Simulated effect of calcification feedback on atmospheric CO2 and ocean acidification

Ocean uptake of anthropogenic CO2 reduces pH and saturation state of calcium carbonate materials of seawater, which could reduce the calcification rate of some marine organisms, triggering a negative feedback on the growth of atmospheric CO2. We quantify the effect of this CO2-calcification feedback by conducting a series of Earth system model simulations that incorporate different parameterization schemes describing the dependence of calcification rate on saturation state of CaCO3. In a scenario with SRES A2 CO2 emission until 2100 and zero emission afterwards, by year 3500, in the simulation without CO2-calcification feedback, model projects an accumulated ocean CO2 uptake of 1462 PgC, atmospheric CO2 of 612 ppm, and surface pH of 7.9. Inclusion of CO2-calcification feedback increases ocean CO2 uptake by 9 to 285 PgC, reduces atmospheric CO2 by 4 to 70 ppm, and mitigates the reduction in surface pH by 0.003 to 0.06, depending on the form of parameterization scheme used. It is also found that the effect of CO2-calcification feedback on ocean carbon uptake is comparable and could be much larger than the effect from CO2-induced warming. Our results highlight the potentially important role CO2-calcification feedback plays in ocean carbon cycle and projections of future atmospheric CO2 concentrations.

. A warmer ocean could also accelerate respiration/remineralization rate of organic carbon and cause a reduction in the vertical flux of particulate organic carbon (POC) to the abyssal ocean. This reduced vertical transport of POC would weaken the ocean biological pump and decrease the ocean's ability to take up carbon, providing a positive feedback to the growth of atmospheric CO 2 6 . On the other hand, increased C:N:P stoichiometry in seawater due to rising partial pressure of CO 2 in the ocean could enhance extracellular organic matter production, strengthening the ocean biological carbon pump and providing a negative feedback to rising atmospheric CO 2 [7][8][9] .
In addition to global warming, ocean acidification, through its effect on the ocean carbon cycle, could provide feedbacks to atmospheric CO 2 . Global mean ocean surface pH, which can be used to quantify the degree of ocean acidification, has dropped by 0.1 units 1 , representing a 26% increase in hydrogen ion concentration since the industrial revolution. The rise of hydrogen ion concentration would consequently lower carbonate ion concentration ( ) − [CO ] 3 2 and in turn causes a reduction of seawater CaCO 3 (aragonite or calcite) saturation state, which is defined as Here, ⁎ K sp is the stoichiometric solubility product for aragonite or calcite, which are two different polymorphs of CaCO 3 10 .
Calcifying organisms that use CaCO 3 to precipitate their shells or skeletons may not be able to acclimate to the reduction of CaCO 3 saturation state 11 . Gattuso et al. detected a nonlinear relationship between calcification rate and CaCO 3 saturation state from experimental results of certain types of coral species, and suggested calcification rate may drop substantially as a result of decreasing aragonite saturation state 12 . Langdon et al. through mesocosm experiment results, argued that declining aragonite saturation state is a primary factor that attenuates coral reef calcification 11 . Laboratory experiments with coccolithophorids 13,14 and analyses of foraminiferal shell weight observational record across glacial-interglacial Termination 15 indicated a reduction in CaCO 3 production with increasing CO 2 concentration and resulting ocean acidification. More recent observational or experiment results also provided evidences of the negative response of calcification to ocean acidification [16][17][18] . In spite of the abundant observational and experimental evidence, the sensitivity of the response of calcification to acidification varies dramatically between experiments using different species of calcifying groups or manipulation methods.
The process of calcification decreases − [CO ] 3 2 , suppressing the ocean's ability to absorb atmospheric CO 2 . Therefore, the potential reduction of calcification as a result of CO 2 -induced ocean acidification could enhance the ocean's uptake of carbon, providing a negative feedback for rising atmospheric CO 2 19 , which is termed as CO 2 -calcification feedback here. Recently, a few modeling studies have been conducted to examine effects of the CO 2 -calcification feedback. In these studies, the parameterization schemes that link CaCO 3 production to CaCO 3 saturation state (Ω ) are based on different results of experimental studies, and as a consequence, estimates of the effect on oceanic uptake of atmospheric CO 2 from CO 2 -calcification feedback varies among studies [20][21][22][23][24] .
As an extension of previous studies, here we further examine the effect of CO 2 -calcification feedback on the oceanic uptake of atmospheric CO 2 . We incorporate the calcification-Ω dependence into an Earth system model of intermediate complexity to quantify the strength of the CO 2 -calcification feedback in mitigating the growth of atmospheric CO 2 and ocean acidification. Usually, previous studies on CO 2 -calcification feedback assume a single type of calcification-Ω parameterization scheme. Here, different types of calcification-Ω parameterization schemes are used, which enables us to assess the importance in the parameter value (parameter uncertainty) and the equation form (model structural uncertainty) of the calcification response to ocean acidification. Also, we compare the strength of CO 2 -calcification feedback with the feedback induced by climate change. Furthermore, we assess the effect of different parameterization of CO 2 -calcification feedback on projected ocean acidification. This study aims to further our understanding in the role of the CO 2 -calcification feedback on the ocean carbon cycle and atmospheric CO 2 , which is important for a reliable projection of future atmospheric CO 2 concentrations and climate change.

Results
To quantify the effect of CO 2 -calcification feedback, we conduct a series of Earth system model simulations that incorporate different parameterization schemes representing the dependence of calcification rate on saturation state of CaCO 3 (refer to Method section and Fig. 1). All model simulations last from year 1800 to 3500 with SRES A2 CO 2 emission scenario until year 2100 and zero CO 2 emission afterwards. Total cumulative anthropogenic CO 2 emissions amount to 2270 PgC.
To quantify the effect of climate change on the carbon cycle, we also conduct additional simulation experiments in which CO 2 -induced warming does not affect the carbon cycle. A detailed description of the model and simulation experiments can be found in the Method section.

Model-observation comparison.
To test the performance of the UVic model in simulating present day carbon cycle, the model-simulated distributions of key variables of the ocean carbon cycle are compared with GLODAP observations 25 . As shown in Fig. 2, the simulated vertical profiles of alkalinity and dissolved inorganic Future projections. In the following, we first present results from simulations including the effect of CO 2 -induced warming on the ocean carbon cycle. Then we compare the effect of CO 2 -calcification feedback with that from CO 2 -induced warming, which is obtained by the differences between the simulations with and without CO 2 -induced warming.
As shown in Fig. 3, in the control simulation (S0), by year 2100, the global ocean has absorbed 581 PgC of anthropogenic CO 2 . After the cessation of CO 2 emission at 2100, the ocean continues to absorb CO 2 , and by year 3500, the global ocean has a total CO 2 uptake of 1462 PgC (Fig. 3b, also see Supplementary Table S1). As for the carbon uptake by the terrestrial biosphere, by year 2100 and 3500 the land has absorbed 629 and 2236 Pg C, respectively (Fig. 3d). Atmospheric CO 2 reaches a peak value of 897 ppm at year 2100 ( Fig. 3e). A cessation of CO 2 emission leads to a gradual decline of atmospheric CO 2 . By year 3500, atmospheric CO 2 concentration is 612 ppm with a global mean surface warming of 3.9 °C (Fig. 3f). The ocean's absorption of anthropogenic CO 2 acidifies the global ocean (Fig. 4). By year 2100, relative to the preindustrial values, surface mean pH drops by 0.42 units, corresponding to a 50.6% reduction in surface C affects the ocean carbon cycle through its impact on the production of CaCO 3 . In the S0 simulation, CaCO 3 production increases with time ( Fig. 5) mainly as a result of increasing ocean temperature that boosts the growth of phytoplankton, which CaCO 3 production depends on according to equation (2). In the simulations with Ω C -dependent / R CaCO POC 3 , CaCO 3 productions are under the influence of both changing temperature and Ω C . As shown in Fig. 5, except for S1, relative to the preindustrial value, there is a general decrease of CaCO 3 production with time, indicating the dominant influence of decreasing Ω C . After around year 2150, there appears to be a recovery of CaCO 3 production as a result of the recovery of surface greatly decreases the production of CaCO 3 , which has great implication for the oceanic uptake of CO 2 as discussed below.
The change in CaCO 3 production has great impact on the ocean alkalinity. In the simulation of S0, an increase in CaCO 3 production (Fig. 5) leads to a decrease in ocean-mean alkalinity (Fig. 4). In the simulations with the Ω C -dependent / R CaCO POC 3 , ocean-mean alkalinity generally increases with time. For example, in the simulation of R3, by year 3500, ocean-mean and surface-mean alkalinity have increased by 25 and 24 μ mol kg −1 respectively (see Supplementary Table S1 online). Meanwhile, in the simulations that include the Ω C -dependent / R CaCO POC 3 , the vertical gradient of alkalinity diminishes due to the reduced CaCO 3 production rate and the consequent weaker CaCO 3 pump (Fig. 6, see Supplementary Fig. S1 online).  Table 1.
Scientific RepoRts | 6:20284 | DOI: 10.1038/srep20284 As a result of modification of the ocean alkalinity, the Ω C -dependent / R CaCO POC 3 affects the ocean's uptake of CO 2 . By year 2100, compared to the S0 simulation, the inclusion of Ω C -dependent / R CaCO POC 3 increases accumulated oceanic CO 2 uptake by 1 PgC (0.1%) to 36 PgC (6.2%), depending on the exact form of CaCO 3 production parameterization (Fig. 3). By year 3500, the increase in accumulated oceanic CO 2 uptake relative to the simulation of S0 ranges from 9 PgC (0.6%) to 285 PgC (19.5%) across different CaCO 3 production parameterization schemes (Fig. 3, see Supplementary Table S1 online). As a consequence, by year 3500, simulated atmospheric CO 2 ranges from 608 to 542 ppm with the inclusion of dependence of / R CaCO POC 3 on Ω C , compared with 612 ppm in the S0 simulation (Fig. 3). Moreover, by year 3500, the inclusion of Ω C -dependent / R CaCO POC 3 acts to reduce the amount of surface warming by 0.04 to 0.6 K relative to the S0 simulation, depending on the / R CaCO POC 3 parameterization scheme used (see Supplementary Table S1 online).  The inclusion of / R CaCO POC 3 dependence on Ω C also has a great influence on ocean acidification (Fig. 4  is mainly a result of increased alkalinity, which dominates the effect of increased DIC on ocean acidification. In the above, we have discussed model-simulated results with the inclusion of CO 2 -induced warming. To test the importance of CO 2 -induced warming on the ocean carbon cycle, we have performed additional simulations that do not include the radiative effect of increasing atmospheric CO 2 . Our simulations show that, by year 3500, in the S0 case, in the absence of CO 2 -induced warming effect, model-simulated cumulative ocean's uptake of CO 2 is 122 PgC greater than that in the simulation with CO 2 -induced warming (Fig. 7, Table S1 online). This comparison shows that in terms of the magnitude of oceanic CO 2 uptake, the effect of CO 2 -calcification feedback could be comparable to or even much larger than that from the feedback of CO 2 -induced warming.

Discussion
Here, we use the UVic model to quantify the effect of potential CO 2 -calcification feedback on the projections of the ocean carbon cycle and climate change. To evaluate the effect of CO 2 -calcification feedback on the ocean carbon cycle and associated uncertainties, we include two different types of parameterization schemes that link CaCO 3 production with saturation state of calcite. In each scheme, a set of different parameters is used. As atmospheric CO 2 increases and the ocean becomes more acidic, the introduction of Ω C -dependent / R CaCO POC 3 decreases the production of CaCO 3 , increasing ocean alkalinity and enhancing the oceanic uptake of atmospheric CO 2 . Therefore, it triggers negative feedbacks on the growth of atmospheric CO 2 and curbs global warming to a certain degree. Under SRES A2 CO 2 emission scenario with zero emission after year 2100 and a total cumulative emission of 2270 PgC, relative to the simulation with fixed CaCO 3 : POC production ratio, by year 2100, the simulations that include CO 2 -calcification feedback decrease modeled atmospheric CO 2 by 0.1 to 7 ppm; by year 3500, the simulations that include CO 2 -calcification feedback decrease modeled atmospheric CO 2 concentration by 4 to 70 ppm. The magnitude of the CO 2 -calcification feedback depends on the calcification-Ω C parameterization scheme used and parameter values used, demonstrating the importance of both the model structure uncertainty and parameter uncertainty of the CO 2 -calcification feedback in regulating the ocean carbon cycle. While the inclusion of the CO 2 -calcification feedback enhances the ocean's uptake of atmospheric CO 2 , it acts to mitigate ocean acidification mainly as a result of increased ocean alkalinity. For example, by year 3500, the inclusion of Ω C -dependent / R CaCO POC 3 increases surface mean pH and − [CO ] 3 2 by 0.8% and 15.0% in R3 relative to S0. Furthermore, our simulations show that the effect of CO 2 -calcification feedback on ocean's uptake of atmospheric CO 2 is comparable to, and in some cases, much larger than the effect from CO 2 -induced warming.
Our study shows a noticeable CO 2 -calcification feedback on atmospheric CO 2 . Different estimates of this feedback are reported in previous modeling studies [21][22][23][24] Table 1.
Scientific RepoRts | 6:20284 | DOI: 10.1038/srep20284 the difference is associated with different representations of CO 2 -calcification feedback. Uncertainties in our model results here reflect uncertainties in modeled parameterization of CO 2 -calcification feedback, which actually reflects uncertainties in our understanding of the calcification response to changing ocean chemistry. The reported response of the CaCO 3 production rate to ocean acidification varies dramatically between experiments using different species of calcifying groups or manipulation methods 20,26 . Therefore, modeling simulations based on different results of experimental studies would result in different estimates of the effect of CO 2 -calcification feedback. More coordinated experimental and observational studies on the CaCO 3 production response to ocean acidification are needed for a more reliable appraisal of the CO 2 -calcification feedback.
In this study, we have investigated the response of calcification to acidification and its feedback to the ocean carbon cycle. Other processes relevant to CaCO 3 cycle that are not included in this study could also have important effect on the ocean carbon cycle. For example, inclusion of the dependence of CaCO 3 dissolution rate on CaCO 3 saturation state would further alter the ocean carbon cycle 19,27 . In addition, the ballast effect, i.e., the link between the fluxes of particulate organic carbon (POC) and particulate inorganic carbon (PIC) to the abyssal ocean 28,29 , is not included in the model. It is possible that reduced CaCO 3 production could result in a decrease in PIC export rate, which consequently lowers POC export rate. This reduced vertical transport of POC would weaken the oceanic carbon pump and decrease the capacity for the global ocean to absorb atmospheric CO 2 , acting as a positive feedback to atmospheric CO 2 22,30 . The feedback from the ballast effect could partly counteract the CO 2 -calcification feedback, which merits further study.
This study demonstrates the potential important effect of CO 2 -calcification feedback on the ocean carbon cycle and atmospheric CO 2 on the timescale from centuries to millennia. Further experimental and modeling studies are needed to acquire a better understanding of the CO 2 -calcification feedback, which is crucial for a reliable projection of future atmospheric CO 2 concentrations and climate change. represents the ratio of CaCO 3 production to the production of particulate organic carbon, and R C N : is the carbon to nitrogen Redfield ratio 31 .

Methods
Parameterization of CaCO 3 production. In the original model, CaCO 3 : POC production ratio / R CaCO POC 3 in equation (2) is fixed at a constant value of 0.018. In this study, two types of parameterization functions of / R CaCO POC 3 that link CaCO 3 production with saturation state of calcite (Ω ) C are introduced into the UVic model.
The first type of parameterization follows the Michaelis-Menten function based on Pinsonneault et al. 37 : denotes the specified maximum value of / R CaCO POC 3 (the CaCO 3 : POC production ratio), and K max is a half-saturation constant 37,38 . Different model versions based on this parameterization are denoted as "series S" (Table 1). In series S, the values of K max are selected to be 0.07, 1.5 and 20, which covers the range of K max values used by Pinsonneault et al. 37  (S1 to S3 and R1 to R3 in Table 1). In addition, we have the original model configuration with fixed / R CaCO POC 3 (S0 in Table 1). For each parameterization, the dependence of CaCO 3 : POC production ratio on Ω C are presented in Fig. 1.
Simulation experiments. All of the model versions mentioned above are integrated for 10,000 model years with fixed preindustrial atmospheric CO 2 concentration of 280 ppm to reach a quasi-equilibrium preindustrial state of global climate and carbon cycle. Using the preindustrial climate state as initial condition for the nominal year of 1800, two sets of 1700-year transient simulations are performed (from year 1800 to 3500). The first set of simulation includes the feedback from CO 2 -induced warming on the ocean carbon cycle, whereas the second set of simulation does not include the radiative effect of increasing CO 2 on global climate. Each set of experiments includes seven simulations, corresponding to the seven model versions listed in Table 1