An equilibrium last interglacial simulation (lig127k) is performed with the Australian Community Climate and Earth System Simulator Earth System Model, ACCESS-ESM1.5 , which includes interactive land and ocean carbon cycles and is Australia's submission to the Paleoclimate Modeling Intercomparison Project 4 (PMIP4) – Coupled Model Intercomparison Project (CMIP6) . ACCESS-ESM1.5 differs from the previous version, ACCESS-ESM1 mostly in the land and ocean components. ACCESS-ESM1.5 is built upon the ACCESS1.4 physical model and includes the UK Met Office Unified Model (UM) version 7.3 as an atmospheric component, which is directly coupled to the updated land Community Atmosphere Biosphere Land Exchange model (CABLE) version 2.4 , both with a horizontal resolution of 1.875∘ × 1.25∘. UM has 38 vertical levels. The ocean model is the NOAA/GFDL Modular Ocean Model (MOM) version 5 , which is coupled to the Los Alamos National Laboratory sea ice model (LANL CICE) version 4.1 . MOM5's resolution is 1∘ × 1∘ with 50 vertical levels. The coupler is the Ocean Atmosphere Sea Ice Soil – Model Coupling Toolkit (OASIS-MCT) . The current version of the land model CABLE also includes biogeochemistry (BGC) implemented using the CASA-CNP module . The ocean carbon cycle is simulated using the Whole Ocean Model of Biogeochemistry And Trophic-dynamics (WOMBAT) model .

WOMBAT is a nutrient–phytoplankton–zooplankton–detritus (NPZD) model, with one class of phytoplankton and one class of zooplankton. It includes dissolved inorganic carbon (DIC), alkalinity (ALK), phosphate (PO4), oxygen (O2), and iron. The biogeochemical tracers are coupled using the stoichiometric C (carbon):N (nitrogen):P (phosphorous):O2 ratio of 106:16:1:172. The air–sea gas exchange is based on the square of the wind speed , and the seawater partial pressure of CO2 (pCO2) is calculated following the third phase of the Ocean Carbon-Cycle Model Intercomparison Project (OCMIP) protocol using temperature, salinity, DIC, ALK, and PO4. WOMBAT simulates production in and export from the photic zone and remineralization and dissolution at depth for both organic and inorganic (CaCO3) particulate matter, with parameters adjusted for inorganic export to be ∼8 % of organic export. The remineralization of organic matter is calculated based on depth- and temperature-dependent parameters, while the dissolution of CaCO3 uses a constant dissolution rate. ACCESS-ESM1.5 does not include a sediment component, and all particulates reaching the bottom ocean layer are instantly remineralized following the relevant remineralization rates. Other specific details of the model can be found in , , and .

A pre-industrial 1850 simulation (piControl) is run in accordance with the CMIP6 protocol , with a constant atmospheric CO2 forcing of 284.3 ppm but using CMIP5 solar irradiance (1365.65 Wm-2, as explained in ). This simulation is run for 1000 years. Initialized from this PI control run, a LIG simulation (lig127k) is performed using orbital parameters following the PMIP4 protocol but with the solar constant adjusted to the CMIP5–PMIP3 value to be comparable to the piControl simulation (1365.65 Wm-2). CO2 concentration for the LIG is set at 275 ppm. Vegetation is kept constant at 1850 PI levels. The lig127k experiment is run for 650 years. The run is at equilibrium over the SO, although there is a small drift in globally averaged DIC of ∼1 µmol kg-1/100 yr-1 below 3 km. The analysis presented here is based on the average of the last 100 years of the lig127k simulation compared to the last 100 years of the piControl simulation. Further details on these experiments can be found in , , and .

WOMBAT includes two DIC tracers, one being forced by the prescribed atmospheric CO2 concentration (PI: 284.3 ppm; LIG: 275 ppm), and the other forced by a constant atmospheric CO2 concentration of 280 ppm. Unless otherwise stated, all of the analyses presented here are based on the tracers forced with CO2 concentrations of 280 ppm, while the climate response is forced with the radiative forcing of 284.3 ppm for PI and 275 ppm for the LIG. This allows quantification of the effects of the LIG climate on the carbon cycle independently of the difference in background CO2 concentrations.

Changes in surface pCO2, which ultimately control the direction and magnitude of air–sea fluxes, can further be decomposed into contributions from SST, SSS, DIC, and ALK . The SST contribution is calculated as 1ΔpCO2SST=e(γSST×ΔSST)×pCO2PI-pCO2PI, where ΔpCO2SST represents the contribution of SST change to surface pCO2 change and γSST=0.0423 ∘C-1 is the sensitivity of pCO2 changes to changes in SST . ΔSST is the difference in SST between LIG and PI, and pCO2PI is the surface seawater pCO2 from the piControl simulation.

The SSS, DIC, and ALK contributions are calculated as 2ΔpCO2X=γX×ΔXX‾×pCO2PI, where ΔpCO2X represents the contribution of change in variable X (SSS, DIC, and ALK) to surface pCO2 change, ΔX is the change in X between LIG and PI, X‾ is the mean value of X at PI, and γX is the sensitivity of pCO2 changes to changes in variable X (γDIC is known as the Revelle factor), with γSSS=1, γDIC=15,iflat≥60∘13,lat<60∘, and γALK=-12,iflat≥60∘-11,lat<60∘. These values are based on previous estimates of the meridional profiles of DIC and ALK buffer factors , although uncertainties still remain regarding these estimates. This decomposition helps shed light on the independent effects of different physical variables to the net surface pCO2 change.

To better quantify changes in the carbon cycle between the LIG and PI experiments, we split the total DIC into its remineralized (Corg), dissolved carbonate (CCaCO3), and preformed (Cpre) components using the equations listed below.

Corg can be estimated from regenerated phosphate (PO4Reg) using the C:P stoichiometric ratio (rC:P=106) : 3ΔCorg=rC:P×ΔPO4Reg.

PO4Reg can be approximated using apparent oxygen utilization (AOU) and the P:O2 stoichiometric ratio (rP:O2=1/172) : 4ΔPO4Reg=rP:O2×ΔAOU.

AOU estimates the oxygen consumed during respiration and can be calculated as the difference of dissolved oxygen concentration (O2) from the saturated concentration of oxygen (O2sat) : 5AOU=O2sat(T,S)-O2.

This can then be used to infer the efficiency of the biological pump (BPEff) as per : 6BPEff‾=PO4Reg‾/PO4‾, where X‾ is the mean value of X.

The contribution to DIC from the carbonate pump is estimated by 7ΔCCaCO3=0.5×ΔALK+rN:P×ΔPO4Reg, where the term rN:P×ΔPO4Reg accounts for the reduction in ALK from production of nitrate (NO3-) , which is estimated using PO4Reg and the N:P stoichiometric ratio (rN:P=16). Finally, the preformed carbon concentration (Cpre) is obtained as follows: 8ΔCpre=ΔDIC-ΔCorg-ΔCCaCO3.

As a result of the insolation anomalies and associated feedbacks, the global mean annual SST anomaly at the LIG compared to PI as simulated by the ACCESS-ESM1.5 is equal to 0.17 ∘C, with a pronounced warming at high latitudes and maximum positive SST anomalies over the North Atlantic (up to 4 ∘C, ). Our simulated temperatures are in line with the range of PMIP4 lig127k simulations . A model–data comparison of the LIG climate state is presented in .

A mean 0.53 ∘C warming is simulated over the SO south of 40∘ S (Fig. a). Warmer conditions are simulated everywhere south of 50∘ S apart from a ∼ 1 ∘C cooling centred at 58∘ S in the South Atlantic and up to a 1.5 ∘C cooling in the subantarctic eastern Pacific. The strongest warming is simulated over the Southeast Atlantic and Indian Ocean sectors (up to 3 ∘C). Regional SSTs up to 4 ∘C higher are simulated around 60∘ S for both austral spring (Fig. b) and summer (not shown). The higher SSTs over the SO compared to PI are accompanied by a marked reduction in sea ice extent over both austral summer and winter (Fig. a), peaking at 41 % reduction in austral winter (Fig. a).

A 1.5∘ equatorward shift of the SH westerlies is simulated at the LIG (Fig. e), with a 10 % weakening of the winds south of 50∘ S. This leads to ∼ 10 % weaker upwelling south of 55∘ S and up to ∼ 20 % stronger upwelling north of 55∘ S (Fig. f). Seasonally, the largest changes in upwelling are found for the winter and spring seasons (Fig. c and d). The equatorward shift of the westerlies reduces the northward Ekman transport south of 55∘ S, leading to warming, while the higher Ekman transport north of 55∘ S induces a cooling around 50∘ S (Fig. a and b). The Antarctic Circumpolar Current is weaker, and the transport through Drake Passage is reduced from PI by ∼ 55 Sv (not shown).

These physical changes in the SO impact the carbon cycle and the air–sea CO2 exchange (Fig. ). It is worth reiterating here that we analyse changes in the carbon cycle using tracers for a constant atmospheric CO2 concentration of 280 ppm for both simulations (Sect. 2.1), which enables us to solely analyse the impact of climatic and oceanic circulation changes on the carbon cycle. The CO2 flux over the SO at PI shows a carbon uptake near the Antarctic coast, an outgassing band between 65–45∘ S, and another uptake zone further north (Fig. a and blue line in d). This CO2 outgassing is due to the upwelling of DIC-rich deep waters (Fig. f). At the LIG, the upwelling region widens over the sectors of the Atlantic and Indian oceans, and narrows over the eastern Pacific Ocean sector (Fig. ). There is a strong increase in outgassing in the sectors of the Atlantic and western Indian oceans, as well as at ∼ 67∘ S in the eastern Pacific Ocean sectors (Fig. c and red line in d). An increase in CO2 uptake is simulated in the Amundsen, Bellinghausen, Weddell, Lazarev, Riiser-Larsen, and Ross seas and the subantarctic eastern Pacific (Fig. c). The zonal mean CO2 flux in Fig. d shows a small increase in uptake south of 62∘ S, possibly due to reduced mixing and reduced winter sea ice cover. Overall, there is a net increase in the mean SO outgassing by ∼ 150 % at the LIG compared to PI (Fig. e). This increase in outgassing mostly occurs during the austral winter and spring seasons (Fig. e and f).

This increased outgassing over the SO is compensated by increased uptake over other ocean basins, especially the North Pacific subpolar gyre, the South Atlantic Ocean, and the Northwest Indian Ocean, while higher outgassing is simulated in the North Atlantic (Fig. ). In this paper, we focus primarily on the SO. To better understand the SO changes, we decompose the oceanic pCO2 changes into their different components.

The air–sea gas exchange is primarily controlled by the seawater partial pressure of CO2 (pCO2). Using the equations presented in Sect. 2.2, Fig.  shows a decomposition of the changes in pCO2 into their SST, SSS, DIC, and ALK contributions. In line with the net increase in SO outgassing at the LIG (Fig. ), the net pCO2 over the SO is 1.2 µatm higher at the LIG compared to PI (black circle) (Fig. a). In agreement with the changes in air–sea CO2 flux, surface pCO2 is higher in a zonal band centred at ∼ 55∘ S, while it is lower in coastal regions (contours in Fig. b and black line in Fig. g). The contributions from individual components add up to reflect the simulated differences reasonably well in terms of both magnitude (1.4 µatm, grey square in Fig. a) and spatial distribution (Fig. b), affirming the validity of the decomposition method. The overall slightly positive pCO2 anomaly at the LIG compared to PI results from the competing effects of lower CO2 solubility (red triangle, +5.65 µatm) and changes in surface DIC and ALK (blue diamond, -4.25 µatm).

The largest contributor to solubility is SST (brown triangle), while the largest contributor of the combined DIC and ALK effect is DIC (cyan diamond in Fig. a). Higher SSTs in the SO at the LIG lead to a 5.8 µatm pCO2 increase (brown triangle in Fig. a). This SST-induced increase is present over most of the SO and is highest over the South Atlantic, Indian, and West Pacific regions (Fig. c). Changes in SSS do not contribute significantly to the pCO2 anomalies (magenta triangle in Figs. a and d). Changes in DIC cause the largest single contribution to the overall pCO2 change, with the net decrease in surface DIC leading to a -10.9 µatm pCO2 change (cyan diamond in Fig. a). As an exception to this, the higher DIC concentrations in the Atlantic Ocean around 55∘ S and in the eastern Pacific sector between 60–70∘ S (Figs. g, h, and d) lead to higher pCO2 (Fig. e). This higher DIC is due to increased upwelling in these regions (Figs. b and c, d), resulting from the equatorward shift of the SH westerlies. The contributions based on changes in ALK and DIC have very similar patterns, albeit of opposite signs, with a small difference over the West Pacific around 50∘ S. The combined effect of ALK and DIC changes leads to a total net decrease in pCO2 by 4.25 µatm due to the higher impact of changes in DIC (-10.9 µatm, cyan diamond) compared to ALK (+6.65 µatm, green diamond). Overall, south of 45∘ S, changes in SST lead to an overall increase in pCO2, with the highest contributions over the Indian and West Pacific sectors of the SO, while changes in DIC lead to an overall decrease of pCO2, with the strongest reductions in the South Atlantic, Indian, and West Pacific sectors of the SO but an increase in the Amundsen Sea sector (Fig. ).

The zonal mean patterns of these contributions are presented in Fig. g. Coastal regions around Antarctica south of ∼ 75∘ S show that the lower ΔpCO2, both simulated (black line) and calculated (grey line), primarily arise from the combined DIC and ALK component (blue line) that is mainly attributed to the lower DIC (dashed cyan line) compensated by changes in ALK (dashed green line). Between 70 and 60∘ S, ΔpCO2 anomalies are close to zero as the reduced CO2 solubility due to higher SSTs (red line) is compensated by the combined DIC and ALK changes (blue line). Between 60 and 45∘ S, the simulated surface ocean ΔpCO2 is positive (∼ 2 µatm, black line) as the effect of higher SSTs dominates over the combined ALK and DIC components. North of ∼ 45∘ S, the positive ΔpCO2 signal arises from the DIC + ALK component, while solubility mitigates the anomaly (Fig. g).

Figure g shows that while solubility is mostly controlled by changes in SST, contributions from SSS changes are important south of 72∘ S. Similarly, DIC has the dominant control over the DIC + ALK component. The contributions from both the solubility and (DIC + ALK) components reach their maxima (solubility being positive and DIC + ALK being negative) around 62∘ S, which corresponds to the maximum divergence in wind-driven surface currents (Fig. f). To summarize, the higher overall pCO2 (and hence lower air–sea CO2 flux) at the LIG compared to PI can be attributed to higher SSTs south of 45∘ S, and the combined DIC + ALK component between 45 and 35∘ S due to the equatorward shift of the upwelling regions (Fig. g). The SST patterns have already been discussed in Sect. 3.1. DIC patterns (Fig. d) can result from changes in circulation and the biological pump. These are investigated in the next section.

Figure c and d show that North Atlantic Deep Water (NADW) formation is ∼5 Sv higher in our LIG simulation compared to the PI simulation, leading to colder waters at depths of 1–2 km between 0–60∘ N, accompanied by higher SSTs and SSSs in the Labrador Sea (not shown). We also simulate a ∼ 4 Sv reduction in AABW formation (Fig. c and d). This reduction in AABW formation results in a warming of SO deep waters by up to 2 ∘C compared to PI. Due to a northward shift and weakening of winds (Fig. e, f), the Antarctic Intermediate Water (AAIW) formation regions shift northward and the AAIW formation rate slows down . The higher SST, equatorward-shifted upwelling, and reduced sea ice extent lead to an increase in net primary production (NPP) and export production over the SO of ∼ 17 % and ∼ 11 % respectively (Fig. a and b).

These circulation changes impact the DIC distribution in the ocean (Fig. b). For instance, the reduced formation rate of AAIW (Fig. d) increases residence times and leads to lower dissolved oxygen (Fig. a), higher PO4 (Fig. c), higher remineralized carbon (Fig. d), and higher DIC concentrations (Fig. b) at intermediate depths of the SO north of 55∘ S. Similarly, the increased accumulation of nutrients and DIC, as well as oxygen depletion in deep and abyssal waters (Fig. a, b and c), can be attributed to a weaker AABW formation rate (Fig. d), which results in a higher efficiency of the biological pump, as detailed below.

The ACCESS-ESM1.5 simulates a ≥50 µmol kg-1 decrease in oxygen everywhere below 3 km at the LIG compared to PI. A decrease in oxygen can also be seen in the SO across all depths, including a northward-reaching tongue attributable to AAIW at ∼ 1 km depth extending to the Equator (Fig. a). DIC anomalies follow the patterns of dissolved oxygen, albeit with the opposite sign. The simulated DIC concentrations are ∼ 50 µmol kg-1 higher across all depths in the SO and globally below 3.5 km depth (Fig. b).

DIC anomalies are decomposed into contributions from remineralized organic carbon, dissolved calcium carbonate, and preformed carbon components (Fig. d, f and h, Sect. 2.3). A total of 60 % of the DIC increase in the SO and abyssal ocean can be attributed to an increase in remineralized carbon (Fig. d) resulting from a 10 % more efficient biological pump (Eq. 6, Sect. 2.3). This increase in remineralized carbon can be attributed to increased residence times, which are in turn due to weaker bottom- and intermediate-water formation rates (Fig. c and d). Weaker AABW and AAIW indeed lead to positive apparent oxygen utilization (AOU) and regenerated PO4 anomalies (Fig. e). In addition, a 2 ∘C warming of bottom waters leads to a ∼ 6 % increase in remineralization rates, thus further contributing to the higher remineralized carbon. Around 25 % of the abyssal increase in DIC is attributed to the carbonate pump (Fig. f). This reflects changes in NPP, export production (Fig. ), and water mass residence times, given that carbonate production is a constant percentage of total NPP in this model setup and carbonate dissolution is constant and independent of water chemistry or temperature. Preformed DIC represents only ∼ 5 % of the DIC changes (Fig. h). This increased sequestration of DIC in the deep ocean reduces the surface DIC concentration (Figs.  and g, h), thus contributing to a lowering of surface pCO2 (Fig. e and g).

For the Northern Hemisphere, stronger NADW (Fig. c and d) results in decreased DIC concentrations in intermediate and deep waters of the North Atlantic by up to 75 µmol kg-1 (Fig. b). This negative DIC anomaly can be explained by a decrease in both remineralized and preformed DIC. Stronger NADW subducts more DIC-depleted surface waters into the deep ocean and reduces residence times, thus leading to a decrease in remineralized carbon. This is also associated with a ∼ 40 µmol kg-1 increase in dissolved oxygen content (Fig. a) and an up to 0.75 µmol kg-1 decrease in PO4 concentration (Fig. c). The reduction in preformed carbon (Fig. h) is most likely due to a new NADW formation site in the Labrador Sea and a slight northward shift of the deep-water formation regions in the Norwegian and Greenland seas.