This study uses the fully-coupled GFDL-ESM2M Earth system model developed at the NOAA Geophysical Fluid Dynamics Laboratory . The individual components of the model are the modular ocean model version MOM4p1 , the atmospheric model version AM2 , the land model version LM3 and the sea-ice model from . The atmosphere and land components have a horizontal resolution of 2° × 2.5°. The ocean model has a horizontal resolution of 1°, which increases to 0.3° near the equator. It has 50 vertical levels, with a layer thickness of 10 m in the upper 200 m gradually increasing to 300 m at greater depths. Additionally, the ocean model is coupled to the biogeochemistry model TOPAZv2, which simulates 30 biogeochemical tracers to represent cycling of carbon and alkalinity, as well as active pelagic calcite and aragonite and their sediment interaction . Three phytoplankton groups and implicit zooplankton grazing actively influence the production and remineralisation of carbon and alkalinity. Pelagic calcite and aragonite production depends on the respective saturation state and the associated detritus sinks through the water column and can be dissolved or deposited in the sediments, from where it can either be stored or redissolved. Ocean carbonate chemistry follows routines from the OCMIP2 protocol . The land model simulates the cycling of carbon, including different vegetation types, soils, wildfires and harvesting.

The model has been shown to perform well in representing the global carbon cycle . It simulates the uptake and storage of anthropogenic carbon by the marine and terrestrial carbon sinks close to observations , and it represents alkalinity well by actively simulating both calcite and aragonite cycling in comparison to other climate models .

A set of different model simulations with and without ocean alkalinity enhancement were performed for this study (Fig. ). After an initial spin-up , an emission-driven historical simulation was performed over the period 1861 to 2005, which follows fossil fuel and land-use change CO2 emissions as well as prescribed non-CO2 radiative forcing agents from the CMIP5 protocol. This simulation was continued for 20 simulation years in which fossil fuel CO2 emissions followed observed emissions until 2020 and national determined contributions until 2025 . From 2026 on, the adaptive emission reduction approach (AERA) was applied to reach a specified global warming level. The AERA algorithm determines the emissions pathway by estimating anthropogenic warming, calculating the forcing-equivalent remaining emissions budget for the prescribed warming target using the transient climate response to cumulative CO2-forcing-equivalent emissions (TCRE-fe), and then distributing the forcing-equivalent emissions budget over future years according to a third-order polynomial that leads to zero emissions. The timing of net-zero emissions, and thus the time window for reaching the target temperature, is an emergent property of the algorithm rather than an externally imposed constraint. If temperatures overshoot, the AERA responds by prescribing negative emissions. The pathway is recalculated every 5 years based on updated estimates of global mean temperature and the remaining CO2-forcing-equivalent emission budget at the time of the stocktake . Three target warming levels of 1.5, 2.0 and 3.0 °C above pre-industrial were chosen to cover a range of possible global warming scenarios. These simulations were run until 2500 to investigate changes on multi-centennial timescales under temperature stabilization (labelled as “Ref” in figures). Non-CO2 radiative forcing agents and land-use change were following the RCP2.6 scenario after 2005 in all future warming simulations and are kept constant after 2100 .

The Ocean Alkalinity Enhancement (OAE) experiments follow the same CO2 emission and non-CO2 forcing pathways as the reference simulations (labelled as “OAE” in figures). OAE is implemented following the protocol defined by the Carbon Dioxide Removal Model Intercomparison Project (CDRMIP) . Additional CO2 emissions due to the OAE intervention, associated with the mining, processing, and transport of alkaline materials or solutions , are not included in this study. A total of 0.14 Pmolyr-1 of alkalinity is continuously and homogeneously added to the global surface ocean, excluding regions south of 60° S and north of 70° N, as they are seasonally sea-ice covered and therefore a year-round input of alkalinity to the surface ocean may not be possible. Although this magnitude and spatial distribution of alkalinity input is unlikely to be achievable with current or near-future societal means , it provides a useful framework for exploring impacts of global-scale OAE and facilitates comparison with other model-based studies using the same framework .

Following the approach of , we perform a third set of simulations to assess feedback processes that influence the efficiency of OAE (labelled as “Ref∗” in figures). These simulations are driven by prescribed atmospheric CO2 concentrations from the “OAE” run (which are lower than in the REF run) and therefore simulate the same climate trajectory as the “OAE” simulations, but without the addition of the alkalinity. The non-CO2 forcing is the same as in the OAE and Ref simulations. The simulations thus exclude any additional atmospheric CO2 removal by the ocean due to OAE. These additional simulations help to isolate the OAE effect by removing global carbon feedback processes. We conducted those simulations only for the 2 °C warming level, as it is sufficient to illustrate the relevant processes, which can be generalized to other warming levels.

For all scenarios, we have conducted a five-member perturbed initial-condition ensemble. An initial-condition ensemble enables the separation of the forced climate response from uncertainties arising from natural internal variability . We set up the ensembles with a sea surface temperature perturbation on the order of 10⁻⁵ °C at one grid point in the Weddell Sea on 1 January 1861 .

To investigate the effect of OAE termination on both the efficiencies and the ocean acidification response, we extended one ensemble member of the 2 °C OAE simulations to the year 3000, but terminating OAE in 2500, and extended both reference simulations as well.

Several complementary metrics can be used to assess OAE efficiency, each capturing different aspects. These metrics are explained in the following and illustrated, along with the associated carbon fluxes, in Fig. .

Three processes drive the total pH response to OAE: reduction in atmospheric CO2, the pH change at the new pCO2 equilibrium, and the remaining pCO2 disequilibrium: 5ΔpHtotal≃ΔpHCDR+ΔpHequilibrated+ΔpHdisequilibrium

The three processes are described in the following.

In the Ref simulations without OAE, global surface air temperature stabilizes at the prescribed global warming levels (solid lines in Fig. a). In each case, CO2 emissions decline to zero and subsequently become slightly negative (Fig. ), which is required to maintain the stabilized warming levels through the year 2500, as the GFDL-ESM2M model has a positive zero emission commitment on multi-centennial timescales . The 31 year mean atmospheric CO2 concentration peaks at 484 ppm [455–504] in 2074 [2052–2102] in the 1.5 °C scenario, at 573 ppm [551–589] in 2096 [2081–2112] in the 2 °C scenario and at 758 ppm [742–777] in 2176 [2164–2186] in the 3 °C scenario (Fig. b). Thereafter, atmospheric CO2 decreases due to the continuous uptake of carbon by the terrestrial biosphere and predominately by the ocean.

In the OAE simulations, the continuous addition of alkalinity to the surface ocean causes additional carbon uptake from the atmosphere and reduces atmospheric CO2 concentrations compared to the reference simulations without OAE (dashed lines in Fig. b). The 31 year mean peak atmospheric CO2 is lower and earlier with OAE than without on the ensemble average, albeit with overlapping ensemble ranges: at 468 ppm [437–488] in year 2059 [2041–2088] in the 1.5 °C scenario, at 546 ppm [524–560] in year 2085 [2076–2097] in the 2 °C scenario and at 689 ppm [677–708] in year 2167 [2158–2178] in the 3 °C scenario. The decrease in atmospheric CO2 increases with the amount of global warming. Atmospheric CO2 concentrations are 73 ppm [72–74] lower in the 1.5 °C scenario, 88 ppm [86–88] lower in the 2 °C scenario, and 130 ppm [130–131] lower in the 3 °C scenario relative to their respective reference simulations between 2470 and 2500 (Fig. a–c). The greatest reductions occur within the first 100 (1.5 °C scenario) to 200 (3 °C scenario) years following the start of alkalinity addition (Fig. a–c), after which the differences increase less towards the end of the simulations (explained in Sect. ).

The reduction in atmospheric CO2 under OAE reduces global surface air temperature relative to the reference scenarios (Fig. a). Global surface air temperature is 0.64 °C [0.57–0.71] lower in the 1.5 °C scenario, 0.75 °C [0.68–0.80] lower in the 2 °C scenario, and 0.79 °C [0.65–0.96] lower in the 3 °C scenario compared to their respective references between 2470 and 2500. Unlike atmospheric CO2 concentrations, which showed the largest differences developing shortly after alkalinity addition, the surface temperature difference increases approximately linearly throughout the simulation in all scenarios (Fig. d–f). In the 2 °C scenario, the global surface air temperature decreases linearly by -0.16 °C per century [-0.15 to -0.17]. The temperature trend is slightly lower for the 1.5 °C scenario with a cooling of -0.14 °C per century [-0.13 to -0.15] and largest in the 3 °C scenario with -0.17 °C per century [-0.16 to -0.19]. This near-linear trend reflects the logarithmic relationship between radiative forcing and CO2. Using the formulation of , ΔF = -5.35 Wm-2 ln⁡(CO2Ref/CO2OAE), we estimate a near-linear reduction in radiative forcing of -0.29 to -0.34 Wm-2 per century (R2 > 0.95 for all scenarios) with a higher reduction in higher warming levels. The trends in radiative forcing show only small variations between scenarios, even though the reductions in atmospheric CO2 concentrations due to OAE differ much more. This is due to similar ratios of CO2 between the reference scenario and the OAE simulations across warming levels, but slightly higher for a warmer climate. Consequently, the three scenarios exhibit comparable cooling trends despite their differing CO2 trajectories, although the exact temperature response may also be influenced by scenario-dependent climate feedbacks and ocean heat uptake.

The regional temperature response to global OAE deployment (Fig. a–c) closely mirrors the spatial pattern of greenhouse gas-induced warming (cf. Fig. 4.19 in ), but with the opposite sign. Cooling is strongest over continents and at high northern latitudes, and weaker over the ocean, particularly in the Southern Ocean. An exception to the overall cooling occurs in the northern North Atlantic, where a localized warming develops in the higher warming scenarios. This North Atlantic warming reflects an earlier and stronger recovery of the Atlantic Meridional Overturning Circulation under OAE, which enhances ocean heat transport into the region (not shown). This effect is most pronounced in the 3 °C scenario.

Ocean alkalinity enhancement substantially increases the oceans' alkalinity inventory (Fig. ). The added alkalinity in turn leads to additional ocean carbon uptake and lowers atmospheric CO2, but part of this drawdown is offset by carbon release from the land biosphere (Fig. ). In the reference simulations, the total cumulative ocean carbon uptake since pre-industrial increases rapidly initially with increasing CO2 emissions and slows markedly toward the end of the simulation when emissions reach near-zero or become negative (solid blue lines in Fig. ). OAE leads to a persistent additional increase in cumulative ocean carbon uptake relative to the reference scenario (dashed blue lines in Fig. ). By year 2500, cumulative ocean carbon uptake is higher by 362 PgC [356–366], 380 PgC [377–383] and 407 PgC [403–412] in the 1.5, 2, and 3 °C scenarios, respectively. When feedbacks that redistribute carbon among reservoirs are neglected, the cumulative ocean uptake is substantially larger. In the 2 °C scenario, this gross cumulative ocean carbon uptake attributable to OAE (the difference between OAE and Ref∗) amounts to 628 PgC [625–632], approximately 65 % larger than the net uptake. This difference arises because reduced atmospheric CO2 concentrations due to OAE result in an anomalous outgassing of carbon stored in the ocean, thereby offsetting the gross ocean carbon uptake due to OAE. The reduction in atmospheric CO2 is modulated by anomalous carbon release from the land biosphere. Note that the historical carbon release from land is due to land-use change emissions, which dominate over the land carbon uptake due to increases in atmospheric CO2 . In the 2 °C scenario, this land carbon release totals 194 PgC [189–198], resulting in a net atmospheric carbon reduction of 189 PgC [185–192]. The land carbon response varies across warming levels. The net carbon release from land is largest in the 1.5 °C scenario at 209 PgC [196–218] and smallest in the 3 °C scenario at 129 PgC [122–134]. Consequently and following the ocean carbon uptake, the net atmospheric CO2 reduction is largest in the 3 °C scenario (279 PgC [271–284]) and smallest in the 1.5 °C scenario (156 PgC [149–164]).

Regionally, OAE leads to enhanced cumulative net carbon uptake in most ocean regions (Fig. a), with particularly strong uptake along western boundary currents and north of the Antarctic Circumpolar Current. However, some regions become anomalous cumulative carbon sources over time (i.e., these regions lose carbon to the atmosphere relative to the reference simulation without OAE). Most notably, the Southern Ocean south of 60° S shows net outgassing, as no alkalinity is added in these regions and the lower atmospheric CO2 leads to anomalous outgassing. A similar signal emerges south of the eastern tropical Pacific. There, added alkalinity produces only a weak local reduction in ocean pCO2 due to high buffer capacity and low accumulation of alkalinity, while atmospheric pCO2 is reduced more strongly by global OAE, leading to a negative air–sea pCO2-gradient and hence outgassing in these regions. The land biosphere exhibits widespread carbon loss, with strong spatial variability in magnitude (Fig. a). The gross cumulative ocean carbon uptake attributable to OAE (difference between OAE and Ref∗) is positive everywhere (Fig. b) and closely resembles the spatial pattern of cumulative anthropogenic carbon uptake . Spatial uptake patterns are relatively uniform during the first century but become increasingly heterogeneous over time. As the atmospheric CO2 levels in the REF∗ and OAE simulations are identical, the land shows no significant changes in carbon fluxes (Fig. b).

The simulated carbon cycle response and differences between warming levels relate to the efficiency of OAE. The four metrics to assess the global efficiency of OAE (Sect. ) differ strongly in their absolute values, temporal evolution, and sensitivity to the global warming stabilization level and emission pathway (Fig. ). The maximum ocean capture efficiency at the surface, determined solely by the response of seawater carbonate chemistry to alkalinity addition, remains nearly constant over time and across the three warming levels at 0.81–0.86 (black lines in Fig. ). However, it increases slightly at higher atmospheric CO2, where the ocean buffer capacity is reduced (Fig. ). At lower buffer capacities, the sensitivity of ocean pCO2 to changes in alkalinity increases relative to that with respect to dissolved inorganic carbon, allowing for an increased uptake of carbon to balance an alkalinity-induced reduction of pCO2. As a result, efficiencies are slightly higher at peak atmospheric CO2 and in the 3.0 °C scenario compared to the 1.5 °C case.

The net ocean capture efficiency, defined as the change in net air–sea CO2 flux per unit of alkalinity added, is consistently lower than the maximum efficiency and declines over time (blue lines in Fig. ). Before peak atmospheric CO2, efficiencies range between 0.64–0.68 with little variation between scenarios, but slightly higher in warmer scenarios. After the peak, they drop to 0.35–0.36 by 2400–2500, reflecting a gradual long-term decrease. This temporal decline is a result of four interacting processes that are qualitatively discussed: the redistribution of alkalinity within the ocean interior, changes in the ocean buffer capacity, carbon effluxes from the land biosphere and ocean internal changes of alkalinity and DIC by biological activity. First, the added alkalinity undergoes vertical mixing and thereby, the volume in exchange with the atmosphere, over which the equilibration with the atmosphere occurs, increases steadily over time. The larger the volume, the more old carbon-rich waters are being upwelled to the surface and see an atmosphere with reduced atmospheric CO2 concentration, resulting in anomalous outgassing to the atmosphere and hence lowering net ocean carbon uptake. We use the effective mixed layer depth for OAE from , characterizing the ocean volume under influence of alkalinity addition. Note that this is not the depth of the commonly used wind-driven mixed layer. Scenarios show negligible differences in magnitude and temporal change in this volume, with the effective mixed layer depth increasing near-linearly from around 140 m in 2026 to about 1400 m in 2500 (not shown), leading to a steady decrease in efficiency. Second, a low ocean buffer capacity means that the change in oceanic pCO2 per change in dissolved inorganic carbon is large compared to the sensitivity of atmospheric pCO2 to the atmospheric carbon content. In this case, a lot of carbon is taken up by the ocean until both systems equilibrate at an intermediate pCO2 and efficiency is high. The buffer capacity is lower in the high warming scenarios and decreases before peak atmospheric CO2 and increases afterwards in each scenario (Fig. ). As the timing of peak CO2 differs between scenarios, so does the influence of buffer capacity onto the efficiency. Third, the reduction in atmospheric CO2 by ocean carbon uptake is buffered by a relative efflux of carbon from the land biosphere. This efflux increases atmospheric pCO2 and allows for more ocean carbon uptake until equilibrium is reached, increasing efficiency. We find a steady increase in the cumulative land efflux over time and largest effluxes in the 1.5 °C scenario and lowest in the 3 °C scenario (Sect. ). The first three processes only occur with interactive atmospheric CO2 and are the main drivers of the temporal changes in the net ocean capture efficiency. Fourth, biological activity in the ocean, most prominently calcification, is depending on the ocean state, mostly pH and can influence OAE . In our Earth system model, calcification responds to changes in the calcite and aragonite saturation states and we find more calcification under OAE than in the reference scenarios, leading to an overall reduction in additional alkalinity within the ocean (Fig. c). Calcification reduces DIC, allowing for further ocean carbon uptake and increasing efficiency, but also removes alkalinity and thereby decreases efficiency. Overall, calcification leads to a net reduction in efficiency since calcification removes twice as much alkalinity as DIC, leaving a net increase in pCO2. The first three effects dominate the temporal change in net ocean capture efficiency: Before peak atmospheric CO2, a decline in buffer capacity increases the efficiency of alkalinity-driven CO2 uptake. However, this gain is largely compensated by ocean outgassing across a larger ocean volume in response to lower atmospheric CO2, mediated by an additional carbon efflux from the land-biosphere. As a result, net ocean capture efficiency remains nearly constant. Afterwards, as the increase in buffer capacity now reduces efficiency as well, the efficiency generally decreases. The differences in buffer capacity and the land effluxes between warming scenarios balance each other in the long-term, resulting in similar efficiencies between scenarios around 2500.

The gross ocean capture efficiency is close to the maximum ocean capture efficiency, with a mean value of 0.79 [0.78–0.79] in the 2 °C scenario (green line in Fig. b). It does not change significantly over time along the emission pathway. By definition, the gross efficiency isolates the direct oceanic response to alkalinity addition, excluding the effects of changes in atmospheric CO2 due to changes in ocean and land carbon uptake. The gross efficiency is limited by incomplete equilibration following continuous OAE.

The net atmospheric CO2 reduction efficiency is lower than the net ocean capture efficiency (red lines in Fig. ), with pre-peak values of 0.52–0.60 with higher efficiencies in the higher warming levels. After peak atmospheric CO2, efficiencies decline to 0.11–0.12 between 2400 and 2500. The difference between atmospheric CO2 reduction efficiency and the net ocean capture efficiency arises because the land biosphere releases carbon back to the atmosphere when atmospheric CO2 levels decline, with larger effluxes in the lower warming scenarios (Fig. ). These effluxes are directly accounted for in the net atmospheric reduction efficiency, while also influencing the net ocean capture efficiency indirectly as described above. Variability across individual ensemble members in atmospheric CO2 reduction efficiency reflects the strong year-to-year variability in land-atmosphere carbon exchange. The near-constant pre-peak efficiencies explain the near-linear reduction of atmospheric CO2 shown in Fig. , with a reducing decrease after peak atmospheric CO2 as the efficiency drops as well.

To assess the temporal evolution of efficiencies after OAE termination, we evaluate the cumulative efficiency, defined as total carbon uptake divided by total alkalinity addition, since the rate of alkalinity addition is zero after 2500 (Fig. ). Our results show that the cumulative gross ocean capture efficiency remains below the maximum efficiency over the subsequent 500 years. A persistent ocean–atmosphere pCO2 difference of approximately 2 µatm after OAE termination (not shown) indicates that full equilibration has not yet been achieved. While surface waters continue to equilibrate, previously subducted alkalinity resurfaces through upwelling and mixing, allowing unrealized alkalinity potential to further reduce oceanic pCO2. In contrast, the cumulative net ocean capture efficiency continues to decrease, as the ocean begins to outgas CO2 in response to lower atmospheric CO2 concentrations. Upwelled waters, no longer supported by ongoing alkalinity addition, equilibrate with the reduced atmospheric CO2 and release carbon. This drives a redistribution of carbon among reservoirs, with the land biosphere taking up carbon again relative to the reference simulation. The net atmospheric CO2 reduction efficiency decreases further but begins to stabilize, indicating a balance of ocean and land carbon fluxes or a convergence toward a new steady-state carbon balance.

Regional differences in maximum ocean capture efficiency at the surface reflect the sensitivity of surface pCO2 to alkalinity relative to dissolved inorganic carbon, which is lowest in the tropics and highest at high latitudes (Fig. a). This is consistent with regional differences in the buffer capacity . In some regions such as the North Atlantic, North Pacific and parts of the Southern Ocean, gross uptake efficiency locally exceeds the theoretical maximum expected from surface alkalinity addition alone (Fig. b). This is due to lateral redistribution of un-equilibrated alkalinity by ocean circulation, which adds local carbon uptake potential and explains the larger than maximum efficiency. Additionally, these regions exhibit higher gas transfer velocities . Therefore, less un-equilibrated alkalinity is transported away from these regions, while incoming un-equilibrated alkalinity can equilibrate faster locally. However, show no specifically high efficiency in these regions when alkalinity is only added within the region itself, and find a strongly increased efficiency in these regions due to resurfacing of alkalinity that was added below the surface in other regions, both indicating that the redistribution of un-equilibrated alkalinity is the dominant process.

The addition of alkalinity leads to a pronounced increase in global mean ocean surface pH relative to the reference scenarios (Fig. a). In the absence of alkalinity enhancement (Reference scenarios), global mean surface pH initially declines and subsequently recovers during the global warming stabilization phase, reflecting the initial rise and later decline in atmospheric CO2. In the OAE scenarios, global mean surface pH during the period 2470 to 2500 is elevated relative to the reference scenario by 0.105 [0.103–0.106] in the 1.5 °C scenario, 0.111 [0.109–0.112] in the 2 °C, and 0.124 [0.123–0.126] in the 3 °C scenario (Fig. b). Under the 1.5 °C scenario, this increase is sufficient to raise global mean surface pH above pre-industrial levels, thereby fully reversing historical surface ocean acidification. The temporal evolution of surface pH closely follows that of atmospheric CO2, with the largest pH changes occurring in the high-warming scenario, where OAE leads to the strongest atmospheric CO2 drawdown. This is also evident at the regional scale. All regions experience increases in surface pH due to OAE, with slightly larger changes in the 3 °C scenario compared to the lower warming scenarios (Fig. d–f).

The increase in global surface pH is primarily driven by the reduction in atmospheric CO2, with smaller additional contributions from OAE-specific direct chemical effects, such as the equilibrated pH response and transient pCO2 disequilibrium (Fig. c; see Sect.  for further details on the method). In 2100 in the 2 °C global warming scenario, the combined direct chemical effects of OAE account for about 48 % of the total additional pH increase. Over time, the ocean–atmosphere system progressively equilibrates, and the contribution from atmospheric CO2 drawdown becomes increasingly dominant, while both direct chemical effects diminish. Averaged over 2470 to 2500, approximately 72 % of the total global surface pH increase can be attributed to reduced atmospheric CO2, 17 % due to equilibrated pH response and 9 % due to remaining pCO2 disequilibrium. The residual is a result of approximations in calculating the three contributions.

Following OAE termination in 2500, the pH difference between the OAE simulation and reference simulation gradually declines (Fig. b), resulting in a remaining pH response of 0.071 in the year 3000. This reduction in OA mitigation is initially dominated by a decrease in pCO2 disequilibrium (Fig. c), as surface waters continue to equilibrate in the decades following termination. However, ongoing upwelling and mixing of previously subducted alkalinity continue to lower oceanic pCO2, maintaining a nearly constant pCO2 offset of around 2 µatm between the OAE and Ref^∗ simulations. Consequently, pCO2 disequilibrium still explains 2 % of the total pH change 500 years after OAE termination. The contribution of the equilibrated pH response also decreases after 2500, reflecting the declining surface alkalinity difference between the OAE and reference simulations, and accounts for approximately 7 % of the pH response in the year 3000. In contrast, while the carbon dioxide removal effect weakens after OAE termination (Fig. b), its relative importance increases, explaining about 88 % of OA mitigation by the year 3000. This weakening of the CDR effect is due to continued outgassing of oceanic CO2 after OAE termination, driving an increase in atmospheric CO2 and a decrease in surface ocean pH.

In the subsurface ocean, the largest global mean ocean acidification mitigation from OAE is simulated at a depth of around 250 m, with a maximum pH increase of 0.137 [0.135–0.139] by year 2500 in the 2 °C scenario (Fig. a). As at the surface, the subsurface pH mitigation response is larger in the 3 °C scenario and smaller in the 1.5 °C scenario (not shown). The subsurface maximum reflects the competing effects of two opposing vertical gradients: the OAE-induced alkalinity enhancement and associated DIC reduction weakens with depth, whereas the sensitivity of pH to changes in alkalinity and DIC increases with depth. Their combined effect yields a maximum pH response below the surface. Similar subsurface maxima have also been reported for historical ocean acidification . Consistent with the surface response, the direct chemical effects of OAE (Fig. c) contribute substantially early on, while the CDR effect becomes increasingly dominant over time (Fig. b). In 2100, the CDR effect explains 54 % of the total interior pH change, with the direct chemical effects accounting for the remaining 46 %. By 2500, the CDR contribution increases to 69 %. After OAE termination in 2500, the total subsurface pH mitigation gradually declines and the relative contribution of the CDR effect increases to 74 % by the year 3000. Below 1000 m, however, pH mitigation continues to increase despite the OAE termination, because ongoing ocean ventilation progressively propagates the accumulated pH deficit into the deep ocean on multi-centennial timescales.