The motivation reaches deep into the history of Earth’s climate. The triple isotopic composition of atmospheric oxygen, reported as 17Δ and measured in air bubbles trapped in polar ice cores, acts as a global integrator of past biospheric oxygen fluxes and can provide indirect constraints on associated carbon-cycle processes. To decode that signal, scientists must separate the contribution of the oceanic biosphere, 17Δocean, from those of the terrestrial biosphere and the stratosphere. Until now, ocean-focused efforts had relied on box models, one-dimensional and two-dimensional models, productivity-based schemes, and an implementation in the ocean component of the Community Earth System Model, but 17Δocean had never been implemented in a fully coupled Earth system model.

The new implementation explicitly represents the three processes that control 17Δocean: photosynthesis, respiration and air–sea gas exchange. Three prognostic tracers corresponding to the bulk abundances of the isotopologues 32O2, 33O2 and 34O2 are carried through the model and updated at each time step. Biological oxygen production and consumption are derived from the marine carbon-cycle model embedded in iLOVECLIM, which uses a single phytoplankton species, a single zooplankton species and a single nutrient, phosphate, with carbon-to-oxygen conversions following Redfield stoichiometric ratios. Air–sea exchange of oxygen is calculated independently of the air–sea CO2 flux, using oxygen-specific formulations that include both equilibrium and kinetic isotopic fractionation factors, and a piston velocity parameterized as a quadratic function of wind speed.

Some simplifications shape the results. Because the ecosystem model does not explicitly simulate gross primary productivity of oxygen or autotrophic respiration, the authors assume autotrophic respiration equals phytoplankton production, giving a gross-to-net ratio of 2 and an overall conversion factor of roughly 2.6 between carbon-based production and gross oxygen productivity. That figure sits close to the ratio of 2.7 reported by Marra in 2002, though observational estimates range from about 1.8 to 4.5 across the ocean, so a spatially constant ratio may lead to regional over- or underestimation. Photosynthesis is implemented without isotopic fractionation, respiration applies an 18O/16O fractionation factor of 0.980 with a triple-isotope exponent of 0.518, and the model is initialized to reproduce a 17Δocean of 41.5 ppm and a δ18O–O2 of 0.151 per mil.

The simulations were run under preindustrial boundary conditions, with atmospheric CO2 fixed at 280 ppm, until the deep ocean reached equilibrium after 3000 years of model time. The authors stress that the observational datasets were collected under modern oceanic conditions at specific locations, depths, seasons and times, so point-by-point agreement is not expected; the comparison is intended to evaluate large-scale consistency. Within that framing, the model performs well. Total dissolved oxygen in the photic zone is generally consistent with the World Ocean Atlas 2023 climatology, with a slight depletion of roughly 30 µmol kg−1 in polar regions and modest enrichment in some equatorial upwelling areas. Global gross primary productivity of oxygen reaches 9.68×10^15 mol O2 per year, with the equator hosting optimum values of 150 mol O2 per square metre per day, elevated values in upwelling zones and high-latitude spring blooms, and low values at the centers of subtropical gyres.

The isotopic tracers follow physically coherent patterns. Simulated δ18O–O2 is about −0.5 per mil at the equatorial surface and about 1 per mil at the poles, and increases with depth by an average of 7 per mil in the Atlantic and 10 per mil in the Pacific, reaching roughly 16 per mil in the Atlantic oxygen minimum zone and up to about 22 per mil in the Pacific equivalent, where older water masses allow greater cumulative respiratory fractionation. Simulated 17Δocean reaches about 50 ppm at the equator, around 40 ppm in upwelling regions, and between −5 and 20 ppm at the centers of the gyres, while near Antarctica values climb to 100 ppm where seasonal sea ice blocks air–sea exchange and allows the tracer to accumulate beneath the ice.

Not every region matches the observations. In the Atlantic, modeled subsurface values tend to be overestimated: at 20° S and 3 kilometres depth, simulated δ18O–O2 reaches 7 per mil against an observed value closer to 3 per mil, a relative error of approximately 133 percent, and at 2 kilometres depth the simulated 17Δocean shows a relative error of about 30 percent, with discrepancies growing below 4 kilometres. The authors suggest several explanations, including excessive surface oxygen production exported to depth via Antarctic Bottom Water, an inaccurate representation of that water mass’s dynamics, the choice of respiratory fractionation values, and inherent uncertainties in the observational datasets. They also note that observed values in the region may be partly influenced by Labrador Sea Water, including the Iceland-Scotland and Denmark Strait overflow waters, whereas the model’s region is largely dominated by Antarctic Bottom Water.

The Southern Ocean emerges as the model’s clearest weakness. Comparing monthly values in the photic layer at the exact locations of available measurements, the model consistently overestimates 17Δocean there, with all simulated points lying above the 1:1 line: observed values range from 0 to 80 ppm while the model spans 20 to 160 ppm. The most likely cause, the authors conclude, is an overrepresentation of primary production, plausibly linked to the model’s lack of iron limitation, since iron constrains phytoplankton growth in high-nutrient, low-chlorophyll regions such as the Southern Ocean. A complementary comparison in temperature–salinity space shows that the positive bias persists even for waters occupying comparable hydrographic ranges, suggesting a systematic model behaviour rather than an artifact of comparing dissimilar water masses. In the Pacific, by contrast, simulated values generally fall within the observed range.

At fixed stations, the picture is more encouraging but incomplete. Year-round measurements at the Bermuda Atlantic Time-series and Hawaii Ocean Time-series sites show that the model captures the timing of seasonal variability in 17Δocean within the upper photic layer, with the strongest seasonal variation between 50 and 100 metres in both observations and simulations. The amplitude, however, is underestimated: during September to November, measured 17Δocean peaks at 120 ppm while the model reaches only 45 ppm. The modeled tracer also stabilizes much shallower, around 100 metres, because the prescribed photic layer is limited, whereas observations stabilize below 250 metres at BATS and around 300 metres at HOTS. Limited seasonal variability in simulated phytoplankton productivity at these subtropical sites appears to be a primary driver of the muted amplitude.

Oxygen minimum zones pose another challenge. In iLOVECLIM these zones tend to expand excessively, so the team introduced an ad hoc lower threshold of 30 µmol kg−1 of total oxygen, below which remineralization is halted to ensure numerical stability. The constraint prevents unrealistic anoxia but also truncates the zones and may weaken the biogeochemical contrast needed for realistic isotopic gradients, producing a truncated oxygen minimum zone alongside an exaggerated isotopic response that complicates interpretation. A sensitivity experiment using the revised air–sea gas-exchange fractionation factors of Li et al. (2019) brought measurable improvements: within oxygen minimum zones the revised factors reduce δ18O–O2 by up to 2.5 per mil and 17Δocean by up to 24 ppm, with interior decreases reaching about 16 ppm, while the coefficient of determination for δ18O–O2 remained 0.70 in both simulations and rose slightly for 17Δocean, from 0.02 to 0.03. The authors attribute these low coefficients to comparing climatological model fields with sparse observations rather than to a point-by-point evaluation, and caution that the revised factors cannot by themselves explain the remaining Southern Ocean discrepancies.

For the researchers, the payoff is a diagnostic as much as a simulation tool. Implementing 17Δocean exposed biases in the model’s seasonal cycles and oxygen minimum zones that would otherwise have remained hidden, and the framework now opens the way to reconstructing past variations in ocean productivity. The next step is to extend the implementation to the terrestrial biosphere, which would allow reconstruction of the past global biosphere and direct comparison with 17Δ records from air bubbles in ice cores, completing the chain from dissolved oxygen in the modern ocean to the isotopic archive of Earth’s biological past.

Subject of Research: First implementation of the triple isotopic composition of dissolved oceanic oxygen in the iLOVECLIM coupled climate model

Article Title: Including the triple isotopic composition of dissolved oxygen in the ocean into the iLOVECLIM model (version 1.1.7): development and evaluation

Article References: Clermont, E., Yang, J.-W., Extier, T., & Roche, D. M. (2026). Including the triple isotopic composition of dissolved oxygen in the ocean into the iLOVECLIM model (version 1.1.7): development and evaluation. Geoscientific Model Development, 19(20), 10203-10222. https://doi.org/10.5194/gmd-19-10203-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-10203-2026

Keywords: 17Δocean, triple oxygen isotopes, iLOVECLIM, Earth system model, ocean productivity, dissolved oxygen, ice cores, oxygen minimum zones, air-sea gas exchange, Southern Ocean, biogeochemistry, model evaluation