a. Idealized forced experiments

The ocean model used in this study is identical to that used in Karnauskas et al. (2007): the Gent and Cane (1989) reduced gravity, primitive equation model of the tropical oceans with a hybrid vertical mixing scheme (Chen et al. 1994). The Chen et al. (1994) scheme accounts for mixed layer entrainment/detrainment, shear flow instability, and free convection in the thermocline by combining the physics of the Kraus and Turner (1967) mixed layer model with the Price et al. (1986) dynamical instability model. Surface fluxes are calculated interactively by coupling the OGCM to a thermodynamic atmospheric mixed layer (Murtugudde et al. 1996), thus allowing for feedbacks between SST and surface fluxes. The OGCM is structured vertically on sigma coordinates and includes a mixed layer plus 19 subsurface layers. The zonal boundaries of the model grid are 30°N and 20°S, along which a sponge layer is used (see Chen et al. 1994). Meridional boundaries are represented by the approximate coastlines of Asia, Indonesia, and Australia in the western pacific (Indonesian Throughflow is closed off) and the Americas in the eastern Pacific. All experiments are run on a grid with higher resolution in the eastern tropical Pacific: zonally stretching from ¼° in the east to ∼1° at the western boundary, and meridional stretching from ¼° near the equator to ∼1° at 20°S and 30°N. Since the OGCM is a reduced gravity model, all topography is constant in the vertical. Thus, there are no realistic continental shelves, implying that the experiments including the Galápagos essentially represent the islands as a simple vertical barrier extending from the ocean floor to the sea surface. However, based on subsurface bathymetric measurements (Fig. 1), this is a reasonable approximation in the upper few hundred meters of the ocean, where most of the interesting features of the equatorial circulation are at play.

In the forced setup, two experiments were performed—designed to reveal the total effect of the Galápagos Islands on an idealized El Niño event. Prior to both experiments, the model was spun up from the climatology of Levitus and Boyer (1994) for 60 years. The first of the two forced experiments, NoG_F, was initialized with conditions that were produced by spinning up the model with the Galápagos excluded from the model grid. Thus, the mean state in NoG_F is the mean state that would exist if the Galápagos Islands did not exist. The model was then integrated for two years, saving 10-day mean fields. The second of the two forced experiments, G_F, was initialized with initial conditions produced by spinning up the model with the Galápagos Islands included in the model grid. Thus, the mean state in G_F is the appropriate mean state accounting for the existence of the Galápagos. As mentioned in section 1 and explained in detail in Karnauskas et al. (2007), the Galápagos mean state contains a reduction of the equatorial cold bias over the non-Galápagos mean state. The model was then integrated for two years, saving 10-day mean fields.

Each forced experiment (e.g., NoG_F) comprises two model runs: one forced by climatological forcing and one forced by climatological forcing plus a westerly wind stress anomaly designed to induce a reasonable El Niño event. Thus, anomalies for each experiment are computed as the difference between the perturbation and climatological runs of that experiment. A westerly wind stress anomaly of 0.072 N m⁻² (equivalent to about 6 m s⁻¹) was added to the climatological zonal wind stress over a box centered on the equator in the western Pacific (maximum at the equator tapered outward to 8°N–8°S, beween 160°E and 160°W) from September through November of the first year of integration. The location, magnitude, timing, and duration of the idealized wind stress anomaly are reasonably similar to those preceding significant El Niño events in recent observed history.

Climatological forcing is derived from ECMWF operational analysis surface wind stress, 2° spatial resolution, base period 1985–2003 (Bengtsson et al. 1982); Xie and Arkin (1996) precipitation, 2.5° spatial resolution, base period 1979–2003; International Satellite Cloud Climatology Project (ISCCP) cloud cover, 280-km spatial resolution, base period 1983–94 (Rossow and Schiffer 1991); and Earth Radiation Budget Experiment (ERBE) shortwave radiation, 2.5° spatial resolution, base period 1984–90 (Barkstrom 1984).

Two distinct types of ENSO modes, which can be identified by the direction of propagation of the SST anomaly, are the SST mode (or S mode), which exhibits westward propagation and involves local SST–wind interactions, and the thermocline mode (or T mode), which exhibits eastward propagation and involves remote wind–thermocline interactions (Guilyardi 2006). In that light, the forced El Niño events analyzed in this study are of the T-mode variety, although it is understood that ENSO events in nature are not as straightforward. Since the purpose of this study is, not to understand the genesis of ENSO events but, to quantify and understand the effect of the Galápagos Islands on key ENSO characteristics, we suggest that the method used to initiate El Niño events in the forced experiments is of secondary importance. Furthermore, the amplitude of the idealized El Niño event produced by the model is arbitrary. The greater the westerly wind stress anomaly we impose, the higher the amplitude of the ensuing El Niño event. Again, the goal is, not to accurately simulate a specific El Niño event as seen in observations but, rather, to study characteristics that are dependent on the Galápagos Islands.

b. Hybrid coupled experiments

For the second set of experiments, we use an identical model setup as that of the idealized forced experiments with the exception that the ocean is coupled to the zonal wind stress field. The coupling is similar to that used in Sun et al. (2004) wherein the zonal wind stress anomaly is proportional to the zonal SST gradient in the equatorial Pacific:

View Expanded

where τ_x is zonal wind stress, λ is longitude and φ is latitude, τ_xclim is the monthly climatological mean zonal wind stress, μ is the coupling coefficient, ΔT is the difference between area-averaged SST between the western (5°S–5°N, 130°E–180°) and eastern (5°S–5°N, 130°–80°W) equatorial Pacific Ocean, and ΔT_clim is the monthly climatological mean ΔT. Thus, a positive SST anomaly in the eastern equatorial Pacific will result in a positive (westerly) zonal wind stress anomaly. In our model, we use a coupling coefficient of μ(0) = 0.0065 N m⁻² K⁻¹, decreasing with cosφ, diminishing at 15°N and 15°S. This value of μ(0) is reasonably close to the range suggested by Sun et al. (2004).

Two 75-yr simulations were produced, with the sole difference being the existence of the Galápagos Islands in the model grid. Experiment NoG_C was initialized with the spunup state without Galápagos (the same initial conditions as those used in idealized forced experiment NoG_F) and integrated forward with the island absent from the model grid. Experiment G_C was initialized with the spunup state that included the Galápagos Islands (the same initial conditions as those used in idealized forced experiment G_F) and integrated forward with the Galápagos present in the model grid. After removing the brief spinup period due to the introduction to the SST–wind stress coupling, 68 years of output are available for analysis.

a. Idealized forced experiments

In this brief subsection, we examine the effect of the Galápagos Islands on an idealized El Niño event and the mechanisms responsible for the differences. We first examine indices that highlight the ocean response at key locations in the tropical Pacific Ocean. Niño-4 (5°S–5°N, 160°E–150°W) represents the anomaly in the west-central equatorial Pacific, Niño-3 (5°S–5°N, 150°–90°W) in the eastern Pacific, and Niño-1 + 2 (10°S–0°, 90°–80°W) in the coastal upwelling region off the west coast of South America but east of the Galápagos Islands.

Area-averaged SST anomalies were computed for each of the aforementioned Niño indices and are presented in Fig. 2. According to Fig. 2, the peak SST anomaly in the Niño-3 region is reduced by approximately 0.5°C by the Galápagos, or a 15% reduction in the anomaly over experiment NoG_F. The reduction of the anomaly in Niño-4 is small. In Niño-1 + 2, however, the SST anomaly is reduced by approximately half. The lag between the three Niño indices is relatively unchanged between NoG_F and G_F, with both experiments exhibiting Niño-4 leading, followed by Niño-3, and finally Niño-1 + 2.

Given that the mean state plays a dominant role in the overall effect of the Galápagos on ENSO amplitude, it is reasonable to guess that subsurface physics are important in explaining the overall effect of the Galápagos on ENSO owing to the efficiency of entrainment/mixing to affect the mixed layer heat budget and SST depends strongly on the thermocline. The complete mixed layer heat budget is given by

View Expanded

where Q_t is the rate of heat storage (ρC_pH∂T/∂t), and so on. The mixed layer heat budget anomalies throughout the evolution of the idealized El Niño events are shown in Fig. 3. In both experiments, the processes responsible for the warming are Q_ZA and Q_EMX, with Q_MA also making a modest contribution and the surface heat fluxes acting to damp the anomaly. That Q_ZA and Q_EMX are the dominant terms contributing to the positive Q_t and that the net surface heat flux acts against the warming is consistent with a calculation of the interannual heat budget from in situ observations (Wang and McPhaden 2000). However, it is the Q_EMX term that is different between the two experiments. In experiment NoG_F, the peak Q_EMX anomaly is 55 W m⁻², compared to 42 W m⁻² in experiment G_F. On the mean, entrainment/mixing is a process that cools the mixed layer. Thus for an El Niño event, the entrainment/mixing anomalies are positive; entrainment/mixing is cooling the mixed layer less. This implies that entrainment/ mixing is the process primarily responsible for the difference in the amplitude of the SST anomaly between experiments NoG_F and G_F and that the anomalous net heat flux to the atmosphere, primarily by Q_LAT, is reduced as well.

b. Hybrid coupled experiments

In this section, we discuss the effect of the Galápagos Islands on internal ENSO variability with the ocean model coupled to the atmosphere through zonal wind stress. Specifically, we are interested in any changes in the time scale of ENSO, as the idealized experiments served to show that, given an identical perturbation in the wind field, the amplitude of the SST response should be damped in the case with the Galápagos Islands. Recall that the sole difference between hybrid coupled experiments NoG_C and G_C is the existence of the island in the ocean model grid. It is well known that ENSO events tend to peak in boreal winter, with Niño-3 SST anomaly (SSTA) variance peaking strongly in November and December. In this regard, both the NoG_C and G_C experiments are in excellent agreement with observations (not shown).

The Niño-3 SST time series from both experiments, with the seasonal cycle removed, are shown in Fig. 4 (top). Also shown in Fig. 4 (bottom) are the power spectra computed by discrete Fourier transform (DFT) for Niño-3 SSTA in experiments NoG_C, G_C, and 68 years (1935–2002) of the NOAA Extended Reconstructed SST version 2 dataset (Smith and Reynolds 2004). To avoid spurious spectral features (i.e., those of period close to the length of the time series), we only show power spectra from the fourth harmonic and beyond (i.e., period not longer than one quarter the length of the time series). Examining the power spectrum for the NoG_C experiment, there is one dominant spectral peak, which is found at the 2-yr period. This can be confirmed intuitively by counting the number of warm events in any 10-yr interval of the NoG_C Niño-3 time series; there will be exactly 5. This is classic “biennial ENSO” behavior exhibited by some ocean and coupled climate models. Such strictly biennial regularity is not realistic, as ENSO in nature at present is neither perfectly regular nor significantly biennial. In our coupled model simulations, the Galápagos Islands have the effect of significantly reducing the biennial peak and giving rise to a preferred time scale near 3 yr. Moreover, a closer look at the G_C Niño-3 time series reveals that the system also has the ability to transition into and out of a biennial ENSO regime; between years 20 and 40, the ENSO periodicity in the G_C experiment is nearly biennial. It is also interesting to note that during, that period, the amplitude is smaller than surrounding periods. In NoG_C, ENSO is strictly biennial throughout with relatively consistent amplitude.

As evident by comparing simulated and observed power spectra (Fig. 4, bottom), the overall effect of reducing the biennial peak and increasing spectral power at lower interannual frequencies represents considerable improvement in the representation of ENSO, even within a relatively simple coupled context. An alternative technique for computing power spectra, which readily lends itself to understanding the statistical significance of spectral peaks, is the multitaper method (MTM) (Ghil et al. 2002). From the MTM method, the basic features of the NoG_C and G_C Niño-3 power spectra are very similar to those resulting from DFT (Fig. 4). In the NoG_C Niño-3 power spectrum, the 2-yr peak is the only interannual peak significant at the 99% confidence level. In the G_C power spectrum, a broad swath of three peaks (from ∼3 to 1.5 yr) is significant at the 99% confidence level.

To examine why the Galápagos Islands leads to a shift in the ENSO time scale from biennial to quasi quadrennial, we begin by analyzing composite ENSO events. We perform an empirical orthogonal function analysis on model SST for the 68 years of output, which is presented in Fig. 5. The only real mode possible in our free-running hybrid coupled model is ENSO, as there is no interannual forcing being applied to the system that would justify otherwise. The first EOF of SSTA in both NoG_C and G_C experiments represents a fully mature ENSO event, portrayed in its warm phase in Fig. 5. This is evident by the fact that the first principal component (PC1) and the Niño-3 index in each experiment are correlated >0.99 for years 42–68 (Fig. 5, bottom left). Despite the requirement of orthogonality, an EOF analysis can lead to “propagating” modes, that is, individual statistical modes representing different aspects or phases of a single phenomenon. The second (and third, in the case of G_C) mode of SSTA represents the transition between positive and negative phases of ENSO, that is, the warming period preceding an El Niño event and cooling that follows. The warming and cooling periods surrounding mature ENSO events are akin to the process described as recharging and discharging heat from the equatorial Pacific (e.g., Jin 1997); these secondary modes of SSTA variability could be said to correspond with the recharge–discharge process. Spatially, there are only minor differences in SST modes between experiments NoG_C and G_C. The exception is a break in the EOF loading at the western edge of the Galápagos Islands, which is found farther toward the coast of South America in the NoG_C case.

Temporally, differences in the EOF analysis shed light on an important distinction in the nature of ENSO in the coupled experiments. In both experiments, the secondary principal components (PC2) are maximum while the PC1/Niño-3 signal is growing, and minimum while the PC1/Niño-3 signal is decaying (again, because they represent the warming and cooling phases). However, in the G_C experiment, the positive phases in the PC2 are smaller in magnitude and longer lasting, consistent with a slower growth of PC1/Niño-3 signal. Conversely, the PC2 signal in the NoG_C experiment grows as rapidly as it decays, and does not persist in either phase. Hence, the ENSO events themselves in the NoG_C experiment grow and decay similarly, resulting in a shorter period than in the case with the Galápagos Islands. The present discussion is equally applicable to model sea level anomaly (SLA), as in Fig. 6. The total SLA is subject to the effects of propagating Kelvin and Rossby waves characteristic of the delayed oscillator paradigm of ENSO, and the heat content changes associated with the recharge–discharge process of ENSO, particularly in the eastern equatorial Pacific where the broader thermocline variations are large (e.g., McPhaden and Yu 1999). As with SSTA, the leading PC of SLA corresponds with the phase of ENSO in both experiments (Fig. 6, bottom), but the warming phase is slower evolving in the G_C experiment.

To examine the differences in the time evolution of composite ENSO events between the two hybrid coupled experiments, Fig. 7 portrays the composite (over years 42–68) evolution of Niño-3 SSTA, SLA PC, SSTA PCs, as well as the time rate of change of the Niño-3 mixed layer heat content anomaly d(HC′_ml)/dt and the zonal wind stress anomaly averaged from 5°S to 5°N. The basic message is that Niño-3 SSTA, the sea level pattern, SST PC1, and zonal wind stress anomaly track very closely but grow much more rapidly in the NoG_C experiment. This is consistent with the fact that the maxima in the secondary SSTA mode, which represents the warming preceding El Niño events, and the change in mixed layer heat content itself, are shorter lived in the NoG_C experiment and essentially mirror the minima following the event. In the G_C case, the warming mode persists longer and changes the mixed layer heat content more gradually (i.e., a smaller rate of change over a longer period of time). A direct comparison of the composite SST anomaly in the eastern equatorial Pacific is shown in Fig. 7c and reveals a slower growth but, ultimately, the same amplitude. When this slower and more subtle growth of the warm anomaly is compared to a composite of observed El Niño events (Fig. 7d), it is clearly an improvement in the representation of ENSO evolution. The mechanisms behind the slower growth of warm ENSO events in the hybrid coupled model, which lead to a longer preferred time scale for ENSO, and how they relate the Galápagos Islands are discussed in the following subsection.

c. The delayed oscillator and Bjerknes feedback

In this subsection, we discuss the mechanism for the change in ENSO time scale. There are several important theories for the oscillatory nature of ENSO. The concept of the delayed oscillator (DO) mechanism as proposed by Suarez and Schopf (1988) implies that the ENSO period is set by the basin crossing times of the equatorially trapped oceanic waves. Strictly within the context of the DO mechanism and its constituent equatorial waves, there are two ways by which the introduction of the Galápagos Islands to the hybrid coupled model could lead to a change in the ENSO period: 1) the speed of the Kelvin or Rossby waves change, thereby changing the basin crossing times or 2) the Galápagos Islands changes the effective width of the ocean basin by acting as an effective eastern boundary, thereby changing the basin crossing times of the equatorial waves. The latter is not a plausible explanation for two reasons: While the Galápagos Islands are wide enough in the meridional direction to significantly obstruct the EUC and cause a shift in the mean state (Karnauskas et al. 2007), they are not wide enough to cause any appreciable reflection of the Kelvin wave. Second, if the Galápagos Islands were wide enough to act as an effective eastern boundary and reflect Kelvin waves, this would result in shorter basin crossing times for the Kelvin wave and translate into a shorter ENSO period—the opposite of which is the case.

The other possible explanation for the Galápagos Islands leading to a change in the period of ENSO within the DO framework—that the basin crossing time is changed because the speed of the waves are changed—is also not plausible. Examining the solution to the equatorial Kelvin wave (Gill 1982),

View Expanded

where

View Expanded

u_Ke is the geostrophic zonal current perturbation, ′ is reduced gravity (the acceleration due to gravity, g*, reduced by the percent difference between upper and lowerlayer density), β is the beta coefficient (the change in the Coriolis parameter with latitude), G is a sinusoidal function, and H is the fluid depth. Equatorial Kelvin waves have approximately the dispersive characteristics of ordinary shallow water gravity waves (Matsuno 1966; Holton 1992) and, thus, travel at a speed proportional to the square root of depth H [Eq. (4)]. In the case of a Kelvin wave, H is the effective upper-layer depth, which we will approximate using the depth of the thermocline. To examine the dependence of zonal current u_Ke and phase speed c on the upper-layer depth H, we simply graph the solutions to Eqs. (3) and (4) using reasonable values for g′ (0.003g*, where g* = 9.81 m s⁻²), and β (2.3 × 10⁻¹¹ m⁻¹ s⁻¹) (Gill 1982). We also drop the G function from Eq. (3) and use y² = 1 m. The results, shown in Fig. 8, illustrate how a deeper H will lead to a faster, yet damped, Kelvin wave signal. Karnauskas et al. (2007) demonstrated that the introduction of the Galápagos Islands into an OGCM leads to a change in the equatorial mean state, specifically a deeper thermocline and mixed layer. Thus, in the case of G_C, we would expect a greater H and faster Kelvin wave phase speed. This is, indeed, the result in both the idealized forced and the hybrid coupled experiments: Kelvin wave speed is slightly higher in the simulations that included the Galápagos Islands in the model. Referring to the waves generated in the forced mode, Fig. 9 shows that the Kelvin wave generated by an identical wind stress perturbation is faster in the experiment G_F than NoG_F (as evident by the positive difference at the leading edge of the Kelvin wave signal) and damped (as evident by the negative difference in the core of the Kelvin wave signal). Thus, the change in Kelvin wave crossing time cannot explain the change in ENSO period. Once again, the period of ENSO would shorten, as it would have if the Galápagos were an effective eastern boundary to Kelvin waves.

Rather than explain the change in ENSO period as a result of the Galápagos Islands from the standpoint of the DO set by the wave speed the waves, we consider the behavior of the fully coupled system and its response to Kelvin waves under different mean states, particularly the role of the Bjerknes feedback (Bjerknes 1969), while invoking aspects of other leading theories for the oscillatory nature of ENSO, including the recharge–discharge mechanism (Jin 1997). From Eq. (3) and its solution, illustrated in Fig. 8, the solution for the zonal geostrophic current associated with an equatorial Kelvin wave is damped in the presence of a larger H, which is the case in the G_C mean state. Whether an equatorial Kelvin wave is of the downwelling or upwelling variety is defined by the u_Ke field; a convergent (divergent) u_Ke field results in downwelling (upwelling) signal propagating along with the wave energy. Similarly, a weaker convergence of u_Ke imparts weaker downwelling and therefore results in a smaller SST anomaly. This, in part, helps to explain the damped SST anomaly response to identical wind stress perturbations in the idealized forced experiments. However, in the coupled context, this has further implications for the growth of the SST anomaly in time.

To analyze the behavior (amplitude and temporal characteristics) of Kelvin waves in the hybrid coupled model experiments, we compute the projection of Kelvin waves onto the geostrophic current anomaly. Figure 10 (top) shows the ratio of the variability of the Kelvin wave signal along the equator from 180° to 100°W in experiments NoG_C and G_C. Also shown in Fig. 10 (bottom) is the difference in the mean state (equatorial temperature) without the Galápagos Islands (annual mean). Because the interannual variability inherent to the coupled experiments would be aliased into any long-term mean calculations, we rely on the difference in annual means derived from the climatological runs of the idealized forced experiments. As expected from theoretical considerations (Fig. 8), the Kelvin wave variability is damped by up to 28% by the Galápagos, which appears to be stronger whereas the thermocline difference is greater (e.g., note similarities at 105° and 125°W). Note that the standard deviation is dependent only on amplitude and not frequency: thus, the result in Fig. 10 is independent of the fact that there are fewer full ENSO cycles in the G_C case. East of the Galápagos Islands, the thermocline is much deeper in the Galápagos mean state. The Kelvin wave may therefore become more damped in that region, although this cannot be confirmed due to problems with calculating projections near a concave coastline.

Using theoretical calculations and analysis of idealized forced and hybrid coupled experiments, we have shown that the equatorial Kelvin waves involved in ENSO in experiments that included the Galápagos Islands are damped. We have also shown that the longer ENSO period in the G_C simulation is a result of a longer, more gradual warming period leading to the peak El Niño event. The key question is thus: Why is there slower anomaly growth and hence longer ENSO period with the Galápagos Islands included? The answer is directly related to the fact that the SST anomaly imparted by the Kelvin wave is damped initially and relies on the Bjerknes feedback for growth, which is easily extracted from our model given the simplified ocean–atmosphere coupling scheme. The feedback hypothesized by Bjerknes (1969) is that an SST anomaly in the eastern equatorial Pacific leads to a positive zonal wind stress anomaly, which alters the thermocline and positively feeds back onto the SST anomaly. From the coupling mechanism [Eq. (1)], the rate of growth of the SST anomaly beyond the initial SST perturbation by the downwelling Kelvin wave is strongly dependent on the amplitude of the initial SST perturbation. To illustrate this, Eq. (1) can be rewritten in terms of anomalies and expanded:

View Expanded

where primes denote anomalies, ΔT_clim is the average ΔT _clim in experiments G_C and NoG_C (5.7°C), and, similarly, T_west is the average T_west (32.8°C) and T_east is the average T_east (27.1°C). To illustrate this dependence, we initialize Eq. (5) with a T ′_east of 0.5°C, calculate the resulting zonal wind stress anomaly τ′_x, and assume that the growth of the SST anomaly by the next time step will be proportional to that zonal wind stress anomaly. We then carry this out 12 time steps and repeat the process for alternative values of T ′_east of 1.0°, 1.5°, and 2.0°C. The results of the sensitivity exercise are presented in Fig. 11. Noting the difference between the growth rate lines for an initial SST anomaly of 0.5° versus 1.0°C, for example, the growth rate via the Bjerknes feedback alone is over four times faster for the larger initial SST anomaly. Recall that the idealized forced results led to a damping of the SST anomaly of about 0.5°C, yet the rate of the growth of the anomaly did not change, because there was no dynamical coupling with the atmosphere and therefore no Bjerknes feedback. Hence, the dependence of the strength of the Bjerknes feedback on the amplitude of the initial anomaly explains why the growth phase of El Niño events would be slower in the G_C simulation while the rate of decay of the El Niño events would remain unchanged.

d. Subtropical cells and recharge–discharge

With the rate of anomaly growth and thus time scale difference reconciled, the curiosity remains: Why, if the Kelvin wave SST response is damped, would the amplitude of the fully mature warm ENSO event be similar (e.g., the composite events shown in Fig. 7c)? As previously alluded, ENSO and its oscillatory nature have been described as a series of recharge and discharge phases (Jin 1997). The meridional circulation in the tropical–subtropical Pacific Ocean is defined by wind-driven subtropical cells, vigorous upwelling at the equator, poleward surface transport, subduction, and equatorward return flow along the thermocline. Variations in transport convergence in the upper pycnocline by Pacific subtropical cells have recently gained attention due to their possible role in modulating decadal variability (McPhaden and Zhang 2002, 2004). Greater transport convergence is associated with cooler equatorial SST, and vice versa, because the equatorward flow in the upper pycnocline is a branch in the same circuit as the equatorial upwelling. Likewise, suppressed transport convergence indicates reduced equatorial upwelling of cold water and would lead to a warmer equatorial SST. In this context, we examine the role of mass transport convergence in the upper pycnocline in affecting differences between the simulated ENSO in our hybrid coupled experiments. Our methodology is identical to that of McPhaden and Zhang (2002, 2004). Based on the vertical structure of the upper pycnocline described by McPhaden and Zhang and vertical variations of mass transport, in either hemisphere and for either coupled simulation, we choose density surfaces between which to vertically integrate mass transport. In experiment NoG_C we use 21.18–25.19 kg m⁻³ (21.71–25.06 kg m⁻³) for the Northern (Southern) Hemisphere and in experiment G_C, respectively, 21.16–24.85 kg m⁻³ (21.71–24.68 kg m⁻³). Finally, along 6°N (6°S), we integrate from 145°E–90°W (160°E–85°W) and obtain upper pycnocline convergence (UPC) by taking the difference between the northward transport at 6°S and the southward transport at 6°N.

We again construct a picture of the composite El Niño event in terms of UPC (Fig. 12). The unfolding of events in the NoG_C composite is as follows: a large SSTA develops at an early stage because the mean state is one in which SSTA is highly sensitive to Kelvin waves, a strong Bjerknes feedback drives a rapid growth of both the SSTA and zonal wind stress anomaly (recall from Figs. 7a,b that the SSTA and zonal wind stress anomaly lines are nearly identical), and UPC is immediately and strongly suppressed. The UPC minimum occurs at the same time as the SSTA maximum. In effect, the DO and Bjerknes feedback are operating in close synchrony, and the UPC quickly responds to the large fluctuations in zonal wind stress. Over the 68 years simulated, the UPC anomaly and SSTA time series are anticorrelated −0.82. This is indicative of a strong positive feedback between UPC and SST during the warming phase: the SST warming reduces the zonal winds, which reduces UPC and upwelling, which contributes to the SST warming. Also in the NoG_C composite (Fig. 12a), UPC is in quadrature with d(HC′_ml)/dt; as UPC decreases, HC′_ml increases.

The evolution of UPC in the G_C composite is quite different: a smaller SSTA begins to develop because the mean state is one in which SSTA is less sensitive to Kelvin waves, and a weaker Bjerknes feedback drives a slower growth of the SSTA and zonal wind stress anomaly. That wind stress anomaly is also apparently too weak throughout most of the warming period to cause a change in the wind-driven UPC. As a result, UPC does not begin to decrease until later than in the NoG_C case. At the end of the warming phase, the zonal wind stress anomaly is strong enough to suppress the UPC, which drives the Niño-3 SSTA to approximately the same amplitude as in the NoG_C composite. As evident in the G_C composite, there is strong association between UPC and d(HC′_ml)/dt, meaning that UPC is having a negative effect on the warming throughout the composite event.

The recharge–discharge paradigm of ENSO was also discussed by Kessler (2002) who, using observations, portrayed the ENSO cycle as an evolution of the state of the equatorial system through phase space defined by basin-wide mean 20°C isotherm depth and Niño-3 SST. In a similar fashion, Fig. 13 depicts the coupled ENSO cycle as an evolution through phase space defined by Niño-3 SSTA and zonal wind stress anomaly (averaged from 5°N to 5°S). In the NoG_C case (Fig. 13a), the system undergoes a rapid growth/warming (quadrant IV to quadrant I) and immediately undergoes a rapid decay/cooling (quadrant II to quadrant III), with little time spent near the origin. Conceptually, this is the DO mechanism operating on a more sensitive mean state with the Bjerknes feedback further driving the anomaly growth. In Kessler’s phase diagram depicting the recharge–discharge process, this would resemble a constant circular orbit through the four quadrants of phase space with nearly equal time spent in each state, rather than an abundance of points near the origin or in the recharge phase. In contrast and more consistent with observations, the G_C phase diagram (Fig. 13b) depicts a DO mechanism operating on a less sensitive mean state, with the Bjerknes feedback acting more subtly until the equatorial system is completely discharged and promptly returns to gradually recharge (i.e., more data points are found near the origin and in quadrant III). Such an evolution through the coupled phase space is more realistic with respect to observations (Fig. 13c).