Revisiting ocean carbon sequestration by direct injection: A global carbon budget perspective

In this study we look beyond the previously studied effects of oceanic CO2 injections on atmospheric and oceanic reservoirs, and also account for carbon cycle and climate feedbacks between the atmosphere and the terrestrial biosphere. Considering these additional feedbacks is important since backfluxes from the terrestrial biosphere to the atmosphere in response to 10 reducing atmospheric CO2 can further offset the targeted reduction. To quantify these dynamics we use an Earth-system model of intermediate complexity to simulate direct injection of CO2 into the deep ocean as a means of emissions mitigation during a high CO2 emission scenario. In three sets of experiments with different injection depths, we simulate a 100-year injection period of a total of 70 GtC and follow global carbon cycle dynamics over another 900 years. Simulated seawater chemistry changes and marine carbon storage effectiveness are similar to previous studies. As expected, by the end of the 15 injection period avoided emissions fall short of the targeted 70 GtC by 16% to 30% as a result of carbon cycle feedbacks and backfluxes in both land and ocean reservoirs. An unexpected feature are effects of the model’s internal variability of deepwater formation in the Southern Ocean, which, in some model runs, causes additional oceanic carbon uptake after injection termination relative to a control run without injection and therefore with slightly different atmospheric CO2 and climate. These results of a model that has very low internal climate variability illustrate that attribution of carbon fluxes and 20 accounting for injected CO2 may be very challenging in the real climate system with its much larger internal variability. Earth Syst. Dynam. Discuss., doi:10.5194/esd-2016-20, 2016 Manuscript under review for journal Earth Syst. Dynam. Published: 28 April 2016 c © Author(s) 2016. CC-BY 3.0 License.

geostrophic wind anomalies, which are a first-order approximation of dynamical feedbacks associated with changing winds in a changing climate (Weaver et al., 2001), are also applied. Simulated CO 2 injections into different ocean regions are based on the Ocean Carbon Cycle Model Intercomparison Project that since these simulations are forced with historical emissions and the RCP 8.5 scenario until year 2020, the model is not in steady state in 2020 and some climatic change occurs. Also, because the injected CO 2 is withdrawn from the atmosphere so that total carbon is conserved, the CM injection runs essentially have negative emissions of 0.7 GtC yr -1 .
To determine how long the injected carbon stays in the ocean, we follow the IPCC [2005] and calculate a fraction 125 retained ( = ! * ! !! * 100), which is the percentage ratio between the total mass of the injected carbon that remains in the ocean (M o , determined using the diagnostic marker tracer) and the total cumulative mass injected into the ocean (M i ) since the start of the injection period (year 2020). This metric accounts for the injected carbon atoms and does not include possible adjustments of fluxes of other carbon in the Earth system.
To assess the global carbon cycle response to the injections, we use another metric, the net fraction stored 130 ( = ΔC !"#$% * ! !! * 100, in %) that measures total carbon reservoir changes. The netFS is defined as the ratio between the absolute change in globally integrated total oceanic carbon (ΔC ocean ), relative to the RCP 8.5 control run, and the total cumulative mass injected into the ocean (M i ) since the start of the injection period. In contrast to FR that counts only the injected carbon atoms, netFS accounts for all potential feedbacks of carbon fluxes into and out of the ocean in response to the injection of CO 2 into the ocean. 135 To investigate if the targeted atmospheric carbon reductions in the WE simulations, differ from what would happen if CO 2 was never emitted (avoided emissions) or first emitted and subsequently removed from the atmosphere, e.g., via technology such as direct air capture (DAC, see section 3.4.1) [Lackner, 2009] and subsequent safe and permanent storage, presumably in geological reservoirs, we performed another simulation where the atmospheric CO 2 concentration was 0.7 By the end of the simulation in year 3020, about 6,000 GtC have been added to the global carbon cycle.
Consequently, atmospheric CO 2 has increased substantially, leading to a total atmospheric carbon content of about 4620 GtC at the end of the simulation (Figs. 2 a, b).
By the end of the extended RCP8.5 control run about 58 % of the emitted CO 2 remains in the atmosphere. The rest 150 of the carbon has been taken up by oceanic and terrestrial reservoirs (Figs. 2 d,f). Oceanic carbon uptake is highest during the first few decades of the simulation, when emissions are highest, and then decreases thereafter (Fig. 2 c). The decrease in net oceanic carbon uptake is particularly caused by a reduction in the ocean buffering capacity [Prentice et al., 2001], leading to a decrease in ocean carbon uptake even under increasing atmospheric CO 2 levels; a response also seen in other model simulations [Zickfeldt et al., 2013]. 155 Simulated terrestrial carbon uptake is initially high as well, but then declines rapidly, with the terrestrial reservoir becoming a source for atmospheric carbon in the year 2139 before leveling off at very little net exchange between the terrestrial reservoir and the atmosphere after about year 2280 (Fig. 2 e). The initial increase in total land carbon uptake is due to the simulated CO 2 fertilization effect on vegetation [Matthews, 2007]. However, as temperatures become higher, terrestrial net primary productivity (NPP) is reduced due to water stress. Moreover soil respiration increases with 160 temperature until it eventually becomes the dominant processes, leading to a net loss of carbon from the terrestrial reservoir to the atmosphere. Projections of future net terrestrial carbon uptake or loss processes are highly uncertain (Carvalhais et al., 2014;Hagerty et al., 2014;van der Sleen et al., 2014;Sun et al., 2014), which is also reflected in the large variability between the CMIP5 (Coupled Model Intercomparison Project Phase 5) model results, with changes in terrestrial carbon budgets ranging from -0.97 to +2.27 GtC yr -1 between 2006 and 2100 [Ahlström et al., 2012]. 165

Changes in seawater chemistry
Here, we compare the WE simulations to the RCP 8.5 control run to assess injection-related seawater chemistry changes. By the final year of the injection period (year 2119), a total of 10 GtC is injected at each site (Fig. 1). The respective increases in DIC and reductions in pH depend on how quickly the injected carbon is transported away from the injection sites by local ocean currents and mixing [see Orr, 2004]. Our model-predicted changes in DIC and pH at the 170 injection sites (relative to the control run) are within the range of Orr [2004] (Table S1-2).
Simulated ocean surface pCO 2 is lower in the CO 2 injection runs because of lower atmospheric CO 2 levels and the related decrease in air-sea carbon fluxes, which results in lower surface DIC concentrations and a slightly higher surface pH (by 0.008 to 0.01 units compared to the control run).

Fractions retained 175
Here, we assess to which extent the simulated CO 2 injections are effective in keeping the injected carbon out of the atmosphere. This is described by the fractions retained (FR). The global FR of our CM and WE simulations (Table 1) are within the full range of the GOSAC-OCMIP results [Orr et al., 2001;Orr, 2004]. The simulated FR (Table 1) increases with the depth of injection because it generally takes longer for deeper waters to again come into contact with the atmosphere, something also shown in previous studies [e.g., Caldeira et al. 2001;Orr et al., 2001;Orr, 2004;Jain and Cao, 2005]. 180 By comparing the WE and CM simulations at all depths, we can determine how climate change affects FR. As in previous studies, our results show that FR is enhanced by climate change [Jain and Cao, 2005;Ridgwell et al., 2011]. In the WE simulations values of FR are always higher than in the CM runs (Table 1). For I-800 and I-1500, the FR increase due to climate change is largest in the Pacific, whereas for I-3000, Atlantic sites show the highest FR increase due to a larger ocean response to climate change (Table 1). However, in all simulations more of the injected carbon is retained in the Pacific 185 compared to injections in other ocean basins.
We also assess whether the enhanced FR in our WE simulations are affected by changes in the Atlantic Meridional Overturning Circulation (AMOC). Relative to preindustrial, which has a maximum AMOC intensity of 15.98 Sv, we find AMOC decreases by 8%, 29%, 40%, 34% in the years 2020, 2120, 2520, 3020, respectively in the WE simulations. AMOC in the CM simulations, relative to preindustrial, shows smaller decreases of about 7.6%, 21%, 8.6%, 8.6% in the years 2020, 190 2120, 2520, 3020, respectively. These differences partially explain why FR is enhanced in the WE simulations, since a reduced AMOC slows the transport of deep water masses and prolongs the time until they again come into contact with the atmosphere. As in other climate change studies [e.g., Doney, 2010;Bopp et al., 2013], we also find an increase in ocean stratification (not shown) in all respective basins in our WE runs, relative to the CM runs, which has also led to reduced points where FR is high, points to carbon cycle and climate feedbacks, which are directly related to changes in atmospheric 220 CO 2 concentrations (i.e. ocean-atmosphere pCO 2 differences and CO 2 fertilization effects) and changes in temperature. Other studies have also shown that these feedbacks occur and affect the size of the global carbon reservoirs (Arora et al., 2013).
The curve progression of the atmospheric reduction in the DAC run is very similar for I-1500 and I-3000, which is due to the occurrence of most of the same carbon cycle and climate feedback mechanisms. However, due to no carbon injections in the DAC run, the atmospheric reduction is higher as soon as injected carbon starts leaking in the WE simulations as presented in 225 Figure 3. In the UVic model (version 2.9), the atmospheric carbon reduction of the DAC run (Fig. 3) can also be referred to as the true atmospheric carbon reduction target. Depending on depth of injection, this implies further that direct injection of CO 2 would not be able be 100% efficient and provide 100% of the true atmospheric reduction target on decadal to centennial timescales (Fig. 3). Due to the occurrence of an ocean deep convection event in the DAC run after the year 2120 (see section

3.4.2), we cannot easily compare the DAC run to the WE simulations after the injection period. 230
While ocean feedbacks in response to CO 2 injection and reduced atmospheric CO 2 levels have been discussed extensively in previous studies [e.g. Orr 2004;IPCC, 2005, Ridgewell et al., 2011, we here additionally consider land feedbacks with the purpose of accounting for the entire Earth system's response to potential marine CO 2 injections.
By the last year of the injection period (year 2119), I-800 shows the highest divergence from GIC (Fig. 4 c) with an atmospheric carbon reduction of only 48 GtC, which is 22 GtC less than targeted. Since from the dye tracer it is known that 235 25% (i.e. 17.8 GtC) of the injected CO 2 has leaked to the atmosphere (Table 1), C-cycle and temperature feedbacks must be responsible for the other 4.2 GtC that remained in the atmosphere. This remaining amount can partially be explained by the reduced pCO 2 difference between the atmosphere and the ocean, which leads to a smaller carbon flux into the ocean (Fig. 4   d). Plus, relative to the control run, there is a lower atmosphere-to-land carbon flux until approximately the year 2075 (Fig. 4 f), leading to 1.2 GtC less total land carbon by the end of the injections (Fig. 4 e). After the injections start (year 2020), both 240 NPP and soil respiration are lower in I-800 than in the control run, leading to a maximum reduction in land carbon of about 4.2 GtC in year 2075 (Fig. 4 e). Thereafter, total land carbon in I-800 increases. By the end of the injections in year 2120, the terrestrial carbon pools have taken up 1.2 GtC less than the control run without CO 2 injection.