Evaluating the Formation Mechanisms of the Equatorial Pacific SST Warming Pattern in CMIP5 Models

The warming patterns of equatorial Pacific SST due to rising greenhouse gas concentrations is one of the most important problems in projecting regional climate change and has thus been paid considerable attention in the research community for decades (Clement et al., 1996; Collins, 2005; Liu et al., 2005; Xie et al., 2010; Ma and Yu, 2014). The patterns of equatorial Pacific SST warming (EPSW) affect various aspects of regional and global climate change. For example, they dominate the changes in annual-mean precipitation, with increased (decreased) rainfall over the areas of large (small) SST warming, and play a more important role in the changes in tropical cyclone intensity than the local absolute SST increases (Vecchi and Soden, 2007; Knutson et al., 2008; Xie et al., 2010; Huang et al., 2015). Moreover, the uncertainties of EPSW also dominate the uncertainties of the changes in atmospheric circulation over the equatorial Pacific (Ma et al., 2012; Ma and Xie, 2013).

Two well-known features of EPSW patterns have been obtained from the multi-model ensemble (MME) of CMIP3 and CMIP5, and from individual model simulations (Fig. 1a): the zonal El Niño-like warming pattern (simply referred to as the El Niño-like pattern hereafter), with more warming in the eastern than western Pacific (Ramanathan and Collins, 1991; Meehl and Washington, 1996; Collins, 2005; Vecchi and Soden, 2007; Song and Zhang, 2014); and the meridional equatorial peak warming (EPW) pattern (Liu et al., 2005; Xie et al., 2010). However, these patterns remain controversial in different scenarios (DiNezio et al., 2009; Zhang and Li, 2014) and different models (Huang and Ying, 2015). For instance, a few studies have suggested a La Niña-like warming (Clement et al., 1996; Cane et al., 1997) or a zonal uniform warming (DiNezio et al., 2009) for the zonal structure of the SST warming over the equatorial Pacific.

Several distinct mechanisms have been proposed to explain the discrepant SST warming patterns. For the zonal structure, the weakened Walker circulation associated with a slower increase in rainfall than in moisture (Held and Soden, 2006) can reduce the zonal SST gradient to promote an El Niño-like pattern by reducing the westward surface wind stress and the westward oceanic current as well as the cold upwelling in the eastern Pacific (Vecchi and Soden, 2007). The zonal SST gradient can also be weakened by a greater evaporative cooling in the western Pacific than in the eastern Pacific (Knutson and Manabe, 1995) and by the stronger cloud radiation regulation in the western Pacific (Ramanathan and Collins, 1991). On the other hand, the zonal SST gradient can be enlarged by the increased ocean vertical temperature gradient in the eastern Pacific with upwelling colder subsurface water, known as the ocean dynamical thermostat effect (Clement et al., 1996; Cane et al., 1997), favoring a La Niña-like warming pattern. Moreover, the zonal warming pattern can be enlarged by the Bjerknes feedback of zonal air-sea coupling (Bjerknes, 1969; Song and Zhang, 2014).

Figure 1.  The (a) MME SST warming pattern and (b) mixed layer ocean temperature warming pattern in the equatorial Pacific. Stippling indicates that more than 80% of models have the same sign.

In terms of the meridional pattern, (Seager and Murtugudde, 1997) attributed the EPW pattern to the weaker trade wind at the equator than that in the subtropics, and (Liu et al., 2005) to the changes in latent heat, shortwave cloud forcing and ocean vertical mixing. (Xie et al., 2010) further emphasized the dominant role of the climatological minimum of evaporative cooling at the equator.

All of these formation mechanisms seem theoretically reasonable. However, some mechanisms can be found merely in individual model experiments. For example, the ocean dynamical thermostat as a damping effect to the El Niño-like pattern was found in the Zebiak-Cane CGCM with a uniform heat flux forcing situation (Clement et al., 1996). Based on hybrid CGCM experiments, the EPW pattern was attributed to the stronger trade wind speed in the subtropics than at the equator (Seager and Murtugudde, 1997), whereas the effect of evaporative cooling was suggested based on the simulations of the GFDL's CGCM (Knutson and Manabe, 1995). However, the performances of these mechanisms in a large group of models remain unclear.

In the present study, we analyze the changes in the ocean mixed layer energy budgets in 32 CMIP5 models to evaluate the importance of these mechanisms on the formation of the equatorial Pacific SST warming pattern. To quantify the importance of these mechanisms, we decompose the ocean mixed layer energy budgets into various terms to represent the respective mechanisms. The paper is organized as follows: Section 2 describes the models, variables and methods. Section 3 presents the results. Conclusions are given in section 4.

Outputs from 32 CMIP5 models are used in the present study. Table 1 lists the names and relevant organizations of the 32 models. The details of the models can be found at http://www-pcmdi.llnl.gov/ (Taylor et al., 2012). The historical runs for the period 1981-2000 and the RCP8.5 runs for 2081-2100 are used to represent the current and future climate, respectively.

The variables include the monthly mean SST, total cloud fraction (its standard variable name in CMIP5 is clt), surface latent heat flux (Q _E), sensible heat flux (Q _H), net longwave radiation (Q _LW), net shortwave radiation (Q _SW), surface zonal (uas) and meridional (vas) wind velocity, surface scalar wind speed (sfcWind), ocean temperature (thetao), and ocean 3D mass transport (umo, vmo, and wmo). The net longwave/shortwave radiation is defined as the difference between upward and downward longwave/shortwave radiation. The sign of the flux is defined such that a positive flux warms the ocean. Some variables not archived in a few models are marked in Table 1. Moreover, the ocean vertical mass transport not well described in CSIRO Mk3.6.0, BNU-ESM and MIROC5 is also excluded (http://cmip-pcmdi.llnl.gov/cmip5/ errata/cmip5errata.html). Ocean 3D currents are obtained from the ocean 3D mass transports. All of the model outputs are interpolated onto a 2.5^°× 2.5^° grid.

The change under global warming is first defined as the difference between the 20-year long-term mean of the RCP8.5 run and that of the historical run. Changes in each model are normalized by their respective tropical SST warming averaged between 60^°S to 60^°N, in order to remove the influence of tropical mean SST change. Then, the regional mean SST increase is removed to define the EPSW pattern. As shown in Fig. 1a, the sign agreement test indicates that most of the CMIP5 models (more than 80% of the 32 models) show some universal patterns of EPSW.

The formation mechanisms of the EPSW patterns can be detected from the surface energy budget changes. For instance, the effect of evaporative cooling can be represented by the latent heat changes (Xie et al., 2010), the effect of cloud-shortwave-radiation-SST feedback by the shortwave radiation changes (Ramanathan and Collins, 1991), and the effect of the ocean dynamical thermostat is implied in the ocean heat transport changes (Clement et al., 1996; DiNezio et al., 2009).

For the change in long-term mean, the energy budget balance in the ocean mixed layer can be expressed as (Xie et al., 2010) \begin{equation} \Delta Q_{E}+\Delta Q_{H}+\Delta Q_{LW}+\Delta Q_{SW}+\Delta D_{O}=0 ,(1) \end{equation} where ∆ denotes future change. ∆ Q _E,∆ Q _H,∆ Q _LW,∆ Q _SW and ∆ D _O represent changes in latent heat flux, sensible heat flux, net longwave radiation, net shortwave radiation and ocean dynamical processes, respectively. The D _O can be decomposed as \begin{equation} \Delta D_{O}=\Delta Q_{u}+\Delta Q_{v}+\Delta Q_{w}+\Delta R , (2)\end{equation} where ∆ Q _u,∆ Q _v and ∆ Q _w represent changes in the ocean 3D heat transports, and ∆ R is a residual term representing changes in heat transports due to sub-grid scale processes such as vertical mixing and lateral entrainment (DiNezio et al., 2009).

Because the ∆ Q _u,∆ Q _v and ∆ Q _w include both the effects of changes in ocean currents and changes in ocean temperature gradients associated with different mechanisms, we decompose them into two components: \begin{eqnarray} \Delta Q_{u}&\approx&-\rho_{o}c_{\it p}\int\limits_{-H}^0\Delta u\dfrac{\partial T}{\partial x}dz-\rho_{o}c_{\it p}\int\limits_{-H}^0u \dfrac{\partial\Delta T}{\partial x}dz\nonumber\\ &=&\Delta Q_{u1}+\Delta Q_{u2} ,\nonumber\\ \Delta Q_{v}&\approx&-\rho_{o}c_{\it p}\int\limits_{-H}^0\Delta v\dfrac{\partial T}{\partial y}dz-\rho_{o}c_{\it p}\int\limits_{-H}^0v \dfrac{\partial\Delta T}{\partial y}dz\nonumber\\ &=&\Delta Q_{v1}+\Delta Q_{v2} ,\\ \Delta Q_{w}&\approx&-\rho_{o}c_{\it p}\int\limits_{-H}^0\Delta w\dfrac{\partial T}{\partial z}dz-\rho_{o}c_{\it p}\int\limits_{-H}^0w \dfrac{\partial\Delta T}{\partial z}dz\nonumber\\ &=&\Delta Q_{w1}+\Delta Q_{w2} ,(3)\nonumber \end{eqnarray}

where ρ _o is sea water density; c _p is specific heat at constant pressure; H is mixed layer depth, chosen as a constant of 30 m; and u,v,w and T are ocean zonal, meridional and vertical current, and temperature, respectively. ∆ Q _u1,∆ Q _v1 and ∆ Q _w1 represent the effect of changes in ocean currents, which mainly reflect the role of changes in surface wind stress and in atmospheric general circulation (Vecchi and Soden, 2007); and ∆ Q _u2,∆ W _v2 and ∆ Q _w2 represent the effect of changes in ocean temperature gradients. The patterns of mixed layer temperature changes in Fig. 1b are close to the EPSW patterns (Fig. 1a), with a spatial correlation coefficient near 0.97, indicating that the mixed layer energy budget is reasonable for studying the SST change pattern and that the mixed layer depth (30 m) is properly chosen.

Another important variable involving multiple processes is latent heat flux (Xie et al., 2010). The surface latent heat flux in models is calculated using the bulk formulas: \begin{equation} \label{eq1} Q_{E}=\rho _{a}LC_{E}Vq_{s}(T_{ss})(1-{RH}e^{-\alpha T'}) , (4)\end{equation} where ρ _a is surface air temperature; L is latent heat of evaporation; C _E is the exchange coefficient; V is surface wind speed; q _s(T) is the saturated specific humidity, following the Clausius-Clapeyron relationship; T _ss is SST; and T' is the difference between SST and surface air temperature, known as the stability parameter. RH is the relative humidity, α=L/(R _vT²)≈ 0.06 K^-1, and R _v is the ideal gas constant for water vapor.

From Eq. (2), changes in latent heat flux can be influenced by changes in SST, surface wind speed, surface stability and RH, related to different processes (Xie et al., 2010; Huang, 2015). Thus, ∆ Q _E is decomposed into two parts: ∆ Q _E=∆ Q _EO+∆ Q _EA, where ∆ Q _EO=α Q _E ∆ T _ss is the response of SST change (Newtonian cooling) and ∆ Q _EA contains the effects of changes in wind speed, RH and surface stability (Du and Xie, 2008; Xie et al., 2010). In ∆ Q _EA, the effect due to surface wind speed change can be written as ∆ Q _EW=Q _E∆ V/V, which is the key aspect in the wind-evaporation-SSTfeedback (Xie and Philander, 1994) and important to the SST warming pattern formation (Xie et al., 2010). The residual of ∆ Q _EA, ∆ Q _ER=∆ Q _EA-∆ Q _EW, represents both the effect of changes in RH and surface stability.

The ∆ Q _EO=α Q _E∆ T _ss, including the effects of the climatological evaporation Q _E and the SST change, can be divided into two terms, following (Huang, 2015): \begin{equation} \label{eq2} \Delta Q'_{EO}=\alpha\langle{Q_{E}}\rangle\Delta T'_{ss}+\alpha Q'_{E}\langle{\Delta T_{ss}}\rangle=\Delta Q'_{EO1}+ \Delta Q'_{EO2} , (5)\end{equation} where the angled brackets denote the tropical Pacific mean, the prime represents the deviations, the term ∆ Q' _EO1 represents the response of the spatially non-uniform SST change, and ∆ Q' _EO2 the effect of the spatial distribution of the climatological latent heat flux.

Figure 2.  Changes in (a) latent heat flux ($\Delta Q_E$), (b) net longwave radiation ($\Delta Q_LW$), and (c) net shortwave radiation ($\Delta Q_SW$). (d-f) As in (a-c) but with the respective tropical Pacific mean removed.

Figure 3.  Regional changes in the (a) zonal, (b) meridional and (c) vertical heat transport, and (d) the residual term in Eq. (2).

Figures 2a-c exhibit the changes in latent heat flux (∆ Q _E), net longwave radiation (∆ Q _LW) and net shortwave radiation (∆ Q _SW). Changes in sensible heat flux (∆ Q _H) are omitted due to relatively small values. SST warming is mainly contributed by increases in net downward longwave radiation, while changes in latent heat and net shortwave radiation suppress surface warming. The regional deviations of these surface energy budgets are shown in Figs. 2d-f. Changes in latent heat flux and net shortwave radiation exhibit pronounced spatial patterns (Figs. 2d and f), indicating more important influences on the EPSW pattern; whereas, the increases in net longwave radiation (Fig. 2e) are mainly spatially uniform, contributed by the near uniform increases in greenhouse gases.

For the ocean dynamics (Fig. 3), the 3D heat transports are mainly located in the equatorial Pacific, except the meridional heat transport, which cools the NH and warms the SH off the equator. The horizontal heat advection (Figs. 3a and b) warms the surface of the equator, while the vertical heat advection (Fig. 3c) cools SST in the eastern Pacific. In addition, the residual term mainly warms the equatorial eastern Pacific and cools the off-equatorial flanks of the eastern Pacific (Fig. 3d).

In the MME, the SST warming in the eastern Pacific is larger than that in the western Pacific, exhibiting an El Niño-like pattern. The difference between the regional mean of (5^°S-5^°N, 145^°-85^°W) and (5^°S-5^°N, 125^°E-175^°W), denoted by the dashed green boxes in Fig. 1a, is around 0.12^°C per 1^°C of global warming.

Figure 4.  Components of the regional changes in latent heat flux: (a) $\Delta Q_EO$, (b) $\Delta Q_EA$, (c) $\Delta Q_EO1$ and $\Delta Q_EO2$

Figure 5.  (a) Cloud-shortwave-radiation-SST feedback index in the historical run. (b) Changes in total cloud fraction.

Figure 6.  Regional changes in the ocean heat transports induced by changes in (a) zonal current ($\Delta Q'_u1$), (b) meridional current ($\Delta Q'_v1$), (c) vertical current ($\Delta Q'_w1$), (d) zonal gradients of temperature ($\Delta Q'_u2$), (e) meridional gradients of temperature ($\Delta Q'_v2$), and (f) vertical gradients of temperature ($\Delta Q'_w2$).

Four mechanisms are suggested to influence the zonal pattern formation. The total effect of evaporative cooling, represented by the changes in latent heat flux, causes warmer SST in the eastern than the western Pacific, favoring an El Niño-like pattern (Fig. 2d). Figures 4a and b show the Newtonian cooling effect (∆ Q _EO) and the atmospheric forcing effect (∆ Q _EA). The ∆ E _EO near the equator is similar to the EPSW pattern (Fig. 4a), indicating a favorable factor for the El Niño-like pattern. On the contrary, the atmospheric adjustment effect (Fig. 4b) appears to damp the El Niño-like warming. In ∆ E _EO, the effect of the spatial distribution of climatological latent heat flux (∆ E _EO2, Fig. 4d) is the dominant contributor to the total effect of evaporative cooling, favoring an El Niño-like pattern (Knutson and Manabe, 1995), while the effect of non-uniform SST change (∆ Q _EO1, Fig. 4c) plays a damping role.

The cloud-shortwave-radiation-SST feedback is suggested to be another factor favoring the El Niño-like pattern, which can be represented by the changes in shortwave radiation. As shown in Figs. 2c and f, there is more decreased net shortwave radiation over the western Pacific than the western Pacific, favoring an El Niño-like pattern. To illustrate the role of cloud-shortwave-radiation-SST feedback, a cloud-shortwave-radiation-SST feedback index (CSFI) is defined by regressing monthly net shortwave radiation anomalies to SST anomalies (Sun et al., 2003; Sun et al., 2006), to quantify the strength of shortwave feedbacks in the climate system. Figure 5a shows the spatial distribution of the CSFI in the historical run. The CSFI is negative in most parts near the equator, suggesting a negative convective cloud-shortwave-radiation-SST feedback, and positive over the eastern Pacific, indicating a positive stratus cloud-shortwave-radiation-SST feedback (Ramanathan and Collins, 1991; Song and Zhang, 2014). The negative (positive) cloud-SST feedback will suppress (enhance) the local SST warming. This process can be demonstrated by the changes in cloud amount (Fig. 5b). Thus, the cloud-shortwave-radiation-SST feedback weakens the zonal gradient of SST, contributing to an El Niño-like pattern.

The changes in ocean heat transports associated with the ocean current changes (Figs. 6a-c) indirectly reflect the effect of the changes in atmospheric general circulation connected by the surface wind stress changes. The effects of changes in ocean zonal and vertical currents both warm the SST along the equator (Figs. 6a, c), which is associated with the weakened Walker circulation (Vecchi and Soden, 2007). However, the zonal current changes do not contribute much to the zonal gradient of SST changes (Fig. 6a) because of the near uniform zonal current changes (Fig. 7a). Meanwhile, the downwelling changes in the eastern Pacific (Fig. 7b)——weakening the cold upwelling and warming the SST——mainly represent the effect of weakened Walker circulation on the zonal gradient of SST changes (Fig. 6c and ∆ Q _w1). The effect of changes in meridional current also warms the SST in the eastern Pacific around 5^°N (Fig. 6b) with a relatively weak magnitude, which could be attributed to the weak weakening of the meridional overturning circulation (Vecchi and Soden, 2007; Ma and Xie, 2013).

The ocean dynamical thermostat effect can be represented by changes in the ocean heat transports due to changes in ocean vertical temperature gradients (Figs. 6f) (Cane et al., 1997; Seager and Murtugudde, 1997; An and Im, 2014). Under global warming, the ocean vertical temperature gradients will increase (Fig. 7b), with less solar radiation absorbed in the subsurface than at the surface. Thus, the background upwelling pulls up cooler subsurface water to cool the surface in the eastern Pacific, damping the El Niño-like pattern (Fig. 6f).

The energy budget analyses basically verify that the previous suggested mechanisms are pronounced in the MME of the 32 CMIP5 models. However, they also exhibit great discrepancies in spatial structure and strength (Figs. 2f, 4c and d, and 6). The effects of weakened Walker circulation (Fig. 6c) and ocean dynamical thermostat (Fig. 6f) are confined near the equator (2.5^°S-2.5^°N), with great horizontal gradients, because of the narrow upwelling and stratification region in the eastern Pacific. Whereas, the effects of climatological evaporation and cloud radiation feedback extend to 5^°S-5^°N, close to the structure of the SST change pattern.

The effect of climatological evaporation, cloud radiation feedback, the weakened Walker circulation, and the ocean dynamical thermostat can be represented by the zonal differences between the eastern (5^°S-5^°N, 145^°-85^°W) and western (5^°S-5^°N, 125^°E-175^°W) Pacific of ∆ Q _EO2, ∆ Q _SW, ∆ Q _w1 and ∆ Q _w2, respectively. The climatological evaporation contributes the most to the El Niño-like pattern with the east-west difference exceeding 2 W m^-2 (around 2.03 W m^-2), while the ocean dynamical thermostat contributes a comparable damping to the El Niño-like pattern formation (-1.96 W m^-2). The cloud-shortwave radiation-SST feedback (0.92 W m^-2) and the weakened Walker circulation (0.59 W m^-2) play a positive but relatively small role.

Figure 7.  (a) Changes in horizontal currents averaged in the mixed layer (vectors less than 0.02 m s$^-1$ are omitted). (b) Vertical gradients of changes in ocean temperature (color shading) and the zonal overturning current (vectors; m s$^-1$) at the equator (averaged between 2.5$^\circ$S and 2.5$^\circ$N). Changes in vertical velocity are multiplied by 100 for display, and vectors less than 0.05 are omitted.

Figure 8.  (a) Zonal mean of $\Delta T'_ss$ (multiplied by 10; units: K) and the terms with positive contribution to the EPW pattern (units: W m$^-2$ K$^-1$). (b) The terms with negative contribution to the EPW pattern (units: W m$^-2$ K$^-1$).

The meridional EPSW exhibits a peak warming at the equator (Fig. 8a). Three terms of the zonal-mean heat budgets peak at the equator, favoring the EPW pattern (Fig. 8a): \(\alpha Q'_E\langle \Delta T_ss\rangle\), representing the effect of climatological evaporation; ∆ Q' _u, representing the effect of changes in ocean zonal heat transport; and ∆ R', representing the effect of changes in the ocean residual term. On the other hand, the changes in the shortwave radiation (∆ Q' _SW), RH and stability (∆ Q' _ER), the meridional heat transport (∆ Q' _v), and the vertical heat transport (∆ Q' _w), damp the EPW pattern (Fig. 8b).

Among the mechanisms, the latent heat changes due to the effect of the climatological evaporative cooling is the greatest positive contribution to the EPW pattern (Fig. 8a), which was first mentioned by (Liu et al., 2005) and emphasized by (Xie et al., 2010). Another important positive factor in the present analysis, which has not been emphasized, is the effect of changes in the ocean zonal heat transport due to the weakened Walker circulation (yellow curve in Fig. 8a), as demonstrated in Figs. 6a and 7a. This result is inconsistent with that in (Liu et al., 2005), suggesting the changes in oceanic circulation are not important. The residual term (∆ R') involving sub-grid scale processes, such as the ocean vertical mixing, also has a positive contribution to the EPW pattern, although its meridional range is relatively small. Meanwhile, these favorable mechanisms are balanced mainly by the effects of changes in the ocean vertical heat transports due to enhanced oceanic vertical temperature gradients and the latent heat changes due to changes in the atmospheric RH and stability (Fig. 8b). It should be noted that the effects of changes in shortwave radiation (Liu et al., 2005) and surface wind speed (Seager and Murtugudde, 1997), believed to be positive in forming the EPW pattern, do not contribute to the EPW pattern positively. The former damps the EPW pattern, while the latter mainly affects the off-equatorial patterns.

This paper analyzes the changes in the mixed-layer energy budget using 32 CMIP5 models, to investigate the formation mechanisms of the annual-mean equatorial Pacific SST warming patterns. Discussed are two patterns that are pronounced but whose mechanisms are unclear: the zonal El Niño-like pattern and the meridional equatorial peak pattern.

For the El Niño-like pattern, we examined the effects of climatological evaporation, the cloud-shortwave-radiation-SST feedback, the weakening of the Walker circulation, and the ocean dynamical thermostat. The quantitative energy budget analyses, based on the MME of the CMIP5 models, revealed that the effect of climatological evaporation plays a major role, while the cloud-shortwave-radiation-SST feedback and the weakened Walker circulation play relatively small roles. On the contrary, the effect of the ocean dynamical thermostat plays a major negative role, damping the El Niño-like pattern formation, with comparable magnitude to the effect of climatological evaporation. The effects of climatological evaporation and the cloud-radiation feedback on the equator extend much wider meridionally than those of the effects associated with ocean dynamics.

For the meridional EPW pattern, the dominant role of the climatological latent heat flux is also apparent in the MME of the 32 CMIP5 models, as in (Xie et al., 2010). Nevertheless, the performances of some mechanisms evaluated in the present study are different from those in some previous studies. The changes in the zonal heat transport due to the weakened Walker circulation make a considerable positive contribution to the EPW pattern, which is inconsistent with the result in (Liu et al., 2005). Moreover, the effect of changes in shortwave radiation damps the EPW pattern, which is inconsistent with the positive role proposed by (Liu et al., 2005), while the effect of surface wind speed mainly influences the off-equatorial patterns, which is also inconsistent with the positive role proposed in (Seager and Murtugudde, 1997).

The present study is based on the MME of 32 CMIP5 models' outputs. The inter-model spreads in the EPSW are quite large in current CMIP models (DiNezio et al., 2009; Huang and Ying, 2015), with great impacts on the uncertainties in projecting regional climate changes (Huang et al., 2013; Ma and Xie, 2013). The present energy budget analysis provides a useful method to study the importance of the mechanisms to the inter-model uncertainty in the EPSW, which is worthy of study in the future.