Variability of the Pacific subtropical cells under global warming in CMIP6 models*

Xue HAN, Junqiao FENG, Yunlong LU, Dunxin HU

1 Key Laboratory of Ocean Circulation and Waves, Chinese Academy of Sciences, Qingdao 266071, China

2 Institute of Oceanology, Chinese Academy of Sciences, Qingdao 266071, China

3 University of Chinese Academy of Sciences, Beijing 100049, China

4 Center for Ocean Mega-Science, Chinese Academy of Sciences, Qingdao 266071, China

5 Laboratory for Ocean Dynamics and Climate, Pilot National Laboratory for Marine Science and Technology (Qingdao),Qingdao 266237, China

Abstract The Pacific subtropical cells (STCs) are shallow meridional overturning circulations connecting the tropics and subtropics, and are assumed to be an important driver of the tropical Pacific decadal variability.The variability of STCs under global warming is investigated using multimodal outputs from the latest phase of the Coupled Model Inter-comparison Project (CMIP6) and ocean reanalysis products.Firstly, the volume transport diagnostic analysis is employed to evaluate how coupled models and ocean reanalysis products reproduce interior STC transport.The variation of heat transport by the interior STC under the high-emissions warming scenarios is also analyzed.The results show that the multimodal-mean linear trends of the interior STC transport along 9°S and 9°N are -0.02 Sv/a and 0.04 Sv/a under global warming, respectively, which is mainly due to the combined effect of the strengthened upper oceanic stratification and the weakening of wind field.There is a compensation relationship between the interior STC and the western boundary transport in the future climate, and the compensation relationship of 9°S is more significant than that of 9°N.In addition, compared with ocean reanalysis products, the coupled models tend to underestimate the variability of the interior STC transport convergence, and thus may lose some sea surface temperature (SST) driving force, which may be the reason for the low STC-SST correlation simulated by the model.The future scenario simulation shows that the heat transport of interior STC is weakened under global warming, with a general agreement across models.

Keyword: interior subtropical cell (STC); global warming; Coupled Model Inter-comparison Project(CMIP6); western boundary transport

1 INTRODUCTION

The Pacific subtropical cells (STCs) are oceanic channels connecting the tropical and subtropical ocean (Liu and Alexander, 2007).From the climatological mean perspective, STCs include poleward Ekman current on the ocean surface,equatorward western boundary current and interior flow (interior STC) at the upper pycnocline layers,and upwelling along the equator and at the eastern boundary (McCreary and Lu, 1994; Rothstein et al.,1998; Schott et al., 2004; Capotondi et al., 2005).STCs provide the main mechanism of tropicalsubtropical water mass exchange through interior STC and equatorial upwelling, play an important role in the redistribution of mass and heat between tropical and subtropical Pacific (Klinger and Marotzke, 2000), and significantly affect the interannual-decadal change of the tropical Pacific SST (e.g., McPhaden and Zhang, 2002, 2004;Nonaka et al., 2002; Lohmann and Latif, 2005).

Most studies have divided the mass transport of STCs to the equator into two parts: the western boundary and the interior STC (e.g., Fine et al.,1981, 1987; Capotondi et al., 2005; Cheng et al.,2007; Hong et al., 2014).On average, western boundary transport is much larger than interior STC transport, but the variation of western boundary transport is less than that of interior STC transport.The interior STC transport dominates the interannual-decadal variability of the tropical Pacific (Lee and Fukumori, 2003; Cheng et al.,2007; Lübbecke et al., 2008).

Previous studies proposed two mechanisms to explain the interdecadal variability in the tropical Pacific associated with STCs.Gu and Philander(1997) proposed that the advection of the thermal anomalies originating in the North Pacific is carried to the equator by STCs and amplified by Bjerknes feedback (calledvˉT′ mechanism).However, several studies have suggested that such thermal signals cannot reach the equatorial region (Schneider et al.,1999; Nonaka and Xie, 2000; Pierce et al., 2000;Hazeleger et al., 2001).Kleeman et al.(1999)proposed an alternative mechanism, suggesting that wind stress forcing in the subtropics can drive equatorial SST anomalies by influencing the strength of STCs (calledv′Tˉ mechanism).The results from the observational analysis also validate this mechanism: the internal STC transport convergence is highly correlated with the SST in the tropical central-eastern Pacific (e.g., McPhaden and Zhang, 2002, 2004; Capotondi et al., 2005; Zhang and McPhaden, 2006; Chen et al., 2015a), and also for the Atlantic Ocean (Rabe et al., 2008; Tuchen et al., 2020).

Numerous studies have focused on the forcing mechanism of STCs.McCreary and Lu (1994) used a two-layer model to show that the strength of STCs is related to subtropical zonal wind stress.Liu and Philander (1994) used a set of idealized rectangular basin ocean simulations to show that subtropical wind stress forcing can significantly change the tropical temperature but with limited effects on equatorial undercurrent (EUC) transport.Graffino et al.(2019) studied the impact of wind stress in different regions of the Pacific on STCs through global ocean models, and proved that subtropical wind stress anomalies are the primary forcing mechanism of Pacific STCs.

On the interannual scale, the interannual variability of interior STC can influence the variability of ENSO (Huang and Wang, 2001; Schott et al., 2008;Zilberman et al., 2013).The interior STC transport is weakened during El Ni?o and enhanced during La Ni?a (McPhaden and Zhang, 2002; Izumo, 2005).On interdecadal scales, STC transport convergence at 9° weakened by about 11 Sv during the 1970s and 1990s (McPhaden and Zhang, 2002; Capotondi et al., 2005), while STC transport increased by about 10 Sv from the 1990s to early 2000s (McPhaden and Zhang, 2004; Feng et al., 2010).Many scholars have obtained similar interdecadal variability through ocean models (Cheng et al., 2007; Farneti et al.,2014b), coupled ocean-atmosphere models (Lohmann and Latif 2005; Zhang and McPhaden, 2006), and ocean reanalysis products (Schott et al., 2008; Zeller et al., 2019).

Anthropogenic climate change caused by the increase of greenhouse gases in the atmosphere has become a major scientific and socio-economic problem (Xie et al., 2010).Through the coupled models, a series of changes in the tropical Pacific under global warming are found in the climate prediction simulation.Some scholars also pay attention to the possible response of STCs to the future greenhouse effect.Merryfield and Boer(2005) found that the STCs convergence transport was weakened.Lohmann and Latif (2005)suggested that the STCs strength measured by meridional overturning stream function decreased(strengthened) in the northern (southern) hemisphere.Park et al.(2009) and Yang et al.(2014) also showed the different behavior in two hemispheres.Graffino et al.(2021) used model data to calculate the heat transport change of STCs under 1pctCO2 runs: the heat transport in the northern (southern)hemisphere was weakened (strengthened).In addition, several studies proposed that under global warming the time-mean STC change can lead to the change in ENSO variability (e.g., Chen et al.,2015b, 2019), indicating the importance of investigating the future change in the STC.

Due to the inherent systematic errors of the climate model and the existence of a large number of processes that may affect the future STC changes(England et al., 2020), in many cases, the models cannot reach an agreement on the response of the tropical Pacific to global warming (e.g., Knutson and Manabe, 1998; Collins, 2005; Park et al., 2009),so predicting future trends of STC is challenging.

The studies show that there is partial compensation correlation between the interior STC and the west boundary transport on the interannualdecadal time scale through reanalysis products and model data (Capotondi et al., 2005; Lübbecke et al.,2008).With the addition of observational hydrographic data, Tuchen et al.(2020) further confirmed the existence of this compensation relation (in the Atlantic Ocean) using Argo data.In addition, the level of compensation for interannual variation between western boundary transport and interior STC transport strongly affects the level of compensation on a decadal timescale (Zeller et al.,2019).The varying levels of compensation between western boundary and interior transport are thought to be important for the tropical Pacific heat budget(Hazeleger et al., 2004).However, the compensation between the two in the future climate is controversial: Luo et al.(2009) found that in most coupled models, the contribution of western boundary transport (interior STC) increased significantly (decreased), the compensation between the two still exists.Wang and Cane (2011) also found a strong weakening of interior STC transport convergence, but found no evidence of compensation for the western boundary transport.

Recognizing the importance of Pacific STCs in shaping local and global climate, our goal is to further analyze the changes in STCs and their components under global warming.To achieve this,we compared the volume and heat transport of STCs and their relationship with equatorial Pacific SST,which are reproduced under historical conditions and SSP585 scenarios.The data and methods used in this study are shown in Section 2.Section 3.1 evaluates the changes of STCs transport under global warming.In Section 3.2 the relationship between interior STC and equatorial Pacific SST is discussed.Section 3.3 pays attention to the changes in interior STC heat transport under SSP585 scenarios.Section 4 gives the main results,discusses and compares them with previous work,and draws conclusions.

2 DATA AND METHOD

This section includes details of all coupled models and ocean reanalyses used in this study.Table 1 lists the information of the models: the institute and its references.All data are remapped on a 1°×1° grid.

2.1 Coupled model

The coupled models play an essential role in the study of climate science and the prediction of future climate change.Compared with CMIP5, CMIP6 has been greatly improved in terms of dynamicparameterization scheme and model resolution(Eyring et al., 2016).All CMIP6 simulations can be obtained through Earth System Grid Federation(ESGF).We used the ocean temperature and salinity data of historical simulation and SSP585 test to calculate the STCs in the Pacific Ocean.

Table 1 Main characteristics of selected coupled models

Historical simulation experiment is a historical climate simulation driven by various external forces based on observation and changing with time,including natural (such as solar energy change and volcanic eruption) and man-made forcing (such as greenhouse gases and aerosols).The experiment is used to evaluate the model’s ability to simulate climate change.For all CMIP6 historical runs, the historical period ranges from 1850 to 2014.

Compared with the four radiative forcing pathways in the Representative Concentration Pathways (RCPs) scenario test in CMIP5, CMIP6 has developed a new set of emission scenarios driven by different socio-economic models-Shared Socioeconomic Pathways (SSPs).It describes the possible future development of society without the influence of new climate policies.SSP585 is the updated RCP8.5 scenario (Kriegler et al., 2017).It represents a combination scenario of energyintensive socioeconomic development path driven by fossil fuel economy (O’Neill et al., 2016; Riahi et al., 2017).Under this scenario, the radiation forcing will stabilize at 8.5 W/m2in 2100 (O’Neill et al., 2016).

2.2 Ocean reanalysis and observation

We compare the results of historical simulations with three ocean reanalysis products: GODAS,SODA3.4.2, and ORAS5.

The Global Ocean Data Assimilation System(GODAS) is a real-time ocean analysis and reanalysis system from 1979 to the present.The GODAS is based on a quasi-global configuration of the Geophysical Fluid Dynamics Laboratory(GFDL) MOM.V3, forced by the momentum flux,heat flux, and fresh water flux of NCEP atmospheric Reanalysis 2, and using the 3DVAR scheme to assimilate the temperature and synthetic salinity profile.The model has a resolution of 1°×1° enhanced to 1/3° in the N-S direction within 10° of the equator (Behringer and Xue, 2004).The model has 40 levels with a 10-m resolution in the upper 200 m.

Simple Ocean Data Assimilation, V3.4.2(SODA3.4.2) is an ocean unit modular ocean model V5 based on the coupling model of the GFDL CM2.5 of the National Oceanic and Atmospheric Administration.The sea surface wind field and heat flux forcing of the model are from ERA-interior(European Center for Medium Range Weather Forecasts Reanalysis Interior) dataset.The time range of the product is from 1980 to 2019, and the horizontal resolution is 0.25°×0.25°, divided into 50 uneven layers vertically, with high resolution in the upper layer of the ocean.

European Centre for Medium-Range Weather Forecasts (ECMWF) ocean reanalysis system 5(ORAS5) contains five groups of members generated by initial conditions, observations, and force disturbances.The ocean model used by the system is Nucleus for European Modeling of the Ocean Version 3.4.1 (NEMO V3.4.1).The horizontal resolution selected is 1°×1°.There are 75 vertical levels, with level spacing increasing from 1 m at the surface to 200 m in the deep ocean (Zou et al., 2019).

In addition, we added observation data for comparison when calculating the time mean of interior transport.Argo (Array for real-time Geostrophic Oceanography) data is the gridded product made by Scripps Institution of Oceanography based on Argo observation profile (Roemmich and Gilson, 2009).The horizontal resolution is 1°×1°,and the vertical direction is 58 layers.

2.3 Method

Our analysis is based on STCs transport to the equator (1 Sv=106 m3/s), including the interior STC and the western boundary transport.First, 1 500 m is selected as the level of no motion, the meridional geostrophic currents are calculated by the dynamic height method using ocean temperature, salinity data, and then by vertically integrating meridional geostrophic currents from the mixed layer depth to a depth of 26σθ(potential density=26 kg/m3), and zonally integrating from the eastern edge of the western boundary current to the eastern boundary(140°E-80°W at 9°N and 160°E-80°W at 9°S) to estimate interior STC transport (Chen et al., 2015a;Feng et al., 2018).The western boundary transport is from the meridional geostrophic velocity integral of the sea surface to 26σθ, and latitudinal integration from the western edge of the western boundary current to the eastern edge (126°E-140°E at 9°N and 148°E-160°E at 9°S).

In addition, we also calculated the heat transport of interior STC, and the expression is as follows

wherecp=3 992.1 (J·℃)/kg is the specific heat capacity of seawater (assumed constant),ρis a reference density,θandvare sea water potential temperature and meridional geostrophic flow,respectively, and?is latitude.

3 RESULT

3.1 STCs transport and its future change

In this section, we calculate the interior STC transport at 9°S and 9°N.9°N is selected because there is a potential vorticity ridge at this latitude,which hinders the mass and heat transport between extratropical and tropical regions (McPhaden and Zhang, 2002), while 9°S is selected because of hemispheric symmetry.

We start by looking at the interior STC transport under historical simulation (Fig.1), in which the observational period of Argo was from 2004 to 2014,and the reanalysis products were from 1980 to 2014.From the perspective of time average, the results of reanalysis products and Argo data are consistent.At 9°N, the transport values of Argo, GODAS, SODA,and ORAS5 are -7.4, -6.3, -7.2, and -8.2 Sv,respectively.At 9°S, the time mean of interior transport are 11.6, 11.4, 13.5, and 13.4 Sv,respectively.In addition, we calculated interior transport from 2004 to 2019 using Argo (-6.2 Sv at 9°N, 10.8 Sv at 9°S).Except for some models (e.g.,FGOALS-f3-L, INM-CM4-8, and MRI-ESM2-0),the time-mean transport of most models under historical simulation is relatively in a good agreement with the ocean reanalysis products.Most of the interior STC transport of 9°N are in the range of -10--5 Sv.The multimodal-mean value is -9.3 Sv.The maximum (-20.9 Sv) and minimum (-2.5 Sv)are simulated by INM-CM4-8 and FGOALS-f3-L respectively.The multimodal-mean value at 9°S is 11.5 Sv.The transport value of most models is 5-15 Sv,the maximum value (24.1 Sv) still simulated by INM-CM4-8, and the transport value simulated by ACCESS-ESM1-5 (3.1 Sv) and MRI-ESM2-0(3.7 Sv) is smaller.

In addition to the mean-state of transport, we also investigate the variability of interior STC simulated by the models.The studies show that STC transport convergence weakened from the 1970s to 1990s(McPhaden and Zhang, 2002; Capotondi et al.,2005), and increased from the 1990s to early 2000s(McPhaden and Zhang, 2004; Feng et al., 2010).As for the trend of interior transport, the selection of different time periods may cause great differences.Since the time period of Argo only started in 2004, it is shorter than the reanalysis products and model data.In addition, the time series (not shown) of the reanalysis products and Argo are consistent, with the correlation above 0.8 and passing the 95%significance test.Therefore, we mainly study the linear trend of interior transport by reanalysis products and model data.First, we calculate the linear trend of interior STC transport (Fig.2a & c).In this case, the trend has not been removed.The ocean reanalysis products show the same trend:enhancement, while there is no uniform trend in models.Some models can simulate well the linear trend of 9°N and 9°S transport (BCC-CSM2-MR,CMCC-CM2-SR5, FIO-ESM-2-0, INM-CM4-8,MRI-ESM2-0), while a few models simulate the trend contrary to the reanalysis data (FGOALS-G3,NORESM2-LM, MIROC6, MPI-ESM1-2-HR).This suggests that although the models can simulate well the climatology of interior STC transport, they are still insufficient in the simulation of time variability.

Fig.1 Time-mean interior STC transport at 9°N (a) and 9°S (b) under historical simulation

Fig.2 Linear trend (Sv/a) of interior STC under historical simulation (a, c) and SSP585 runs (b, d)

We further focus on the linear trend of interior STC transport under global warming (Fig.2b & d).As shown in the figure, most models agree on a general weakening of the interior STC under SSP585 scenarios at 9°S and 9°N, although the weakening strength is different.The linear trend of the multimodal-mean of interior STC at 9°N and 9°S is about 0.04 Sv/a and -0.02 Sv/a, respectively.Unlike most of the models, interior STC simulated by INM-CM4-8 (-0.02 Sv/a at 9°N, 0.13 Sv/a at 9°S),and INM-CM5-0 (0.08 Sv/a at 9°S, 0.01 Sv/a at 9°N) shows an enhanced trend, and the opposite trend is simulated by CMCC-CM2-SR5 (0.03 Sv/a)at 9°S.

Furthermore, we can see that the trend uncertainty in future scenario is reduced compared with historical simulation.We think that trend uncertainty may be related to the size of the sample.The more the sample, the smaller the uncertainty.In the historical simulation and SSP585 scenario, the time period is from 1980 to 2014 and from 2015 to 2100, respectively, so there is smaller uncertainty in the projected trends.

Then we explore the reasons for the weakening of interior STC transport under global warming.Firstly, the changes of mixing layer depth, the 26σθisopycnal and meridional geostrophic velocity are analyzed (Figs.3-4).It can be seen that under the SSP585 scenario, the depth of the mixed layer changes less, while the 26σθisopycnal becomes deeper.In addition, most of models show a weakening of meridional flow.Combining the calculation method of interior STC, we believe that the weakening of interior STC transport is mainly caused by the weakening of meridional flow.The reasons for the weakening of meridional flow are considered from two aspects: the oceanic stratification and the forcing of wind field.

Fig.3 Meridional geostrophic velocity change (cm/s) between historical simulation and SSP585 scenario in the upper Pacific Ocean (0-500 m) along 9°N

Fig.4 Same as Fig.3, but for 9°S

STCs are wind-driven circulation features.According to the wind-driven circulation theory, the meridional geostrophic transport can be calculated by using the wind field data.Since the integral depth is not a sensitive factor of the meridional geostrophic transport distribution (Yuan et al.,2014), the interior STC transport can be expressed as (Feng et al., 2018):

whereβis the meridional derivative of the Coriolis parameter, andτis the surface wind stress.

Therefore, we believe that the enhancement of the wind stress curl will lead to the enhancement of interior STC.The wind stress curl and zonal wind stress in tropical Pacific (Fig.5) are calculated.At 9°N, the wind stress curl shows negative anomaly,indicating that under SSP585 scenario, the weakening of wind stress curl leads to the weakening of interior STC; it can also be obtained that the wind stress curl in the central and western Pacific decreases at 9°S.In addition, the easterly near-equatorial zonal wind stress is weakened (Xie et al., 2010), which weakens poleward Ekman transport, resulting in the weakening of the 9°interior STC transport convergence.Further, due to global warming, the heat absorbed by the ocean increases (Levitus et al., 2000; Cheng et al., 2021),the Pacific SST increases, the density of the upper layer of the ocean decreases, the stratification is strengthened (Li et al., 2020), and the subtropical seawater subsidence weakens (Peng et al., 2022),resulting in the weakening of the interior branches of STCs.Therefore, the combined effect of the strengthened upper oceanic stratification and the weakening of wind field leads to the weakening of the interior STC.

Considering the compensation relationship between the western boundary transport and the interior STC transport and the uncertainty of the relationship between the two global warming issue studies by Luo et al.(2009) and Wang and Cane(2011), we calculated the linear trend of western boundary transport under SSP585 scenarios (Fig.6).Different from interior STC transport, western boundary transport does not show consistent trends in the northern and southern hemispheres.At 9°S,most of the models show enhanced western boundary transport, the multimodal-mean of the enhancement trend is 0.02 Sv/a, which is consistent with the weakening trend of interior STC; while at 9°N, there is no uniform inter-model variation, with half of the models showing a weakening trend of the western boundary transport.In addition, there is an asymmetry in the variability of western boundary transport in the two cells: the trend of western boundary transport at 9°S is significantly stronger than that at 9°N (Fig.6).The linear correlation coefficients between western boundary transport and interior STC (Table 2) are calculated.We find that most of the models show that there is still a compensation between western boundary transport and interior STC under SSP585 scenarios, which is consistent with the results of Luo et al.(2009), and the compensation was more significant for 9°S compared with 9°N (Figs.2b, 2d, & 6).Lee and Fukumori (2003) have noticed this phenomenon and explained it by using the different magnitude of the off-equator wind stress curl in the northern and southern hemispheres.

Fig.5 The left panel: the difference of multimodal-mean SSP585 minus historical simulation: wind stress curl (a) (N/m3);zonal wind stress (c) (N/m2); the right panel: wind stress curl (b); zonal wind stress (d) under historical simulation

Fig.6 Linear trend (Sv/a) of the western boundary transport at 9°N (a) and 9°S (b) under SSP585 scenario

Table 2 Linear correlation coefficient of interior STC and western boundary transport simulated by coupled model under SSP585 scenario

3.2 Relationship between interior STC transport and equatorial Pacific SST

The relationship between interior STC transport and SST in the equatorial Pacific has been well confirmed in observations and models (e.g.,McPhaden and Zhang, 2002, 2004; Farneti et al.,2014a; Farneti, 2017).On the interannual and interdecadal time scales, the interior STC transport and equatorial SST are anti-correlated, in agreement with studies for the Atlantic Ocean (Tuchen et al.,2020).In order to explore this connection,we calculated the variance contribution of the interannual-decadal variability of the interior STC(not shown), and find that the interannual variability(6.8 Sv for reanalysis products; 3.1 Sv for models)of the interior STC was much larger than the decadal variability (3.9 Sv for reanalysis products;2.0 Sv for models).We show the time series of interior STC transport convergence under historical simulation (evaluated as the sum of transport southward at 9°N and transport northward at 9°S)and the equatorial Pacific SST anomaly (9°S-9°N,180°-90°W), and compare them with the time series obtained from ocean reanalysis products (Fig.7).Here, linear trend is removed and the Butterworth filter (2-8-year band-pass filter) is applied to obtain interannual variability of interior STC transport convergence and SST.

Observing the time series, it can be found that the variability of the interior STC transport convergence obtained from the ocean reanalysis is relatively consistent, while the coupled models show a large inter-model variation.It can be seen from Fig.7 that different models simulate different period of interior STC.The interannual period of interior STC transport convergence simulated by FGOALS-f3-L, and TaiESM1 is about 3-4 years.The simulation period of MPI-ESM1-2-HR, MPIESM1-2-LR and MIROC6 is longer, about 6 years.In addition, the period simulated by INM-CM4-8 and INM-CM5-0 is shorter, about 2 years.In addition, CMIP6 also shows a large inter-model variation in the amplitude of interior transport.The amplitude simulated by FGOALS-f3-L, NorESM2-LM and MIROC6 was larger, while the amplitude simulated by CanESM5, INM-CM4-8, and INMCM5-0 was smaller.The SSP585 simulation also shows the same phenomenon (Fig.8).In order to better evaluate this aspect, we calculated the standard deviation of interior STC transport convergence and SST and the correlation between the two (Fig.9).Values obtained from ocean reanalysis products are shown in historical runs plots for comparison with models.As shown in the figure, the standard deviation (6-8 Sv) of STC from ocean reanalysis products under historical simulation is larger than that of most models (2-5 Sv).As for equatorial SST anomaly, the multimodal-mean standard deviation of CMIP6 models in historical simulation (0.56 ℃) is consistent with that of ocean reanalysis products (0.54 ℃).In addition, the models showing a large variation in STC transport also exhibit large variations in equatorial Pacific SST on interannual time scales, which exists in both historical and SSP585 scenarios.

From Fig.9, we can also see how the different models reproduce the STC-SST relationship.For most models, there is little change in the correlation coefficient from historical runs to SSP585 runs.On average, the correlation coefficient of STC-SST is-0.87 for ocean reanalyses, -0.76 for historical simulation, and -0.79 for SSP585 scenario.The average STC-SST relationship of the models is similar to that of the ocean reanalysis products,which is different from that of the interdecadal scale: the multimodal-mean STC-SST relationship of the models is weaker than that of the ocean reanalysis products (Graffino et al., 2021).We considered the reasons for this phenomenon.Before calculating the correlation between STC and SST,linear trend is removed and interannual bandpass filtering are applied, which may be the main reason for the enhancement of the correlation between the two.

Fig.7 Interior STC convergence transport anomaly (Sv, red line) and equatorial Pacific SST anomaly (℃, blue line) are simulated by ocean reanalysis products and CMIP6 historical simulation

Fig.8 The same as shown in Fig.7, but for CMIP6 SSP585 simulation

3.3 Interior STC heat transport

In this section, we calculated the interior STC heat transport using Eq.1 under SSP585 runs and compared it with historical simulation.In this case,the trend has not been removed.Figure 10 shows the interior STC heat transport simulated by CMIP6 models.It can be found that there is a minimum value of equatorial heat transport at 9°N and 10°S.Compared with historical simulations,heat transport is weakened at 10°S-5°S and 5°N-15°N, and enhanced at 11°S-20°S and 16°N-20°N under SSP585 runs.We again focused on the timemean interior STC heat transport at 9°N and 9°S(Fig.11).As shown in Fig.11, about 2/3 of the models show a weakening of the interior STC heat transport, which also corresponds to the result obtained in Section 3.1: the weakening of interior STC transport at 9°N and 9°S.The strength of interior STC heat transport weakening is quantified.The multimodal-mean STC heat transport is reduced by 3.2% and 6.9% at 9°S and 9°N, respectively.At 9°N, the maximum (-38.2%) and minimum (-0.9%)of the weakening trend are simulated by NorESM2-LM and FGOALS-g3, respectively.At 9°S, the maximum (-52.6%) and minimum (-0.9%) values are simulated by ACCESS-CM2 and FGOALS-g3,respectively.In addition, some models show enhanced interior STC heat transport under SSP585 operation (CanESM5, INM-CM4-8, INM-CM5-0,and MIROC6).

4 DISCUSSION AND CONCLUSION

In this study, we evaluated CMIP6 model simulations of interior STC and analyzed the change of interior STC volume transport and heat transport under global warming.We used the state-of-the-art coupled models from CMIP6 listed in Table 1.As a reference, we used the Argo and three ocean reanalysis products.

Fig.10 The multimodal-mean interior STC meridional heat transport (PW=1015 W) vs.latitude

Fig.11 The multimodal-mean interior STC meridional heat transport (PW) at 9°N (a) and 9°S (b) under historical (blue frame) and SSP585 scenarios (red frame)

First, by calculating the time-mean transport of interior STC under historical simulation, we found that the time-mean transport values calculated by the Argo and ocean reanalysis products are relatively consistent.However, some models do not reproduce interior transport well (e.g., FGOALS-f3-L, INM-CM4-8, and MRI-ESM2-0).Fu et al.(2022)believes that this is related to the zonal density structure, and by comparing with the observations, it is found that the ensemble warm bias in the model will weaken the local density gradient, thus leading to the weaker Atlantic STCs interior transport.Whether this is also the case in the Pacific Ocean,and why some specific models simulate stronger interior transport, will be further studied in the follow-up work.Except for some models, the timemean transport of most models under historical simulation is relatively in a good agreement with the ocean reanalysis products.The multimodal-mean values of interior STC transport are -9.3 Sv and 11.5 Sv at 9°N and 9°S, respectively.In addition, the ocean reanalysis products show an enhanced trend of interior STC.At 9°N, the linear trends of GODAS,SODA, and ORAS5 are -0.08, -0.11, and -0.14 Sv/a,respectively; at 9°S, the trends are 0.15, 0.10, and 0.18 Sv/a respectively.The multimodal-mean trend is 0.02 and 0.01 Sv/a at 9°N and 9°S, respectively,with a large difference compared to the reanalysis products.From Fig.2, we see that only a few models(5/20) can well simulate the linear trend of the transport in two cells.This shows that although the models can simulate the time-mean interior STC transport well, there are still insufficiencies in the simulation of time variability.To better simulate the interior STC, it is very important to clarify the relationship between the forcing mechanism (such as zonal wind stress, wind stress curl) and the final response (interior STC transport).Solomon and Zhang (2006) and Zhang and McPhaden (2006)suggested that the variability of subtropical wind stress in the models is too weak to drive strong STCs response (especially Ekman pumping).However, Graffino et al.(2021) did not find such a feature in the simulation, although the ocean reanalysis tends to give larger meridional Ekman transport than the coupled models.How the subtropical wind stress reproduced by the coupled models affects the variability of STC requires further research.

Then we continued to pay attention to the changes of STCs under global warming and their causes.Most of the models show that the interior STC in the northern and southern hemispheres will weaken in the future climate.This is consistent with previous results (Luo et al., 2009; Wang and Cane,2011; Graffino et al., 2021).The linear trend of the multimodal-mean of interior STC at 9°N and 9°S is about 0.04 Sv/a and -0.02 Sv/a, respectively.The reasons for the weakening of meridional flow are considered from two aspects: the oceanic stratification and the forcing of wind field.According to Sverdrup theory, the meridional transport in the inner ocean region is related to the wind stress curl.In the northern (southern) hemisphere, the positive(negative) wind stress curl leads to upwelling, which accelerates the circulation and enhances the interior STC transport.From Fig.5a & b, we see that the wind stress curl in the northern hemisphere will be weakened in the future scenario, as will the western and central Pacific in the southern hemisphere.The near-equatorial easterly zonal wind stress is also weakened, which weakens poleward Ekman transport and equatorial pycnocline transport, resulting in the weakening of the 9° interior STC transport convergence.In addition, due to global warming,the density of the upper layer of the ocean decreases, the stratification is strengthened, and the subtropical seawater subsidence weakens, resulting in the weakening of the interior branches of STCs.Therefore, the combined effect of the strengthened upper oceanic stratification and the weakening of wind field leads to the weakening of the interior STC.

Different from interior STC transport, western boundary transport shows different trends in the northern and southern hemispheres.At 9°S,most models show western boundary transport enhancement; at 9°N, there is no uniform intermodel variation in western boundary transport, and only half of the models show a weakening trend.In addition, it is well known that there is a compensation between the interior STC and the western boundary transport, which has been emphasized by Zhang and McPhaden (2006).In the future climate, most models (15/20) still show the compensation of interior STC and western boundary transport.What is more interesting is the difference in compensation between the northern and southern hemispheres: compared with 9°N, the western boundary transport at 9°S has a more significant compensation for interior STC (Figs.2 & 6; Table 2).

Lee and Fukumori (2003) proposed through numerical experiments that the relative contribution of zonal near-equatorial wind stress and offequatorial wind stress curl would dictate the relative variability of interior and boundary transport.The larger the off-equatorial wind stress curl, the stronger the compensation between the interior and western boundary transport.We calculated the standard deviation of wind stress curl and found that the standard deviation near 9°N (5.9×10-9) was about 19% smaller than that near 9°S (7.27×10-9) (mainly caused by interannual variation).This is consistent with the larger boundary and interior transport compensation near 9°S.In addition, Lee and Fukumori (2003) also considered the role of ITF.It is found that ITF was not sensitive to the compensation of interior and boundary transport.However, the numerical experiment did not consider the influence of ITF on the wind field, so it may be necessary to conduct sensitivity experiments using the coupled ocean-atmospheric model to further study the influence of ITF on the Pacific STCs.

Turning to the interior STC transport convergence variation, we analyzed the standard deviation of interior STC transport convergence and equatorial Pacific SST, as well as the STC-SST negative correlation.The time variability of ocean reanalysis products is relatively consistent, while the coupled models show a large inter-model variation, which can be clearly seen from the time series of STC and SST anomalies (Figs.7-8).In addition, the models showing a large variation in interior STC transport also show a large variation in equatorial Pacific SST on an interannual scale.There is agreement among most of the models on the STC-SST relationship.It is worth noting that INM-CM4-8 (small negative correlation).Indeed, thev′Tˉ mechanism (Kleeman et al., 1999) explaining the STC-SST connection suggests that by underestimating the STC transport variability, the coupled models may lose part of its SST driving force.This may explain the low correlation of the simulated STC-SST for some models (e.g., INM-CM4-8).In the follow-up work,we can further analyze the reasons for the poor performance of the models, and then provide a reference for the improvement of the models.

We then calculated the change of the interior STC heat transport under future climate and compared it with historical simulation.The future scenario simulation shows that the heat transport of interior STC is weakened under global warming,with a general agreement across models.The multimodal-mean STC heat transport is reduced by 3.2% and 6.9% at 9°S and 9°N, respectively.

Overall, we evaluated the simulation of interior STC by CMIP6 models, as well as quantitatively analyzed the changes of STCs under future climate and made some progress in explaining the reasons for this phenomenon.Some previous studies have discussed the impact of interior STC on ENSO based on reanalysis data (Huang and Wang, 2001;Schott et al., 2008; Zilberman et al., 2013).We believe that the change of interior STC will have some impact on the cycle or amplitude of ENSO in the future climate.Using new model data to analyze the change of the relationship between the two and how the future change of STC will affect the tropical climate will be the focus of future work.

5 DATA AVAILABILITY STATEMENT

We thank the Earth System Grid Federation(ESGF) for archiving the data and providing access,and the Climate Modeling Group for producing and providing its model output.CMIP6 data are available at the ESGF website (https://esgf-node.llnl.gov/projects/cmip6).The GODAS dataset is available at the Physical Science Laboratory (https://psl.noaa.gov/data/gridded/data.godas.html).The SODA3.4.2 reanalysis datasets are available through www.soda.umd.edu.ORAS5 datasets are available free of charge at the University of Hamburg (https://www.cen.unihamburg.de/en/icdc/data/ocean/easy-init-ocean/ecmwforas5.html).Argo data are available throughhttps://sio-argo.ucsd.edu/RG_Climatology.html.

6 ACKNOWLEDGMENT

The authors would like to thank Dr.Qingye WANG of the Institute of Oceanology, Chinese Academy of Sciences for their useful discussion during this study.We also thank the editor and anonymous reviewers, whose constructive suggestions and comments lead to a great improvement of the manuscript.