Journal of Geophysical Research: Oceans

Decadal variability of shallow cells and equatorial sea surface temperature in a numerical model of the Atlantic

Authors


Abstract

[1] The relative role of extraequatorial mechanisms modulating decadal sea surface temperature anomalies (SSTA) in the equatorial Atlantic is investigated using a suite of sensitivity experiments based on an ocean general circulation model. The model is forced by observed wind stress and/or computed heat flux from an associated advective atmospheric mixed layer model. In addition, the surface forcing is optionally applied on the equator or in off-equatorial regions. The long-term response of equatorial SST is dominated by local forcing. However, a weak but significant part of the response that is not in phase with the locally induced SST variability is caused by remote forcing. Subtropical cells (STCs) provide the oceanic bridging of the climate signals. The dynamical forcing leads to a spin-up and spin-down of the shallow cells, which, in the case of local forcing included, coincides with cold and warm SSTA. The local heat flux forcing reveals an overall damping tendency on the dynamical SST response. When excluding the local forcing, the isolation of the effect of the northern remote forcing from the one in the south appears to be essential in understanding the respective mechanisms at work. In the Northern Hemisphere, the spin-up and spin-down of the STC is highly correlated with the (lagging) SSTA, and the effect of off-equatorial heat flux forcing on SSTA is negligible. In the Southern Hemisphere, both momentum and heat fluxes in the subtropics lead to a significant SST response on the equator.

1. Introduction

[2] Sea surface temperature (SST) in the tropical oceans is of central importance for the global climate. In the past, most attention was drawn to the Pacific, where interannually varying SST anomalies (SSTA) on the equator are known to be an essential part of the coupled ocean-atmosphere ENSO (El Niño/Southern Oscillation) phenomenon. On longer timescales, the equatorial SSTA in the eastern basin (i.e., the “cold tongue”) is characterized by significant decadal/interdecadal modulations, which lead to irregularities of the ENSO, both in amplitude and frequency [e.g., An and Wang, 2000; Fedorov and Philander, 2000]. In the tropical Atlantic, a similar zonally oriented climate mode has been found [e.g., Servain et al., 1982; Philander, 1986; Zebiak, 1993; Carton and Huang, 1994; Ruiz-Barradas et al., 2000; Murtugudde et al., 2001; Xie and Carton, 2004] which is sometimes referred to as the Atlantic Niño.

[3] The cause of the decadal and longer-term variability of equatorial SST is not yet clear. This is subject to intensive ongoing research, with again the main focus on the Pacific. It is believed by some, that the modulation of ENSO can be understood as a solely local chaotic process, arising from the nonlinearity of the tropical ocean-atmosphere system alone [e.g., Timmermann and Jin, 2002]. Others suggest the modulation to be caused by variability within the shallow meridional overturning circulation; namely within the subtropical cells (STCs) [McCreary and Lu, 1994; Liu et al., 1994] which, in general, bridge cold water from the subduction regions in higher latitudes to the tropics and thereby maintain the sharp equatorial thermocline.

[4] In the context of the latter hypothesis, which has to take off-equatorial forcing processes into account, two recently proposed mechanisms of an oceanic teleconnection are favored candidates for remotely driven SSTA on the equator: (1) equatorward advection of temperature anomalies formed by subtropical subduction by the mean flow of the STCs [e.g., Gu and Philander, 1997; Zhang et al., 1998], and (2) changes in the strength of the STCs, which lead to varying amounts of cold water that is transported to the tropics [e.g., Kleeman et al., 1999; Nonaka et al., 2002]. These two scenarios are commonly referred to as the equation imageT′ mechanism and the vequation image mechanism, respectively. Previous model studies have shown that, in the Pacific, the impact of oceanic advection of thermal anomalies from the mid latitudes to the equator is small [e.g., Schneider et al., 1999; Hazeleger et al., 2001b]. In their ocean general circulation model (OGCM) Nonaka et al. [2002] focused on the spin-up and spin-down of the shallow cells and found a profound impact on the equatorial SST variability on the decadal scale; the authors emphasize that the vequation image mechanism appears to be more viable in the Pacific than the equation imageT′ mechanism. The model results are consistent with observational findings. On the basis of their analysis of historical hydrographic data, McPhaden and Zhang [2002] report a gradual increase in equatorial SST in recent decades and a simultaneous decrease in the equatorward geostrophic transport across 9°S and 9°N, indicating decelerated STCs. However, it has to be kept in mind that the analysis of McPhaden and Zhang [2002] does not take transport by the western boundary currents and the Indonesian Throughflow into account that, according to recent modeling studies, have a significant effect on the tropical heat transport in addition to STCs.

[5] In the Atlantic, the picture is even more complicated because of the fact that the shallow subtropical-tropical connections in both hemispheres are superimposed upon the northward warm water return flow of the deep reaching Meridional Overturning Circulation (MOC). This leads to a rather asymmetric mean transport of the STCs; that is, a strong equatorward subsurface flow in the south coincides with weaker contributions from the northern subduction areas. In other words, the majority of the water in the Equatorial Undercurrent (EUC) seems to be of South Atlantic origin [Metcalf and Stalcup, 1967; Wilson et al., 1994; Zhang et al., 2003]. Recent model studies corroborate the observational findings, although the estimated contributions from the northern STC to the ventilation process of the equatorial thermocline may differ significantly from case to case [e.g., Malanotte-Rizzoli et al., 2000; Fratantoni et al., 2000; Harper, 2000; Jochum and Malanotte-Rizzoli, 2001; Hazeleger et al., 2003; J. Kröger and C. Böning, Pathways and seasonality of interhemispheric transport of South Atlantic water into the Caribbean Sea, manuscript in preparation, 2005, hereinafter referred to as Kröger and Böning, manuscript in preparation, 2005]. However, model results are known to strongly depend on parameterizations (e.g., the representation of the MOC in regional configurations) or the forcing (e.g., the applied wind product [see, e.g., Inui et al., 2002]). Therefore it is difficult to evaluate the STCs in individual realizations, when, at the same time, the observational database is sparse.

[6] To elucidate the respective roles of both (1) local and remote and (2) dynamical and thermodynamical forcing on decadal SSTA in the equatorial Atlantic in connection with the transient behavior of the shallow cells we are going to analyze a set of sensitivity runs which were conducted with a regional OGCM of the tropics and subtropics. The model and the experiments will be introduced in section 2. The particular forcing functions that characterize each of the experiments are based on one single data set of interannually varying winds (section 2.1). We start this study by performing a fully forced reference run. Thereafter, a set of sensitivity experiments is performed by restricting the imposed low-frequency forcing in two independent ways (see section 2.2). The results of the sensitivity experiments with respect to the equatorial SST response, the variability of the shallow overturning circulation and the meridional heat transport, and the relation between these properties are presented in section 3. Finally, we summarize and discuss the main findings of this paper in section 4.

2. Model

2.1. Configuration

[7] A reduced gravity, primitive equation, sigma coordinate OGCM [Gent and Cane, 1989; Murtugudde et al., 1996] is utilized for this study in the tropical and subtropical Atlantic. With a coupling to an atmospheric mixed layer model (AML) [Seager et al., 1995] and with freshwater fluxes treated as a natural boundary condition [Huang, 1993], the model explicitly accounts for a complete upper ocean hydrology [Murtugudde and Busalacchi, 1998]. The AML calculates (1) air temperature and air humidity together with the surface fluxes of latent and sensible heat and longwave radiation, and (2) evaporation (E), which is used to calculate the fresh water flux. This requires inputs of solar radiation (Earth Radiation Budget Experiment (ERBE); see, e.g., http://asd-www.larc.nasa.gov/erbe/ASDerbe.html), cloud cover (International Satellite Cloud Climatology Project (ISCCP) [Rossow and Schiffer, 1991]), precipitation (P) [Xie and Arkin, 1995], winds (National Centers for Environmental Prediction (NCEP) [Kalnay et al., 1996]) and the modeled SST. The fluxes at the top of the atmospheric mixed layer are parameterized according to Seager et al. [1995], and the surface fluxes are determined by bulk transfer formulae. Note that there is neither an explicit relaxation to prescribed sea surface temperatures (SST) nor salinities (SSS) in the regional model. Near the southern and northern boundaries salinity and temperature are relaxed to vertical profiles of the Levitus et al. [1994a, 1994b] climatology. In addition, the Levitus et al. [1994a, 1994b] database is used to initialize the model.

[8] The model domain extends over 40°S–40°N, 100°W–20°E with a resolution of 1/2° in longitude and a stretched grid in latitude (1/3° within 10°S–10°N and 1° near the boundaries). In the vertical, the model consists of 24 active sigma coordinate layers beneath a variable thickness, surface mixed layer. All relevant properties of the surface mixed layer and of the deepest active layer are computed prognostically. The remaining layers are computed diagnostically according to their prescribed fixed ratio in layer thickness. The application of sigma coordinates allows for a persistent enhanced resolution just below the mixed layer [see, e.g., Inui et al., 2002; Lazar et al., 2002]. The embedded hybrid mixing scheme of Chen et al. [1994] represents the main processes of vertical turbulent mixing in the upper ocean: The bulk mixed layer model [Kraus and Turner, 1967] relates entrainment and detrainment to wind and buoyancy forcing. Below the mixed layer, the gradient Richardson number closure is added to a background diffusivity of 10−5m2/s to derive vertical exchange of mass and water properties between layers. A simple convective adjustment accounts for static instabilities. Lateral mixing of water properties, layer thickness, and momentum is computed along sigma layers using a high-order Shapiro filter. The deepest active layer is bounded at the bottom by an isopycnal surface and an inactive bottom layer below, both of which with a density of 1027.6 chosen because it corresponds to the 6° isotherm in the Levitus et al. [1994a, 1994b] data, considered the lower limit of the thermocline.

[9] Although the model lacks a deep circulation and therefore a southward spreading of North Atlantic Deep Water (NADW) as the lower branch of the Meridional Overturning Circulation (MOC), it is able to mimic the warm water return flow of the MOC quite reasonably by restoring to the monthly resolved Levitus et al. [1994a, 1994b] climatology in the northern and southern sponge layers. This leads to a mean cross equatorial flow of O[12 Sv] in the upper 1200 m (see section 3.1). For a further discussion on the representation of the upper limb of the MOC in the reduced gravity model, see Lazar et al. [2002]. Overall, all forcing components that drive the regional model are based on climatological fields, namely precipitation, cloud cover, and solar radiation, except for those computed from the NCEP wind data set (i.e., latent and sensible heat, and momentum fluxes which are derived from wind speed, and wind stress, respectively). The parameterization of the spatially variable penetration of the shortwave radiation has been shown to significantly improve the SST of the model [Murtugudde et al., 2002]. This parameterization, based on the coastal zone color scanner (CZCS) derived attenuation depths, has also been included here.

2.2. Experiments

[10] By spinning up the model from a state of rest for 100 years, using climatological NCEP wind fields (computed from the years of 1948 to 2001), we not only gain an adequate adjustment in the dynamical fields but also assure that the residual asymptotic drift in SST (extreme values <.05°C/50 years) is negligible compared to the respective low-latitudinal response of that quantity to each of the specified forcing functions in the sensitivity experiments. All experiments start from the end of the climatological spin-up. As it was mentioned in section 2.1, the only forcing capable of introducing interannual and longer-term variability into the model is based on atmospheric winds and wind stresses derived from the NCEP data set. To isolate the proposed mechanisms that potentially drive SST variability along the equator, we decompose the wind fields in two ways as follows.

[11] First of all, the thermodynamical and dynamical surface forcing are separated by either applying only the observed interannual varying wind speed (“HFLUX”), leading to latent and sensible heat flux variations (plus a wind advection component in the AML), or only wind stress (“STRESS”). In each case the respective other field is replaced by its climatological counterpart. The standard experiment, which allows for both observed forcing components, is indicated by “FULL”. Secondly, we perform a regional decomposition in a similar way as was done by Nonaka et al. [2002] in the Pacific. Climatological and interannual forcings are blended using the filter function:

equation image

leading to suppressed interannual variability either on or off the equator with smooth transition zones between 5° and 15° in both hemispheres, indicated by “NE” (No Equator) and “OE” (Only Equator), respectively. The standard case, which applies the observed forcing everywhere, is denoted by “WD” (Whole Domain). Eventually we complement and refine the set of remotely forced experiments by inducing interannual variability in the subtropics of either one hemisphere, indicated by “ON” (Only North) and “OS” (Only South), respectively. Overall, the forcing scenarios lead to 15 different model realizations, which are summarized in Table 1.

Table 1. Sensitivity Experiments Based on a 100 Year-Long Climatological Spin-up Comprising the Years 1948–2001a
Wind-Based Forcing ComponentsRegion of Applied Forcing
Whole DomainOnly EquatorNo EquatorOnly NorthOnly South
  • a

    For further explanation on naming conventions of experiments, see section 2.2.

Full interannual setFULL_WDFULL_OEFULL_NEFULL_ONFULL_OS
Heat flux onlyHFLUX_WDHFLUX_OEHFLUX_NEHFLUX_ONHFLUX_OS
Momentum flux onlySTRESS_WDSTRESS_OESTRESS_NESTRESS_ONSTRESS_OS
Full climatological setCLIM    

3. Results

3.1. Meridional Flow in the Model

[12] A common means to get insight into net meridional transport accompanied by the shallow and the deep reaching overturning cells in the model is to compute the zonal integral of the flow field. Figure 1 (top) shows the accordant vertical stream function (VSF) of an exemplary year (1984, a year with anomalous strong easterly trade winds over the entire Atlantic basin between about 5°S and 20°N) in the fully interannual reference run (FULL_WD). The displayed VSF is based on the low-pass-filtered (weighted 10-year running mean) time series of the volume transport which was integrated from Africa to the Americas and, important to note, accumulated from the surface downward over depth levels, described by

equation image

with longitude λ and latitude ϕ, and a being the radius of the earth and v the meridional velocity component.

Figure 1.

(top) Low-pass-filtered (weighted 10-year running mean) vertical stream function (VSF) in Cartesian coordinates, (middle) related anomalies (VSFA), and (bottom) VSF in density coordinates in FULL_WD in 1984 (in Sv). The thick black line indicates the local extreme values of the VSF in the upper 200 m. Note the different scaling and the respective exaggerated vertical axes of the upper 200 m in the top plots.

[13] The projection of the flow field into the meridional-vertical plane reveals strong tropical cells (TCs) in both hemispheres, which are confined to narrow bands of 5° width on either side of the equator and to about the upper 100 m. The southern subtropical cell extends approximately from 20°S to the equator and from the surface to 200 m. The northward return flow of the MOC encompasses the whole domain. It is not clear if, in this model, an additional feeding from the northern extratropics into the equatorial thermocline exists (via the northern STC) since, in the integral picture, an efficient northern subtropical-tropical connection might simply be hidden by the superimposed MOC. In contrast the MOC might successfully block the northern STC from supplying the Equatorial Undercurrent (EUC) as was shown by Lagrangian tracking of subducted water masses in z level models [Jochum and Malanotte-Rizzoli, 2001; Kröger, 2001; Hazeleger et al., 2003].

[14] The 1984 anomalies of the vertical stream function (VSFA) in Figure 1 (middle), that is, deviations from the overall mean of the applied 54-year NCEP forced period, already indicate the potential importance of a northern STC in conveying transient signals from higher latitudes to the equatorial region. Whereas there exists a broad consensus among numerical models that the mean supply of the equatorial Atlantic thermocline is predominantly accomplished from the south [Fratantoni et al., 2000; Harper, 2000; Jochum and Malanotte-Rizzoli, 2001; Hazeleger et al., 2003; Kröger and Böning, manuscript in preparation, 2005], this may not necessarily be the case for the propagation of long-term anomalies. Indeed, in the following, we will identify a significant role of the northern STC in bridging extratropical induced signals to lower latitudes and, by that, in modulating equatorial SST variability on the decadal timescale.

[15] Furthermore, it is necessary to focus on the strong tropical cells (TCs), since several studies based on z level OGCMs have suggested that, in the mean, these cells do not contribute to the ventilation process of the equatorial thermocline and therefore do not feed the equatorial “cold tongue” [Hazeleger et al., 2001a, 2003; Kröger and Böning, manuscript in preparation, 2005]. The TCs in z level models disappear when the zonal integration of the flow field is conducted within density layers (σθ) instead of Cartesian coordinates, described by

equation image

In this regard, it is important to note that, unlike in the previous mentioned studies, both of the TCs in the present model do not vanish when the zonal integral is computed along isopycnals (Figure 1 (bottom)). After the transformation of the vertical coordinates, a mean northern TC with a maximum of still more than 11 Sv is apparent. The northern TC clearly connects the mixed layer to the upper thermocline that is located at about 24.5 σθ in our model. On the contrary, the southern TC, that rotates with the same magnitude as the northern TC in the Cartesian framework, has totally disappeared. It has to be noted that we were using monthly data, in order to assess the impact of higher-frequency eddy fluxes we were integrating over daily sampled density fluxes for two years which led to a slight weakening of the northern TC by less than 3 Sv. The persistence of the northern TC can probably be attributed to the fact that the ocean mixed layer is explicitly represented in our model, which, if one regards this as an advantage over the fixed z level models with generally too coarse vertical resolution in surface and thermocline layers (O[10 m]), suggests a mean contribution of the northern TC to the equatorial ventilation process. However, in the remainder of the paper, we will, in addition to the STCs, focus on the role of the variability of the TCs with respect to long-term SSTA on the equator.

3.2. Long-Term SST Response on the Equator

[16] A comparison of phase (Hovmoeller) diagrams of long-term anomalies of equatorial SST for nine experiments is shown in Figure 2 (cf. the first three columns in Table 1). First of all, the forced response of SSTA in each particular experiment is significant with respect to the internal long-term variability of equatorial surface temperatures due to instability processes. Internal variability becomes potentially important when going to higher, eddy-resolving horizontal resolution as is indicated by the study of Jochum et al. [2004]. However, against the background of the eddy-permitting grid size in our experiments, the fully climatological reference run (CLIM) reveals far weaker instability signals along the equator than the interannual forced runs. These signals are not capable of showing up in the chosen range of contours in Figure 2 (absolute values below 0.02°C, not shown). Secondly, the SST responses in the sensitivity runs are to first-order linear, both in direction of the thermodynamical/dynamical and of the regional decomposition. Adding SSTAs of complementary experiments, in all cases, leads to about the same response that is found in the respective fully forced run.

Figure 2.

(top) Low-pass-filtered (weighted 10-year running mean) SSTA on the equator in (left) FULL_WD versus (center) FULL_OE versus (right) FULL_NE. (middle) Low-pass-filtered (weighted 10-year running mean) SSTA on the equator in (left) HFLUX_WD versus (center) HFLUX_OE versus (right) HFLUX_NE. (bottom) Low-pass-filtered (weighted 10-year running mean) SSTA on the equator in (left) STRESS_WD versus (center) STRESS_OE versus (right) STRESS_NE. Values are in degrees Celsius. Note the different scaling in *_NE (in the right plots): [−0.26 0.26] instead of [−0.9 0.9].

[17] In the model, equatorial SSTA is dominated by local forcing (compare left, center, and right plots in Figure 2, indicative of forcing over the whole domain, and only on and off the equator, respectively). Furthermore, in this case (with the observed atmospheric functions on the equator applied) the signature and magnitude of the SST response is dominated by wind stress (dynamical) forcing (compare the top, middle, and bottom plots in either the left or the center plots in Figure 2, in both cases a lineup of the full, the thermodynamical, and the dynamical runs). The heat flux forcing acts on a slightly different timescale and, especially in the eastern half of the basin, it seems to damp the dynamical response until the beginning of the 1990s and to amplify it afterward.

[18] Unlike in the Pacific model of Nonaka et al. [2002], the effect of off-equatorial forcing on decadal SSTA is weaker than the local forcing, but is still significant compared to model drift and internal long-term variability as was already mentioned above (right plots in Figure 2). Here, the dynamical and thermodynamical remote forcing functions lead to rather equally pronounced long-term modulations of equatorial SST (compare top and middle plots in Figure 2 (right)). Moreover, the particular response patterns to the respective variations of stress and heat flux in the extratropics are, in contrast to those in the local forced experiments, positively correlated over nearly the entire investigation period. Note the rather zonally uniform response in the remote experiments; a similar equatorial response to slow extratropical forcing was found to be consistent in a linear and a nonlinear model of the Pacific and attributed to linear wave dynamics [Hazeleger et al., 2001b].

3.3. Variability of Shallow Cells

[19] In the following, we want to focus on the spin-up and spin-down of the shallow cells and their relation to the long-term SST variability on the equator (the vequation image mechanism). A key issue is to define a metric for the strength of the water mass transport within the cells and especially their temporal variability. For this purpose, we track the local extremum of the VSF in the upper 200 m as a function of latitude and time (thick black line in Figure 1 (top)). The Hovmoeller diagram of the resulting (low-pass-filtered) VSF index points to the latitudes of strongest variability in the shallow cells and related timescales.

[20] In FULL_WD the most pronounced signals are located at about 2° on either side of the equator (Figure 3), which corroborates the dominance of local forcing. Weak TCs in both hemispheres between about 1960 and 1965 are followed by enhanced cell strengths in the early 1980s (with up to 6 Sv peak to peak VSF index difference in the north). Note that the variability of both TCs is not in phase. The local maximum in the Northern Hemisphere follows the minimum after 22 years, whereas it is 16 years in the south. A comparison of the time evolution of the VSF index in STRESS_WD and HFLUX_WD in Figure 4 shows that the spin-up and spin-down of the shallow cells is almost entirely induced by the wind stress. The VSF response in HFLUX_WD is negligible (about one order of magnitude lower than in STRESS_WD). The unaltered VSF in HFLUX_WD is most likely due to an unaltered density structure of the underlying flow field and is therefore strongly indicative of salt compensation when temperature anomalies subduct in the subtropics and propagate/advect within the thermocline, a process that was already identified (in a similar model configuration) by Lazar et al. [2001]. The decomposition into dynamical and thermodynamical surface forcing successfully isolates the vequation image mechanism from the equation imageT′ mechanism.

Figure 3.

Hovmoeller of low-pass-filtered (10-year running mean) VSF index (anomalies of local extreme values) in FULL_WD (in Sv). Note that positive (negative) values in the Northern (Southern) Hemisphere denote strengthening of the shallow cells and vice versa.

Figure 4.

Hovmoeller of low-pass-filtered VSF index anomalies in (top) HFLUX_WD and (bottom) STRESS_WD (in Sv). Note that positive (negative) values in the Northern (Southern) Hemisphere denote strengthening of the shallow cells and vice versa.

[21] Figure 5 shows Hovmoeller diagrams of the accompanying meridional heat transport in both experiments, integrated from Africa to the Americas and over all active model layers. The variability of heat transport is much stronger in STRESS_WD than in HFLUX_WD with most pronounced peak to peak differences of about 0.3 PW in the northern tropics at 2–3°N. The distinctive coherence of heat transport and cell strength variability at this latitude points to the important role of the northern TC in communicating long-term variability into the equatorial thermocline. At about 2°S, the pronounced VSF index variability of the southern TC does not project onto a significantly varying heat transport, indicative that the southern TC and its variability are confined to the upper mixed layer or fully compensated by eddy fluxes (cf. Figure 1 (bottom)). In HFLUX_WD patterns of anomalous heat transport prevail in the Southern Hemisphere, but their magnitude is about one order below that one in STRESS_WD.

Figure 5.

Hovmoeller of low-pass-filtered heat transport anomalies in (top) HFLUX_WD and (bottom) STRESS_WD (in PW).

[22] In the case of remote forcing only (FULL_NE), unlike in FULL_WD, the most pronounced response of the shallow cell variability is located off the equatorial region, at about 10 to 15° in both hemispheres, indicative of STCs with rather weak peak to peak VSF index differences ranging between −1 and 1 Sv (Figure 6). Until the mid-1970s the southern and northern cells compensate in strength simultaneously on the decadal scale (the relative weak northern STC in 1960 is strengthened after 1970 while at the same time the initially strong southern STC decelerates), whereas later on, this correlation falls apart. The northern STC between 1980 and 1990 and the southern STC in the 1990s, both spin-up and, together, lead to enhanced transport toward the equator. The transition from warm SST anomalies to cold anomalies from 1980 to 2000 in STRESS_NE (bottom right plot in Figure 2) together with the concomitant increase in transport of the shallow cells is indicative of the tight relationship between the equatorial SST and the transport anomalies, and points to the vequation image mechanism. Although the VSF anomalies are characteristic of coherent patterns from the subtropics to the equator, the overall latitudinal decrease of the magnitude of the anomalies toward the equator strongly indicates local off-equatorial upwelling, most likely within the adjacent countercurrents, that is, the Northern and the Southern Equatorial Countercurrent (NECC, SECC).

Figure 6.

Hovmoeller of low-pass-filtered VSF index anomalies in FULL_NE (in Sv). Note that positive (negative) values in the Northern (Southern) Hemisphere denote strengthening of the shallow cells and vice versa.

3.4. Relation Between VSF Index, Heat Transport, and Equatorial SSTA

[23] With the focus on the vequation image mechanism, in the following we investigate the relation between the VSF index, meridional heat transport and equatorial SSTA in the wind driven only experiments. By further isolating the remote from the local forcing cases, we expect to identify the subtropically induced modulation of SST variability on the equator, together with the transient behavior of the shallow cells, which, in the vequation image scenario, by spinning up and down, may communicate the wind driven dynamical changes into the tropics in the form of an “oceanic bridge”.

[24] Figure 7 shows low-pass-filtered (weighted 6-year running mean) time series of the equatorial SSTA in the eastern half of the basin (box mean from 1°S to 1°N and 25°W to Africa) and of anomalies of the “combined” VSF index (the difference of cell strengths) at 2°N and 2°S and of the heat transport convergence between these latitudes from experiment STRESS_WD. The acceleration and deceleration of the local TCs are highly correlated with the equatorial divergence and convergence of heat transport, and both together are anticorrelated with the SST response in the upwelling region of the EUC. Strong cells coincide with enhanced poleward transport of heat and anomalous cold SST on the equator and vice versa, indicative of the vequation image mechanism. A peak to peak overturning difference of 6 Sv is related to changes of about 0.2 PW in the net heat transport and 1°C in the east equatorial surface temperature. The best correlations of the combined VSF index and the heat transport convergence with the SST response can be found at no lag, with values of −0.87 and −0.76, respectively. A Monte Carlo test based on the Random Phase method of Ebisuzaki [1997] provides an estimate of the significance levels. All the correlations reported in this study are significant at 99%.

Figure 7.

Time series of low-pass-filtered (6-year running mean) anomalies of the combined VSF index (difference of local extreme values) at 2°S and 2°N (in Sv, solid curve), of the heat transport convergence between these latitudes (in PW, dotted curve), and of east equatorial SST (box mean from 1°S to 1°N and 25°W to Africa) (in degrees Celsius, dashed curve) in STRESS_WD.

[25] In order to understand the relative roles of the Northern and the Southern Hemisphere in the vequation image scenario, we again calculate the correlations of the equatorial SST response with the heat transport and cell strength variability in STRESS_WD, but this time at isolated latitudes (Figure 8). The coherent mass and heat transport variabilities at 2°N look very similar to the anomalous convergences in Figure 7. However, the heat transport peak to peak difference across 2°N is almost twice as large as it is for the convergence between 2°N and 2°S, whereas the magnitude of the varying VSF index at 2°N is slightly reduced. The highest correlations of the cell strength and heat transport variability at 2°N with the SST response are found to be −0.79 and −0.83, with the SSTA lagging by 3 and 12 months, respectively. The VSF index is lagging behind the heat transports. This fact indicates these two parameters have different adjustment timescales to the applied wind forcing. The relatively fast Ekman response at the surface seems to predominantly set the changes in heat transport, whereas our integral definition of the VSF index apparently includes the slower geostrophic adjustment in the thermocline. At 2°S, in contrast to the north, neither a significant correlation between VSF index and heat transport (as was discussed before in section 3.3) nor between these two properties and the equatorial SST response can be found.

Figure 8.

Time series of low-pass-filtered (6-year running mean) anomalies of VSF index (in Sv, solid curve), of heat transport (in PW, dotted curve), and of east equatorial SSTA (box mean from 1°S to 1°N and 25°W to Africa) (in degrees Celsius, dashed curve) at (top left) 2°N, (top right) 2°S, (bottom left) 9°N, and (bottom right) 9°S in STRESS_WD. Note that positive (negative) values of VSF index in the Northern (Southern) Hemisphere denote strengthening of the shallow cells and vice versa.

[26] When going to higher latitudes, namely to 9°N and 9°S, again we find coherent mass and heat transport variabilities in the Northern Hemisphere. At 9°N, the transports of mass and heat are highly correlated with the equatorial SSTA when leading by 6 and 18 months, respectively (with values of −0.80 and −0.72). In comparison to 2°N, both transports are weakened by a factor of about 2. No correlations can be found at 9°S. The spin-up and spin-down of the northern shallow cell circulation plays a key role in the long-term modulation of equatorial surface temperature in the upwelling region. The coherent picture of local momentum fluxes driving the mass and heat transport variability, and eventually the SST response at the equator, is hereby not confined to the localized northern TC, but is shown to be detectable up to 9°N. However, the SSTAs in the experiments where the local forcing was suppressed are not only considerably weaker, but also far from coherent compared to the signals in the runs that include equatorial forcing (cf. Figure 2). To filter out the dominant effect of the local wind stress, we need to repeat the correlation analysis with the only remotely forced experiments.

[27] The only significant correlation in STRESS_NE can be found between the VSF index and the heat transport variability at both 2°N and 9°N (0.94 and 0.73 with 4 and 6 months lag, respectively; not shown). The equatorial SST response in STRESS_NE is neither correlated with the mass and heat transport variability at isolated latitudes (i.e., at 2° and 9°N/S) nor with their low-latitudinal anomalous convergences. In order to understand the relative roles of the northern and the Southern Hemisphere in the remotely driven vequation image scenario, we repeat the correlation analysis in the runs that were only forced in the extratropics of either one hemisphere. Figure 9 shows time series of equatorial SSTA in the eastern half of the basin and mass and heat transport variability at the same latitudes as in Figure 8, but this time for experiments STRESS_ON (Figure 9 (left)) and STRESS_OS (Figure 9 (right)). The equatorial SST response in STRESS_ON is highly correlated with both the anomalous mass and heat transport at both investigated latitudes in the Northern Hemisphere. At 2°N and at 9°N, highest values can be found with the VSF index leading by 10 and 17 months, respectively (−0.91 and −0.86) and with the changes in heat transport leading by 17 and 22 months (−0.89 and −0.82). A peak to peak overturning difference of 0.7 Sv (0.9 Sv) at 2°N (9°N) is related to changes of about 0.1 PW in the net heat transport at these latitudes and 0.15°C in the east equatorial surface temperature. In contrast, no significant correlation can be found between both the mass and heat transport at 2°S and 9°S and the equatorial SST response to the forcing in the Southern Hemisphere in STRESS_OS.

Figure 9.

Time series of low-pass-filtered (6-year running mean) anomalies of VSF index (in Sv, solid curve) of heat transport (in PW, dotted curve), and of east equatorial SSTA (box mean from 1°S to 1°N and 25°W to Africa) (in degrees Celsius, dashed curve) in (top left) STRESS_ON at 2°N, (top right) STRESS_OS at 2°S, (bottom left) STRESS_ON at 9°N, and (bottom right) STRESS_OS at 9°S. Note that positive (negative) values of VSF index in the Northern (Southern) Hemisphere denote strengthening of the shallow cells and vice versa.

[28] The decomposition of the off-equatorial forcing functions into their application either only in the northern or in the southern subtropics turns out to be essential for understanding the underlying bridging mechanism in each hemisphere. Figure 10 compares the Hovmoeller diagrams of long-term anomalies of equatorial SST for all experiments with the forcing being applied in either one hemisphere (cf. the last two columns in Table 1). First of all, note that the respective SST responses to the northern (*_ON) and the southern (*_OS) forcing are not coherent, no matter if you compare the full (FULL_*), the thermodynamical (HFLUX_*), or the dynamical (STRESS_*) runs (compare top, middle, and bottom plots, respectively, in Figure 10). Secondly, it appears that the dynamical forcing in the north (STRESS_ON) almost exclusively sets the equatorial SST response relative to HFLUX_ON, whereas in the Southern Hemisphere the thermodynamical contribution (HFLUX_OS) leads to more pronounced SST anomalies than the interannual momentum fluxes applied in the southern subtropics in STRESS_OS (compare left and right plots, respectively, in Figure 10). The SSTA shown in Figure 10, together with the results of the correlation analysis, are strongly indicative of different mechanisms that govern the ocean bridging of climate signals in each hemisphere. A coherent relation between equatorial SSTA and mass and heat transport in the Southern Hemisphere, which would point to a dominating role of vequation image in communicating long-term variability from the southern subtropics toward the equator could not be found. Thus in the south we should rather expect a combination of the vequation image mechanism and the equation imageT′ mechanism acting in concert. In the Northern Hemisphere, on the other hand, a predominant role of vequation image in modulating SST variability in the equatorial upwelling region is strongly suggested by our model.

Figure 10.

(top) Low-pass-filtered (weighted 10-year running mean) SSTA on the equator in (left) FULL_ON versus (right) FULL_OS. (middle) Low-pass-filtered (weighted 10-year running mean) SSTA on the equator in (left) HFLUX_ON versus (right) HFLUX_OS. (bottom) Low-pass-filtered (weighted 10-year running mean) SSTA on the equator in (left) STRESS_ON versus (right) STRESS_OS. Values are in degrees Celsius.

4. Summary and Discussion

[29] We have used an ocean GCM to investigate the respective role of dynamical and thermodynamical atmospheric forcing on decadal SST variability in the equatorial Atlantic. Furthermore, the forcing functions have been regionally partitioned to gain insight into the relative roles of locally and remotely induced variability on the low-latitudinal response. The model was driven by observed momentum and heat fluxes based on NCEP wind analyses, with the heat fluxes being computed from an associated advective atmospheric mixed layer model. Our sensitivity studies show that, in contrast to model results in the Pacific by Nonaka et al. [2002], the relative contributions of equatorial and off-equatorial winds in driving the long-term SST variability are significantly different. The equatorial SST response in the Atlantic is dominated by local forcing. This dominance is a robust model result with respect to the individual NCEP-based forcing components, which means that it either holds for applying only wind stresses (dynamical forcing), or only heat fluxes (thermodynamical forcing), or for applying both components. Extreme values of the long-term SST variability in the upwelling region of the east equatorial basin differ by up to 2°C, with the local surface heat flux forcing revealing an overall damping tendency on the dynamical response to local wind stresses over almost the entire investigated period. However, extratropical wind forcing does significantly modulate equatorial SSTA on the decadal scale, even though the magnitude of the remotely induced SST variability is about 1/5 of the local response. When applied simultaneously in the Northern and the Southern Hemispheres, both the dynamical and the thermodynamical components are effective in exciting a distinctive, rather coherent long-term response on the equator that is not in phase with the particular SSTAs in the locally forced runs.

[30] The communication of the remotely induced variability toward lower latitudes is provided by the subtropical cells (STCs). By decomposing the forcing into its dynamical and thermodynamical components we have successfully isolated two recently proposed mechanisms of the oceanic bridging of climate variability, which are commonly referred to as equation imageT′ and vequation image. In the equation imageT′ scenario, temperature anomalies are formed by the anomalous heat fluxes in the subtropics, subduct, and experience equatorward advection by the mean flow of the STCs. In the vequation image scenario, off-equatorial wind stress variability spins up and down the STCs and thus leads to varying amounts of cold water that is transported to the tropics. Both mechanisms appear to be effective in the model. In the context of the equation imageT′ scenario, Lazar et al. [2001] investigated the propagation of a synthetic temperature anomaly in the subsurface branch of the southern STC by conducting a process study with a similar configuration of the model applied here (exclusively driven by climatological fields). Since salinity compensation tempers the density perturbation in their experiment, the temperature anomaly does not lead to significant anomalous velocities in the thermocline flow, and the propagation of the temperature signal is explained by the authors to first approximation by time-mean flow advection. Our experiments driven by heat fluxes display negligible variability of the shallow cells and thus corroborate the effective salt compensation in the model. The spin-up and spin-down of the TCs and STCs has to be attributed to the dynamical forcing component. The experiments which include the observed momentum fluxes indeed produce significant variability in the shallow overturning circulation.

[31] For the remainder of the paper the focus was on vequation image and therefore on the experiments which were driven by the wind stresses only. In order to be able to measure the variability of the shallow cells we computed the vertical stream function (VSF) of the zonally integrated flow field and tracked its local extreme values in the upper 200 m, leading to the “cell strength” or “VSF” index, as a function of latitude and time. Hovmoeller diagrams of the VSF index, which point to the latitudes of strongest variability in the shallow overturning, reveal most pronounced signals at about 2° on either side of the equator in the cases where equatorial forcing was included, and at about 10 to 15° in each hemisphere in the experiments where only off-equatorial forcing was applied, indicative of strong variability within the TCs and STCs, respectively. Hovmoeller diagrams of the accompanying meridional heat transport variability reveal coherent patterns with the VSF index in the Northern Hemisphere that are absent in the south. Hereafter a correlation analysis was performed to gain more detailed insight into the relation between the spin-up and spin-down of the shallow cells, the heat transport variability, and the variability of SST in the east equatorial basin.

[32] In the experiments which included local (equatorial) forcing, across 2°N and 2°S a high correlation was found between the (combined) spin-up and spin-down of the TCs, the anomalous convergence of heat transport, and the equatorial SST response. Weak (strong) cell strengths coincide with enhanced (reduced) convergence of heat transport and warm (cold) SST anomalies on the equator. The same significant relation between the equatorial SST response, VSF index, and anomalous heat transport could be found at isolated latitudes in the Northern Hemisphere. Hereby the SSTA is lagging behind the mass and heat transport variability at 2°N by 3 and 12 months and at 9°N by 6 and 18 months, respectively. At 2°S and 9°S, in contrast to the north, neither a significant correlation between VSF index and heat transport nor between these two properties and the equatorial SST response can be found. In the remotely, simultaneously in both hemispheres driven experiment no significant correlations between the equatorial SST response and the mass and heat transports at any latitude could be found. In the case in which the forcing was applied only in the northern extratropics, we found high correlations between the equatorial SST response and both the anomalous mass and heat transport in the Northern Hemisphere, with the VSF index leading by 10 and 17 months and the changes in heat transport leading by 17 and 22 months, at 2°N and 9°N, respectively. On the other hand, when the forcing was applied only in the southern extratropics, no significant correlations could be found at all.

[33] It was shown by the hemispheric decomposition of the wind fields that the combined effect of the remote forcing in both hemispheres on the SST response is misleading and obscures how and to what extent the variability of the individual shallow cells maps onto the respective SSTA. The spin-up and spin-down of the northern STC is highly (anti) correlated with the equatorial SST response and leads at the order of 1 to 2 years, with increasing lag toward higher latitudes. Furthermore, the effect of heat flux forcing is negligible in the Northern Hemisphere, and thus the oceanic bridging of climate variability is dominated by the vequation image mechanism. In the Southern Hemisphere the situation is not as clear. Remotely induced variability in the form of both, long-term heat and momentum fluxes significantly contributes to equatorial SSTA. Hence the way by which the southern STC communicates climate signals toward lower latitudes is a combination of the vequation image mechanism and the equation imageT′ mechanism.

[34] The focus here was on the long-term SST response on and around the equator, since, in the context of a possible coupled mode, the low latitudes are the most likely region to see a significant influence on the atmosphere variability by the SST. It has to be noted that this study was based on a regional ocean GCM. Because of the prescribed climatological forcing at the northern and southern boundaries and the NCEP wind-based forcing on top, the role of interactions of the shallow cells with the long-term variability of the MOC or the effect of full coupling to an unconstrained atmosphere could not be addressed here. However, we believe that our sensitivity runs were fully capable of discerning the respective roles of the shallow cells in modulating the equatorial SST variability. Currently we are extending the analyses to other tropical upwelling (and obduction) regions, namely the Guinea Dome and the Angola Dome. Hereby, special attention will be drawn to the Northern Hemisphere, where, in the mean picture, the upwelling of the NECC most likely constitutes the southern edge of the subtropical-tropical connection, since the interhemispheric return flow of the MOC is commonly meant to block the northern STC away from the equator, a well known asymmetric characteristic that is unique to the Atlantic Ocean. Note that in the regional model of the tropical and subtropical Atlantic of Kröger and Böning (manuscript in preparation, 2005) a mean cross-equatorial flow of the MOC of less than 10 Sv is sufficient in suppressing the supply of the equatorial thermocline from the north. The model used in this study superimposes a MOC return flow of more than 12 Sv onto the shallow cells and is very likely to provide the same effective blocking mechanism.

[35] Another asymmetry in the Atlantic was discovered in this study. The communication of long-term variability from the subtropics to the tropics is subject to different underlying mechanisms in each hemisphere. Whereas in a model study of the Pacific it was shown that the combined cell strength variability of the northern and the southern STC is anticorrelated with the equatorial SST response [Nonaka et al., 2002] and therefore the spinning up and spinning down of the shallow overturning cells in both hemispheres acts in concert, in the Atlantic only the isolated transport variability of the northern STC is able to map its signature significantly onto the respective SSTA. The relatively short timescale that it takes for communicating the remotely induced variability to the equatorial band, together with the rather uniform zonally elongated SST response, resembles the behavior of response experiments to extratropical dynamical forcing in the Pacific [Hazeleger et al., 2001b] and indicates that the communication, as in the Pacific runs, is accomplished by propagation of Rossby and Kelvin waves. A hypothesis for the asymmetry of the governing transfer mechanisms of long-term variability in the Atlantic could be that the MOC return flow in our model is effective in blocking southward advection of anomalous signals away from the equator while, at the same time, it is exclusively permeable for wave propagation. Hence only a spin-down of the MOC would allow for temperature anomalies of northern origin to reach the equatorial thermocline and subsequently superimpose their signature onto the prevailing SST variability. To test this hypothesis we plan to repeat our sensitivity runs with a suppressed MOC return flow.

Acknowledgments

[36] We are grateful for comments and suggestions of Raghu Murtugudde, Eric Hackert, and Markus Jochum. This research was funded with NOAA grant NA16GP1576, NSF grant 523964, and NASA grant NAG5-7194.

Ancillary