The large atmospheric CO2 concentration in 2100 under RCP8.5 reflects the large projected increase in emissions during this century, while under RCP2.6 a similar atmospheric concentration of CO2 is seen in 2100 as at the beginning of the simulation (2020) (Fig. 2a). We note that atmospheric CO2 levels in our CSIRO-MK3L-COAL for the control cases are greater than for their respective concentration driven RCPs due to nutrient limitation in the land, leading to reduced carbon uptake (Q. Zhang et al., 2014).

Under all emissions scenarios and experiments, AOA leads to reduced atmospheric CO2 levels (Fig. 2a). Under RCP8.5, AOA reduces atmospheric concentration by 82–86 ppm; representing a ∼ 16 % decrease in atmospheric concentration. In contrast to RCP8.5, AOA under RCP2.6 leads to a smaller reduction in atmospheric concentration (53–58 ppm). Fig. 2a shows that, by the end of the century, AOA compensates for the projected increase in atmospheric CO2 due to RCP2.6.

Over the 2020–2100 period, the reduction in atmospheric CO2 levels associated with AOA is primarily due to increased ocean carbon uptake, offset by small decreases in the land surface carbon uptake (Table 1). In the ocean, RCP8.5 leads to much greater net uptake than RCP2.6, about 50 % more, due to the larger (and growing) disequilibrium between the atmosphere and ocean.

In the ocean, the relative increase in carbon uptake in response to AOA is primarily abiotic in nature. Consistent with Keller et al. (2014) and Hauck et al. (2016) the simulated changes in ocean export production were very small (∼ 0.2 PgC) under RCP8.5, which was due to small changes in ocean state, e.g. stratification. Under RCP2.6, it was slightly larger at 1.2 PgC, but still less than 1 % percent of the total ocean uptake increase simulated under AOA, due to small changes in ocean state in a more stratified ocean. In contrast, the relative decreases in land carbon uptake were biotic in nature. The simulated cooling drove both a reduced net primary production, leading to reduced carbon uptake, and an increase in carbon retention associated with a reduction in heterotrophic respiration. However, overall, the net decrease in land carbon uptake means that in the response to AOA globally the reduced net primary production dominated. On the land, in the RCP8.5 simulation there was a smaller reduction in carbon uptake than in RCP2.6 (Table 1), due to larger decreases in surface air temperature (SAT) over land in RCP2.6 than RCP8.5 (∼ 2×; see Sect. 3.1.2). The land carbon cycle response was also smaller under high than low emissions due to nutrient limitation being reached, thereby limiting the effect of CO2 fertilization (Q. Zhang et al, 2014).

For both emissions scenarios, the four AOA experiments all produced similar reductions in atmospheric CO2 concentrations (Fig. 2) with less than a 5 % difference in the total land and ocean carbon uptake. The global changes in land and ocean carbon uptake are not very sensitive to where we add the alkalinity to the surface ocean. This is consistent with Kohler et al. (2013) who saw little difference in adding olivine along existing shipping tracks, versus uniformly adding it to the surface ocean. It is also consistent with regional addition studies of Ilyina et al. (2013), Feng et al. (2016), and Feng et al. (2017) which demonstrated a global impact.

Our simulated total increased carbon uptake under AOA_G with RCP8.5 (179 PgC) is comparable to the 166 PgC reported by Keller et al. (2014). Their cumulative increase in ocean carbon uptake by 2100 of 181 PgC is in very good agreement with our value of 184 PgC. However, they simulated a reduction in land uptake nearly twice the -5.8 PgC reduction in our AOA_G simulation. These differences reflect both the lower sensitivity of the simulated climate feedbacks in our ESM, and differences in land surface models.

In the control simulations, the global mean surface air temperature (SAT; 2 m) increased in the period 2020–2100 with RCP2.6 simulating a net warming of 0.4 ± 0.1 K while RCP8.5 warmed by 2.7 ± 0.1 K (2081–2100). AOA experiments simulated a reduction in global mean SAT relative to their corresponding control simulation (Fig. 2b). Within each emissions scenario the global mean SAT decline associated with AOA is always greater and more variable over the land than ocean (Table 1). In the period 2081–2100 we see larger mean changes in SAT under RCP2.6 than RCP8.5 primarily due to differences in atmospheric CO2 growth rate. Krasting et al. (2014) showed that the slower rate of emissions, the lower the radiative forcing response. This occurs in response to the timescales associated with the uptake of heat and carbon. Consequently, under RCP8.5 the atmospheric CO2 growth rate is much faster than RCP2.6, leading to a strong radiative forcing response. This explains why, despite a larger reduction in atmospheric CO2 concentration under RCP8.5, the biggest reduction in global mean SAT occur under RCP2.6. These mean changes are also associated with large inter-annual variability.

Under RCP2.6, all the AOA experiments keep global warming levels much closer to values in 2020 than RCP2.6 by the end of this century (2100; Fig. 2b). In contrast, under the RCP8.5 scenario, none of the AOA experiments have a significant impact on the projected warming by the end of this century (less than 10 %) reflecting the large warming projected under high emissions.

Within each of the scenarios, there are some differences in the magnitude of the cooling within the four different AOA experiments; however, these are smaller than the inter-annual variability over the last two decades of the simulations. Therefore, it appears that the global mean SAT decline with AOA is not very sensitive to where the alkalinity is added under either emission scenario.

The global mean cooling associated with AOA_G under RCP8.5 (-0.16 ± 0.08 K; 2081–2100) is close to the mean surface air temperature cooling of -0.26 K reported by Keller et al. (2014) for similar levels of AOA. These differences may reflect the simplified atmospheric representation of the University of Victoria (UVic) Earth system model of intermediate complexity and different climate sensitivities.

Here, we quantify changes in ocean acidification in terms of pH and aragonite saturation state changes. We consider these two diagnostics because they are associated with different biological impacts and are not necessarily well correlated (Lenton et al., 2016). In the future, the global mean changes in pH and aragonite saturation state will be proportional to the emissions trajectories following Gattuso et al. (2015), with the largest changes associated with the higher emissions (RCP8.5) (Fig. 2c–d). By 2100, despite the return to 2020 values of atmospheric CO2 concentration under RCP2.6 (Fig. 2), neither pH nor aragonite saturation state return to 2020 values, consistent with Mathesius et al. (2015).

In the 2020–2100 period, AOA under RCP2.6 led to much larger increases in surface pH and aragonite saturation state, more than 1.3 times, and 1.7 times that of RCP8.5, respectively (Table 2). These changes reflect the differences in the mean state associated with high and low emissions, specifically the difference between alkalinity and dissolved inorganic carbon (ALK-DIC), a proxy for ocean acidification (Lovenduski et al., 2015). As the values of DIC in the upper ocean are larger under RCP8.5 than RCP2.6, the difference between ALK and DIC (ALK-DIC) is smaller and the chemical buffering capacity of CO2 or Revelle factor (Revelle and Suess, 1957) is less. This means that, for a given addition of ALK the increase in the upper ocean DIC will always be greater under RCP8.5 due to its reduced buffering capacity. Consequently, the changes in ALK-DIC with AOA are greater under RCP2.6 than RCP8.5, which translates to greater increases in pH and aragonite saturation state.

While there was a significant difference in pH and aragonite saturation state changes with AOA between high and low emissions cases, the global mean changes for different AOA experiments within each scenario are quite similar (Table 2). The exception to this is the AOA_SP experiment, where the pH and aragonite saturation state changes are only ∼ 75 % of the change in the other AOA experiments. This reduced change in the polar region is consistent with the smaller changes in the surface ocean alkalinity values associated with AOA_SP (Table 1). These differences at higher latitudes reflect the enhanced subduction of alkalinity away from the surface ocean into the ocean interior that occurs in the high latitude oceans (Groeskamp et al., 2016).

AOA_G under RCP8.5 leads to a relative increase in pH of 0.06, which is consistent with Keller et al. (2014), while the relative increase in aragonite saturation state (0.28) is also very close to their simulated value (0.31). To put these changes into context, the estimated decrease in pH since the pre-industrial period is 0.1 units (Raven et al., 2005), and is already responsible for detectable changes in the marine environment (Albright et al., 2016).

For both scenarios, the greatest surface alkalinity changes occur where the alkalinity is added (Fig. 3). Spatially, under either emission scenario, the relative differences in 2090 are very similar; consequently, we only show the changes under RCP2.6 (Fig. 3). The only significant differences occur in the Arctic, reflecting larger longer-term changes in alkalinity projected under higher emissions (Yamamoto et al., 2012).

Overall, the greatest increases are seen in the tropical ocean (AOA_T) suggesting this is the most efficient region in retaining the added alkalinity in the upper ocean. This reflects the fact that subduction processes in the tropical ocean are less efficient than in other regions such as the higher latitudes. The (ice-free) subpolar oceans (AOA_ SP) produced the smallest relative increase in alkalinity, and this reflects the strong and efficient surface to interior connections through subduction occurring at higher latitudes (Groeskamp et al., 2016). The global mean relative increase associated with AOA in the subtropical gyres (AOA_ST) and globally (AOA_G) fall between the tropical (AOA_T) and higher latitude (AOA_SP) values. In the case of AOA_ST, this reflects the timescales associated with the longer residence time of upper ocean waters in the subtropical gyres.

The most modest relative increase in alkalinity occurs in the ice-covered regions where alkalinity is not explicitly added. Interestingly, even when alkalinity is added in the very high latitude Southern Ocean, it is carried northward by the Ekman current which explains the very modest increase in the region where AOA occurs between 50 and 60∘ S. In terms of the total alkalinity added to the surface ocean, about one-third remains in the upper 200 m by 2100 (Fig. 4). Specifically, for AOA_G we see that 31 % remains in the upper ocean, and for AOA_T and AOA_ST that 34 % remains in the upper ocean, while for AOA_SP the figure is 22–24 % which (as anticipated) is lower than in other regions.

Spatially, AOA in the higher latitude regions (AOA_SP) leads to very large relative increases in alkalinity (> 1000 µmol kg-1; 2090) occurring along the northern most boundary of the northern subpolar gyres, particularly the North Pacific. Clearly, in this region the rate of AOA exceeds the rate of subduction allowing alkalinity to build up. Large relative increases in alkalinity also occur in the Southern Ocean under AOA_SP, particularly along western boundary currents. However, in contrast to northern high latitudes the values still remain low suggesting that the rate of addition does not exceed the rate of subduction even under the highest emission scenario.

AOA_ST shows a large relative increase of ∼ 300 µmol kg-1 (2081–2100) in the subtropical gyre regions. Overall, we find that these relative increases are quite homogenous across the entire subtropical gyres, with strong mixing with tropical waters leading to significant relative increases in tropical Atlantic, western Pacific and Indian Oceans. Within the tropical ocean, under AOA_T the largest relative changes are found across the entire tropical Indian Ocean (∼ 400 µmol kg-1) with large relative increases also seen in the Indonesian seas (∼ 280 µmol kg-1; 2081–2100). Away from the tropical Indian Ocean, we find that relatively homogenous increases occur in the western Pacific and the Atlantic, with much more modest relative increases in the eastern Pacific reflecting the dominant east to west upper ocean circulation. AOA_T leads to relative increases in surface alkalinity that are consistent with the response to AOA_ST – in the region of ∼ 130 µmol kg-1 (2081–2100).

In the case of AOA_G, a relatively uniform net increase in alkalinity occurs in all regions with the exception of the upwelling regions such as the tropical Pacific, which showed a more modest relative increase. In AOA_G there is little evidence of any of the very large increases in alkalinity seen in the more regional AOA experiments. This spatial pattern of relative increase is broadly consistent with the pattern of global alkalinity increase simulated by Ilyina et al. (2013) and Keller et al. (2014) for AOA in the (ice-free) global ocean.

As only about 30 % of the total AOA remains in the upper 200 m, we explore the fate of this alkalinity in the interior ocean in the zonal sections of alkalinity (Fig. 4). As the pattern is very similar between RCP2.6 and RCP8.5, we only show RCP2.6, noting that in the North Atlantic the projected ocean stratification is stronger under higher emissions (not shown) leading to slightly decreased subsurface values. This increased stratification is consistent with other studies (e.g. Yool et al., 2015).

Unlike the surface plots of AOA, the relative increases in subsurface alkalinity due to AOA are very similar across all experiments. This heterogeneous spatial pattern of alkalinity increase is associated with water entering the interior ocean along specific surface to interior pathways. Alkalinity also moves into the interior ocean along the poleward boundaries of the subtropical gyres, associated with the formation and subduction of mode waters, and an increase in the subtropical gyres associated with large-scale downwelling and deep mixing in the North Atlantic. The changes in alkalinity are mainly found in the upper ocean (< 1000 m) which reflects the relatively short period of alkalinity addition. Given the short period, this is analogous to present-day observed distributions of anthropogenic carbon (Sabine et al., 2004).

As the changes in export production are very small, the large changes in the interior alkalinity concentrations primarily reflect the physical transport, rather than the sinking and remineralization of calcium carbonate. Clearly other biological processes, not represented in our model, have the potential to impact the surface and interior values of alkalinity (Matear and Lenton, 2014). One such process is the reduction in the (rain) ratio of PIC : POC under higher emissions (Riebesell et al., 2000). However, it has been shown that even a very large reduction in PIC production (50 %) would not significantly impact our results (Heinze, 2004). Unfortunately, at present the magnitude and sign of many of these other feedbacks remain poorly known (Matear and Lenton, 2014); consequently, quantifying their impact on our results is very difficult, and beyond the scope of this study.

The similarity in global ocean carbon uptake associated with all AOA experiments for a given emission scenario hides the large spatial differences between simulations. Given that the largest carbon cycle response occurs in the ocean (Table 1), we focus on this response for RCP8.5 and RCP2.6 (Figs. 5 and 6). As expected, ocean carbon uptake is strongly enhanced in the regions of AOA. Away from regions of AOA, there is a reduction in carbon uptake, associated with the weakening of the gradient in CO2 between the atmosphere and ocean due to AOA. Interestingly, the largest increase spatially occurs in the Southern Ocean under AOA_SP for RCP2.6, while in contrast the largest changes under RCP8.5 occur in the tropical ocean under AOA_T. The very small changes in export production in RCP2.6 were located in the Arabian Sea (not shown), likely driven by enhanced mixing in this region. While these changes are < 1 % of the total change in carbon uptake, they may nevertheless be important regionally.

The decrease in global mean SAT associated with all AOA experiments for a given emission scenario again hides the large spatial differences between the simulations. The response of surface temperature is spatially very heterogeneous (Figs. 7 and 8) and the regional surface temperature changes are very similar between the two emissions scenarios. The exception to this is the Arctic which did not show a consistent response across the different AOA experiments, reflecting the period over which the mean changes were calculated, and the simulated large variability in SAT in this region. Under both emission scenarios, the largest cooling associated with AOA occurs over northern Russia and Canada, and Antarctica (greater than a -1.5 K cooling) with a larger cooling in these regions under RCP2.6.

AOA in the RCP2.6 scenario brings about a net cooling of the surface ocean with the exception of the North Atlantic, east of New Zealand, and off the southern coast of Alaska, which show a very modest warming. A similar pattern is evident in RCP8.5; however, there is a greater cooling in the high latitudes, and less cooling in the lower latitudes than under RCP2.6.

Globally, the response of pH and aragonite saturation state associated with AOA are similar; however, large spatial and regional differences are present (Figs. 9–14). To aid in the interpretation of changes in aragonite saturation state, overlain on the aragonite saturation state maps are the contours corresponding to the value of 3 – the approximate threshold for suitable coral habitat (Hoegh-Guldberg et al., 2007). On these surface maps and subsequent section plots we plot the saturation horizon, i.e. the contour corresponding to the transition from chemically stable to unstable (or corrosive), i.e. aragonite saturation state is equal to 1 (Orr et al., 2005).

The largest relative changes in pH and aragonite saturation state were associated with regions of AOA (Figs. 9–12), reflecting increases in the surface values of alkalinity (Fig. 3). All simulations increase pH and aragonite saturation state in the Arctic despite no direct addition in this region, with the largest changes here associated with AOA_G and AOA_SP. Interestingly, all simulations show little to no increase in the high latitude Southern Ocean, consistent with more efficient transport of the added alkalinity into the ocean interior.

The changes in pH associated with AOA experiments under RCP8.5, while spatially very different particularly when added in the subpolar ocean, are still much less than the decreases associated with RCP8.5 with no AOA (Fig. 9). In terms of aragonite saturation state (Fig. 10), the conditions for coral growth in the tropical ocean remain very unfavourable by the end of century (i.e. aragonite saturation state < 3) under all regional and global experiments, with the exception of AOA_T, where a very small region in the central Pacific Ocean exhibits suitable conditions.

Consistent with Feng et al. (2016), we find that this level of AOA under RCP8.5 is insufficient to ameliorate or significantly alter the large-scale changes in ocean acidification. More positively, at the higher latitudes the saturation horizon is moved poleward with the largest shift associated with AOA_SP, and the smallest shift at the high latitudes occurring under AOA_T. Consistent with these changes, we see a deepening of the saturation horizon everywhere, and little difference spatially between AOA experiments, consistent with zonal mean changes in alkalinity for the four AOA experiments (Fig. 11).

The spatial pattern of changes associated with AOA under RCP2.6 is broadly consistent with that seen under higher emissions; however, the magnitude of the response is much larger – again, due to the larger differences between Alkalinity and DIC with AOA under RCP2.6 (Figs. 12 and 13). In terms of aragonite saturation state, the area of tropical ocean favourable for corals is considerably expanded. As anticipated the largest changes in the area favourable for tropical corals is associated with AOA_T, closely followed by AOA_ST. As the saturation horizon does not reach the surface under RCP2.6, we can only look at the changes in the interior ocean. Here, there is a deepening in the saturation horizon of a very similar magnitude in all experiments (Fig. 14), with the exception of the Arctic. Here, the response of the saturation horizon is more sensitive to the location of the AOA, varying between ∼ 100 m under AOA_T and ∼ 280 m under AOA_SP (Fig. 14).

Spatially, the large changes in ocean acidification in response to AOA under RCP2.6 more than compensate for the changes in ocean chemistry due to low emissions in the period 2020–2100. Globally, the changes in the period 2020–2100 are sufficient to reverse or compensate for the changes since the pre-industrial period (1850). However, spatially in some regions such as equatorial upwelling, an important area of global fisheries (Chavez et al., 2003), AOA in fact leads to higher values of aragonite saturation state and pH than the ocean experienced in the pre-industrial period (Feely et al., 2009). We can only speculate on the potential impact of a reduction in aqueous CO2 and elevated pH levels on marine biota in these regions. For a recent review of the potential impact of rising pH and aragonite saturation state on marine organisms, we direct the reader to Renforth and Henderson (2017).

In this paper, while we have focused on year-round AOA, as a sensitivity experiment we also explored whether AOA added in summer or winter was more efficient. To do this, we focused on the higher latitudes regions where the largest seasonal changes in mixing are found (de Boyer Montegut et al., 2004; Trull et al., 2001). Here, we tested whether AOA in either summer or winter was more effective than year-round addition. To test this for RCP8.5, we add alkalinity only during the summer at half of the annual rate (or 0.125 PmolALK yr-1) in the AOA_SP region.

Our results showed that the response to AOA in summer was very close to 50 % of the response of the year-round addition associated with AOA_SP (or 0.25 PmolALK yr-1). This suggests that the response of AOA appears invariant with regard to when the alkalinity is added. This also suggests, consistent with published studies (e.g. Keller et al., 2014; Feng et al., 2016; Kohler et al., 2013), that the response of the ocean to different quantities of AOA is scalable under the same emissions scenario. Whether this is true under very much larger additions of alkalinity, as simulated by Gonzalez and Ilyina (2016), is less clear.