A model-based interpretation of low-frequency changes in the carbon cycle during the last 120,000 years and its implications for the reconstruction of atmospheric Δ14C



A main caveat in the interpretation of observed changes in atmospheric Δ14C during the last 50,000 years is the unknown variability of the carbon cycle, which together with changes in the 14C production rates determines the 14C dynamics. A plausible scenario explaining glacial/interglacial dynamics seen in atmospheric CO2 and δ13C was proposed recently (Köhler et al., 2005a). A similar approach that expands its interpretation to the 14C cycle is an important step toward a deeper understanding of Δ14C variability. This approach is based on an ocean/atmosphere/biosphere box model of the global carbon cycle (BICYCLE) to reproduce low-frequency changes in atmospheric CO2 as seen in Antarctic ice cores. The model is forced forward in time by various paleoclimatic records derived from ice and sediment cores. The simulation results of our proposed scenario match a compiled CO2 record from various ice cores during the last 120,000 years with high accuracy (r2 = 0.89). We analyze scenarios with different 14C production rates, which are either constant or based on 10Be measured in Greenland ice cores or the recent high-resolution geomagnetic field reconstruction GLOPIS-75 and compare them with the available Δ14C data covering the last 50,000 years. Our results suggest that during the last glacial cycle in general less than 110‰ of the increased atmospheric Δ14C is based on variations in the carbon cycle, while the largest part (5/6) of the variations has to be explained by other factors. Glacial atmospheric Δ14C larger than 700‰ cannot not be explained within our framework, neither through carbon cycle-based changes nor through variable 14C production. Superimposed on these general trends might lie positive anomalies in atmospheric Δ14C of ∼50‰ caused by millennial-scale variability of the northern deep water production during Heinrich events and Dansgaard/Oeschger climate fluctuations. According to our model, the dominant processes that increase glacial Δ14C are a reduced glacial ocean circulation (+∼40‰), a restricted glacial gas exchange between the atmosphere and the surface ocean through sea ice coverage (+∼20‰), and the enrichment of dissolved inorganic carbon with 14C in the surface waters through isotopic fractionation during higher glacial marine export production caused by iron fertilization (+∼10‰).

1. Introduction

New measurements from which the radiocarbon signal (14C) of the atmosphere during the past thousands of years can be reconstructed were published steadily throughout the last years [Bard et al., 1998; Stuiver et al., 1998; Voelker et al., 1998; Goslar et al., 2000; Hughen et al., 2000, 2004; Kitagawa and van der Pflicht, 2000; Schramm et al., 2000; Beck et al., 2001; Reimer et al., 2004; Fairbanks et al., 2005]. The reconstructions of past changes in the atmospheric 14C concentration have a high accuracy for the Holocene period where the data are based on measurements of the 14C/12C ratios measured in dendrochronologically dated tree rings [Reimer et al., 2004]. For the pre-Holocene period the various Δ14C data sets (Figure 1) come from different natural archives. They exhibit increasing differences the further back in time they extend. The data for the last 10,000 years (10 kyr) show a moderate long-term decrease of the order of 100‰. Despite the differences between the Δ14C reconstruction for the periods of the last glacial and deglaciation, a rather sharp increase of approximately 600‰ around 44 kyr BP to high values of the order of 400–800‰ from 40 to 20 kyr BP and a strong decrease from 20 to 10 kyr BP are consistent features of most of the records.

Figure 1.

A compilation of published reconstructions of atmospheric Δ14C during the last 50 kyr.

The most important factor that can influence the atmospheric 14C concentration is the variable 14C production rate [Siegenthaler et al., 1980] which is caused by changes in solar activity and the geomagnetic dipole field intensity provided that the galactic cosmic ray flux to the solar system stayed constant [Lal and Peters, 1967]. To our knowledge there are no clear indications for a variable galactic cosmic ray intensity during the last 50 kyr. Assuming that also the average solar modulation of the cosmic ray flux was constant, estimates of past changes in the 14C production rate can be based on geomagnetic field records [Laj et al., 2002]. Another approach is to use other radionuclide records from ice cores to estimate the 14C production rate history [Muscheler et al., 2004]. Especially 10Be records from the polar ice caps have the potential to document solar and geomagnetic field changes [see, e.g., Beer et al., 1990; Wagner et al., 2000].

Both approaches to estimate the 14C production rate have advantages and disadvantages. As mentioned 10Be records theoretically include all potential changes that also influence the 14C production rate. However, climatic influences on these records could partly mask the production signal. In the cases of the Greenland Summit 10Be records it is clear that the variable accumulation rate has a dominant influence on the 10Be concentration [Wagner et al., 2001]. At least during fast climate fluctuations in the northern hemisphere, the so-called Dansgaard/Oeschger events [Johnsen et al., 1992], this obvious climatic influence can be at least removed partly by calculating the 10Be flux [Yiou et al., 1997; Muscheler et al., 2000; Wagner et al., 2001]. However, there still could be an influence of changes in atmospheric 10Be transport or deposition on the 10Be flux [Field et al., 2006]. In addition, uncertainties in the accumulation rate estimates directly translate into uncertainties in the 10Be flux.

Geomagnetic field reconstructions could also contain uncorrected climatic influences as for example due to a climatic influence on the acquisition of remanent magnetization by sediments [Kok, 1999]. In addition, since such records are also influenced by higher moments of the geomagnetic field, which are of minor importance for the 14C production rate, several records are usually combined to remove or diminish local effects [Laj et al., 2000; Yang et al., 2000]. There are no continuous records for the last 60 kyr and different data sets have to be normalized and combined in order to use such a compiled record to estimate changes in the 14C production rate. Differences between different records (e.g., in the chronology) lead to relatively large uncertainties also in the geomagnetic field-based reconstruction of the 14C production rate [Laj et al., 2002, 2005]. On longer timescales the 10Be and geomagnetic field-based 14C production rate reconstructions agree well. Nevertheless, there are differences that affect the interpretation of the causes for the Δ14C changes due to the high sensitivity of Δ14C to small production rate changes [Muscheler et al., 2004].

Another part of the climate system that changes atmospheric 14C concentrations is the carbon cycle itself. Thus one caveat in the interpretation of atmospheric Δ14C variations over the past thousands of years was always the lack of precise knowledge of changes in the carbon cycle. Some 14C studies proposed significant changes in the global carbon cycle [e.g., Beck et al., 2001; Hughen et al., 2004; Muscheler et al., 2004]. Especially changes in ocean circulation, gradually over time, or abrupt declines in northern deep water formation were proposed as hypotheses to explain either the long-term trend or millennial-scale variability in Δ14C. The positive anomaly in atmospheric Δ14C during the Younger Dryas cold events (∼12 kyr BP), for example, was suggested to be caused by a reduction of the Atlantic thermohaline circulation (THC) [e.g., Marchal et al., 2001; Delaygue et al., 2003]. However, most of these model-based scenarios were not challenged to reconstruct atmospheric CO2 concentrations. As CO2 was measured in Antarctic ice cores 650 kyr back in time [Fischer et al., 1999; Petit et al., 1999; Indermühle et al., 1999, 2000; Monnin et al., 2001; Kawamura et al., 2003; Siegenthaler et al., 2005] with an uncertainty of several ppmv only, it can be understood as the ultimate paleo record. Therefore any interpretation of paleo records which assume massive climatic changes and as consequence impacts on the global carbon cycle should be discussed with respect of their implications for the temporal evolution of atmospheric CO2.

During the last four glacial cycles atmospheric CO2 rose from its minima of 180 ppmv during glacial cold periods to its preindustrial maximum of 280–300 ppmv during interglacials [Fischer et al., 1999; Petit et al., 1999; Monnin et al., 2001; Kawamura et al., 2003]. The older CO2 data measured recently in the EPICA Dome C ice core show similar glacial but lower interglacial values of about 250 to 260 ppmv connected to cooler interglacials during that time [Siegenthaler et al., 2005]. The interpretation of this glacial/interglacial rise in CO2 was challenging the scientific community throughout the last two decades. Various hypothesis were tested with models of different complexity to explain these observations (see reviews by Archer et al. [2000] and Sigman and Boyle [2000]), but a widely excepted interpretation remained elusive. The most recent approach [Köhler et al., 2005a] applied a global carbon cycle box model for the first time in a transient mode and reproduced observed variations in atmospheric CO2 and its carbon isotopes over the last glacial/interglacial transition from the LGM to the Holocene. In a second application forced with a different extended data set this approach was able to reproduce the low-frequency changes observed in CO2 over the last 650 kyr (r2 = 0.75) [Köhler and Fischer, 2006b; Wolff et al., 2005]. Accordingly, this model study was able to identify important processes contributing to the glacial/interglacial change in CO2 and proposed a scenario which might explain the observed changes in the carbon cycle. However, due to the simplicity of the model and the data uncertainties, alternative scenarios explaining atmospheric CO2 for other reasons cannot be excluded. We therefore believe that carbon cycle dynamics over the last eight glacial cycles can nowadays be estimated by applying the same modeling approach. Thus one main caveat in the interpretation of atmospheric 14C might be overcome and new conclusions can be drawn from a reinterpretation of the radionuclide records.

In the following we will use the carbon cycle box model of Köhler et al. [2005a] to estimate the fraction of Δ14C variability based on carbon cycle dynamics. We will first introduce the model and the temporal forcing used here over the last glacial cycle, and describe then how 14C production rates were calculated from 10Be or the geomagnetic field strength. We show simulation results of the global carbon cycle and compare reconstructed CO2 with ice core data for the last 120 kyr. This will be followed by simulated changes in Δ14C based on either constant or variable 14C production rates. We discuss our simulation results with respect to the available Δ14C data covering the last 50 kyr, but concentrate on the later half of this time window (26 kyr) due to the high uncertainties in the measurements. Our modeling approach is focused on glacial/interglacial variability. Therefore impacts on both atmospheric CO2 and Δ14C caused by a largely reduced THC in the Atlantic on a multicentennial to millennial timescale cannot be addressed straight forward with our model. However, its potential impacts on the carbon and 14C cycle will be investigated in a detailed sensitivity analysis and discussed within the range of results coming from other studies.

2. Methods

2.1. Carbon Cycle Box Model BICYCLE

BICYCLE is a global carbon cycle box model including a globally averaged atmospheric box and modules of the terrestrial and oceanic components of the carbon cycle (Figure 2). Carbon and the carbon isotopes 13C and 14C are prognostic variables of the model. Perturbations of the climate signals driving our model from the preindustrial situation are prescribed using available information obtained from various paleoclimatic records derived from ice cores and sediment records. The model descriptions of the terrestrial [Köhler and Fischer, 2004] and oceanic [Köhler et al., 2005a] parts were published previously. The model was used recently to participate in the EPICA challenge as the most complex approach [Wolff et al., 2004, 2005; Köhler and Fischer, 2006b] to propose variations in atmospheric CO2 in the pre-Vostok period prior to the publication of the new EPICA Dome C CO2 data [Siegenthaler et al., 2005]. This application of BICYCLE on the long timescale of the EPICA challenge (740 kyr) shows that the process understanding gained from Termination I seems to be sufficient to reconstruct low-frequency changes in the global carbon cycle during the late Pleistocene. A downloadable version of the BICYCLE model is in preparation.

Figure 2.

A sketch of the BICYCLE model including boundary conditions and preindustrial ocean circulation fluxes (in Sv = 106 m2 s−1) in the ocean module. Ocean circulation fluxes which change over time (NADW formation and subsequent fluxes, Southern Ocean vertical mixing) are depicted in bold arrows. The globally averaged terrestrial biosphere distinguishes ground vegetation following different photosynthetic pathways (C4, C3), nonwoody (NW), and woody (W) parts of trees, and soil compartments (D, FS, SS) with different turnover times.

It should be noted that carbon dioxide measured in ice cores is in general given as volume mixing ratio in parts per million by volume (ppmv), while carbon cycle models calculate the atmospheric partial pressure (pCO2 in μatm). Only in dry air and at standard pressure, they are numerically equal [Zeebe and Wolf-Gladrow, 2001]. For reasons of simplicity, we use throughout this article for both carbon dioxide data and simulation results the first nomenclature (CO2 in ppmv) and assume equality between both. This simplification neglects a relatively constant offset between both quantities of a few ppmv.

2.2. Time-Dependent Forcing of the Model

The model is perturbed from the preindustrial reference state by time-dependent boundary conditions based on various paleoclimatic records (Figure 3). Although the focus of this study is on the last 50 kyr (for which data-based reconstructions of atmospheric Δ14C exists), we apply all forcings for a full glacial cycle of approximately 120 kyr in order to investigate how changes in the carbon cycle as seen in the atmospheric CO2 records are covered with our modeling approach. We shortly introduce the paleoclimatic archives used here, but refer the reader interested in more details to earlier publications [Köhler and Fischer, 2004; Köhler et al., 2005a]. The model is equilibrated for 3 kyr into a steady state (for 50 kyr if an increased 14C production rates was assumed initially) for the environmental conditions found in 123 kyr BP and then run forward in time using the following time-dependent forcings. All paleoclimatic records were used on their individual age scales.

Figure 3.

Paleoclimatic records which were used to force the BICYCLE model (Figures 3a–3g) and measured and simulated CO2 (Figure 3h). (a) Planktic δ18O of ODP677 [Shackleton et al., 1990] (1°12′N, 83°44′W). (b) Changes in sea level [Bintanja et al., 2005] based on stacked benthic δ18O compiled by Lisiecki and Raymo [2005]. (c) Changes in the depth of the Pacific lysocline [Farrell and Prell, 1989]. (d) Changes in atmospheric temperature over land in the northern hemisphere (40–80°N) [Bintanja et al., 2005]. (e and f) Original (thin) and sea level corrected (bold) δ18O from (E) the Greenland ice core of NorthGRIP [NorthGRIP Members, 2004] and (f) the deuterium δD record from the Antarctic ice core at EPICA Dome C [EPICA Community Members, 2004]. (g) Atmospheric dust content as measured in the EPICA Dome C ice core [EPICA Community Members, 2004]. (h) Measured (points, 10 points running mean through all measurements: dashed line) and simulated (solid line) atmospheric CO2. Data from Vostok (blue circles) [Petit et al., 1999] plotted on the orbitally tuned age scale [Shackleton, 2000], Dome Fuji (green diamonds, open symbols: dry extraction, closed symbols: wet extraction techniques) [Kawamura et al., 2003], Taylor Dome (cyan squares) [Indermühle et al., 1999, 2000] on the timescale of Brook et al. [2000], EPICA Dome C (red triangles) [Monnin et al., 2001, 2004]. Data prone to high-frequency fluctuations (Figures 3a, 3e, 3f, and 3g) were used as 3 kyr running means.

2.2.1. Sea Level

We take the modeling results of Bintanja et al. [2005] for variations in sea level (Figure 3b). They are based on a compiled stack of 57 globally distributed benthic δ18O records [Lisiecki and Raymo, 2005]. This approach neglects fast fluctuations caused by meltwater pulses.

2.2.2. Ocean Temperatures

Ocean temperature changes are taken from isotopic temperature proxies as recorded in ice core records drilled in neighboring continental ice sheets in the high latitudes and as recorded in planktic and benthic foraminifera for low latitudes and deep ocean boxes. For equatorial surface waters δ18O data from ODP677 [Shackleton et al., 1990] (1°12′N, 83°44′W) were taken to change SST scaled to a glacial/interglacial amplitude of 3.75 K [Visser et al., 2003] (Figure 3a). Temperature changes in the intermediate and deep sea boxes are forced by data from Labeyrie et al. [1987] (not shown). Northern high-latitude SST were changed by δ18O from the Greenland ice core NorthGRIP [NorthGRIP Members, 2004] (Figure 3e), while the Southern Ocean SST was forced by the deuterium δD record from the Antarctic EPICA Dome C ice core [EPICA Community Members, 2004] (Figure 3f). The two ice core temperature proxies were corrected for their embedded sea level change information [Jouzel et al., 2003] and normalized to a glacial/interglacial amplitude in SST of 4 K [e.g., Pflaumann et al., 2003; Becquey and Gersonde, 2003].

2.2.3. Gas Exchange Rates

Changes in sea ice cover in the high latitudes will change the gas exchange rates there. Estimates of annual averaged sea ice cover change from their preindustrial 10 × 1012 m2 in each hemisphere [Cavalieri et al., 1997] to 14 × 1012 m2 in the North and 22 × 1012 m2 in the South [Crosta et al., 1998a, 1998b; Sarnthein et al., 2003; Gersonde et al., 2005] are linearly coupled to variations in SST in the North Atlantic and Southern Ocean surface boxes in our model.

2.2.4. Ocean Circulation

Measurements of the age difference between coexisting benthic and planktic foraminifera from western equatorial Pacific deep-sea cores suggest that the glacial overturning in the Pacific Ocean was comparable to the present-day situation [Broecker et al., 2004]. Data- and model-based evidence, however, suggests a different circulation scheme in the Atlantic and the Southern Ocean during glacial times [e.g., Meissner et al., 2003; Hodell et al., 2003; Knorr and Lohmann, 2003; McManus et al., 2004; Watson and Naveira-Garabato, 2006]. Glacial/interglacial changes in ocean circulation implemented in our scenarios are (1) a reduced glacial strength of the North Atlantic Deep Water (NADW) formation (10 Sv versus 16 Sv in preindustrial times), triggered by the North Atlantic SST proxy (glacial strength of NADW formation once δ18O in NorthGRIP falls below −37‰, Figure 3e), and (2) reduced glacial mixing in the Southern Ocean (31% of preindustrial exchange flux). The changes in the latter are linearly coupled to variation in SST in the Southern Ocean. Fluxes which change over time are depicted in bold blue arrows in Figure 2.

2.2.5. Iron Fertilization of the Marine Export Production

In the preindustrial reference scenario, we prescribe an upper biological export production of organic matter at 100 m water depth of globally 10 PgC yr−1 [Gnanadesikan et al., 2002] which is coupled via a constant rain ratio to the export of 1 PgC yr−1 of CaCO3. This leads to un-utilized macro-nutrients in the Southern Ocean. In our model, these macro-nutrients are mobilized for the supply of additional export production once the dust concentration in the EPICA Dome C ice core [EPICA Community Members, 2004] (Figure 3g) exceeds 310 ppbv, a threshold which is deduced from carbon cycle dynamics seen during Termination I [Köhler et al., 2005a]. Here, dust is an indirect proxy for the iron concentrations and forces the marine biota via the iron fertilization hypothesis [Martin, 1990; Ridgwell, 2003]

2.2.6. Terrestrial Carbon Storage

Changes in the carbon storage in the terrestrial biosphere depend on the variations in the internal calculated CO2 concentration and the global average temperature. The latter consists of a 3:1 mixture (representing the latitudinal distribution of land area excluding Antarctica) of hemispheric wide temperature estimates with amplitudes of 8 K in the North and 5 K in the South [Kutzbach et al., 1998]. Here we take again the sea level corrected EPICA Dome C δD temperature proxy for the South [EPICA Community Members, 2004] (Figure 3f) and modeling results from Bintanja et al. [2005] for the North (Figure 3d). One example of an ensemble of possible forcing combinations was chosen here, when terrestrial carbon storage increases by more than 1000 PgC during the last glacial/interglacial transition, which is on the upper range proposed by other studies. The terrestrial module and scenarios with different amplitude of the glacial/interglacial change were analyzed elsewhere [Köhler and Fischer, 2004, 2006b].

2.2.7. CaCO3 Chemistry

All changes in the global carbon cycle lead to changes in the CO32− ion concentration ([CO32−]) in the deep ocean, which would then lead to variations in the sedimentation and dissolution rates of CaCO3 and thus to fluxes of dissolved inorganic carbon (DIC) and alkalinity between the deep ocean and the sediments, a process known as carbonate compensation [Broecker and Peng, 1987]. In the absence of a process-based model of early diagenesis, which would calculate these fluxes, we prescribe the observed variations in the depth of the Pacific lysocline [Farrell and Prell, 1989] as another boundary condition (Figure 3c). The lysocline is the oceanic depth below which sedimentary calcite dissolves and is approximated here with the saturation depth of calcite, which is calculated as a function of depth (pressure) in vertical steps of 200 m and interpolated in-between. The lysocline in the Atlantic and Southern Ocean [Crowley, 1983; Howard and Prell, 1994] varied in its depth differently than in the Pacific, but sensitivity studies with our model [Köhler and Fischer, 2006a] have shown that these differences are of minor importance on the proposed atmospheric CO2 and Δ14C and are therefore neglected. From these additional boundary conditions the model generates DIC and alkalinity fluxes in order to bring changes in the simulated lysocline in line with the observations. This approach calculates net effects on the overall budgets of DIC and alkalinity and thus implicitly includes terrestrial weathering inputs of bicarbonate [Munhoven, 2002]. A data-based study [Vecsei and Berger, 2004] estimated that the growth of coral reefs during the last 20 kyr mainly occurred during the Holocene warm period and was especially restricted to times, in which the sea level was higher than 70 m below present. Its impact on the glacial/interglacial rise in CO2 in the context of our model was investigated to be of minor importance and is neglected here. An approach which incorporates implicitly a temporal delay of the sedimentary response to changes in deep ocean [CO32−] as proposed by data and models [Archer et al., 1997; Marchitto et al., 2005] is used in an updated version of the model [Köhler and Fischer, 2006b]. The consequences of this improvement on carbon and 14C are discussed in section 4.1.

2.3. Inferring Data-Based Information on the 14C Production Changes

Under the assumption that our knowledge on natural changes in the global carbon cycle has improved through the studies performed with the BICYCLE model, we can now calculate its implications on the 14C cycle.

We first simulate carbon cycle dependent changes in atmospheric Δ14C with BICYCLE based on constant 14C production rates of either the modern level (scenario S1) or at twice the modern level (S2).

A third scenario (S3) is based on inferred changes calculated from 10Be measured at the GRIP and GISP2 ice cores [Muscheler et al., 2004, 2005] covering the last 60 kyr (Figure 4). Finally, a recent high-resolution reconstruction of the geomagnetic field strength (GLOPIS-75) covering the last 75 kyr [Laj et al., 2005] is taken as an alternative basis for the time-dependent changes in the 14C production rates (S4). The 14C production rates are calculated after Masarik and Beer [1999] using the normalized relative geomagnetic field strength. As described in Muscheler et al. [2004] the GISP2 timescale used for parts of the 10Be record and originally for GLOPIS-75 was transferred to the GRIP timescale by synchronizing the δ18O data sets of the GRIP and GISP2 ice cores.

Figure 4.

Time-dependent 14C production rates as used in different simulation scenarios. Both the time-dependent 14C production rates based on 10Be and GLOPIS-75 are synchronized to the GRIP age scale. Shaded background shows the original data (light gray: area covered by lower and upper range based on 10Be; dark gray: GLOPIS-75 (mean ±1 SD)) which are used for the simulations. Also plotted for clarity are a low-pass filtered version of the data sets (cutoff frequency = 1/3000 yr−1) (yellow: 10Be; blue: GLOPIS-75). The blue curve is overlayed on the yellow one. The thin yellow lines in the blue area mark the lower or upper end of the range based on 10Be in the case of an overlap of the two curves.

For both time-dependent approaches the range covered by either an lower and upper estimate (10Be) or the standard deviation (GLOPIS-75) is used in our simulations. Two alternatives are proposed to analyze the importance of the initial 14C production rates on the simulation results. First, the 14C production rates between 123 kyr BP and the start of the reconstruction (60 kyr BP in S3; 75.8 kyr BP in S4) is set to modern values. This initial 14C production rate is increased to 1.5 × modern values in a second set of experiments (S3x, S4x).

For our simulations we use the original reconstructions of the relative 14C production rates based on 10Be and the GLOPIS-75 data sets, but for comparison and discussion a low-pass filtered version (cutoff frequency = 1/3000 yr−1) is much more robust (Figure 4). GLOPIS-75 is based on a stack of 24 records, while the 10Be-based records consists of the combined data sets from two ice cores. The short-term variability in the original data is larger in 10Be than in the GLOPIS-75 record. The low-pass filtered reconstructions of the relative 14C production rates vary between 0.8 and 2.4, but fluctuate most of the time between 1.0 and 1.5. They match each other remarkably well in the timing and the amplitude of the Laschamp event (∼39 kyr BP at the GRIP ss09sea age scale), in which they rise to their maximum values of 2.0. Apart from one time window around 22 kyr BP the two reconstructions agree within their uncertainties or are larger if based on the 10Be reconstruction than on the geomagnetic field. As discussed in detail by Muscheler et al. [2004], Δ14C is very sensitive to small differences in the 14C production rate. Long-term differences by 10% lead to modeled Δ14C differences of 100‰.

Regarding the 10Be-based 14C production rate, changes in climate could influence the 10Be deposition and the assumption that it reflects global changes in the 14C production rate could introduce errors. In the case of the geomagnetic-field based estimate of the 14C production rate, it is possible that the choice and normalization of the included geomagnetic field records, long-term changes in solar activity or long-term changes in the galactic cosmic ray flux produce departures from the real history of the 14C production rate. Therefore we use both approaches and a comparison of the results might give us information about possible biases in the assumed 14C production rates.

2.4. Impact of Millennial-Scale Variability in the Ocean Circulation on the Carbon Cycle and on 14C

Various data- and model-based studies support the hypothesis that the ocean circulation, and here especially the North Atlantic Deep Water (NADW) formation was reduced during glacial times, and probably (nearly) completely shutdown during Heinrich events [e.g., Rahmstorf, 2002; McManus et al., 2004]. In the more general interpretation of North-South teleconnection patterns with the concept of a bipolar seesaw each switch from an interstadial to a stadial period during millennial-scale climate oscillation of the Dansgaard/Oeschger events is connected with a reduced THC in the Atlantic Ocean [Broecker, 1998; Stocker and Johnsen, 2003; Knutti et al., 2004; EPICA Community Members, 2006].

In a forward model these millennial-scale variability in NADW formation can only be analyzed if especially the ice core records from Greenland and Antarctica are synchronized. Major improvements in this synchronization effort were made within the last years [e.g., Blunier et al., 1998; EPICA Community Members, 2006], but so far the various records used here (especially NorthGRIP and EPICA Dome C) are taken on their individual timescales.

We therefore analyze the impacts of abrupt changes in the northern deep water production on both atmospheric CO2 and Δ14C in a sensitivity study. We reduce the strength of the NADW formation (to a residual strength of 0, 2, 4, 6, or 8 Sv). Furthermore the length of the reduced overturning (500, 1000, 1500, 2000 years) and the 14C production rates (constant at modern or 2 × modern level) are varied. It has to be noted that changes in NADW formation do not imply any additional changes in ocean temperature or sea ice formation. These processes are (in contrast to ocean general circulation models (OGCMs)) decoupled in our model. However, the marine export production is depending on nutrient availability in the surface waters and might be reduced during a reduced THC due to macro-nutrient depletion.

We start these experiments from three different initial conditions. Two sets of experiments are performed for steady state background situations with all other processes held constant at their preindustrial level, no CaCO3 exchange fluxes between deep ocean and sediment are considered here. The experiments differ only in the circulation scheme of the ocean, either an interglacial circulation as depicted in Figure 2 (NADW: 16 Sv; maximum net vertical mixing (20 Sv) in the Southern Ocean), or a glacial circulation (NADW: 10 Sv; minimum net vertical mixing (0 Sv) in the Southern Ocean). The third set of experiments is performed during transient simulations with all other processes which effect the carbon cycle active. The reduced strength of the NADW formation starts at 30 kyr BP. The chosen time period falls in a relatively stable climate period and corresponds approximately to the timing of the Heinrich 3 event [Hemming, 2004] and stadial 5 on the ss09sea timescale in the NorthGRIP and GRIP ice cores [Dansgaard et al., 1993; NorthGRIP Members, 2004].

3. Results

Our approach to reinterpret observed atmospheric Δ14C data is based on process understanding gained from the interpretation of the global carbon cycle. Dynamic variations in the carbon cycle as proposed by our model are therefore in our focus before we expand our interpretation to the 14C cycle.

3.1. Carbon Cycle Reconstruction and Atmospheric CO2

In our standard simulation S1 over the last 123 kyr atmospheric CO2 decreases gradually from its interglacial value of 280 ppmv before 120 kyr BP toward 230 ppmv around 105 kyr BP (Figures 3h and 5b). CO2 then slowly decreases to 210 ppmv over most of the glacial period, before it sharply drops to its full glacial minima of 185 ppmv at 30 kyr BP. The slow decline is interrupted by millennial scale variability of up to 10 ppmv and a large negative excursion of 30 ppmv between 70 and 60 kyr BP. Between 18 and 10 kyr BP (Termination I) CO2 rises by 85 ppmv within less than 10 kyr toward Holocene values. If compared with the available CO2 data from Antarctic ice cores (Figure 3h), our simulation covers the low-frequency variations rather well (correlation of the simulation results and the 10-points running mean: r2 = 0.89) while fast fluctuations during MIS 3 and the atmospheric rise in CO2 in the late Holocene are not matched quantitatively in our experiment. Timing inconsistencies between our simulation results and the data during the glaciation and the peaks around 80 kyr BP (Figure 3h) might be due to the use of unsynchronized paleoclimatic records to force our model as mentioned in the section 2.2.

Figure 5.

Simulation results for constant 14C production rates (scenario S1). (a) Estimates of the individual contribution of different processes (ocean temperature, sea level, sea ice, NADW formation, Southern Ocean (SO) vertical mixing, iron (Fe) fertilization, terrestrial biosphere, CaCO3 chemistry) on changes in atmospheric CO2. Estimates are based on the differences of results of the standard scenario combining all processes and simulation with changes in all but the one process in question. (b) Transient dynamic of CO2 in the standard scenario S1. (c and d) Same as in Figures 5a and 5b, but for atmospheric Δ14C.

The time-dependent contributions of eight processes to the rise in CO2 are estimated by the difference from our standard simulation with experiments in which changes in the processes in question are switched off (Figure 5a). However, due to the high nonlinearity of the system this is only a first estimate, and all individual contributions derived by this method will not add up to the dynamics seen in the standard scenario. Changes in terrestrial carbon storage, sea level and sea ice are increasing glacial CO2 by up to 10–20 ppmv each. Variations in the strength of the NADW formation contribute a rather constant 20 ppmv to the glacial reduction in CO2, followed in amplitude by the effect of colder glacial ocean temperatures (25 ppmv). In the contribution of the latter, however, a gradual temporal increase in the CO2 down-draw over time can be observed. Iron fertilization operates mainly during two peak windows, around 65 kyr BP and between 30 and 20 kyr BP, including the LGM, reducing CO2 by about 25 ppmv. Southern Ocean mixing and the effect of processes acting on CaCO3 follow very similar temporal pattern, with pronounced small millennial scale variability and maximum amplitudes of 35 and 40 ppmv, respectively.

3.2. Variations in Atmospheric Δ14C Based on Carbon Cycle Dynamics Under Constant 14C Production Rates

If we keep the 14C production rate constant over time, we can now analyze which variations in atmospheric Δ14C can be expected on the basis of changes in the global carbon cycle only. We can even further identify the temporal evolution of the contributions of different processes to the Δ14C dynamics.

The physical and the biological pump in the ocean bring carbon and 14C from the surface to the deep ocean. Because of the two additional processes influencing 14C (production in the atmosphere and radioactive decay), the vertical gradients in the ocean maintained by the oceanic pumps are different: DIC is enriched in the deep ocean and depleted at the surface while the opposite is the case for 14C. A change in the strength of the pumps which are dominating both cycles therefore leads to opposing dynamics in atmospheric CO2 and Δ14C: The overall dynamics of carbon cycle-based variations in Δ14C calculated with constant 14C production rates are in anti-phase with those seen in CO2 (Figure 5d), e.g., atmospheric Δ14C increases during times in which atmospheric CO2 decreases and vice versa.

Throughout most of the time of the last glacial cycles (110–20 kyr BP), Δ14C is elevated by about 30‰ based on carbon cycle dynamics. This background increase is modulated by millennial scale variations of ±10‰ and punctuated by two periods with higher carbon cycle contributions (50–60‰) during a 10 kyr long window around 65 kyr BP and the period between 35 and 18 kyr BP including the LGM. However, the magnitude of these carbon cycle-based variations depends on the chosen modern 14C production rate. If we double this production rate, for example, the carbon cycle-based Δ14C variations increase to 110‰ during their maximum peaks (Figure 6). One has to note that in this scenario the background Δ14C would rise to about 1000‰.

Figure 6.

Carbon cycle-based changes in atmospheric Δ14C based on different but constant 14C production rates in scenarios S1 and S2.

Of course, the 14C production rate cannot be arbitrarily chosen. The constraint on the 14C production rate is that we can reproduce the preindustrial Δ14C values. This can be achieved by normalizing the modeled 14C/12C ratios or by adjusting the absolute 14C production rate to obtain the measured values. However, the discussion above shows that nonlinearities are an important aspect of our calculations. Due to the higher 14C production rate during the last glacial [e.g., Muscheler et al., 2004] the influence of the carbon cycle induced changes on Δ14C at that time is expected to be larger than for the case of the relatively low modern 14C production rate.

The contribution of individual processes to the evolution in Δ14C in scenario S1 with respect to the preindustrial time is estimated with the same approach as for CO2 (subtracting results with the process of interest switched off from results of scenario S1). Again, this is a first order estimate which neglects nonlinearities and the individual contributions to Δ14C (Figure 5c) will not sum up to the overall changes in Δ14C seen in S1 (Figure 5d). SST and sea level do not impact on Δ14C. A reduced glacial terrestrial carbon storage and changes in the CaCO3 chemistry lead to a reduction in Δ14C by up to 25‰ each. The latter is caused by the dilution of the Δ14C signal through the input of carbon depleted in 14C entering the ocean/atmosphere/biosphere system from the sediments. The reduced glacial terrestrial carbon storage implies that oceanic glacial carbon reservoirs are increased. Due to the on average longer turnover time of carbon in the ocean compared to the terrestrial pools (in our model), more 14C is taken up and removed for longer times from the atmosphere. The four remaining processes which all increase glacial Δ14C either restrict glacial gas exchange between the atmosphere and the surface ocean (sea ice: maximum contribution to Δ14C ∼ 20‰), reduce the transport of water and thus 14C from the surface to the deep ocean in glacial times (NADW formation: 15‰; Southern Ocean mixing 30‰), or impact on atmospheric Δ14C through isotopic fractionation during higher marine export production caused by Fe fertilization leaving DIC enriched in 14C in surface waters which would quickly enter the atmosphere through gas exchange (15‰).

For twice as much 14C production (scenario S2) the contribution of the individual processes is approximately doubled (not shown).

3.3. Millennial-Scale Variability of the Atlantic THC

The anomalies in both atmospheric CO2 and Δ14C caused by a large decline in the NADW formation depend on all of the four chosen variables (residual strength of NADW, length of experiment, 14C production rate, background conditions). The results of this sensitivity study are summarized in Table 1, some selected examples of simulated time series are shown in Figure 7. The simulated anomalies vary between −3 to −29 ppmv for atmospheric CO2 and +5 and +105‰ for atmospheric Δ14C.

Figure 7.

Examples for the effect of reduced northern deep water formation on atmospheric Δ14C and CO2. A complete shutdown of the NADW formation from either interglacial ocean circulation (IG THC) or transient experiments at 30 kyr BP with constant 14C production rates (modern or 2 × modern level) was simulated. Four simulations with different length of the shutdown experiment (500, 1000, 1500, 2000 years) are shown.

Table 1. Sensitivity Study on the Effect of a Reduced NADW Formation on Atmospheric Δ14C and CO2a
Length, yearsΔ(Δ14C), ‰, Modern 14C Production RatesΔ(Δ14C), ‰, 2 × Modern 14C Production RatesΔ(CO2), ppmv
Residual NADW, SvResidual NADW, SvResidual NADW, Sv
  • a

    Additionally to the strength of the residual NADW, the length of the experiment, the background ocean circulation and the constant 14C production rates are varied. Maximal anomalies from initial values are summarized here.

Steady State, Interglacial Ocean Circulation
Steady State, Glacial Ocean Circulation
Transient, Start at 30 kyr BP

3.3.1. Atmospheric CO2

A reduced THC leads always to negative anomalies in atmospheric CO2. Less carbon enriched deep water is transported to the surface in the Southern Ocean as a consequence of the reduced THC in the Atlantic Ocean.

In experiments with large residual NADW (6–8 Sv) the anomalies in CO2 are smaller in the glacial ocean than in the interglacial one. This different behavior between small and large residual NADW can be understood by the changing circulation patterns in the deep Atlantic Ocean. In all experiments a flux of 6 Sv is transporting water from the deep Southern Ocean to the deep Atlantic Ocean mimicking the Antarctic Bottom Water (AABW) (Figure 2). For experiments in which the residual NADW is larger than 6 Sv the resulting effect of NADW and AABW is still a net transport from the deep Atlantic to the deep Southern Ocean, while in experiments with smaller residual NADW the opposite is the case. In other words, the reduction in northern deep water production and thus the reduction in the subsequent upwelling of carbon-rich deep water in the Southern Ocean is smaller in the glacial than in the interglacial situation for the experiments with large residual NADW. For those experiments with small residual NADW upwelling in the Southern Ocean does not occur in the glacial ocean (net vertical mixing flux is reduced): no carbon-rich deep waters are entering the Southern Ocean surface box leading to larger drops in CO2 than for the interglacial ocean, in which the deep Southern Ocean is ventilated via its vertical mixing rates. The glacial and the interglacial sets of experiments differ also in the amount in which the export production is reduced due to nutrient depletion. This reduction in marine export of carbon to the ocean interior depends also on the strength of the residual THC and is at maximum 16% in the glacial and 8% in the interglacial situation. A reduced export production leads to outgassing of CO2 and counteracts the effect of the ocean circulation on the atmospheric CO2.

In the transient experiments the anomalies in CO2 are reduced to a fourth to a half (residual NADW: 0–4 Sv) if compared to the glacial background conditions. This reduction of the initial anomalies is mainly caused by reduced marine export (at maximum reduced to a third during a complete THC shutdown) in combination with the negative feedback of the terrestrial biosphere via CO2 fertilization: Lower CO2 leads to less terrestrial photosynthesis, which is then followed by smaller terrestrial carbon storage and a rise in CO2.

The anomalies in CO2 in the transient experiments are small, and additional contributions of other processes are expected. A largely reduced Atlantic THC would lead to a warming in the Southern Ocean [Stocker and Johnsen, 2003], which would then lead to outgassing of CO2 due to a lower solubility of CO2 in warm waters. The North Atlantic region and large parts of Eurasia would experience a cold period. This might lead to a southward migration of the treeline, a loss of carbon in the vegetation, but also to reduced respirational loss of carbon from the soil. During glacials the net effect would be a rise in CO2 by up to 10 ppmv, however, the response depends on the background climate and might also be of opposite sign [Köhler et al., 2005b]. In the marine biosphere, the North Atlantic plankton stocks might collapse and the export production of organic carbon to the deep ocean might drop by 20% [Schmittner, 2005], which would then increase atmospheric CO2. As a consequence of the bipolar temperature anomaly [Stocker and Johnsen, 2003], the sea ice cover in the North might rise while that in the South might shrink. As the North Atlantic is a sink to CO2 while the Southern Ocean is a source, both sea ice anomalies would therefore lead to rising atmospheric CO2 of a few ppmv. Terrestrial photosynthesis is directly dependent on the atmospheric concentration of carbon dioxide. Consequently, the effects of all these processes would not linearly add up to a combined effect but interact via existing feedbacks. Nevertheless, the order of magnitude deduced from these pieces of evidence agrees with the observed anomalies in CO2 of +10–20 ppmv as seen between 60 and 30 kyr BP [Indermühle et al., 2000], which could be explained by the combined effects initiated by a shutdown of the Atlantic THC [e.g., Köhler et al., 2005b; Schmittner, 2005].

3.3.2. Atmospheric Δ14C

The most important variable in this sensitivity study for the evolution of atmospheric Δ14C is the residual strength of the NADW formation. A residual NADW of 8 Sv would imply that the THC is reduced only by 2 Sv in the glacial and in the transient experiment leading to a positive anomaly in Δ14C of 5 to 20‰, while a complete shutdown of the NADW will result in +29 to +105‰. The less northern deep water is produced the more 14C has to stay in the atmosphere.

The size of the anomalies rise with the length of the reduction in northern deep water production. In the shortest experiments (500 years) simulating a complete shutdown only about a half to two thirds of the anomalies gained in the 2000 years experiments are found. This dependency on the length of the experiment is weaker if a larger residual northern deep water production (up to 8 Sv) is allowed.

Twice the modern 14C production rate leads also to approximately twice as large perturbations in Δ14C.

For a small residual NADW (0–4 Sv) the impacts on atmospheric Δ14C are smallest for interglacial ocean circulation, but the effects are twice as large if the experiments are started from either the glacial circulation or during the transient experiments. In experiments with larger residual NADW the effect of the ocean circulation is marginal.

For the comparison of the interglacial and the glacial ocean circulation, the Δ14C signal is tightly coupled to the carbon cycle and CO2. If a similar reduction of the NADW in the glacial ocean leads to a larger (smaller) drop in CO2 than in the interglacial ocean, a similar trend but with opposite sign is observed for Δ14C. A drop in CO2 is caused by reduced upwelling of carbon-rich and 14C-poor deep waters in the Southern Ocean. This process is more important for atmospheric CO2 than the change in the deep water production and thus the downwelling of carbon in the North Atlantic itself, probably because of the four times larger areal extent of the Southern Ocean in our box model configuration. Deep waters are further enriched in carbon and depleted in 14C in a glacial ocean with its reduced overturning compared to the interglacial situation. The loss of the upwelling deep waters through a complete shutdown of the THC in the Atlantic is therefore more important for the carbon and 14C budgets at the surface in the glacial than in the interglacial climate.

The situation is more difficult for the transient experiments. The anomalies in Δ14C are similar to those obtained with glacial circulation, because the combined effect of the additional processes enriching Δ14C at 30 kyr BP (sea ice, iron fertilization) compensate for the effects of those depleting Δ14C at that time (terrestrial biosphere, CaCO3 chemistry) (Figure 5c).

The additional effects impacting on the carbon cycle which would follow a collapse of the Atlantic THC (bipolar anomalies of temperature and sea ice, marine and terrestrial biology) have also to be considered for the 14C cycle. Temperature changes are of minor importance for 14C, the opposing sea ice anomalies in both hemispheres might lead to a neglecting effect, and the reduced photosynthesis in both the marine and terrestrial biosphere would lead to a slight damping of the Δ14C anomalies.

We conclude that a complete shutdown of the NADW formation might at maximum lead to a peak of +100‰ in Δ14C. However, realistic 14C production rates were usually smaller than the 2 × modern values needed to achieve this result. A further damping of the amplitude would be caused by the expected changes in the biospheric response to the THC collapse. Therefore, on the basis of our model understanding, amplitudes of up 50‰ seemed to be realistic. These results are in line with those received recently with an OGCM. In a freshwater perturbation experiment, Butzin et al. [2005] reduced the NADW export at 30°S from 9 Sv to 1 Sv which led to a rise in atmospheric Δ14C of 40‰.

A reduced NADW formation would also lead to positive anomalies in atmospheric δ13C of CO2. This δ13C record derived from ice cores [Smith et al., 1999] is only coarsely resolved and is only going back in time as far as 26 kyr BP. However, the δ13C signal indicates that during the Younger Dryas cold event, also known as stadial 1, (∼12.7 to 11.5 kyr BP) the NADW formation was not largely reduced [Köhler et al., 2005a, 2006].

3.4. Impact of Time-Dependent 14C Production Rates

The initial 14C production rate significantly affects Δ14C for the first 20 kyr in which the 14C production varies (Figure 8). This conclusion can be drawn from both the comparisons of the scenarios driven by 10Be (S3, S3x) and those driven by the geomagnetic field (S4, S4x). The 10Be record and thus the variable 14C production rates cover the last 60 kyr BP, while the geomagnetic field reconstruction goes 75 kyr back in time. Simulations S3 and S3x therefore differ by 500‰ for constant 14C production rates (before 60 kyr BP), converge between 60 and 40 kyr BP, and are nearly identical thereafter. The same is happening for S4 and S4x 15 kyr earlier (500‰ difference before 75 kyr BP, converging between 75 to 55 kyr BP, nearly identical thereafter). This initial effect has to be kept in mind if simulation results are compared with data-based reconstructions.

Figure 8.

Simulated atmospheric Δ14C based on different 14C production rates. Simulations S3 and S4 are initialized with modern 14C production rates; S3x and S4x are initialized with 1.5 × modern 14C production rates. The results of the four scenarios S3, S3x, S4, and S4x span wider ranges because of the uncertainties in the time-dependent 14C production rates shown in Figure 4. Curves are overlayed on each other (S4x on top of S4, S3x on top of S3, both S4 and S4x on top of S3 and S3x). Upper and lower ends of the ranges of the underlying curves are then indicated by the thin lines.

The range of the simulated Δ14C based on 10Be is about 300‰ wide at 50 kyr BP, which is reduced to 200‰ for recent times (Figure 9). The 10Be-based range of variability is inferred from different combinations of the 10Be data from the GRIP and GISP2 ice core. Due to the reservoir effect and the 14C decay Δ14C decreases caused by a diminished 14C production rate are relatively slow [Muscheler et al., 2004]. This means that the climate system cannot transfer within a short period of time from a state with atmospheric Δ14C at the upper end of the range given by the uncertainty of the 10Be-based 14C production rate to a state with Δ14C at the lower end of this range. In other words, even if the Δ14C error range comprises 300‰, the 10Be-based 14C production rate cannot explain most of the steep Δ14C decrease during the last deglaciation.

Figure 9.

Comparison of the simulated atmospheric Δ14C based on different 14C production rates during the past 50 kyr with the INTCAL04 data set [Reimer et al., 2004]. The results of the scenarios S3x and S4x span wider ranges because of the uncertainties in the time-dependent 14C production rates shown in Figure 4. Curves are overlayed on each other (S4x on top of S3x). Upper and lower ends of the ranges of the underlying S3x are then indicated by the thin yellow lines.

The smaller uncertainty in the 14C production rates based on GLOPIS-75 leads to Δ14C results with uncertainties of only ∼200‰ throughout the last 50 kyr. In both simulations (S3, S4) atmospheric Δ14C declines by ∼400‰ over the last 38 kyr. The lowest GLOPIS-75-based Δ14C values are for most of the simulation time about 100‰ lower than the range covered by 10Be-based scenarios. For the time period 50 kyr BP to 16 kyr BP the scatter in the atmospheric Δ14C data is still large (Figure 1). For reasons of simplicity we compare our simulation results only with the INTCAL04 record going 26 kyr back in time [Reimer et al., 2004] (Figure 9). The large variability spanned by our simulation results over the last 50 kyr is in agreement with the variability found in the data sets, only certain data sets with Δ14C larger than 700‰ fall out of the range obtained by our results (e.g., Beck et al. [2001], prior to 35 kyr BP). The GLOPIS-75-based scenario is not covering the INTCAL04 data within its uncertainty range prior to 15 kyr BP, while the 10Be-based scenario is underestimating atmospheric Δ14C only for the earliest part (26–24 kyr BP). After 10 kyr BP the general trend seen in Δ14C in INTCAL04 is covered by the simulations based on GLOPIS-75, while those based on 10Be are slightly higher than the data set. The general trend of a reduction in atmospheric Δ14C seen in the data sets is covered by our scenarios, however, the gradient of ∼500‰ over the last 30 kyr is not covered in any of our scenarios.

Millennial-scale climate variability caused by a reduced Atlantic THC might be responsible for the sharp positive anomalies seen in the Δ14C data sets (e.g., of +50 to +100‰ at 12.5, 18.5, 22 kyr BP in INTCAL04). These changes in ocean circulation are generally assumed to have been taken place during interstadial/stadial climate transitions, especially during the beginning of the Younger Dryas or Heinrich events [e.g., Stocker and Johnsen, 2003; EPICA Community Members, 2006], they might also be caused by meltwater pulses [Clark et al., 2002]. According to our model and the study of Butzin et al. [2005] peaks up to approximately 50‰ might be caused by reduced northern deep water production. However, while the peak at 12.5 kyr BP might fall together with the onset of the Younger Dryas, those at 18.5 and 22 kyr are not connected with stadial climate conditions seen in Greenland ice cores [e.g., NorthGRIP Members, 2004] or a known Heinrich event. According to Hemming [2004] Heinrich events 1 (16.8 kyr BP) and 2 (24 kyr BP) were about 2 kyr earlier respectively later than these Δ14C anomalies. While the peak at 18.5 kyr BP falls together with an abrupt rise in sea level and thus a meltwater pulse at 19 kyr BP [Clark et al., 2004], which might nevertheless indicate a change in the Atlantic THC as cause for this anomaly, similar details on sea level and subsequent changes in the THC are not retrievable for the peak at 22 kyr BP. All highly resolved reconstructions of sea level during Termination I are based on coral reef terraces and start not before 22 kyr BP [Lambeck and Chappell, 2001].

4. Discussion and Conclusions

In this study we combined proxy-based information on the 14C production rates with simulation results of the global carbon cycle box model BICYCLE to interpret the observed atmospheric Δ14C data during the last 50 kyr. To deepen this interpretation we discuss in the following the model limitations of BICYCLE (section 4.1), and compare our results on Δ14C with other studies (section 4.2), before we end with our final conclusions (section 4.3).

4.1. Model Limitations and the Global Carbon Cycle

Our approach is based on a simple global carbon cycle model. We are therefore able to force our model externally with available proxy data and to run it over hundreds of thousands of years in a reasonable computing time. The more complex a model gets (i.e., in terms of spatial resolution) the more difficult it gets to find appropriate paleoclimatic archives to force it forward in time. This simplicity restricts also the kind of processes which are included in the model in an accurate way. However, the natural variability in the global carbon cycle on glacial/interglacial time scales is covered with our model. We can at least propose a scenario which is plausible and in line with various pieces of evidence from different archives. The same processes as in our simulations have to be at least considered in other simulation scenarios to explain the glacial/interglacial CO2 rise, but the quantitative contributions of individual processes may differ (see more detailed discussion by Köhler et al. [2005a]). More complex scenarios such as the “silicic acid leakage hypothesis” [Matsumoto et al., 2002], which involves a time-dependent and spatial heterogenous change in the distribution of nutrients, species composition and export of organic and inorganic matter, cannot be followed up in details with our model.

For the Holocene we did not implement recent theories which try to explain the rise in CO2 by 20 ppmv during the last 8 kyr [Indermühle et al., 1999], and therefore our simulation results do not match the CO2 data very well. It should be mentioned that these theories are so far not mutually exclusive, but each idea follows different data or model constraints [e.g., Broecker and Clark, 2003; Ridgwell et al., 2003; Ruddiman, 2003; Joos et al., 2004].

It should be noted that our simulation results propose a contribution of an increased glacial marine export production to the higher glacial atmospheric Δ14C due to Fe fertilization in the Southern Ocean. This change in the global carbon cycle leads to a 20 ppmv reduction in CO2 and is in the range estimated from sediment cores [Kohfeld et al., 2005]. Furthermore, it has been argued on the basis of the understanding which emerged from present day Fe fertilization experiments that the contribution of this process to the glacial decline in CO2 is probably less than 1 ppmv and at maximum 14 ppmv [de Baar et al., 2005]. However, these conclusions were based on a study of dust and iron fluxes to Antarctica in which the glacial flux was 11-fold larger than at present [Edwards et al., 1998]. Recent iron flux studies from the EPICA Dome C ice core [Gaspari et al., 2006; Wolff et al., 2006] find a 25- to 34-fold increase in glacial iron flux to Antarctica. This would rise the upper limit of the impact of Fe fertilization as given by de Baar et al. [2005] by more than a factor of two, and we thus would find our simulation results in a range possible from the observations.

To base the carbonate compensation mechanism on changes in the lysocline as done here bears also the uncertainties contained in the reconstruction of the Pacific lysocline. Archer [1991] posed the question if the lysocline data of Farrell and Prell [1989] record more a production or a dissolution signal and raised also concerns that there might be a latitudinal bias in the data. This lysocline approach leads also to a more or less instantaneous response of the carbonate compensation to changes in deep ocean [CO32−], while process-based models of early diagenesis [Archer et al., 1997] and paleo records [Marchitto et al., 2005] indicate that this process is time-delayed with a response time which can be approximated with an e-folding time τ of 1.5 to 6 kyr. In an updated version of the model which takes this temporal delay into account [Köhler and Fischer, 2006b] the contribution of the CaCO3 chemistry to changes in atmospheric CO2 is about half as large as here (τ = 1.5 kyr) and the correlation between simulation results and CO2 data over the last 650 kyr is slightly smaller (r2 = 0.75). This different behavior of the carbon cycle might also reduce the contribution of the CaCO3 chemistry to Δ14C by up to a factor of two. The updated model version [Köhler and Fischer, 2006b] considers also a by a factor of two smaller glacial/interglacial change in the terrestrial carbon storage (500 PgC), which is at the lower end of the range covered by data- and model-based studies [Köhler and Fischer, 2004]. This would also reduce the impact of terrestrial processes on atmospheric Δ14C. The uncertainties within these two processes taken together imply that the overall carbon-based influence on atmospheric Δ14C might be up to 30‰ higher in a simulation with the updated model than in scenario S1.

4.2. Comparison of Our Interpretation of Δ14C With Other Modeling Studies

If we compare our results with other approaches to reconstruct atmospheric Δ14C [Beck et al., 2001; Hughen et al., 2004] some similarities and some differences become apparent. Beck et al. [2001] used a similar multi-box model and combined a few assumptions to match their simulation results to their own observations covering 45–10 kyr BP. According to their study the long-term decline in Δ14C can be explained by changes in the 14C production rates, an increase in carbonate sedimentation rates by one order of magnitude and a gradually increase in the THC from 35 to 11 kyr BP. In Hughen et al. [2004] an even simpler model was applied consisting of seven boxes only. Again several assumptions on changes in 14C production rate, ocean circulation and carbonate sedimentation are combined to be able to explain the long-term dynamics seen in the Δ14C observations. With 100‰ the amplitude in Δ14C caused by changes in the THC is in both studies larger than in our scenarios. We calculate a contribution of 45‰ for changes in ocean circulation under constant modern 14C production rates. If we assume a rise in the glacial 14C production rate by 50% this would enlarge the THC contribution to about 70‰. Interestingly, changes in the gas exchange rates (sea ice) and the marine export production have also important impacts on atmospheric Δ14C in our model while these processes were not considered in the other two studies.

Similar to Beck et al. [2001], the sedimentation rates proposed by our model rise also over Termination I. Our approach of simulating CaCO3 chemistry considers only the net changes between terrestrial weathering inputs and losses to the sediments. The simulated ocean/atmosphere/biosphere system is about 2000 PgC smaller in carbon during interglacial times than during the LGM [Köhler and Fischer, 2006a]. This decline of the total active carbon pools in ocean, atmosphere, and biosphere leads to lower Δ14C (−20‰ at modern 14C production rates) between 35 and 15 kyr BP followed by a rising contribution to Δ14C thereafter (Figure 5c). While the declining trend is also seen in Beck et al. [2001], its contribution to glacial Δ14C is an order of magnitude larger in their study (−300‰). Changes in the shallow water sedimentation rates over Termination I were also important in Hughen et al. [2004] accounting for a decline of 150‰ in Δ14C. Beck et al. [2001] assume a rise in sedimentation rates from 0.24 PgC yr−1 until 25 kyr BP to 2.0 PgC yr−1 at 11 kyr BP, identical to a loss of 15,680 PgC to the sediments, but they give no information about assumed weathering rates. In Hughen et al. [2004] the preindustrial ocean is losing as much carbon to the sediments as it gains from the continental weathering. In the reduced glacial carbon cycle assumed by Hughen et al. [2004] the losses to the sediments are approximately half as large as the gain from weathering. This rise in the sedimentation rates from glacial to interglacial conditions would also lead to a loss of carbon of the order of 5,000–10,000 PgC (depending on the assumed temporal evolution). In all three studies (Beck, Hughen, this study) the amount of 14C contained in both the weathering inputs and the sedimentation and dissolution of CaCO3 might be important for the budgets. Since these numbers are not given in the other studies their importance on the results cannot be assessed, however, we tested the effect of the 14C content of these fluxes on atmospheric Δ14C within our modeling framework. In our study the Δ14C of sedimented CaCO3 is similar to that during production in the surface ocean, while the 14C of dissolved CaCO3 might vary due to its residence time in the sediments. A variation of 14C of this dissolution flux between that of newly built sediments and old sediments without any 14C revealed only slight changes in our overall simulation results. Therefore the amount of carbon which is extracted from the ocean to the sediments in a glacial/interglacial transition is most important for the 14C. However, its variability is large between the three studies (2,000 to 16,000 PgC), but difficult to assess without additional information such as carbon cycle dynamics and impacts on atmospheric CO2.

The choice on which approach the 14C production rate is based on is a very important assumption for the ability of a case study to reconstruct the observed decline in atmospheric Δ14C during the last 30 kyr. In a previous study applying the BICYCLE model over Termination I [Köhler et al., 2005a], we used a record of the geomagnetic field strength based on NAPIS-75 [Laj et al., 2002] to reconstruct changes in the 14C production rates. A remarkable better fit of the simulated Δ14C to the data-based reconstruction was achieved. The main difference between NAPIS-75 and GLOPIS-75 is that NAPIS-75 produces higher 14C production rates between 30 and 20 kyr BP and contains therefore a larger negative gradient over the last 20 kyr. As a result simulated Δ14C based on NAPIS-75 would be higher than those based on GLOPIS-75 at 30 kyr BP and decrease faster bringing the simulations better in line with the observations. NAPIS-75 was also used by Hughen et al. [2004], while the study of Beck et al. [2001] is based on SINT200, a similar geomagnetic stack with high uncertainty and very low temporal resolution. The chosen 14C production rate is also important for the conclusions of Robinson et al. [2005]. They attributed the difference between the solely production-driven variability (based on 10Be) and the reconstruction in atmospheric Δ14C between 26 and 10 kyr BP to changes in the carbon cycle, mainly to changes in ocean circulation.

We challenge our model also to reconstruct atmospheric CO2 and base our assumptions how the carbon cycle might have varied over time as much as possible on data constraints. A similar cross-check of the simulations results was not performed in the other studies, but this would help to assess the results of their studies.

4.3. Conclusions

We have proposed in this study that the changes in the 14C production rate have a dominant influence on the Δ14C record. According to our model a changing carbon cycle during glacial times contributes with less than 100‰ about a 1/6 to the observed atmospheric Δ14C variability during the last 50 kyr. Glacial atmospheric Δ14C larger than 700‰ cannot not be explained within our framework, neither through carbon cycle-based changes nor through variable 14C production. It is not clear which processes might be responsible for these observations. An assumed shutdown of the NADW formation which might have occurred during stadial climate conditions and especially Heinrich events might lead to positive anomalies of ∼50‰ in Δ14C. Our proposed scenario is in line with observed atmospheric CO2. A further constraint for its plausibility might come from a comparison of simulated and measured atmospheric δ13C, a new ice core record which is urgently wanted in high precision and good temporal resolution for a better understanding of the global carbon cycle [e.g., Broecker and Stocker, 2006].


We thank Kenji Kawamura for the Dome Fuji CO2 data, Richard Bintanja for the sea level and temperature change data, and Catherine Kissel and Carlo Laj for the GLOPIS-75 record. We also thank Laurent Labeyrie, Jean Lynch-Stieglitz, Tzu-Chien Chiu, and an anonymous reviewer for their constructive comments, which helped shift the focus of this study and improved its quality. P.K. and H.F. were funded by the German Ministry of Education and Research through project RESPIC of the German Climate Research Programme (DEKLIM). This study was in part sponsored by the Swiss National Science Foundation, NSF, and NASA (RM).