Rethinking Tropical Ocean Response to Global Warming: The Enhanced Equatorial Warming (original) (raw)

1. Introduction

It is important to understand the response pattern of the climate system to global warming forcing. These patterns represent powerful fingerprints for the attribution of the causes of the climate change. They also provide practical guidance for the future climate response in specific regions. Furthermore, they present a great challenge to our understanding of the fundamental mechanisms of the response of the climate system to global climate forcing. Previous studies have identified several robust response patterns to global warming (Cubasch et al. 2001), such as the polar amplification associated with the snow–ice albedo feedback, and a stronger warming over land than over ocean due to the smaller heat capacity of the land.

The response of tropical Pacific sea surface temperature (SST) has received great attention because of its potential impact on global climate through atmospheric teleconnections. All the work so far, however, has focused on the zonal SST gradient along the equator, that is, the so called El Niño–like response (e.g., Knutson and Manabe 1995, 1998; Meehl and Washington 1996; Cane et al. 1997; Cane 2005; Yu and Boer 2002; Vavrus and Liu 2002; Boer and Yu 2003a, b; Collins and CMIP Modeling Groups 2005). As stated in the current Intergovernmental Panel on Climate Change (IPCC) report, “A majority of models show a mean El-Niño-like response in the tropical Pacific, with the central and eastern equatorial Pacific SST warming more than the western equatorial Pacific, with a corresponding mean eastward shift of precipitation” (Cubasch et al. 2001). Similar features can also be seen in Fig. 1, which shows the SST trends in response to a 1% transient CO2 increase in 14 fully coupled ocean–atmosphere GCMs, 13 from the current IPCC simulations (http://www-pcmdi.llnl.gov/ipcc/about_ipcc.php), and 1 as an in-house Fast Ocean–Atmosphere Model (FOAM; Jacob 1997). If we define the El Niño–like response as an equatorial warming stronger on the east than the west, we may identify seven experiments as El Niño–like warming: Geophysical Fluid Dynamics Laboratory (GFDL; Fig. 1e), Institute Pierre Simone Laplace (IPSL; Fig. 1l), the Model for Interdisciplinary Research on Climate (MIROC)_medres (Fig. 1j), Max Planck Institute (MPI; Fig. 1k), Meteorological Research Institute (MRI; Fig. 1l), U.K. Met Office (UKMO)_hadcm (Fig. 1m), and UKMO_hadgem (Fig. 1n). These simulations, at first sight, seem to point to the “El Niño like” pattern as a fingerprint for tropical Pacific SST response to global warming.

Here, we suggest that this view that emphasizes the pattern analogy with El Niño is deficient for characterizing tropical climate change. Indeed, in both models and observations, it still remains very controversial if the SST trend is El Niño–like warming (stronger warming in the east) or La Niña–like warming (stronger warming in the west). First, many models do not exhibit the El Niño–like warming. In Fig. 1 three models appear to have the equatorial Pacific SST warming more on the west than on the east or La Niña–like warming: Commonwealth Scientific and Industrial Research Organisation (CSIRO; Fig. 1c), FOAM (Fig. 1d), and Goddard Institute of Space Studies (GISS; Fig. 1f); three models have the maximum warming in the central equatorial Pacific: Canadian Centre for Climate Modelling and Analysis (CCCMA; Fig. 1a), Centre National de Recherches Météorologiques (CNRM; Fig. 1b), and MIROC_hires (Fig. 1i); and one model has a comparable warming in the west and east: Institute of Atmospheric Physics (IAP; Fig. 1g). Therefore, the overall distribution of El Niño–/La Niña–like warming in the current IPCC members is similar to that in the 20 Coupled Model Intercomparison Project (CMIP) experiments analyzed by Collins and CMIP Modeling Groups (2005) with a more sophisticated method. After a comparison of the patterns of the SST trend and ENSO variability in each model, they concluded that “the most likely scenario (p = 0.59) . . . is for no trend towards either mean El Niño-like or La Niña-like conditions,” although there is “ a small probability (p = 0.16) for a change towards El Niño-like conditions . . . .” This conclusion remains true in their analysis if the realism of the model ENSO relative to the observation is also considered.

A similar controversy exists for the trend pattern of the historical SST observation, because of the small trend signal relative to interannual and interdecadal climate variability, and the poor quality of observations before WWII, especially over the tropical Pacific. We have examined the linear SST trend in three SST datasets from 1870 to 2000: the Extended Reconstructed SST (ERSST; Smith and Reynolds 2003), the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST; Rayner et al. 2003), and the Kaplan SST (Kaplan et al. 1998). The SST trend since the 1940s shows a robust El Niño–like pattern in all three datasets (Fig. 2; Schneider and Held 2001). This El Niño–like trend, however, becomes less robust and even reverses the sign for trends starting earlier this century. For example, the SST trend since 1900 exhibits a La Niña–like warming in the Kaplan SST [with actually a cooling trend in the central-eastern equatorial Pacific (Cane et al. 1997)], a slight La Niña–like warming in HadISST, but a clear El Niño–like warming in ERSST (not shown). Finally, the SST trend since 1870 shows an El Niño–like warming in ERSST, a weak El Niño–like warming in HadISST, and a rather neutral (zonally uniform) trend in Kaplan SST (not shown). Knutson and Manabe (1998) have also noticed a discrepancy of SST warming trends since 1900 between the El Niño–like trend in an earlier version of HadISST and a La Niña–like trend in the Kaplan SST. They also found that the long record of sea level pressure at Darwin favors an El Niño–like trend. The message is that the patterns of tropical Pacific SST trend differ substantially among different datasets and for different time periods, especially for earlier times.

The mechanism of Pacific equatorial SST response also remains controversial. For the El Niño-like warming, the weaker warming in the west has been suggested to be regulated by the negative feedback associated with the latent heat cooling (Knutson and Manabe 1995) or cloud radiation forcing (Meehl and Washington 1996) on the warm pool SST. On the other hand, it has also been suggested that, with global warming forcing, the equatorial Pacific should exhibit a La Niña–like warming because of the strong oceanic upwelling cooling, and in turn the wind-upwelling dynamic feedback (Bjerknes 1969) in the eastern Pacific (Clement et al. 1996), or by an enhanced latitudinal SST gradient through equatorward oceanic ventilation (Seager and Murtugudde 1997; Liu 1998). Most importantly, the physical nature of the long-term tropical SST response to global warming could differ significantly from that for El Niño variability, because the former is forced externally by a slow radiative climate forcing of global extent, while the latter is generated internally by fast processes within the equatorial ocean–atmosphere system.

Here, we propose an alternative view with the emphasis on the latitudinal SST gradient between the equator and the subtropics. (Here, equator and subtropics refer roughly to the latitude ranges of 0°–10° and 10°–30°, respectively). We propose, relative to the El Niño–like response, a more robust response pattern of tropical Pacific SST to global climate forcing is an Enhanced Equatorial Response (EER) relative to the subtropics. In the case of global warming (cooling), EER is characterized by a stronger warming (cooling) near the equator than in the subtropical Pacific. We will show that, relative to the El Niño–like response, many more climate models are consistent with the EER warming. There also appears to be some observational evidence consistent with EER, although the observational uncertainty remains large. A further study using FOAM identified three mechanisms responsible for the EER warming: the enhanced latent heat loss in the subtropics due to a strong sensitivity of latent heat flux there (Seager and Murtugudde 1997), the reduction of shortwave radiation over the subtropics because of the enhanced atmospheric gross stability (Miller 1997), and in turn, low clouds (Klein and Hartmann 1993), and the suppression of equatorial oceanic mixing/entrainment due to the increased surface ocean stability. In contrast to previous views that prefer El Niño as an analog and emphasize the zonal SST gradient, the atmospheric Walker circulation and the equatorial wind-upwelling dynamic feedback, our alternative view here focuses on the meridional SST gradient, the atmospheric Hadley circulation, and the wind–evaporation feedback. Our paper is arranged as follows. Section 2 discusses the EER evidence in models. Section 3 investigates the mechanism of the SST response by analyzing the heat budget for a FOAM simulation. Further sensitivity experiments with FOAM are discussed in section 4. Relevance of the EER response to the present and past observations is discussed in section 5, and the conclusions are given in section 6.

2. Modeling evidence for EER

Modeling evidence suggests that EER is more robust than the El Niño–like response. In Fig. 1, all eight simulated El Niño–like responses can also be considered as being consistent with EER because their zonal mean warming is stronger on the equator than in the subtropics. In addition, the EER warming also appears in models inconsistent with El Niño–like conditions. Indeed, clear EER warming emerges in all the models except for IPSL (Fig. 1h), MIROC_medres (Fig. 1j), and UKMO_hadcm (Fig. 1m). The clear EER features can also be seen in the Pacific zonal mean SST trend (Fig. 3a), and, even more clearly, in the SST trend normalized by the Pacific mean SST trend (Fig. 3b). Consistent with Fig. 1, except for IPSL, MIROC_medres, and UKMO_hadcm (in dash), all the models exhibit stronger warming on the equator. Furthermore, it is interesting to notice that most normalized SST trends exhibiting EER are clustered surprisingly tight (Fig. 3b), in spite of their dramatically different climate sensitivity (Fig. 3a). For example, FOAM has a climate sensitivity of 1.5 K while MIROC_hires has a sensitivity of 3.5 K (Fig. 3a); the two normalized SSTs, however, almost coincide with each other (Fig. 3b). This suggests that the response of the latitudinal SST gradient is extremely robust across the models, regardless of their climate sensitivity. A further observation of the latitudinal gradient of SST trend (Fig. 3b, Fig. 3a, or Fig. 1) shows a common feature of all the models, which is characterized by a stronger warming in the northern subtropics than in the southern subtropics. This stronger northern warming is caused by the larger land coverage and a stronger land warming effect in the Northern Hemisphere. As a result, the EER warming is more robust relative to the southern subtropics than to the northern subtropics. Indeed, even in IPSL, MIROC_medres, and UKMO_hadcm, the Pacific SST warms more on the equator than in the southern subtropical SST. This leads one to speculate that the EER warming may exist in all the models in Fig. 1 on both hemisphere subtropics, were the land effect excluded.

The discussion on EER above can be quantified by three EER trend indices using the normalized SST trend as

i1520-0442-18-22-4684-eq1

i1520-0442-18-22-4684-eq1

Here, T_EQ is the normalized SST trend averaged within 10° of the equator, while TN and TS are the normalized SST trends averaged in the Pacific northern and southern subtropics (10°–30°), respectively. A positive EER_N (EER_S_) corresponds to an equatorial warming stronger than in the northern (southern) subtropics. The EER is simply the mean of the northern and southern EER and therefore reflects the overall EER trend in both hemispheres. Figure 4 plots the scatter diagrams of the EER and the corresponding EER_N_ (Fig. 4a) and EER_S_ (Fig. 4b). The robustness of EER is immediately clear: all the models show EER warming (positive EER) except for one model (IPSL). Furthermore, Fig. 4b shows a clear EER warming relative to the southern subtropics (positive EER_S_) in all the models, while Fig. 4a shows an EER warming relative to the northern subtropics (positive EER_N_) in all the models except for IPSL, MIROC_mdres, and UKMO_hadgem. Finally, EER indices fall closely on a straight line with respect to both EERN and EERS, except for one outlier (UKMO-hadcm). The slope dEER/dEER_S_ (0.86 in Fig. 4b) is larger than dEER/dEER_N_ (0.61 in Fig. 4a). This is equivalent to an equator-subtropical gradient of normalized SST larger to the south than to the north, (by about 3 times) as evident from Fig. 3b. As a result, the overall EER extent is determined more by the response in the southern subtropics than in the northern subtropics. These robust features of the EER warming are in contrast to the ambiguous signals of El Niño–/La Niña–like warming. Therefore, modeling evidence strongly supports the notion of a much more robust EER warming than El Niño–like warming.

3. Mechanisms for EER warming

The EER warming is somewhat unexpected. With global warming, one might expect a weak warming in the warm equatorial region, especially over the warm pool, because of several negative feedbacks there. The high SST over the warm equatorial region could be regulated by temperature–evaporation feedback because evaporative cooling increases with temperature due to the Clausius–Clapeyron dependence (Knutson and Manabe 1995); it can also be regulated by reduced high cloud shortwave climate forcing (Ramanathan and Collins 1991), by increased atmospheric heat transport (Wallace 1992; Fu et al. 1992; Hartmann and Michelsen 1993; Pierrehumbert 1995) and by the wind-upwelling ocean–atmosphere dynamic feedback (Sun and Liu 1996; Clement et al. 1996).

Here, we will focus on the FOAM simulation. FOAM is convenient for our study because all the monthly outputs in the ocean, including convective heating and mixing, have been stored such that the surface oceanic heat budget can be reconstructed with high accuracy. Furthermore, FOAM is computationally efficient such that many sensitivity experiments can be performed, as shown in the next section. FOAM consists of an AGCM that uses the physics of the National Center for Atmospheric Research (NCAR) Community Climate Model 3 (CCM3) and the 2D parallel dynamics of the Parallel Community Climate Model 2 (PCCM2), a _z_-coordinate OGCM with a 2D parallel decomposition and an explicit free surface, and a parallel coupler (Jacob 1997). The atmosphere model has a resolution of R15 with 18 levels, while the ocean model has a resolution of 2.8° × 1.4° × 24 levels. In this section, we will investigate the trend in the transient CO2 experiment (CO2T), which is forced by a 1% increase of CO2 for 110 yr. Similar results have been obtained in another two ensemble experiments.

It will be shown that the surface heat flux is a major forcing for the EER warming. Over the tropical–subtropical Pacific, the total heat flux is increased along the equator but decreased in the subtropics (Fig. 5a), due partly to the latent heat flux (Fig. 5b) and partly to the shortwave radiation (Fig. 5c). In comparison, the changes of longwave forcing (Fig. 5d) and sensible heat flux (not shown) are largely uniform across the region.

a. The latent heat flux

The latent heat flux contributes significantly to the enhanced gradient of the total surface heat flux, especially the strong cooling over the western subtropics. Overall, the latent heat loss is increased over almost the entire tropical–subtropical region (Fig. 5b), predominantly because of the increased latent heat loss associated with the perturbation humidity U(q*_S_′ − q_′_A) (where U is the wind speed, q*S is the surface saturation specific humidity, qA is the specific humidity at the lowest model level, and an overbar represents the time mean). In spite of a slightly larger increase of the surface air temperature than the SST (by 0.1°C), the sea–air specific humidity gradient q*_S_′ − q_′_A is increased because the surface saturation specific humidity q*_S_′ increases more than the surface air specific humidity q_′_A (Knutson and Manabe 1995) as a result of the nonlinear dependence of saturation specific humidity on temperature and a relative humidity smaller than 1 (about 0.83). Furthermore, since the mean trade wind speed U is stronger in the subtropics than on the equator, the latent heat loss due to perturbation humidity has a greater sensitivity in the subtropics than on the equator, leading to a stronger cooling in the subtropics than on the equator. Over most of the Tropics and especially across the equator, the seasonal surface wind is reduced slightly, reflecting a weaker seasonal Hadley cell, as will be discussed later. The perturbation wind speed effect U ′(q*S − qA) therefore counters the perturbation humidity U(q*_S_′ − q_′_A) to reduce the latent heat loss near the equator, also favoring the enhanced meridional gradient of the latent heat flux. In the western part of the subtropics, however, the perturbation monthly winds tend to enhance the mean trade wind (not shown). Therefore, the effects of perturbation humidity and perturbation wind speed reinforce each other, generating a strong latent heat flux loss of over −6 to −8 W m−2 there. Finally, the latent heat loss is increased significantly by −4 to −8 W m−2 in the subtropics, but only slightly by 0 to −2 W m−2 on the equator. This contributes significantly to the increased meridional gradient of the total heat flux, and, in turn, the EER warming.

The mechanism of a stronger sensitivity of the latent heat flux in the subtropics than near the equator has been pointed out by Seager and Murtugudde (1997) in a hybrid coupled ocean–atmosphere model study, which coupled an ocean general circulation model with an atmospheric boundary layer model. Forced by a spatially uniform anomalous heat flux, and with the wind speed and wind stress prescribed as the observed climatology, their model generated an enhanced equatorial warming relative to the subtropics, which was attributed to the stronger observed mean trade wind speed in the subtropics.

b. The shortwave cloud forcing

The enhanced meridional gradient of total surface heat flux is also contributed significantly by the shortwave forcing, especially in the central and eastern part of the subtropics. Over the subtropics, the shortwave forcing is reduced (Fig. 5c) because of an increased low cloud cover (Fig. 6a), which reduces the incoming shortwave on the top of the atmosphere (TOA; Fig. 6c) and then over the surface ocean (Fig. 5c). The low cloud cover is increased by an increased mean stability (−∂/∂p; Fig. 7a) because the warming increases from 1.5° in the surface to 5°C in the upper troposphere across the Tropics. This increased stability is due to the increased specific humidity and in turn moisture convection, which keeps the lapse rate close to the moist-adiabatic rate in the warm pool (Knutson and Manabe 1995). The warmer air over the deep convection region is then transported to the subtropics by the mean Hadley cell, increasing stability above the trade inversion and in turn, the low cloud cover there (Klein and Hartmann 1993; Miller 1997). The maximum reduction of shortwave forcing occurs in the central northern subtropics and eastern southern subtropics (Fig. 5c) because of the largest increase of low cloud cover there (Fig. 6a). In these regions, the effect of enhanced stability is further reinforced by an anomalous descending (positive ▵ω in Fig. 6e) and in turn adiabatic warming above the trade inversion. Over most of the eastern subtropics, however, the dominant descending is reduced somewhat (negative ▵ω in Fig. 6e), reminiscent of the somewhat reduced Hadley circulation, as will be discussed later.

Near the equator and especially in the western deep convection region, the anomalous descending increases the high cloud (Fig. 6b) at the expense of low cloud (Fig. 6a). As a result, the outgoing longwave radiation at the top of the atmosphere is reduced mostly near the equator (Fig. 6d). However, the longwave radiation does not contribute to the EER warming because the net surface longwave radiation is largely uniform across the tropical–subtropical region (Fig. 5d).

c. Hadley cell

The strength of the Hadley circulation is not changed significantly in the annual mean and in the transition seasons but is weakened substantially in summer and winter. The maximum overturning flow in summer and winter are reduced by about 10% (Figs. 8a,b). The reduction of the seasonal Hadley circulation is reminiscent of previous studies in coupled GCMs (Knutson and Manabe 1995) and simple box models (Miller 1997; Clement and Seager 1999). The heat balance for the weaker Hadley cell under global warming can be interpreted following Knutson and Manabe (1995). The increased moisture induces an enhanced diabatic heating in the upper troposphere over the ascending region. Part of this heating is balanced by an enhanced radiative cooling of the warmer and wetter convective air, while the other part is enhanced by an increased dynamic cooling associated with the increased atmospheric stability. Over the descending region, the increased radiative cooling is balanced by an increased dynamic warming associated with the increased atmospheric stability.

d. Surface ocean heat budget

To quantify the contribution of different mechanisms on tropical SST response, we analyze the surface heat budget. Using the monthly model output, the surface heat budget is reconstructed for the top model layer temperature as

i1520-0442-18-22-4684-eq2

i1520-0442-18-22-4684-eq2

The SST tendency is seen to be forced by three major terms: the surface heat flux, advection, and vertical mixing. The contribution to SST change by a specific forcing A is

i1520-0442-18-22-4684-eq3

i1520-0442-18-22-4684-eq3

This contribution is first calculated for the transient CO2 simulation and the control simulation separately and is then averaged yearly before the linear trend is calculated. Finally, the difference between the linear trends of the two simulations is taken as the anomalous heat budget responsible for the CO2-induced SST trend. Figure 9a shows the latitudinal distribution of the anomalous heat budget of the three major terms after the zonal average across the Pacific. Each term is further decomposed into the shortwave, longwave, latent, and sensible heat fluxes for the surface heat flux (Fig. 9b), and the zonal, meridional, and vertical advections for the total advection (Fig. 9c) and the vertical diffusion and convection for the vertical mixing (Fig. 9d). The surface heat budget (Fig. 9a) shows the need of high accuracy in the reconstruction of the heat budget because the SST trend, at the order of 1– 2°C century−1, is a small residual of three much larger major terms.

The heat budget shows clearly that the surface heat flux is one major forcing for the EER warming trend. The total heat flux exhibits a significant meridional gradient with a prominent heating near the equator and weak cooling toward the subtropics (Fig. 9a). The mechanism for the change of the surface heat flux has been discussed earlier in Fig. 5 and Fig. 6 and will be summarized here in the context of the heat budget. The increased atmospheric CO2 traps infrared energy flux and eventually leads to a net anomalous longwave heating at the surface (Fig. 9b). The longwave heating is partly balanced by the increased latent heat loss and partly by the increased shortwave cooling, both of which directly force an EER warming. The strong latent heat loss in the subtropics is due to the stronger mean trade wind and in turn sensitivity of the latent heat loss to perturbation humidity there, while the strong shortwave cooling in the subtropics is caused by the increased low clouds, which in turn is caused by the enhanced atmospheric stability there.

The vertical mixing is the other major forcing for EER warming because it exhibits a strong warming near the equator (Fig. 9a). This strong equatorial warming is caused by the positive surface heat flux on the equator, which heats the surface more than the subsurface water and therefore increases the surface ocean stability near the equator (Fig. 7b). The increased surface ocean stability suppresses the vertical diffusivity through a Richardson number–dependent parameterization of the vertical diffusivity in the ocean model (Pacanowski and Philander 1981), and in turn the cold vertical diffusive flux and entrainment, resulting in an anomalously warm vertical diffusion (Fig. 9d), and in turn, vertical mixing (Fig. 9a). The vertical diffusion effect, however, is confined near the equator because of the strong near-surface stratification there.

The combined surface heat flux and vertical entrainment heating for the EER warming is balanced mainly by an increased cold vertical advection. Near the equator, the increased surface ocean stratification results in a greatly enhanced upwelling cooling (Fig. 9c), even though the upwelling velocity is reduced slightly by a small westerly wind anomaly on the equator (Vavrus et al. 2005). This increased upwelling cooling cancels most of the heating due to surface heat flux and the warm diffusive, which otherwise would have warmed the equator by 50°C century−1.

In short, the budget analysis suggests that the increased CO2 changes the sea–air humidity gradient, as well as the stability in both the atmosphere and ocean. These changes lead to a stronger latent heat loss and shortwave reduction in the subtropics and a reduced entrainment cooling on the equator, all favoring a stronger warming on the equator relative to the subtropics, that is, the EER warming. The mechanism for EER warming therefore differs fundamentally from the El Niño–like response, which is associated with the atmospheric Walker circulation and the equatorial wind-upwelling dynamic feedback.

4. Sensitivity experiments

Sensitivity experiments are performed with FOAM to further illustrate the mechanism for EER. All the sensitivity experiments are integrated for 110 yr. First, the EER pattern is found to be insensitive to the magnitude and the time scale of the CO2 forcing. This can be seen in two onset global warming experiments with the atmospheric CO2 suddenly increased by 2 (2CO2) and 3 (3CO2) times. The final equilibrium responses in both experiments show an EER warming, similar to that of CO2T. A global cooling experiment is also performed with the CO2 lowered suddenly by a half (HALFCO2), somewhat reminiscent of the climate at the Last Glacial Maximum (LGM). This experiment shows an EER cooling. Together, these experiments suggest that EER is robust for both the warming and cooling climate in FOAM. Therefore, the nature of the tropical SST change for cold climate, such as the LGM, is fundamentally similar to that for a global warming climate, except for an opposite sign.

One important implication from the heat budget analysis above is that the Bjerknes equatorial wind-upwelling feedback is not essential to EER. This is in contrast to the El Niño–like response, which is thought to depend critically on the Bjerknes feedback. To confirm this point, a transient CO2 experiment (FIXWIND) is performed in a thermally coupled setting: the experiment is integrated the same as the fully coupled experiment CO2T except that the wind stress to the ocean model is prescribed as the climatological seasonal cycle of the control. With the fixed wind stress, we eliminated all dynamic ocean–atmosphere feedbacks, in particular the Bjerknes feedback, because ocean currents no longer respond to anomalous wind stress. In the mean time, thermodynamic ocean–atmosphere interaction remains fully active because the SST and the atmosphere affect each other through the interactive surface heat flux. Figure 10 shows that the SST still exhibits an enhanced equatorial warming as in CO2T. In the context of FOAM, this experiment demonstrates unambiguously that dynamic ocean–atmosphere coupling is unnecessary for EER. A further analysis of the heat budget of FIXWIND also suggests similar mechanisms for the EER warming as in the fully coupled experiment CO2T, with the contribution mainly from the enhanced latent heat loss and reduced incoming shortwave forcing in the subtropics, and the reduced oceanic entrainment cooling on the equator. Notice that, in spite of the less critical role of the large-scale ocean dynamics, oceanic mixing process is still active in FIXWIND. This implies that the subgrid small-scale ocean dynamics can still be important for EER warming.

5. Relevance to observations

Two SST observations available are potentially relevant to EER. One is the historical SST of the last 130 yr (e.g., Smith and Reynolds 2003; Rayner et al. 2003; Kaplan et al. 1998) and the other is the LGM SST reconstructed from paleoproxies (e.g., CLIMAP 1981). Before proceeding, it is important to keep in mind the limitations of these SSTs for the detection of EER. The trend in historical SST pattern is much weaker than interdecadal variability. The trend is also likely to have a substantial error before World War II, when the SST field is reconstructed statistically from much sparser observations, such that different SST datasets could differ substantially (e.g., Knutson and Manabe 1998). The LGM SST has a larger signal but also a larger uncertainty, because each proxy reconstruction method has a large uncertainty, and these methods could differ from each other substantially [see Broccoli (2000) for a more recent comparison]. Our examination below appears to find some evidence for EER. However, due to the large uncertainties, our following assessment of the relevance of the EER to observations is unlikely to be conclusive.

We first examine the historical SST. Because of substantial interdecadal climate variability, the pattern of the SST trend varies with the time period, as discussed earlier for the El Niño–like trend. [This trend sensitivity to time period itself suggests that the trend needs to be treated with extra caution (Percival and Rothrock 2005).] As discussed previously in Fig. 2, the SST trend for the last half-century exhibits a robust EER warming in all the three SST datasets. The trend pattern starting from earlier times, however, becomes less robust due to strong Pacific interdecadal variability (PDV). Figure 11 plots the evolution of Pacific zonal mean SST for the three SSTs. To focus on the low-frequency variability, all SSTs are low passed with a 15-yr running mean. A visual observation of the three datasets shows some common features, notably a weak warming trend embedded in a dominant multidecadal PDV in the extratropics (Minobe 1997). A similar interdecadal variability can also be detected in the Tropics, but with the opposite polarity to the extratropics (Zhang et al. 1997; Deser et al. 2004; in addition to shorter decadal variability). The phase of the PDV is such that the extratropics (Tropics) is warm (cold) near about the 1870s, cold (warm) in the 1900s, warm (cold) again in the 1950s, and cold (warm) again toward the 1980s. Consequently, the SST warming trend tends to be stronger on the equator and weaker in the extratropics (consistent with EER warming) starting from the 1870s, but weaker on the equator and stronger in the extratropics (opposite to EER warming) since the 1900s, and, again, stronger on the equator and weaker in the extratropics after the 1950s. The particularly strong EER warming trend after the 1950s, we speculate, may represent the final emergence of the EER with the increased CO2. This follows because the historical SST response signal should intensify with the increased anthropogenic CO2 forcing and therefore should become more distinguished toward the present. Indeed, in our transient CO2 simulations in FOAM, a clear EER warming does not emerge until about 50 yr after the CO2 increase. Admittedly, the EER warming trend in the last half-century is stronger than the true EER warming response to the CO2 forcing, because of the in-phase superimposition of the PDV for this period. This preliminary speculation, we recognize, remains difficult to prove or disprove without future observations.1 Furthermore, even if the speculation is true, it remains difficult to tell whether the EER warming is more robust than the El Niño–like warming because the historical EER warming tends to cooccur with the El Niño–like condition (e.g., Fig. 2).

The LGM SST change provides another independent test for EER. The 1981 reconstruction of LGM SST (CLIMAP 1981) shows an enhanced equatorial cooling relative to the subtropics (Fig. 12a), consistent with EER cooling. The EER cooling can also be identified in the LGM SST reconstructed using the modern analog method (Fig. 12b; Prell 1985). The response of the LGM zonal equatorial SST gradient, however, remains heavily debated (Lea 2002; Rosenthal and Broccoli 2004), similar to the case of the historical SST trend discussed before. With the cooling stronger in the west than in the east, the 1981 CLIMAP reconstruction (Fig. 12a) appears to show a La Niña–like cooling. The enhanced western cooling is also inferred from a recent SST reconstruction from Mg/Ca and _δ_18O (Koutavas et al. 2002) and may also be consistent with some bioproductivity proxy records, which imply a weaker productivity and in turn weaker upwelling in the eastern Pacific at LGM (Loubere 1999). The enhanced western cooling is also consistent with a recent reconstruction of western Pacific salinity, which increases during cold statidals, implying a reduction of precipitation in the west and, in turn, equatorial trade wind (Stott et al. 2002). In the mean time, there is also substantial evidence suggesting a cooling inconsistent with La Niña–like cooling. An earlier reconstruction of the LGM SST (CLIMAP 1979) exhibits an enhanced equatorial cooling in the central Pacific while the reconstruction from Prell (Fig. 12b) shows a stronger cooling on both the western and eastern sides. An enhanced eastern cooling is also suggested by the faunal-based analysis in the eastern equatorial Pacific (Pisias and Mix 1997) and an SST reconstruction based on Mg/Ca and _δ_18O in both the western and eastern Pacific (Lea et al. 2000). The enhanced eastern cooling seems to be also consistent with a steeper equatorial thermocline at LGM (Andreasen and Ravelo 1997) and an increased trade wind that is implied from an increased dust deposit (Chuey et al. 1987; Liu et al. 2000). In contrast to the highly controversial zonal equatorial SST gradient, so far, there has been little observational evidence against the general pattern of an enhanced cooling in the equator relative to the subtropics, that is, the EER cooling. Therefore, the LGM SST seems to exhibit a more robust EER cooling than the El Niño–/La Niña–like cooling. This impression is also consistent with published LGM model simulations, because some models show an enhanced cooling in the east (Hewitt et al. 2001; Kitoh et al. 2001) and some others in west (Liu et al. 2002; Shin et al. 2002). All of these experiments, however, are consistent with an EER cooling. The LGM SST evidence is important because it provides so far the only paleo-observational evidence that can be used to test the global SST changes. Boer et al. (2004) has also used the LGM SST as a past observational evidence with implication to future SST. However, as in other previous studies, they have focused on the El Niño–like response.

6. Conclusions and discussions

The response pattern of tropical Pacific SST to global warming is studied with a new emphasis on the meridional SST gradient. Based on evidence from transient CO2 experiments, we propose an important tropical Pacific SST fingerprint in response to global climate forcing as an enhanced equatorial warming relative to the subtropics (EER warming). We further argue that EER provides a more robust SST fingerprint than the traditionally studied El Niño–like condition. In contrast to the El Niño–like response that is characterized by the zonal SST gradient, EER is characterized by the meridional SST gradient. Most importantly, the mechanism for EER is associated with the enhanced negative shortwave cloud forcing and latent heat loss in the subtropics, and the reduced surface ocean entrainment on the equator. This differs fundamentally from the mechanism for the El Niño–like response, which is associated with the eastern Pacific oceanic upwelling and Bjerknes ocean–atmosphere dynamic feedback. Therefore, the recognition of EER is important for our understanding of the nature of tropical climate change and its global impact. The identification of EER in the present and past observations still remains a great challenge, because of the small signal of SST gradient and the large uncertainty of the data. Nevertheless, some observations available appear to show an emerging EER warming in recent decades, and an EER cooling at LGM. The LGM EER cooling may also be more robust than the El Niño–like cooling.

The responses of the tropical atmospheric and oceanic circulation, as well as the energy transports, remain to be understood from the coupled atmosphere–ocean perspective. Here, we briefly discuss the implied energy divergence in FOAM. In spite of a reduction of seasonal Hadley circulation (Fig. 8), the divergence of the atmospheric energy transport from the Tropics is increased because of the increase of the atmospheric gross stability (Knutson and Manabe 1995; Miller 1997; Clement and Seager 1999). This increase of energy divergence can be seen in the net radiation absorption by the atmospheric column for the fully coupled (CO2T; Fig. 13a) and thermally coupled (FIXWIND; Fig. 14a) CO2 experiments. Here, the atmospheric energy transport divergence is taken as the energy absorption by all the processes in the atmospheric column, because at seasonal and longer time scales the atmosphere is in quasi equilibrium. Both models show a tendency of energy divergence from the Tropics toward the extratropics. A further comparison of the two simulations shows that the energy diverges mostly in the western deep convection region in CO2T but is largely uniform across the equatorial Pacific over the intertropical convergence zones in FIXWIND.2 This suggests that tropical dynamic ocean–atmosphere coupling may induce energy divergence through the energy redistribution along the equator.

The divergence of oceanic heat transport can also be implied from the total surface heat flux, if the tropical upper ocean is assumed to be in quasi-equilibrium with the transient CO2 forcing. As seen in Fig. 5a (or Fig. 9a) for CO2T and in Fig. 14c for FIXWIND, the net surface heat flux into the ocean is increased along the equator and decreased in the subtropics. This is largely consistent with Watterson (2003). Assuming that the upper ocean is in quasi equilibrium with the CO2 forcing, the net heat gain on the equator needs to be exported to the subtropics by tropical ocean circulation. This oceanic heat transport effect is roughly in the same direction as the atmospheric energy transport divergence (but occurs at a smaller scale near the equator for the ocean). However, the oceanic heat transport effect is weaker than that of the atmosphere because the overall change of the heat divergence over the ocean is weaker than the radiation absorption by the atmosphere column (Fig. 5a versus Fig. 13a, and Fig. 14c versus Fig. 14a). Therefore, the net energy absorption by the coupled ocean–atmosphere system, as implied by the net radiation flux at the top of the atmosphere (Figs. 13b, 14b), is largely carried by the atmosphere. Furthermore, the change of atmospheric heat transport is not compensated by the oceanic heat transport. This differs from the results of some simple coupled ocean–atmosphere box models (Clement and Seager 1999; Held 2001). Part of the discrepancies may stem from their assumption of a fixed energy export toward the midlatitude. In FOAM, the net energy flux changes with the increased CO2 (Figs. 13b, 14b), implying a change of the net energy export toward the extratropics in the coupled system. The largely similar EER responses in the fully coupled and thermally coupled experiments also suggest that the change of the oceanic circulation is not critical for EER warming. However, an increase of surface stability can also increase the heat transport in the ocean, somewhat similar to the atmosphere. The importance of the surface mixing process for EER also implies the potential importance of subgrid-scale mixing dynamics for EER. This may offer a partial explanation for the disappearance of EER in a coupled GCM when the dynamic ocean model is replaced by a slab ocean (Yu and Boer 2002) because the oceanic mixing effect is absent in the slab mixed layer ocean.

We have also studied the EER in a subset of the CMIP simulations (Collins and CMIP Modeling Groups 2005): CCCMA, Center for Climate System Research (CCSR)_NIES, CSIRO_Mk2, NCAR_CSM1, ECHAM4_OPYC3, the Second Hadley Centre Coupled Ocean–Atmosphere GCM (HadCM2), GFDL_R15, and ECHAM3_LSG. The results (not shown) are similar to the current IPCC simulations, with six out of eight exhibiting EER warming. Interestingly, the only two models that do not exhibit EER are the two models with the poorest ocean representation: the GFDL-R15 model (Knutson and Manabe 1995), which has an ocean of a resolution of 4.5° latitude × 3.75° longitude with 12 levels, and the ECHAM3-LSG, which has a geostrophic ocean that does not apply well to the tropical ocean. It however remains unclear if this is simply a coincident. In general, the role of the ocean in EER remains to be clarified.

This study has left many important questions to be understood. Are the mechanisms identified in FOAM also applicable to other models? Furthermore, are there common mechanisms to all the models for the EER warming? It is important to improve the observational data such that the characteristic pattern of the SST change in response to global climate forcing can be identified. Finally, in spite of some crude dynamic considerations, the choice of the latitudinal SST contrast as the target of study here remains somewhat artificial. It is perhaps more revealing to identify the pattern of SST trend according to the dynamic regions, instead of simply the latitude.

In spite of remaining problems and uncertainties, this pilot study, we hope, has accomplished its two objectives. First, we have proposed the EER as an alternative paradigm for the tropical Pacific SST fingerprint, which is more robust than the traditionally studied El Niño–like response. Second, through the study of FOAM simulations, we have demonstrated some viable mechanisms for EER, mainly the latent heat flux, the shortwave cloud forcing, and the oceanic mixing. Understanding the spatial pattern of temperature response (i.e., fingerprint) requires much more sophisticated dynamics than that for the global mean temperature changes. By further exploring the climate dynamics associated with the SST pattern, we hope to stimulate more studies on the tropical climate change and its global impact.

Acknowledgments

We thank Drs. L. Wu, M. Notaro, and R. Jacob for their help in model simulations; Ms. D. Lee, P. Behling, and Mr. K. Wiley for their help with the analysis; Drs. W. Prell, D. Lea, P. Loubere, E. DeWeaver, and G. Petty for helpful discussions; and Drs. M. Cane, M. Collins, and A. Noda for sharing with us their preprints and figures. We are grateful to Dr. A. Broccoli for helpful discussions and for providing us the LGM SST data. We particularly thank the two reviewers, whose insights and constructive criticism have improved the paper significantly. This work is supported by DOE, NSF, and NOAA. Partial support (ZL and NW) is also provided by NSF of China (40333030) and the National Basic Research Program of China (2004CB720208).

REFERENCES

Fig. 1.

Fig. 1.

Fig. 1.

Linear trend of Pacific SST (K century−1) for 14 coupled ocean–atmosphere GCM simulations forced by a transient 1% increase of atmospheric CO2 concentration. The models include 13 IPCC models (http://www-pcmdi.llnl.gov/ipcc/about_ipcc.php) and the FOAM. For each model, the trend is calculated using the first 80 yr of simulation (except for IPSL and MIROC_medres, which only have 70 yr of data). (a) CCCMA, (b) CNRM, (c) CSIRO, (d) FOAM, (e) GFDL_cm (f) GISS_eh_1880–1999, (g) IAP, (h) IPSL_1860–1930, (i) MIROC_hires, (j) MIROC_medres (k) MPI, (l) MRI_1801–1900, (m) UKMO_hadcm, and (n) UKMO_hadgem. The results for IAP, MIROC_medres, and MPI are from a three-member ensemble mean. For each model, the trend above the mean is shaded.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 2.

Fig. 2.

Fig. 2.

Observed tropical SST trend (K century−1) from 1940 to 2000 in the HadISST data (positive shaded). Similar results are obtained with the ERSST and Kaplan SST datasets.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 3.

Fig. 3.

Fig. 3.

Pacific zonal mean SST trend. (a) SST trend (K century−1) zonally averaged across the Pacific Ocean (120°E–80°W) for the 14 simulations in Fig. 1. (b) Normalized zonal mean SST trend, which is the same as (a), but with each SST trend normalized by its Pacific mean trend. All the simulations are in the solid line, except for IPSL, MIROC_medres, and UKMO_hadcm in dashed line; these three exhibit EER in the Northern Hemispheres (negative EER_N_ in Fig. 4a) in dashed line (IPSL, MIROC_medres, and UKMO_hadcm).

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 4.

Fig. 4.

Fig. 4.

Scatter diagram of EER trend indices for the 14 experiments in Fig. 1. (a) EER vs EER_N_ and (b) EER vs EER_S_. The two regression lines are also plotted.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 5.

Fig. 5.

Fig. 5.

Differences in surface heat fluxes over the ocean (positive downward) between the last 30 yr (years 80–110) of CO2T and the control. (a) Total surface heat flux, (b) latent heat flux, (c) net shortwave radiation, and (d) net longwave radiation (contour interval: 2 W m−2).

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 6.

Fig. 6.

Fig. 6.

Differences between the last 30 yr of the fully coupled CO2T and the control. (a) Low and (b) high cloud amounts (contour interval: 1%); (c) net downward shortwave and (d) longwave radiations at the TOA (contour interval: 2 W m−2); and (e) 500-hPa omega (contour interval: Pa h−1).

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 7.

Fig. 7.

Fig. 7.

Potential temperature differences between the last 30 yr of the fully coupled CO2T and the control. Latitude–height plots of the (a) atmospheric (contour interval: 0.5 K) and (b) oceanic (contour interval: 0.1 K) temperatures zonally averaged across the Pacific Ocean (120°E–90°W). Dashed contours represent warming smaller than 3 K for the atmosphere and 1.2 for the ocean.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 8.

Fig. 8.

Fig. 8.

(a) Atmospheric summer [June–July–August (JJA)] overturning streamfunction in the control simulation. (b) The difference of the overturning streamfunctions between the last 30 yr of the fully coupled CO2T and the control. [Contour interval is 50 Sv in (a) and 5 Sv in (b); 1 Sv = 109 kg s−1.] The winter [December–January–February (DJF)] streamfunction is similar, as a mirage image about the equator with the opposite sign.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 9.

Fig. 9.

Fig. 9.

Surface heat budget for the SST calculated in terms of the contribution of each term to SST variability (after time integration). The anomalous heat budget is calculated as the difference of the contributed SST changes between the fully coupled CO2T and the control experiment. Plotted are the contributions due to (a) total surface heat flux, total advection, and total vertical mixing; (b) surface heat flux as decomposed to sensible and latent heat flux and net surface longwave and shortwave radiation (all surface fluxes positive downward); (c) advection as decomposed to zonal, meridional, and vertical advections; and (d) vertical mixing as decomposed to vertical diffusion, convection, and horizontal mixing. The unit is in °C century−1. (For the conversion to heat flux forcing, 100°C century−1 is about 3 W m−2 for the 20-m surface layer here.) The vertical diffusion and convection are saved from the model integration, while other terms are calculated with the monthly data.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 10.

Fig. 10.

Fig. 10.

Pacific zonally mean (120°E–80°W) SST changes induced by CO2 forcing in the FOAM sensitivity experiments. The SST changes are the difference between the final equilibrium state and the control for onset CO2 experiments 2CO2, 3CO2, and HALFCO2 (multiplied by a minus sign), but between the last 30-yr mean and the control for the transient CO2 experiments CO2T and FIXWIND.

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 11.

Fig. 11.

Fig. 11.

Time–latitude plot of the evolution of Pacific zonal mean (120°E–80°W) SST anomalies in the (top) ERSST, (middle) HadISST, and (bottom) Kaplan SST. The SSTs are low passed with a 15-yr running mean (contour interval: 0.1 K).

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 12.

Fig. 12.

Fig. 12.

Difference in annual mean tropical Pacific SST (K) between LGM and present as reconstructed (a) by CLIMAP (1981) and (b) by the modern analog method of Prell (1985). The data are plotted on the R30 grid. The value is marked if that grid contains at least one paleotemperature estimate. (Adapted from Broccoli 2000)

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 13.

Fig. 13.

Fig. 13.

Heat flux changes between the last 30 yr of fully coupled CO2T and the control for (a) the flux difference between the net downward radiative flux at the TOA [in (b)] and the net downward heat flux at the ocean surface (in Fig. 5a), which represents the energy absorption by all the processes in the atmospheric column; (b) the TOA flux difference between the net incoming shortwave radiation (in Fig. 6c) and the net outgoing longwave radiation (Fig. 6d with a minus sign), which represents the net TOA downward energy flux, or, the energy absorption by the entire coupled atmosphere–ocean column (contour interval: 2 W m−2).

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1

Fig. 14.

Fig. 14.

Fig. 14.

Heat flux differences between the last 30 yr of thermally coupled transient CO2 experiment (FIXWIND) and the control. (a) Atmospheric absorption and (b) net downward radiative flux at the TOA (the same as Figs. 13a and 13b, respectively, but for FIXWIND). (c) The net downward heat flux at the ocean surface (similar to Fig. 5a but for FIXWIND; contour interval: 2 W m−2).

Citation: Journal of Climate 18, 22; 10.1175/JCLI3579.1