In presenting our simulations, we focus on the difference between the
HC and LGM. We first show the simulated physical changes in the ocean
and sea-ice, and then compare them to LGM reconstructions. We then
present how the ocean biogeochemical fields differ between the HC and
LGM climates with a fixed ocean BGC parameterization. Finally, we
explore how modifying the BGC parameterization alters the ocean carbon
storage and key ocean BGC fields like dissolved oxygen and aragonite
saturation state.
LGM climate evaluation
Temperature
The LGM-simulated surface ocean is much colder than the HC (3.2 ∘C, see Table ) with the greatest cooling in the annual averaged SST occurring in the North Pacific, North Atlantic and the Pacific sector of the Southern Ocean (Fig. ).
In comparison to paleo-reconstructions, our simulated LGM cooling was more than 1 ∘C colder than the estimates (Table ), and more comparable to reconstructed SST cooling over the tropical ocean (30∘ S–30∘ N) of 2.7±0.5 ∘C.
Our simulation did not produce a large difference in the averaged value of SST cooling between the Atlantic, Pacific and Indian Oceans (less than 0.3 ∘C difference).
In contrast, the paleo-reconstruction shows more than a 1 ∘C greater cooling in the Atlantic Ocean than the global average (Table ), while the reconstructions had regional tropical coolings of 3 ∘C in the Atlantic, 2.5 ∘C in the Indian, and 1 ∘C in the central Pacific Oceans.
Perhaps some of the discrepancies between the simulated LGM cooling and paleo-reconstructions reflect the large spatial variability in SST cooling and the limited samples available for the paleo-reconstructions.
As synthesised by , the paleo-reconstructions of SST difference between the HC and LGM based on foraminiferal assemblages display large spatial variability in the tropical oceans.
In the eastern tropical Atlantic (east of 25∘ W), the LGM cooling is in excess of 3 ∘C with several cores suggesting cooling of greater than 5 ∘C .
In contrast, LGM cooling in the western tropical Atlantic Warm Pool is only 1 to 2 ∘C .
In the tropical Pacific, most cores show LGM cooling of less than 3 ∘C, however several cores suggest cooling of greater than 6 ∘C .
In contrast, the LGM cooling in the Western Pacific Warm Pool is less than 1 to 2 ∘C and several cores in the central tropical Pacific show little or no cooling in the LGM .
The simulated change in SST between the HC and LGM show similar spatial variability to paleo-reconstructions, with greatest cooling in the equatorial oceans and high latitudes, and least cooling in the subtropics (Fig. d).
Like the paleo-reconstructions, the LGM simulation has less cooling in the tropical West Pacific than the East Pacific (Fig. d).
While our simulated LGM cooling was greater than the paleo-reconstructions, it was generally within the existing spread of paleo-reconstructions and the spatial variability they displayed.
Sea-ice extent
Associated with the colder SST, our LGM simulation has much greater
sea-ice cover than the HC. The simulated maximum seasonal sea-ice
cover in the LGM included a large portion of the North Pacific, North
Atlantic and Southern Ocean as far north as 45∘ S
(Fig. ).
The simulated maximum Southern Hemisphere sea-ice cover in the LGM (35×106 km2) is approximately double the simulated HC value (15×106 km2).
A reconstruction of the maximum sea-ice extent (sea-ice concentration greater than 15 %) shows sea-ice extending to 47∘ S in the Atlantic and Indian sector and 57∘ S in the Pacific sector – a northward displacement of 7 to 10 ∘ in latitude , which is comparable to a our LGM simulation (Fig. ).
From the LGM reconstructions, the maximum sea-ice extent in the Southern
Hemisphere expands to about 39×106 km2 ,
which represents an approximate doubling of ice cover from the modern value
(19×106 km2; ).
Importantly, for both HC and LGM the simulated maximum Southern Hemisphere sea-ice cover, 15 and 35×106 km2 respectively, were similar to the observed and reconstructed values.
In the Northern Hemisphere, the simulated maximum sea-ice extent increased from the HC value of 12×106 km2 to a LGM value of 29×106 km2, a more than doubling the sea-ice extent.
LGM reconstructions give a North Atlantic winter maximum sea extent that extends south of Iceland then across to the southern tip of Labrador, which is similar to the simulated LGM maximum sea-ice extent in the North Atlantic (Fig. ).
Both the paleo-reconstructions and our LGM simulation show a dramatic retreat of the minimum sea-ice extent to north of Greenland during the summer (Fig. ).
In the North Pacific, the simulated maximum LGM sea-ice extent went as far south as 45∘ N, with much greater sea-ice cover in the western Pacific than the eastern Pacific.
Consistent with our LGM simulation, paleo-reconstructions suggest that LGM winter sea-ice extent was greater in the western Pacific , with no sea-ice cover at 50∘ N and 167∘ E in the Central North Pacific .
During the summer season, the simulated LGM minimum seasonal sea-ice extent rapidly contracts to cover 6.5 and 11.4×106 km2 in the Southern and Northern Hemispheres, respectively, which is less than 1/3 of the maximum sea-ice extent (Fig. ).
In contrast, the HC simulation had sea ice cover of 1.7 and 5.3×106 km2 in the Southern and Northern Hemispheres, respectively.
The simulated minimum sea-ice extent was much greater in the LGM than the HC with areas in the North Pacific, North Atlantic and Southern Ocean where sea-ice was present throughout the year.
Meridional overturning
Associated with the changes in the surface ocean during the LGM were
changes in the meridional overturning circulation. In the LGM
simulation, the rate of Antarctic Bottom Water (AABW) formation
increased to 15 from 7 Sv in the HC (Fig. , minimum
in high latitude Southern Hemisphere Global overturning). The LGM
simulation also showed greater subduction of Southern Ocean
intermediate water than the HC simulation. The simulated LGM North
Atlantic Deep Water (NADW) overturning (15 Sv) was similar to
the HC simulation (17 Sv), but the cell was shallower in the
LGM simulation (maximum in the North Atlantic overturning). The LGM
simulated meridional overturning showed that the deep water was
dominated by water from the Southern Ocean.
The simulated features of meridional overturning in the LGM are
supported by the paleo-reconstructions. Paleonutrient tracers
indicate that the boundary between NADW and AABW was substantially
shallower during the LGM than today . In
our simulations, the NADW cell rises from a maximum depth of
3000 m in the HC to a depth of 1500 m in the LGM.
During the LGM, paleonutrient tracers also suggest the North Atlantic Ocean was more
stratified, with a shoaling of the NADW cell, and the AABW penetrating
much farther into the North Atlantic than at present. The greater
contribution of AABW to the deep water of the North Atlantic is also
consistent with Atlantic sediment cores that indicate that in the LGM
the Atlantic deep waters were much colder and saltier than modern day
. The paleo-reconstructions imply that the bottom
waters of the LGM were considerably more saline than today and that
the deep ocean was more stratified, with the stratification being due
to salinity rather than temperature as the bottom temperatures were
close to freezing throughout much of the world ocean
. The simulated salinity distribution in the LGM
was consistent with paleo-reconstructions (Fig. ). As
postulated by , the very cold and saline
bottom water was the consequence of more production of brines from
greater sea-ice formation in the Southern Ocean in the LGM simulation.
Originally, the shoaling of the NADW and greater contribution of AABW
to the North Atlantic was interpreted as a significant reduction in
NADW formation , but more recent
paleonutrient tracers suggest that during the LGM, NADW was similar or
slightly reduced compared to HC . This
slight reduction and shoaling of the NADW in the LGM was consistent
with our simulations.
Ocean biogeochemical fields
The colder surface water coupled with greater subduction of AABW in
the LGM simulation filled the deep LGM ocean with water of a much
greater oxygen concentration than the HC (Fig. ). The
LGM-simulated oxygen concentration in the ocean was more than
100 mmolm-3 greater than the HC simulation with the
simulated LGM global oxygen content about 55 % greater than
the HC simulation (Table ).
For dissolved inorganic carbon (DIC), the difference between the LGM
and HC simulations was more complicated than oxygen
(Fig. ). In the deep water (below 2000 m), the DIC
concentrations in the LGM simulation were greater than the HC
simulation, consistent with the subduction of more colder AABW during
the LGM, which, like oxygen, elevated DIC concentrations in the deep
ocean.
However, in the upper ocean during the LGM, the simulated DIC concentrations were much less than the HC simulation, which reflected the lower atmospheric CO2 concentration in the LGM than in the HC: 185 vs. 280 ppm.
The reduction in atmospheric CO2 between the HC and LGM meant the surface water equilibrated with a much lower atmospheric CO2 level in the LGM and subsequently reduced DIC concentrations.
The reduced DIC concentration exceeded 200 mmolm-3 in the surface water of the subtropical and tropical ocean.
The reduced DIC concentrations extended furthest into the interior in the downwelling regions of the subtropical gyres.
There was also a similar decline in the high latitude Northern Hemisphere in the LGM simulation associated with reduced atmospheric CO2 and freshening of surface water.
In the LGM simulation, alkalinity concentrations (like salinity) were much greater in the deep ocean than the HC simulation (Fig. ).
Compared to the HC, LGM-simulated alkalinity decreased in the water above 1500 m and increased in the water below (Fig. ).
The zonal averaged change in the simulated alkalinity was very similar to the salinity change between the LGM and HC (Fig. ).
The large increase in the DIC concentrations in the deep ocean in the
LGM simulation reflects changes in the subduction of DIC-rich polar
water and changes in the remineralization of organic matter in the
ocean interior. From the apparent oxygen utilization (AOU) of the
interior waters and the prescribed ratio of O/P/C of particulate
organic matter (POM) used in the model, we computed the amount of
phosphate and DIC produced from the remineralization of sinking
POM. In the LGM simulation, the remineralized phosphate was much less
than the HC simulation, with the total amount of remineralized
phosphate being only 40 % of the HC simulation. While the
amount of remineralized phosphate declined in the LGM simulation, the
phosphate concentrations in the deep water (below 1500 m) were
greater than the HC simulation (Fig. ). The increase in
phosphate in the LGM simulation reflects the increased subduction of
phosphate-rich AABW which transfers phosphate into the abyss and
reduces the phosphate concentrations in the upper ocean
(Fig. ). Paleo cadmium and δ13C data from the
tropical and North Atlantic show increased nutrient levels in the
abyss, but reduced levels above 2000 m during the LGM
. This
redistribution of nutrients in the LGM was consistent with our
simulated LGM phosphate distribution.
The LGM simulation relies on increased AABW subduction to transfer nutrients from the upper ocean into the abyss and elevate DIC, phosphate, alkalinity and oxygen concentrations below 2000 m.
Associated with the transfer of phosphate from the upper ocean into the abyss in the LGM simulation there was a global reduction in the export of POC from the euphotic zone (Table ).
The LGM simulation (Fig. ) was consistent with paleonutrient data which suggests export production in the Antarctic Zone of the Southern Ocean was lower during the LGM .
Similarly, the simulated decline in North Pacific export production (Fig. ) was consistent with paleonutrient data from the Subarctic Northwest Pacific, which indicates less export production during glacial periods .
The simulated reduction of export production in the Subarctic Northwest Pacific and North Atlantic, in regions not covered by sea ice, is caused by reduced phosphate availability (Fig. ).
The equatorial upwelling region of the east Pacific and east Atlantic were two other regions in the LGM simulation where the surface phosphate concentration declined and there was a reduction in export production in the LGM.
The increased AABW subduction in the LGM simulation supported a greater vertical gradient in DIC, alkalinity and phosphate, while the reduction in POC and particulate inorganic carbon (PIC) export in the LGM simulation partially reduced these vertical gradients.
Comparing the simulated DIC and phosphate concentrations in the ocean interior in the HC to the LGM showed that remineralization of POC in the LGM declined by 60 %.
The decline in remineralized DIC and phosphate (60 %) exceeded the decline in POC export from the euphotic zone (44 %), which implied the LGM circulation had increased resupply of remineralized phosphate and carbon back to the upper ocean.
In general, outside of the Southern Ocean the surface water became phosphate limited.
Carbon budget
As discussed in the introduction, the strength of deep water
formation, the suppression of air–sea gas exchange due to sea-ice
expansion in the Southern Ocean and the increase in POC export, have
all been suggested as playing possible roles in reducing atmospheric
CO2 during glacial periods .
Our LGM simulation had the first two features, and in the following discussion we also explore how increasing POC export could impact the LGM ocean carbon content.
In our coupled simulations, we fixed atmospheric CO2 concentrations in the LGM and HC, and then simulated the carbon content of the ocean.
The ocean carbon content in the LGM simulation was 604 Pg C less than the HC simulation (Table ).
The changes in the simulated LGM ocean carbon reflect changes in the ocean state and the reduction in the atmospheric CO2 level in the LGM.
To separate the impact of changes in the ocean state from changes in atmospheric CO2 we used a LGM ocean-only simulation where the atmospheric CO2 was set at 280 ppm (OLGM-2).
By increasing the atmospheric CO2 in the LGM climate, the carbon content in the ocean was 1130 Pg C greater than the HC simulation (OLGM-2, Table ), which was 1730 Pg C greater than the standard LGM simulation (OLGM-1).
Thus, the ocean state during the LGM was preconditioned to increase the carbon content of the ocean, but this preconditioning was not enough to counter the effect of reduced atmosphere CO2 during the LGM, which caused a loss of carbon from the upper ocean (Fig. ).
While the LGM ocean was conditioned to store more carbon in the deep ocean, changes in BGC cycles appear necessary to get the carbon into the ocean.
To explore potential biogeochemical processes that could increase the carbon storage in the LGM ocean we considered three plausible modifications to the standard BGC formulation: (1) increased POC export, (2) increased depth of POC remineralization, (3) reduced PIC export.
The resulting impacts on the carbon storage in the ocean are summarized in Table and discussed in the next few paragraphs.
To increase POC export, the scaling factor for POC export was increased by 10× (see Eq. in Appendix). The motivation for increased POC export in the LGM is the increased iron supply to the ocean, which relieves iron limitation in regions like the Southern Ocean and increases POC export .
Although the POC export scale factor was increased by 10×, the POC export increase was much smaller (1.3×) because in the LGM simulation most of the oceans outside of the Southern Ocean were phosphate limited (Fig. a) and weakly responded to the increased scaling factor for POC export; the increase in export production mostly occurred on the northern boundary of the Southern Ocean (Fig. d).
By increasing the POC export scaling factor, POC export increased by 1.5 to 5.9 Pg C y-1, which was still less than the HC value of 8.1 Pg C y-1, and the carbon content of the ocean increased by 260 Pg C (Table ).
The POC remineralization depth was increased by changing the power law exponent from -0.9 to -0.7 (see Eq. , Appendix). The motivation for such a change is that a colder ocean would reduce the rate of bacterial remineralization of POC .
The change in remineralization increased the fraction of POC sinking through 1000 m from 12.5 to 20 %.
Increasing the depth of POC remineralization reduced POC export from the photic zone by 1.2 Pg C y-1 (OLGM-5, Table ).
The deeper remineralization depth for POC increased the carbon content of the ocean by 170 Pg C.
The sensitivity of the carbon content of the ocean in the LGM to changes in the depth of POC remineralization was much less than suggested by the simulation of .
The reduced sensitivity of carbon storage in our LGM simulation to the depth of POC remineralization reflects the conclusion that the steady-state response is much smaller than the transient response .
Finally, the PIC export was set to zero, which was motivated by the strong relationship between calcification and temperature , i.e. in a cooler ocean less PIC would be produced.
With no PIC export, the solubility of CO2 in the surface water increased and enabled the ocean to store an additional 260 Pg C.
Independently, none of the BGC modifications comes close to increasing the LGM storage of carbon in the ocean above the HC simulation (OLGM-3, 4 and 5, Table ). By employing all three BGC modifications discussed above in one LGM simulation (OLGM-6), the ocean stored 830 Pg C more than the standard LGM simulation (OLGM-1; Table ).
With all three BGC modifications, the increase in the ocean carbon storage in the LGM simulation was 240 Pg C more than the HC simulation and now comes within the bounds of the combined reduction in atmospheric carbon and terrestrial carbon reported by of 520±400 Pg C (Table ).
Combined, the three BGC modifications are sufficient to explain the change in land and atmospheric carbon between the HC and LGM.
The consequences of these BGC modifications will next be assessed by looking at how they change the stability of calcium carbonate, export production and dissolved oxygen.
However, before exploring these changes we first consider other biogeochemical processes that could account for the additional carbon storage in the ocean.
The carbon stored in the ocean could be increased by increasing alkalinity input from the land due to terrestrial weathering. The 1120±400 Pg C would require an alkalinity increase of 112±40 Pmol Eq (10 % global increase). During the LGM, a drier climate, extensive ice sheets and exposed shelf areas would change terrestrial weathering but such changes are unlikely to have increased alkalinity input from the HC . Therefore, increased alkalinity input into the ocean seems an unlikely explanation by itself for increasing the carbon stored in the ocean.
Further, the additional alkalinity would increase the stability of calcium carbonate in the ocean, which we show in the next section was already too deep in the standard LGM simulation (OLGM-1, Fig. a).
Greater cooling of the ocean in the mid and low latitudes is another potential mechanism to store more carbon in the LGM ocean. Given that our SST cooling was near the paleo-reconstruction maximum cooling this mechanism seems unlikely for storing more carbon in the
ocean in our LGM simulation.
Changes in the large-scale ocean circulation could provide another way
to increase carbon storage. In our LGM simulation, the increased
subduction of AABW dominated the increased carbon content of the deep
ocean as shown by the reduction in the regenerated phosphate
concentrations in the deep ocean. Reducing AABW subduction would
reduce ocean carbon storage by reducing the amount of high-DIC water
subducted into the deep ocean. Therefore, increased subduction of
AABW could increase the solubility pump and ocean carbon storage but,
as explored by , this did not greatly increase
the carbon stored in the ocean. Further, our LGM simulation already
has a much greater increase in AABW than other LGM simulations
e.g.).
Another way of removing carbon from the land–ocean–atmosphere in the
LGM is the net burial of organic carbon in ocean sediments. The
required net organic carbon burial using our LGM simulation and
reported atmosphere and land carbon changes
(1120±400 Pg C) is not large given the long timescale
involved in the drawdown of atmospheric CO2 from the
interglacial to glacial (100 000 years). The burial of
organic carbon would cause a loss of phosphate from the
ocean. Assuming a C/P ratio of 106 this would equate to a loss of
10.5±4 Gmol P y-1. Present estimates of the
burial rate of phosphorus in the ocean sediments is about
320 Gmol P y-1 , which is
an order of magnitude greater than the required net phosphate loss
(10.5 Gmol P y-1). The small additional loss of
phosphate and carbon needed to close the LGM carbon budget requires
either a small increase in organic matter burial or a small decrease
in the continental supply of organic matter to the ocean that is
remineralized in the ocean. Either is a feasible mechanism to account
for the missing carbon in the interglacial–glacial drawdown in
atmospheric CO2.
Importantly, burying carbon in ocean sediments would not change the simulated lysocline depth of the LGM ocean simulation produced by standard BGC formulation (OLGM-1; Fig. ).
The LGM ocean is preconditiond to store carbon, but just the physical changes alone are insufficient to get the required carbon into the ocean. Individually, the BGC modifications also do not get sufficient carbon into the ocean. With all three BGC modifications (increased POC export, reduce POC remineralization and no PIC export), sufficient carbon is stored in the ocean and in the next section we will use the modifications to calcium carbon stability, export production and oxygen levels in the ocean to assess these changes.
Calcium carbonate stability
In comparing the stability of calcium carbonate in the ocean between
the LGM and HC states two key factors need to be considered – the
large reduction in atmospheric CO2 and the large cooling of
the ocean in the LGM. The former factor increases the stability of
calcium carbonate while the latter factor, through its impact on
carbon speciation, would reduce the stability of calcium carbonate.
To represent the surface ocean carbon chemistry we use the aragonite saturation state (Ω), which is a useful indicator of calcification rates .
At the surface, the simulated annual averaged Ω in the LGM was slightly less than the HC between 40∘ S and 40∘ N, but much greater poleward of these two latitudes (Fig. ).
The BGC modifications only slightly alter the surface Ω value and for the simulations with all the BGC modifications (OLGM-6) there is a slight increase in surface aragonite saturation state (Fig. ).
Interestingly, the simulated Ω=3.25 isoline, the value at present used to define the location of viable coral reef conditions , was nearly unchanged between the LGM and HC simulations.
Recent sonar and coring in the southern portion of the Great Barrier Reef detected the presence of drowned coral reefs in the LGM that were as far south as the present-day Great Barrier Reef.
Such observations are consistent with our simulation that the southern extent of coral reefs along the east coast of Australia was not dramatically different between the LGM and HC, which is consistent with the modified BGC simulation (OLGM-6).
The changes in the carbon chemistry also extend into the ocean interior.
The simulated depth of the aragonite lysocline in the LGM simulation (OLGM-1) was much deeper than the HC simulation (Fig. ).
Outside of the tropical regions, the OLGM-1 simulation's entire water column was super-saturated for aragonite, which is unrealistic.
Hence, processes that increase alkalinity input into the ocean and increase the carbon burial in the sediments do not provide a means of shoaling the aragonite lysocline.
Individually, the three BGC modifications still have very deep lysoclines: the Southern Ocean and North Pacific Ocean lysoclines are still near the ocean bottom and more than 1000 m deeper than the HC simulation (Fig. ).
However, in the experiment with the three modified BGC processes employed to increase the carbon storage in the ocean (OLGM-6), the depth of the lysocline was much shallower than the OLGM-1 simulation (Fig. d).
For this simulation (OLGM-6), the lysocline depth in the Pacific Ocean was deeper than the HC simulations but it was shallower in the North Atlantic and along the northern boundary of the Atlantic and Indian sectors of the Southern Ocean (Fig. ).
There is good evidence that the mean position of the lysocline did not change much between the LGM and the HC, with the average lysocline depth being less than 1 km deeper in the LGM than the HC in the North Pacific and Southern Ocean , which is most consistent with the LGM simulation with the three BGC modifications (OLGM-6).
The paleo-reconstructions also suggest the lysocline shoaled in the Atlantic and deepened in the Pacific , consistent with the OLGM-6 simulation.
Further, recent observations from seamounts just south of Tasmania suggest deep cold water coral communities during the LGM occured slightly deeper than at present , consistent with a deepening of the lysocline during the LGM.
To have a LGM simulation agree with the paleo observations for south of Tasmania, all three BGC modifications are required (Fig. ).
Export of particulate organic carbon
In the standard LGM simulation (OLGM-1), the export of POC from the upper ocean generally declines everywhere relative to the HC simulation except on the southern boundary of the eastern equatorial upwelling region of the Pacific (Fig. b).
The setting of the export of PIC to zero had no impact on POC export, while deepening the POC remineralization increased the transfer of carbon and nutrients into the ocean interior and the POC export was generally reduced everywhere relative to the HC (Fig. d).
With the POC scaling factor increased, the total POC export was still less than the HC (Table ), but now the Southern Ocean and small areas of the North Atlantic and Northwest Pacific have increased POC export compared to the HC.
The simulation with all three BGC modifications shows a very similar pattern of change as the simulation with just increased POC export (Fig. a and c).
In this simulation, POC export increased in the Southern Ocean and declined in the subtropics and tropics (Fig. a).
Such changes to POC export are consistent with paleo-reconstructions of POC differences between the LGM and HC .
The increase in POC export in the Southern Ocean reduces the phosphate concentrations in the subducted Antarctic Intermediate Waters, which has been shown to be critical to setting the magnitude of the export production outside of the Southern Ocean .
Dissolved oxygen
The greater subduction of AABW in the LGM dramatically increased the oxygen levels in the ocean and coupled with the decreased remineralization of POC in the ocean there was a large increase in oxygen concentrations in the ocean interior during the standard LGM simulation relative to the HC simulation (55 % increase).
In the standard LGM simulation, oxygen levels increased in both the intermediate and deep water (Fig. ) and the total volume of anoxic water was less than the HC simulation.
With the BGC modifications to POC export and remineralization, the oxygen levels in the intermediate water decrease in the Southern Ocean and increase in the North and Equatorial Pacific (Fig. a, b and d).
In the deep ocean, the large increase in the simulated LGM oxygen levels only start to be eroded in the simulation where POC export scaling was increased to reflect increased iron supply to the surface ocean (Fig. ).
However, it is only in the simulation with increased POC export and deeper remineralization that the oxygen levels in the deep Southern Ocean start to decline compared to the HC simulation (Fig. a).
The consequence of reduced anoxic water in the LGM should be reduced dentrification in these regions.
Recent analysis of sedimentary δ15N records suggest that global aggregate rates of N fixation and water column denitrification rates over the past 200 000 years were less active during glacial periods and more active during interglacial periods . This is consistent with our simulated decreases in the extent of water column anoxia during the LGM, which occurred in all LGM simulations.
As and suggested, changes in ventilation and water characteristics are more important factors to the variability of water column denitrification than regional changes in export production, because the decrease in anoxia is not due to reduced POC export but rather due to increased subduction of oxygen-rich water.
Such a conclusion is consistent with our LGM simulations where the volume of anoxic water showed little difference among the LGM simulations with varying BGC formulations. For example, the simulation with all three BGC modifications (OLGM-6) had only 5 % less volume of anoxic water than the standard LGM simulation (OLGM-1).
While the impact of the three BGC modifications on the volume of anoxic water was small, these modifications did have a significant impact on oxygen levels in the intermediate and deep water.
Paleo-reconstructions suggest oxygen levels in the upper 1500 m were greater in the LGM than HC .
That all LGM simulations show this behaviour again demonstrates that oxygen levels in the upper ocean are primarily set by the ocean dynamics consistent with the change in the volume of anoxic water.
In contrast, the deep ocean trace metal records generally suggest a decline in LGM oxygen levels relative to the HC .
Such a decline in LGM oxygen requires oxygen consumption in the deep ocean to increase by an amount that exceeds the temperature-driven increase of oxygen solubility.
Such a behaviour is only achieved when all three BGC modifications are used in the LGM simulation.