Air-sea disequilibrium enhances ocean carbon storage during glacial periods

Temperature and iron fertilization are more important in driving glacial-interglacial CO2 cycles than previously thought.


This PDF file includes:
Supplementary Methods Comparison of surface carbon and oxygen with observations Fig. S1. Circulation and carbon isotope distribution in the PI and LGM simulations. Briefly, the TMM is a numerical scheme for efficient simulation of ocean biogeochemical and other passive tracers. The TMM represents tracer transport due to advection, diffusion and sub-grid scale parameterizations as sparse matrices, and the time-stepping of such tracers as a sequence of sparse matrix-vector multiplications. For the MOBI-TMM offline simulations, monthly mean transport matrices (52) and other relevant physical forcing fields, including wind speed, insolation, sea ice concentration, temperature, salinity and freshwater flux (evaporation, precipitation and runoff), were extracted from equilibrium PI and LGM runs of UVic ESCM.
All fields, including the transport matrices, were linearly interpolated to the current time step before being applied. Equilibrium and perturbation tracer simulations were run for 10,000 years to a steady-state.
To simulate the preformed tracers, we use the annually-repeating, instantaneous surface field of the corresponding tracer from MOBI-TMM as the boundary condition, which is then advected and diffused into the interior with the TMM. A periodic (seasonally-repeating) equilibrium solution was found using a Newton-Krylov method (53).

Comparison of surface carbon and oxygen with observations
To assess whether there are any systematic biases in the models' ability to simulate air-sea disequlibria in O 2 and CO 2 , we have carried out a comparison of those fields with available observations. For O 2 , we combined data from World Ocean Atlas (WOA) 2013 (54), the University of Washington Argo oxygen renalysis (55), and quality controlled data from floats deployed as part of the SOCCOM program (SOCCOMviz data portal (56); https://soccom.princeton.edu/). Data were binned spatially into the WOA 1 • ×1 • grid boxes, and temporally by month, to generate a seasonal climatology. Grid boxes without any measurements were discarded. Results are given in Fig. S4 (top three rows). Evidently, where there are data the model is consistent with them and suggest significant wintertime (June-July-August) undersaturation, but data in that season at high latitudes remain sparse. Other analysis (57) of SOCCOM float observations also show oxygen is undersaturated by up to 20% in the Seasonal Ice Zone (SIZ), in very good agreement with our model results. This broadly supports our conclusion that AOU may substantially overestimate the inventory of C soft in the ocean. The largest discrepancy with observations is during summer (December-January-February), when the model does not capture the high oxygen concentrations due to photosynthesis along the edge of the SIZ.
Comparison of surface pCO 2 with data is complicated by the fact that modern measurements are contaminated by anthropogenic CO 2 and have a strong temporal trend. To address this, we forced MOBI-TMM with observed historical atmospheric CO 2 from 1765-2018, keeping the circulation and all other physical fields fixed at their PI value, and compared simulated pCO 2 at the time and locations at which observations are available. For the data we combined monthly gridded track data from the Surface Ocean CO 2 Atlas (SOCAT), version 6 (58) and SOCCOM float data (59). As shown in Fig We conclude that the model is broadly consistent with existing measurements of surface pCO 2 and oxygen, and that there is no evidence that the model systematically overestimates disequilibrium. However, it should be emphasized that much more data is needed to better constrain oxygen and carbon disequilibrium, especially in newly-formed dense waters.

PI PI
LGM LGM

PI PI
LGM LGM  LGM equilibrium experiments. OU is the difference between the preformed O 2 concentration (surfaceO 2 transported passively into the interior by the circulation) and in situ concentration. AOU is defined as the saturation O 2 concentration, calculated at the local temperature and salinity, minus the in situ concentration. Data are plotted for all model grid points. AOU values are systematically higher than the corresponding OU values (17-19), which will translate into an erroneously higher estimate of respired carbon using the AOU method (see main text for details). The error is much larger for the LGM than PI. (B) Plot of difference between 14 C and ideal mean age in the PI simulation versus that in the LGM run. Data are plotted for all model grid points. Conventional radiocarbon ages are generally higher than ideal mean ages because of the reservoir age effect (65), but this difference increases significantly during the LGM because of the effect of circulation and sea ice on preformed 14 C (Fig. S6) (36, 66).
. Simulated A and radiocarbon age.
Γ C ) and ideal mean (Γ) age. (A,E) Change in radiocarbon age (∆Γ C ) in response to replacing the PI circulation with the LGM circulation. Top panels are for the Atlantic, bottom for the Pacific. Black solid line is the zero contour. Circulation not only affects how long ago a water parcel was last in contact with the atmosphere, but also the (preformed) concentration of the parcel when it was at the surface (36, 65), i.e., the "reservoir age". The ideal mean age on the other hand depends only on transit time and thus (B,F) the magnitude and pattern of ∆Γ resulting from a more Southern Ocean-ventilated circulation during the LGM are different from those of ∆Γ C , with much of the Pacific and bottom waters in the Atlantic experiencing shorter average transit times (67). This is evident from (C,G), which shows ∆Γ C resulting from propagating the PI surface 14 C field into the interior as a radioactively-decaying tracer with the LGM circulation using the TMM. The pattern is now much more similar to that of ∆Γ. Panels (D,H) illustrate the impact on Γ C due to the surface boundary condition by propagating the LGM surface 14 C field into the interior with the PI circulation. The increase in surface reservoir age due to an altered circulation leads to apparently older bottom waters. Sea ice has a similar effect.
. Effect of circulation changes on radiocarbon (

Fig. S7
Change in ocean carbon storage and atmospheric CO 2 (inset) in response to PI perturbations to the LGM equilibrium state.
. Response of LGM ocean carbon cycle to PI perturbations. and ∆C dis,bio (bottom) in the "SI-CO 2 " sensitivity experiment in which the perturbed sea ice only affects air-sea gas exchange of CO 2 . More sea ice prevents equilibration with the atmosphere, thus decreasing C dis,phy and increasing C dis,bio . Bottom row: (G) Surface, (H) Atlantic zonal mean and (I) Pacific zonal mean distributions of ∆C dis,bio in the "SI-bio" sensitivity experiment in which the perturbed sea ice only affects light penetration. Reduced export production leads to less respired CO 2 and, thence, a decrease in C dis,bio (24). Black solid line is the zero contour.
. Physical and biological impacts of sea ice changes on carbon storage. In this "T-bio" sensitivity experiment, the PI temperature field is replaced by the LGM one only when it affects biology. In MOBI, phytoplankton growth and respiration rate depend on temperature. Cooler glacial temperatures reduce both, with the former leading to less export production and the latter to a deeper remineralization depth and hence increased residence time of C soft . The net effect is an overall increase in the inventory of respired carbon. Bottom row: (C) Surface, (D) Atlantic zonal mean and (E) Pacific zonal mean distributions of C dis,bio in the iron sensitivity experiment. Black solid line is the zero contour. .