Measuring large-scale mCDR using a novel Atmospheric CO₂ Perimeter Flux (ACPF) method

1 Introduction

Achieving Paris Agreement net-zero CO₂ emissions by 2050, consistent with IPCC pathways limiting warming to 1.5 °C (SSP1-1.9, SSP1-2.6), requires CO₂ removal (CDR) at 7–9 Gt CO₂ per year by mid-century (IPCC, 2022; Smith et al., 2024). The current global CDR rate is approximately 2 Gt CO₂ yr⁻¹, dominated by land-based approaches whose potential is constrained to 3–5 Gt CO₂ yr⁻¹ by finite land availability, soil carbon saturation and permanence insufficient to meet the mid-century target (Smith et al., 2024; Fuss et al., 2018; Deprez et al., 2024).

Marine CDR naturally complements land-based efforts. The ocean already absorbs 10.6 Gt of CO₂ annually and stores approximately 38,000 Gt C as dissolved inorganic carbon, roughly 44 times the atmospheric pool, providing significant leverage for deliberate mCDR (Ciais et al., 2013; Friedlingstein et al., 2025). The ocean’s capacity to absorb additional CO₂ through carbonate equilibria at the planetary scale makes it the largest reservoir available to address the deficit that land-based approaches alone cannot meet. Actively investigated mCDR approaches, including ocean iron fertilization, ocean alkalinity enhancement, macroalgae cultivation and sinking, electrochemical seawater CO₂ removal and coastal ecosystem restoration, collectively exhibit theoretical gigaton-scale potential (Boyd et al., 2007; Renforth and Henderson, 2017; NASEM, 2022; Oschlies et al., 2025; Ward et al., 2025; Hooper et al., 2025). Realizing it requires a scalable, independent MRV framework at the megaton-to-gigaton scale. Whether existing methodologies can meet that requirement remains unresolved [Fennel, 2026; Oschlies et al., 2025; European Marine Board (EMB), 2025].

Current mCDR MRV relies on three ocean-domain approaches, each constrained at that scale. First, surface-ocean pCO₂ monitoring estimates air–sea flux using an empirical gas-transfer parameterization that carries at least a 25% uncertainty (Wanninkhof et al., 2009; Doney et al., 2024; Ward et al., 2025). Second, dissolved inorganic carbon and alkalinity budgeting, the most direct accounting variables, are analytically demanding and prone to drift and biofouling, and even dense autonomous arrays can underestimate net carbonate-system change by 35–55% in realistic conditions (Ward et al., 2025; Bach et al., 2024). Biological pump here encompasses both biological and chemical processes, tracked by sediment traps, thorium-234 profiling, and biogeochemical Argo floats, which monitor particulate organic carbon export to depth, but not net atmospheric removal, and trap efficiency and lateral advection remain poorly constrained (Boyd et al., 2019; Buesseler et al., 2024). All three require an estimate of what the ocean sink would have been without the intervention. Because ocean carbon dynamics are nonlinear, that estimate is model-derived, and at megaton scales, its uncertainty can approach the removal signal itself (Fennel, 2026; Ito and Reinhard, 2025; Oschlies et al., 2025).

Given these limitations, the atmosphere offers a more directly integrated measurement domain. The marine boundary layer (MBL), the lowest 0.5–2 km over the ocean, integrates net surface CO₂ exchange across the underlying region into a spatially coherent field, well mixed vertically on timescales of 10–30 min by wind-driven turbulence (Stull, 1988). Exploiting this well-mixed property of the MBL, we present the Atmospheric CO₂ Perimeter Flux (ACPF) method, an MRV framework for marine CDR. ACPF employs CO₂ sensor ASVs with wind-vector measurements around the perimeter of a bounded ocean intervention region and quantifies the net CO₂ mass flux entering and leaving the marine boundary layer across that perimeter and across the capping inversion. We derive a conversion factor that enables comparison of ASV-measured depletion with that from satellite XCO₂ data. Establishing that a drawdown is additional requires a baseline that sets out the design by which it is obtained from the same or similar instruments, over the same region.

2 MRV gap: the Atmospheric CO₂ Perimeter Flux (ACPF) method

2.1 Measurement principle and conceptual framework

No previous MRV methodology verifies ocean-based removal of atmospheric carbon at megaton-to-gigaton scales. ACPF shifts the measurement domain from ocean to atmosphere, quantifying net surface atmospheric CO₂ drawdown over a bounded ocean region rather than tracking carbon through the ocean’s biological pump. Any net carbon removal proportionately reduces the CO₂ mixing ratio within the overlying MBL. The differential architecture compares upwind ambient readings with later readings of the same downwind, depleted sampled air mass. This removes synoptic-scale CO₂ variations from the difference. A network of self-stationing CO₂ sensors surrounds the intervention area, with upwind sensors recording ambient CO₂ and downwind sensors recording depleted air leaving the area (Figure 1). The control volume is a slab of air within the MBL, capped by the trade-wind inversion and bounded laterally by the sensor ring. As an air mass enters one side, transits the intervention region, and exits the other, the method quantifies the resulting change in CO₂ concentration across those boundaries. This measurement architecture is particularly well-suited to warm trade-wind regions. Virtually all ocean nitrogen fixation occurs in warm trade-wind regions with sufficient iron (Stal, 2009), likely allowing multi-gigaton CO₂ removal as may have occurred following the Mt. Pinatubo eruption. Steady trade-wind flow keeps direction and speed generally constant across the perimeter, so relatively few sensors are required. The quantity constrained is the net surface CO₂ flux integrated over the enclosed region during the transit of the sampled air mass, not the intervention-induced removal itself. We therefore distinguish three quantities. F_net is the total net air–sea CO₂ flux over the enclosed region. F_nat is the flux that would have occurred over the same region during the same period in the absence of intervention. F_add is the additional removal, F_add = F_net − F_nat. ACPF constrains F_net directly. Recovering F_add requires attribution, and attribution requires a baseline. The claim made here is therefore narrower than the claim of model independence. ACPF replaces a counterfactual on carbon export through the ocean interior, which can only be simulated, with one on surface CO₂ flux, measurable by the same array in the same units over the same region.

2.2 Baseline and attribution design

Additionality is established by experimental design rather than measurement alone and requires three observational elements.

First, a regional baseline. The array is deployed over the intervention region for a full seasonal cycle before intervention. This establishes the mean and variance of F_nat under the instrument, geometry, and units that are later applied to the intervention. While individual eddies typically last several months (Chelton et al., 2011), similar eddies in a region can be tested, providing appropriate statistics. Low costs of autonomous surface vehicles permit baseline durations that are impractical for ship campaigns.

Second, a concurrent untreated reference region. A second anticyclonic eddy matched in sea surface temperature, mixed-layer depth, surface pCO₂ and chlorophyll is instrumented in parallel, and the additional removal is given by Equation 1:

F add = ( F T , post − F R , post ) − ( F T , pre − F R , pre ) (1)

for treated region T and reference region R. The reference term removes seasonal stratification change, regional advective gradients and shared synoptic variability; the pre-period difference removes the persistent offset between the two.

Third, measurement continued through relaxation. Bloom senescence and remineralization return CO₂ to the surface over weeks to months, so the post-intervention window is the recovery limb, not an unperturbed reference. It is included within the reported integral, so that F_add is quantified over the full perturbation and relaxation cycle rather than at peak drawdown, with the measurement period set by the observed flattening of the cumulative flux curve rather than by a prior bloom model.

An intervention is quantifiable when its areal removal rate exceeds three times the standard deviation of the regional natural flux measured over the preceding baseline season. Interventions below this threshold are not verifiable by perimeter flux alone, whatever the instrument precision. Perimeter flux is agnostic to the origin of the CO₂ crossing it, so non-marine contributions are excluded by co-measurement rather than by the mass balance.

2.3 Marine boundary layer: physical basis

The MBL is maintained in a well-mixed state by wind-driven turbulence, with an overlying temperature inversion that limits exchange with the free troposphere. Turbulent eddies recycle air between the surface and the inversion on timescales of 10–30 min (Stull, 1988; Wood, 2012). A downwind-edge measurement represents the vertically averaged MBL mixing ratio over the past 10–30 min upwind.

MBL height is typically 1.4 km in trade-wind conditions, comprising a 1.3 km base (Bellon and Stevens, 2012; Nuijens and Stevens, 2012; Carrillo et al., 2016) with entrainment deepening (Wood and Bretherton, 2004; Wood, 2012). The corresponding inversion pressure is 866 mbar in the ICAO Standard Atmosphere. Height varies substantially by region and time of day and is 0.85 km over a planned pilot region off the western coast of Mexico during the OCO-2/3 observations. Forecasts and measurements are available from the Global Forecast System. These must be verified periodically by aircraft or radiosonde. The free troposphere above the inversion retains air near its original CO₂ content, entrained into the MBL at the rate.where w_e is the entrainment velocity (m s⁻¹) and H_MBL is MBL height (m). For subtropical and trade-wind regions, Wood and Bretherton (2004) report w_e ≈ 4 mm s⁻¹, giving K_entr ≈ 1.0% per hour. Over the approximately five-hour transit of a parcel across a 100 km region at 20 km h⁻¹, about 5% of the MBL column is exchanged with the undepleted free troposphere. Reported entrainment velocities vary by about a factor of two, so the correction carries an uncertainty near ±50% of its magnitude. This term is carried throughout as an explicit bounded correction of that magnitude and is not treated as negligible.

K entr = w e / H MBL (2)

2.4 Control-volume formulation and uncertainty budget

The governing relation is a conservation statement for the control volume bounded laterally by the sensor perimeter, below by the sea surface and above by the MBL inversion. Taking C as the CO₂ mass within that volume, conservation is expressed by Equation 3:where F_adv is net horizontal advective flux, F_entr is vertical exchange at the MBL top, and F_surf is the air–sea flux to be recovered. Storage variation arises from synoptic-scale variations in ambient CO₂. These are subtracted out by comparing delayed measurements of the same airmass or by long-term averaging downwind. CO₂ variations from the air-sea flux are integrated as the air mass blows across the downwind sensors. Accurate measurements must sum CO₂ differences over a period at least as long as the transit time of an airmass across the intervention region.

∂ C / ∂ t = F adv + F entr + F surf (3)

The advective term is the product of wind speed, perpendicular wind component, MBL height, sensor spacing and air density, applied to the CO₂ mole-fraction difference and summed around the perimeter to give the removal rate in tonnes per hour. The calculation is carried out in molar units, the molar-mass ratio (1.52) applied within the flux expression, so no separate mass-fraction unit is required. Each downwind reading is differenced against the corresponding upwind reading, reducing the influence of ambient CO₂ variability.

2.5 Surface-to-column conversion: the R-factor

Surface ASVs measure the local depletion in mixing ratio (ΔCO₂, surf, ppmv). Satellites, including OCO-2/3, GOSAT and CO₂M, report the column-averaged dry-air mole fraction XCO₂ (Kuze et al., 2009; Eldering et al., 2017; Suto et al., 2021; Sierk et al., 2021), and the R-factor relates the twoTreating the atmosphere as two layers, the well-mixed MBL and a free troposphere retaining background CO₂, and taking the depletion as uniform within the MBL up to typically 866 mbar, the column anomaly is the surface anomaly diluted by the ratio of MBL column mass to total column mass:

R = ΔCO ₂ , surf / Δ XCO ₂ .

R = P surface / ( P surface − P MBL , top ) = 1013 / ( 1013 − 866 ) ≈ 6.9 (4)

The MBL column holds approximately 14.5% of atmospheric mass, so a surface depletion of 6.9 ppmv gives a column anomaly near 1 ppm XCO₂. R depends on the inversion height and is evaluated for each site rather than adopted as a constant, rising to near 10 over the shallower 0.85 km inversion of the planned pilot region. The depth dependence governs the surface depletion that an array must resolve, but not the column anomaly, since for a fixed removal, the swept air mass and R scale with layer depth in opposite senses and cancel. Equation 4 is a first-order slab estimate rather than a universal physical conversion. It assumes uniform depletion throughout the MBL and corrects the surface CO₂ values to dry-air mole fraction, a correction that is typically less than 5%. R should be used as an order-of-magnitude scaling until calibration measurements are available.

Satellite measurements upwind and downwind yield ambient and depleted XCO₂, whose difference cross-checks the ASV mass balance. For OCO-2 and OCO-3, single-sounding uncertainties after quality filtering and bias correction are approximately 1 ppm over land and 0.7 ppm over ocean glint (Taylor et al., 2023), consistent with the single-sounding random errors of 0.5–1 ppm reported for OCO-2 (Eldering et al., 2017), and CO₂M targets below 0.7 ppm (Sierk et al., 2021). Single-sounding precision is a necessary but not sufficient condition for detection. Spatial sampling, cloud screening, systematic retrieval bias, and background gradients also govern whether a coherent anomaly is resolvable, so thresholds should be set based on aggregated soundings rather than single-sounding precision.

3 Signal magnitude: a historical natural analogue

Atmospheric CO₂ growth slowed following the 1991 Mt. Pinatubo eruption, with concentrations stabilizing and then remaining approximately 2.25 ppm below levels projected from emissions through today (Fiekowsky and Burnham, 2025). Fay et al. (2023) isolate the ocean response to the volcanic forcing in a large-ensemble Earth system model and report an air–sea CO₂ flux anomaly of −0.29 ± 0.14 Pg C yr⁻¹ in 1992 and an increase of 0.53 Pg C in upper-ocean dissolved inorganic carbon by 1993. They attribute that part of the carbon anomaly principally to changes in solubility and eruption-generated El Niño–Southern Oscillation variability. We use the Pinatubo episode solely to establish the order of magnitude of the surface signal that an ACPF array must resolve. We calculate an illustrative removal of 18 Gt CO₂ over 20 months across a region 600 km wide orthogonal to a mean wind of 18 km h⁻¹ (Haeckel et al., 2001), with an effective boundary-layer height of 2.0 km, which includes entrainment mixing. This sweeps air with a mass of 3.7 × 10¹⁴ t, implying an average, not instantaneous, mole-fraction depletion near 32 ppmv over the period. Because R depends on layer depth, it is evaluated at the 2.0 km height used here (Huang and Mao, 2015). In the ICAO Standard Atmosphere, 2.0 km corresponds to 795 mbar, yielding R = 1013/(1013 − 795) ≈ 4.6 and a column anomaly of near 6.9 ppm, well above the sensor detection threshold of 0.5–0.7 ppm. An intervention an order of magnitude smaller would still yield a surface depletion near 3 ppmv, the conclusion relevant to ACPF design.

4 Discussion

4.1 Applicability across mCDR approaches

ACPF applies to any mCDR intervention producing net atmospheric CO₂ removal over a bounded, persistent region, subject to two conditions. The surface depletion must exceed the sensor threshold of roughly 0.5–0.7 ppm, and the removal rate must exceed three times the standard deviation of the regional natural flux measured over the preceding baseline season. For high-rate approaches such as ocean iron fertilization within mesoscale eddies, both conditions are met comfortably at the scales considered here. For lower-rate approaches such as open-ocean alkalinity enhancement, detectability improves with increased sensor density, deployment in semi-enclosed seas, or larger intervention areas.

4.2 Relationship to ocean-domain MRV

ACPF constrains a different quantity in a different domain from ocean-based MRV and can be complementary to it rather than a replacement. Its distinguishing features are a mass balance that requires no biological-pump model, an independent satellite cross-check via the R-factor, and continuous monitoring without ship surveys. It carries its own uncertainties (Table 1), and the combined 25–30% is not by itself smaller than the 25% inherent in gas-transfer parameterization or the 35–55% reported for dense carbonate arrays. The stronger case is joint deployment: atmospheric perimeter flux constraining net surface exchange, while ocean-domain observations detail the fate of carbon over time. Atmospheric flux measurement has a terrestrial precedent in eddy covariance (Baldocchi, 2003, 2014), which ACPF extends to the ocean by resolving horizontal advective flux across a perimeter rather than vertical flux at a tower.

Table 1. Classification of control-volume terms and their contributions to the ACPF uncertainty budget.

4.3 Limitations and the research agenda

This perspective sets out a measurement concept and the design required to test it, rather than a validated MRV method. Because several elements remain open, specific operational requirements are still being refined. For example, while wind fields for air–sea flux work is routinely taken from remote sensing or numerical weather models, with platform anemometers serving to validate rather than replace them (Wanninkhof et al., 2019). If the reanalysis and scatterometer winds prove adequate over the perimeter, the ASV specification and hence array cost can be reduced accordingly, though continuous coverage over the transit time remains a requirement that intermittent satellite passes do not meet.

A second open element is long-term durability. ACPF constrains atmospheric exchange during the measurement period but not afterwards. Predicting when sequestered carbon will return to the atmosphere would ordinarily draw on historical data such as analyses of CO₂-removing and non-removing eruptions, from the ice-age cycle, and from OIF tests showing both negligible and high-removal results, but these analyses are still in development. The durability of ice-age carbon removal is estimated at over 1,000 years and preliminary evidence from large eruptions indicates two eruption types: carbon-removing events, with decades-to-centuries durability, and non-carbon-removing, with no detectable atmospheric durability. Pending publication of historical data analyses, depth profiling of DIC, DOC and POC through the intervention volume can predict durability.

A third area of ongoing rather than demonstrated field performance. Furthermore, the surface-to-column factor is a slab estimate that does not account for the vertical sensitivity of the satellite retrieval. Because this sensitivity is lowest in the lowest kilometre, where the ACPF signal lies. The retrieved column anomaly may therefore be smaller than the slab estimate implies, and R requires calibration against measured profiles. Additionally, the attribution design has not been executed, and the magnitude and variability of the natural flux term in a candidate region are unmeasured.

These open elements are the reason for this Perspective. Historically, the verification of ocean-based removal has been constrained by the cost and impracticality of establishing counterfactuals in the ocean interior, where measurements are dispersed in three dimensions and cannot be readily sustained. By shifting the focus to the atmospheric boundary layer, this framework separates the measurement of climate-relevant atmospheric CO₂ removal from the underlying ocean science that explains it. This provides a domain where the counterfactual is reduced to a surface-flux baseline, which can be measured by the same low-cost autonomous array performing the intervention measurement. The degree to which a CO₂ reduction is sustained remains an empirical question, and prototype deployments currently underway are designed to answer it. We present this framework now so the community can examine the design before field results are reported, rather than after.

5 Conclusion

Ocean-domain MRV depends on model-simulated counterfactual baselines whose uncertainties can approach the removal signal at megaton scales. ACPF addresses this by shifting the measurement domain to the overlying atmosphere, which primarily influences Earth’s climate. By integrating the net CO₂ mass flux across a sensor perimeter that encloses the well-mixed MBL, ACPF constrains net surface exchange without a biological pump model. Its baseline is obtained by direct measurement over the same region, with the same array and in the same units, before and alongside the intervention. The R-factor provides a first-order scaling between surface and column anomalies, enabling a satellite cross-check. A historical analogue indicates that gigaton-scale removal generates surface depletions of order 32 ppmv, well above current detection thresholds. At a cost below USD 1 million per deployment, ACPF offers a scalable route to the MRV infrastructure that gigaton-scale marine CDR will require. The priorities for further work are a paired treatment-and-reference field trial, calibration of the surface-to-column conversion against measured vertical profiles and correlation of ACPF removal estimates with satellite XCO₂ retrievals and particulate organic carbon measurements.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. During the preparation of this work, the authors used Grammarly and Claude (Anthropic) to improve the language and style of the manuscript. After using these tools, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Isabel Maria Palma Aleixo Cabrita, Centro de Recursos Naturais e Ambiente, Portugal

Reviewed by: Ting Gu, Tianjin University of Science and Technology, China