similar to those seen in Fig. 11b, albeit the signs are opposite; this difference reflects the fact that the ATL3 index is anticorrelated with the dipole index. Comparison of the latent heat flux anomalies (Figs. 12a,b) with the shortwave heat flux anomalies (Figs. 12c,d) shows that, near the equator, there is some degree of cancellation between these components. This cancellation suggests that the net surface heat flux feedback onto the ocean is small. It is likely in this case that dynamical feedbacks will be much more important than thermodynamic feedbacks. Figure 9 showed that, associated with fluctuations in ATL3 SST, there are substantial wind anomalies in the western Atlantic near the equator in JJA and SON. We may expect these wind anomalies to be effective at exciting equatorial waves in the ocean. The resulting adjustments to the thermocline depth will change ATL3 SST in a manner similar to that which occurs during ENSO in the Pacific.
6. Discussion
It was emphasized in the introduction that, unlike the tropical Pacific, the tropical Atlantic is not dominated by any single mode of climate variability. The analyses presented in this paper have characterized the major features of four different signals—or “elments”&mdash at are important in this region. These are the following: (a) the NAO, which is the dominant component of the internal atmospheric variability; (b) the remote response to ENSO in the Pacific; (c) the response to the Atlantic cross-equator SST dipole pattern; and (d) the response to equatorial Atlantic SST. Various characteristics of these signals have been investigated and these characteristics shed light on the underlying physical mechanisms that are the key things we wish to understand. Below, in section 6a, we summarize and contrast the mechanisms responsible for each of the different elements we have considered. We also consider some of the mechanisms via which the different elements may interact.
An important aim of the study was to quantify the magnitude of the different signals as a function of season. This has been done on a signal-by-signal basis. What remains is to consider how the different signals fit together to determine the total variance, and what are the consequences for the predictability of climate fluctuations. We address these issues in section 6b.
a. Physical mechanisms responsible for climate variability in the tropical Atlantic region
As a part of the internal atmospheric variability the NAO is excited principally by instabilities of the atmospheric flow. The surface heat flux anomalies that arise in association with the NAO force changes in the north and tropical Atlantic SST (Cayan 1992). These SST changes may feedback to force the NAO but the indications are that any such feedbacks are weak, at least in boreal winter (Rodwell et al. 1999; Sutton et al. 2000, manuscript submitted to Atmos. Sci. Lett.).
In contrast with the NAO, fluctuations in SST play a fundamental role in each of the other three elements of climate variability that have been considered. The first of these—ENSO—is unambiguously a coupled ocean–atmosphere mode of variability. It exhibits a prefered interannual timescale and its existence relies upon a well established positive feedback mechanism that involves changes in both the ocean and the atmosphere. This study has focused on the remote response to ENSO over the tropical Atlantic region. This response arises through atmospheric teleconnections, but we have seen that these atmospheric signals can force changes in the Atlantic Ocean. The SST anomalies that subsequently develop then further force the atmosphere.
There have been suggestions (e.g., Chang et al. 1997) that the Atlantic dipole might also be viewed as a coupled ocean–atmosphere mode of variability. This study suggests, however, that it is probably best viewed simply as a sensitivity of the atmosphere to fluctuations in the cross-equator SST gradient. In the first place, SST anomalies to the north and south of the equator are not significantly anticorrelated and there is little evidence of any prefered timescale (Houghton and Tourre 1992;Mehta 1998). Second, contrary to the suggestion of Chang et al. (1997), our results do not indicate that the atmospheric response to this SST gradient feeds back onto the ocean in such a way as to reinforce it. The latent heat flux anomalies suggest a negative feedback.
If coupled atmosphere–ocean feedbacks do not significantly reinforce the cross-Equator SST gradient, then other processes must be responsible for forcing the fluctuations in this gradient that are observed. It is likely that several different processes, operating on a range of timescales, play a role. We have discussed, for example, how the remote response to ENSO can force the northern lobe of the dipole. The same is true of fluctuations in the NAO since it too modulates surface latent heat fluxes north of the equator. Oceanic Rossby waves forced by wind fluctuations can also be expected to modulate the SST dipole by changing the depth of the thermocline, and anomalous oceanic advection (both Ekman and geostrophic) is likely to be important too. Last, it has recently been proposed by Yang (1999) that changes in North Atlantic deep convection could force changes in the tropical Atlantic Dipole via changes in the meridional overturning circulation.
The last element considered was the Atlantic ENSO-like mode described by Zebiak (1993). This study supports the view that the Bjerknes positive feedback mechanism that is fundamental to the Pacific ENSO could also operate in the Atlantic. We found that this feedback is most likely to operate in boreal summer and fall. There is also some evidence from observations of a prefered timescale (Zebiak 1993; Latif and Gröñer 2000). Thus it is probably appropriate to view the Atlantic ENSO as a coupled ocean–atmosphere mode of variability. This said, our results suggest that the Atlantic ENSO is likely to be significantly influenced by (and perhaps sometimes suppressed by) other elements of tropical Atlantic climate variability. In particular, we found that the response to the Atlantic dipole may include significant equatorial zonal wind anomalies, especially in MAM. We would expect these wind anomalies to excite equatorially trapped waves in the ocean raising the possibility of interactions between fluctuations in Atlantic SST Dipole and ENSO-like equatorial Atlantic fluctuations.
b. Contributions to the total variance and potential predictability of atmospheric fluctuations
In section 4b we considered the atmospheric fluctuations associated with a particular SST index. We consider here how some chosen indices of atmospheric variability are influenced by different aspects of the SST field, and by internal variability. Figure 13 shows contributions to the variance in three different indices of wind stress variability. These indices describe fluctuations in: (a) the NE trades in the latitude band 10°– 20°N; (b) the cross-equator flow in the western Atlantic (50°–20°W); (c) the equatorial trades in the central Atlantic (40°–20°W).
Figure 13a shows a strong seasonal cycle in the total variance of the NE trades, with most variance in boreal winter and spring. Throughout the year, around 70% of this variance is associated with internal atmospheric variability. The remaining variance is forced by fluctuations in SST and, as could be anticipated from our earlier results, the dominant influence in DJF and MAM is ENSO. By contrast, in JJA and SON none of the three SST indices accounts for substantially more of the variance than the others. (Note that, because the three SST indices are not uncorrelated the fractions of the total variance explained will not in general sum to 1; see figure caption for more details.)
The ratio of the SST-forced variance to the total variance is a measure of the “potential predictability” (e.g., Rowell 1998), that is the predictability of atmospheric fluctuations assuming a knowledge of the SST. Figure 13a shows that the potential predictability of fluctuations in the NE trades is generally low (∼30%), but note that this is an average value for all years. Since ENSO events occur only every few years, and the figure shows that the potential predictability in this index derives mainly from ENSO, we can infer that if we had knowledge that an ENSO event was occuring then the actual predictability of this index would be substantially higher than is indicated by the long-term average value.
The variance in the cross-equator flow index shows a much less strong seasonal cycle than is seen in the northeast trades (Fig. 13b). The total variance reaches a maximum in boreal spring and a minimum in boreal autumn. In contrast to the NE trades, the variance is predominantly SST forced indicating a much higher level of potential predictability. The internal variance accounts for at most 32% of the total (in SON), and accounts for only 15% of the total in MAM. The absolute level of internal variance is almost constant throughout the year, thus the seasonality in the total variance derives mainly from the SST-forced component.
In all seasons it is, as we expect, the dipole SST index that accounts for the largest fraction of the SST-forced variance with little sign of an important role for the either ENSO or ATL3 SST. In MAM the dipole accounts for 82% of the SST-forced variance, or 70% of the total variance, suggesting the anomalous cross-equator flow is very strongly controlled by the anomalous cross-equator SST gradient in this season. By contrast, it is interesting that in JJA and SON the three SST indices considered together account for less than 50% of the SST-forced variance. This result suggests that, in these seasons, some other features of the SST field have a significant role in forcing cross-equator flow.
The total variance of the equatorial trades peaks much more strongly in boreal spring (Fig. 13c). As with the cross-equator flow, the absolute level of internal variance is almost constant throughout the year, so the seasonality in the total variance derives overwhelmingly from the SST-forced component. In MAM and JJA, when the total variance is greatest, the internal variance contributes less than 20%, indicating a high level of potential predictability in these seasons.
Of the three SST indices, the ENSO index accounts for the smallest fraction of the SST-forced variance, which suggests that forcing of the equatorial Atlantic by ENSO is not particularly important. In DJF, the ATL3 index accounts for the largest fraction but it is interesting that in the other three seasons the dipole index accounts for a little more variance than does the ATL3 index. If fluctuations in the equatorial trades were controlled mainly by fluctuations in equatorial SST, as is the case in the Pacific, then we would expect the ATL3 index to account for most variance. These results suggest that fluctuations in the dipole SST index could have an important role in forcing the equatorial trades. Because we expect the ocean to be especially sensitive to fluctuations in the equatorial trades, such an influence could have important implications for climate variability in the tropical Atlantic region. These implications will be discussed below.
7. Conclusions
The major processes that contribute to atmospheric variability in the tropical Atlantic region have been investigated using an ensemble of integrations of an atmospheric GCM. The ensemble permits a separation between the essentially unpredictable variability that arises solely from processes internal to the atmosphere and the potentially predictable variability that arises in response to changes in SST.
The results support the following conclusions.
Internal variability in the tropical Atlantic region is dominated by the equatorward extension, especially from the Northern Hemisphere, of extratropical patterns (or modes). The North Atlantic oscillation (NAO) is the most important such mode.
There are at least three SST-forced signals that can be identified in the tropical Atlantic region. The importance of these signals varies with season and this seasonality is summarized in Fig. 14. SST anomalies associated with Pacific ENSO events exert maximum influence during DJF and MAM. The Atlantic SST dipole pattern exerts maximum influence in MAM and JJA. Equatorial Atlantic SST anomalies exert maximum influence in JJA and SON.
The main feature of the response to ENSO is a weakening of the NE trades in DJF (especially) and in MAM. The main feature of the response to the Atlantic Dipole is a cross-equatorial flow directed toward the hemisphere in which SST is anomalously high. The main feature of the response to equatorial Atlantic SST is seen in zonal wind anomalies to the west of the SST maximum, and in convergence toward the SST maximum.
An analysis of feedbacks from the atmospheric signals back onto the ocean supports the view that the Bjerknes (1969) instability mechanism that plays a fundamental role in Pacific ENSO events is also likely to operate in the equatorial Atlantic, especially during boreal summer and fall. This analysis did not, however, support the suggestion of Chang et al. (1997) that the Atlantic dipole should be viewed as a coupled ocean–atmosphere mode of variability. The dipole is probably better viewed simply as a sensitivity of the atmosphere to variations in the cross-equator SST gradient.
An analysis of the potential predictability of atmospheric fluctuations in the tropical Atlantic region indicated high potential predictability for fluctuations in zonal and meridional flow near the equator, with maximum potential predictability in boreal spring. Fluctuations in the trade winds in the latitude band 10°– 20°N exhibit substantially lower potential predictability, but the potential predictability is highest during ENSO (El Niño or La Niña) events.
The results of this study highlight the need for considerable further work to understand the nature and predictability of climate fluctuations in the tropical Atlantic region. The results also offer, however, clear guidance as to where attention should be directed. Specifically, effort should focus on improving understanding, simulation, and prediction of the dominant elements that have been discussed. There is considerable scope for further elucidation of the physical mechanisms that underly these elements and their interactions. Future research is needed to further investigate the atmospheric mechanisms and also the dynamical (as opposed to thermodynamic) feedbacks onto the ocean, which were beyond the scope of this study. The role of land surface processes in the variability of tropical Atlantic climate must also be investigated.
An important feature of the present study has been the effort to quantify the comparative importance of different aspects of climate variability in different regions and seasons, and the consequences for predictability. This effort was summarized most succinctly in Fig. 13. Such analyses offer a helpful way to compare results from different models, and the authors therefore hope that this type of approach will be more widely adopted.
Acknowledgments
We would like to thank Myles Allen for his assistance with the optimal detection analysis and Alison Renshaw, John Davies, and David Sexton for running the model. RTS and SPJ were supported by the U.K. Universities Global Atmospheric Modelling Programme (UGAMP). DPR’s work was supported by the U.K. Public Meteorological Service Research and Development Programme, Contract MSG-2/97.
REFERENCES
Allen, M. R., and L. A. Smith, 1994: Invesigating the origins and significance of low frequency modes of climate variability. Geophys. Res. Lett.,21, 883–886.
——, and ——, 1997: Optimal filtering in singular spectrum analysis. Phys. Lett.,234, 419–428.
Bjerknes, J., 1969: Atmospheric teleconnections from the equatorial pacific. Mon. Wea. Rev.,97, 163–172.
Bretherton, C. S., C. Smith, and J. M. Wallace, 1992: An intercomparison of methods for finding coupled patterns in climate data. J. Climate,5, 541–560.
Carson, D. J., 1998: Seasonal forecasting. Quart. J. Roy. Meteor. Soc.,124, 1–26.
Carton, J. A., X. Cao, B. S. Giese, and A. M. da Silva, 1996: Decadal and interannual SST variability in the tropical Atlantic Ocean. J. Phys. Oceanogr.,26, 1165–1175.
Cayan, D. R., 1992: Latent and sensible heat-flux anomalies over the northern oceans—driving the sea-surface temperature. J. Phys. Oceanogr.,22, 859–881.
Chang, P., L. Ji, and H. Li, 1997: A decadal climate variation in the tropical Atlantic Ocean from thermodynamic air&ndash≃a interactions. Nature,385, 516–518.
——, R. Saravanan, L. Ji, and G. C. Hegerl, 2000: The effect of local sea surface temperatures on atmospheric circulation over the tropical Atlanic sector. J. Climate,13, 2195–2216.
Cullen, M. J. P., 1993: The unified forecast/climate model. Meteor. Mag.,122, 81–94.
Curtis, S., and S. Hastenrath, 1995: Forcing of anomalous sea surface temperature evolution in the tropical Atlantic during pacific warm events. J. Geophys. Res.,100, 15 835–15 847.
da Silva, A., A. C. Young, and S. Levitus, 1994: Atlas of surface marine data. Tech. Rep. 6, U.S. Department of Commerce, NOAA, NESDIS, 83 pp.
Davies, J. R., D. P. Rowell, and C. K. Folland, 1997: North Atlantic and European seasonal predictability using an ensemble of multi-decadal agcm simulations. Int. J. Climatol.,17, 1263–1284.
Enfield, D. B., and D. A. Mayer, 1997: Tropical Atlantic SST variability and its relation to El Niño–Southern Oscillation. J. Geophys. Res.,102 (C1), 929–945.
Folland, C. K., T. N. Palmer, and D. Parker, 1986: Sahel rainfall and worldwide sea temperatures: 1901–85. Nature,320, 602–606.
——, J. Owen, M. N. Ward, and A. Colman, 1991: Prediction of seasonal rainfall in the sahel region using empirical and dynamical methods. J. Forecasting,10, 21–56.
Frankignoul, C., and K. Hasselmann, 1977: Stochastic climate models. Part II. Application to sea-surface temperature anomalies and thermocline variability. Tellus,29, 289–305.
Gates, W. L., 1992: AMIP: The atmospheric model intercomparison project. Bull. Amer. Meteor. Soc.,73, 1962–1970.
Goldenberg, S. B., and L. J. Shapiro, 1996: Physical mechanisms for the association of El Niño and West African rainfall with Atlantic major hurricane activity. J. Climate,9, 1169–1187.
Gray, W. M., C. W. Landsea, P. W. Mielke Jr., and K. J. Berry, 1993:Predicting Atlantic basin seasonal tropical cyclone activity by 1 August. Wea. Forecasting,8, 73–86.
Hasselmann, K., 1979: On the signal-to-noise problem in atmospheric response studies. Meteorology of Tropical Oceans, D. B. Shawn, Ed., Royal Meteorological Society, 251–259.
Hastenrath, S., and L. Greischar, 1993: Further work on the prediction of northeast Brazil rainfall anomalies. J. Climate,6, 743–758.
Houghton, R. W., and Y. M. Tourre, 1992: Characteristics of low-frequency sea surface temperature fluctuations in the tropical atlantic. J. Climate,5, 765–771.
Huang, B., and J. Shukla, 1997: Characteristics of the interannual and decadal variability in general circulation model of the tropical Atlantic ocean. J. Phys. Oceanogr.,27, 1693–1712.
Hulme, M., Y. Biot, J. Borton, M. Buchanan-Smith, S. Davies, C. Folland., N. Nicholds, D. Seddon, and N. Ward, 1992a: Seasonal rainfall forecasting for Africa. Part I: Current status and future developments. Int. J. Environ. Stud.,39, 245–256.
——, and Coauthors, 1992b: Seasonal rainfall forecasting for Africa. Part II. Application and impact assessment. Int. J. Environ. Stud.,40, 103–121.
Landsea, C. W., 1993: A climatology of intense (or major) Atlantic hurricanes. Mon. Wea. Rev.,121, 1703–1713.
Latif, M., and A. Gröñer, 2000: The equatorial Atlantic oscillation and its response to ENSO. Climate Dyn.,16, 213–218.
Lindzen, R. S., and S. Nigam, 1987: On the role of sea-surface temperature-gradients in forcing low level winds and convergence in the Tropics. J. Atmos. Sci.,44, 2418–2436.
Lough, J. M., 1986: Tropical Atlantic sea surface temperature and rainfall variations in subsaharan Africa. Mon. Wea. Rev.,114, 561–570.
Mehta, V., 1998: Variability of the tropical ocean surface temperatures at decadal–multidecadal timescales. Part 1: The Atlantic Ocean. J. Climate,11, 2351–2375.
Moura, A. D., and J. Shukla, 1981: On the dynamics or droughts in northeast brazil: Observations, theory and numerical experiments with a general circulation model. J. Atmos. Sci.,38, 2653– 2675.
Nobre, P., and J. Shukla, 1996: Variations of sea surface temperature, wind stress, and rainfall over the tropical Atlantic and South America. J. Climate,9, 2464–2479.
Parker, D. E., C. K. Folland, A. Bevan, M. N. Ward, M. Jackson, and F. Maskell, 1995: Marine surface data for analysis of climate fluctuations on interannual to century timescales. Natural Climate Variability on Decadal to Century Timescales, D. G. Martinson, Ed., National Academy Press, 241–250.
Phillips, T. J., 1994: A summary documentation of the AMIP models. Tech. Rep. 18, PCMDI, Lawrence Livermore National Laboratory. [Available from Lawrence Livermore National Laboratory, 7000 East Ave., Livermore, CA 94550-9234.].
Potts, J. M., C. K. Folland, I. T. Jolliffe, and D. Sexton, 1996: Revised “leps” scores for assessing climate model simulations and long-range forecasts. J. Climate,9, 34–53.
Rajagopalan, B., Y. Kushnir, and Y. M. Tourre, 1998: Observed midlatitude and tropical Atlantic climate variability. Geophys. Res. Lett.,25, 3967–3970.
Renshaw, A. C., D. P. Rowell, and C. K. Folland, 1998: Enso responses and low frequency weather variability in the North Pacific/American sector 1943–93. J. Climate,11, 1073–1093.
Rodwell, M. J., D. P. Rowell, and C. K. Folland, 1999: Oceanic forcing of the winter North Atlantic oscillation and European climate. Nature,398, 320–323.
Rowell, D. P., 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J. Climate,11, 109–120.
——, and F. W. Zweirs, 1999: The global distribution of sources of atmospheric decadal variability and mechanisms over the tropical Pacific and southern North America. Climate Dyn.,15, 751– 772.
——, C. K. Folland, K. Maskell, and M. N. Ward, 1995: Variability of summer rainfall over tropical North Africa (1906–92): Observations and modelling. Quart. J. Roy. Meteor. Soc.,121, 669– 704.
Saravanan, R., and P. Chang, 2000. Interaction between tropical Atlantic variability and El Niño–Southern Oscillation. J. Climate,13, 2177–2194.
Venzke, S., M. R. Allen, R. T. Sutton, and D. P. Rowell, 1999: The atmospheric response over the North Atlantic to decadal changes in sea surface temperature. J. Climate,12, 2562–2584.
Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north nordeste of Brazil using eigenvectors of sea-surface temperature. Int. J. Climatol.,11, 711–743.
——, ——, K. Maskell, A. Colman, D. P. Rowell, and K. Lane, 1993:Experimental seasonal forecasting of tropical rainfall at the U.K. Meteorological Office. Prediction of Interannual Climate Variations, J. Shukla, Ed., NATO ASI Series, Vol. 16, Springer-Verlag, 197–216.
Willmott, C. J., C. M. Rowe, and T. Minz, 1985: A global archive or land cover and soils data for use in general circulation models. J. Climatol.,5, 119–143.
Yang, J., 1999: A linkage between decadal climate variations in the Labrador Sea and the tropical Atlantic Ocean. Geophys. Res. Lett,26, 1023–1026.
Zebiak, S. E., 1993: Air&ndash≃a interaction in the equatorial Atlantic region. J. Climate,6, 1567–1586.
Seasonal mean SST (°C, shaded) and surface wind stress (N m−2) from the model
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Percentage of total interannual variance that can be explained as a response to fluctuations in SST: (a)–(d) zonal wind stress, (e)–(h) meridional wind stress. White areas over the ocean show where values do not significantly exceed zero at the 5% significance level
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Dominant EOF of internal variability in tropical Atlantic surface wind stress for each of the four seasons. The vectors are scaled so as to indicate the typical (one standard deviation) magnitude (in N m−2) of fluctuations in wind stress associated with this mode
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Dominant EOF of SST-forced variability in tropical Atlantic surface wind stress, and related patterns of SST, for each of the four seasons: (a) DJF, (b) MAM, (c) JJA, and (d) SON. (top) The dominant EOF (vectors) and the tropical Atlantic part of the related SST pattern (shaded). (middle) The global extent of the related SST pattern (shaded). (bottom) The optimal filter pattern corresponding to the dominant EOF (vectors) and the tropical Atlantic part of the related SST pattern is again reproduced (shaded). For each season the SST pattern is derived by regressing the SST field on the optimized principal component (see text for details). Regions where the regression is not significant are shown blank. As in Fig. 3 the wind stress vectors are scaled so as to indicate the typical magnitude (N m−2) of fluctuations associated with the dominant mode
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
(Continued)
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Regions used to define SST indices. The solid lines indicate the regions used to define the ENSO index in the Pacific (5°S–5°N, 180°–110°W) and the ATL3 index in the Atlantic (2.5°S– 2.5°N, 20°W–0°). The dashed lines indicate the two regions used to define the Atlantic dipole index (10°–20°N, 50°–20°W, and 0°–20°S, 40°W–15°E). The dipole index is defined as the SST averaged over the northern region minus the SST averaged over the southern region
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Time series of the SST indices defined in the caption of Fig. 5 for each of the four seasons: DJF (solid lines), MAM (dashed lines), JJA (dotted lines), SON (dash&ndash⋄tted lines). (a) ENSO index, (b) Atlantic dipole index, and (c) ATL3 index. Note that the vertical scale in (a) differs to that in (b) and (c)
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Surface wind stress obtained by regression on the ENSO SST index. The shading shows the fraction of the variance in the ensemble mean that can be explained by this index, and blank regions indicate where the regression is not significant
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
As Fig. 7 but for the Atlantic dipole SST index
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
As Fig. 7 but for the ATL3 SST index
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Surface heat fluxes (into the ocean) obtained by regression on the ENSO SST index. (a) DJF latent heat flux, (b) MAM latent heat flux, (c) DJF shortwave heat flux, and (d) MAM shortwave heatflux. Blank areas indicate where the regression is not significant. Latent heat fluxes are shown only over the ocean. Shortwave heat fluxes are shown over the land as well because they give information about changes in convection
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
As in Fig. 10 but for the Atlantic dipole SST index. (a) MAM latent heat flux, (b) JJA latent heat flux, (c) DJF shortwave heat flux, (d) MAM shortwave heat flux, (e) JJA shortwave heat flux, and (f) SON shortwave heat flux
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
As in Fig. 10 but for the ATL3 SST index. (a) JJA latent heat flux, (b) SON latent heat flux, (c) JJA shortwave heat flux, and (d) SON shortwave heat flux
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Contributions to the variance in three indices of surface wind stress variability: (a) NE trades index (zonal wind stress averaged over the region 10°–20°N; 50°–10°W); (b) cross-equator flow index (meridional wind stress averaged over the region 4°S–4°N; 50°– 20°W); (c) equatorial trades index (zonal wind stress averaged over the region 4°S–4°N; 40°–20°W). (upper) The total interannual variance (solid lines) and the internal variance (dotted lines) of the index as a function of season. [Units are 10−5(N m−2)2.] The difference between these two lines is the SST-forced variance. (lower) The fraction of the SST-forced variance that can be accounted for by a linear relationship to one of the SST indices: ENSO SST index (solid lines), dipole SST index (dotted lines), ATL3 SST index (dashed line). For the upper panels the fraction of forced variance is calculated as in the ANOVA analysis of Rowell et al. (1995); Rowell (1998). For the lower panels for each SST index the fraction of variance explained, f, is computed according to: f = b2 (variance of SST index)/(total variance of wind stress index), where b is the regression coefficient. Because the three SST indices are not uncorrelated these fractions do not in general sum to 1. Their relative magnitudes should therefore be viewed only as a rough guide to the comparative importance of the associated SST fluctuations
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Seasonality of the dominant elements of climate variability in the tropical Atlantic region. The figure indicates the seasons during which each of the elements discussed in the text accounts for the largest fraction of the total atmospheric variance in its region of influence
Citation: Journal of Climate 13, 18; 10.1175/1520-0442(2000)013<3261:TEOCVI>2.0.CO;2
Percentage of the total internal variance explained by the first three empirical orthogonal functions of surface windstress
Estimated ensemble-mean and within-ensemble variances,
Because we do not expect variability in sea ice to have a major influence on the tropical Atlantic region we henceforth refer simply to “SST-forced variability,” although by so doing we do not rule out a possible role for sea ice.
Consequently, optimal filters are of fundamental importance in any problem involving the detection of weak signals in the presence of contaminating noise, see, e.g., Hasselmann (1979).
