5.1. Sea Surface Temperature
 El Niño- or La Niña-like SST patterns are generally interpreted as a weaker or stronger Walker circulation in GCM experiments [e.g., Knutson and Manabe, 1995; Meehl et al., 2007], observations [e.g., Cane et al., 1997], or paleoreconstructions of tropical Pacific climate changes [Lea et al., 2000; Koutavas et al., 2002]. Recent studies indicate that an enhanced equatorial warming (EEW) pattern is a more robust indication of a weaker Walker circulation in AGW experiments rather than the zonal gradient [Liu et al., 2005; DiNezio et al., 2009]. The six models analyzed here simulate EEW evidenced in the zonal mean warming (Figure 9b) but do not agree in the changes in the zonal gradient (Figure 9a).
 The opposite response, enhanced equatorial cooling (EEC), would indicate a stronger Walker circulation in LGM experiments. However, the LGM experiments do not exhibit a consistent relationship between EEC and a stronger Walker circulation. In contrast to the 2xCO2 experiments, two different patterns associated with changes in the winds can be identified in the zonal (Figure 9c) or meridional (Figure 9d) SST changes: (1) EEC resulting from increased ocean dynamical cooling due to stronger zonal currents in response to a stronger Walker circulation and (2) a north-south gradient in cooling resulting from enhanced (reduced) ocean cooling in the NH (SH) equatorial Pacific due to the anomalous northward meridional currents forced by the anomalous cross-equatorial winds.
 Two out of the four models with a stronger Walker circulation simulate an EEC pattern. In the remaining two models, the interhemispheric gradient may dominate over the EEC response despite the stronger equatorial trade winds. In FGOALS-g1.0 and IPSL-CM4, the ocean response to cross-equatorial winds is much weaker compared with the zonal changes, especially in FGOALS-g1.0, which shows the largest increase (20%) in equatorial zonal wind stress, explaining why the EEC pattern is evident in the SST changes, just opposite to the 2xCO2 experiments. In contrast, CCSM3.0, a model that simulates a stronger Walker circulation, exhibits unchanged zonal and meridional equatorial gradients in SST, and no indication of EEC. MIROC3.2 simulates a 30% stronger Pacific Walker circulation (measured by the east-west SLP difference between the western and eastern Pacific), but the trade winds strengthen by just 10%, perhaps explaining why the EEC pattern cannot be distinguished from the interhemispheric pattern.
 In both 2xCO2 and LGM experiments, the changes in subsurface thermal structure on the equatorial Pacific exhibit changes related to changes in wind forcing; however, the link is not straightforward because the equatorial thermocline responds to both dynamical and thermodynamical forcing, local and remote. The six models simulate robust changes in thermal structure in the equatorial Pacific in response to 2xCO2, with a minimum in warming in the western Pacific at depths of about 150 m where the thermocline is located (Figure 10a). The changes in thermal stratification, ∂T/∂z, in response to 2xCO2 forcing (Figure 10b) show a sharper thermocline and a zonal mean shoaling of the thermocline(Figure 10b, reds).
Figure 10. Response of the equatorial thermocline to 2xCO2 and LGM forcing. Multimodel change in equatorial Pacific Ocean (a) temperature and (b) vertical temperature gradient (∂T/∂z) simulated by six coupled GCMs in response to 2xCO2 forcing. Multimodel change in equatorial Pacific ocean (Figures 10c and 10e) temperature and (Figures 10d and 10f) ∂T/∂z simulated by (Figures 10c and 10d) MIROC3.2, FGOALS-g1.0, and IPSL-CM4 (three-model ensemble mean) and by (Figures 10e and 10f) CCSM3.0 in response to LGM forcing. These four models simulate a stronger Walker circulation in response to LGM forcing. The changes in CCSM3.0 are shown separately because this model's subsurface response is substantially different than that of the remaining three models with a stronger Walker circulation. Stippling is as in Figure 1. In Figures 10a, 10c, and 10e the dash-dotted lines indicate the 20°C (2xCO2) and 18°C (LGM) isotherms in the control (black) and forced (red) experiments. In Figures 10b, 10d, and 10f the dash-dotted lines indicate the depth of the thermocline, i.e., the maximum of ∂T/∂z, in the control (black) and forced (blue) experiments. The equatorial sections are averaged over the 2°S and 2°N latitude band. Contours show multimodel annual mean temperature and ∂T/∂z simulated in the control experiment. CI = 2 K (temperature) and CI = 2.5 10−2 K m−1 (∂T/∂z).
Download figure to PowerPoint
 The depth of the 20°C isotherm, Z20, is often used to explore interannual changes in the depth of the thermocline, ZTC, in the tropical Pacific. However, this is problematic in multimodel climate change analyses because (1) the isotherm that is representative of the thermocline varies from model to model and (2) the isotherms that are representative of the thermocline will change as the climate warms or cools. A more general method is to define ZTC as the vertical location of the maximum vertical temperature gradient. To illustrate this difference we plot Z20 in the control and 2xCO2 experiments overlaid on the temperature changes (Figure 10a, blue and black dashed lines) and ZTC in the control and 2xCO2 experiments overlaid on the ∂T/∂z changes (Figure 10b, blue and black dashed lines). The changes in Z20 are quite different from the ZTC changes, with changes in Z20 having a somewhat larger signal in the eastern basin, whereas the simulated ZTC changes are larger in the western Pacific.
 The 2xCO2 changes in ZTC (Figure 10b, dashed lines) are consistent with the equilibrium response to weaker zonal winds, consisting of a zonal mean shoaling of the thermocline in response to the curl of the wind, in addition to the relaxation of the thermocline tilt [DiNezio et al., 2010; Clarke, 2010]. In the eastern Pacific, the zonal mean shoaling of the equatorial thermocline opposes the deepening due to a reduced tilt resulting in negligible changes. The two effects contribute to a shoaling of the thermocline in the western Pacific, bringing cooler waters upward, thus explaining the minimum in subsurface warming (Figure 10a).
 In contrast, the LGM changes in the equatorial thermocline do not exhibit a robust minimum in cooling in the western Pacific expected from a stronger Walker circulation. Three out of the four models that simulate a stronger Walker circulation in the LGM experiments simulate a minimum in subsurface cooling in the western Pacific (Figure 10c) consistent with a long-term response of the thermocline to stronger trade winds, much like the subsurface minimum in warming in the 2xCO2 experiments (Figure 10a). The remaining model, CCSM3.0, simulates a near opposite response, with a subsurface maximum in cooling with a magnitude about 1 K larger than the surface (Figure 10e). The fact that the subsurface cooling is greater than the tropical mean surface cooling strongly suggests that this anomaly originated in the extratropics, where surface cooling is stronger (Figure 1b). Nonetheless, this model simulates a stronger thermocline tilt (Figure 10f, dashed lines) in equilibrium with the stronger wind forcing in this model (Figure 7, blue dot number 3).
 In models with a stronger LGM Walker circulation the vertical temperature gradient ∂T/∂z, exhibits increased (decreased) thermal stratification below (above) the thermocline (Figure 10d). These changes in stratification reflect a zonal mean deepening of the thermocline, which according to equatorial adjustment theory, occurs once Rossby waves adjust the equatorial ocean establishing a Sverdrup response to the WSC changes [Vecchi et al., 2006; Vecchi and Soden, 2007a; DiNezio et al., 2010; Clarke, 2010]. In the eastern Pacific, this response opposes the shoaling due to a stronger tilt resulting in negligible ZTC changes there. This is clearly shown by the dash-dotted lines in Figure 10d indicating ZTC in the control (black) and LGM (blue) climates. This has implications for the coupled response, because in contrast with La Niña events, the thermocline does not shoal in central and eastern Pacific. Therefore the positive feedback loop between winds, SST, and thermocline depth (i.e., the Bjerknes feedback) breaks down.
 In the LGM experiments the depth of the 18C isotherm, Z18 (Figure 10c, dash-dotted lines) does not change in the west due to competing effects between the surface cooling shoaling this isotherm, and the dynamical response to the winds, deepening the thermocline. In the eastern Pacific, the shoaling of Z18 is due to the surface cooling, because the thermocline changes are negligible there. In general, due to competing dynamical and thermodynamical effects the changes in ZTC do not agree with the changes in Z18, or any other isotherm that lies the in thermocline in the present climate.
 Apart from the minima in warming or cooling in the western Pacific, the 2xCO2 changes in temperature and thermal stratification exhibit warming that is intensified near the surface. In contrast, the LGM experiments show cooling that is more uniformly distributed throughout the water column, with no differences between the surface and the deep ocean. The intensification of the 2xCO2 ocean warming is not unexpected because the deep ocean takes centuries to adjust to the surface forcing [Hewitt et al., 2003]. The AGW experiments are too short to achieve a completely equilibrated deep ocean. In contrast, much longer integrations are performed to ensure that the deep ocean is equilibrated in the LGM experiments. This explains why the LGM experiments do not exhibit surface intensified cooling or a sharper thermocline. In other words, in the 2xCO2 experiments, the surface enhanced warming can be thought of as a transient response as proposed by DiNezio et al.  and not as a result of anomalies subducted from the extratropics as proposed by Seager and Murtugudde . Moreover, the lack of anomalous stratification in the LGM experiments, suggests that the ocean dynamical thermostat mechanism proposed by Clement et al.  does not play a dominant role in the response of the equatorial Pacific to LGM forcing. In contrast, this mechanism operates in the AGW experiments due to the transient increase in stratification, but without driving the equatorial Pacific into a La Nina-like state [DiNezio et al., 2009, 2010].
 Another difference arising from the different time scales of the AGW and LGM responses is that changes from remote areas of the ocean have sufficient time to reach the equatorial Pacific in the LGM experiments. This could explain the enhanced subsurface cooling in CCSM3.0 (Figure 10e), which could originate from surface anomalies subducted from the extra tropics. Analysis of the temperature changes suggests that this anomaly originates in the NH (not shown), in disagreement with model and observational analysis that have shown that only anomalies subducted in the SH can influence the equatorial thermocline [Liu et al., 1994; Lu et al., 1998; Shin and Liu, 2000]. The spatial features of this anomaly, which can be traced back to the NH midlatitude thermocline, suggest that is a response to the surface forcing and unlikely to be an artifact of the deep ocean adjustment. This is the only model that simulates this type of remote response, but the implications for the equatorial climate warrants further investigation of the underlying mechanisms.
 According to the GCM experiments neither the La Niña analog nor the EEC pattern provide a framework from which to infer LGM changes in the Walker circulation from SST proxies. For instance, in each model the relative changes in zonal SST gradient are generally uncorrelated with relative changes in the equatorial trade winds (Figure 11a) both for 2xCO2 and LGM experiments. The intermodel changes in the Z18 tilt, a measure of the depth of the thermocline frequently used in the paleoceanographic literature [e.g., Andreasen and Ravelo, 1997], are less consistent with the changes in the trade winds (Figure 11c) than the changes in ZTC tilt. The changes in Z18 tilt are consistent with the changes in zonal winds in only three models. Among the models where the tilt of Z18 is not a good proxy for the trade winds, HadCM3M2 exhibits a large departure from the one-to-one relationship. In contrast, five out of the six models simulate changes in ZTC tilt that are consistent with the wind changes (Figure 11b).
Figure 11. Ocean proxies for equatorial Pacific trade winds. Relative changes in equatorial Pacific trade winds (τx) in response to 2xCO2 forcing (red) and to LGM (blue) forcing versus relative change in (a) zonal sea surface temperature gradient (dSST), (b) zonal gradient of the depth of the thermocline (dZTC), (c) zonal gradient of the depth of the 18°C isotherm (dZ18), and (d) depth of the thermocline in the western Pacific (ZTC west). In Figures 11a–11d, the changes in τx, dSST, dZTC, and dZ18 are zonal averages over the equatorial band 2°S–2°N, 150°E–90°W. The changes in ZTC on the western Pacific are averaged over 2°S–2°N, 150°E–180°.
Download figure to PowerPoint
 On the equator, where the Coriolis force vanishes, and in the absence of friction, the changes in thermocline tilt and zonal winds have to balance according to:
where g′ is the reduced gravity, a measure of density contrast between the upper ocean and the deep ocean, hx is zonal gradient of the depth of thermocline, τx is the zonal component of the wind stress, ρ0 is a reference density of seawater, and H is the depth of the Ekman layer. For the LGM experiments, the changes in thermocline sharpness, i.e., the maximum ∂T/∂z, are negligible (not shown) because the ocean climate is fully equilibrated; thus, g′ can be taken as constant. The following one-to-one relation ship between the relative changes in thermocline tilt, Δhx/hx, and wind stress, Δτx/τx, results from (1) assuming a constant Ekman layer, H:
The changes in the zonally averaged zonal component of the surface wind stress and the zonally averaged tilt of the thermocline simulated by the GCMs agree with this scaling exhibiting a very close one-to-one relationship (r2 = 0.84, Figure 11b, blue dots). The 2xCO2 experiments simulate an increase in thermocline sharpness, g′ which according to (1) requires a smaller thermocline tilt to balance a given wind forcing. Thus, if the wind weakens, the thermocline tilt has to relax further to remain in balance with the wind. Conversely, due to the increased stratification, the thermocline tilt could relax, even in the absence of weaker winds. The GCM experiments exhibit larger (relative to the one-to-one relationship) changes in thermocline tilt than in winds consistent with this argument (Figure 11b, blue circles). Because the changes in thermocline depth are negligible in the east, the relative changes in the thermocline depth in the western Pacific also follow a one-to-one relationship with wind changes (Figure 11d), although they are still biased about the one-to-one line for the 2xCO2 case. Therefore, observational or proxy estimates of the thermocline depth in the western equatorial Pacific could provide information to constrain the changes in the Walker circulation.
5.3. Sea Surface Salinity
 The changes in precipitation, ΔP, simulated by all six models in response to LGM forcing (Figures 12c and 12d) are closely linked to the changes in atmospheric overturning circulation diagnosed by Δω500 (Figures 12a and 12b). Precipitation changes are both due to radiative, thermodynamical, and dynamical processes [Held and Soden, 2006]. The former is responsible for a tropical mean precipitation reduction (increase) in precipitation by about 2.7% K−1 (1.5% K−1) associated with tropical mean cooling (warming) (Figure 4b), but in some regions precipitation increases due to changes in circulation. In the LGM experiments, precipitation increases in regions where ascending motion increases (Δω500 < 0), such as the SPCZ, and decreases in regions where subsidence increases (Δω500 > 0), such as the central and eastern Pacific. The spatial correlation between the multimodel mean Δω500 and ΔP over the tropical Indo-Pacific is –0.75 ± 0.12 with values ranging from −0.6 to −0.89 (Table 2) confirming that the spatial patterns of Δω500 and ΔP agree quite well.
Figure 12. LGM changes in atmospheric overturning circulation, precipitation, and sea surface salinity. Two-model ensemble mean change in (a and b) midtropospheric vertical velocity (Δω500), (c and d) precipitation (ΔP), and (e and f) sea surface salinity (ΔSSS) simulated by coupled general circulation models (GCMs) in response to LGM forcing. Figures 12a, 12c, and 12e correspond to the GCMs that simulate a much stronger (30%) Pacific Walker circulation in the LGM experiment (FGOALS-g1.0 and MIROC3.2). Figures 12b, 12d, and 12f correspond to the GCMs that simulate no changes in the Pacific Walker circulation in the LGM experiment (HadCM3M2 and GFDL-CM2.1). Stippling indicates where the two models agree in the sign of the changes. Contours show multimodel annual mean fields simulated in the control experiment. Contour intervals are 10 hPa d−1, 1 mm d−1, and 0.2 psu for ω500, precipitation, and SSS, respectively. Note that the Δω500 and ΔP color scales are not linear.
Download figure to PowerPoint
Table 2. Spatial Correlation Coefficient Between the Changes in Midtropospheric Vertical Velocity (Δω500), Precipitation (ΔP), Evaporation Minus Precipitation [Δ(E − P)], and Sea Surface Salinity (ΔSSS) Over the Tropical Indo-Pacific (40°E–90°W, 25°N–25°N)
|Model||r(Δω500, ΔP) 40°E–90°W, 25°S–25°N||r(Δ(E − P), ΔP) 40°E–90°W, 25°S–25°N||r(ΔP, ΔSSS) 40°E–90°W, 25°S–25°N|
 Substantial differences in the spatial patterns of Δω500 and ΔP can be identified among the models. HadCM3M2 and GFDL-CM2.1 simulate large reductions of precipitation over the Maritime continent (Figure 12d) coincident with the regions with anomalous subsidence over exposed land due to lowered sea level (Figure 12b). This weakening of the ascending branch of the Walker circulation could force remote changes in the tropical circulation, such as the eastward shift in the SPCZ [Kodama, 1999] and the increase in ascending motion over the western Indian ocean as suggested by Δω500 (Figure 12b). In contrast, the models that exhibit the largest (30%) strengthening of the Walker circulation (FGOALS-g1.0 and MIROC3.2) do not simulate large changes in precipitation over the Maritime continent (Figure 12c), possibly due to compensating thermodynamical (i.e., drying due to the cool surface) and dynamical (i.e., a strengthened Walker circulation) effects. Moreover, these models do not simulate the large zonal shifts in precipitation in the Indian ocean and in the SPCZ, possibly due the lack of east-west shifts in convection over the Maritime continent.
 The close relationship between circulation and precipitation changes suggests the possibility for detecting changes in the Walker circulation in precipitation proxies. However, the simulated patterns of precipitation change fail to translate into consistent patterns of sea surface salinity change, ΔSSS (Figures 12e and 12f), a variable that can be reconstructed from proxy data. This is not unexpected since SSS is influenced by evaporation and ocean advection, in addition to precipitation [Stott et al., 2002; Oppo et al., 2007]. In all models, the changes in evaporation minus precipitation, Δ(E − P), are dominated by ΔP as indicated by the large spatial correlations (r2 > 0.90, Table 2) suggesting that the ΔSSS results either from changes in precipitation or from changes in ocean circulation, with a much lesser role for evaporation.
 The spatial correlation between the multimodel mean ΔP and ΔSSS over the tropical Indo-Pacific is −0.47 ± 0.20 with values ranging from −0.31 to −0.67 (Table 2). The fact that a large fraction of the spatial variance of ΔSSS is not explained by the Δ (E − P) suggests that changes in ocean salt advection are equally important in determining the patterns of ΔSSS in the LGM. However, some features of ΔSSS could be directly linked to ΔP. For instance the large reduction in precipitation over the Maritime Continent simulated by HadCM3M2 and GFDL-CM2.1 in the LGM experiments has a clear signature in the salinity of the southeastern Indian Ocean where SSS increases by more than 1 psu (Figure 12f). These models also simulate a fresher western Indian ocean where precipitation increases (Figure 12d). Overall these patterns indicate that SSS proxies could be used to constrain the changes in the Indian Walker circulation and shifts in the SPCZ.