The simulated climate of the Last Glacial Maximum and insights into the global carbon cycle

. The ocean’s ability to store large quantities of carbon, combined with the millennial longevity over which this reservoir is overturned, has implicated the ocean as a key driver of glacial-interglacial climates. However, the combination of processes that cause an accumulation of carbon within the ocean during glacial periods is still under debate. Here we present simulations of the Last Glacial Maximum (LGM) using the CSIRO Mk3L-COAL Earth System Model to test the contribution of physical and biogeochemical processes to ocean carbon storage. For the LGM simulation, we ﬁnd a signiﬁcant global cooling 5 of the surface ocean (3.2 ◦ C) and the expansion of both minimum (Northern Hemisphere: 105 %; Southern Hemisphere: 225 %) and maximum (Northern Hemisphere: 145 %; Southern Hemisphere: 120 %) sea ice cover broadly consistent with proxy reconstructions. Within the ocean, a signiﬁcant reorganisation of the large-scale circulation and biogeochemical ﬁelds occurs. The LGM simulation stores an additional 322 Pg C in the deep ocean relative to the Pre-Industrial (PI) simulation, particularly due to a strengthening in Antarctic Bottom Water circulation. However, 839 Pg C is lost from the upper ocean via equilibration 10 with a lower atmospheric CO 2 concentration, causing a net loss of 517 Pg C relative to the PI simulation. The LGM deep ocean also experiences an oxygenation ( > 100 mmol O 2 m − 3 ) and deepening of the aragonite saturation depth ( > 2,000 m deeper) at odds with proxy reconstructions. Hence, physical changes cannot in isolation produce plausible biogeochemistry nor the required drawdown of atmospheric CO 2 of 80-100 ppm at the LGM. With modiﬁcations to key biogeochemical processes, which include an increased export of organic matter due to a simulated release from iron limitation, a deepening of remineralisation 15 and decreased inorganic carbon export driven by cooler temperatures, we ﬁnd that the carbon content in the glacial oceanic reservoir can be increased (326 Pg C) to a level that is sufﬁcient to explain the reduction in atmospheric and terrestrial carbon at the LGM (520 ± 400 Pg C). These modiﬁcations also go some way to reconcile simulated export production, aragonite saturation state and oxygen ﬁelds with those that have been reconstructed by proxy measurements, thereby implicating changes in ocean biogeochemistry as an essential driver of the climate system. for the analysis undertaken in this work. The matplotlib package (Hunter, 2007), and Cartopy packages (http://scitools.org.uk/), cmocean package et producing visualisations.

In numerous other regions, however, productivity appears to have been reduced during glacial climates. The affected regions include waters south of the Antarctic Polar Front (Francois et al., 1997;Jaccard et al., 2013), the North Pacific (Crusius et al., 2004;Jaccard et al., 2005;Kohfeld and Chase, 2011;Ortiz et al., 2004), tropical Indian Ocean (Singh et al., 2011) and the Equatorial Pacific (Costa et al., 2016;Herguera, 2000;Loubere et al., 2007). A weaker export production in these regions would have offset a strengthened biological pump in the Sub-Antarctic, thereby weakening the ability of the ocean to store 5 carbon at the LGM. Whether the strengthening of the biological pump in the glacial sub-Antarctic was able to outweigh losses in productivity in other regions requires further testing. This has led some authors to look for alternative biological mechanisms, notably temperature-dependent remineralisation (Chikamoto et al., 2012;Matsumoto, 2007;Menviel et al., 2012) and an altered CaCO 3 :C organic export production ratio (Archer and Maier-Reimer, 1994;Lerman and Mackenzie, 2005;Sigman et al., 1998), to explain the net flux of carbon into the ocean. 10 Therefore, numerous physical and biogeochemical changes have been associated with a glacial ocean and all have been identified in some respect as important drivers of the carbon cycle. Now, recent insights into the distributions of dissolved oxygen  and carbonate species (Yu et al., 2014) within the glacial ocean provide new opportunities to identify which combination of physical and biogeochemical changes could have realistically sequestered carbon within the ocean at the 15 LGM. Following the experiments conducted by Tagliabue et al. (2009), we use an Earth System Model with attached biogeochemistry, CSIRO Mk3L-COAL, to test current theories against these new insights. Using our simulated LGM ocean state, we provide a new perspective on the mechanisms responsible for the 80-100 ppm drawdown in atmospheric CO 2 during glacial cycles and demonstrate the importance of marine biogeochemistry to global climate. 20

Model and experiments
The model simulations were performed using the CSIRO Mk3L climate system model version 1.2 (Phipps et al., 20111.2 (Phipps et al., , 2012, which includes components that describe the atmosphere, land, sea ice and ocean. The horizontal resolution of the atmosphere, land and sea ice models are 5.6 • ×3.2 • in the longitudinal and latitudinal dimensions, respectively, with 18 vertical levels. The ocean model has a horizontal resolution of 2.8 • ×1.6 • with 21 vertical levels. For this study, we conduct simulations using both 25 the full climate system model and the stand-alone ocean model. Two fully coupled model experiments were undertaken to simulate the Pre-Industrial (Cpl-PI) and Last Glacial Maximum (Cpl-LGM) climates. The Cpl-PI climate was obtained by forcing the model with an atmospheric CO 2 concentration of 280 ppm and by prescribing 1950 CE values for the orbital parameters. This experiment was integrated for a total of 10,000 years 30 (Phipps et al., 2013). The Cpl-LGM simulation followed the protocol developed by Phase III of the Palaeoclimate Modelling Intercomparison Project (PMIP3), with the exception that no changes were made to terrestrial topography, oceanic bathymetry or the positions of the coastlines. The atmospheric CO 2 equivalent concentration was set to 167 ppm, providing a radiative forcing equivalent to the specified reductions in the atmospheric concentrations of CO 2 , CH 4 and N 2 O from 280 ppm/760 ppb/270 ppb for pre-industrial simulations to 185 ppm/350 ppb/200 ppb for LGM simulations. The orbital parameters were set to values for 21 ka BP. The Cpl-LGM simulation was initialised from the state of the Cpl-PI simulation at the end of model year 100. The model was then integrated for a total of 3,900 model years, until it had reached quasi-equilibrium. Over this integration the ocean experienced a slow drift in whole-ocean salinity so that 0.5 psu was added, reflecting the coupling between a cooler 5 atmosphere and the ocean.
With the Cpl-LGM climate state a suite of different ocean biogeochemical simulations were made with slightly different parameterisations to explore the effect on the carbon cycle (Table 1). These experiments utilised Mk3L-COAL (Carbon-Ocean-Atmosphere-Land), an enhanced version of the Mk3L climate system model which includes biogeochemical modules embed- 10 ded within the ocean, atmosphere and terrestrial models. For a description of the ocean biogeochemistry the reader is directed towards Appendix A of Matear and Lenton (2014) and the experiments of Duteil et al. (2012).
A total of 8 ocean-only simulations with on-line biogeochemistry were undertaken. All experiments were forced by key boundary conditions (wind stresses, temperature, salinity, incident radiation, sea ice and atmospheric pressure at sea level), which O-LGM3. The scaling factor (S O npp ) was increased by a factor of 10 (Eq. (1)) to increase the export of Particulate Organic 25 Carbon (POC) from the surface ocean, and therefore strengthen the biological carbon pump. Increasing POC export in the LGM ocean was motivated by an enhanced delivery of iron to the surface ocean via aeolian dust at the LGM (Delmonte et al., 2004;Kawahata et al., 2000;Lambert et al., 2012;Martin, 1990;Martínez-Garcia et al., 2009;Martinez-Garcia et al., 2014;Watson et al., 2000).
LGM4. The POC remineralisation depth was increased by changing the power law exponent (Rem pwl ) in Equation (2) from -0.9 to -0.7, which replicated a bulk shift of POC from the upper to the deep ocean. The motivation for increasing the amount of POC that reaches deeper levels is the expectation that a cooler ocean would reduce the rate of bacterial remineralisation of POC in the upper ocean (Rivkin and Legendre, 2001;Matsumoto, 2007). This change increased the simulated POC that reaches the 1,000 m depth level from 12.5 % to 20 %.
O-LGM5. Export production of Particulate Inorganic Carbon (PIC) was turned off by setting the rain ratio (R P IC ) of PIC:POC to zero in Equation (3). A reduction in PIC export at the LGM would increase the carbon content of the ocean, as marine calcifiers release CO 2 during formation of their shells and enhanced carbon loss via outgassing. The motivation for reducing 5 PIC export in the glacial ocean is also temperature related, as there is a strong positive relationship between temperature and calcification (Lough and Barnes, 2000).
O-LGM6. All three modifications to ocean biogeochemistry were employed. All ocean-only simulations were integrated for 10,000 years to ensure that the ocean carbon cycle reached a steady state. 10 To assess whether the behaviour of the biogeochemical tracers within the coupled model differed from those in the oceanonly model, we ran the coupled model with online ocean biogeochemistry for a further 1,000 years using the steady-state biogeochemical fields from the ocean-only experiments. This assessment was made using both the PI and LGM climates. For key diagnostics, such as the meridional overturning circulation, ocean carbon content and global export production, the be- 15 haviour of the ocean-only simulation differed by less than 1 % from the coupled simulations. Given the computational speed of the ocean-only model, these experiments provide an ideal platform to test the sensitivity of the ocean biogeochemical fields to the parameterisations used in the biogeochemical model.
3 Results and discussion 20 In the following, we first discuss the simulated physical changes to the ocean observed between the Cpl-PI and Cpl-LGM simulations. Second, we discuss how the ocean biogeochemical fields differed between the O-PI1 and O-LGM1 simulations, which were forced with the output of the coupled simulations. Finally, we explore how modifying biogeochemical parameterisations alters the biogeochemistry, including changes to carbon storage, export production, aragonite saturation state and dissolved oxygen, that reconcile our simulated glacial ocean with what is considered realistic according to palaeoclimate proxy records. The simulated change in SST between the Cpl-PI and the Cpl-LGM simulations shows a similar magnitude and spatial structure to proxy reconstructions and prior modelling studies, with greatest cooling in the equatorial oceans, high latitudes and 30 5 Clim. Past Discuss., doi:10.5194/cp-2016-73, 2016 Manuscript under review for journal Clim. Past Published: 11 July 2016 c Author(s) 2016. CC-BY 3.0 License. eastern boundary currents, and the least cooling in the subtropics and western boundary current regions ( Fig. 1; Table 2). The global SST mean of the Cpl-LGM was 3.2 • C cooler than the Cpl-PI. This change falls within the range of estimates (∼2-4 • C) produced by other climate models (Alder and Hostetler, 2015;Annan and Hargreaves, 2013;Braconnot et al., 2007;Ganopolski et al., 1998;Kitoh et al., 2001;Shin et al., 2003;Smith and Gregory, 2012), but sits towards the cooler limits of previous multiproxy SST reconstructions that estimate a change of 2 ± 1.8 • C (Ballantyne et al., 2005;McIntyre et al., 5 1976;Waelbroeck et al., 2009). However, a recent reanalysis of the proxy data presented by Waelbroeck et al. (2009) showed past estimates may have underestimated cooling by as much as 50 % (Ho and Laepple, 2015). This finding reconciles some disagreement between climate models and palaeoproxies, and places our simulated cooling of 3.2 • C well within the bounds of uncertainty in reconstructions. 10 Regionally, the greatest cooling took place in the high latitudes and in the Equatorial Pacific, where temperatures were in excess of 4 • C cooler than the Cpl-PI climate. Meanwhile, the Western Pacific Warm Pool, subtropical gyres and western boundary currents cooled less (0.5-3.0 • C). Again, proxy (Bostock et al., 2013;Gersonde et al., 2003Gersonde et al., , 2005Kaiser et al., 2005;Kucera et al., 2005;Lamy et al., 2004;Lüer et al., 2009;Martínez-Garcia et al., 2009;Waelbroeck et al., 2009) and climate model simulations (Alder and Hostetler, 2015;Hargreaves, 2013, 2015;Shin et al., 2003) are consistent with 15 both the magnitude and spatial pattern of cooling. Enhanced cooling in the high latitudes and in the eastern boundary currents generated strong zonal and meridional temperature gradients relative to Cpl-PI SST. There is a consistent regional pattern to SST cooling in the LGM emerging from proxy and model simulations (Annan and Hargreaves, 2013;Braconnot et al., 2007;Felis et al., 2014) that is broadly consistent with our simulated cooling. 20 Where there is still large uncertainty in SST change at the LGM is in the tropical ocean (see Annan and Hargreaves, 2015, for a review). The Cpl-LGM cooling of 3.3 • C across the tropical ocean (15 • S -15 • N) is greater than other simulations (Annan and Hargreaves, 2013;Ballantyne et al., 2005;Braconnot et al., 2007;Ganopolski et al., 1998;Kitoh et al., 2001;Smith and Gregory, 2012), but falls well within the −5.1 to −2.17 • C estimated by Ho and Laepple (2015). Regionally, climate models and proxies both agree that cooling in the tropical Atlantic Ocean probably exceeded cooling in the tropical Pacific and Indian

Sea ice extent
While true estimates of sea ice coverage in the PI climate can only be inferred from whaling records, our Cpl-PI sea ice extents are consistent with estimates made using satellite measurements during the 1979-1987 period (Gloersen et al., 1993, Table 2).
These measurements represent the first global estimates of sea ice coverage, and although some evidence indicates that sea ice has declined by 20 % since the 1950's (Curran et al., 2003), the strong agreement between the Cpl-PI sea ice fields and the observations of Gloersen et al. (1993) provide a good benchmark for assessing LGM sea ice changes.
Associated with cooler SSTs, sea ice coverage (areas with sea ice concentration at or in excess of 15 %) was greatly ex-5 panded in the Cpl-LGM for both hemispheres relative to the Cpl-PI (Fig. 2, Table 2). In the Southern Hemisphere, total sea ice coverage increased by ∼120 % and ∼225 % at its seasonal maximum and minimum, respectively, relative to the Cpl-PI. In the Northern Hemisphere, total sea ice coverage increased by ∼145 % and ∼105 % at its seasonal maximum and minimum, respectively, relative to the Cpl-PI. These increases correspond to equatorward expansions of the sea ice field of between 5-10 • Ocean (Gersonde et al., 2005, Fig. 2). This magnitude of growth in the Atlantic and Indian sectors has been tested and largely supported by a few subsequent studies (Collins et al., 2012;Xiao et al., 2016), and is consistent with our Cpl-LGM sea ice field. In the Pacific sector, however, the simulated maximum sea ice edge extends well equatorward of the 57 • S suggested by Gersonde et al. (2005) (Fig. 2). By comparing the coverage of sea ice in the Southern Hemisphere of the Cpl-LGM (∼46×10 6 20 km 2 ) with that estimated by Gersonde et al. (2005) (∼39×10 6 km 2 ), we can attribute the simulated excess of sea ice in the glacial Southern Ocean to a possible overestimate in the Pacific sector.
In the North Atlantic, perennial sea ice cover was present in the Greenland Sea and Fram Strait during the LGM (Müller et al., 2009;Telesiński et al., 2014). There is also evidence that winter sea ice reached south of Iceland to fill much of the 25 Labrador Sea (Pflaumann et al., 2003) and extended along the eastern Canadian margin (De Vernal et al., 2000, 2005. Meanwhile, the central and eastern parts of the subpolar North Atlantic are thought to have been largely ice-free (Pflaumann et al., 2003). In the North Pacific, proxy reconstructions suggest strong cover in the Okhotsk Sea (Sakamoto et al., 2005;Nürnberg and Tiedemann, 2004;Yamazaki et al., 2013), the Japan Sea (Ikehara, 2003) and the western Bering Sea (Ovsepyan et al., 2013;Riethdorf et al., 2013b, a), with seasonally ice-free conditions in the central west (Jaccard et al., 2005). Sea ice reconstructions 30 are lacking in the central and eastern North Pacific Ocean, but climate models that completed the PMIP3 LGM experiment simulate stronger sea ice presence in the western margins of the Northern Hemisphere basins (Fig. 2). The Cpl-LGM sea ice in this study is broadly consistent with the palaeo evidence in the North Atlantic, but an intense and year-round cover developed over the central North Pacific that contrasts directly with the findings of Jaccard et al. (2005), whom argued for ice-free conditions during the summer. Furthermore, the expansion of Cpl-LGM sea ice is greater than other PMIP3 climate models, which places our simulated sea ice field towards the higher bounds predicted by climate system models for the LGM.

Meridional overturning circulation
The changes observed in the surface ocean within the Cpl-LGM climate were accompanied by changes in the global meridional 5 overturning circulation ( Fig. 3; Table 2). The rate of Antarctic Bottom Water (AABW) formation, defined here as the strongest negative cell south of 60 • S, was greater in the Cpl-LGM than the Cpl-PI. AABW formation rates approached 25 Sv in the Cpl-LGM, about double that of the Cpl-PI. The greater formation rate increased the transport of AABW out of the Southern Ocean (northwards of 50 • S) by roughly a factor of four from 17,440 to 71,664 Sv. This indicates that the increase in AABW formation was also associated with an increase in the proportion of this water mass that was carried out of the Southern Ocean. The changes to AABW and NADW circulation were conducive to the development of a global overturning circulation domi-20 nated by dense water from the Southern Ocean. An altered global overturning circulation is now a widely recognised component of glacial climate states across the Pleistocene (Broecker, 2013). In fact, several authors now attribute at least half of the glacialinterglacial atmospheric CO 2 difference to circulation changes in the ocean (Broecker et al., 2015;Kohfeld et al., 2005;Sigman et al., 2010). Theoretical, proxy and model-based research is now beginning to converge on the large-scale characteristics of the glacial ocean circulation, where AABW was more dominant due to a combination of expanded sea ice, enhanced brine 25 rejection causing denser bottom waters, and reduced diapycnal mixing (Adkins, 2013;De Boer and Hogg, 2014;Ferrari et al., 2014;Skinner et al., 2010;Watson and Naveira Garabato, 2006). The neutral density boundary between a northward flowing AABW and a southward flowing NADW was also shoaled substantially at the LGM in comparison with the current climate.
Numerous palaeonutrient tracers support the presence of AABW within the deep North Atlantic Ocean at the LGM (Curry and Oppo, 2005;Duplessy et al., 1988;Keigwin, 2004;Marchitto and Broecker, 2006;Oliver et al., 2010). The maximum depth 30 of NADW flow was displaced to above 2,000 m as a direct result, and the shoaling of NADW facilitated the development of a saltier, more stratified glacial deep ocean (Adkins, 2013). Ferrari et al. (2014) have interpreted these changes as inextricably linked to the expansion of sea ice in the Southern Ocean, which caused a greater proportion of Circumpolar Deep Water to rise into a zone of negative buoyancy flux and thereby produce greater quantities of denser AABW. Imposing only the orbital parameters and atmospheric radiative forcing of the LGM, our Cpl-LGM simulation was able to reproduce these features of a glacial ocean circulation.

LGM climate: biogeochemical fields
The physical changes in the ocean between the Cpl-PI and the Cpl-LGM, as described above, caused significant changes in 5 ocean biogeochemistry within the ocean-only simulations (Table 3). To assist in the discussion of the large-scale biogeochemical changes we divide the upper and the deep ocean based on the 2,000 m depth. This approach also allows for more clearly distinguishing between changes to the global overturning circulation, air-sea exchange and biological processes on the biogeochemical fields.

Carbon
For the ocean-only experiment O-LGM1, the carbon content of the ocean was 517 Pg C less than the O-PI1 experiment ( Fig. 4; Table 3). This change in carbon content reflects the combined effects of an altered ocean physics, which increased carbon in the deep ocean by 322 Pg C, and a lower atmospheric partial pressure, which caused 839 Pg C of carbon to be lost from the upper ocean. The air-sea gas exchange scheme in CSIRO Mk3L-COAL is based on the parameterisation given by Wanninkhof 15 (1992), and therefore regardless of the increase in solubility in the upper ocean that is caused by cooling, the decrease in partial pressure in the atmosphere ensured that the upper ocean DIC equilibrated at a lower concentration.
To quantify the carbon gain caused by a cooler ocean with expanded sea ice and an altered overturning circulation, the oceanonly experiment, O-LGM2, forced by the Cpl-LGM output was run with a PI atmospheric CO 2 concentration of 280 ppm. The 20 ocean carbon content of O-LGM2 was increased by 1,127 Pg C relative to O-PI1. The gain of carbon in O-LGM2 confirmed that the glacial ocean was indeed conducive to partitioning greater quantities of carbon in the deep ocean, which can largely be attributed to the increased formation rate of AABW. AABW has recently been identified as eliciting the greatest response in atmospheric CO 2 of all major ocean water masses (Menviel et al., 2015). However, total carbon content decreased in ex-

Nutrients and export production
Phosphate (PO 4 ) concentrations in experiment O-LGM1 also declined in the upper ocean and increased in the deep ocean ( Fig. 5). Like DIC, this reorganisation was driven by a strengthened AABW circulation cell and is consistent with proxy reconstructions. Cadmium and δ 13 C measurements from the Atlantic Ocean show increased nutrient concentrations in the deep ocean, but reduced levels above 2,000 m at the LGM (Boyle, 1992;Gebbie, 2014;Marchitto and Broecker, 2006;Tagliabue 5 et al., 2009).
A direct consequence of the redistribution of PO 4 was the reduction in the production of particulate organic matter across many regions of the O-LGM1 ocean (Fig. 5). With the exception of the South Pacific and isolated areas in the subtropics, export production in the O-LGM1 experiment decreased relative to the O-PI1 experiment, so that global export production 10 was 56 % of O-PI1. The global reduction was also illustrated by a decrease in regenerated carbon (DIC org ), which indicates a weakened biological carbon pump (Table 3). The magnitude of the reduction in export production for the O-LGM1 experiment sits outside the range of 76-83 % estimated using oxygen isotopic measurements (Blunier et al., 2002) and in the opposite direction to the conclusions of Galbraith and Jaccard (2015), who argued for a net strengthening of the glacial biological pump at the LGM. The strong reduction of export production can be attributed, in part, to a large decrease in export production from 15 the sub-Antarctic zone. This feature is in direct conflict with palaeoproductivity proxies in the Atlantic and Indian sectors of the sub-Antarctic Ocean (Anderson et al., 2002Chase et al., 2001;Jaccard et al., 2013;Nürnberg et al., 1997), and some parts of the Pacific sector (Bradtmiller et al., 2009;Lamy et al., 2014). Outside of the Southern Ocean, the reduction in export production in the O-LGM1 experiment is consistent with palaeoproductivity evidence (see Introduction).  (Fig. 8). There is good evidence that the mean position of the lysocline was not appreciably different at the LGM as compared to the Late Holocene (Yu et al., 2014). This information places the simulated lysocline (Ω = 1) of O-LGM1 as unrealistic. ocean by ∼80 and ∼120 mmol m −3 , respectively, which constitutes a global increase of 55 % ( Fig. 9; Table 3).
The increase in dissolved oxygen in O-LGM1 was considerable, but agreed well with proxy reconstructions for the upper ocean. The oxygen-poor intermediate waters of the western North Pacific (Ishizaki et al., 2009;Shibahara et al., 2007), eastern North Pacific (Cannariato andKennett, 1999;Cartapanis et al., 2011;Chang et al., 2014;Dean, 2007;Nameroff et al., 2004;20 Pride et al., 1999;Ohkushi et al., 2013;van Geen et al., 2003), eastern South Pacific (Martinez et al., 2006;Muratli et al., 2010;Salvatteci et al., 2016), Equatorial Pacific (Leduc et al., 2010) and Indian Ocean (Reichart et al., 1998;Suthhof et al., 2001;van der Weijden et al., 2006) were better oxygenated at the LGM relative to the PI climate. An important consequence of oxygenating the upper ocean is a reduction in the strength of denitrification in the these regions. Sedimentary δ 15 N records suggest that global aggregate rates of water column denitrification rates over the past 200 kya were lower during glacial periods 3.3 Importance of ocean biogeochemistry for climate 5 The spatial pattern of export production, the depth of the aragonite saturation horizon and the dissolved oxygen field of experiment O-LGM1 are outstanding in their disagreement with proxy evidence. Notably, experiment O-LGM1 lost a substantial quantity of carbon from the upper ocean in excess of the additions to the deep ocean caused by an altered physical ocean state. Experiment O-LGM1 was therefore unable to explain the glacial-interglacial drawdown of atmospheric CO 2 , despite the presence of a physical ocean state within realistic bounds, generated by the Cpl-LGM experiment. If we are to reconcile the 10 biogeochemistry of the glacial ocean with that inferred from proxy evidence, we must consider altering ocean biogeochemistry.

Reconciling the carbon budget
Three plausible modifications to the ocean biogeochemistry (see methods) were considered: (1) increased POC export production, (2) increased depth of POC remineralisation, and (3) reduced PIC export. In the following we step through the changes to 15 carbon content caused by each modification, and the reader is directed to Table 3 for reference.
(1) Experiment O-LGM3. Although the the scaling factor controlling the export production of organic matter was increased 10-fold, the actual increase in POC export production averaged over the global ocean was more modest at roughly 30 %.
Because most of the ocean within the O-LGM1 experiment became phosphate limited as greater quantities of nutrients were 20 redistributed into the deep ocean, the increase in export production in experiment O-LGM3 was only felt in those regions where PO 4 was not limiting. The sub-Antarctic zone of the Southern Ocean experienced the greatest increase in export production (∼250 %) due to the increase in the scaling factor, followed by a few small regions along the Chilean margin and in the Northwest Pacific (Fig. 7). These regional responses caused the global net export production rate to increase from 4.48 to 5.92 Pg C yr −1 . Although this rate of POC export production was still lower than the O-PI1 experiment of 8.01 Pg C yr −1 , this increased 25 carbon content by 188 Pg C.
(2) Experiment O-LGM4. The shift of organic matter to depth was associated with a global reduction in POC export production of ∼1.2 Pg C yr −1 as remineralisation released PO 4 and DIC further from the photic zone. Despite the reduction in the biological pump, the bulk transfer of POC to depth generated an increase in ocean carbon storage of 150 Pg C.  (Table 4).

Reconciling export production
Experiment O-LGM1 generated a strongly reduced POC export production across almost all regions of the global ocean  (Fig. 10). 15 The increase in the scaling factor dominated the change in export production produced when combining all three biogeochemical modifications (O-LGM6). The strong increase in export production observed in the sub-Antarctic was clearly replicated within this experiment and reconciles our simulated export production field with current evidence of productivity at the LGM.
In the Southern Ocean, the Atlantic and Indian sectors of the sub-Antarctic zone experienced a greater flux of organics to the 20 sediments ( Anderson et al., 2002Chase et al., 2001;Jaccard et al., 2013;Nürnberg et al., 1997). Whether this was also the case for the Pacific sector remains under debate, with some evidence for increase (Bradtmiller et al., 2009;Lamy et al., 2014) conflicting with evidence for no change (Bostock et al., 2013;Chase et al., 2003). Meanwhile, it is widely accepted that waters south of the Antarctic Polar Front were reduced in their productivity (Bostock et (Sarmiento et al., 2004).
This mechanism coupled with cooler temperatures caused reductions in export production across much of the mid and lower latitude oceans in experiment O-LGM6, which maintains the qualitative agreement between simulated and proxy observations 5 (see Introduction). Hence, the good spatial agreement between O-LGM6 and palaeoproductivity proxies at the LGM gives confidence that a stronger biological pump in sub-Antarctic waters, combined with an expanded sea ice cover that limited airsea exchange in the Antarctic Zone, was a key component for transferring carbon from the atmosphere to the ocean at the LGM.

10
There is good evidence that the mean position of the lysocline was not appreciably different at the LGM as compared to the Late Holocene (Yu et al., 2014). Because much of the ocean was saturated for aragonite in the O-LGM6 experiment, additional processes are required to shoal the aragonite saturation horizon (Ω = 1) and thereby reconcile proxy evidence.
One mechanism to shoal the aragonite saturation horizon would be to reduce continental inputs of alkalinity at the LGM. 15 However, the presence of glaciers, drier atmospheric conditions and the exposure of continental shelves due to lower sea level would have increased the supply of carbonates to the ocean (Gibbs and Kump, 1994;Riebe et al., 2004), thereby increasing ocean alkalinity and further deepening the aragonite saturation horizon. This mechanism has been largely refuted as having a significant effect on the glacial-interglacial difference in the carbon budget (Brovkin et al., 2007;Foster and Vance, 2006;Jones et al., 2002), and can therefore be ignored. 20 The individual biogeochemical modifications were also insufficient to effectively shoal the depth of aragonite saturation to be consistent with palaeo evidence. However, combining all three modifications in experiment O-LGM6 shoaled the aragonite saturation horizon significantly (Fig. 11)   Modifying ocean biogeochemistry did, however, have a large effect on the oxygen concentrations of the deep ocean (Fig. 13).
Increasing export production (O-LGM3) and deepening the remineralisation depth (O-LGM4) both reduced oxygen concentrations by 28 and 13 mmol m −3 , respectively. The combination of these modifications (O-LGM6) amplified their individual 25 effects, so that deep ocean oxygen was reduced by 63 mmol m −3 relative to the O-LGM1. The increased sensitivity of deep ocean oxygen to the combination of increased export production and a deeper remineralisation depth was due to an increase in the quantity of regenerated nutrients (DIC org and P org ) that resulted (Table 3). A greater proportion of regenerated nutrients relative to preformed nutrients at the LGM has been identified as a key driver of interior ocean deoxygenation (Jaccard and Galbraith, 2012;Sigman et al., 2010), and this process was captured in experiment O-LGM6. ative to proxy records may be resolved by (1) reducing the strength of the AABW circulation cell, and/or (2) further increasing export production, and/or (3) altering the spatial pattern of export production. These possibilities strongly indicate that global export production at the LGM may be underestimated by our simulations. Further investigation is required to reconcile our experiments with palaeo evidence of deep ocean deoxygenation.

4 Conclusions
In this study we have shown that physical changes to the ocean state, including an expanded sea ice field and an altered circulation, are not sufficient to explain the drawdown of 80-100 ppm CO 2 in the atmosphere of the LGM. While we demonstrate that the physical ocean state at the LGM is indeed highly conducive to storing carbon, owing largely to an increased subduction 10 of Antarctic Bottom Water, the tendency for the upper ocean to lose carbon through its equilibration with a lower atmospheric CO 2 concentration outweighs these gains. Thus, various biogeochemical modifications are necessary to overcome these losses and produce net gains of carbon in the ocean. The marine biogeochemical changes explored in this study were (1) an increase in export production consistent with greater iron fertilisation, (2) a shift of remineralisation to the deep ocean consistent with cooler temperatures, and (3) a decrease in the production of Particulate Inorganic Carbon consistent with cooler temperatures.

15
Only when all three changes were applied to a glacial ocean does the ocean carbon content increase sufficiently to account for the combined loss of carbon from the atmosphere and land at the LGM. Furthermore, their addition helps to reconcile unrealistic fields of export production, aragonite saturation state and dissolved oxygen produced by a simulation of the LGM without biogeochemical changes.

20
A key limitation of this study was the inability of biogeochemical modifications to truly deoxygenate the deep ocean on a global scale, as shown in palaeoclimate reconstructions. Either one or a combination of the following could resolve this inconsistency in our simulations: (1) the strength of the AABW circulation cell was too strong; (2) global export production was too weak; (3) the true spatial pattern of export production was not captured. Importantly, points (2) and (3) further implicate ocean biogeochemical processes as a strong influence on climate. New focus should be applied to investigate these possibilities, 25 including the use of multiple representations of the LGM physical ocean state and its effect on ocean biogeochemistry. Future work should also aim to include sedimentary processes, including carbonate burial, considering the importance of sediment processes to ocean biogeochemistry and climate.   b Power law exponent for POC remineralization changed from −0.9 to −0.7. Table 2. Changes in sea surface temperature (SST), sea ice extent and large scale circulation between the LGM and PI simulations. SST changes are compared with proxy reconstructions of SST generated by Waelbroeck et al. (2009) and Ho and Laepple (2015) whom use different proxies for their reconstructions to produce the differences depicted here and discussed in the text. The range of formation rates of Antarctic Bottom Water (AABW) and North Atlantic Deep Water (NADW) depicted here were calculated by Otto-Bliesner et al. (2007)   where Ω  Where the aragonite saturation horizon exceeds the depth of the ocean, the deepest grid box was included in the averaging process.  LGM -unmodified BGC −1037 ± 400 LGM -unmodified BGC (CO2 of 280 ppm) 607 ± 400 LGM -modified BGC (O-LGM6) c −194 ± 400 a Estimate of increase in ocean carbon content during the LGM made by Ciais et al. (2011), whereby atmospheric carbon was reduced by 194 ± 2 Pg C and terrestrial carbon was reduced by 330 ± 400 Pg C.
b BGC refers to biogeochemistry.
c Assumes all three biological modifications that were postulated (see Table 1,