The elimination of carbon dioxide (RDC) will be necessary to achieve ambitious climate objectives (IPCC, 2022: Chapter 12). Fertilization of oceanic iron (IOF) stimulates CO binding in iron-limited surface water, reducing the partial pressure of CO on the sea surface (pCO) and promoting the absorption of atmospheric CO, while a fraction of the fixed carbon sinks deep, where it is temporarily stored (Buesseler et al., 2026). Most evaluations place its potential higher than 0.1–1 Gt CO Year−1, in idealized scenarios assuming almost complete relief from iron limitation, some reaching several gigatons (IPCC, 2022; NASEM, 2022), i.e. a maximum of a few percent of current anthropogenic CO Emissions (42 Gt CO Year−1; Friedlingstein et al., 2025). The OIF is nevertheless still under development as a CDR option.Buesseler et al. (2026) advocate a new generation of OIF field trials, larger (>1000 km) and longer (3 to 6 months) than any other to date (25 to 300 km and 10 to 40 days; Boyd et al., 2007), to quantify both the effectiveness of iron fertilization in sustainable and additional carbon removal and its impacts on marine ecosystems. The OIF is already the subject of ethical and governance debates that need to be addressed (Boettcher et al., 2021; Cooley et al., 2023; Loomis et al., 2022); here, we question the technical feasibility and scientific value of the tests themselves, central considerations in international frameworks designed to assess whether such research is scientifically justified (e.g., the London Protocol; In my opinion, 2010). Buesseler et al. (2026) identify several major uncertainties to be resolved before the OIF can progress towards proof of concept. However, 30 years of studies in oceanic biogeochemistry (Boyd et al., 2007; Cullen and Boyd, 2008; de Baar et al., 2005; Martin, 1990; Strong et al., 2009; Yoon et al., 2018) Propose two conclusions. First, the structural uncertainties surrounding the potential for the OIF’s suppression cannot be resolved by these trials. Second, the trials can neither assess the impacts of the IF at relevant climate scales (i.e. global and continuous) or mitigate the harmful effects they could cause. A single common thread crosses the two: the trials would transfer their central questions to global biogeochemical models, which have repeatedly shown that this potential is low even under unenforceably favorable assumptions (Aumont and Bopp, 2006; Tagliabue et al., 2023; Zahariev et al., 2008).
Field trials cannot limit central uncertainties regarding the potential for withdrawal from the OIF
To limit the additivity and durability of carbon export, a test should not only detect export and its depth, but isolate the fraction caused by fertilization. However, only a small portion of the additional biomass produced after the addition of iron is likely to collapse: proliferations can be limited by other nutrients, such as silicic acid, and a large part of the biomass is consumed first by grazers (Boyd et al., 2004; Martin et al., 2013). Even when carbon flows, little reaches greater depths: in the region underlined by the authors, only 20% to 50% of the particle flow at 150 m arrives at 500 m, with great differences between sites (Buesseler et al., 2007). The authors themselves recognize that the expected increase to 500 m could be « in the range of natural variability of the flow of people of color », making the fertilization signal difficult to distinguish from the background. More fundamentally, the increase in exports is not guaranteed: among the 13 historical trials of the OIF, only EIFEX (the European Iron Fertilization Experiment of 2004) has achieved an unambiguous increase in deep carbon exports (Yoon et al., 2018).Oceanic physics adds to the challenge. A fertilized plot is not a sealed test tube; The turbulent movement stretches it into fine filaments over hundreds of kilometers (Lévy et al., 2018), more strongly at the surface than in depth, spatially decoupling the surface pCO A reduction in any deep export. During SOIREE, the area extended from about 7 km to more than 150 km in six weeks (Abraham et al., 2000), and in six months an area could reach 100 to 400 times its initial surface (Groeskamp et al., 2020), moving away from any realistic instrument configuration. The same stirring mixes fertilized and surrounding waters, blurring the boundary between disturbed regions and control regions (Boyd et al., 2023), while fine-scale vertical flows can bring some of the exported material back to the surface (Resplandy et al., 2019). Untangling this signal from natural variability, seasonal ventilation, ongoing climate change and extreme events would be extremely difficult (Fennel, 2026; Tagliabue et al., 2023). These limitations weaken the scientific justification of new field trials at the scale currently proposed.The uncertainties that govern the net climate benefit of the OIF come from processes covering local scales to those of the basin, then from days to years, an inevitable tangle of local and short tests by nature (Gnanadesikan and Marinov, 2008). The authors propose to focus the tests on the depth of export, on the grounds that it governs the durability of storage. But the length of time exported carbon remains outside the atmosphere also depends on the circulation and where it is remineralized (Siegel et al., 2021): in the Southern Ocean, two-thirds of exported carbon up to 1000 m can be re-exposed in a century (Robinson et al., 2014). The depth of export is therefore useful but insufficient to quantify what the authors call a « centennial ton »: carbon exported and preserved from the atmosphere for at least 100 years. This will require relatively high-resolution global models, whose circulation representations differ considerably and remain a major source of uncertainty in the carbon biocycle.Measuring export is not measuring CO Suppression, let alone net climate profits. Export reduces sea surface pCO locally, but atmospheric absorption follows months to years and may remain incomplete if fertilized waters are subducated before rebalancing (Bach et al., 2023; Zhou et al., 2025); The authors recognize that this air-sea flow cannot be assessed solely by local observation. A « centennial ton » of export is therefore not a « centennial ton » of abduction. The benefit also depends on far-field effects that a local trial can neither observe nor attribute with certainty: fertilization increases productivity within the plot while depletening the nutrients that would have supported it downstream (Jin et al., 2008; Noh et al., 2025; Tagliabue et al., 2023), and can modify the production of greenhouse gases such as NO, partially compensating for the expected radiative benefit (Law, 2008). The uncertainties that determine the real effectiveness of the OIF’s climate therefore operate precisely at scales that field trials cannot cover.Such experiments would not necessarily solve the modeling problem they are supposed to solve. The authors present new trials as a way to improve oceanic biogeochemical models, arguing that they misrepresent many processes controlling the carbon biological cycle (Henson et al., 2022). But improving the representation of processes by models requires observations that isolate specific processes and constrain them one by one, from particle aggregation to grazing, through remineralization and air-to-sea exchange (Siegel et al., 2016). A fertilization trial does the opposite: it would impose a disturbance on an already variable and incompletely characterized ecosystem, producing an integrated response in which these processes are disturbed. Such an answer can show what happened in the same place at a given time, but it would not reveal which process is poorly represented in the model or how its representation should be corrected.The authors see « a considerable merit to learning through practice », but learning requires experience to isolate what needs to be learned. Here, the amount of interest is precisely that which the test cannot separate from natural variability, cannot follow up to its centenary horizon, and cannot distinguish it from an unmeasured and highly dynamic distant field. In this model, « doing » does not solve uncertainty. It produces a global signal whose large-scale interpretation has yet to be transmitted to global biogeochemical models that the tests were supposed to improve.More importantly, the limitations of current models do not make the general assessment of the limited potential for OIF suppression in the peer-reviewed literature unreliable. Although global ocean biogeochemical models necessarily simplify many processes, they play a central role in the evaluation of the ocean carbon cycle (Friedlingstein et al., 2025) and probably represent the main controls on carbon removal of the OIF well enough to estimate the order of magnitude. Through a range of model structures and idealized fertilization scenarios, they systematically find limited elimination potential, even under unrealistic and optimistic assumptions such as prolonged iron limitation or large-scale fertilization of the Southern Ocean (Aumont and Bopp, 2006; Oschlies et al., 2010, 2025; Tagliabue et al., 2023; Yu et al., 2026; Zahariev et al., 2008). This consensus is corroborated by the paleo-oceanographic archives: even millennia of natural iron fertilization at the scale of the basin by dust during the ice ages produced only a small additional flow of atmosphere-ocean carbon (0.01 year Gt C−1; Boyd, 2008).
Trials can neither assess nor mitigate the impacts of the OIF
The trials are also presented as a way to determine, through a « go/no-go » framework, whether the OIF can cause long-term alterations of oceanic ecosystems. They can fail for two reasons: they cannot assess the impacts that count the most on climate scales, and the framework cannot mitigate the damage they could cause.Impacts that determine whether the OIF is safe for climate-relevant CDR operate at scales that cannot be achieved by a field trial. A trial is local and lasts for months, while the consequences are very distant and take place over years or even decades. Fertilization depletes nutrients that would otherwise support downstream productivity, with consequences for fishing and higher trophic levels (Gnanadesikan and Marinov, 2008;Tagliabue et al., 2023); These responses are also intermittent and irregular, making them particularly difficult to detect and assign. Large-scale OIF could further intensify deoxygenation and acidification under fertilized regions and in remote waters connected by traffic (Oschlies et al., 2010, 2025), acting in the same direction as climate change and thus amplifying existing stressors (Tagliabue et al., 2023). None of this can be observed on a single area over a few months, as themselves acknowledged by Buesseler et al.It is crucial to note that these are not speculative risks: they are systematically identified through the same global models on which the authors rely elsewhere, and they evolve with the intervention. Effects that seem negligible in a small area can reach, for example, a 10% loss of oxygen from the Southern Ocean on the relevant climate scale (Oschlies et al., 2010, 2025). A trial never measures this scale-impact relationship, cannot demonstrate that large-scale deployment is safe and, if perceived as successful, could provide false assurance. The hypothesis that the system simply « resets » after stopping the addition of iron is equally unverifiable: the recovery of slow processes such as subsoil deoxygenation or ecosystem reorganization cannot be observed on the time scale of a trial and is postulated rather than shown; climate-relevant OIF would in any case imply continuous fertilization without restart. We recognize that the gradual expansion of experiences is a standard practice, and that each CDR option involves compromises that decision-makers must weigh on them. But the argument in favor of larger trials depends on the resolution of the uncertainty that justifies them, and here they cannot: the decisive impacts go beyond the scope of the proposed trials, so that their diagnosis would again be delegated to the same models whose limits the authors invoke elsewhere. These limits of the model should not be used to rule out ecosystem risks that are difficult to observe and quantify but remain highly plausible.Even for the impacts that a test can detect, the proposed exit is reactive rather than preventive. Monitoring toxins, oxygen, pH or food network variations can help decide when to stop an experiment, but cannot reverse the changes already triggered, and some of the relevant harms are local and observable but potentially irreversible. Iron enrichment stimulated the production of domoic acid by Pseudo-nitzschia in the seawater of the Papa ocean station (Trick et al., 2010), and concentrations reached ecologically worrying levels during SOFeX (Silver et al., 2010); Food web displacements and underwater deoxygenation caused by iron could also persist after the cessation of iron addition. A go/no-go rule can therefore limit additional exposure, but cannot make the intervention safe once an impact has been triggered, not to mention remedying distant consequences that the test never respects. This limitation is accentuated with a sustained and climate-relevant deployment.
Conclusion
Taken together, these limitations constitute the scientific argument against a new generation of OIF field trials. Such experiments can document the responses of local ecosystems, but they cannot resolve the uncertainties that determine whether climate-relevant OIF could ensure sustainable and additional net carbon removal, nor demonstrate its environmental safety at this scale. Instead, their interpretation would again depend on global biogeochemical and ecosystem models whose limits are invoked to justify the tests from the start. This creates an unfavorable balance between scientific performance and environmental risk. Uncertainty should not allow for larger and larger interventions when they cannot answer the questions on which their acceptability depends.Instead, the priority should be to improve our understanding of the fundamental mechanisms governing the export, processing and storage of carbon in the oceans. This knowledge is necessary independently of the OIF: it is essential to anticipate how the ocean’s biological carbon cycle will react to current and future pressures. It would be better to continue it with targeted observations and the development of models. For example, research should focus on targeted process studies and sustained observations of naturally fertilized systems, including waters around the Kerguelen and Crozet archipelagos, as well as iron-enriched waters surrounding the Pacific islands (Bourdin et al., 2026). At the same time, models should be used to test how uncertainties in carbon export, mesoscale physics and biogeochemical feedback modulate the estimated carbon yield of the OIF. Such efforts would provide a more solid empirical basis for improving biogeochemical models without relying on increasingly large deliberate disturbance experiments whose interpretation remains structurally ambiguous.We make these comments in a constructive spirit and with great respect for the authors of the proposal put forward by Buesseler et al., many of whom are leading scientists whose contributions have greatly improved our understanding of the oceanic carbon cycle. We hope that this criticism will contribute to an open and constructive scientific dialogue and help guide future research efforts and investments towards approaches most likely to produce strong and policy-relevant knowledge. In the end, we share the same goal: to identify effective, scientifically sound and socially responsible avenues for climate mitigation.
