1. Introduction

On interannual time scales, the climate of the tropical Pacific Ocean is dominated by the El Niño–Southern Oscillation (ENSO) phenomenon. This phenomenon results from the interaction between the surface winds and the sea surface temperature (SST), as highlighted by Bjerknes (1969). The oscillatory nature of ENSO is linked to the subsurface adjustment of the ocean that provides the memory of the coupled ocean–atmosphere system. Wyrtki (1975, 1985) and Cane and Zebiak (1985), using respectively, sea level observations and a simple coupled ocean–atmosphere model, found that the period between two El Niño events corresponds to the time required to recharge the equatorial band with warm water. In Wyrtki's (1975) view, the buildup of heat in the western Pacific is a precursor of El Niño. This mechanism contrasts with the delayed oscillator theory (Schopf and Suarez 1988; Battisti and Hirst 1989), which became the leading paradigm for ENSO during the Tropical Ocean Global Atmosphere (TOGA) program. In this theory, the adjustment period is controlled by the time required for equatorial Rossby waves to propagate westward, reflect at the western boundary, and propagate eastward as equatorial Kelvin waves. Jin (1997a,b) developed a conceptual recharge model that embodies the delayed oscillator, but ignores the explicit role of wave propagation. This model combines the dynamics of the SST–wind interaction with the nonequilibrium response between the mean thermocline depth and the surface wind stress. Using a compilation of different in situ datasets including subsurface temperature, Meinen and McPhaden (2000) demonstrate that the concept developed by Jin (1997a,b) is in agreement with observations over the period 1980–99. While variations in SST, zonal wind stress, and the east–west slope of the thermocline are all nearly in phase within the equatorial band, the heat buildup leads each of these variables by an average of 7 months. These studies suggest that the equatorial heat buildup is an intrinsic element of El Niño.

The detailed mechanisms involved in the heat buildup in the western equatorial Pacific have received, paradoxically, little consideration since the pioneering work of Wyrtki (1975, 1985). The presence of meridional boundaries in the oceans, in addition to the stress of the trade winds, is regarded as the major reason for the accumulation of warm water. Zebiak (1989) argues that variations in heat buildup are dynamically rather than thermodynamically induced. He also shows that equatorial dynamics are more effective in controlling the evolution of heat content than the dynamics at higher latitudes and the dynamics of the western boundary region. The variability in the heat buildup prior to the development of El Niño has been used to explore the predictability of the tropical Pacific climate (e.g., Latif and Flügel 1991). By analyzing forecasts from a simple coupled model, Perigaud and Cassou (2000) found that the impact of westerly wind bursts (WWBs) on ENSO predictions also depends on the ocean heat buildup, especially during recent strong El Niño events. The relative importance of these different mechanisms in the development of an El Niño is still subject to debate. For example, it is not guaranteed that the buildup amplitude must reach a given threshold before an El Niño and even if this condition is necessary and/or sufficient. A detailed analysis of physical mechanisms involved in the buildup and a demonstration of their pertinence for El Niño prediction are yet to be established.

An increase in the strength of easterly winds in the central Pacific generates upwelling equatorial Rossby waves characterized by westward currents and by westward propagation. These waves are associated with a positive pressure perturbation that raises the sea level in the western Pacific. Li and Clarke (1994) confirm that a 3-month lag between the trade winds in the central Pacific and the sea level in the western Pacific is associated with the time required for Rossby waves to propagate into the region. They also noted that significant correlations between these phenomena could occur with lags as long as one year. Their result is consistent with the time evolution of positive sea level anomalies that has been described as a precursor mode for the onset of an El Niño event. This mode has been derived from different datasets and several statistical methods (e.g., Latif and Flügel 1991; Xue et al. 2000; Meinen and McPhaden 2000). A question that comes to mind is why local wind forcing or other local processes do not dissipate the buildup. In the western Pacific, observations have revealed that the mixed layer is insulated from deep and cold waters by strong salinity stratification within the warm isothermal surface waters (Lindstrom et al. 1987; Lukas and Lindstrom 1991; Sprintall and Tomczak 1992; Ando and McPhaden 1997). The difference in depth between the bottom of the isothermal layer and the bottom of the mixed layer is the barrier layer thickness (BLT). The barrier layer serves to inhibit the entrainment of cold water from below into the mixed layer. The effects of the barrier layer on the dynamics of the warm pool have been shown to be important during the onset of El Niño in a coupled model where the removal of the barrier layer was able to reduce or abort El Niño (Maes et al. 2002b). However, the role of the barrier layer during El Niño has not been fully investigated.

The main purpose of this study is to explore whether the salinity barrier layer is important in maintaining the heat buildup in the western Pacific prior to El Niño. The relationship between the barrier layer and the buildup could be important, as suggested in Fig. 1, which displays longitude–time plots, averaged between 2°N and 2°S, of SST and thermosteric anomalies together with an estimate of the BLT. Because of sparse observations of salinity profiles, the salinity variability is determined by an indirect approach that combines the temperature data of the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON) array, TOPEX/Poseidon sea level anomalies, and a set of vertical empirical orthogonal functions derived from conductivity–temperature–depth profiles (Maes and Behringer 2000). The thermosteric anomalies are computed by replacing the variable salinity profiles by their mean profile in the computation of dynamic height anomaly referenced to 500 dbar. These thermosteric anomalies are better suited than the standard dynamical height anomalies to study the heat buildup in the western Pacific because the halosteric contribution to dynamic height is not negligible and is of opposite sign to the thermosteric contribution (Maes et al. 2002a). The BLT is computed as the difference between the depth where the temperature differs from SST by 0.5°C and the depth of the mixed layer where the density differs from the surface density by 0.125 kg m⁻³. As shown in Fig. 1, the El Niño events over the period 1993–2002 are characterized by a warming in the east together with an eastward displacement of the warm pool as defined by the 28.5°C isotherm. Prior to these events, a buildup of warm water is evident and simultaneous with the presence of a large BLT in the western Pacific. A detailed analysis of this particular relationship requires basinwide information on both surface and subsurface ocean states that is currently impossible to assemble from observations. The outputs of general circulation models (GCMs) provide a useful alternative if such models are able to reproduce similar features to those shown in Fig. 1.

The rest of the paper is organized as follows. Section 2 starts with a description of the coupled model with an emphasis on the simulated mean state and the seasonal cycle for the control experiment. Section 3 examines the relationship between the model heat buildup and the BLT in relation to El Niño. The analyses are focused on a strong El Niño simulated by the model that will highlight the combined role of the heat buildup and the BLT prior to this event. The importance of the relationship between the heat buildup and the BLT is tested in two-stage perturbed experiments. In the first stage, the dependency of the ocean vertical mixing scheme on the salinity stratification is removed within the equatorial warm pool during a 1-yr period. This year corresponds approximately to the heat buildup period prior to the El Niño event. In the second stage, the salinity dependency in the vertical mixing scheme is restored and the experiment continues. The idea is to perturb the model only during the buildup year so that subsequent analyses of model variability can be tied unambiguously back to that period. Because the tropical atmosphere exhibits significant internal variability that could disrupt El Niño, six-member ensembles of both the perturbed and control experiments have been generated to test the robustness of the effects of heat buildup and the BLT on El Niño. The physical processes involved during the buildup period prior to El Niño are analyzed in section 4. Comparable experiments are run for two other El Niño events and their similarities with the main experiment are examined before presenting conclusions in the last section.

2. The coupled model and simulated variability

The coupled model used in this study consists of the Météo-France atmospheric general circulation model (AGCM) and of the oceanic GCM (OGCM) developed at the Laboratoire d'Océanographie Dynamique et de Climatologie (LODYC). The AGCM was adapted for climate studies from the ARPEGE forecast model and is described by Déqué et al. (1994). In this study, a spectral T31 triangular horizontal truncation is used that corresponds to a resolution of 3.75° in the Tropics. The vertical discretization uses a hybrid sigma-pressure coordinate with 19 levels extending up to 10 hPa. The physical parameterizations include the radiation scheme developed by Morcrette (1990); deep and shallow convection are based on the mass-flux scheme of Bougeault (1985) and on a simple modification of the Richardson number (Geleyn 1987). The different coefficients employed in the convection and cloudiness treatment are derived from the sensibility analyses detailed in Terray (1998).

The OGCM is based on the Océan Parallélisé (OPA) code (Madec et al. 1998), and the present version is a tropical Pacific basin version adapted by Maes et al. (1997). The domain covers the tropical Pacific between 120°E and 70°W with 1° zonal resolution and is closed at 30°N and 30°S. The latitudinal resolution varies gradually from 0.5° at the equator to 2° at the northern and southern boundaries. Poleward of the 15°N–15°S band and below the mixed layer, a linear restoring term damps the solution toward the seasonal-varying climatology of Levitus (1982). The model has 25 layers in the vertical with 10-m resolution in the upper 150 m. The maximum ocean depth is limited to 4500 m, but both bottom topography and coastlines are realistic. The solar radiation is allowed to penetrate the upper layers. The parameterization of vertical diffusion is based on a 1.5 turbulent kinetic energy closure scheme (Blanke and Delecluse 1993). The lateral diffusion uses a classical second-order operator whose coefficient corresponds to a medium value as discussed in Maes et al. (1997).

The coupling between the ocean and atmosphere models is achieved through the exchange, on a daily mean basis, of the fields of SST, surface wind stress, solar and net heat fluxes, and precipitation minus evaporation. The seasonal-varying climatology from Reynolds and Smith (1995) over the period 1979–88 is used to prescribe SST outside the OGCM domain. A smoothing procedure is applied to the SST near the poleward boundaries of the ocean model in order to adjust the net heat flux component exchanged between models. No flux correction is applied within the 15°S–15°N band and the coupling between the ocean and the atmosphere is completely free within this region. The initial conditions are obtained from independent forced integrations of each model and, because of the initial shock when the components are coupled, only the outputs from year 8 to year 40 are analyzed in the rest of this paper. During this period, the surface fields exchanged between the models do not exhibit significant long-term tendencies. The climate drift of the coupled model is relatively low with a negative tendency in the upper 300-m ocean temperature of less than 0.3°C over the 40 years (Belamari 2002).

Although ENSO modeling has made significant progress during the last decade, several intercomparisons of coupled GCMs have revealed that errors in SST could be large and systematic in the equatorial Pacific and that ENSO variability reproduced by models is often weak and/or incorrectly simulated. A complete description of the variability simulated by the present model is beyond the scope of this study, but it is important to verify that the mean state, the seasonal cycle, and the interannual variability of the key variables are consistent with the observations. The annual-mean simulated SST along the equator exhibits an east–west gradient with SSTs greater than 28°C in the western Pacific and SSTs around 26°C in the eastern Pacific. In this latter region, the model SST is characterized by a warm bias of about 1°C compared to observed climatologies, which is reasonable for models with coarse resolution (e.g., Mechoso et al. 1995). The annual-mean simulated zonal wind stress along the equator shows a more complex structure in its biases with stronger than observed wind stress in both the far western and the eastern Pacific and lower than observed wind stress in the central Pacific. These biases result in the underestimation of the southeastern trade winds in the Southern Hemisphere and in an overly strong penetration of the northeastern trade winds in the Northern Hemisphere (see Fig. 2 in Belamari et al. 2003).

The seasonal cycles of SST and zonal wind stress represent other key elements, as they serve to define the interannual anomaly. The left and center panels of Fig. 2 display the mean seasonal cycles of SST and zonal wind stress along the equatorial band. The main features of the SST observed seasonal cycle are well captured by the coupled model, with an annual harmonic that peaks in magnitude at around 100°W and displays a maximum development of the cold tongue in August– September. It must be noted, however, that the cold tongue does not persist long enough into the boreal winter and that the aforementioned mean warm bias in the eastern Pacific occurs during the warm season in April– May when the trade winds relax. In late summer, the strong development of the cold tongue induces a westward displacement of the warm pool that is too pronounced in the model. In the western Pacific, the waters warmer than 29°C are not always present throughout the year but this cold bias is less than 0.5°C, as shown by the permanent presence of waters warmer than 28.5°C westward of the date line (see the dark line in Fig. 2). The presence of these warm waters is necessary to trigger deep convection in the atmosphere as shown by the relationship between the SST and the outgoing longwave radiation (e.g., Webster et al. 1998). Despite the aforementioned differences, it is clear that the model captures the main features of the SST seasonal cycle quite realistically.

Similar conclusions hold for the zonal wind stress evolution, which shows the east-to-west phase propagation with a clear dominance of a semiannual cycle in the east between 180° and 120°W and an annual cycle west of the date line (Fig. 2b). The position of the southeasterly winds in late boreal winter is, however, located too far to the east and the strength of the easterly winds in summer is too strong as compared to the analyses of Yu and McPhaden (1999). In the western Pacific, the presence of westerly winds in winter is related to strong intraseasonal activity. It is important to note that this coupled model is able to naturally generate WWBs, whose the role in triggering a warm event was studied by Belamari et al. (2003).

The right-hand side of Fig. 2 displays the seasonal cycle along the equatorial band of the BLT defined by the same criteria as in Fig. 1. As expected, the BLT is significant only in the western Pacific and barely extends east of the date line. The largest values (around 20 m) occur west of the date line during the boreal winter. The variability in the west exhibits a clear annual cycle. The relationship between a significant BLT and the warmest waters (>28.5°C) is also evident in Fig. 2, in agreement with recent observations reported in Delcroix and McPhaden (2002). The eastward displacement of the barrier layer in boreal winter is associated with strong eastward currents generated by WWBs. This behavior is consistent with the present understanding of barrier layer formation along the equator (Roemmich et al. 1994; Vialard and Delecluse 1998; Cronin and McPhaden 2002). The eastern edge of the warm waters is also characterized by a strong sea surface salinity (SSS) front (not shown) that exhibits an ENSO-related variability in agreement with the results of Picaut et al. (1996) and Delcroix and Picaut (1998). Along the equator, the simulated SSS, with typical mean values less than 33.5 psu in the far western Pacific and around 35 psu in the central Pacific, presents a “fresh” bias as compared to the observed climatology (Delcroix et al. 1996). The excessive precipitation over the Maritime Continent and the weakness of the southeasterly trade winds in the central Pacific explain, in large part, these model deficiencies.

Figure 2 suggests that the main processes of the interaction between the SST and the wind are well captured by the model at seasonal time scales. Nevertheless, this does not guarantee that the level of interannual variability will be satisfactory (e.g., Mechoso et al. 1995). Maes et al. (2002b) have already noted the weak amplitude of typical indices such as the Southern Oscillation index (SOI) and SST anomalies in the Niño-3.4 box (5°N–5°S, 120°–170°W) in the central Pacific. However, the time series of such indices shows that the present model is able to reproduce ENSO events of different amplitudes characterized by a spectral peak at around 4 years, in agreement with observations. Most importantly, the level of interannual variability in SST is closely related to the wind stress anomalies. For example, the ratio between the variance of these variables is within 20% of the value derived from observations in the central Pacific by Davey et al. (2002).

3. Relationships between the buildup, the barrier layer, and El Niño

4. Physical mechanisms of the heat buildup–BLT relationship

The systematic disappearance of El Niño conditions following the reduction of the heat buildup highlights the importance of its relationship with the barrier layer during the preceding year, but it also raises some questions about the physical mechanisms at work. It should be kept in mind that, although the modifications are applied only in the equatorial western Pacific, the response is basinwide.

Figure 8 exhibits the response of the model for both control and perturbed experiments during the 1-yr period when the modifications to the mixing scheme were applied. The SST field, the zonal wind stress, and the anomaly of the 20°C isotherm depth are displayed together with the 34.25-psu surface isohaline (white line). The 20°C depth anomaly is used to summarize the thermocline behavior but it is also strongly related to the heat content and to the sea level (e.g., Rebert et al. 1985). The 34.25-psu SSS value represents a useful way to identify the salinity front, which is associated with the zonal displacements of the eastern edge of the warm pool (Picaut et al. 1996). Hereafter, the term “warm pool” will refer to the region extending from the western boundary to the longitudinal position of this SSS value. In the control experiment, a series of WWBs appears over the warm pool from November of year 14 to the end of April of year 15 (Fig. 8, top). These westerly winds maintain locally the warm SSTs through Ekman convergence in the upper layers and the generation of downwelling equatorial Kelvin waves. The propagation of the Kelvin waves is shown in the 20°C depth anomaly. If the first series of WWBs results in an eastward displacement of the warm pool, only the waves associated with the second series in January–February of year 15 significantly depress the thermocline and propagate to the eastern part of the basin by April. At that time, positive thermosteric anomalies have already spread throughout the entire equatorial band (Fig. 3). The Kelvin waves associated with the third series of WWBs in late April of year 15 maintain the depression of the thermocline, but more importantly, they arrive in July–August at the beginning of the upwelling season in the east (Fig. 2). Figure 3 shows that this period corresponds to the switch from negative to positive SST anomalies. Around this period, the equatorial band is fully charged and the conditions are favorable for the development of a warm event during the next boreal winter. In the control experiment, the development of such conditions is clearly associated with the generation and timing of downwelling equatorial Kelvin waves.

In the perturbed experiment (Fig. 8, bottom), the relationship between the heat buildup, the Kelvin waves, and the seasonal cycle is clearly different, especially in the central and eastern part of the basin. The period from October of year 14 to January of year 15 is still characterized by a series of WWBs over the warm pool but, while the intensity of the bursts is similar to the control experiment, the fetch of the WWBs extends farther along the equator. In November of year 14 for instance, westerly winds blow over the western Pacific out to the date line, a limit that is far beyond the eastern edge of the warm pool. It should be noted that the eastward displacement of the warm pool itself is not significantly different when compared to the control experiment. Within the warm pool, it takes less than 3 months to suppress the positive anomalies in the 20°C depth isotherm, and the absence of variability in these anomalies indicates that the seasonal conditions prevail during the rest of the perturbed experiment. Associated with the different wind stress conditions, the response of the ocean is different as well in the central and eastern parts of the basin. Downwelling equatorial Kelvin waves are generated earlier in the year and due to the extended wind fetch they propagate throughout the equatorial band. By the end of January of year 15, the thermocline has been depressed by 10–20 m in the eastern Pacific. At the end of April the heat buildup is mostly gone and it will not recover. A few WWBs still appear in March and April, but it is clear that they do not generate strong Kelvin waves and, consequently, they do not influence the central and eastern Pacific. Compared to the control experiment, the timing of the Kelvin waves is different in the perturbed experiments and the coincidence of their arrival with the development of the upwelling season is not so marked. Starting in July of year 15, the upwelling season in the perturbed experiment is more pronounced (Fig. 8, left) and closer to the mean seasonal conditions in terms of westward extension and intensity than in the control experiment (Fig. 2). These contrasting results suggest that the sensitivity of the coupled response to the removal of the barrier layer is strongest from October to March and that the dynamics of the warm pool has an important influence during this period. It justifies a posteriori the choice of the 1-yr period for the first stage of the perturbed experiment in order to test the effects of the heat buildup–BLT relationship on the development of El Niño.

Figure 8 shows that the series of WWBs in the perturbed experiment is associated with the presence of SSTs warmer than 29°C within and beyond the warm pool. In order to investigate these SST differences, the heat budget of the mixed layer in two regions of the 2°N–2°S band is considered. The first region is the warm pool as defined by the 34.25-psu SSS criterion previously mentioned; the second region is located in the east of the warm pool and extends 25° eastward from the 34.25-psu isohaline. Tables 1 and 2 give different parameters of interest averaged over these two regions from October to December of year 14. For the mixed layer heat budget, the methodology consists of integrating separately the terms in the equation of temperature over the time-variable depth of the mixed layer. The total tendency of the budget results from the contributions of the horizontal and vertical advections, the combined effect of shortwave radiation and surface heat flux (referred to hereafter as atmospheric forcing), and the entrainment of water from below associated with the turbulent mixing of heat. The horizontal diffusion always represents a sink of heat, but its amplitude remains negligible when compared to the dominant terms.

In the warm pool region of the perturbed experiment, the BLT is reduced by a factor of 3 and the mixed layer depth consequently increases (Table 1). As expected, the reduction of the BLT induces a strong modification of the vertical heat transfer. In the control experiment, the water parcels exchanged at the bottom of the mixed layer have the same or an even warmer temperature, resulting in a positive source for the heat budget. In the perturbed experiment, the entrainment cooling of the mixed layer is operative and becomes the dominant term of the heat budget (Table 2). The process that opposes this cooling is provided by the atmospheric heating, which acts as a thermostat (0.28 against −0.38). The positive contribution of the atmosphere is linked to the increase in depth of the mixed layer that is thus able to gain more solar heat flux. It is also related to the importance of the atmospheric convection over the warm pool. The projection of wind stress over a deepening mixed layer results in reduced surface currents and a slight reduction in the advective contribution to the heat budget. Finally, although the total tendency of the heat content is reduced due to the deepening of the mixed layer, it remains positive, and this largely explains the presence of anomalously warm SST in the perturbed experiment (Fig. 8; Table 1).

Tables 1 and 2 show that quite similar dynamics operate in the eastern region for both the perturbed and control experiments. The differences are small and are related to the dynamics of the warm pool. For instance, the SST is slightly larger in the perturbed experiment and it results mainly from the stronger contribution of advection to the mixed layer heat budget. The main terms of the advection are the zonal and meridional components, and their amplitudes are related to the stronger zonal fetch of the wind initiated over the warm pool. The larger fetch of the wind, resulting from its coupling with the warm SST, induces a larger momentum supply into the ocean (Table 1). This coupling induces not only a local response in the convective activity of the atmosphere but also a response in the planetary-scale divergent circulation. In the perturbed experiment, the strong convective center is displaced eastward by 20° to 40° of longitude compared to the control experiment and explains the positive equatorial differences near the date line region shown in Fig. 9. In this region, this corresponds to a difference of order to 30% in the 1000-hPa potential velocity field.

5. Discussion and conclusions

These analyses of oceanic and atmospheric responses to the removal of the barrier layer in the coupled model confirm the importance of the warm pool dynamics for El Niño. The switching on or off of the entrainment cooling associated with the barrier layer is responsible for mixed layer changes that, in return, control SST changes. The high sensitivity of the atmospheric response to small SST changes over the warm pool results in local and basin scale oceanic responses. During the particular period of heat buildup, the deepening of the mixed layer associated with the removal of the barrier layer and the net surface heat flux results in the maintenance of warm SST. It favors the subsequent zonal extension of the fetch of the WWBs. The enhanced vertical mixing below the mixed layer dissipates one part of the heat buildup in the western Pacific, whereas downwelling Kelvin waves generated by WWBs insure the discharge of the remaining excess of heat toward the eastern Pacific. The failure of El Niño to develop also confirms the necessary condition of the heat buildup for the development of El Niño.

Our presentation so far has focused on a particular El Niño in order to detail the physical mechanisms at work. It is necessary to examine similar experiments on other El Niño events in order to gain confidence in the results and to isolate the most important features. The El Niño events of year 11 and year 22, as indicated by the arrows in the bottom of Fig. 4, may be useful in this regard. These events are characterized by different intensities and by different preconditions, as compared to the strong event of year 16 (Fig. 4). For example, while the warm event of year 22 is also preceded by a cold event, the buildup prior to the event of year 11 is a period of near-zero SST anomaly. These two El Niño events were explored by means of perturbed experiments with mixing scheme modifications during a first stage covering, respectively, the 09/10–10/10 and 20/10– 21/10 one-year periods. During these periods, the BLT in the western Pacific in the control experiment is smaller with values around 35 m as compared to 46 m for the strong event. Nevertheless, the different mechanisms at work during both perturbed experiments are not fundamentally changed. The net result of removing the barrier layer's influence on vertical mixing is to reduce the heat buildup by a factor of 2 to 4 in the western Pacific and to discharge the heat toward the eastern Pacific before the onset of El Niño. This decrease occurs in the thermocline of the equatorial band. The increase in the mixed layer depth, the maintenance or even slightly increase of warm SST within and eastward of the warm pool, as well as the thermostat role of the atmosphere are also confirmed. The coupled response of the warm pool is characterized by a wind fetch extension that generates a remote response in the rest of the basin, through downwelling equatorial Kelvin waves. At the end of each buildup period (i.e., the first step of the perturbed experiments), six-member ensembles are generated following the same methodology as discussed in the strong El Niño (section 3c). The annual mean SST anomaly along the equator for each ensemble, control and perturbed, is displayed in Fig. 10. For both El Niño events (years 11 and 22), the mean difference is significant over most of the central and eastern equatorial band. Three members of the perturbed ensemble for year 11 do show some warm anomalies associated with an east–west tilting of the thermocline (not shown). These three members, as well as one member associated with the perturbed ensemble of the strong event (Fig. 7), have a mean SST anomaly larger than 0.5°C. These four members could be thus identified as El Niño events. However, these 4 members are a small fraction of the total of 18 members, and the suppression of the warm events due to the modifications of the relationship between the heat buildup and the BLT is considered to be significant.

It is important to use fully coupled models to explore the feedbacks between the barrier layer, the mixed layer, and the SST. Maes et al. (2002b) have recently investigated the impact of the barrier layer on the onset of El Niño through similar perturbed experiments. In both that study and the present one the barrier layer was removed, enabling entrainment cooling and deepening of the mixed layer. However, the timing and duration of the perturbed stages of the experiments were chosen differently in these two studies. In Maes et al. (2002b), the removal of the barrier layer was done during the series of WWBs just before the onset of El Niño (starting in October of year 15) whereas, in the present study, this removing is done during the 1-yr buildup period prior to the onset of El Niño (from October of year 14 to September of year 15). The results show different effects on the heat budget of the mixed layer and on the SST. Maes et al. (2002b) reported a slight decrease in the warm SST as compared to the maintenance of the warm SST of the warm pool in the present study. This difference in SST is mainly explained by the difference in the westerly wind fetch and in the overall response of the coupled system at the basinwide scale. Because of this sensitivity to the timing of the presence of the barrier layer, it would be worthwhile to conduct similar barrier layer experiments with other coupled models, which may have different mean states and different levels of interannual variability.

An understanding of processes that regulate SST in the western equatorial Pacific was one major motivation of the 1992–93 Coupled Ocean–Atmosphere Response Experiment (COARE). On the list of potential feedbacks that were uncertain at the time of the COARE implementation, Webster and Lukas (1992) mentioned the possibility that the salinity barrier layer processes could provide a long-term memory mechanism between El Niño events. The present study offers a variation on that idea in that it establishes a precise role for the barrier layer in the heat buildup prior to El Niño events.

As previously stated, the relationship between the heat buildup and the barrier layer appears to be a necessary condition to maintain the heat buildup in the equatorial Pacific over a 1-yr period. The heat buildup then provides the potential for an El Niño to develop. Of course, it does not establish that the relationship is sufficient because other processes could also be necessary for SST changes in the warm pool independent of the barrier layer effects. In order to evaluate the importance of the barrier layer in the intrinsic variability of the system, an additional experiment was performed. The modifications in the vertical mixing scheme were applied in the same manner, that is, in the 4°N–4°S band and over waters warmer or equal to 28°C, but they were continuously applied during a 30-yr period. This simulation exhibits a mean state and a seasonal cycle similar to those in the control experiment, but it also supports a weak ENSO variability. It suggests the idea that the heat buildup–salinity barrier layer relationship is not a sufficient condition for El Niño, but it also seems to suggest that this condition is not necessary either. A closer look at the results of the additional experiment can resolve the apparent contradiction. Recall that the modifications in the vertical mixing scheme used in the present study can only begin to reduce or to remove the barrier layer once it has been formed. Some residual BLT values of about 10 m are present in this additional simulation and they persist for a few weeks before being destroyed. This short period represents, however, a sufficient time to sustain the heat buildup in the western central Pacific, so the model can develop weak to moderate El Niño events. To further extend the analysis of the role of the barrier layer, it will be important to fully consider the mechanisms that lead to its formation such as discussed by Roemmich et al. (1994) and Cronin and McPhaden (2002). The coupled response of the Pacific Ocean in a simulation where one or several of such mechanisms are suppressed has not yet been explored and further investigations are required.

To conclude, there is more and more evidence that the salinity barrier layer plays an important role in the dynamics of El Niño. The relationship between the heat buildup and the barrier layer over long periods of time represents a necessary condition for a sustained accumulation of heat in the western Pacific. In this region, the heat buildup is a direct consequence of the suppression of the entrainment cooling below the mixed layer due to the presence of the barrier layer. The subsequent development of El Niño in the central and eastern equatorial band is the remote response to Kelvin waves generated by the SST–wind coupling over the warm pool. The removal of the barrier layer in the western Pacific causes the reduction and discharge of the heat buildup in the western Pacific before the onset of El Niño and thus prevents the development of El Niño. These results are an argument for the development of real-time in situ observations of the barrier layer both for assessing the applicability of these results and potentially for providing observations to initialize El Niño forecasts.