Ocean Fertilization Just Got a Reality Check: Why Patch Size and Physics Matter for Carbon Capture

Only 5% of the carbon absorbed by a fertilized ocean patch is taken up in the first two weeks — the rest accumulates slowly over months, long after most monitoring ends. That’s the striking implication of a new modeling study that rethinks how we measure the effectiveness of ocean iron fertilization, a controversial climate intervention strategy once hailed as a quick fix for atmospheric CO₂.

For decades, scientists have debated whether seeding the ocean with iron could trigger massive phytoplankton blooms, pulling down carbon like a biological vacuum. Early experiments like the Southern Ocean Iron Enrichment Experiment (SOIREE) suggested it worked: satellite images showed green swirls spreading across the sea. But translating those blooms into reliable carbon removal has been maddeningly uncertain. Estimates of how much CO₂ was actually absorbed varied wildly — sometimes by a factor of four — even for the same event.

Now, a team led by Stefano Campagnola, Stephanie Dutkiewicz, Michael J. Follows, and Enrico Ser-Giacomi has built a new kind of model that tracks not just biology, but the full three-dimensional physics of a fertilized water patch as it stretches, thins, and mixes with surrounding waters (Campagnola et al., 2026). Their framework treats the ocean not as a uniform vat, but as a dynamic, evolving parcel — a Lagrangian ecosystem — whose shape and movement fundamentally alter how much carbon it can pull from the air.

And the results flip conventional wisdom on its head: dilution, long seen as a problem, can actually boost productivity. The size of the initial iron injection matters more than expected. And most surprisingly, the bulk of carbon capture happens long after the bloom peaks — a finding that challenges how we design and evaluate future climate interventions.

The Science

At the heart of this work is a shift in perspective: from Eulerian (fixed-location) to Lagrangian (moving-patch) modeling. Most ocean models look at what happens at a given point in space over time. But when you add iron to the sea, that water doesn’t stay put. It moves, deforms, and mixes. To understand the true carbon budget, you need to follow the patch itself.

The researchers built an idealized 3D model of a Lagrangian water parcel — approximated as an ellipsoid — that evolves under four key physical forces: horizontal strain (stretching), divergence (spreading), and horizontal and vertical diffusivity (mixing). These processes change the patch’s width $W(t)$, length $L(t)$, and thickness $T(t)$ over time, altering its volume $V(t)$ and surface area $A(t)$.

Crucially, the model doesn’t assume the patch is well-mixed. Instead, it uses Reynolds decomposition to track not just average tracer concentrations (like iron or phytoplankton), but also their spatial variances and covariances — the heterogeneity within the patch. This allows the model to capture how localized hotspots of nutrients and biology interact, a detail most previous models ignored.

The biochemical engine is tied to carbon dynamics through the partial pressure of CO₂, $pCO_2$. The model tracks $\Delta p(\mathbf{x},t)$, the anomaly in $pCO_2$ between the patch and the atmosphere, governed by three main terms:

$${\left[\frac{d\Delta p(\mathbf{x},t)}{dt}\right]}_{bio}=-\varphi \frac{K_o{K}_g}{h}\Delta p(\mathbf{x},t)+\frac{\varphi }{\rho }\frac{1}{R_{N:C}}{S}_{upt}(\mathbf{x},t)-\varphi {\Sigma }_{\mathcal{T}}\frac{d\mathcal{T}(\mathbf{x},t)}{dt}$$

The first term represents air-sea CO₂ exchange, the second biological uptake, and the third solubility changes due to temperature. From this, the authors define two integrated metrics:

• Lagrangian Carbon Flux (LCF): total carbon absorbed by the evolving patch over time.
• Lagrangian Primary Production (LPP): total organic carbon produced by phytoplankton in the patch.

They tested their model against SOIREE, the first major iron fertilization experiment in the Southern Ocean, which released sulfur hexafluoride (SF₆) to track a 50 km² patch for over 40 days.

What They Found

The model successfully reproduces key features of SOIREE. It captures the observed expansion of the patch (

), the rise and fall of iron and phytoplankton concentrations, and the development of spatial heterogeneity. Phytoplankton biomass peaks around day 21, reaching 16 times background levels — consistent with satellite and in situ data.

But where it diverges — and delivers its most consequential insight — is in carbon uptake. The model estimates that over the full 42-day simulation, the patch absorbed 690 tons of carbon. Yet, during the first 13 days — the period covered by ship-based measurements — only 32 tons were taken up. That’s just 4.6% of the total.

This slow, prolonged drawdown contradicts the assumption that carbon capture happens rapidly during the bloom peak. It also explains why past estimates varied so much: studies relying on short-term data missed the bulk of the signal.

The model further shows that a competing model, SWAMCO — which assumes a constant entrainment rate and neglects patch deformation — overestimates early carbon uptake by more than fourfold (130 tons vs. 32). This suggests that simplified models may be systematically biased.

Even more surprising is the role of dilution. In a homogeneous model (where the patch is assumed uniformly mixed), dilution always reduces carbon uptake — more mixing means more dilution of nutrients, weaker blooms, less CO₂ drawdown. But in the heterogeneous model, moderate dilution enhances primary production by creating spatial gradients that boost phytoplankton growth through covariance effects.

The size of the initial injection also matters critically. When the same amount of iron is spread over different initial areas, the model reveals an optimal range: 70 to 150 km². Smaller patches (e.g., 10 km²) dilute too quickly; larger ones (e.g., 300 km²) evolve too slowly due to scaling laws in ocean turbulence. Remarkably, SOIREE’s actual injection size — about 50 km² — falls near the lower edge of this sweet spot, making it surprisingly efficient for carbon capture.

Vertical processes add another layer of complexity. Positive horizontal divergence (upwelling) increases LCF by about 40% across all conditions, likely by thinning the patch and enhancing surface exposure. But it has little effect on LPP, which depends on volume. Meanwhile, higher vertical diffusivity reduces LCF (by exchanging carbon with deeper layers) but increases LPP (by bringing up more iron).

Lagrangian Carbon Flux Over Time

Label	Value
Day 0	0
Day 10	25
Day 20	150
Day 30	400
Day 40	690

Source: Campagnola et al. (2026). Most carbon uptake occurs after day 14.

Why This Changes Things

These findings aren’t just academic — they reshape how we think about ocean fertilization as a climate tool. For years, the debate has centered on whether it works at all. This study says: yes, but not in the way we thought.

First, the timescale problem. If 95% of carbon uptake happens after day 14, then short-term monitoring — the norm in past experiments — is fundamentally inadequate. Any future trial aiming to quantify carbon removal would need autonomous floats or gliders to track the patch for months, not weeks. Without that, we’re measuring the tip of the iceberg.

Second, the patch size dilemma. Most past experiments used small injections (10–50 km²), assuming bigger is better. But this model suggests there’s a Goldilocks zone. Too small, and dilution kills the bloom. Too large, and the patch evolves too slowly to respond efficiently. This has direct implications for scaling: if we ever deploy ocean fertilization at scale, we can’t just dump iron everywhere. We’d need to optimize injection geometry based on local physics.

Third, the physics-biology coupling. The old view treated biology as the driver and physics as noise. This work shows they’re inseparable. Dilution isn’t just a loss mechanism — it can amplify productivity. Vertical upwelling doesn’t just supply nutrients — it reshapes the entire carbon budget. Any credible assessment of ocean fertilization must now account for this coupling.

And finally, the heterogeneity factor. By resolving spatial variances and covariances, the model captures emergent properties that homogeneous models miss. This isn’t just a technical upgrade — it’s a conceptual shift. Ecosystems aren’t uniform; their patchiness is part of what makes them work.

What's Next

The study opens as many questions as it answers. The model is idealized — it doesn’t include zooplankton grazing, multiple phytoplankton types, or light limitation. It assumes constant external forcing, whereas real ocean patches encounter eddies, fronts, and storms. And it doesn’t track how much carbon actually gets sequestered at depth — the ultimate goal of fertilization.

But its power lies in its efficiency. Because it’s computationally lightweight, the authors could run ensembles of 3,000 simulations to map sensitivities across physical parameters. That kind of exploration would be impossible with full 3D biogeochemical models.

Next steps include coupling this Lagrangian framework to real ocean circulation data, testing it against other fertilization experiments (like LOHAFEX or EIFEX), and extending it to include export efficiency — how much of the produced organic carbon sinks below the mixed layer.

Perhaps most urgently, the model underscores the need for long-duration observational campaigns. Autonomous Lagrangian floats, like those used in the Global Drifter Program, could be adapted to carry biogeochemical sensors. Paired with satellite tracking, they could finally provide the ground truth needed to validate models like this one.

The broader lesson extends beyond iron fertilization. As we explore other ocean-based carbon removal strategies — from alkalinity enhancement to macroalgal sinking — we’ll face the same challenge: how to measure efficacy in a dynamic, three-dimensional environment. This study offers a template: follow the water, respect the physics, and never assume uniformity.

Ocean fertilization remains controversial. It could disrupt ecosystems, alter food webs, and create oxygen-depleted zones. The ethical and governance questions are profound. But if we’re going to test it — or any marine carbon dioxide removal method — we need tools that reflect reality, not convenience. This model is a step toward that rigor. It doesn’t tell us whether we should fertilize the ocean. But it does tell us how we’d measure it — honestly, thoroughly, and without illusion.