Global warming may change the thermocline water pathways and transports from the subtropics to the tropics in the Pacific Ocean, which are known to have profound implications for the El Niño-Southern Oscillation (ENSO) and thereby global climate. This study investigates the changes by comparing solutions between a present-day climate and a future, warmer climate from a set of Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) models. As the climate warms, although the total transport from the subtropics to the tropics exhibits no significant change, transport via western boundary pathways increases and via interior pathways decreases. This shift is due to high potential vorticity (PV) zones that extend further westward, thus dynamically guiding thermocline water away from interior pathways to prefer western boundary pathways from the subtropics to the tropics. Additionally, a warmer climate induces a large temperature increase near the sea surface in the eastern tropics and a significantly enhanced Equatorial Undercurrent (EUC) in the western and central Pacific; the former is related to the decreased transport through interior pathways and the latter is linked to the increased transport through western boundary pathways. Implications of the results of this study are also discussed.
 The Pacific Subtropical Cells (STCs) are shallow meridional circulation cells in which water flows out of the tropics within the surface layer, subducts in the subtropics, flows equatorward within the thermocline, and upwells in the eastern equatorial ocean [e.g., McCreary and Lu, 1994]. The subtropical-tropical thermocline water pathways (i.e., subsurface branches of the STCs) include a direct path via the ocean interior and an indirect path via the western boundary [e.g., Rothstein et al., 1998], and are asymmetrical between the North and South Pacific with more water reaching the equator through interior pathways from the South Pacific [Luo et al., 2005].
 Decadal variations of Pacific STC transports since the 1950s have been observed and are believed to be responsible in part for decadal changes of observed SSTs in the eastern tropics [McPhaden and Zhang, 2002]. This relationship has been recently explored using eighteen model simulations of 20th century climate from fourteen models of the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4). A significant correlation exists between the STC transports and tropical SST over the last half century in the majority of the models [Zhang and McPhaden, 2006]. Meanwhile, climate model studies have shown that global variability in water temperature during the 20th century is caused by a combination of changes in external forcing, such as fluxes associated with changes in greenhouse gas concentrations and solar activity, as well as intrinsic internal variability of the climate system [e.g., Barnett et al., 2005; Meehl et al., 2004].
 Using a set of IPCC AR4 model simulations of 21th century climate, Vecchi and Soden  recently examined the response of the tropical atmospheric and oceanic circulation to increasing greenhouse gases. They found that modeled changes over the tropical Pacific consist of a weakening of the Walker circulation and eastward shift of convection over the central equatorial Pacific, a reduction in sea level pressure gradients across the equator, a reduction in the tilt of the Pacific thermocline and shoaling of the thermocline depth in the western Pacific, and a preferential warming of SSTs along the eastern equatorial Pacific.
 Based upon coupled climate models under future anthropogenic forcing scenarios, significant changes in the Pacific Ocean also include a rise in sea surface level [e.g., Landerer et al., 2007], an increase in water temperature [e.g., Overland and Wang, 2007], and a speedup of planetary waves [e.g., Fyfe and Saenko, 2007]. These changes clearly impact the thermocline structure and thus the water pathways and transports within the thermocline from the subtropics to tropics. However, it is not clear how the thermocline water pathways and transports respond to global warming. In the present research we investigate this subject by comparing solutions from a set of IPCC AR4 coupled models between a present-day climate and a future, warmer climate.
2. Data and Method
 Simulations from the IPCC 20C3M and the SRESA1B experiments are selected to represent the present-day climate and the future, warmer climate, respectively. The 20C3M experiments simulate the 20th century climate where atmospheric CO2 concentrations and other input data are based upon historical records; the SRESA1B scenarios represent a climate change in which atmospheric CO2 concentrations double their present-day level in the year 2100 and are then held fixed and maintained to years 2200 or 2300. For this study, the 1951–2000 means are used for 20C3M and the 2151–2200 means are used for SRESA1B. Eleven climate models (CGCM3.1_T63, CNRM-CM3, CSIRO-MK3.0, FGOALS-g1.0, GFDL-CM2.0, GFDL-CM2.1, HADCM3, IPSL-CM4, MIROC3.2_medres, MRI-CGCM2.3.2, and PCM1) are examined in this study, chosen because they have sufficient output available for the SRESA1B scenarios at the Program for Climate Model Diagnosis and Intercomparison (PCMDI) IPCC data center. For each of the models only one member run is analyzed; usually ‘run1’.
 The volume transport in the models is calculated as the equatorial convergence in the upper pycnocline across 9°S and 9°N, following Zhang and McPhaden . The total STC transport is integrated across the entire basin, the interior transport is the sum of the equatorial transport integrated from the eastern boundary to 165°E at 9°S and to 135°E at 9°N, and the western boundary transport is the difference between the total transport and the interior transport. The upper thermocline is defined as the range of density classes that encompasses equatorward moving water masses below the surface mixed layer or the Ekman layer, whichever is deeper. Density classes defined this way are generally 22.5 < σt < 26.0 Kg m−3 for 20C3M and 21.5 < σt < 25.5 Kg m−3 for SRESA1B.
 In the tropics, an isotherm is often used as a proxy for the thermocline. For instance, in the studies of Philip and van Oldenborgh , the 20°C isotherm for 20C3M and 24°C isotherm for SRESA1B are used for the tropical Pacific. However, as pointed out by Vecchi and Soden  this type of metric is problematic in multi-model climate analyses. For this study we define the thermocline depth more traditionally as the location of the maximum vertical gradient of temperature.
 Because model results depend on how well the models represent the pycnocline structure as well as the surface forcing, a large variation of STC transports is found among models (Figure 1). For the 20C3M runs, the CGCM3.1_T63, CNRM-CM3, CSIRO-MK3., GFDL-CM2.1, and PCM1 models have the best representation of the STC transports in both the interior and the western boundary with respect to observations. The other models also simulate the STC transports reasonably well, with the notable exception of the HADCM3 model.
 As the climate warms there is a significant redistribution of equatorward transport between interior and western boundary pathways, but a relatively small change in total transport. An enhancement of transport through western boundary pathways is seen in all of the eleven models, and a reduction of transport through the interior pathways in most of them (except there is no change for CSIRO-MK3.0 and some increase for IPSL-CM4). By averaging the results from the eleven models, it is found that there is ∼3.0 Sv (from 26.0 to 29.0 Sv) more water channeled through the western boundary pathways and ∼3.2 Sv (20.0 to 16.8 Sv) less water through the interior pathways towards the equator in the thermocline, resulting in no significant change in the total transport (only 0.2 Sv reduction from 46.0 to 45.8 Sv). In terms of hemispheric changes in the warmer climate, in the North Pacific there is an increase of ∼0.7 Sv (from 13.2 to 13.9 Sv) for the western boundary transport and a decrease of ∼1.0 Sv (from 8.7 to 7.7 Sv) for the interior transport, resulting in a decrease of ∼0.3 Sv (21.9 to 21.6 Sv) in the total transport. In the South Pacific there is an increase of ∼2.3 Sv (from 12.8 to 15.1 Sv) for the western boundary transport and a decrease of ∼2.2 Sv (from 11.3 to 9.1 Sv) for the interior transport, resulting in an increase of 0.1 Sv (24.1 to 24.2 Sv) in the total transport.
 We next examine the changes of the thermal and density structures in order to understand why there are such changes in the thermocline water pathways and transports. The future warming scenario forces greater ocean warming near the surface around the equator as well as in the latitudes poleward of 30° in both hemispheres (Figure 2). Generally, the warming decreases with depth. However, the least warming is found subsurface in the latitudes between 10°S and 10°N, and the subsurface water in the latitudes between 10° and 20° in both hemispheres is warmer than the surrounding water in the same layer, with the maximum warming surprisingly found in the North Pacific centered around 12°N at 70m depth.
 The uneven warming both in the vertical and in the horizontal plays a large role in shaping the density field. As an example, significant pycnocline structural changes can clearly be seen for the density fields along 180° (Figure 3a). In the warmer climate, the density contrast between the top and bottom of the pycnocline is larger (e.g., the upper pycnocline range around 10°N is 22.5 < σt < 26.0 Kg m−3 for 20C3M and 21.5 < σt < 25.5 Kg m−3 for SRESA1B, respectively) and the pycnocline is therefore intensified. Additionally, around 10–20° in both hemispheres, the pycnocline becomes thicker and the isopycnal surfaces are somewhat steeper, resulting mainly from the subsurface water there gaining more heat (Figure 2). These changes increase the tendency of thermocline water in the subtropics to move further west before reaching the tropics through interior pathways, consistent with our findings above on the STC pathways and transports.
 Potential vorticity (PV) can be used to understand dynamical constraints on STC water pathways [e.g., Rothstein et al., 1998]. As the thermocline structure is significantly altered in the SRESA1B, high PV zones around 10° in both hemispheres are extended further westward (Figure 3b). These zonally extended high PV zones constrain water pathways to also extend westward because geostrophic inviscid flow follows PV contours. Along with these changes on isopycnal surfaces, the current response to the warming indeed provides evidence for more water reaching the equator through western boundary pathways and less water through interior pathways. In addition, the Equatorial Undercurrent (EUC) is found to be intensified in the western and central Pacific and be weakened in the eastern Pacific. This change in the EUC may be linked to changes in the STC pathways and transports, which will be further discussed together with Figure 3d.
 Under the warmer climate scenario, the thermocline structure along the equator also experiences a considerable shift (Figure 3c). The temperature difference shows warming at all levels for the SRESA1B experiments, but with greater warming near the surface (as high as 3.1°C in the eastern Pacific) as well as beneath the thermocline, and less warming within the thermocline (as low as 0.52°C in the western Pacific). Since the thermocline water along the equator originates from the subtropics where the surface water temperature is relatively lower than in the tropics, the warming minimum in the western Pacific thermocline may be related to the increased EUC there which, in turn, may be linked to the increased western boundary transport (see discussions with Figures 3b and 3d). According to McPhaden and Zhang  and Zhang and McPhaden , the interior transport has a greater impact on eastern Pacific tropical SST than the western boundary transport. Comparing SRESA1B with 20C3M, associated with this change in the upper ocean temperature is an intensification of the stratification, a sharpening of the thermocline, and a shoaling of the mean depth of the thermocline by about 25 m in the western and central Pacific, and hence a reduction in tilt of the thermocline. These findings are in good agreement with previous studies [Vecchi and Soden, 2007].
 Accompanying these changes in upper ocean thermal structure is a change in the ocean circulation (Figure 3d). The major current system in the tropical Pacific is captured well by the models, including the South Equatorial Countercurrent (SECC) between 11°S and 6°S, the South Equatorial Current (SEC) between 6°S and 4°N, the North Equatorial Countercurrent (NECC) between 4°N and 9°N, the North Equatorial Current (NEC) between 9°N and 20°N, and the EUC between 3°S and 3°N. However, the eastward Subsurface Countercurrents (SCCs), or Tsuchiya Jets, are not obvious in the models, whereas in direct observations at the dateline they appear as lobes off the EUC with a maximum speed of ∼20 cm/s [Johnson et al., 2002].
 As the climate warms, the SECC, SEC, NECC, and NEC are all significantly weakened (also see Figure 3b), in response to the tropical atmospheric changes such as the weakened Walker circulation and the reduced sea level pressure gradients [e.g., Vecchi and Soden, 2007]. However, a complicated change can be seen in the EUC where there is an increase of eastward transport in the upper and central parts of the current but a decrease in its lower part. This structure of the circulation change suggests that the EUC anomaly can not be simply attributed to a local response of the equatorial ocean to the relaxation of the trade winds along the equator. If local dynamics were the sole determining factor, one would anticipate a weakening of the EUC as a result of the reduction in the zonal pressure gradient. However, the models show weakening only below the core of the EUC with an enhancement in the upper and central portions of the EUC. Along 180°, the EUC transport is increased by ∼2.6 Sv from 24.7 Sv in 20C3M to 27.3 Sv in SRESA1B. Together with the evidence from the current response on isopycnal surfaces (Figure 3b), we therefore suggest that the increased EUC may be linked to enhanced STC transports through the western boundary. In addition, a direct comparison between Figure 3d and Figure 2 suggests the temperature changes are related to the changes in the zonal currents. For example, the weakening of the westward SEC and NEC may have a relation with the greater warming in the upper ocean around the equator and around 10–15°N, respectively, and the intensification of the eastward EUC with the lower amplitude warming in the thermocline around the equator.
 Since there are considerable differences in the model configurations (e.g., resolutions and vertical mixing schemes), variations in the EUC and STC transports are indeed large among models, resulting in large uncertainties (Table 1). However, the responses in the EUC and STCs to global warming seem to be unanimous and robust across the models; these models are in general agreement on the changes in the future climate. For example, comparing SRESA1B with 20C3M there is an increase of STC transport through the western boundary in all of the eleven models; this increase is ∼12% on the average, and the transport has a trend of ∼1.4 Sv/100 yr in the SRESA1B experiments.
Table 1. Multimodel Ensemble Means and Standard Deviations of Transports Through the Western Boundary and Interior Pathways, and in the Equatorial Undercurrent Along 180°a
Means and standard deviations are in Sv.
 Changes in the Pacific STCs are believed to be one of many factors impacting the El Niño-Southern Oscillation (ENSO). Observations and previous studies based upon the 20C3M simulations have shown that the interior STC transport is strongly anti-correlated with SST anomalies in the central and eastern tropical Pacific [McPhaden and Zhang, 2002; Zhang and McPhaden, 2006]. Therefore, these changes in STC pathways and transports in the warmer climate, i.e., the subtropical-tropical thermocline water more likely moving through the western boundary than the interior, may influence future ENSOs.
 In addition, the thermocline change in the western tropical Pacific is also thought to have a profound influence on ENSO properties. After the 1976–77 climate shift, for example, intensification of the thermocline and shoaling of the mean depth of the thermocline in the western tropical Pacific were observed. These changes are consistent with a period of more active ENSOs [e.g., Zhang et al., 2007]. In the warmer climate, however, while mean state changes in the western tropical Pacific (i.e., the intensification of the thermocline and the shoaling of the mean depth of the thermocline) are largely consistent among these different IPCC AR4 models, the changes in ENSO properties are quite different with some models showing larger amplitude events, some showing smaller amplitude events, and some showing little change, the causes of which are in question [e.g., Guilyardi, 2006; Philip and van Oldenborgh, 2006].
 The key forcing mechanisms known to be important for the STC pathways and transports are wind stress and buoyancy forcing, both of which change significantly under the warmer climate. Comparing SRESA1B with 20C3M, we find that most of the eleven models show a weakening of trade winds in the North Pacific but a strengthening of trade winds in the eastern South Pacific. However, changes in surface heat and freshwater fluxes are quite different among the models. As discussed in section 3, in the warmer climate the PV around 10° in both hemispheres is increased due to intensified stratification from changes in buoyancy flux, which tends to reduce surface density. In addition to the buoyancy forcing, winds definitely help to set up the interior ocean density structure, and changes in Ekman pumping can be directly related to changes in stretching vorticity. Therefore, changes in winds are also likely to have a strong effect on the PV changes. The relative importance of buoyancy forcing and wind stress for the changes of the large-scale ocean circulations associated with the STCs needs to be examined further.
 We acknowledge the international modeling groups for providing their data for analysis, the Program for Climate Model Diagnosis and Intercomparison and the IPCC Data Archive at Lawrence Livermore National Laboratory for collecting, archiving, and making the data available. The authors wish to thank anonymous reviewers for their numerous comments that helped to improve the original manuscript. Y. Luo acknowledges the support of NSF grant OCE-0752194. R.-H. Zhang acknowledges the support of NSF grant ATM-0727668.