Abstract

Many previous studies point to a connection between the annual cycle and interannual variability in the tropical Atlantic Ocean. To investigate the importance of the annual cycle in the generation of tropical Atlantic variability (TAV) as well as its associated coupled feedback mechanisms, a set of controlled experiments is conducted using a global coupled ocean–atmosphere general circulation model (GCM) in which the climatological annual cycle is modified. An anomaly coupling strategy was developed to improve the model-simulated annual cycle and mean sea surface temperature (SST), which is critical to the experiments. Experiments include a control simulation in which the annual cycle is present and a fixed annual cycle simulation in which the coupled model is forced to remain in a perpetual annual mean state. Results reveal that the patterns of TAV, defined as the leading three rotated EOFs, and their relationship to coupled feedback mechanisms are present even in the absence of the annual cycle, suggesting that the generation of TAV is not dependent on the annual cycle. Each pattern of variability arises from an alteration of the easterly trade winds. Results suggest that it is the presence of these winds in the mean state that is the determining factor for the structure of the coupled ocean–atmosphere variability. Additionally, the patterns of variability persist longer in the simulation with no annual cycle. Most remarkable is the doubling of the decay phase related to the north tropical Atlantic variability, which is attributed to the persistence of the local wind–evaporation–sea surface temperature (WES) feedback mechanism. The author concludes that the annual cycle acts to cut off or interrupt conditions favorable for feedback mechanisms to operate, therefore putting a limit on the length of the event life cycle.

1. Introduction

Sea surface temperature (SST) variability about a time-independent background climate state in the tropical Atlantic Ocean is hypothesized to not be self sustaining (Chang et al. 1997, 2001; Xie 1999; Xie et al. 1999), and thus, external forcing is needed. One possibility is that the annual cycle, and in particular the migration of the intertropical convergence zone (ITCZ), provides that forcing. Due to its position, the ITCZ influences either the northern or southern equatorial trade winds. A perturbation about this seasonal state could generate the interannual variability associated with tropical Atlantic variability (TAV). Additionally, it may be that these seasonal states provide the necessary background conditions for SST variability in that region to grow. Specifically, the feedback mechanisms related to TAV, which act to enhance and sustain the variability, could depend on the background state and thus only operate during favorable mean conditions. Therefore, the variability exhibited in the tropical Atlantic could simply be an enhancement or reduction of the annual cycle, as has been suggested by previous research (Nobre and Shukla 1996; Hastenrath 1984). Whether the annual cycle is a necessary component to the existence of TAV is the question addressed in this paper.

Some of the most compelling evidence of a connection between TAV and the annual cycle is the similarity in the spatial structure of seasonal variability compared to interannual variability. This similarity has previously been discussed in Mitchell and Wallace (1992) and Chiang and Vimont (2004). Both of these studies suggest that the processes involved in annual and interannual variability may be similar. To illustrate this relationship, we provide a brief description of the features of the annual cycle in its two extreme seasons and compare that to interannual variability, defined by the rotated empirical orthogonal functions (REOFs) of SST anomaly (SSTA) over the tropical Atlantic Ocean.

SST in the deep tropics is at its warmest during boreal spring, when the ITCZ is at its southernmost position. Seasonal SST anomalies (from an annual mean) are characterized by a strong cross-equatorial gradient with a dipole structured pattern, which is accompanied by cross-equatorial winds. The winds flow from the cold hemisphere (north of the equator) toward the warm hemisphere (south of the equator) and are turned by the Coriolis force so that during this season, the northeast trade winds are enhanced and the southeast trade winds are weakened. Due to the more southerly position of the ITCZ, the enhanced northern trade winds dominate the majority of the North Atlantic region. During boreal late summer and early fall, when the ITCZ is at its northernmost position, the gradient of anomalous SST (as a departure from an annual mean) across the equator is at its weakest seasonal state. Strong easterlies are present along the equator driving upwelling, which leads to the coldest seasonal SST in the eastern tropical Atlantic. This season favors strong easterly trade winds on and to the south of the equator, while the northeastern trade winds are weaker. A more detailed description of the tropical Atlantic annual cycle is provided in Wang et al. (2004) and Okumura and Xie (2006).

To illustrate interannual variability, we examine the SSTA REOF patterns and correlate their principal component time series with the wind stress field to find their associated wind patterns (Fig. 1). The data used for this analysis for both SST and wind stress is the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996). The similarities between the leading pattern (Fig. 1, top panel) with the boreal summer season and the second leading pattern (Fig. 1, middle panel) with the boreal spring season are evident. Note that the variability associated with the leading pattern is an equatorial tongue coincident with weakened equatorial and southeast trade winds, both of which are the characteristics of the boreal late summer/early fall. Variability associated with the second leading pattern is a near-equatorial gradient in SSTA coincident with winds flowing up gradient and turned by the Coriolis force such that they affect the northeast trade winds. These characteristics correspond to the seasonal variations of the boreal spring season. The fact that interannual variability resembles variability found within the annual cycle is a suggestion that the interannual variability might simply be an enhancement or reduction of the annual cycle and may be dependent on its variability for existence.

Fig. 1.

Correlation of observed zonal and meridional wind stress with the (top) ATL3 index, (middle) the NTA, and (bottom) SSA REOF PC time series. The shading represents the corresponding pattern of SSTA for reference (in °C). For the top panel, the pattern displayed is the regression of the ATL3 index onto SSTA, while for the middle and bottom panels, the pattern is the corresponding EOF pattern of SSTA. The units for SSTA are °C. The vectors are the vector representation of the zonal and meridional correlation values. All data are from NCEP–NCAR reanalysis (Kalnay et al. 1996).

Fig. 1.

Correlation of observed zonal and meridional wind stress with the (top) ATL3 index, (middle) the NTA, and (bottom) SSA REOF PC time series. The shading represents the corresponding pattern of SSTA for reference (in °C). For the top panel, the pattern displayed is the regression of the ATL3 index onto SSTA, while for the middle and bottom panels, the pattern is the corresponding EOF pattern of SSTA. The units for SSTA are °C. The vectors are the vector representation of the zonal and meridional correlation values. All data are from NCEP–NCAR reanalysis (Kalnay et al. 1996).

Evidence from previous research also points to the influence of the annual cycle in the tropical Atlantic. For example, Weare et al. (1976) and Weare (1977) reveal the potential dominance of the annual cycle in the Atlantic. They found that while variability in both the tropical Pacific and Atlantic Oceans contains seasonal variability, the Pacific exhibits much more interannual variability than the Atlantic, which suggests that the annual cycle may play a larger role and interannual variability a lesser role in the Atlantic than in the Pacific. Servain et al. (1999) derived indices to represent the equatorial pattern, dipole-structured pattern, and the movement of the ITCZ. By performing lagged correlations between the indices, they found that both patterns of variability are linked with the ITCZ on 1–2-yr and decadal time scales. Because the migration of the ITCZ in the tropical Atlantic exhibits a strong annual cycle, these results suggest a relationship between the Atlantic patterns of variability and the annual cycle. Nobre and Shukla (1996) found that anomalous atmospheric circulation patterns over the northern tropical Atlantic region are locked to the annual cycle, exhibiting variability only during a portion of the calendar year. Additionally, this study found that wet or dry years over the Nordeste region of Brazil are associated with late or early withdrawal of the seasonal ITCZ migration. Similarly, Hastenrath (1984) concluded that rainfall variability over the Central American/Caribbean, northeast Brazil, sub-Sahara Africa, and Angolan coast all appear as a reduction or enhancement of the annual cycle. All of these studies suggest an influence of the annual cycle on TAV as a potential driving or modulating mechanism.

Two feedback mechansims, which may be tied to the annual cycle, are well known in the tropics: (i) a Bjerknes-type feedback mechanism involving the zonal wind, thermocline, and SST and (ii) a wind–evaporation–sea surface temperature (WES) feedback mechanism. In particular, the WES feedback was introduced in association with the dipole-structured variability operating cross-equatorially in the tropical Atlantic (Chang et al. 1997; Xie 1999), while the Bjerknes feedback mechanism is related to the eastern Atlantic equatorial variability (Zebiak 1993). Seasonality is exhibited by these two mechanisms with the Bjerkenes feedback mechanism active in boreal spring and summer (Keenlyside and Latif 2007; Okumura and Xie 2006) with a secondary peak in November–December (Okumura and Xie 2006) and the WES mechanism active in boreal spring (Okajima et al. 2003; Chiang et al. 2002). This seasonality suggests that the feedback mechanisms may depend on the background seasonal state and therefore may only be able to influence TAV in certain seasons.

There are numerous avenues for seasonal remote forcing on the tropical Atlantic Ocean. The largest remote forcing is associated with the El Niño–Southern Oscillation (ENSO; Huang 2004), which can influence SST in all regions of the tropical Atlantic Ocean (Horel et al. 1986; Carton and Huang 1994; Latif and Barnett 1995; Nobre and Shukla 1996; Lau and Nath 1996; Enfield and Mayer 1997; Penland and Matrosova 1998; Latif and Grötzner 2000; Mo and Häkkinen 2001; Huang 2004) via the Pacific North American (PNA)/Pacific South American (PSA) patterns, among other atmospheric teleconnections. The regional influence of the North Atlantic Oscillation (NAO) is suggested to affect SST in the extratropical North Atlantic (Marshall et al. 2001; Visbeck et al. 1998; Chang et al. 2001) and eastern equatorial region (Czaja et al. 2002). Additionally, SST in the southern Atlantic Ocean is thought to be connected to the South Atlantic convergence zone (Robertson and Mechoso 2000) and the Antarctic Oscillation (AAO; Huang and Shukla 2005). Previous research revealed seasonal relationships between variability in the tropical Atlantic and the PNA and NAO (Visbeck et al. 1998; Chang et al. 2001; Czaja et al. 2002), the PSA pattern (Mo and Häkkinen 2001; Huang 2004), and ENSO (Hastenrath and Heller 1977; Hastenrath 1978; Hastenrath et al. 1987; Aceituno 1988; Hameed et al. 1993; Enfield 1996; Enfield and Mayer 1997; Horel et al. 1986; Carton and Huang 1994; Latif and Barnett 1995; Latif and Grötzner 2000; Huang 2004; Nobre and Shukla 1996; Barreiro et al. 2005).

Though it is quite clear from previous research that TAV itself as well as its forcing and feedback mechanisms exhibit seasonality, it is unclear whether the annual cycle is the necessary condition for the growth and development of TAV or if, in fact, it is the annual mean state of the system that is the key component.

Though the majority of the previously mentioned studies remove the annual cycle from their data (either observation or model output) to investigate the seasonality of interannual variability, the data still contain the influence of the annual cycle. The only way to truly remove the annual cycle influence is through a modeling study in which variability is generated from a state containing no annual cycle. Three modeling studies, Chang et al. (1997), Xie (1999), and Chang et al. (2001), in which there is no annual cycle present, find the dipole-structured pattern of variability present and related to the local WES feedback mechanism. However, these studies use either idealized or empirical atmosphere models and constrain their domain in latitude. In addition, these studies focus on the cross equatorial variability and do not include discussion of the zonal, equatorial pattern. An additional study by Xie et al. (1999) does find both patterns of variability, the zonal and meridional patterns, present in the absence of an annual cycle; however, the mean state in their model is zonally symmetric and symmetric about the equator.

We investigate TAV both in the presence and absence of an annual cycle within a fully coupled, global simulation. The experiment designed to test this idea is the perpetual annual mean simulation, in which the annual cycle was completely removed by perpetuating an annual mean state for the duration of the simulation. The only difference between this simulation and the control simulation is the absence of the annual cycle. If TAV does not depend on the annual cycle, it is expected that the patterns of variability and their associated feedback/forcing mechanisms will be unchanged from the control simulation. It will be shown that the annual cycle is not necessary for exciting the observed patterns of variability, and that once the anomalous SST pattern is generated, the local feedback mechanisms begin to operate despite the absence of an annual cycle. This is not to say that the seasons do not play a role in influencing which patterns are dominant in a particular season. Here we suggest that any perturbation to the mean state of the tropical easterly winds can bring about TAV. In the presence of an annual cycle, the seasonal fluctuations and forcings can provide those perturbations. Seasonal influence on TAV is investigated in Bates (2007) and a future paper.

The paper is organized as follows: Section 2 provides the author’s definition of TAV. Section 3 provides a description of the model and experiments. Section 4 contains the results of the study including descriptions of the similarity in spatial structure and time scales of variability (section 4a), ENSO and NAO influence (section 4b), and the coupled mechanisms and their relation to pattern evolution (section 4c). The paper is then concluded in section 5 with a discussion of the results.

2. Definition of TAV

REOFs of monthly mean SSTAs are used to define TAV in this study following Huang (2004). The patterns resulting from an REOF analysis of observed monthly mean SSTA from Smith and Reynolds (2003) are illustrated in Fig. 2 (left column). The leading pattern is named the southern tropical Atlantic (STA) pattern, which includes variability along the equator as well as the coastal region of Angola, the second leading pattern is called the northern tropical Atlantic (NTA) pattern, and the third is called the southern subtropical Atlantic (SSA) pattern.

Fig. 2.

The leading patterns from (left) an REOF and (right) an EOF analysis of monthly SST anomalies from observations (Reynolds and Smith 1994). The patterns have been normalized so that the associated PC time series has a variance of 1. The contour interval is 0.1°C, and values greater than 0.1°C and less than −0.1°C are shaded. The percentages in the top right of each panel is the percent of total variance within that data associated with that pattern.

Fig. 2.

The leading patterns from (left) an REOF and (right) an EOF analysis of monthly SST anomalies from observations (Reynolds and Smith 1994). The patterns have been normalized so that the associated PC time series has a variance of 1. The contour interval is 0.1°C, and values greater than 0.1°C and less than −0.1°C are shaded. The percentages in the top right of each panel is the percent of total variance within that data associated with that pattern.

REOFs are chosen specifically for this study for two reasons. First, the regions represented by the three leading REOFs highlight the areas in the tropical Atlantic containing the highest standard deviation of SST. Second, because rotation separates the variability spatially, it allows for the examination of each region’s variability and associated coupled mechanisms individually. This is an important distinction from regular EOFs in which the variability of interest is contained within the first two EOFs, which define the variability with an equatorial pattern (EOF1) and a meridional, dipole-structured pattern (EOF2) as illustrated in Fig. 2 (right column). This representation groups variability in the equatorial region with variability both in the North Atlantic in pattern 2 and in the southern subtropical Atlantic in pattern 1. Because each region of the tropical Atlantic is influenced by different forcings, both local and remote, it was important for this study to split the variability regionally to isolate influences for each region.

An REOF analysis of the SSTA from the control simulation (Fig. 3, left column) tends to split the STA pattern into two patterns: one capturing equatorial variability (not shown, REOF 4) and the other capturing the coastal Angola variability (middle panel). Additionally, this simulation tends to emphasize the coastal Angola variability over the equatorial variability since it appears as the second leading EOF pattern. Our experiment simulation, the perpetual annual mean simulation, does not exhibit this split, and displays a pattern more like observations (Fig. 3, right column). Analysis of the development and related mechanisms associated with the STA REOF from the control simulation revealed that it was only capturing the coastal Angola variability. To make sure we are comparing the same variability in both simulations and that we capture both regions involved with observed STA variability, we have chosen not to use the STA pattern and accompanying principal component (PC) time series, but rather to use the ATL3 index (SSTA averaged over 3°S–3°N, 20°W–0°). A plot of this index regressed onto the SST from the control and annual mean simulations (Fig. 4) reveals that using this index does capture the full region contained within the STA pattern from observations in both simulations. Therefore, all composite analyses discussed later in the paper use the NTA and SSA PC time series and ATL3 index.

Fig. 3.

The leading patterns from an REOF analysis of SST anomalies from (left) the control ACGCM and (right) the perpetual annual mean ACGCM. The REOF patterns have been normalized so that the associated PC time series has a variance of 1. The contour interval is 0.1°C, and values greater than 0.1°C and less than −0.1°C are shaded. The percentage in the top right of each panel is the percent of total variance within that simulation associated with that pattern.

Fig. 3.

The leading patterns from an REOF analysis of SST anomalies from (left) the control ACGCM and (right) the perpetual annual mean ACGCM. The REOF patterns have been normalized so that the associated PC time series has a variance of 1. The contour interval is 0.1°C, and values greater than 0.1°C and less than −0.1°C are shaded. The percentage in the top right of each panel is the percent of total variance within that simulation associated with that pattern.

Fig. 4.

Regression of the ATL3 index onto anomalous SST (contours), heat content (shading), and wind stress (vectors) for (left) the control simulation and (right) the perpetual annual mean simulation. Anomalous heat content is calculated as the integrated anomalous temperature over the upper 275 m and thus the units are °C m−1 (values greater than 0.2 shaded). The units for SSTA are °C, and the contour interval is 0.2, while the units for wind stress are N m−2.

Fig. 4.

Regression of the ATL3 index onto anomalous SST (contours), heat content (shading), and wind stress (vectors) for (left) the control simulation and (right) the perpetual annual mean simulation. Anomalous heat content is calculated as the integrated anomalous temperature over the upper 275 m and thus the units are °C m−1 (values greater than 0.2 shaded). The units for SSTA are °C, and the contour interval is 0.2, while the units for wind stress are N m−2.

Because the rotation maximizes spatial coherence, a dipole-structured pattern is not present in this representation. The WES feedback mechanism was originally proposed in relation to the cross-equatorial nature of this structure (Chang et al. 1997; Xie 1999); thus, it was necessary to determine if this type of feedback was present in the REOF representation. For this purpose, we again turn to Fig. 1, which illustrates the anomalous wind stress patterns associated with each REOF pattern. Wind stress patterns consistent with the WES mechanism (i.e., winds flowing from colder to warmer SSTA along the region of the highest SST gradient that are deflected by the Coriolis force such that they act to diminish the easterly trade winds in that hemisphere) are related to the NTA and SSA patterns of variability (Fig. 1, middle and bottom panels). The wind pattern related to the NTA pattern has a cross-equatorial influence, though it is weaker for the SSA pattern. This is most likely due to the fact that the SSA pattern is located farther from the equator than the NTA pattern. This is also an indication that a cross-equatorial SST gradient is not a requirement for this mechanism to operate. It appears here to operate on the local SST gradient. Because it may be operating only in one hemisphere, it may only act to enhance the local SST anomaly.

3. Model and experiments

a. Modeling strategy

Because this study involves an investigation of coupled processes and coupled feedback mechanisms, the model chosen is a coupled general circulation model (CGCM). The two components are the Poseidon ocean GCM (OGCM; Schopf and Loughe 1995) and the atmospheric GCM (AGCM) developed at the Center for Ocean–Land–Atmosphere Studies (COLA; Kinter et al. 1997). Since the purpose of this study is to investigate the influence of the mean annual cycle on the interannual variability, a prerequisite is that the model represents the mean state and annual cycle accurately. This is not satisfied by current CGCMs. Thus, the strategy of anomaly coupling was employed to avoid these errors.

The essence of anomaly coupling is that both components only exchange information about the departures from a prescribed annual cycle. Before a variable is passed to the companion model to be used as a boundary condition, it first has a “model” climatology subtracted, which is obtained from an uncoupled integration of that model. This creates an anomaly and has, in theory, removed that model’s bias. A desired climatology, here we use observations, is then added so that a full field is passed to the companion model. This modeling strategy has been proven in previous studies to be successful at reducing known fully coupled modeling errors (Kirtman et al. 2002) and is found in this study to capture TAV sufficiently (cf. Fig. 3, left column, with Fig. 2, left column). Not only is this process ideal for these studies by reducing tropical mean state and annual cycle errors, but it is also advantageous because it allows for the control of the annual cycle through the added climatology field. A detailed examination of anomaly and fully coupling strategies is described by Kirtman et al. (2002).

b. Control simulation

The purpose of the control simulation is to include the full annual cycle for comparison with various perturbed annual cycle experiments. Since the anomaly coupling strategy is employed, the experiment involves three separate model integrations. In addition to the anomaly coupled simulation, an atmosphere-only and ocean-only simulation must be conducted first to obtain the model climatologies used in the anomaly coupling procedure. The atmosphere-only simulation was previously conducted by Kirtman et al. (2002), and thus, the model climatologies for wind stress and surface fluxes were taken from that study. The forcing for that simulation was an observed SST annual cycle from Reynolds and Smith (1994).

Because the climatological fields of both heat and salinity fluxes contain large uncertainties, the “observed” climatologies for these variables are calculated from the output of the ocean-only simulations. Within the uncoupled ocean model, these surface fluxes are parameterized as relaxation terms so that SST and sea surface salinity (SSS) are damped to observations. Following the procedure used by Schopf and Loughe (1995), two ocean model iterations were performed. The first iteration uses a very strong damping coefficient on heat and salinity flux. This is done to create the fluxes that the ocean model needs to maintain the SST and SSS close to observations. Because the strong damping term is unrealistic, the model is run a second time, damping to a climatology from the first iteration and using a smaller damping coefficient. The time scale of damping on the heat flux, using 50 m for the mixed layer depth, is approximately 23 days in iteration 1 and approximately 115 days for iteration 2. The damping time scale on SSS, also using 50 m as the mixed layer, is approximately 45 days for iteration 1 and 230 days for iteration 2. Each iteration was integrated for 56 yr. Initial conditions for iteration 1 were obtained from an ocean spinup simulation forced with an annual cycle, while instantaneous states from the end of iteration 1 were used as initial conditions for iteration 2.

Since the observed heat flux is calculated within the ocean-only simulation, the atmospheric forcing for this simulation is a bit more involved. Ideally, we would like to force the ocean model with climatological atmospheric forcing; however, using that method, the resulting SST variability would be small. Because the heat flux is calculated using a damping term, this damping term would be underestimated and the resulting heat flux would not be realistic. Essentially, there would be no SSTA to damp, which in turn leads to an unrealistic heat flux. Therefore, we adopted the technique of adding interannual variability to the climatological wind stress field to obtain a more realistic heat flux climatology for use in the anomaly coupled simulation. The annual cycle, or climatology, is from da Silva et al. (1994) for wind stress, incoming solar radiation, and surface fluxes, and the interannually varying wind stress is taken from 56 yr (1948–2003) of NCEP–NCAR reanalysis data (Kalnay et al. 1996). This is consistent with the subsequent anomaly coupled simulation because we want there to be an annual cycle as well as interannual variability present.

Having obtained the model climatology for SST, the anomaly coupled control simulation could then be performed. Initial conditions for the anomaly coupled GCM (ACGCM) simulation were taken from the control atmosphere- and ocean-only simulations as were the model climatologies for the anomaly coupling procedure. The ocean model climatology was calculated from the last 35 yr of the ocean-only control simulation, while the atmosphere model climatology is the same as that used by Kirtman et al. (2002). Each variable is anomaly coupled in the same manner (see section 3a); however, as described above, iteration 2 of the ocean model is also used to calculate the observed climatologies for both heat and salinity flux. The observed climatologies for incoming solar radiation and wind stress are from da Silva et al. (1994). Observations used for the SST climatology are from Smith and Reynolds (2003). This simulation resulted in 250 yr of data for analysis.

c. Annual mean simulation

Unlike the control simulation, which was allowed to progress through the annual cycle, the annual mean simulation maintained a time-independent mean state throughout the simulation. To create this mean environment, the model was forced to integrate one particular day perpetually, thus maintaining a diurnal cycle while removing the seasonal change. The particular day was chosen by finding the day whose incoming solar radiation most closely resembled the annual mean solar radiation. This calculation used the NCEP–NCAR reanalysis top of the atmosphere incoming solar radiation data (Kalnay et al. 1996). The day chosen for the annual mean simulation is 22 September.

For the perpetual annual mean experiment we repeated the computations to generate the anomaly coupling, with the exception that each of the model simulations, including the two uncoupled simulations and the anomaly coupled simulation, perpetuates 22 September. Additionally, annual mean states are used for the boundary conditions in the uncoupled simulations and for the observed and model climatologies in the ACGCM rather than allowing progression through the annual cycle. Therefore, the 25-yr atmosphere-only simulation conducted for this experiment used an observed annual mean SST (Smith and Reynolds 2003) as its boundary condition. As in the control simulation, we needed to create a realistic heat flux climatology to be used as the observed climatology in anomaly coupling. Therefore, in iteration 2 of the ocean-only simulations, we added wind stress anomalies, taken from the corresponding atmosphere-only simulation (perpetual annual mean simulation), to be used as boundary conditions. As a result, the ocean-only iterations are each 25 yr in length. Again, this is consistent with the subsequent anomaly coupled simulation to be performed. In this simulation, we want to remove any influence of the annual cycle; therefore, we obtain the interannually varying wind stress from the atmosphere only simulation forced with climatological SST. This essentially yields atmospheric noise with no annual cycle. This creates the variability needed to generate a realistic heat flux climatology without influence from the annual cycle.

As in the control simulation, the model climatologies used in anomaly coupling are calculated from the component-only simulations. However, because the component-only simulations perpetuate an annual mean state, the climatology is simply a mean from each simulation, using the last 15 yr from the ocean simulations and the last 10 yr from the atmosphere simulations. Each model was spun up before these times; however, we felt that 15 yr of data from the ocean and 10 yr of data from the atmosphere were sufficient to create a representative model climatology. An annual mean of the observed climatologies used in the control simulation are used here as the observed climatology for the anomaly coupling procedure. Because the mean state of the system is not changing, the anomaly coupled simulation comes to equilibrium faster than the control simulation, and therefore was integrated for 100 yr.

d. Model mean state verification

Comparing the mean SST from the perpetual annual mean simulation with the annual mean from the control simulation (Fig. 5), we find that the annual mean simulation maintains a realistic mean state with a similar structure as in the control simulation. This is not a surprising result because the anomaly coupling is designed to keep the model from straying too far from the observed climatology.

Fig. 5.

Annual mean SST from (a) the control simulation and mean SST from the perpetual (b) annual mean simulation. SST above 26°C are shaded. The contour interval is 1°C. (c) Mean from the annual mean simulation minus the mean from the control simulation. The contour interval is 0.3°C. Differences with a magnitude larger than 0.5°C are shaded.

Fig. 5.

Annual mean SST from (a) the control simulation and mean SST from the perpetual (b) annual mean simulation. SST above 26°C are shaded. The contour interval is 1°C. (c) Mean from the annual mean simulation minus the mean from the control simulation. The contour interval is 0.3°C. Differences with a magnitude larger than 0.5°C are shaded.

The main difference between these two means is a larger zonal gradient of SST along the equator in the annual mean simulation. This difference, however, is consistent with having perpetuated a mean condition. The annual mean state consists of strong easterly trade winds throughout the tropics. By perpetuating this state, the equatorial winds are not relaxed seasonally, and thus more continuous upwelling can occur and therefore create a larger zonal SST gradient.

4. Results

a. Similarity in spatial structure and time scales of variability

Both a comparison of standard deviation of monthly mean SST (Fig. 6) and the three leading patterns resulting from a rotated EOF analysis of monthly mean SSTA (Fig. 3) in the control and perpetual annual mean simulations reveal that the regions containing the highest variability in the tropical Atlantic are basically the same in the presence or absence of an annual cycle. Though there are differences, in general, variability in each of the three regions of concern to TAV, the equatorial, northern tropical, and southern subtropical regions, is captured by both simulations.

Fig. 6.

Std dev of monthly mean SST from (a) the control simulation and (b) the perpetual annual mean simulation. Magnitudes above 0.5°C are shaded. The contour interval is 0.1°C. (c) Std dev from the annual mean simulation minus the std dev from the control simulation. The contour interval is 0.1°C.

Fig. 6.

Std dev of monthly mean SST from (a) the control simulation and (b) the perpetual annual mean simulation. Magnitudes above 0.5°C are shaded. The contour interval is 0.1°C. (c) Std dev from the annual mean simulation minus the std dev from the control simulation. The contour interval is 0.1°C.

The percentages in the top right of each panel in Fig. 3 indicate the amount of variance for which each pattern of variability accounts within that simulation. However, the control simulation contains more total variance in SSTA than in the annual mean simulation (total variance equals 0.18636 for the control simulation and 0.13370 for the annual mean simulation). When this difference is considered along with the percentages provided in the plot, we note that the total variance associated with each pattern is very similar between the two simulations.

Although few of the peaks in the power spectra of the PCs from either simulation are significant at the 90% level (Fig. 7), a comparison gives an idea of how the time scales of variability associated with each pattern have changed in the annual mean setting, although the differences noted here have not been subject to a significance test. From Fig. 7, it appears that perpetuating an annual mean state has, at the least, maintained the interannual variability observed in the control simulation, but has somewhat diminished longer time-scale variability. The main concern of this paper is what happens to the variability on interannual time scales with and without an annual cycle present, and thus, the changes in decadal variability are not addressed. The STA pattern variability in both simulations (top panels) contains a 3–4-yr period peak, while SSA variability is enhanced in the annual mean simulation in the 2–3-yr time scale, and NTA variability is enhanced in the 4–6-yr time range. To truly understand the changes in the periodicities of these time series, longer simulations and further analysis would have to be conducted. The only conclusion we draw for this study is that interannual variability exists in the absence of an annual cycle.

Fig. 7.

Spectral density of the PC time series associated with the three leading REOFs from SSTA output from (left) the control simulation and (right) the annual mean simulation. The top panels correspond to the STA pattern, the middle panels to the NTA pattern, and the bottom panels to the SSA pattern PC time series. The dashed line represents a red noise spectrum, the bold solid line corresponds to the time series spectrum, while the thin solid line indicates the lower limit of the 90% confidence interval as calculated by a chi-squared test.

Fig. 7.

Spectral density of the PC time series associated with the three leading REOFs from SSTA output from (left) the control simulation and (right) the annual mean simulation. The top panels correspond to the STA pattern, the middle panels to the NTA pattern, and the bottom panels to the SSA pattern PC time series. The dashed line represents a red noise spectrum, the bold solid line corresponds to the time series spectrum, while the thin solid line indicates the lower limit of the 90% confidence interval as calculated by a chi-squared test.

It is appropriate at this point to discuss whether the method of rotating the EOFs aided in indentifying the same three regions of enhanced variability since this method maximizes spatial coherence. We again refer to the plot of SST standard deviation resulting from each simulation (Fig. 6) to argue that the three regions of concern are, in fact, the regions of the tropical Atlantic containing the largest standard deviation. In addition, a composite analysis of indexes created to capture variability in each of the three regions was performed. These indexes are box averages of SSTA over the three regions of largest variability. Not only are the resulting patterns similar to those found in the REOF analysis, but the spectral density of each time series is remarkably similar to the spectral density of the PC time series from the REOF analysis (not shown), thus indicating that the patterns found in the REOF analysis are real and not forced by the analysis method.

To provide evidence that the choice of the annual mean state was not fortuitous in generating TAV, results from another perpetual mean simulation are presented here. This time, a boreal spring mean condition was perpetuated. This provides an independent simulation in which there is no annual cycle while perpetuating a different mean state. The leading three REOFs of SSTA from this simulation are presented in Fig. 8. Note that, as with the annual mean simulation, this simulation also produced the three patterns of variability associated with TAV.

Fig. 8.

The leading patterns from an REOF analysis on SST anomalies from an ACGCM in which a boreal spring mean (MAM) condition was perpetuated. The REOFs have been normalized so that the associated PC time series has a variance of 1. The contour interval is 0.1°C, and values greater than 0.1°C and less than −0.1°C are shaded.

Fig. 8.

The leading patterns from an REOF analysis on SST anomalies from an ACGCM in which a boreal spring mean (MAM) condition was perpetuated. The REOFs have been normalized so that the associated PC time series has a variance of 1. The contour interval is 0.1°C, and values greater than 0.1°C and less than −0.1°C are shaded.

b. ENSO and NAO influence

ENSO and the NAO have remote influence on TAV, and we may have altered their influence within this framework. As noted in the introduction, both of these mechanisms exhibit seasonal preferences that may be affected by the removal of the annual cycle. First, we note that the control simulation does capture the seasonality of both mechanisms sufficiently. Both observations (Kalnay et al. 1996) and the control simulation exhibit the most variability associated with ENSO in the December–February (DJF) and September–November (SON) seasons; however, the control simulation favors SON while observations favor DJF. The NAO in both observations (Jones et al. 1997) and the control simulation exhibits the most variability in the DJF season. Second to DJF, the observations favor variability almost equally in March–May (MAM) and SON while the control simulation clearly favors MAM. For the purposes of this study, we view these two forcings as part of the annual cycle forcing, and therefore, any difference in their variability within the annual mean simulation is viewed as being a part of the altered state that forces TAV in the absence of an annual cycle. We expect that the variability associated with ENSO and the NAO should change and are not concerned with any discrepancy between the two simulations.

Nevertheless, for completeness, we present here how these two forcing mechanisms have been altered in the absence of an annual cycle. To investigate, we use EOF and spectral analysis on Pacific SSTA and North Atlantic sea level pressure anomalies (SLPA). A feature of the control simulation is that it tends to mix higher latitude, decadal variability in the northern equatorial Pacific (similar to the Pacific decadal oscillation) with equatorial, ENSO variability. For this reason, we use rotated EOFs for the Pacific SSTA analysis to isolate ENSO variability, while we are able to pick out the NAO from a regular EOF analysis of SLPA in the North Atlantic. This particular coupled model contains a weaker ENSO in both variance explained and magnitude than other models and observations; therefore, its influence over the Atlantic Ocean may be inherently weaker. Nonetheless, the ENSO and NAO patterns are the leading patterns of variability in both simulations (Figs. 9 and 10). A spectral analysis of the corresponding principal component time series reveals that both mechanisms do retain interannual time scales of variability. With the exception of longer time-scale variability being absent from the annual mean simulation, the reproduction of ENSO variability on interannual time scales is quite remarkable. It is obvious from Fig. 10 that the time scales of variability associated with the NAO are changed in the presence and absence of an annual cycle, but this is to be expected since the NAO has been shown to be pronounced in the November–February (NDJF) season and the annual mean simulation is similar in construction to a boreal fall simulation. Because the interannual variability is still present in the perpetual annual mean simulation and explains a similar amount of total variance in the system, we believe that NAO fluctuations may still influence TAV.

Fig. 9.

(left) The leading rotated EOF pattern and (right) spectral density of the corresponding principal component time series on SSTA data from the control simulation (top) and the perpetual annual mean simulation (bottom). The contour interval in the top left panel is 0.1° and 0.2°C in the bottom left panel. In the right panels, the dashed line represents a red noise spectrum, the bold solid line corresponds to the time series spectrum, while the thin dotted line indicates the lower limit of the 90% confidence interval as calculated by a chi-squared test. The percentages in the bottom left of the left panels indicates the amount of SSTA variance for which each pattern accounts within its respective simulation.

Fig. 9.

(left) The leading rotated EOF pattern and (right) spectral density of the corresponding principal component time series on SSTA data from the control simulation (top) and the perpetual annual mean simulation (bottom). The contour interval in the top left panel is 0.1° and 0.2°C in the bottom left panel. In the right panels, the dashed line represents a red noise spectrum, the bold solid line corresponds to the time series spectrum, while the thin dotted line indicates the lower limit of the 90% confidence interval as calculated by a chi-squared test. The percentages in the bottom left of the left panels indicates the amount of SSTA variance for which each pattern accounts within its respective simulation.

Fig. 10.

(left) The leading rotated EOF pattern and (right) spectral density of the corresponding PC time series on SLPA data from the control simulation (top) and the perpetual annual mean simulation (bottom). The contour interval in the left panels is 2 hPa. In the right panels, the dashed line represents a red noise spectrum, the bold solid line corresponds to the time series spectrum, while the thin dotted line indicates the lower limit of the 90% confidence interval as calculated by a chi-squared test. The percentages in the bottom right of the left panels indicate the amount of SLPA variance for which each pattern accounts within its respective simulation.

Fig. 10.

(left) The leading rotated EOF pattern and (right) spectral density of the corresponding PC time series on SLPA data from the control simulation (top) and the perpetual annual mean simulation (bottom). The contour interval in the left panels is 2 hPa. In the right panels, the dashed line represents a red noise spectrum, the bold solid line corresponds to the time series spectrum, while the thin dotted line indicates the lower limit of the 90% confidence interval as calculated by a chi-squared test. The percentages in the bottom right of the left panels indicate the amount of SLPA variance for which each pattern accounts within its respective simulation.

c. Coupled mechanisms and pattern evolution

This section will focus on the characteristics of the patterns, their evolution, and associated local coupled mechanisms in the presence and absence of an annual cycle. To infer these relationships, a composite analysis of various variables including anomalous SST, wind stress, heat flux, and sea level pressure is performed on the ATL3 index and NTA and SSA principal component time series. Warm and cold events are defined in each time series as events whose amplitude was larger than 1.5 times the standard deviation of the time series, with the warm events having positive sign and the cold events negative sign. To ensure that the same event was not considered more than once, the event had to be the largest within a 2-yr range. An average of the desired variable was made for all warm and all cold events, and the composite is displayed as the warm minus the cold averages. This composite is considered the peak month in the composite since it corresponds to the peak of the event. A composite is also calculated for 12 months before the peak to 12 months after the peak so that the evolution of all fields leading up to and following the peak in SSTA can be investigated. The three time series contain a similar number of peak events in both the control and annual mean simulation with anywhere from 50 to 80 events averaged for each composite. Significance of the difference (warm minus cold) is calculated using Student’s t test for each composite and is discussed below as appropriate.

1) Equatorial variability

Even in the absence of an annual cycle, the variability present in the eastern equatorial region of the Atlantic Ocean retains its most prominent characteristics. The peak phase in both simulations is characterized by warming in the equatorial and coastal Angola regions. In both simulations, variability in the equatorial region is related to fluctuations of the equatorial trade winds while variability in the coastal Angola region is related to fluctuations in the southeast trade winds. Two separate processes are involved with these two wind fluctuations.

The first of these processes has the characteristics of the Bjerknes mechanism, which has previously been associated with equatorial variability (Zebiak 1993). Figure 11 illustrates that a warming in the eastern equatorial region is preceded by a relaxation of the easterly equatorial trade winds in the mid- to western portion of the basin. The anomalous wind pattern appears significantly in both simulations approximately 2 months preceding the peak phase of SSTA and peaks approximately 1 month prior to the SSTA peak. Figure 4 shows this relationship as well as the increase in heat content, which peaks simultaneously with the SSTA peak, in the eastern equatorial basin. This relationship with zonal wind and heat content are consistent with the Bjerknes mechanism.

Fig. 11.

Composite values of anomalous zonal wind stress (contours) and SSTA (shading) from the ATL3 index composite analysis along the equator for (left) the control simulation and (right) the perpetual annual mean simulation. Values of SSTA greater than 0.1°C are shaded. The contour interval for wind stress is 0.004 N m−2. The y axis values indicate the month relative to the peak month in the composite.

Fig. 11.

Composite values of anomalous zonal wind stress (contours) and SSTA (shading) from the ATL3 index composite analysis along the equator for (left) the control simulation and (right) the perpetual annual mean simulation. Values of SSTA greater than 0.1°C are shaded. The contour interval for wind stress is 0.004 N m−2. The y axis values indicate the month relative to the peak month in the composite.

The second process, first noted in Huang and Shukla (2005), though not in its relationship with equatorial variability, begins as a weakening of the subtropical high in the South Atlantic basin leading to anomalous cyclonic flow around the anomalous low pressure. These anomalous sea level pressure and wind patterns lead the peak in SSTA by approximately 4 months in each simulation and migrate northward. Equatorward migration of SST anomalies due to the WES mechanism has been found in Xie (1999), though he also noted that the migration is stunted by Ekman transport. However, northward migration of this feature can also occur without the WES feedback as follows: consider a warm anomaly as an example. The wind anomaly on the northern edge diminishes the trade winds thus reducing evaporation causing an anomalous flux of heat into the ocean in this region. Consequently, the anomalous warm SST shifts northward. The northern edge of the feature, containing anomalous westerly winds, acts to diminish the trade winds, thus causing a warming in the eastern basin. It first affects the southeasterly trade winds bringing about variability along coastal Angola and as it moves farther north, can influence equatorial winds causing equatorial variability in SST.

Figure 12 illustrates the northward migration of the SLP and wind stress anomalies and their influence on SSTA. In both simulations, an anomalous low pressure forms in the southern subtropical Atlantic around 40°S 4 to 5 months before the peak in SSTA. The migration of this feature is clearly seen in the control simulation. Note the warming response of the SSTA is on the northern flank of the feature as is expected, and that as the feature nears the equator it begins to affect SSTA in the coastal Angola and equatorial regions. The feature in the annual mean simulation appears simultaneously with a relaxation of easterly winds along the equator (the first process described above), and therefore, the northward migration is less clear. The feature, however, does extend northward and begins to affect easterly winds south of the equator in the coastal Angola region, but it is not as easy to see its individual influence on equatorial SSTA since it is appearing simultaneously with the first process.

Fig. 12.

Composite values of anomalous SLP (contours) and wind stress (vectors) from the composite analysis of the ATL3 index. (top) The average of the fourth and fifth composite months preceding the peak, (middle) the average of the second and third months preceding the peak, and (bottom) the average of the peak month and the month before the peak. The shading indicates where SSTA values are larger than 0.2°C. SLP is in units of hPa with a contour interval of 0.5 and negative contours are dashed.

Fig. 12.

Composite values of anomalous SLP (contours) and wind stress (vectors) from the composite analysis of the ATL3 index. (top) The average of the fourth and fifth composite months preceding the peak, (middle) the average of the second and third months preceding the peak, and (bottom) the average of the peak month and the month before the peak. The shading indicates where SSTA values are larger than 0.2°C. SLP is in units of hPa with a contour interval of 0.5 and negative contours are dashed.

The first mechanism, corresponding to the Bjerknes-type feedback, mostly appears simultaneously with the second mechanism, and the response of SST cannot be sufficiently separated between the two mechanisms. However, the second mechanism is noted to occur independently of the first and in some cases may lead to the first mechanism by influencing the equatorial trade winds. The life cycle of an anomalous SST event is prolonged by approximately 3 months in the annual mean simulation, as seen from the autocorrelation of the ATL3 index in Fig. 13. It begins 2 months earlier and ends approximately 1 month later in the composite than the warming in the control simulation. We attribute this to prolonged weakened easterlies, which begin before and continue longer than those in the control simulation (Fig. 11).

Fig. 13.

Autocorrelation for the (top) ATL3 index and the PC time series for the (middle) NTA and (bottom) SSA REOF patterns. The bold line represents observations, the long dashed line is the control simulation, and the dotted–dashed line is the perpetual annual mean simulation.

Fig. 13.

Autocorrelation for the (top) ATL3 index and the PC time series for the (middle) NTA and (bottom) SSA REOF patterns. The bold line represents observations, the long dashed line is the control simulation, and the dotted–dashed line is the perpetual annual mean simulation.

2) North tropical Atlantic variability

Variability associated with the NTA pattern in both simulations is associated with a fluctuation of the northeast trade winds. As shown in Fig. 14, the peak of a warming event in the region of the NTA pattern follows, by approximately 1 month, a peak in anomalous westerly wind stress in both simulations. These anomalous winds appear significantly approximately 7 (control) and 10 (perpetual annual mean) months prior to the peak in SSTA and either simultaneous (control) or 2 months prior (perpetual annual mean) to a significant change in SST. They lead to a warming in SST by diminishing the easterly trade winds, thus reducing evaporation and increasing heat flux into the ocean. In addition, weakened easterlies may act to decrease the wind-induced upwelling along the coast. This anomalous wind structure occurs farther to the north in the annual mean simulation resulting in an SST anomaly centered farther to the north than that in the control simulation (see Fig. 3). The anomalous SSTA does migrate equatorward, which results in the appearance of a weakening of the wind stress shortly after the peak in SSTA (Fig. 14). Even though this migration is occurring, we can still see a difference in the duration of the weakened easterlies between the two simulations. In the control simulation, the wind stress changes sign after the peak and does not show any more activity for the duration of the composite months. In contrast, the wind stress continues to indicate anomalous westerlies in the mid- to western Atlantic throughout the annual mean simulation composite.

Fig. 14.

Composite values of anomalous zonal wind stress (contours) and SST (shading) along latitude 15°N for each composited month from the NTA PC composite analysis. The y axis values indicate the month relative to the peak month in the composite. The units for wind stress are N m−2 and the contour interval is 0.01 for the control simulation and 0.05 for the annual mean simulation. SSTA values are in units of °C.

Fig. 14.

Composite values of anomalous zonal wind stress (contours) and SST (shading) along latitude 15°N for each composited month from the NTA PC composite analysis. The y axis values indicate the month relative to the peak month in the composite. The units for wind stress are N m−2 and the contour interval is 0.01 for the control simulation and 0.05 for the annual mean simulation. SSTA values are in units of °C.

The wind pattern consistent with the WES mechanism are winds flowing from cooler to warmer regions being deflected by the Coriolis force such that they diminish the trade winds in the warm region further enhancing the warm anomaly. In the case of a warming in the NTA pattern region, this results in cross-equatorial winds flowing from south to north. We therefore use meridional wind stress at 5°N as an indication of the presence and strength of the WES mechanism (Fig. 15). In the control simulation an increase in meridional wind stress (indicating northward flowing wind) appears approximately 3–5 months after significant warming in SST begins but 2–3 months before the peak phase, indicating that the winds may not be the excitation mechanism for the SST variability, but the WES mechanism may act to sustain or enhance the SST anomaly. However, in the annual mean simulation, this picture is not as clear; however, it is clear that the SST and wind stress anomalies persist much longer than in the control simulation.

Fig. 15.

Composite values of anomalous meridional wind stress (contours) at 5°N and SSTA (shading) at 15°N for each composited month from the NTA PC composite analysis. (left) The results from the control simulation and (right) the perpetual annual mean simulation. The y axis values indicate the month relative to the peak month in the composite. The units for wind stress are N m−2 and the contour interval is 0.002 for the control simulation and 0.001 for the annual mean simulation, while the SSTA is in units of °C with a contour interval of 0.1.

Fig. 15.

Composite values of anomalous meridional wind stress (contours) at 5°N and SSTA (shading) at 15°N for each composited month from the NTA PC composite analysis. (left) The results from the control simulation and (right) the perpetual annual mean simulation. The y axis values indicate the month relative to the peak month in the composite. The units for wind stress are N m−2 and the contour interval is 0.002 for the control simulation and 0.001 for the annual mean simulation, while the SSTA is in units of °C with a contour interval of 0.1.

Though the developing and peak phases occur at similar times in the composite, the length of the decay phase is very different. In the control simulation, the anomalous SST decays 5–6 months after the peak, while in the perpetual annual mean simulation, this anomaly persists beyond 12 months after the peak, as illustrated in Fig. 15 in which significant SST anomalies in the perpetual annual mean simulation are shown to persist longer than those in the control simulation. This difference in the length of the life cycle is also illustrated in the plot of the NTA time series autocorrelation (Fig. 13, middle panel). The perpetual annual mean line (dotted–dashed line) does not cross the first e-folding point, approximately 0.4, until approximately 18 months while the control line (long dashed line) crosses around 7 months. The difference is thought to be due to the presence of anomalous wind stress consistent with the WES feedback mechanism. From Figs. 14 and 15, we deduce that anomalous zonal (at the site of the warming) and meridional (at 5°N) wind stress peak either before or simultaneously with the peak in SST warming in the control simulation and quickly diminish after the peak, while in the perpetual annual mean simulation both continue to be present after the peak, and in fact, the meridional wind stress peaks after the peak in SSTA. It appears in Fig. 14 that the anomalous zonal wind stress in the annual mean simulation disappears in the second month after the peak month; however, the feature is migrating southward and, as mentioned above, looking at the data along 15°N with time will falsely indicate the end of this feature’s life cycle. From these figures plus inspection of composite maps of the wind stress field (not shown), we conclude that wind stress consistent with the WES mechanism no longer exists in the control simulation the month after the peak in SSTA at which point the entire structure of the wind changes and the decay phase of SSTA begins. In the annual mean simulation, however, wind stress consistent with the WES mechanism persists beyond 12 months after the peak and may be the reason for the persistence of the SST anomaly.

In the annual mean simulation, variability in the region of the STA pattern appears 1 month after the peak phase of the NTA pattern; however, this does not occur in the control simulation. We believe this is due to the presence of the WES mechanism in the annual mean simulation. This feature acts cross-equatorially in relation to the NTA pattern, as described in section 1, with anomalous winds flowing south to north. South of the equator these winds are deflected by the Coriolis force such that they enhance the trade winds in that region. The feedback to the ocean due to enhanced evaporation is to cause anomalously cold SST, which is what is observed. Because the feature persists so much longer in the annual mean simulation, it appears to be able to influence the SST south of the equator as a result of variability north of the equator. This indicates that there is communication between the two hemispheres; however, in reality, it may not be observed because the progression of the annual cycle cuts the interaction off before it has a significant impact.

Both simulations support the conclusion that SSTA in the NTA region is related to a weakening of the northeast trade winds and that the wind stress anomalies, consistent with the WES mechanism, act to enhance and maintain the variability. The persistence these anomalous wind stress patterns also appears to influence SST south of the equator with opposite sign SSTA occurring after the appearance of the anomalous cross-equatorial wind.

3) South tropical Atlantic variability

In both simulations, the two peak phases of positive SSTA related to SSA variability are preceded by a decrease in the southern subtropical high, which is associated with cyclonic flow in the wind field (Fig. 16). Both the anomalous winds and sea level pressure peak 1–2 months prior to the peak in anomalous SST in both of the simulations, providing evidence that these changes induce the SST anomaly. The anomalous sea level pressure migrates northward during the development phase in the same manner as discussed above with the pattern associated with the equatorial variability. The warm anomalies are coincident with the northern side of this flow, which corresponds with the region of the southeast trade winds. This occurs because the anomalous westerly winds are diminishing the southeast trade winds, therefore decreasing evaporation subsequently creating a warm anomaly. After the peak, the anomalous sea level pressure continues to migrate north until it reaches the equatorial region along the coast of northeast Brazil. As it moves north, it begins to influence the equatorial trade winds causing anomalous westerlies along the equator. It is at this time in both simulations that the SSTA anomaly in the equatorial region begins to appear along with the anomalies in the southern subtropical region.

Fig. 16.

Composite maps of anomalous SLP (contours, units = hPa), wind stress (vectors, units = N m−2), and SSTA (shading, units = °C) from the SSA PC time series composite analysis for (left) the control simulation and (right) the perpetual annual mean simulation. The top panels are an average of the third and second months preceding the peak month, the middle panels are the average of the month prior and the peak month, while the bottom panels are the average of the two months following the peak month. The contour interval for SLP is 0.5, and SSTA values greater than 0.2 are shaded.

Fig. 16.

Composite maps of anomalous SLP (contours, units = hPa), wind stress (vectors, units = N m−2), and SSTA (shading, units = °C) from the SSA PC time series composite analysis for (left) the control simulation and (right) the perpetual annual mean simulation. The top panels are an average of the third and second months preceding the peak month, the middle panels are the average of the month prior and the peak month, while the bottom panels are the average of the two months following the peak month. The contour interval for SLP is 0.5, and SSTA values greater than 0.2 are shaded.

Though they peak in the same composite month, the anomalies of sea level pressure and wind remain present and significant for 2 months past the peak in SSTA in the perpetual annual mean simulation, whereas they decay in the control simulation after the peak month. This may contribute to the SST anomaly persisting through 6 months after the peak in the perpetual annual mean simulation while it decays at 2 months after the peak in the control simulation. This increase in the length of the life cycle is illustrated in Fig. 13 (bottom panel). Comparing the autocorrelation of the SSA time series in the control simulation (long dashed line) with the perpetual annual mean simulation (dotted–dashed line), we note that the perpetual mean simulation anomaly persists a few months longer than in the control simulation. With the exception of an extended decay phase in the perpetual annual mean simulation, each phase in the life cycle of the SSA pattern shares many characteristics with the pattern in the control simulation.

5. Discussion

A series of model simulations have been conducted to assess the influence of the annual cycle on tropical Atlantic variability. These experiments were designed specifically to address whether the annual cycle is necessary for the generation of the patterns of variability and their development. An anomaly coupled version of the COLA atmospheric GCM coupled to the Poseidon ocean GCM was developed expressly for this investigation. Anomaly coupling was chosen to minimize the errors fully coupled models have in representing the tropical annual mean state and annual cycle. This model was successful in improving the representation of the mean state of SST as well as in removing the erroneous semiannual signal in the equatorial SST annual cycle while maintaining a good simulation of the three patterns of variability involved in TAV.

By removing the annual cycle, the perpetual annual mean simulation created a unique way to look at the influence of the annual cycle on the three patterns of variability. Examination of standard deviation and a rotated EOF analysis of SST anomalies revealed that the interannual variability resulting from the annual mean simulation occurs in the same regions as in the control simulation. All three patterns of variability are identified in the REOF analysis and are found to have very similar characteristics to those resulting from the control simulation. In addition to the similarity in the location and appearance of the patterns, the local feedback mechanisms, wind–evaporation–SST (WES) and the equatorial Bjerknes-type feedback, were associated with their respective patterns in agreement with the control simulation. The implication of this result is that the structure of the variability is not related to the annual cycle. Furthermore, once the anomalous SST patterns are generated, the same local feedback mechanisms are responsible for maintaining the anomaly even in the absence of the annual cycle and therefore are not dependent on a particular season to be active.

Because the generation of the three patterns of variability is related to fluctuations in the trade winds, we conclude that it is the presence of these winds in the mean state, which when perturbed can excite these patterns of variability. In the control simulation, seasonal variations modulate the dominant atmospheric forcing mechanisms, whereas in the absence of the annual cycle, these seasonal changes are not occurring. It is speculated that stochastic atmospheric forcing alone (no particular pattern is needed) could alter the easterly winds, and therefore, all that is necessary for these three patterns to appear is the presence of tropical easterly winds. Tropical easterly winds occur in the annual mean, and therefore, it is conjectured that in this study the mean state is the key factor to the occurrence of TAV.

One role of the annual cycle uncovered in these experiments is that it may act to shorten the life cycle of these events. Without a change of season in the perpetual annual mean simulation, the ATL3-related variability appeared earlier and persisted longer than in the control simulation, while both the NTA and SSA patterns persisted longer past the peak phase. Having a mean state that does not change seasonally may allow for more prolonged interactions between the atmosphere and ocean, which continually reinforce the SST anomaly. For example, in relation to the NTA pattern of variability, the lengthening of the pattern’s existence appeared to be due to the persistent wind stress patterns consistent with the local WES feedback mechanism. Our results suggest that the annual cycle cuts off this air–sea interaction. It is likely that the seasonal migration of the ITCZ alters the wind pattern such that the WES mechanism can only operate for part of the year. Similarly, we find that prolonged ATL3 variability is related to a prolonged weakening of the equatorial winds, which is related to the Bjerknes feedback mechanism. Though the variability related to the SSA pattern is not associated with a coupled feedback mechanism for its prolonged life cycle, we believe that the unchanging forcing field allowed it to persist uninterrupted. We therefore conclude that the annual cycle acts to disrupt the variability associated with each of these three regions and thus puts a limit on the length of their life cycle.

Additionally, the annual cycle may act to cut off influences between the patterns of variability themselves, specifically those relying on cross-equatorial air–sea interaction. Because the wind stress patterns consistent with the WES mechanism operate for a longer period in the annual mean simulation, a relationship between the STA and NTA patterns is revealed. Specifically, variability in the north tropical Atlantic can lead to variability in the region of the STA pattern through the influence of anomalous cross-equatorial winds. Likewise, in the spring mean simulation, the NTA pattern can develop following the presence of the STA pattern, again as a result of the anomalous winds operating cross-equatorially. While the cross-equatorial relationship is present in the control simulation, the annual cycle inhibits the teleconnection such that variability in the opposite hemisphere does not develop.

Present in both simulations is a relationship between the ATL3 index variability and SSA pattern variability. In both the ATL3 and SSA pattern composites, variability in the southern subtropical Atlantic can lead to equatorial and coastal Angola variability. A warming accompanied by cyclonic flow in the wind field begins in the southern subtropical region and migrates northward, consistent with the presence of the WES mechanism, in both simulations. Depending on the geographic position of the feature, it can influence southeast trade winds leading to variability along the coast of Angola, or if positioned farther north, it alters the equatorial trade winds causing variability in the eastern equatorial region. This relationship is not changed by the annual cycle. The WES mechanism associated with the SSA variability, as described in section 4c acts locally on the SST gradient and does not involve cross-equatorial flow. As a result, the seasonal migration of the ITCZ, which interrupts WES feedback in relation to NTA variability, is not a factor in the southern subtropical Atlantic and the migration of the feature northward remains undisturbed.

Though interannual variability remained present in the annual mean simulation, longer time-scale variability associated with TAV, ENSO, and the NAO was not as pronounced or was completely removed from the system. This raises the question of the role of the annual cycle in this decadal and longer time-scale variability for all of these patterns. Additionally, we found that ENSO retained its interannual variability even in the absence of an annual cycle suggesting that it too does not depend on seasonal variations or the progression through the annual cycle for its existence. This is in contrast to its apparent phase locking with the annual cycle. The time scales of variability related to the NAO, however, were greatly changed in the absence of an annual cycle (though interannual variability still existed), which suggests its dependence on seasonal variations for its oscillation periods. The questions surrounding these relationships are currently unresolved.

The question addressed in this paper is whether the existence of the annual cycle is a necessary condition for the variability observed in the tropical Atlantic Ocean. Based on the evidence presented, we conclude that the annual cycle is not a necessary component for TAV. In the absence of an annual cycle, the three patterns of variability associated with TAV are still present and are related to similar feedback mechanisms as they are in the presence of an annual cycle. Both an annual mean and a boreal spring mean state were perpetuated and support this statement, therefore providing evidence that the choice of the mean state to perpetuate was not simply fortuitous. This does not suggest that the annual cycle does not play a role in modulating the SSTA variability. In fact, we suggest that the position of the seasonal ITCZ could cause the fluctuations of the easterly trade winds in the mean state therefore tying TAV to particular seasons. This and other seasonal influences on TAV are discussed in a future paper.

Acknowledgments

This research was supported by grants from the National Science Foundation (ATM0332910), the National Oceanic and Atmospheric Administration (NA04OAR4310034), and the National Aeronautics and Space Administration (NNG04GG46G). I would like to thank J. Shukla and B. Huang for their support and guidance throughout this project. I am grateful to B. Kirtman for his assistance in model development and his constructive comments. Thanks also to J. Manganello and two anonymous reviewers for their constructive reviews.

REFERENCES

REFERENCES
Aceituno
,
P.
,
1988
:
On the functioning of the Southern Oscillation in the South American sector. Part I: Surface climate.
Mon. Wea. Rev.
,
116
,
505
524
.
Barreiro
,
M.
,
P.
Chang
,
L.
Ji
,
R.
Saravanan
, and
A.
Giannini
,
2005
:
Dynamical elements of predicting boreal spring tropical Atlantic sea-surface temperatures.
Dyn. Atmos. Oceans
,
39
,
61
85
.
Bates
,
S. C.
,
2007
:
Seasonal influences on coupled ocean-atmosphere variability in the tropical Atlantic Ocean. COLA Tech. Rep. 254, 39 pp. [Available from COLA, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.]
.
Carton
,
J. A.
, and
B.
Huang
,
1994
:
Warm events in the tropical Atlantic.
J. Phys. Oceanogr.
,
24
,
888
903
.
Chang
,
P.
,
L.
Ji
, and
H.
Li
,
1997
:
A decadal climate variation in the tropical Atlantic Ocean from thermodynamic air-sea interactions.
Nature
,
385
,
516
518
.
Chang
,
P.
,
L.
Ji
, and
R.
Saravanan
,
2001
:
A hybrid coupled model study of tropical Atlantic variability.
J. Climate
,
14
,
361
390
.
Chiang
,
J. C. H.
, and
D. J.
Vimont
,
2004
:
Analogous Pacific and Atlantic meridional modes of tropical atmosphere–ocean variability.
J. Climate
,
17
,
4143
4158
.
Chiang
,
J. C. H.
,
Y.
Kushnir
, and
A.
Giannini
,
2002
:
Deconstructing Atlantic Intertropical Convergence Zone variability: Influence of the local cross-equatorial sea surface temperature gradient and remote forcing from the eastern equatorial Pacific.
J. Geophys. Res.
,
107
.
4004, doi:10.1029/2000JD000307
.
Czaja
,
A.
,
P. V. D.
Vaart
, and
J.
Marshall
,
2002
:
Variations in low-latitude circulation and extreme climatic events in the tropical Americas.
J. Climate
,
15
,
3280
3290
.
da Silva
,
A.
,
A. C.
Young
, and
S.
Levitus
,
1994
:
Algorithms and Procedures.
Vol. 1, Atlas of Surface Marine Data 1994, NOAA Atlas NESDIS 6, 83 pp
.
Enfield
,
D. B.
,
1996
:
Relationships of inter-American rainfall to tropical Atlantic and Pacific SST variability.
Geophys. Res. Lett.
,
23
,
3305
3308
.
Enfield
,
D. B.
, and
D. A.
Mayer
,
1997
:
Tropical Atlantic sea surface temperature variability and its relation to El Niño-Southern Oscillation.
J. Geophys. Res.
,
102
,
929
945
.
Hameed
,
S.
,
A.
Meinster
, and
K. R.
Sperber
,
1993
:
Teleconnections of the Southern Oscillation in the tropical Atlantic sector in the OSU coupled upper ocean-atmosphere GCM.
J. Climate
,
6
,
487
498
.
Hastenrath
,
S.
,
1978
:
On modes of tropical circulation and climate anomalies.
J. Atmos. Sci.
,
35
,
2222
2231
.
Hastenrath
,
S.
,
1984
:
Interannual variability and annual cycle: Mechanisms of circulation and climate in the tropical Atlantic sector.
Mon. Wea. Rev.
,
112
,
1097
1107
.
Hastenrath
,
S.
, and
L.
Heller
,
1977
:
Dynamics of climatic hazards in Northeast Brazil.
Quart. J. Roy. Meteor. Soc.
,
103
,
77
92
.
Hastenrath
,
S.
,
L. C.
Castro
, and
P.
Acietuno
,
1987
:
The Southern Oscillation in the tropical Atlantic sector.
Contrib. Atmos. Phys.
,
60
,
447
463
.
Horel
,
J. D.
,
V. E.
Kousky
, and
M. T.
Kagaro
,
1986
:
Atmospheric conditions in the tropical Atlantic during 1983–1984.
Nature
,
322
,
243
245
.
Huang
,
B.
,
2004
:
Remotely forced variability in the tropical Atlantic Ocean.
Climate Dyn.
,
23
,
133
152
.
Huang
,
B.
, and
J.
Shukla
,
2005
:
Ocean–atmosphere interactions in the tropical and subtropical Atlantic Ocean.
J. Climate
,
18
,
1652
1672
.
Jones
,
P.
,
T.
Jonsson
, and
D.
Wheeler
,
1997
:
Extension to the North Atlantic Oscillation using early instrumental pressure observations from Gibraltar and South-West Iceland.
Int. J. Climatol.
,
17
,
1433
1450
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Keenlyside
,
N. S.
, and
M.
Latif
,
2007
:
Understanding equatorial Atlantic interannual variability.
J. Climate
,
20
,
131
142
.
Kintner
,
J. L.
, and
Coauthors
,
1997
:
The COLA atmosphere–biosphere general circulation model. Volume 1: Formulation. COLA Tech. Rep. 51, 46 pp. [Available from COLA, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705.]
.
Kirtman
,
B. P.
,
Y.
Fan
, and
E. K.
Schneider
,
2002
:
The COLA global coupled and anomaly coupled ocean–atmosphere GCM.
J. Climate
,
15
,
2301
2320
.
Latif
,
M.
, and
T. P.
Barnett
,
1995
:
Interactions of the tropical oceans.
J. Climate
,
8
,
952
968
.
Latif
,
M.
, and
A.
Grötzner
,
2000
:
The equatorial Atlantic oscillation and its response to ENSO.
Climate Dyn.
,
16
,
213
218
.
Lau
,
N-C.
, and
M. J.
Nath
,
1996
:
The role of the “atmospheric bridge” in linking tropical Pacific ENSO events to extratropical SST anomalies.
J. Climate
,
9
,
2036
2057
.
Marshall
,
J.
, and
Coauthors
,
2001
:
North Atlantic climate variability: Phenomena, impacts and mechanisms.
Int. J. Climatol.
,
21
,
1863
1898
.
Mitchell
,
T. P.
, and
J. M.
Wallace
,
1992
:
The annual cycle in equatorial convection and sea surface temperature.
J. Climate
,
5
,
1140
1156
.
Mo
,
K. C.
, and
S.
Häkkinen
,
2001
:
Interannual variability in the tropical Atlantic and linkages to the Pacific.
J. Climate
,
14
,
2740
2762
.
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
.
Okajima
,
H.
,
S-P.
Xie
, and
A.
Numaguti
,
2003
:
Interhemispheric coherence of tropical climate variability: Effect of the climatological ITCZ.
J. Meteor. Soc. Japan
,
81
,
1371
1386
.
Okumura
,
Y.
, and
S-P.
Xie
,
2006
:
Some overlooked features of tropical Atlantic climate leading to a new Niño-like phenomenon.
J. Climate
,
19
,
5859
5874
.
Penland
,
C.
, and
L.
Matrosova
,
1998
:
Prediction of tropical Atlantic sea surface temperatures using linear inverse modeling.
J. Climate
,
11
,
483
496
.
Reynolds
,
R. W.
, and
T. M.
Smith
,
1994
:
Improved global sea surface temperature analyses using optimum interpolation.
J. Climate
,
7
,
929
948
.
Robertson
,
A. W.
, and
C. R.
Mechoso
,
2000
:
Interannual and interdecadal variability of the South Atlantic convergence zone.
Mon. Wea. Rev.
,
128
,
2947
2957
.
Schopf
,
P. S.
, and
A.
Loughe
,
1995
:
A reduced-gravity isopycnal ocean model: Hindcasts of El Niño.
Mon. Wea. Rev.
,
123
,
2839
2863
.
Servain
,
J.
,
I.
Wainer
,
J.
Julian
,
P.
McCreary
, and
A.
Dessier
,
1999
:
Relationship between the equatorial and meridional modes of climatic variability in the tropical Atlantic.
Geophys. Res. Lett.
,
26
,
485
488
.
Smith
,
T. M.
, and
R. W.
Reynolds
,
2003
:
Extended reconstruction of global sea surface temperatures based on COADS data (1854–1997).
J. Climate
,
16
,
1495
1510
.
Visbeck
,
M.
,
H.
Cullen
,
G.
Krahmann
, and
N.
Naik
,
1998
:
An ocean model’s response to North Atlantic Oscillation-like wind forcing.
Geophys. Res. Lett.
,
25
,
277
280
.
Wang
,
C.
,
S-P.
Xie
, and
J. A.
Carton
,
2004
:
A global survey of ocean-atmosphere interaction and climate variability.
Earth’s Climate: The Ocean-Atmosphere Interaction, Geophys. Monogr., Vol. 147, Amer. Geophys. Union, 1–19
.
Weare
,
B. C.
,
1977
:
Empirical orthogonal analysis of Atlantic Ocean surface temperatures.
Quart. J. Roy. Meteor. Soc.
,
103
,
467
478
.
Weare
,
B. C.
,
A. R.
Navato
, and
R. E.
Newell
,
1976
:
Empirical orthogonal analysis of Pacific sea surface temperatures.
J. Phys. Oceanogr.
,
6
,
671
678
.
Xie
,
S-P.
,
1999
:
A dynamic ocean–atmosphere model of the tropical Atlantic decadal variability.
J. Climate
,
12
,
64
70
.
Xie
,
S-P.
,
Y.
Tanimoto
,
H.
Noguchi
, and
T.
Matsuno
,
1999
:
How and why climate variability differs between the tropical Atlantic and Pacific.
Geophys. Res. Lett.
,
26
,
1609
1612
.
Zebiak
,
S. E.
,
1993
:
Air–sea interaction in the equatorial Atlantic region.
J. Climate
,
6
,
1567
1586
.

Footnotes

Corresponding author address: Susan Bates, University of Washington, Department of Atmospheric Sciences, Box 351640, Seattle, WA 98195-1640. Email: bates@atmos.washington.edu