Impact of marine carbon removal on atmospheric CO2

A computer simulation of Earth’s climate is used to study if marine carbon removal will lead to a reduced atmospheric carbon dioxide concentration, and if there are potential secondary impacts on marine life and chemistry. We find that for stationary carbon removal plants the ocean cannot supply sufficient carbon rich water to allow a meaningful reduction of atmospheric CO 2 . This also means that outside the location of carbon removal there is no noticeable impact on plankton concentrations. It can be speculated that putting carbon removal plants on ships would lead to a significant increase in removal efficiency, although the engineering and energy aspects of this approach would need to be investigated.


Introduction
The urgent challenge of stabilizing Earth's climate system in the face of global warming necessitates not only the reduction of greenhouse gas (GHG) emissions but also the active carbon dioxide removal (CDR) from the atmosphere (IPCC 2023).According to the IPCC 6th report, CDR is indispensable for counterbalancing 'hard-to-abate' residual emissions (Buck et al 2023), which are particularly challenging to eliminate in sectors such as agriculture, aviation, shipping, and industrial processes.There are significant variations in CDR methods, encompassing their development stages, removal mechanisms, carbon storage duration, and overall impacts (Mathesius et al 2015, Bach et al 2023).This spectrum ranges from emerging technologies like ocean alkalinisation (Eisaman et al 2023), which are still in their nascent stages, to more established practices like reforestation (Melnikova et al 2023).Understanding the benefits, risks, and trade-offs of each method is crucial for assessing the feasibility and effectiveness of different CDR approaches in the context of global climate goals.
This study focuses on atmospheric implications of marine carbon removal (mCR) approach in light of the recent developments in electrochemical processes (Digdaya et al 2020, Kim et al 2023) for removing carbon from the ocean.For instance, Digdaya et al (2020) suggested to acidify ocean water through electrodialysis.This increases the partial pressure of CO 2 in water and leads to release of CO 2 .Such CO 2 can then be captured; the water can be re-alkalinized, which then leads to an renewed draw-down of CO 2 from the atmosphere.This process works in laboratory settings, and could be used energy efficient in large scales in desalination plants.However, current atmospheric CO 2 concentration growth rate equals approximately 4.6 Gt C yr −1 (Friedlingstein et al 2023), roughly the equivalent of 2.18 ppm yr −1 of CO 2 , so if this mCR technique is to make any contribution to solving the global warming problem, it has to remove more than the current CO 2 growth rate, i.e. at least 10 Gt C yr −1 .
The goal of the present study is to investigate with a numerical simulation the possible regional or global implication of such an undertaking.The numerical simulations are done with a state-of-the-art climate model, one that also contributes to the IPCC reports, and the process described above is represented by removing at every model time-step, from one grid cell, a fraction of its dissolved inorganic carbon.The model then computes everywhere the new state of the planet.The main shortcoming of this experimental design is that an individual model grid cell has an area of some 10 000 square kilometers, whereas the direct source area of a desalination part would be to the order of a square kilometer.However, as we shall see below, this mostly means that our results are an upper bound of the effect of stationary decarbonators.

Experimental design
We use the coarse-resolution version of the Community Earth System Model (CESM1 Shields et al 2012, Hurrell et al 2013).It consists of models of the ocean, the atmosphere, sea ice, and land; they are connected through a coupler, which passes and interpolates fluxes between the various models.The ocean resolution varies smoothly with location, with resolutions of some 20 km around Greenland, 100 km in the Southern Ocean, and 400 km in the subtropical North Pacific.There are 60 vertical layers with nonuniform thickness, ranging from 10 m at the surface to 250 m at the bottom.The ocean model is run with the Gent and McWilliams (1990) representation for mesoscale mixing, and uses a stratification-dependent thickness and isopycnal diffusivity (Ferreira et al 2005).The atmospheric model uses a T31 spectral truncation in the horizontal (3.75 • resolution) with 26 vertical layers.In all the present simulations, changing CO 2 will change radiative forcing.The biogeochemistry component (Moore et al 2013, Lindsay et al 2014) is coupled to the climate system and actively exchanges carbon between ocean, atmosphere, and land.It includes diatoms, small phytoplankton, and diazotrophs, with phytoplankton growth controlled by temperature, light, and available nutrients (nitrate, phosphate, silicate, and iron).CESM1 has a fixed lysocline, i.e. the depth at which calcium carbonate remineralizes is fixed.Thus, the present set-up does not allow for carbonate compensation, a feedback between ocean acidity and calcium carbonate sedimentation that is speculated to affect CO 2 on timescales of millennia or longer (e.g.Archer et al 2000, Brovkin et al 2007).The land component prognostically computes leaf and stem area indices and vegetation height using a prescribed spatial distribution of plant functional types (Lawrence et al 2011).
The initial distribution of all ocean tracers are based on a 2000 year long pre-industrial integration of the 1 degree version of CESM1 (Lindsay et al 2014).We then integrated the present configuration another 5100 years on our own HPC to ensure equilibrated simulations.Since small drifts cannot be avoided, our results will show the difference between the last 100 years of this 5100 year integration, and the mCR simulation, branched off at year 5001 of this simulation.The present-day coarse-resolution climate of CESM1 is in many aspects similar (i.e. it has the same biases) to its 10 times more expensive one degree resolution version (Shields et al 2012); using 160 CPU cores yields a simulation rate of 90 years per day.
A meaningful numerical experiment requires the numerical representation of the laws of nature, boundary conditions, forcing, and initial conditions.We are confident about our choices of the first three since CESM1 is a well tested ESM and much is known about orbital forcing, orography and bathymetry.
Much less is known about carbon pools and fluxes, but a recent study by Jochum et al (2022) shows that the present model represents well not only the preindustrial (PI) carbon pools, but also the atmospheric and terrestrial carbon pools during the last glacial maximum.
The global mean marine carbon concentration is 2300 mmol m −3 (Williams and Follows 2011), so that an ocean volume of 100 000 km 2 (≈ 320 × 320 km, standard grid cell area) and a depth of 10 m contains approximately 0.03 GtC.Removing 10% of this every hour (the model time-step) would lead to an annual mCR of 26 Gt, the equivalent of 12 ppm atmospheric CO 2 .Thus, removing 10% of marine carbon every hour from one model grid cell is a removal rate that would lead to a meaningful reduction of atmospheric CO 2 .Other implementations of mCR are possible as we will see below they are unlikely to lead to significantly different results.

Results
Figure 1 shows for different scenarios the impact of mCR on oceanic dissolved inorganic carbon (DIC, the sum of carbonate, bicarbonate and CO 2 ) and atmospheric CO 2 .The reference case is based on Earth's condition in 1850 and is described above (PI) and shows a stable CO 2 concentration with some minor natural fluctuations.In the first experiment we removed 10% DIC per hour from a surface cell in the middle of the ocean.The expectation is a significant reduction of DIC and CO 2 -if sufficient carbonrich water can be supplied to the carbon sink, and the atmosphere 'sees' the low-carbon patch.The green line shows that this does obviously not happen, the average removal rate is only 0.5 Gt yr −1 , in fact similar to the 0.2 Gt yr −1 removed by only a tenth of the removal rate (orange line).This suggests that the resupply of fresh carbon-rich water is a bottleneck.And indeed, if we spread the removal rate across the whole ocean (reducing it appropriately to account for the increased area) we do achieve a much higher rate of 2.4 Gt yr −1 and a CO 2 reduction of 90 ppm (purple line).
We chose the original removal site to be in the middle of the tropical Pacific, away from strong currents (figure 2(a)).The sea-surface expression of the mCR is rather small, to the order of several thousand kilometers, something that is expected from typical ocean velocties on the order of cm s −1 or hundreds of kilometers per year.Choosing removal sites that are embedded in strong currents like the Gulfstream or the South Equatorial Current (chosen here) will lead to a bigger surface expression and more efficient mCR, but not significantly so (figure 1, cyan line; figure 2(b)).One could remove carbon at depth (brown line), which is more efficient because fresh  water can be mixed into the grid-cell from above as well.But then, of course, there is no impact on CO 2 .
Lastly, we examined the variations in chlorophyll levels, an essential indicator of primary production in marine ecosystems (Williams and Follows 2011).Chlorophyll, primarily found in phytoplankton, is critical for photosynthesis and signifies the health of the marine food web.Our modeling results suggest no significant changes beyond natural fluctuations (figures 2(c) and (d)).This finding aligns with the understanding that in vast ocean areas, factors like iron, phosphate, or other nutrients, rather than carbon, predominantly limit marine life (Williams and Follows 2011).However, it is important to acknowledge that to accurately estimate the long-term impact on marine ecosystems, a longer period of model integration beyond the current 100-year is necessary for a more comprehensive understanding.

Discussion
The fundamental problem with removal strategies based on desalination plants or any other stationary facility appears to be twofold: the ocean cannot supply enough carbon rich water, and the atmosphere only sees a small footprint of reduced surface DIC.A look at the tracer and continuity equations will illustrate the relevant processes and scales: C t + ⃗ u • ∇C = k∇ 2 C + sources/sinks (a tracer with concentration C is modified by advection, diffusion, and by sources and/or sinks) In the ocean the horizontal advection of properties is typically on the same order as their turbulent diffusion: for velocities of 1 cm s −1 , turbulent diffusivities of 1000 m 2 s −1 and length scales of 100 km.However, in the surface mixed layer, vertical diffusion is much stronger than horizontal diffusion: for vertical scales of 10 m, and vertical turbulent diffusivities of 0.1 m 2 s −1 (for oceanic scales and scale analysis, see Pedlosky (1996)).Thus, the mCR is limited by the amount of DIC that can be mixed into the surface waters from below, and the atmospheric CO 2 reduction is furthermore limited by the speed by which the area of the low-DIC waters can be increased.In other words, it means that over some 30 years only a patch of a million square kilometer will show increased uptake of CO 2 (Time ≈ L 2 / k h , see the scaling analysis above and figure 2), less than a percent of the ocean surface and not enough to allow large drawdowns.On the other hand, deploying mCR facilities on mobile units, such as ships, could potentially resolve these issues altogether.This approach would enable strategic and dynamic positioning of the units, allowing them to adapt to varying carbon concentrations and environmental conditions.However, such shift necessitates a reevaluation of the energy considerations originally associated with stationary removal facilities and logistical management of by-products.For instance, electrolytic and electrodialytic processes of generating alkalinity from brine, by-products include gases and aqueous acids (Eisaman et al 2023).Managing these by-products, particularly in large quantities, poses environmental and safety risks.The economic feasibility of scaling these processes is uncertain.While the chloridemediated system shows promise in terms of cost (ranging from $50-$100 per ton CO 2 ), this is based on small-scale models and does not account for additional costs related to large-scale deployment (Kim et al 2023).Energy consumption is also a concern, especially for processes that require high electrical potentials.
The present results are arrived at by using simple scaling arguments and a state-of-the-art Earth system model.We see the strongest weakness in the coarse resolution of the ESM, which is expected to exacerbate supply bottlenecks at higher resolutions, because the supply and the footprint visible to the atmosphere both scale with the square of the resolution.The fact that the carbon enters the removal volume through it bottom surface area means that, at least in a non-turbulence resolving climate model, increasing the resolution reduces the efficiency of mCR.Of course, this may not necessarily be true on the scale of desalination plant.However, there we find that, independent of the efficiency of turbulence or advection, mCR is limited by the number and capacity of current desalination plants: The current annual mCR potential can be estimated by considering the mean marine carbon concentration (approximately 2300 mmol m −3 ) and the processing capacity of approximately 16 000 desalination plants worldwide (Jones et al 2019), which handle around 237 million m 3 d −1 of feed-water.Based on these numbers, the estimated maximal mCR is 0.002 Gt C yr −1 .
Additionally, we recognize the necessity of further exploring the long-term impacts of mCR on marine chemistry and ecosystems.Our current modeling results, with its 100 year integration length, may not fully encapsulate such complex and longterm dynamics.The challenge is compounded by the non-linear nature of atmospheric and oceanic physics and the intricate biogeochemistry of Earth's systems.The fine-scale interactions and feedback mechanisms add layers of complexity, hindering accurate predictions.
Looking forward, we emphasize the importance of enhancing model resolution and extending integration length.By addressing these areas, we aim to navigate the current limitations and challenges in mCR research, paving the way for more informed and effective strategies in the future.

Figure 1 .
Figure 1.The impact of mCR on atmospheric CO2 (top) and total ocean carbon (bottom).The individual colors indicate different removal strengths and locations (see also next figure).

Figure 2 .
Figure 2. (a) and (b) Differences (mCR simulations-control) of sea surface concentrations of DIC.Apart from a natural background variability one can identify the sites of the mCR and its farfield effect by their dark patches.Note the larger footprint (b) when the sites are chosen in a strong current.(c) and (d) Mean surface chlorophyll concentration in the control simulation and its difference in the case of the most efficient removal (in the South Equatorial Current), respectively.No changes beyond the natural variability can be found.