Ocean biology could control atmospheric δ13C during glacial-interglacial cycle



[1] Estimates of changes in the global carbon budget are often based on the assumption that the terrestrial biosphere controls the isotopic composition of atmospheric CO2 since terrestrial plants discriminate against the 13C isotope during photosynthesis. However, this method disregards the influence of 13C fractionation by the marine biota. Here an interpretation of the glacial-interglacial shifts in the atmospheric CO2 concentration and δ13CO2 measured in the Taylor Dome ice core [Smith et al., 1999] is given by accounting for possible changes in the ocean biology based on sensitivity simulations undertaken with the intermediate complexity model CLIMBER-2. With a combined scenario of enhanced biological and solubility pumps, the model simulates glacial atmospheric CO2 and δ13CO2 similar to those inferred from the ice core. The simulations reveal that a strengthening of the oceanic biological carbon pump considerably affects the atmospheric δ13CO2.

1. Introduction

[2] Analysis of the global carbon-isotope budget is a powerful method for estimating carbon transfer between ocean, land, and atmosphere on decadal to millennium timescales [see, e.g., Tans et al., 1993; Bird et al., 1996]. Terrestrial plants discriminate against 13C during photosynthesis with averaged fractionation factor αLB about −18‰ and −5‰ for C3 and C4 plants, respectively [Lloyd and Farquhar, 1994]. By contrast, ocean seawater is enriched in 13C during CO2 exchange between the sea surface and the atmosphere with an averaged fractionation factor αO of ∼8.5‰ [Siegenthaler and Münnich, 1981]. An average sample of terrestrial carbon has a relatively low δ13C value of about −23‰ compared to the δ13CDIC of ∼1–2‰ and 0‰ for ocean surface and deep waters, respectively (δ13CDIC is for δ13C of dissolved inorganic carbon, or DIC). The gradient in δ13CDIC between the surface and the deep ocean is caused by the so-called biological pump [Volk and Hoffert, 1985]. Similar to terrestrial plants, marine phytoplankton discriminates 13C during photosynthesis with a fractionation factor αMB varying from −18‰ in the tropics to −36‰ in the polar regions [Goericke and Fry, 1994]. Decomposition of 13C–depleted organic matter reduces the δ13CDIC values in the deep ocean and increase the δ13CDIC in the upper part of the water column. Ignoring the spatial and temporal variations in αO, αLB and marine productivity, the changes in the terrestrial carbon storage can be estimated from changes in the atmospheric δ13CO2, or δ13CATM (see appendix A, equation (A6)). Calculated in this way, δ13CATM is solely controlled by changes in the terrestrial carbon storage alone.

[3] The assumptions that oceanic and terrestrial carbon storages have a constant but distinct 13C fractionation factor are applied, for example, to distinguish between the oceanic and terrestrial carbon sink when trying to track the pathways of anthropogenic CO2 [Ciais et al., 1995; Joos and Bruno, 1998; Battle et al., 2000]. However, this approach disregards changes in the biological carbon pump, which affect δ13CATM as well [Tans et al., 1993]. For example, strengthening of the biological productivity leads to an increased export flux Np that leaves surface waters enriched in 13C. Consequently, the CO2 flux across the sea surface shifts δ13CATM toward higher values. The question is, how significant δ13CATM is influenced by changes in the marine biota.

[4] Recently, Smith et al. [1999] published a record of the atmospheric CO2 and δ13CATM inferred from the Taylor Dome ice core for the last 26,000 years. These data constrain the reconstruction of the global carbon isotopic budget during the glacial-interglacial transition. The similarity between the Taylor Dome δ13CATM record and the previously reported measurements from the Byrd ice core [Leuenberger et al., 1992] supports the δ13CATM ice core records, although the technical uncertainty in measuring δ13CO2 remains significant (around 0.1‰).

[5] The interpretation of the glacial-interglacial transitions in δ13CATM inferred from these ice cores poses a problem for the approach that assumes that the land carbon exerts control on δ13CATM. Estimations of changes in the terrestrial carbon cycle based on foramenifera δ13C [Shackleton, 1977], the paleo-biome distribution [Adams et al., 1990; Crowley, 1995; Maslin et al., 1995; Guiot et al., 2001], or the simulation of terrestrial biosphere models [Prentice et al., 1994; Francois et al., 1998; Beerling, 1999] reveal a decline in the carbon storage capacity of the land biosphere by 440–1350 Pg C during the last glacial maximum (LGM) compared to the present day terrestrial carbon storage, with a best estimate of about 750 Pg C. Adding additional 750 Pg of terrestrial carbon to the atmosphere would have lead to a change in δ13CATM of about −0.4‰ (without accounting for changes in distribution of C3/C4 plants) which is significantly higher than the value of −0.16‰ inferred from the Taylor Dome ice core data [Smith et al., 1999]. Another potential source of carbon to the atmosphere during the LGM is associated with peat storage. The peat accumulation in the northern peatlands during the Holocene is estimated in 300–500 Pg C [Laine et al., 1996; Gajewski et al., 2001]; presumably, during the glacial period this carbon was stored in the ocean. If one increases the changes in terrestrial carbon during the glacial-interglacial transition by additional 300–500 Pg C, the discrepancy between the results by the approach based on land carbon control on δ13CATM and the ice core data becomes even larger.

[6] Here we examine the contribution from the marine and the terrestrial biota to changes in the global carbon budget during the last glacial termination. The primary goal of our study is to investigate the possible influence of changes in the functioning of the marine biota during the glacial-interglacial transition on the atmospheric CO2 level and δ13CATM. For this purpose the climate system model CLIMBER-2 is used, which allows a fully interactive simulation of the glacial-interglacial changes in the physical and the chemical components of atmosphere, ocean, and biosphere.

[7] The paper is organized in the following way. First we analyze the role of the marine biota in the global carbon cycle by performing a set of sensitivity experiments under present day conditions. Then a similar set of sensitivity experiments are carried out under LGM conditions. Assuming a scenario with a stronger biological cycling of carbon in the ocean, the model reproduce the glacial atmospheric CO2 level and δ13CATM in accordance with the values inferred from the ice cores. Finally, we suggest that the abrupt changes in δ13CATM inferred from the Taylor Dome ice core record are a consequence of changes in the marine biology.

2. Methods

[8] We performed a set of experiments with a climate system model of intermediate complexity CLIMBER-2 [Petoukhov et al., 2000; Ganopolski et al., 2001], version 2.3. It includes a 2.5-dimensional dynamical-statistical atmosphere model, a multibasin, zonally averaged ocean model (including a sea ice model) and a terrestrial vegetation model (including a model for the terrestrial carbon cycle) with a coarse spatial resolution of 10° in latitude and 51° in longitude. In version 2.3 of CLIMBER-2, the latitudinal resolution of oceanic model is increased to 2.5°. The time step differs among the model components (1 day, 5 days, and 1 year for atmospheric, ocean, and vegetation models, respectively). Results of CLIMBER-2 compare favorably with data of present-day climate and with paleoclimatic reconstructions [Ganopolski et al., 1998; Claussen et al., 1999]. The model is recently upgraded by the implementation of a marine carbon cycle model, which accounts for the cycling of inorganic and organic carbon in ocean [Brovkin et al., 2002]. The marine biota model is similar to the model by Six and Maier-Reimer [1996] and includes the distribution of phytoplankton, zooplankton, dissolved organic carbon (DOC), detritus (POC), phosphate, and the stable carbon isotopes.

[9] Regarding setup of the boundary conditions, the simulations are subdivided into three groups (see Table 1). The first group includes control simulations for present-day (PD) and LGM climates (LGM-Ctrl) with fixed atmospheric CO2 concentration (280 and 200 ppmv for PD and LGM-Ctrl, respectively). Models for terrestrial and oceanic biogeochemistry simulate the global carbon cycle in equilibrium with prescribed atmospheric CO2. Simulation PD serves as a reference simulation for biogeochemistry; it sets initial conditions for the total carbon storage in the system (ocean, land, and atmosphere) for the LGM scenarios with interactive carbon cycle. Let us note that in simulation LGM-Ctrl the total amount of carbon in the system differs from the carbon storage in the PD simulation and carbonate compensation is not considered. The purpose of the simulation LGM-Ctrl is to estimate the effects of changes in sea surface temperatures (SSTs) and oceanic circulation on the oceanic carbon storage.

Table 1. Simulation Summary
AcronymSimulation PurposeCarbon ConservationAtm. CO2, ppmvδ13CATM, ‰Averaged ALK, μeq/kgAveraged PO4, μeq/kgExport flux Np, Pg C/yr
  • a

    Boundary conditions as for 21,000 yr before present: Insolation following Berger [1996], ice sheet distribution in accordance with Peltier [1994], salinity, alkalinity, and PO4 concentration increased by 3.3% to account for sea level changes.

  • b

    Atmospheric CO2 concentration is constant; total carbon amount in the system is not conserved.

  • c

    Phytoplankton growth rate is doubled under constraint of PO4 limitation.

  • d

    PO4 concentration is higher than in LGM-Ctrl by 33%.

  • e

    Total carbon storage (land, atmosphere, and ocean) equals to a sum of the total carbon storage in the PD-simulation and dissolved CaCO3 sediments (970 Pg C).

  • f

    Transfer of terrestrial carbon is increased by 360 Pg C. Phytoplankton growth rate is increased 2.5 times under constraint of PO4 limitation.

  • g

    Total carbon storage (land, atmosphere, and ocean) equals to a sum of the total carbon storage in the PD-simulation, dissolved CaCO3 sediments (1330 Pg C), and peat carbon (360 Pg C).

  • h

    Rain ratio is decreased from 0.1 to 0.07. PO4 concentration is higher than in LGM-Ctrl by 20%.

  • i

    Total carbon storage (land, atmosphere, and ocean) equals to a sum of the total carbon storage in the PD-simulation and dissolved CaCO3 sediments (1220 Pg C).

LGM-Ctrlaeffect of SSTs and circulation changes on COnob20024512.156.7
Strangelove Ocean
S-PDCOB estimationnob2802373
S-LGMaCOB estimationnob2002576
LGM Scenarios
LGMa,c,denhanced bio− pumpyese199−6.5625762.8611.1
LGM-peata,d,flower land carbon storageyesg198−6.6526212.8611.6
LGM-rra,c,hdecreased rain ratioyesi200−6.5926072.5710.0

[10] The second set of simulations of the oceanic carbon cycle without oceanic biology (frequently called as the “Strangelove ocean”) is aimed to estimate COB in the ocean for present-day (S-PD) and LGM climates (S-LGM). In the simulations, the physical climate state (temperature, precipitation, atmospheric, and oceanic circulation, etc.) is identical to the state from the control simulations. Similar to the LGM-Ctrl simulation, the total amount of carbon in the system is not conserved, and carbonate compensation is not accounted for. This limits the application of the Strangelove ocean simulations to the estimate of COB.

[11] The third group of simulations consists of several LGM scenarios (LGM, LGM-peat, and LGM-rr) that result in a draw down of the atmospheric CO2 concentration from 280 to 200 ppmv. The purpose of these scenarios is to estimate changes in δ13CATM in accordance with different hypothesis. The carbon cycle in these simulations is interactive; that is, the total amount of carbon stored in the ocean, land, and atmosphere is the same as in the PD-simulation. Because dissolution of the oceanic carbonate sediment plays an important role in the glacial biogeochemistry, a difference in the carbonate storage is included into the total carbon balance (see Table 2). The amount of dissolved carbonates is determined on a base of keeping the constant concentration of carbonate ion in the deep Pacific. Atmospheric and ocean models are interactive in the simulations; they respond to the atmospheric CO2 as greenhouse gas forcing. If atmospheric CO2 concentration in simulations were significantly different from the control level (200 ppmv), simulated climate state would be different from the control simulation (LGM-Ctrl), causing undesirable feedback loops to the carbon cycle. Therefore parameters of scenarios, for example, degree of enhancing of biological productivity in the ocean, are chosen under condition that the atmospheric CO2 concentration is ∼ 200 ppmv. This approach guarantees that the climate state (e.g., temperature, precipitation, circulation) in the LGM scenarios is the same as in the LGM-Ctrl simulation.

Table 2. Global Carbon Balance in the Simulations PD and LGM
CompartmentCarbon, Pgδ13C
Preindustrial (PD)
Atmosphere (280 ppmv CO2)600−6.5
Ocean, DIC and DOC38,3300.4
Oceanic carbonate sediments9701.5
Last Glacial Maximum (LGM)
Atmosphere (199 ppmv CO2)420−6.6
Ocean, DIC and DOC40,1200

3. Results

3.1. Preindustrial Carbon Cycle

[12] The control simulation (PD) corresponds to preindustrial boundary conditions with an atmospheric CO2 value of 280 ppmv and a δ13CATM value of −6.5‰ (see Table 1). The model is integrated for 10,000 years achieving a steady state. The amount of carbon stored in the terrestrial pool is ∼1920 Pg C (around 840 Pg C in the biomass and 1080 Pg C in the soil), with most of the carbon (around 85%) allocated to the C3-photosynthesis pathway and the remaining carbon (15%) to the C4-pathway. Owing to the fact that atmospheric CO2 is depleted in 13C by −6.5‰, the average δ13C of the terrestrial carbon cycle is about −22.6‰ (see Table 2). The ocean carbon storage consists of 38,250 and 80 Pg C in the form of DIC and DOC, respectively (the refractory DOC with thousand years timescale is not accounted for in the model). The marine primary productivity is ∼44 Pg C yr−1 and the export flux Np is ∼7 Pg C yr−1. These fluxes are comparable with the model estimates by Six and Maier-Reimer [1996]. The 13C fractionation between DIC and marine phytoplankton is simulated as a function of dissolved molecular carbon dioxide [CO2(aq)] [Rau et al., 1989; Hofmann et al., 2000] which yields a globally averaged fractionation of −23‰. Because the surface ocean δ13CDIC values are close to 2‰ in the model, the average δ13C of the POC and DOC (δ13CPOC and δ13CDOC, respectively) is about −21‰. The averaged δ13CDIC in the ocean equals 0.4‰ (see Table 2).

[13] To estimate the potential of the marine biological carbon pump in sequestering atmospheric CO2, we conducted the Strangelove ocean experiment (S-PD). The marine primary productivity was set to zero while the atmospheric CO2 was kept at a constant value of 280 ppmv and δ13CATM at value of −6.5‰. After 10,000 years of integration the system approaches an equilibrium where the ocean has released around 2400 Pg C into the atmosphere and the averaged ocean δ13CDIC has increased to 2‰. If the atmospheric CO2 were interactive with the ocean, the corresponding increase of the atmospheric CO2 level would be around 220 ppmv; this estimate is consistent with the increase of around 230 ppmv obtained in the Strangelove ocean experiment with a three-dimensional (3-D) oceanic carbon cycle model HAMOCC [Maier-Reimer et al., 1996].

[14] To illustrate the results, a simplified view on the functioning of the global carbon cycle within the model is shown in Figure 1a. In the ocean, the total dissolved carbon, CO, is separated into two pools: Abiotic carbon pool absorbed by the ocean due to the solubility pump, COA, and carbon cycled by the biological pump, COB, which consists of DOC of marine origin (river transport of terrestrial DOC is not accounted for in the model) as well as of DIC originated from remineralization of POC and DOC. The oceanic carbon storage in the S-PD-simulation is taken as an estimate of COA; the rest of the total oceanic carbon in the PD-simulation is assumed to be an estimate for COB. In this simplification, the indirect influence of marine biota on the solubility pump through changes in the surface alkalinity is neglected. With this separation, the carbon pools with homogeneous 13C fractionation can be considered separately; the average δ13CDIC from S-PD-simulation and δ13CDOC from PD-simulation are taken as estimates for δ13C of COA and COB, respectively. The abiotic carbon cycle contains ∼35,900 Pg C with a δ13C value of 2‰, and the biological cycle has a capacity of ∼2400 Pg C and a δ13C value of −21‰. The total oceanic carbon content is ∼38,300 Pg C, with an averaged δ13CDIC of ∼0‰, both in line with GEOSECS inventory [Bainbridge, 1981; Broecker et al., 1982; Kroopnik, 1985]. The terrestrial carbon storage is ∼1900 Pg C and possesses an average δ13C value of about −23‰, in line with the other estimations for the preindustrial terrestrial carbon cycle [Melillo et al., 1996].

Figure 1.

A simplified view of the global carbon cycle. (a) Simulation PD is shown. The oceanic carbon cycle is separated into an abiotic cycle (DIC absorbed by the solubility pump) and a biological cycle (DOC and DIC originated from remineralization of organic matter). Organic carbon of terrestrial and marine origin is isotopically much lighter (δ13C of −21 to −23‰) than the DIC absorbed by the solubility pump (δ13C ∼2‰). (b) Simulation LGM is shown. The terrestrial carbon storage is lower by 640 Pg C in comparison with the PD-simulation. This amount of isotopically light carbon enters the oceanic carbon cycle without substantial effect on the atmospheric δ13C because of enhanced biological cycle in the ocean. The δ13C of the oceanic biological cycle increases by 2‰ due to a drop in [CO2(aq)].

[15] The two biological branches of the global carbon cycle, the terrestrial and the oceanic one, have very similar carbon isotopic signatures and carbon storage capacities. Therefore it is nearly impossible to separate their influence on the atmospheric isotope signature as well as the CO2 concentration using atmospheric data records alone. Any change in the atmospheric gas composition could be equally interpreted as changes in either the marine or the terrestrial biota branch of the global carbon cycle.

3.2. LGM Carbon Cycle

[16] In the experiments for the LGM, the model was driven by the orbital forcing corresponding to the insolation 21,000 yr before present [Berger, 1996]. The ice sheet distribution and changes in the land area were taken from the PMIP reconstruction [Peltier, 1994]. Ocean volume and concentrations of oceanic tracers (salinity, alkalinity, nutrients) were adjusted to changes in the global sea level (see Table 1) in accordance with the SPECMAP reconstruction [Imbrie et al., 1984]. In the first simulation (LGM_Ctrl), the atmospheric CO2 level was prescribed to a value of 200 ppm [Barnola et al., 1987; Smith et al., 1999] for the atmospheric model and the terrestrial biogeochemistry model while the atmospheric CO2 used for calculating the air-sea exchange into the ocean model was kept at the pre-industrial level of 280 ppm. Corresponding changes in the oceanic circulation and atmospheric fields were described by Ganopolski et al. [1998] and Ganopolski and Rahmstorf [2001]. In comparison with PD-simulation, the global averaged annual mean air temperature and sea surface temperatures (SSTs) decrease by 5°C and 3°C, respectively, and the oceanic carbon storage is lower by 270 Pg C. Most of this decrease is explained by a decrease in the export flux, Np, by 0.5 Pg C corresponding to around 8% of the preindustrial level. The draw down of the atmospheric CO2 concentration caused by the increased solubility of the relatively cold surface waters [see, e.g., Broecker and Peng, 1982; Hofmann et al., 1999; Archer et al., 2000b; Schulz et al., 2001] is not significant in our simulation. The model sensitivity of the oceanic carbon uptake on changes in the SSTs is critical dependent upon the parameterization of vertical and horizontal mixing [Archer et al., 2000a]. In accordance with other ocean circulation models with a relatively large mixing, the carbon uptake in the CLIMBER model is not significantly influenced by changes in the SSTs during the LGM-simulations. The sea surface temperatures, in the polar convection regions where the deep water is formed, are not considerably changed because the temperature here is locked to the freezing temperature of sea water. The impact of Antarctic sea-ice changes on the atmospheric CO2 concentration [Stephens and Keeling, 2000; Maqueda and Rahmstorf, 2002] is negligible in the CLIMBER model similar to the HAMOCC model (D. Archer et al., Effect of Antarctic sea ice and stratification on atmospheric pCO2: A box model artifact?, manuscript submitted to Paleoceanography, 2001). Moreover, the ocean should absorb additional 640 Pg C released from the terrestrial biosphere because the terrestrial carbon storage capacity is drastically reduced due to the decrease in the forest area as well as in the plant primary productivity.

[17] Currently, most of the hypothesis proposed to explain glacial-interglacial transitions in the global carbon cycle can be subdivided into two types: changes in (1) the marine solubility pump (carbonate or silicate oceanic chemistry) or in (2) the marine biological pump (biological productivity) (for review, see Archer et al. [2000a, 2000b] and Sigman and Boyle [2000]). Iron fertilization experiments recently conducted in the Southern Ocean (SO) (SOIREE [Boyd et al., 2000], EisenEx [Smetacek, 2001]), spot this region as a mainly iron limited area with respect to primary production of phytoplankton. Hence the iron fertilization hypothesis by Martin [1990], which explains the lower atmospheric CO2 levels during LGM by a stimulated growth of phytoplankton owing to a much higher aeolian dust supply to the SO (hypothesis 2), becomes again favored in the understanding of glacial-interglacial transitions. The assumption of a strengthened biological productivity during glacial periods is also in line with SO4 proxies [Mayewski et al., 1996; Broecker and Henderson, 1998] while in disagreement with some other proxies, for example, Cd/P ratio [Elderfield and Rickaby, 2000].

[18] To achieve an atmospheric CO2 level drop of 80 ppmv, we carried out a simulation (LGM) with a mixture of the two hypotheses 1 and 2 discussed above (Table 1). The global mean nutrient concentration in the ocean was increased by 33% imitating nutrients washout from the continents [see, e.g., Shaffer, 1990]. Additionally, a growth rate of phytoplankton was increased globally by a factor of 2 simulating an enhancement of the biological pump due to iron fertilization [Hofmann et al., 1999; Watson et al., 2000]. This results in a 55% increase in the export flux. The increased flux of organic matter to the interior of the ocean and the subsequent remineralization results in lower oxygen levels during the LGM than in the present ocean (see Figure 2). In particular, oxygen level in the intermediate waters in the equatorial Pacific becomes very low, owing to the high export flux and the relatively low oxygen concentration in the deep waters there. The lysocline constraint imposed by sediment cores [Sigman et al., 1998] is considered by keeping a constant carbonate ion concentration in the deep Pacific. Additional dissolution of calcium carbonate from the sediments is modelled by adding 970 Pg carbon to the system and correspondingly increase the oceanic alkalinity in a proportion of 2 equivalent per 1 mole of carbon. This simplified approach neglect, to a first approximation, the influence of increased downward flux of organic matter and carbonate shells on the complex remineralization processes in the sediments which influence the pH of the deep ocean and the process of carbonate dissolution [Archer, 1991].

Figure 2.

Oxygen concentration, μmol/kg. (a) Observations GEOSECS. (b) Simulation PD. (c) Simulation LGM.

[19] In response to the stronger biological and solubility carbon pumps, the oceanic DIC pool increases by around 1800 Pg C (Table 2). Because most of this increase is due to an enhanced biological pump, the average oceanic δ13CDIC drops considerably. However, δ13CPOC increases by 2‰ because of the drop in [CO2(aq)]. Surface ocean δ13CDIC is higher by 0.2‰ because of the enhanced biological production. The gradient in δ13CDIC between the surface and the deep ocean increases by 0.7‰ (see Figure 3). The fractionation of terrestrial carbon decreases by 0.1‰ because of a relative advance of C4 plants [Francois et al., 1999; Beerling, 1999]. Finally, atmospheric CO2 is 199 ppmv and δ13CATM is −6.56‰. The latter value is only by 0.06‰ lower than in simulation (PD). To explore an effect of changes in marine biota on the carbon cycle, a simulation (S-LGM) without a marine biota similar to simulation (S-PD) is carried out. Oceanic carbon storage in the simulation (S-LGM) is 36,000 Pg C, and a difference in oceanic carbon between simulations (LGM) and (S-LGM) is 4,100 Pg C (see Figure 1b). We conclude that most of the increase in the oceanic carbon storage, or 1700 Pg C, is due to the enhanced biological pump in the ocean, and that the increase in the oceanic carbon storage due to solubility pump is only 100 Pg C. In other words, the biological carbon cycle within the model should be enhanced by ∼70% for glacial conditions in order to obtain glacial-interglacial changes in CO2 similar to the ones inferred from the ice cores.

Figure 3.

Depth profile of globally averaged changes in δ13CDIC, difference between simulations LGM and PD.

[20] The terrestrial biogeochemistry model in CLIMBER-2 does not simulate the accumulation of carbon in peatlands. The current estimates of peat carbon accumulated in the northern peatlands during the Holocene range from 300 to 500 Pg C [Laine et al., 1996]. Accounting for the glacial-interglacial changes in the peat carbon leads to an additional transfer of terrestrial carbon into the ocean during the LGM. Addressing the uncertainties in the terrestrial carbon budget, a simulation (LGM-peat) was carried out where additional 360 Pg C of terrestrial carbon with δ13C of −22.6‰ (averaged δ13C of terrestrial carbon in simulation PD) were added to the model. The total transfer of terrestrial carbon from the land into the ocean is then 1000 Pg C. The growth rate for oceanic phytoplankton was additionally increased (2.5 times in comparison with the PD-simulation) which corresponds to a hypothesis of stronger enhancement of the biological pump than in the LGM-simulation. As result, the biological productivity increased additionally by 0.7 Pg C and the biological carbon storage, COB, increased by around 170 Pg C. Additional 360 Pg C of carbonate sediments were dissolved to account for the carbonate compensation in the ocean. In response to these forcings, the mean ocean alkalinity increased by 45 μmol/kg (see Table 1) and the carbonate ion concentration in the deep ocean was not significantly changed. Equilibrium atmospheric CO2 in this simulation is 198 ppmv and δ13CATM is −6.65‰, similar to the value of δ13CATM for the LGM inferred from the Taylor Dome data (−6.67‰, see section 4 below). Adding 360 Pg C of organic carbon into the system resulted in a decrease of δ13CATM by 0.09‰, similar to a decrease by 0.1‰ expected from the simplified calculations (see the third paragraph in appendix A).

[21] In the simulations (LGM) and LGM-peat, the rain ratio (ratio of carbonate flux to the export flux) is the same as in the PD-simulation. Assuming the same rain ratio, a strengthening of the biological pump in the LGM scenarios would also enhance formation of calcareous shells by marine biota. Hence the burial rate of CaCO3 into the sediments would increase similarly, shifting the deep ocean to more acid conditions. This would counteract the initial atmospheric CO2 draw down, even were the CO32- concentration in the deep ocean to be kept constant. To address this potential shortcoming of the scenarios, another simulation (LGM-rr) was done with enhanced biological pump but with the rain ratio decreased to such a level that CaCO3 flux is the same as in the PD-simulation. A rationale for the latter assumption is a possible shift in glacial marine ecosystems toward dominance of silica-limited phytoplankton (diatoms) over nonsiliceous species including coccoliths, which have calcareous shells: increased dust deposition rates during glacial times presumably provided the oceans with a higher silica input, releasing diatoms from SiOH4 stress [Harrison, 2000; Treguer and Pondaven, 2000]. Decrease in the rain ratio leads to a more efficient biological pump, which does not need to be as strongly enhanced as in the LGM-simulation. In the LGM-rr-simulation, a control level of the atmospheric CO2 concentration (200 ppmv) was achieved by increasing of the PO4 inventory by 20% and by doubling of the phytoplankton growth rate. The export flux is 10 Pg C/yr and the amount of dissolved carbonate is 1220 Pg C. Decrease in the oceanic biological pump results in smaller δ13CATM (−6.59‰) than in the LGM-simulation. The decrease in δ13CATM in response to decrease in biological productivity is in line with simulations by Marino et al. [1992] who attributed decrease in δ13CATM during glacial period to reduction in terrestrial biomass and decreased oceanic productivity.

4. Discussion: Interpretation of the Ice Core Data

[22] Recently, Smith et al. [1999] presented an atmospheric record inferred from the Taylor Dome ice core, covering the time period from 26 kyr BP until 2 kyr BP. The original data, as depicted in Figure 4 (filled circles with a 1σ error bar), were interpolated within a time interval of 1000 years (solid line). The end of the LGM is taken to be at 17.5 kyr BP following Smith et al. [1999], and the onset of the Holocene is chosen at 11.2 kyr BP in accordance with Monnin et al. [2001]. For the time period between 26 and 17.5 kyr BP, which includes the LGM, the mean atmospheric CO2 level is of ∼192 ppmv and the δ13C value of about −6.65‰. During the Holocene (11.2 to 2 kyr BP) the mean atmospheric CO2 level as well as the δ13CATM value has increased with respect to the LGM (26–17.5 kyr BP) by 79 ppmv and 0.17‰, respectively. This is in agreement with the observations by Leuenberger et al. [1992], who found a ΔCO2 shift of 80 ppmv and a shift in δ13CATM of 0.19‰ between LGM and Holocene in the Antarctic ice core.

Figure 4.

Interpretation of the Taylor Dome ice core data [Smith et al., 1999]. The original data for 26–2 kyr BP are shown by filled circles with a 1σ error bar. The upper and lower curves are 1000 year moving average for atmospheric CO2 concentration and δ13CATM, respectively. Vertical dashed lines indicate boundaries between the LGM, transition, and the Holocene. Averaged atmospheric CO2 and δ13CATM for LGM and Holocene are shown in the rounded boxes. Red arrows point to the simultaneous increase in the atmospheric CO2 and decrease in δ13CATM at the beginning of the transition, interpreted as a reduction in the oceanic biological pump. Blue arrows are for increase in the atmospheric CO2 at the end of the transition and during the middle to late Holocene, presumably due to the carbonate compensation mechanism. Magenta arrows point on simultaneous slowdown of the atmospheric CO2 growth and increase in δ13CATM at the early Holocene, interpreted as an increase in the terrestrial carbon storage.

[23] The averaged shift in δ13CATM of around 0.17‰ inferred from the Taylor Dome ice core is significantly smaller than the value one could expect assuming that the terrestrial biosphere controls δ13CATM values alone. Smith et al. [1999] argued that, due to an SST lowering by 5°C during LGM, the stable carbon isotope fractionation of the air-sea gas-exchange was enhanced by around 0.6‰ [Mook, 1987]. This could explain the higher value of oceanic δ13CDIC and the corresponding low value of atmospheric δ13C at 16.5 kyr BP. However, our simulation does not support this hypothesis because of two reasons. First the global averaged SST is lower in the simulation (LGM) by 3°C, while the SSTs in the areas of deep water formation is not considerably changed. Second, the 13C fractionation associated with the air-sea gas exchange depends linearly on the fraction of the bicarbonate ion [Siegenthaler and Münnich, 1981; Zhang et al., 1995], which is lower by 4% in the LGM simulation because of the lower CO2 and higher alkalinity (see the second paragraph in appendix A). As a result, changes in the fractionation in the solubility pump cannot explain the observed shift in δ13CATM within the model.

[24] Considering the sensitivity study above, we suggest the following scenario for the last glacial-interglacial transition: During the LGM the oceanic biological and solubility pumps worked in an enhanced mode, compared with the Holocene conditions. At the beginning of the deglaciation, around 17.5 kyr BP, a decrease in the aeolian iron supply into the SO [Petit et al., 1990] led to a reduction of the strength of the ocean biological pump. A simultaneous decrease in the oceanic overturning due to an increased freshwater flux from the melting ice sheets could enhance this oceanic carbon release. The subsequent outgassing of CO2 depleted in 13C was followed by a simultaneous increase in CO2 and a decrease in δ13CATM (shown as red arrows in Figure 4). Decreases in the terrestrial carbon storage due to shrinking tropical land areas could potentially be an additional source of atmospheric carbon. However, the amplitude of the latter source cannot be considerable because of rather slow changes in sea level. Moreover, the general trend in the glacial-interglacial transition is a reestablishment of forest and soil cover and thereby a corresponding uptake of carbon (640 Pg C within the model, see Table 2).

[25] We hypothesize that the oceanic biological pump reached an interglacial level ∼13–14 kyr BP when the dust supply to the SO was significantly diminished. The impact of CaCO3 compensation on atmospheric CO2 starts several thousands of years after beginning of transition and results in higher accumulation rates of CaCO3 (blue arrows in Figure 4). Additionally, the sea level rise started to contribute to enhanced carbonate and nutrient sedimentation in the continental shelf areas (blue arrows in Figure 4). With the onset of the Holocene about 11 kyr BP, the terrestrial biosphere began to accumulate carbon depleted in 13C, which resulted in a constant or declined level in atmospheric CO2 and increased δ13CATM values (shown as magenta arrows in Figure 4). Later during the Holocene, carbon storage in the terrestrial biosphere was not changed significantly [Beerling, 1999; Brovkin et al., 2002]. A further increase of atmospheric CO2 of ∼20 ppmv from 8 to 2 kyr BP could be explained by the carbonate compensation due to excessive accumulation of carbonate sediments in the course of the Holocene [Milliman, 1993; Broecker et al., 1999].

5. Conclusions

[26] We have simulated the glacial-interglacial difference in the global carbon budget with a fully interactive model of the climate system, which provides consistent changes in the atmospheric, oceanic, and terrestrial biogeochemistry. With a combination of an enhanced biological production and dissolution of CaCO3, the model simulates changes for glacial atmospheric CO2 and δ13CATM similar to those inferred from the ice cores.

[27] On the basis of these results, we suggest correcting the approach, which assumes that the land carbon exerts control on the atmospheric δ13C. The marine biology affects δ13CATM value as well. Because both terrestrial and oceanic branches of the biological carbon cycle have very similar carbon isotopic signatures and carbon storage capacities, it is nearly impossible to separate their influence on the atmospheric isotope signature as well as the CO2 level using atmospheric data records alone.

[28] An elevation of the δ13CATM can also be caused by an increased storage of carbon in the terrestrial ecosystem. When changes in the terrestrial carbon pool are compensated by changes of the biological carbon pump in the ocean, the transfer of organic matter from land into the ocean and vice versa cannot be inferred from the records of the stable carbon isotope composition of the atmosphere. An increase of the terrestrial carbon pool would increase the δ13CATM, while the simultaneous weakening of the marine biological carbon pump would drive the δ13CATM values in the opposite direction.

[29] The evidence from the ice core data supports this result. Particularly, a drop of 0.5‰ in δ13CATM at the beginning of the transition period (around 16.5 kyr BP) combined with a simultaneous increase in the atmospheric CO2 [Smith et al., 1999] is very difficult to explain without including the marine biology. We interpret this drop as a substantial reduction in the oceanic biological pump, for example, as a consequence of a decrease in the biological production in the Southern ocean.

Appendix A: Sensitivity of δ13CATM to Changes Terrestrial and Marine Biology

A1. Conventional Approach: δ13CATM is Controlled by Land Carbon Storage

[30] The global carbon budget is constrained by conservation equations for total carbon in the system, A, and total 13C, A13, expressed in δ13C terms:

display math
display math

where Ci and δi are for carbon and δ13C of ith compartment for the preindustrial state, index i = {A,O,L} is for atmosphere, ocean, and land, respectively. If ΔCi and Δδi are changes in carbon and δ13C between the glacial and the reference states, then equations (A1)–(A2) can be written in a form

display math
display math

Let us assume that there are no changes in the isotopic fractionation on the atmosphere-ocean and atmosphere-land boundaries, as well as in the oceanic biology. In this case an equilibrium approximation could be applied:

display math

and from equation (A4) follows

display math

From equations (A3) and (A6) one can estimate sensitivity of δ13CATM to changes in terrestrial carbon storage ΔCL. With preindustrial values of δi and Ci in the compartments taken from the Table 2 for (PD) simulation, ΔCA taken from the observations and ΔCO = − ΔCL, 640 Pg C decrease in the terrestrial carbon storage corresponds to 0.36‰ decline in δA (ignoring changes in carbonate sediments). In this approach, atmospheric δ13C is fully controlled by the changes in the terrestrial carbon storage. Sensitivity of δ13CATM to changes in land carbon, ΔδACL, is 0.1‰ per 180 Pg C. In other words, if additional 180 Pg C of terrestrial carbon are dissolved in the ocean, the atmospheric δ13C is decreased by 0.1‰. Accounting for additional carbonate dissolution due to carbonate compensation leads to negligible changes in the sensitivity.

A2. Correction for Inorganic Fractionation

[31] Equation (A5) is valid in case of constant oceanic fractionation factor, αo. It is relatively easy to correct equation (A5) for the changes in inorganic fractionation by assuming

display math

where Δδof is a change in αo due to changes in temperature and carbonate ion composition. The αo increases by 0.1‰ with decrease in water temperature per °C [Mook et al., 1974]. The factor αo depends also on a fraction of bicarbonate ion [HCO3] in DIC species, with decrease ∼0.1‰ per 1% decrease in [HCO3] fraction. If surface alkalinity increases, like in LGM-simulation, then [HCO3] fraction decreases, leading to a decrease in αo. In LGM simulation, globally averaged SST decreases by 3°C, and the fraction of [HCO3] decreases by 4%. Consequently, Δδof is about −0.1‰. Let us note that the correction for inorganic fractionation does not change the sensitivity of δ13CATM to the changes in terrestrial carbon cycle.

A3. Accounting for Changes in Marine Biology

[32] If marine biology is altered, equation (A5) is no longer valid. In this case, the ocean carbon cycle is affected by changes in biological export flux, Np, and fractionation during marine photosynthesis, αMB (see Figure 1). The effect of terrestrial carbon emission on δ13CATM depends on whether this carbon is allocated into abiotic or biological cycle. In a scenario of additional carbon emission, for example, by peat mineralization like in the simulation (LGM-peat), the model can simulate its absorption by ocean by taking approximately 50% into biological cycle and 50% into abiotic cycle due to necessity to fulfil the lysocline constraint. This would lead to halving sensitivity of δ13CATM to changes in land carbon, ΔδACL, to 0.1‰ per 360 Pg C.


[33] The authors thank Martin Claussen, Miguel Maqueda, Stefan Rahmstorf, Colin Prentice, and an anonymous reviewer for the thoughtful comments. We appreciate help by David Archer whose suggestions and remarks lead to considerable improvement of the manuscript. Jørgen Bendtsen was funded by the Danish National Science Research Council.