CPClimate of the PastCPClim. Past1814-9332Copernicus PublicationsGöttingen, Germany10.5194/cp-13-1695-2017Simulation of climate, ice sheets and CO2 evolution during the last four glacial cycles with an Earth system model of intermediate complexityGanopolskiAndreyganopolski@pik-potsdam.deBrovkinVictorhttps://orcid.org/0000-0001-6420-3198Potsdam Institute for Climate Impact Research (PIK), Potsdam, GermanyMax Plank Institute for Meteorology, Hamburg, Germanyalso a guest scientist at: Potsdam Institute for Climate Impact Research (PIK), Potsdam, GermanyAndrey Ganopolski (ganopolski@pik-potsdam.de)29November201713121695171627March201719April201715September20179October2017This work is licensed under the Creative Commons Attribution 3.0 Unported License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/3.0/This article is available from https://cp.copernicus.org/articles/13/1695/2017/cp-13-1695-2017.htmlThe full text article is available as a PDF file from https://cp.copernicus.org/articles/13/1695/2017/cp-13-1695-2017.pdf
In spite of significant progress in paleoclimate reconstructions and modelling
of different aspects of the past glacial cycles, the mechanisms which
transform regional and seasonal variations in solar insolation into long-term
and global-scale glacial–interglacial cycles are still not fully understood –
in particular, in relation to CO2 variability. Here using the Earth system model of
intermediate complexity CLIMBER-2 we performed simulations of the co-evolution of
climate, ice sheets, and carbon cycle over the last 400 000 years using the
orbital forcing as the only external forcing. The model simulates temporal
dynamics of CO2, global ice volume, and other climate system
characteristics in good agreement with paleoclimate reconstructions. These
results provide strong support for the idea that long and strongly asymmetric
glacial cycles of the late Quaternary represent a direct but strongly
nonlinear response of the Northern Hemisphere ice sheets to orbital
forcing. This response is strongly amplified and globalised by the carbon
cycle feedbacks. Using simulations performed with the model in different
configurations, we also analyse the role of individual processes and
sensitivity to the choice of model parameters. While many features of
simulated glacial cycles are rather robust, some details of CO2
evolution, especially during glacial terminations, are sensitive to the
choice of model parameters. Specifically, we found two major regimes of
CO2 changes during terminations: in the first one, when the recovery of
the Atlantic meridional overturning circulation (AMOC) occurs only at the end
of the termination, a pronounced overshoot in CO2 concentration occurs
at the beginning of the interglacial and CO2 remains almost constant
during the interglacial or even declines towards the end, resembling Eemian
CO2 dynamics. However, if the recovery of the AMOC occurs in the middle
of the glacial termination, CO2 concentration continues to rise during the
interglacial, similar to the Holocene. We also discuss the potential contribution of
the brine rejection mechanism for the CO2 and carbon isotopes in the
atmosphere and the ocean during the past glacial termination.
Introduction
Antarctic ice cores reveal that during the past 800 kyr, the atmospheric
CO2 concentration (Petit et al., 1999; Jouzel et al., 2007) varied
synchronously with the global ice volume (Waelbroeck et al., 2002;
Spratt and Lisiecki, 2016). The most straightforward
explanation for this fact is that CO2 drives glacial cycles together
with orbital variations, and the longest, 100 kyr component of the late
Quaternary glacial cycles, which is absent in the orbital forcing, is the
direct response to CO2 forcing where the 100 kyr component is the dominant
one. However, simulations with climate–ice-sheet models of different
complexity (e.g. Berger et al., 1999; Crowley and Hyde, 2008;
Ganopolski and Calov, 2011; Abe-Ouchi et al., 2013)
show that long glacial cycles (i.e. cycles with typical periodicity of ca.
100 kyr) can be simulated even with a constant CO2 concentration if the
latter is sufficiently low. Moreover, these model simulations show that not
only the dominant periodicity but also the timing of glacial cycles, can be
correctly simulated without variable CO2 forcing. This fact strongly
suggests an opposite interpretation of close correlation between global ice
volume and CO2 during Quaternary glacial cycles – namely that glacial
cycles represent a strongly nonlinear response of the Earth system to the
orbital forcing (Paillard, 1998), while variations in CO2
concentration are directly driven by ice sheet fluctuations. In turn,
CO2 variations additionally strongly amplify and globalise the direct
response of ice sheets to the orbital forcing.
In spite of the significant number of studies aimed at explaining low glacial
CO2 concentrations (e.g. Archer et al., 2000; Sigman and Boyle,
2000; Watson et al., 2000), the influence of ice sheets on the carbon cycle
remains poorly understood. It is also unclear the extent to which CO2
variations represent a direct response to ice sheet forcing and how much
is the result of additional amplification of CO2 variations through
the climate–carbon cycle feedback. Indeed, although radiative forcing of ice
sheets contributes about 50 % to glacial–interglacial variations in
global temperature (Brady et al., 2013), most of the
cooling associated with ice sheets is restricted to the area covered by ice
sheets and their close proximity. Thus, the direct contribution of ice sheets
to glacial ocean cooling is rather limited, and therefore the effect of ice
sheets on CO2 drawdown through the solubility effect can explain only a
fraction of the reduction in glacial CO2. At the same time, the direct
effect of ice sheets on atmospheric CO2 concentration through ca. 3 %
changes in the ocean volume and global salinity is rather well understood
but works in the opposite direction and leads to a glacial CO2 rise of
about 10–20 ppm (Sigman and Boyle, 2000; Brovkin et al., 2007).
Another direct effect of ice sheet growth on the carbon cycle through
reducing the area covered by forest (e.g. Prentice et al., 2011) also
operates in the opposite direction. However, several other processes could
potentially contribute to glacial CO2 drawdown through ice sheet
growth and related lowering of sea level. One such mechanism is enhanced
biological productivity in the Southern Ocean due to the iron fertilisation
effect (Martin, 1990; Watson et al., 2000). The latter is attributed to
enhanced dust deposition over the Southern Ocean seen in the paleoclimate
records (Martinez-Garcia et al., 2014; Wolff et al., 2006). At least part
of this enhanced deposition is associated with the dust mobilisation from
exposed Patagonian shelf and glaciogenic dust production related to the
Patagonian ice cap (Mahowald et al., 1999; Sugden et al., 2009). A
number of studies on the effect of iron fertilisation suggested a
contribution of 10 to 30 ppm to the glacial CO2 decrease
(e.g. Watson et al., 2000; Brovkin et al., 2007). Another
effect is related to the brine rejection mechanism, more specifically, to a
much deeper penetration of brines produced during sea ice formation in the
Southern Ocean during glacial time. The latter is explained by shallowing
and significant reduction of the Antarctic Shelf area. According to
Bouttes et al. (2010) this mechanism, in combination with enhanced
stratification of the deep ocean, can contribute up to 40 ppm to the glacial
CO2 lowering.
Apart from the mechanisms mentioned above, many other processes have been
proposed to explain low glacial CO2 concentration. Among them are
changes in the ocean circulation (Watson et al., 2015) and an increase in
Southern Ocean stratification (e.g. Kobayashi et al., 2015),
increase in sea ice area in the Southern Ocean (Stephens and
Keeling, 2000) and a shift in the westerlies (Toggweiler et al., 2006),
increase in nutrient inventory or change in the marine biota stoichiometry
(Sigman and Boyle, 2000; Wallmann et al., 2016), changes in coral reefs
accumulation and dissolution (Opdyke and Walker, 1992),
accumulation of carbon in the permafrost regions (Ciais et al., 2012;
Brovkin et al., 2016), variable volcanic outgassing (Huybers and Langmuir,
2009), and several other mechanisms. Most of these processes are not directly
related to ice sheet area or volume, and thus should be considered as
amplifiers or modifiers of the direct response of CO2 to ice sheets
operating through the climate–carbon cycle feedbacks. Although paleoclimate
records provide some useful constraints, the relative role of particular
mechanisms at different stages of glacial cycles remains poorly understood.
Most studies of glacial–interglacial CO2 variations performed to
date have been aimed at explaining the low CO2 concentration at the Last Glacial
Maximum (LGM, ca. 21 ka). In these studies, both continental ice sheets and
the radiative forcing of low glacial CO2 concentration were prescribed
from paleoclimate reconstructions. Only a few have attempted to explain CO2
dynamics during a part of (usually the glacial termination) or the entire last
glacial cycle, with models of varying complexity – from simple box-type models
(e.g. Köhler et al., 2010; Wallmann et al., 2016), to models of
intermediate complexity (Brovkin et al., 2012; Menviel et al., 2012),
or a stand-alone complex ocean carbon cycle model (Heinze et al., 2016). In
all these studies, radiative forcing of CO2 (or total GHGs) was
prescribed based on paleoclimate reconstructions. Similarly, ice sheets'
distribution and elevation were derived from paleoclimate reconstructions
or model simulations where radiative forcing of GHGs has been prescribed.
Thus, in all these studies, CO2 was treated as an external forcing
rather than an internal feedback. Here we for the first time perform
simulations of the Earth system dynamics during the past four glacial cycles
using fully interactive ice sheet and carbon cycle modelling components, and
therefore the only prescribed forcing in this experiment is the orbital forcing.
The model and experimental setupCLIMBER-2 model description
In this study we used the Earth system model of intermediate complexity
CLIMBER-2 (Petoukhov et al., 2000; Ganopolski et al., 2001). CLIMBER-2
includes a 2.5-dimensional statistical–dynamical atmosphere model, a three-basin
zonally averaged ocean model coupled to a thermodynamic sea ice model, the
three-dimensional thermomechanical ice sheet model SICOPOLIS (Greve, 1997), the dynamic model of the
terrestrial vegetation VECODE (Brovkin et al., 1997), and
the global carbon cycle model (Brovkin et al., 2002, 2007). Atmosphere and
ice sheets are coupled bidirectionally using a physically based energy
balance approach (Calov et al., 2005). The ice sheet model is only
applied to the Northern Hemisphere. The contribution of the Antarctic ice
sheet to global ice volume change is assumed to be constant during glacial
cycles and equal to 10 %. The model also includes parameterisation of the
impact of aeolian dust deposition on snow albedo (Calov et al., 2005;
Ganopolski et al., 2010). The CLIMBER-2 model in different configurations
has been used for numerous studies of past and future climates – in
particular, simulations of glacial cycles (Ganopolski et al., 2010, 2016;
Ganopolski and Calov, 2011; Willeit et al., 2015)
and carbon cycle operation during the last glacial cycle (Brovkin et al., 2012).
As has been shown by Ganopolski and Roche (2009), the temporal dynamics of
the Atlantic meridional overturning circulation (AMOC) during glacial
terminations in CLIMBER-2 are very sensitive to the magnitude of freshwater
flux to the North Atlantic. To explore different possible deglaciation
evolutions, together with the standard model version, we performed an
additional suite of simulations, in which the component of freshwater flux into
the ocean originated from melting of ice sheets was uniformly scaled up or
down by up to 10 %. This rather small change in the freshwater forcing
(typically smaller than 0.02 Sv) does not affect AMOC dynamics appreciably
most of the time but does induce a strong impact during deglaciations
(see below). Other modifications of the climate–ice-sheet component of the
model are described in the Appendix.
The ocean carbon cycle model includes modules for marine biota, oceanic
biogeochemistry, and deep ocean sediments. Biological processes in the
euphotic zone (the upper 100 m in the model) are explicitly resolved using
the model for plankton dynamics by Six and Maier-Reimer (1996). The sediment
diagenesis model (Archer, 1996; Brovkin et al., 2007) calculates burial of
CaCO3 in the deep sea, while shallow-water CaCO3 sedimentation is
simulated based on the coral reef model (Kleypas, 1997) driven by sea level
change. Silicate and carbonate weathering rates are scaled to the runoff
from the land surface; they are also affected by sea level change (Munhoven,
2002). Compared to Brovkin et al. (2012), the carbon cycle model has been
modified in several aspects. Similar to Brovkin et al. (2012), the
efficiency of nutrient utilisation in the Southern Ocean is set to be
proportional to the dust deposition rate (see Appendix), which in the case
of one-way coupling is prescribed to be proportional to the dust deposition
in the EPICA ice core. However, in the fully interactive experiment, the
dust deposition rate over the Southern Ocean has been computed from
simulated sea level (see Appendix). This means that in the fully interactive
experiments (see below) we did not explicitly use any paleoclimate data to
drive the model, and the orbital forcing was the only prescribed forcing. In
the marine carbon cycle component, we also account for a dependence of the
remineralisation depth on ocean temperature following Segschneider and
Bendtsen (2013) (see Appendix). In our previous studies the remineralisation
depth was kept constant.
The CLIMBER-2 model used in earlier studies of glacial carbon cycle did not
include long-term terrestrial carbon pools such as permafrost carbon, peat,
and carbon buried beneath the ice sheets. In the present version of the
model, these pools are included. The model also accounts for peat
accumulation. Modification of the terrestrial carbon cycle components is
described in detail in the Appendix. For simulation of atmospheric
radiocarbon during the last glacial termination we used the rate of
14C production following the scenario of Hain et al. (2014), which is based on the
production model of Kovaltsov et al. (2012).
In our previous experiments performed with the CLIMBER-2 model (Brovkin et
al., 2012; Ganopolski and Brovkin, 2015) we prescribed not only variations
in the Earth's orbital parameters (eccentricity, precession, and obliquity)
but also the radiative effect of GHGs (CO2, CH4, and N2O)
computed using their concentrations from the ice cores records (Lüthi et
al., 2008; Petit et al., 1999). In these experiments, which we denote
hereafter as “one-way coupled” (Fig. 1a, Table 1), atmospheric CO2
concentration was computed by the carbon cycle model but not used as the
radiative forcing for the climate component. Similarly, in these experiments, the
CO2 fertilisation effect on vegetation was computed using reconstructed
CO2 concentration. Therefore, in one-way coupled experiments there were
no feedbacks of the simulated atmospheric CO2 concentration on climate.
In the present study, we performed a suite of one-way coupled experiments for
the last four glacial cycles but we also performed simulations in which the
orbital forcing was the only prescribed external forcing. Since CLIMBER-2
does not include methane and N2O cycles and does not account for these
GHGs in its radiative scheme, we made use of the fact that CO2 is the
dominant GHG and that temporal variations of the other two closely follow
CO2. To account for the effect of methane and N2O forcings, we
computed the effective CO2 concentration used in the radiative scheme
of the model in such a way that radiative forcing of effective CO2
exceeds radiative forcing of simulated CO2 by 30 % at any time. This
type of experiment we refer to as “fully interactive” (Fig. 1b). In
the fully interactive experiment we also use computed CO2 concentration
in the terrestrial component to account for the CO2 fertilisation effect. As
stated above, the dust deposition rate over the Southern Ocean which is used
in the parameterisation of the iron fertilisation effect was computed from the
global sea level. The radiative forcing of aeolian dust and dust deposition
on ice sheets (apart from the glaciogenic dust sources) in both types of
experiments were obtained identically to Calov et al. (2005) and
Ganopolski and Calov (2011) by scaling the fields computed with global climate models, where the
scaling parameter was proportional to the global ice volume.
Model spin-up
The model spin-up and the proper choice of model parameters for simulation
of multiple glacial cycles represent a challenge when using the model with
long-term components of the carbon cycle because inconsistent initial
conditions or even a small imbalance in carbon fluxes could lead to a large
drift in simulated atmospheric CO2 concentration (in the case of
one-way coupling) or the state of the entire Earth system (climate, ice
sheets, CO2) in the case of the fully interactive experiments. Note
that, in the latter case, the negative climate–weathering feedback will
eventually stabilise the system but this occurs at a timescale of several
glacial cycles and over this time climate could drift far away from its
realistic state. To avoid such drift, volcanic outgassing should be
carefully calibrated. Based on a set of sensitivity experiments, we found
that the value of 5.3 Tmol C yr-1 allows us to simulate quasiperiodic cycles
without a long-term trend in atmospheric CO2. Note that even a
±10 % change in volcanic outgassing leads to a significant (order of 100 ppm)
drift in CO2 concentration simulated over the last four glacial cycles.
When the carbon cycle model incorporates such long-term processes as
terrestrial weathering, marine sediment accumulation, and permafrost carbon
burial, the assumption that the system is close to equilibrium at the
pre-industrial period or at any other moment of time is not valid even if the
CO2 concentration was relatively stable during a certain time interval.
To produce proper initial conditions at 410 ka we performed a sequence of
410 kyr long one-way coupled runs with the identical forcings. In the first
run we used as the initial conditions the final state obtained in simulation
of the last glacial cycles (Brovkin et al., 2012). Then we launched each
410 kyr experiment from the final state obtained in the previous model run. The
results of such sequence of experiments reveal a clear tendency to converge
to the solution with similar initial and final states of the Earth system.
We then used the state of climate and carbon cycle obtained at the end of
the last run as the initial conditions for all experiments presented in this
paper. In the analysis of all experiments described below, we exclude the
first 10 000 years.
Simulations of the last four glacial cycles
Realistic simulation of climate and carbon cycle evolution during the last
four glacial cycles is more challenging in the case of the fully interactive
configuration, because in this case a number of additional positive
feedbacks tend to amplify initial model biases. Therefore we begin our
analysis with the one-way coupled simulations similar to that performed in
Brovkin et al. (2012). This configuration was also used for the
calibration of new parameterisations (see Sect. 4) and a set of
sensitivity experiments for the last glacial termination (Sect. 5).
Experiments with one-way coupled climate–carbon cycle model
Simulated climate and ice sheet variations in the one-way coupled
experiments are similar to the ones presented in
Ganopolski and Calov (2011), which is not surprising
since the only difference between model versions used in these studies is
related to the coupling between ice sheet and climate components (see
Appendix). Simulated glacial cycles are characterised by global surface air
temperature variations of about 5 ∘C (not shown) and maximum sea level
drops by more than 100 m during several glacial maxima. Simulated
global ice sheet volume most of the time is close to the reconstructed
one (Spratt and Lisiecki, 2016) (Fig. 2d). In general,
differences between simulated and reconstructed global sea level are
comparable to the uncertainties in sea level reconstructions obtained using
different methods.
Transient simulations of the last four glacial cycles forced by orbital
variations, observed concentration of well-mixed GHGs, and dust deposition rate
(one-way coupled experiments). (a) Maximum summer insolation at
65∘ N, W m-2; (b) radiative forcing (relative to
pre-industrial) of well-mixed GHGs, W m-2; (c) Antarctic dust
deposition rate in relative units; (d) global ice volume expressed in
sea level equivalent (m); (e) atmospheric CO2 concentration (ppm).
Dark red colour in (a)–(c) represents prescribed forcings.
Black dashed line in (d) is the sea level stack from Spratt and
Lisiecki (2016), and in (e) is the compiled Antarctic CO2 record from
Lüthi et al. (2008). Radiative forcing of GHGs in (b) is from
Ganopolski and Calov (2011). Antarctic dust is from Augustin et al. (2004).
Blue lines in (d, e) correspond to the baseline experiment ONE_1.0
and purple lines to experiment ONE_1.1, where meltwater flux into the Atlantic
was scaled up by a factor of 1.1.
Simulated CO2 concentration (Fig. 2e) is also in a good agreement with
reconstructions based on several Antarctica ice cores (Barnola et al.,
1987; Monnin et al., 2004; Petit et al., 1999; Lüthi et al., 2008). The
model correctly reproduces the magnitude of glacial–interglacial CO2
variability of about 80 ppm. Results of simulations with the standard model
version (ONE_1.0) and model with 10 % enhanced meltwater
flux (ONE_1.1) are essentially identical most of the time,
except for glacial terminations. During glacial terminations even small
differences in the freshwater forcing cause pronounced differences in the
temporal evolution of the AMOC, and as a result, of CO2 concentration.
As seen in Fig. 2d, in the experiment ONE_1.0, CO2
concentration grows monotonically during the last glacial termination (TI,
midpoint at ca. 15 ka) and Termination IV (ca. 330 ka) while it rises faster
and overshoots the interglacial level during Terminations II (ca. 135 ka)
and III (ca. 240 ka). In contrast, in experiment ONE_1.1,
similar overshoots occur during Terminations I and III but not
Termination IV. In all cases, simulated CO2 lags behind the
reconstructed one but this lag is smaller in the cases when overshoot is
simulated. Experiments with CO2 overshoots are clearly in better
agreement with empirical data for MIS7 and MIS9. Analysis of model results
shows that pronounced CO2 overshoot occurs in the case when the AMOC is
suppressed during the entire glacial termination and recovers only after the
cessation of meltwater flux (Fig. 3). In contrast, if the AMOC recovers
well before the end of deglaciation, simulated CO2 experiences only
local overshoot and continues to rise during most of the interglacial. The
latter behaviour is similar to that was observed in reality during MIS11
and the Holocene, while the former is typical for MIS5, 7, and 9. Thus our
model is able to reproduce both types of CO2 dynamics during recent interglacials.
Temporal evolution of the AMOC, Sv (a), and atmospheric CO2
concentration, ppm (b) during the last four glacial terminations. Blue
lines correspond to experiment ONE_1.0 and purple lines to experiment
ONE_1.1, where meltwater flux into the Atlantic was scaled up by a factor of 1.1.
Simulated CO2 and δ13C with the one-way coupled model
(ONE_1.0). (a) CO2 concentration (ppm) (Lüthi et al, 2008) (b) atmospheric
δ13CO2 (‰), (c) deep South Atlantic δ13C
(‰); (d) deep North Pacific δ13C (‰). Colour
lines – model results. Empirical data (black dashed lines): (a) Lüthi
et al. (2008); (b) Eggleston et al. (2016); (c, d) Lisiecki et al. (2008).
The rise of CO2 by 10–20 ppm on the millennial timescale during AMOC
shutdowns is a persistent feature of CLIMBER-2 and the mechanism of this
rise has been explained in Brovkin et al. (2012) by a weakening of the reverse
cell of the Indo-Pacific overturning circulation during periods of reduced
AMOC. A similar rise in atmospheric CO2 concentration during periods of
AMOC shutdown has been simulated in some other (but not all) similar
modelling experiments. Incorporation of the temperature-dependent
remineralisation depth additionally contributes to the CO2 overshoots
at the beginning of several interglacials (see below) but the mechanism
described in Brovkin et al. (2012) remains the dominant one.
Comparison of simulated deep ocean δ13C with paleoclimate
reconstructions (Fig. 4) show that the model correctly simulates larger
δ13C variability in the deep Atlantic in comparison to the deep
Pacific but underestimates the amplitude of glacial–interglacial
δ13C variability. Simulated atmospheric δ13CO2 shows
a rather complex behaviour with an amplitude of variability of
0.6 ‰. The agreement between simulated and
reconstructed (Eggleston et al., 2016) atmospheric δ13CO2
is rather poor. Both model and data show a drop in atmospheric δ13CO2
during the last and penultimate deglaciations but the data
also suggest the strong drop at the end of Eemian interglacial while the
model simulated a continuous rise of δ13CO2 during this
interval. In addition, temporal variability of the reconstructed
δ13CO2 is significantly larger than the simulated one. A more
detailed comparison with empirical data during the last deglaciation is
presented in Sect. 5.
(a) Simulated CO2 concentration (ppm); (b) prescribed
Antarctic dust deposition rate in relative units; (c) simulated annual
mean sea ice area in the Southern Hemisphere (106 km2);
(d) simulated oxygen concentration in the deep Southern Ocean in
(µmol kg-1) in the ONE_1.1 experiment (solid line) and the
identical experiment but without iron fertilisation effect (dashed line).
Transient simulations of the last four glacial cycles forced by orbital
variations only (fully interactive experiments). (a) Maximum summer
insolation at 65∘ N (W m-2); (b) radiative forcing
(relative to pre-industrial) of well-mixed GHGs (W m-2); (c) Antarctic
dust deposition rate in relative units; (d) global ice volume expressed
in sea level equivalent (m); (e) atmospheric CO2 concentration,
ppm. Black line in (b) is radiative forcing of GHGs from Ganopolski
and Calov (2011). Black dashed line in (c) is Antarctic dust from
Augustin et al. (2004), in (d) is sea level stack from Spratt and
Lisiecki (2016), and in (e) is compiled Antarctic CO2 record from
Lüthi et al. (2008). Blue lines in (d, e) correspond to the fully
interactive experiment INTER_1.0 and purple lines to the experiment INTER_1.1,
where meltwater flux into the Atlantic was scaled up by a factor of 1.1.
Changes in the ocean oxygenation is considered to be an important indicator
of respired carbon storage in the deep ocean, and therefore the proxy for
the strength of ocean biological pump. Jaccard et al. (2016) inferred a
significant decline in the deep Southern Ocean oxygenation and interpreted it
as the result of a combined effect of iron fertilisation by dust and decreased
deep ocean ventilation. Our results (Fig. 5) are fully consistent with such an
interpretation. The model simulates significant reduction in the dissolved
oxygen in the deep Southern Ocean during glacial period. Roughly two-thirds of this
reduction is already simulated in the experiment without iron fertilisation
and can be solely attributed to reduced deep ocean ventilation. It is
noteworthy that changes in the oxygen concentration in this experiment are
strongly anticorrelated with the area of sea ice in the Southern Hemisphere
(Fig. 5c). This explained by the fact that sea ice directly and
indirectly (through stratification of the upper ocean layer) affects gas
exchanges between the ocean and the atmosphere. Oxygen concentration is
additionally reduced during periods with high dust deposition rate in the
experiment, which accounts for the iron fertilisation effect (Fig. 5d).
Experiments with the fully interactive model
In the fully interactive experiments, orbital forcing is the only prescribed
forcing and the model does not use any time-dependent paleoclimatological
information (such as the Antarctic dust deposition rate used in the one-way
coupled experiment). Results of the fully interactive experiment
INTER_1.0 are shown in Fig. 6. For the first experiment of
such type ever, the agreement between model simulations and empirical
reconstructions is reasonably good. The model simulates correct magnitude
and timing of the last four glacial cycles in respect to both sea level and
CO2 concentration. It also reproduces the strong asymmetry of glacial
cycles. Naturally, the mismatch between the simulated and reconstructed
characteristic in the fully interactive experiments is larger than in the
one-way coupled experiments. In particular, in the fully interactive
experiment, simulated ice volume is underestimated by 10–20 m in sea
level equivalent compared to the reconstructed one. Although the magnitude
of glacial–interglacial CO2 variability in the fully interactive
experiment INTER_1.0 is similar to that in the one-way
coupled experiment ONE_1.0 and in the reconstructions, the
lag between simulated and reconstructed CO2 during glacial terminations
increases additionally in comparison to the one-way coupled experiments.
Interestingly, the last glacial cycle and the first 150 kyr of the
INTER_1.0 and ONE_1.0 experiments are in very
good agreement, while during the time interval between 300 and 150 ka
discrepancies are larger. This period corresponds to higher eccentricity and
therefore to the larger magnitude of orbital forcing. Similarly to the
results of one-way coupled experiments, the fully interactive experiments
also show strong sensitivity to magnitude of freshwater flux during glacial terminations.
Comparison of simulated ice sheet spatial distribution and elevation (Fig. 7)
shows that the results of one-way coupled (ONE_1.0, Fig. 6a) and the fully
interactive experiments (INTER_1.0, Fig. 7b) are almost identical during the
LGM (the same is true for the previous
glacial maxima, not shown) and in reasonable agreement with the
reconstructions. During glacial terminations, the difference between two
experiments increases since in the fully interactive experiment the
radiative forcing of GHGs lags considerably behind the reconstructed one
used in the one-way coupled experiment. As a result, at 7 ka continental ice
sheets melted completely in the one-way coupled experiment (Fig. 7c) while
in the fully interactive experiment, a relatively large ice sheet is still
present in northeastern Canada (Fig. 7d).
It is instructive to compare frequency spectra of simulated and
reconstructed global ice volume in the one-way and fully interactive
experiments (Fig. 8). In addition, Fig. 8 shows results from the experiment
ONE_240 performed with constant radiative forcing of GHGs
corresponding to equivalent CO2 concentration of 240 ppm. As already
shown by Ganopolski and Calov (2011), even with constant CO2, the model
computes pronounced glacial cycle with 100 kyr periodicity, although it has
much weaker amplitude than the reconstructed ice volume. Both model
experiments with varying CO2 radiative forcing (ONE_1.0
and INTER_1.0) reveal much stronger 100 kyr periodicity,
which has only a slightly weaker amplitude than the spectrum of the
reconstructed ice volume. Interestingly, frequency spectra of ice volume
simulated in the one-way and fully interactive experiments have similar
powers in 100 kyr and obliquity (40 kyr) bands, but in the precessional band
(ca. 20 kyr) the one-way coupled experiment reveals a much higher spectral
power. This cannot be explained by the prescribed radiative forcing of GHGs
because the latter contain very little precessional variability. The
stronger precessional component in the ONE_1.0 experiment is
explained by the fact that in the one-way coupled experiment, the model
simulates faster ice sheet growth during the initial part of each glacial
cycle and the modelled ice volume variability at the precessional frequency
is very sensitive to the global ice volume.
Simulated ice sheet elevation (m) at 21 ka (a, b) and
7 ka (c, d) in the one-way coupled experiment ONE_1.0 (a, c)
and fully interactive experiment INTER_1.0 (b, d). Blue lines represent
Ice-5g reconstruction at the LGM (Peltier, 2004).
The composition of “the carbon stew” and factor analysis
In this section, we discuss the contribution of different factors to
simulated variations in CO2 concentration. Because none of the
mechanisms could explain the CO2 dynamics in isolation from the other
factors (e.g. Sigman and Boyle, 2000; Archer et al., 2000), we call the
composition and timing of the mechanisms leading to the glacial CO2
cycle the “carbon stew”. As has been shown in Brovkin et al. (2012), the role of
different mechanisms controlling CO2 concentration at different phases
of glacial cycles is different. However, even if we consider only the LGM
(as most previous works did), the composition of the “carbon stew”
remains highly uncertain, even though there is a growing awareness that
both physical and biological processes must have played a comparably
important role in glacial CO2 drawdown (e.g. Schmittner and Somes, 2016;
Galbraith and Jaccard, 2015). Obviously, the choice of the “carbon stew”
is crucially important for successful simulations of glacial cycles. The aim
of our paper is not to present the ultimate solution for the “carbon stew”
problem since at present this is impossible. Rather we want to demonstrate
that with a reasonable representation of physical, geochemical, and
biological processes in the model, it is possible to reproduce the main
features of Earth system dynamics over the past 400 kyr, including the
magnitude and timing of climate, ice volume, and CO2 variations.
Similar to the study by Brovkin et al. (2012), we performed a set of
experiments using the one-way coupling technique (see Table 1 for details).
We use this approach instead of fully interactive coupling to exclude
complex and strongly nonlinear interactions associated with the ice sheet
dynamics, which significantly complicates the factor analysis. In the case of
the one-way coupled experiments, climate, ice sheets, and other external
factors are identical, and these experiments only differ from each other by the
parameters of the carbon cycle model. Since CO2 simulated in the
one-way coupled experiment with 10 % enhanced meltwater flux
(ONE_1.1) is in a slightly better agreement with
observational data than the standard one (ONE_1.0), for the
factor analysis we used the experiment ONE_1.1 as the reference one and
performed all sensitivity experiments with 10 % enhanced meltwater flux.
The standard carbon cycle model setup
We begin our analysis with the experiment that incorporates only the
standard ocean biogeochemistry as described in Brovkin et al. (2007)
(Fig. 9). This experiment does not include the effect of the terrestrial carbon cycle. In
this configuration, the model is able to explain only about 45 ppm of
CO2 reduction during glacial cycles. Note that this experiment accounts
for glacial–interglacial changes in ocean volume of ca. 3 % and
corresponding changes in the total biogeochemical inventories including
salinity. These volume changes are often neglected in simulations with
3-dimensional ocean models (e.g. Heinze et al., 2016), although these changes
have substantial impact on the global carbon cycle, and in our simulations they
counteract glacial CO2 drawdown by ca. 12 ppm. Without the effect of
ocean volume reduction, the combination of physical processes and carbonate
chemistry can explain up to 57 ppm at the LGM and on average 38 ppm
during the entire 400 kyr time interval (see Table 2). This is consistent
with the recent results by Buchanan et al. (2016) and Kobayashi et al. (2015).
Note that simulated changes in silicate weathering and its impact on
atmospheric CO2 are small, as has already been shown in Brovkin et al. (2012).
Accounting for the land carbon changes does not help to explain the CO2
concentration changes, since terrestrial carbon contains ca. 350 Gt less
carbon at the LGM compared to the pre-industrial state. This reduces the
glacial–interglacial CO2 difference by 10–15 ppm compared to the
ocean-only experiment (Fig. 9b). Enabling the parameterisation for the
iron fertilisation effect in the Southern Ocean results in additional
glacial CO2 drawdown of up to 30 ppm (22 ppm at the LGM), mostly
towards the end of each glacial cycle (Fig. 9c). This value is close
to that reported by Lambert et al. (2015). With all these processes
considered in our previous study (Brovkin et al., 2012), we are still
short of ca. 25 ppm to explain the full magnitude of glacial–interglacial variability.
Transient simulations of the last four glacial cycles forced by orbital
variations, with prescribed, interactive, and fixed concentrations of well-mixed
GHGs. (a) Maximum summer insolation at 65∘ N (W m-2);
(b) temporal evolution of reconstructed and simulated sea level (m);
(c) frequency spectra of the global ice volume; (d) frequency
spectra of boreal summer insolation. Black line is for the data (Spratt and
Lisiecki, 2016), blue line corresponds to the one-way coupled experiment
ONE_1.0, red line to the fully interactive experiment INTER_1.0, and green
line to the ONE_240 experiment with constant (240 ppm) CO2 concentration.
Results of factor separation analysis. (a) Simulated CO2
(ppm) in one-way coupled ONE_1.1 experiment (purple line) and reconstructed
CO2 concentrations (black dashed line, Lüthi et al., 2008).
(b)–(d) Contributions to simulated atmospheric CO2
(ppm) of terrestrial carbon cycle (b), ONE_S4–ONE_S3; iron
fertilisation (c), ONE_S3–ONE_S2; variable volcanic outgassing (d),
ONE_S2–ONE_S1; temperature-dependent remineralisation depth (e),
ONE_S1–ONE_1.1.
Additional processes included in the carbon cycle model
There are a number of other proposed mechanisms which potentially can explain
several tens of ppm of glacial CO2 decline. Our choice of two processes to
obtain the observed magnitude of glacial–interglacial CO2 variations is
somewhat subjective. The chosen mechanisms are explained below, while an
alternative one (brine rejection) is discussed in Sect. 4.3.
The first additional mechanism added to the mechanisms already described in Brovkin et al. (2012) is
temperature-dependent remineralisation depth. In the standard CLIMBER-2
version, remineralisation depth is spatially and temporally constant. Since
in the colder ocean remineralisation depth increases, this enhances the
efficiency of the carbon pump and contributes to a decrease in atmospheric
CO2 concentration (e.g. Heinze et al., 2016; Menviel et al., 2012;
Matsumoto, 2007). Details of the mechanism implementation are described in
the Appendix. As seen in Fig. 9e, making remineralisation depth
temperature dependent introduces additional glacial–interglacial variability
of CO2 with a magnitude of about 20 ppm. Roughly half of this value
is clearly attributed to the CO2 overshoots seen at the
beginning of some interglacials. The reason is that the AMOC shutdowns due
to meltwater flux that happened during glacial terminations lead not only
to surface cooling in the North Atlantic but also to significant
thermocline warming that occurs over the entire Atlantic Ocean (e.g. Mignot
et al., 2007). This subsurface warming causes
pronounced shoaling of the remineralisation depth and the release of carbon
from the ocean into the atmosphere. This process reverses after recovery of
the AMOC at the beginning of interglacials.
Model experiments performed in this study. P denotes prescribed
characteristic; I – interactive; STD – standard model configuration; RD – variable
remineralisation depth; VO – variable volcanic outgassing; IF – iron
fertilisation in the Southern Ocean; TC – terrestrial carbon cycle; BR – brine
rejection mechanism. A minus sign means that the process is excluded and a plus
sign means that process is included. Ice sheets are interactive in all simulations.
“The carbon stew” at the LGM and the entire 400 kyr period in the
ONE_1.1 experiment. Positive (negative) sign indicates that the process
contributes to glacial CO2 reduction (increase) as compared to the
pre-industrial concentration simulated at the end of the experiment.
* Deep ocean and shallow water carbonate sediments, carbonate
and silicate weathering.
Burley and Katz (2015) and Huybers and Langmuir (2009) proposed that the
rate of volcanic outgassing varies during glacial cycle due to variable load
of the ice sheet and ocean on the Earth crust. Therefore we assume that
volcanic outgassing has a variable component (about 30 % of its averaged
value of 5.3 Tmol yr-1), which represents the delayed response to the change in
global ice volume. This simple parameterisation explained in the Appendix does
not affect cumulative volcanic outgassing over the glacial cycle, but
contributes to glacial–interglacial CO2 variability by an additional 10 ppm
(Fig. 9d). With varying volcanic outgassing and temperature-dependent
remineralisation depth, CLIMBER-2 reproduces glacial–interglacial CO2
cycles in a good agreement with paleoclimate records (Fig. 9a).
Brine rejection mechanism
Using a different version of CLIMBER-2, Bouttes et al. (2010) proposed that
a significant fraction of glacial–interglacial CO2 variations can be
explained by the mechanism of brine rejections – more specifically, by a
large increase in the depth to which brines can penetrate under glacial
conditions without significant mixing with ambient water masses. Such
increase in brine efficiency under glacial conditions would result in a
large transport of salinity, carbon, and other tracers from the upper ocean
layer into the deep ocean. By choosing the efficiency coefficient close to 1, Bouttes et al. (2010) demonstrated that brines are able to explain up
to 40 ppm CO2 decrease. We have implemented this mechanism in
combination with stratification-dependent vertical diffusivity in our
version of the CLIMBER-2 model and obtained results qualitatively similar to
Bouttes et al. (2010). While we believe that the brine rejection mechanism
belongs to a class of plausible mechanisms contributing to glacial CO2
drawdown, we did not use brine parameterisation in our simulations for
several reasons. Firstly, the parameterisation for brine rejection cannot be
tested against observational data. For present-day climate conditions, brine
rejection efficiency should be below 0.1, otherwise modern Antarctic bottom
water becomes saltier than the North Atlantic deep water, which is at odds
with reality. This means that to be an efficient mechanism for glacial
CO2 drawdown, the brine efficiency should increase under glacial
conditions, at least by an order of magnitude. Whether this is physically
plausible is not clear. The only paleoclimate constraint on the brine
efficiency is the reconstruction of paleosalinity based on the pore water
(Adkins et al., 2002), which suggests an increase in the deep water salinity in
the Southern Ocean by more than 2 psu during the LGM. Such an increase in
salinity is indeed difficult to reproduce without contribution of brines.
However, the accuracy of salinity reconstructions based on such a method remains
uncertain (Wunsch, 2016). Second, there is a problem with
the temporal dynamics of brine rejection efficiency. Mariotti et al. (2016)
assumed an abrupt decrease in brine rejection efficiency from 0.7 to 0 in a
very short interval between 18 and 16 ka. However, both sea level and the
size of the Antarctic ice sheets remained essentially constant during this
period and therefore there is no obvious reason for such large variations in
the brine rejection efficiency. According to the interpretation of
Roberts et al. (2016), brine rejection remained efficient during most of the
glacial termination and ceased only after 11 ka, when most of the
glacial–interglacial CO2 rise had already been accomplished. In the
view of these uncertainties, we decided not to include parameterisations of the
brine rejection mechanism in simulations of glacial cycles. However, for
simulations of the last glacial termination discussed below, we analysed the
potential effect of brine rejection on radiocarbon and other paleoclimate proxies.
Simulations of Termination ISimulation of climate, CO2, and carbon isotopes during the last termination
The last glacial termination provides a wealth of paleoclimate records with
a potential to better constrain the mechanisms of glacial CO2
variability. In this section, we discuss the last glacial termination in
more detail. Similarly to the previous section, to exclude nonlinear
interaction with ice sheets, we discuss here only one-way coupled
experiments. To reduce computational time, we performed experiments only for
the last 130 000 years starting from the Eemian interglacial and using the
same initial conditions as in the experiments discussed above.
In the standard ONE_1.0_130K experiment, the
model simulates climate variability across Termination I rather
realistically. In particular, it reproduces the temporal resumption of the AMOC
in the middle of the termination resembling the Bølling–Allerød warm event
(Fig. 10a). However, the timing of this event in our model is shifted by
ca. 1000 years compared to the paleoclimate records. Results of our experiments
reveal high sensitivity of the timing of the AMOC resumption to the
magnitude of freshwater flux. A change of the flux by 10 % in the
ONE_1.1_130K experiment significantly alters
millennial-scale variability during the last glacial termination (Fig. 10).
This result suggests that simulated millennial-scale variability during the
Termination I is not robust – i.e. it is unlikely that a single model run
through the glacial termination would reproduce the right timing or even the
right sequence of millennial-scale events.
Although simulated CO2 concentration at the LGM and pre-industrial
state are close to observations, simulated CO2 appreciably lags behind
reconstructed CO2 during the termination (Fig. 10b). This is primarily
related to the fact that simulated CO2 does not start to grow at
ca. 18 ka BP as reconstructed, but only after the end of the simulated analogue of the
Bølling–Allerød event. At the same time, in agreement with
paleoclimate reconstructions, CO2 concentration reaches a local maximum
at the end of the North Atlantic cold event, which resembles the Younger
Dryas event. Simulated CO2 concentration also reveals a continuous
CO2 rise during the Holocene towards its pre-industrial value of 280
ppm. This result confirms that such CO2 dynamics could be explained by
natural mechanisms alone and do not require early anthropogenic
CO2 emissions until ca. 2 ka (Kleinen et al.,
2016). This result also demonstrates that the temporal dynamics of CO2
during interglacials critically depend on the timing of final AMOC
recovery. Late recovery during glacial termination causes a strong overshoot
of CO2 at the beginning of the interglacial followed by some decrease or a
stable CO2 concentration. However, if the complete AMOC recovery occurs
well before the end of termination, only temporal CO2 overshoot occurs
and CO2 continues to rise during the entire interglacial.
Simulation of Termination I with the set of one-way coupled models,
which differ only in the scaling of freshwater flux. Blue line corresponds to the
ONE_1.1_130K experiment with scaling factor 1.1, red line to the ONE_1.0_130K
experiment with scaling factor 1.0. (a) AMOC strength (Sv);
(b) atmospheric CO2 (ppm); (c) atmospheric δ13CO2 (‰); (d) atmospheric Δ14CO2 (‰). Dashed lines:
(b, c) ice core data (Lüthi et al., 2008; Schmitt et al., 2012);
(d) IntCal13 radiocarbon calibration curve (Reimer et al., 2013).
δ13C and radiocarbon ventilation age distribution in the
Atlantic Ocean in the ONE_1.0_130K simulation of Termination I.
(a, b)δ13C (‰), (c, d) radiocarbon ventilation
age in yr 14C. (a, c) Differences between LGM (21 ka) and
pre-industrial in the Atlantic Ocean. (b, d) Temporal evolution of
anomalies during the past 20 kyr at 20∘ N in the Atlantic.
It is generally believed that atmospheric δ13C provides useful
constraint on the mechanisms of deglacial CO2 rise (Schmitt et al.,
2012; Joos et al., 2004; Fischer et al., 2010). Simulated atmospheric
δ13C drops from the LGM level of about
-6.4 ‰ to the minimum value of -6.7 ‰ between
16 and 14 ka (Fig. 10c). This is primarily related to the reduction
of marine biological productivity which, in turn, is explained by the
decrease in iron fertilisation effect over the Southern Ocean during the
first part of Termination I. The magnitude of the δ13C drop is
in a good agreement with empirical data (Fig. 10c). The model is also able
to simulate W-shaped δ13C evolution associated with
reorganisation of the AMOC. However, this W shape is shifted in time
compared to the reconstructed one by ca. 1000 years because the model analogue
of the Bølling–Allerød event occurs earlier than the real one by the same
amount of time. Note that this local maximum in δ13C is
completely absent in the experiment ONE_1.1_130K, where temporal resumption of the AMOC during glacial termination does
not occur. δ13C rise after 12 ka is primary attributed to the
accumulation of carbon in terrestrial carbon pools (forest regrowth and peat
accumulation). At the same time, simulated present-day atmospheric δ13C
is underestimated compared to ice core data by ca. 0.15 ‰.
Simulation of Termination I in the standard ONE_1.0_130K experiment
(solid blue) and the ONE_BRINE_130K (dashed blue) experiment, which includes
brine parameterisation and stratification-dependent vertical mixing.
(a) Atmospheric Δ14C (in % ∘). (b) Deep
tropical Atlantic radiocarbon ventilation age (in yr 14C). (c) Deep
tropical Atlantic δ13C (% ∘). (d) Deep Southern
Ocean salinity (psu); (c, d) for the depth 4 km. Black dashed line
is IntCal13 radiocarbon calibration curve (Reimer et al., 2013).
The model simulates an almost monotonic decrease in atmospheric Δ14C
from the LGM to present. Most of this decrease (ca. 200 ‰)
is caused by a prescribed production rate which was
about 20 % higher during LGM. Only about 80 ‰ of
Δ14C is attributed to a difference in climate state between LGM
and the present, primarily due to a less ventilated deep ocean. As shown in Fig. 10d,
simulated atmospheric Δ14C is significantly underestimated
before 12 ka compared to the reconstruction by Reimer et al. (2013) and at
the LGM this difference reaches more than 100 ‰. It is
possible but unlikely that such big differences can be attributed to
uncertainties in reconstructed production rate. An alternative hypothesis
for explaining this mismatch is discussed below.
Figure 11 shows LGM time slice anomalies and temporal evolution of δ13C
and radiocarbon ventilation age during Termination I in the
Atlantic Ocean simulated in experiment ONE_1.0_130K. The spatial distribution of glacial anomalies and
temporal dynamics of δ13C and radiocarbon ventilation age
during termination are qualitatively very similar. Glacial δ13C
in the deep Atlantic at the LGM is 0.6–1 ‰ lower than
at present; this is primarily related to a shoaling of the AMOC and reduced
ventilation in the Southern Ocean. The vertical distribution of δ13C
anomalies at the LGM is consistent with the paleoclimate
reconstructions (e.g. Hesse et al., 2011).
Simulated ventilation age at the LGM can be directly compared with Skinner
et al. (2017) (their Fig. 4a and c). Both models and data show significant
increase in radiocarbon ventilation age in the deep Atlantic. However, the
spatial patterns of ventilation age changes are rather different. In the
model, the largest increase in the ventilation age occurs in the deep
North Atlantic, which is explained by shoaling of the AMOC cell and
increased presence of the poorly ventilated Southern Ocean water masses. At
the same time, the reconstructed radiocarbon ages in the deep North Atlantic
are characterised by very large scattering (from 1000 to 3000 14C years) and it is unclear whether their average values can be directly
compared to the results of the zonally averaged ocean model.
During glacial termination, both δ13C and radiocarbon
ventilation age show pronounced response at all depths to the millennial-scale reorganisations of the AMOC (Fig. 12b and d). The ventilation age in the
deep Atlantic, which is about 2000 years prior to the model analogue of the
warm Bølling–Allerød event, rapidly reaches nearly the modern level after
the AMOC resumption and drops again to the glacial level during the model
analogue of the cold Younger Dryas event. Such evolution of ventilation age
in the North Atlantic is in agreement with paleoclimate reconstructions
(Robinson et al., 2005; Skinner et al., 2014).
Vertical profile of ventilation age in 14C years for Atlantic (a),
Pacific (b), and Southern Ocean (c). Red line represents modern
conditions; solid blue – LGM in ONE_1.0_130K experiment using the standard
version of the model; dashed blue – LGM in ONE_BRINE_130K experiment with
the model version, which includes brine parameterisation and stratification-dependent
vertical mixing. Red (blue) squares represent basin-averaged radiocarbon age
for the modern (LGM) state based on the data of Skinner et al. (2017).
Dynamics of terrestrial carbon pools (Gt C) in the one-way coupled
ONE_1.0 simulation. Left panels: the whole 400 kyr period; right panels: the
Termination I period. (a) Black line – total carbon storage; magenta
line – conventional carbon pools (biomass and mineral soils).
(b)–(d) Peat, permafrost, and buried carbon storages, respectively.
Brine rejection mechanisms and radiocarbon in the ocean and atmosphere
As discussed above, our version of the CLIMBER-2 model is not able to
accurately reproduce atmospheric 14C decline during the first part of
the last glacial termination. At the same time, Mariotti et al. (2016)
demonstrated that their version of CLIMBER-2, which incorporates the
mechanism of brine rejection, is able to simulate larger atmospheric 14C
decrease from the LGM to the present, which is in better agreement
with the observational data (Reimer et al., 2013). By introducing a similar
parameterisation for brine rejection and stratification-dependent vertical
diffusivity in our model, we are able to reproduce the results of Mariotti et
al. (2016) (Fig. 12). It is noteworthy that we use different temporal
dynamics of the efficiency of brine rejections during termination. Instead
of abrupt and non-monotonic changes in the brine efficiency prescribed in
Mariotti et al. (2016), in the ONE_BRINE_130K
experiment we assume that this efficiency is 0.75 at the LGM, 0 at present,
and in between it follows the global ice volume. We do not claim that this
scenario is more realistic, but at least it is more consistent with the
findings of Roberts et al. (2016). Figure 11 shows that the model with the
brine rejection parameterisation and stratification-dependent vertical
diffusivity simulates atmospheric Δ14C in better agreement with
empirical data then the standard version. This is explained by the fact that
the brine rejection mechanism in combination with stratification-dependent
vertical mixing produces very salty and dense deep water masses which are
almost completely insulated from the surface. The comparison of the vertical
profiles of ventilation age (Fig. 13) with the basin-averaged data of
Skinner et al. (2017) shows that in the Atlantic and Pacific oceans, even
the standard model version overestimates the radiocarbon ventilation age of
glacial water masses. In turn, the model version with the brine rejection
parameterisation simulates water masses which are 500 to 1000 years older
than in the standard version. Only in the Southern Ocean is the reconstructed
ventilation age consistent with both model versions. As a result, the
standard model version simulates ca. 800 14C yr increase in glacial
ocean ventilation age at the LGM, which is in good agreement with the 689
±53 14C yr reported in Skinner et al. (2017). At the same time,
the model with the brine rejection parameterisation simulates an increase in
the ventilation age of more than 1300 14C yr.
Interestingly, the two model versions do not differ much with respect to
simulated deep ocean δ13C. Finally, the two model versions
differ significantly with respect to the deep Southern Ocean salinity. Change
in salinity in the standard model version is only about 1 psu, which is close
to the global mean salinity change due to ice sheet growth. The model
version with the brine rejection parameterisation simulates glacial deep
South Ocean salinity of 37 psu, which is in good agreement with the
reconstruction by Adkins et al. (2002). Thus we found that
including additional effects (brines and stratification-dependent diffusion)
helps to bring atmospheric Δ14C and the deep Southern Ocean
salinity into better agreement with available reconstructions, but at the expense
of very old (likely to be at odds with paleoclimate data) water masses in
the deep ocean. Of course, these results are obtained with a very simple
ocean component and it is possible that more realistic ocean models would be
able to resolve this apparent contradiction.
Changes in the terrestrial carbon cycle
The evolution of the carbon cycle in the one-way coupled simulation is
presented in Fig. 13. The conventional components of the land carbon
cycle (vegetation biomass, soil carbon stored in non-frozen and non-flooded
environment) change between 1400 Gt C during the LGM and 2000 Gt C during
interglacial peaks. Such an amplitude of 600 Gt C of glacial–interglacial
changes is typical for the models of the land carbon cycle without long-term components (Kaplan et al., 2002; Joos et al., 2004; Brovkin et al.,
2002). However, when we account for permafrost, peat, and buried carbon, the
magnitude decreases to 300–400 Gt C. This is due to the counteracting effect of
the permafrost and buried carbon pools relative to the conventional
components. Both these pools vary between 0 and 350 Gt C and reach their
maxima during glacials. The peat storage also reaches about 350 Gt C, but it
grows only during interglacials or warm stadials. Let us note that during
glacial inceptions, while biomass and mineral soil carbon decrease,
terrestrial carbon storage increases due to an increase in buried and
permafrost carbon. As a result, total land carbon did not change much during
the period of large ice sheet initiation.
During the last deglaciation (Fig. 14, right panels), the peat storages increase
monotonically reaching ca. 350 Gt C at the pre-industrial period. The conventional
carbon pools increase from 1400 to 1800 Gt C at the peak of interglacial
(ca. 9 ka), and then start to decline due to the orbital forcing effect on climate in
the Northern Hemisphere. The permafrost and buried carbon pools show the
opposite behaviour, experiencing minimum at 10 and 5 ka, respectively, and
grow afterwards. The combined effect on the total land carbon is a monotonic
increase during interglacials, mostly because of peat accumulation.
Conclusions
We present here the first simulations of the last four glacial cycles with
the one-way and two-way coupled carbon cycle model. The model is able to
reproduce the major aspects of glacial–interglacial variability of climate,
ice sheets, and of atmospheric CO2 concentration even when driven by the
orbital forcing alone. These results provide strong support to the idea that
long and strongly asymmetric glacial cycles of the late Quaternary represent
a direct but strongly nonlinear response of the Northern Hemisphere ice
sheets to the orbital forcing which, in turn, is amplified and globalised by
the carbon cycle feedback.
The model simulates correct timing of the past glacial terminations in terms
of ice volume while the simulated CO2 concentration lags behind
the reconstructed one by several thousand years. The model is also able to
simulate temporal evolution of the stable carbon isotope in the ocean. At
the same time, the agreement between simulated and reconstructed
atmospheric δ13C is rather poor. Similarly, the magnitude of simulated
atmospheric 14C decline during the last glacial termination is
underestimated. Introducing the brine rejection parameterisation and
stratification-dependent vertical diffusivity allows us to improve the
agreement for atmospheric 14C but leads to unrealistically “old”
glacial deep ocean water masses.
The temporal dynamics of CO2 during interglacial depend strongly on the
timing of the AMOC recovery during glacial termination. If the AMOC
recovers only at the end of glacial termination, CO2 concentration
experiences an overshoot at the beginning of interglacial and then CO2
declines. In contrast, early recovery of the AMOC leads to a monotonic
rise of CO2 during interglacials. In our simulations, millennial-scale
variability during the last glacial termination is very sensitive to the
magnitude of meltwater flux, and the sequence and timing of simulated
millennial-scale events are not robust even when the model is forced by
prescribed radiative forcing of GHGs.
Adding new long-term carbon pools (peat, buried and permafrost carbon)
decreases the amplitude of glacial–interglacial changes in the total land
carbon storage. It helps to reduce an effect of terrestrial biosphere on the
CO2 change during glacial inception and to a lesser extent during glacial terminations.
This work demonstrates that simulation of glacial cycles with Earth system
models driven by orbital forcing alone is possible. This does not mean that
we have presented here the ultimate recipes for all processes and feedbacks and,
in particular, for “the carbon stew”. The understanding of how the global
carbon cycle operates on orbital and suborbital timescales still remains
incomplete, and large uncertainties remain in the choice of individual
processes and their parameterisations. Paleoclimate data provide some useful
constraints but the proxy data syntheses are far from being in a
perfect state, with some proxies telling contradictory stories and others being
difficult to interpret.
The CLIMBER-2 model is a rather simple and coarse resolution Earth system
model. This allows us to perform a large ensemble of model simulations on
orbital and even longer timescales. Obviously, such a fast model has
significant limitations. In particular, it employs the zonally averaged
ocean model. Many essential processes, such as iron fertilisation effect,
are parameterised. The development of a high-resolution state-of-the-art
Earth system model suitable for simulation of the interaction between
climate, ice sheets, and carbon cycle at orbital timescales is
crucial to make the next step forward in understanding of Earth system dynamics during the Quaternary.
The CLIMBER-2 model output that was employed for this
study is available by request.
Appendix AModifications of terrestrial carbon cycle model
The old version of the CLIMBER-2 carbon cycle model described in Brovkin et
al. (2002) considers two vegetation types – trees and grass. Each of the two
vegetation types has four carbon pools – leaves, stems, and fast and slow soil
carbon. Each of these four pools occupies the same fraction of grid cell as
the respective vegetation type. Crichton et al. (2014) in their version of
permafrost carbon implementation in CLIMBER-2 have not changed the pool
structure but modified turnover time, assuming that it is increasing under
permafrost conditions. In the new version of the carbon cycle model which
we use in the present work, we have introduced three new carbon pools: boreal
peat, permafrost, and carbon buried under ice sheets. The
fractions of land covered by grass and trees are computed in the vegetation
model following Brovkin et al. (1997), the fraction of land covered by ice
sheets is computed by the ice sheet model and the fraction of permafrost fpm
for the temperature range -5 ∘C <Tts< 5 ∘C
is computed in the land surface module as
fpm=0.5-0.1Tts,
where Tts is annual mean top soil layer temperature. It is assumed that
grass (in boreal latitudes this mean tundra) is located north of forest and
therefore freezes first. Only if permafrost exceeds grass fraction can the
permafrost expand over the area covered with trees. During ice sheet
growth, all carbon under ice sheets apart from the living biomass is
re-allocated into the buried carbon pool. Buried carbon remains intact until
it is covered by ice sheets. During deglaciation, this buried carbon is
transformed into the permafrost pool. The fraction of land covered by peat is
defined as
fpt=fpt*1-fgc-fpm,
where fpt* is the potential fraction of peat in each grid cell
prescribed from modern observational data and fgc is the fraction of land
covered by ice sheets. Note that we do not consider peatlands in low
latitudes. Although peat and permafrost have certain areal fractions, they
are considered to be parts of grid cells covered by vegetation. Net primary
production and fluxes between the fast carbon pools (leaves, stems, and fast
soil pool) are computed the same way as in Brovkin et al. (2012). The
downward flux of carbon from the fast soil is partitioned between slow soil
pool and permafrost proportionally to their relative fractions. The rate of
peat accumulation is equal to a fixed fraction of net primary production in
the respective vegetation type. Evolution of carbon content pi in slow
carbon pools is described by the equation
dpidt=qifi+dfidtbi′,
where pi=bifi, the bi is the concentration of carbon in
the ith carbon pool (in kgC m-2), and fi is the fraction of the
ith pool. The value qi represents the difference between local accumulation
and decay of carbon in the pool, bi′ is carbon concentration in
the pool by which the ith pool is expanded, and bi′=bi if the
ith pool is shrinking. For peat, bi′= 0. For the permafrost, the
situation is more complex, because it can gain/lose carbon from/to slow
soil, peat, and buried carbon pools. The source term for the permafrost pool
qpm consists of the sum of fluxes from the fast grassland and tree soil
pools into the respective slow pools (see Brovkin et al., 2002, for details)
minus the decay term, where the decay timescale is set to 20 000 years. Apart
from carbon, the terrestrial carbon model also computes carbon isotope
(13C and 14C) contents in all carbon pools. Since carbon isotopes
are also computed in the oceanic carbon cycle model, we can compute
δ13C and δ14C in the atmosphere and compare modelling
results with available paleoclimate data.
Modifications of the ocean carbon cycle modelDust deposition in the Southern Ocean
In the one-way coupled experiments, similar to Brovkin et al. (2012), we
used the concentration of aeolian dust in the Antarctic ice cores as the
proxy for iron deposition over the Southern Ocean. Such choice is supported
by the recent measurements of iron content in Southern Ocean sediment
cores (Lamy et al., 2014). In the fully interactive run, the iron flux
over the Southern Ocean (D) in arbitrary units is parameterised through the
global sea level change as
D=100dSdt+10max(S-50;0)+1.5S,
where S is the ice volume expressed in metres of sea level equivalent and
time t is in years. This formula gives significant increase in iron flux for
the case when sea level drops below 50 m, which is likely related to the fact
that Patagonian dust source is very sensitive to the area of exposed shelf
and glacial erosion processes. Numerical parameters in this formula were
obtained by fitting simulated D to dust concentration in the Antarctic
record. This allows us to use the same parameterisation for the iron
fertilisation effect in the one-way and fully interactive experiments. To
prevent large fluctuations in the iron flux related to fluctuations of the time
derivative of S, the dust deposition D computed by this equation has been
smoothed by applying a relaxation procedure. Namely, at each time step n, the
dust deposition Dn is computed as
Dn=(1-ε)D+εDn-1,
where ε= 0.001, which is approximately equivalent to introducing
a 1000-year filter.
Dependence of remineralisation depth on temperature
In CLIMBER-2 the vertical profile of carbon below the euphotic zone is given
by
f(z)=zzr-0.858,
where remineralisation depth zr is held constant equal to 100 m. To take
into account the dependence of remineralisation rate on ambient temperature,
following Segschneider and Bendtsen (2013), we now use the dependence of zr
on the thermocline temperature (300 m) T:
zr=zro2-T-T010,
where T0= 9 ∘C and zro= 100 m. The value of T0 was
selected in such a way that introducing the temperature-dependent
remineralisation depth does not affect atmospheric CO2 concentration
under pre-industrial climate conditions. During glacial times, temperature in
the thermocline decreases by 2–3 ∘C, which causes an increase in zr by
20–30 %. This results in additional CO2 drawdown by ca. 15 ppm.
Parameterisation of the iron fertilisation effect
The rate of dust deposition which is prescribed from the ice cores in the
one-way coupled experiment or computed from global ice volume in the fully
interactive experiments is considered to be a proxy for iron flux and is
used in the parameterisation of iron the fertilisation mechanism. This
parameterisation is only applied to the Southern Ocean (south of 40∘ S).
As described in Brovkin et al. (2002), net primary production of
phytoplankton Π in the model is computed as
Π=c(D)r(T,R)PP+P0Cp(1-f),
where Cp is the phytoplankton concentration, P is the phosphorus
concentration in the euphotic zone, r is a function of temperature T and
photosynthetic active insolation R, f is the fraction of grid cell covered by
sea ice, P0 is a constant, and c is a function of normalised dust
deposition rate D. Note that in the case of prescribed dust deposition rate,
D was obtained by multiplying observed dust concentration in mg g-1 units by
factor 10-3. North of 40∘ S, parameter c is set to 1 and south of 40∘ S
c=min1,0.1+cdD,
where cd= 2. During glacial maxima the value of c reaches a value of about 1, which implies that during these periods there is no iron limitation in the
Southern Ocean. During interglacials, when D is much smaller than 100, c is
close to 0.1. The parameters of this parameterisation have been selected
(i) to reproduce present-day nutrient concentration in the Southern Ocean, and
(ii) to obtain 20–30 ppm additional CO2 drawdown during glacial maxima
due to the iron fertilisation effect.
Variable volcanic outgassing
Following the idea of Huybers and Langmuir (2009), which has been tested
already in Roth and Joos (2012), we introduced a dependence of volcanic
CO2 outgassing O on the rate of sea level change. Namely, we assume that
volcanic outgassing linearly depends on sea level derivative with the time
delay of about 5000 years:
O(t)=O11-O2dS(t-5000)dt.
Here O1= 5.3 Tmol C yr-1 and O2= 50 Tmol C m-1. With
these parameters, volcanic outgassing does not change by more than 30 %
during all glacial cycles. Note that over one glacial cycle the average
value of O is very close to O1.
Calculation of radiocarbon ventilation age
For the direct comparison of model results with the empirical reconstruction
of marine radiocarbon age offset (Skinner et al., 2017), we calculate the model
analogue of this characteristic using simulated relative concentration of
the radiocarbon in the atmosphere
catm=14Catm/12Catm and in the ocean
cocn=14Cocn/12Cocn using the formula
t=-8033lncocncatm.
Modifications of the energy and surface mass balance interface
In our previous simulations with CLIMBER-2 we found that if maximum ice
sheet volume in the Northern Hemisphere exceeds 100 m, the AMOC remains in
the off mode during the entire deglaciation. Although this may be realistic
for some recent deglaciations, such long AMOC shutdown prevents simulation
of complete deglaciation of North America. This is related to the fact that,
due to a very coarse spatial resolution of CLIMBER-2, linear interpolation
of surface temperature between neighbouring sectors (American and Atlantic)
causes a strong cooling over the eastern part of the Laurentide ice sheet,
due to the AMOC shutdown (see for example Arz et al., 2007). In the high-resolution climate models, the effect of AMOC shutdown on North America is
rather limited compared to Europe (e.g. Zhang et al., 2014; Swingedouw
et al., 2009), which is explained by the predominantly eastward direction of air
mass transport. To compensate for this resolution-related problem, we made the
magnitude of temperature anomaly correction over eastern North America (see
Fig. 2 in Ganopolski et al., 2010) dependent on AMOC
strength. Namely, for the AMOC strength below 10 Sv, the amplitude of
temperature correction is scaled down by a factor of Ψmax/10, where
Ψmax is the maximum of the meridional overturning stream function
(in Sv, 1 Sv = 106 m3 s-1) in the Atlantic Ocean. With this
parameterisation, during complete shutdown of the AMOC, cooling over eastern
North America is compensated for by reducing the temperature correction.
Introducing this procedure minimises the impact of the AMOC on Laurentide
ice sheet mass balance. As the result, now even prolonged AMOC shutdown
does not prevent complete melting of the Laurentide ice sheet during glacial terminations.
The authors declare that they have no conflict of interest.
Acknowledgements
The authors thank Edouard Bard and Fortunat Joos
for helpful discussion of the atmospheric and oceanic 14C dynamics and
Luke Skinner for useful comments and suggestions. The authors acknowledge
support of the German Ministry of Education and Research (PalMod project).
Edited by: André Paul
Reviewed by: Luke Skinner and one anonymous referee
ReferencesAbe-Ouchi, A., Saito, F., Kawamura, K., Raymo, M. E., Okuno, J., Takahashi, K.,
and Blatter, H.: Insolation-driven 100,000-year glacial cycles and hysteresis
of ice-sheet volume, Nature, 500, 190–194, 10.1038/nature12374, 2013.Adkins, J. F., McIntyre, K., and Schrag, D. P.: The salinity, temperature, and
δ18O of the glacial deep ocean, Science, 298, 1769–1773,
10.1126/science.1076252, 2002.Archer, D.: A data-driven model of the global calcite lysocline, Global Biogeochem.
Cy., 10, 511–526, 10.1029/96gb01521, 1996.Archer, D., Winguth, A., Lea, D., and Mahowald, N.: What caused the glacial/interglacial
atmospheric pCO2 cycles?, Rev. Geophys., 38, 159–189, 10.1029/1999rg000066, 2000.Arz, H. W., Lamy, F., Ganopolski, A., Nowaczyk, N., and Patzold, J.: Dominant
Northern Hemisphere climate control over millennial-scale glacial sea-level
variability, Quaternary Sci. Rev., 26, 312–321, 10.1016/j.quascirev.2006.07.016, 2007.Augustin, L., Barbante, C., Barnes, P. R. F., Barnola, J. M., Bigler, M.,
Castellano, E., Cattani, O., Chappellaz, J., DahlJensen, D., Delmonte, B.,
Dreyfus, G., Durand, G., Falourd, S., Fischer, H., Fluckiger, J., Hansson, M. E.,
Huybrechts, P., Jugie, R., Johnsen, S. J., Jouzel, J., Kaufmann, P., Kipfstuhl,
J., Lambert, F., Lipenkov, V. Y., Littot, G. V. C., Longinelli, A., Lorrain, R.,
Maggi, V., Masson-Delmotte, V., Miller, H., Mulvaney, R., Oerlemans, J., Oerter,
H., Orombelli, G., Parrenin, F., Peel, D. A., Petit, J. R., Raynaud, D., Ritz,
C., Ruth, U., Schwander, J., Siegenthaler, U., Souchez, R., Stauffer, B.,
Steffensen, J. P., Stenni, B., Stocker, T. F., Tabacco, I. E., Udisti, R.,
van de Wal, R. S. W., van den Broeke, M., Weiss, J., Wilhelms, F., Winther, J.
G., Wolff, E. W., and Zucchelli, M.: Eight glacial cycles from an Antarctic ice
core, Nature, 429, 623–628, 10.1038/nature02599, 2004.Barnola, J. M., Raynaud, D., Korotkevich, Y. S., and Lorius, C.: Vostok ice
core privides 160,000-year record of atmospheric CO2, Nature, 329, 408–414,
10.1038/329408a0, 1987.Berger, A., Li, X. S., and Loutre, M. F.: Modelling northern hemisphere ice
volume over the last 3 Ma, Quaternary Sci. Rev., 18, 1–11, 10.1016/s0277-3791(98)00033-x, 1999.Bouttes, N., Paillard, D., and Roche, D. M.: Impact of brine-induced stratification
on the glacial carbon cycle, Clim. Past., 6, 575–589, 10.5194/cp-6-575-2010, 2010.Brady, E. C., Otto-Bliesner, B. L., Kay, J. E., and Rosenbloom, N.: Sensitivity
to Glacial Forcing in the CCSM4, J. Climate, 26, 1901–1925, 10.1175/jcli-d-11-00416.1, 2013.Brovkin, V., Ganopolski, A., and Svirezhev, Y.: A continuous climate–vegetation
classification for use in climate-biosphere studies, Ecol. Model., 101, 251–261,
10.1016/s0304-3800(97)00049-5, 1997.Brovkin, V., Bendtsen, J., Claussen, M., Ganopolski, A., Kubatzki, C., Petoukhov,
V., and Andreev, A.: Carbon cycle, vegetation, and climate dynamics in the
Holocene: Experiments with the CLIMBER-2 model, Global Biogeochem. Cy., 16,
1139, 10.1029/2001gb001662, 2002.Brovkin, V., Ganopolski, A., Archer, D., and Rahmstorf, S.: Lowering of glacial
atmospheric CO2 in response to changes in oceanic circulation and marine
biogeochemistry, Paleoceanography, 22, PA4202, 10.1029/2006pa001380, 2007.Brovkin, V., Ganopolski, A., Archer, D., and Munhoven, G.: Glacial CO2
cycle as a succession of key physical and biogeochemical processes, Clim. Past.,
8, 251–264, 10.5194/cp-8-251-2012, 2012.Brovkin, V., Bruecher, T., Kleinen, T., Zaehle, S., Joos, F., Roth, R., Spahni,
R., Schmitt, J., Fischer, H., Leuenberger, M., Stone, E. J., Ridgwell, A.,
Chappellaz, J., Kehrwald, N., Barbante, C., Blunier, T., and Jensen, D. D.:
Comparative carbon cycle dynamics of the present and last interglacial, Quaternary
Sci. Rev., 137, 15–32, 10.1016/j.quascirev.2016.01.028, 2016.Buchanan, P. J., Matear, R. J., Lenton, A., Phipps, S. J., Chase, Z., and
Etheridge, D. M.: The simulated climate of the Last Glacial Maximum and insights
into the global marine carbon cycle, Clim. Past., 12, 2271–2295, 10.5194/cp-12-2271-2016, 2016.Burley, J. M. A. and Katz, R. F.: Variations in mid-ocean ridge CO2 emissions
driven by glacial cycles, Earth Planet. Sc. Lett., 426, 246–258, 10.1016/j.epsl.2015.06.031, 2015.Calov, R., Ganopolski, A., Claussen, M., Petoukhov, V., and Greve, R.: Transient
simulation of the last glacial inception. Part I: glacial inception as a
bifurcation in the climate system, Clim. Dynam., 24, 545–561, 10.1007/s00382-005-0007-6, 2005.Ciais, P., Tagliabue, A., Cuntz, M., Bopp, L., Scholze, M., Hoffmann, G.,
Lourantou, A., Harrison, S. P., Prentice, I. C., Kelley, D. I., Koven, C., and
Piao, S. L.: Large inert carbon pool in the terrestrial biosphere during the
Last Glacial Maximum, Nat. Geosci., 5, 74–79, 10.1038/ngeo1324, 2012.Crichton, K. A., Bouttes, N., Roche, D. M., Chappellaz, J., and Krinner, G.:
Permafrost carbon as a missing link to explain CO2 changes during the last
deglaciation, Nat. Geosci., 9, 683–686, 10.1038/ngeo2793, 2014.Crowley, T. J. and Hyde, W. T.: Transient nature of late Pleistocene climate
variability, Nature, 456, 226–230, 10.1038/nature07365, 2008.Eggleston, S., Schmitt, J., Bereiter, B., Schneider, R., and Fischer, H.:
Evolution of the stable carbon isotope composition of atmospheric CO2
over the last glacial cycle, Paleoceanography, 31, 434–452, 10.1002/2015pa002874, 2016.Fischer, H., Schmitt, J., Luthi, D., Stocker, T. F., Tschumi, T., Parekh, P.,
Joos, F., Köhler, P., Volker, C., Gersonde, R., Barbante, C., Le Floch, M.,
Raynaud, D., and Wolff, E.: The role of Southern Ocean processes in orbital and
millennial CO2 variations – A synthesis, Quaternary Sci. Rev., 29, 193–205,
10.1016/j.quascirev.2009.06.007, 2010.Galbraith, E. D. and Jaccard, S. L.: Deglacial weakening of the oceanic soft
tissue pump: global constraints from sedimentary nitrogen isotopes and
oxygenation proxies, Quaternary Sci. Rev., 109, 38–48, 10.1016/j.quascirev.2014.11.012, 2015.Ganopolski, A. and Brovkin, V.: The last four glacial CO2 cycles simulated
with the CLIMBER-2 model, in: Deglacial Changes in Ocean Dynamics and Atmospheric
CO2. Modern, Glacial, and Deglacial Carbon Transfer between Ocean, Atmosphere,
and Land, edited by: Sarnthein, M. and Haug, G., Deutsche Akademie der Naturforscher
Leopoldina, Halle (Saale), 75–80, 2015.Ganopolski, A. and Calov, R.: The role of orbital forcing, carbon dioxide and
regolith in 100 kyr glacial cycles, Clim. Past., 7, 1415–1425, 10.5194/cp-7-1415-2011, 2011.Ganopolski, A. and Roche, D. M.: On the nature of lead-lag relationships during
glacial-interglacial climate transitions, Quaternary Sci. Rev., 28, 3361–3378,
10.1016/j.quascirev.2009.09.019, 2009.Ganopolski, A., Petoukhov, V., Rahmstorf, S., Brovkin, V., Claussen, M., Eliseev,
A., and Kubatzki, C.: CLIMBER-2: a climate system model of intermediate complexity.
Part II: model sensitivity, Clim. Dynam., 17, 735–751, 10.1007/s003820000144, 2001.Ganopolski, A., Calov, R., and Claussen, M.: Simulation of the last glacial cycle
with a coupled climate ice-sheet model of intermediate complexity, Clim. Past,
6, 229–244, 10.5194/cp-6-229-2010, 2010.Ganopolski, A., Winkelmann, R., and Schellnhuber, H. J.: Critical insolation-CO2
relation for diagnosing past and future glacial inception, Nature, 529, 200–204,
10.1038/nature16494, 2016.Greve, R.: A continuum-mechanical formulation for shallow polythermal ice sheets,
Philos. T. Roy. Soc. A, 355, 921–974, 10.1098/rsta.1997.0050, 1997.Hain, M. P., Sigman, D. M., and Haug, G. H.: Distinct roles of the Southern
Ocean and North Atlantic in the deglacial atmospheric radiocarbon decline,
Earth Planet. Sc. Lett., 394, 198–208, 10.1016/j.epsl.2014.03.020, 2014.Heinze, C., Hoogakker, B. A. A., and Winguth, A.: Ocean carbon cycling during
the past 130 000 years – a pilot study on inverse palaeoclimate record modelling,
Clim. Past., 12, 1949–1978, 10.5194/cp-12-1949-2016, 2016.Hesse, T., Butzin, M., Bickert, T., and Lohmann, G.: A model-data comparison
of δ13C in the glacial Atlantic Ocean, Paleoceanography, 26, PA3220,
10.1029/2010PA002085, 2011.Huybers, P. and Langmuir, C.: Feedback between deglaciation, volcanism, and
atmospheric CO2, Earth Planet. Sci. Lett., 286, 479–491, 10.1016/j.epsl.2009.07.014, 2009.Jaccard, S. L., Galbraith, E. D., Martinez-Garcia, A., and Anderson, R. F.:
Covariation of deep Southern Ocean oxygenation and atmospheric CO2 through
the last ice age, Nature, 530, 207–210, 10.1038/nature16514, 2016.Joos, F., Gerber, S., Prentice, I. C., Otto-Bliesner, B. L., and Valdes, P. J.:
Transient simulations of Holocene atmospheric carbon dioxide and terrestrial
carbon since the Last Glacial Maximum, Global Biogeochem. Cy., 18, GB2002,
10.1029/2003gb002156, 2004.Jouzel, J., Masson-Delmotte, V., Cattani, O., Dreyfus, G., Falourd, S., Hoffmann,
G., Minster, B., Nouet, J., Barnola, J. M., Chappellaz, J., Fischer, H., Gallet,
J. C., Johnsen, S., Leuenberger, M., Loulergue, L., Luethi, D., Oerter, H.,
Parrenin, F., Raisbeck, G., Raynaud, D., Schilt, A., Schwander, J., Selmo, E.,
Souchez, R., Spahni, R., Stauffer, B., Steffensen, J. P., Stenni, B., Stocker,
T. F., Tison, J. L., Werner, M., and Wolff, E. W.: Orbital and millennial
Antarctic climate variability over the past 800,000 years, Science, 317, 793–796,
10.1126/science.1141038, 2007.Kaplan, J. O., Prentice, I. C., Knorr, W., and Valdes, P. J.: Modeling the
dynamics of terrestrial carbon storage since the Last Glacial Maximum, Geophys.
Res. Lett., 29, 2074, 10.1029/2002gl015230, 2002.Kleinen, T., Brovkin, V., and Munhoven, G.: Modelled interglacial carbon cycle
dynamics during the Holocene, the Eemian and Marine Isotope Stage (MIS) 11,
Clim. Past., 12, 2145–2160, 10.5194/cp-12-2145-2016, 2016.Kleypas, J. A.: Modeled estimates of global reef habitat and carbonate production
since the last glacial maximum, Paleoceanography, 12, 533–545, 10.1029/97pa01134, 1997.Kobayashi, H., Abe-Ouchi, A., and Oka, A.: Role of Southern Ocean stratification
in glacial atmospheric CO2 reduction evaluated by a three-dimensional ocean
general circulation model, Paleoceanography, 30, 1202–1216, 10.1002/2015pa002786, 2015.Köhler, P., Fischer, H., and Schmitt, J.: Atmospheric δ13CO2
and its relation to pCO2 and deep ocean δ13C during the late
Pleistocene, Paleoceanography, 25, PA1213, 10.1029/2008pa001703, 2010.Kovaltsov, G. A., Mishev, A., and Usoskin, I. G.: A new model of cosmogenic
production of radiocarbon 14C in the atmosphere, Earth Planet. Sc. Lett.,
337, 114–120, 10.1016/j.epsl.2012.05.036, 2012.Lambert, F., Tagliabue, A., Shaffer, G., Lamy, F., Winckler, G., Farias, L.,
Gallardo, L., and De Pol-Holz, R.: Dust fluxes and iron fertilization
in Holocene
and Last Glacial Maximum climates, Geophys. Res. Lett., 42, 6014–6023,
10.1002/2015gl064250, 2015.Lamy, F., Gersonde, R., Winckler, G., Esper, O., Jaeschke, A., Kuhn, G., Ullermann,
J., Martinez-Garcia, A., Lambert, F., and Kilian, R.: Increased Dust Deposition
in the Pacific Southern Ocean During Glacial Periods, Science, 343, 403–407,
10.1126/science.1245424, 2014.Lisiecki, L. E., Raymo, M. E., and Curry, W. B.: Atlantic overturning responses
to Late Pleistocene climate forcings, Nature, 456, 85–88, 10.1038/nature07425, 2008.Luthi, D., Le Floch, M., Bereiter, B., Blunier, T., Barnola, J. M., Siegenthaler,
U., Raynaud, D., Jouzel, J., Fischer, H., Kawamura, K., and Stocker, T. F.:
High-resolution carbon dioxide concentration record 650,000–800,000 years
before present, Nature, 453, 379–382, 10.1038/nature06949, 2008.Mahowald, N., Kohfeld, K., Hansson, M., Balkanski, Y., Harrison, S. P., Prentice,
I. C., Schulz, M., and Rodhe, H.: Dust sources and deposition during the last
glacial maximum and current climate: A comparison of model results with paleodata
from ice cores and marine sediments, J. Geophys. Res.-Atmos., 104, 15895–15916,
10.1029/1999jd900084, 1999.Mariotti, V., Paillard, D., Bopp, L., Roche, D. M., and Bouttes, N.: A coupled
model for carbon and radiocarbon evolution during the last deglaciation, Geophys.
Res. Lett., 43, 1306–1313, 10.1002/2015gl067489, 2016.Martin, J. H.: Glacial-Interglacial CO2 change: The iron hypothesis,
Paleoceanography, 5, 1–13, 10.1029/PA005i001p00001, 1990.Martinez-Garcia, A., Sigman, D. M., Ren, H. J., Anderson, R. F., Straub, M.,
Hodell, D. A., Jaccard, S. L., Eglinton, T. I., and Haug, G. H.: Iron
Fertilization of the Subantarctic Ocean During the Last Ice Age, Science,
343, 1347–1350, 10.1126/science.1246848, 2014.Matsumoto, K.: Biology-mediated temperature control on atmospheric pCO2
and ocean biogeochemistry, Geophys. Res. Lett., 34, L20605, 10.1029/2007GL031301, 2007.Menviel, L., Joos, F., and Ritz, S. P.: Simulating atmospheric CO2,
13C and the marine carbon cycle during the Last Glacial-Interglacial
cycle: possible role for a deepening of the mean remineralization depth and
an increase in the oceanic nutrient inventory, Quaternary Sci. Rev., 56, 46–68,
10.1016/j.quascirev.2012.09.012, 2012.Mignot, J., Ganopolski, A., and Levermann, A.: Atlantic subsurface temperatures:
Response to a shutdown of the overturning circulation and consequences for its
recovery, J. Climate, 20, 4884–4898, 10.1175/jcli4280.1, 2007.Monnin, E., Steig, E. J., Siegenthaler, U., Kawamura, K., Schwander, J.,
Stauffer, B., Stocker, T. F., Morse, D. L., Barnola, J. M., Bellier, B., Raynaud,
D., and Fischer, H.: Evidence for substantial accumulation rate variability in
Antarctica during the Holocene, through synchronization of CO2 in the
Taylor Dome, Dome C and DML ice cores, Earth Planet. Sc. Lett., 224, 45–54,
10.1016/j.epsl.2004.05.007, 2004.Munhoven, G.: Glacial-interglacial changes of continental weathering: estimates
of the related CO2 and HCO3-flux variations and their uncertainties,
Global Planet. Change, 33, 155–176, 10.1016/s0921-8181(02)00068-1, 2002.Opdyke, B. N. and Walker, J. C. G.: Return of the coral-reef hypothesis – basin
to shelf partition of CaCO3 and its effect on atmospheric CO2, Geology,
20, 733–736, 10.1130/0091-7613, 1992.Paillard, D.: The timing of Pleistocene glaciations from a simple multiple-state
climate model, Nature, 391, 378–381, 10.1038/34891, 1998.Peltier, W. R.: Global glacial isostasy and the surface of the ice-age earth:
The ice-5G (VM2) model and grace, Annu. Rev. Earth Planet. Sci., 32, 111–149,
10.1146/annurev.earth.32.082503.144359, 2004.Petit, J. R., Jouzel, J., Raynaud, D., Barkov, N. I., Barnola, J. M., Basile,
I., Bender, M., Chappellaz, J., Davis, M., Delaygue, G., Delmotte, M., Kotlyakov,
V. M., Legrand, M., Lipenkov, V. Y., Lorius, C., Pepin, L., Ritz, C., Saltzman,
E., and Stievenard, M.: Climate and atmospheric history of the past 420,000 years
from the Vostok ice core, Antarctica, Nature, 399, 429–436, 10.1038/20859, 1999.Petoukhov, V., Ganopolski, A., Brovkin, V., Claussen, M., Eliseev, A., Kubatzki,
C., and Rahmstorf, S.: CLIMBER-2: a climate system model of intermediate complexity.
Part I: model description and performance for present climate, Clim. Dynam.,
16, 1–17, 10.1007/pl00007919, 2000.Prentice, I. C., Harrison, S. P., and Bartlein, P. J.: Global vegetation and
terrestrial carbon cycle changes after the last ice age, New Phytol., 189,
988–998, 10.1111/j.1469-8137.2010.03620.x, 2011.
Reimer, P. J., Bard, E., Bayliss, A., Beck, J. W., Blackwell, P. G., Ramsey,
C. B., Buck, C. E., Cheng, H., Edwards, R. L., Friedrich, M., Grootes, P. M.,
Guilderson, T. P., Haflidason, H., Hajdas, I., Hatte, C., Heaton, T. J., Hoffmann,
D. L., Hogg, A. G., Hughen, K. A., Kaiser, K. F., Kromer, B., Manning, S. W.,
Niu, M., Reimer, R. W., Richards, D. A., Scott, E. M., Southon, J. R., Staff,
R. A., Turney, C. S. M., and van der Plicht, J.: Intrcal13 and marine13
radiocarbon age calibration curves 0–50,000 years cal BP, Radiocarbon,
55, 1869–1887, 2013.Roberts, J., Gottschalk, J., Skinner, L. C., Peck, V. L., Kender, S., Elderfield,
H., Waelbroeck, C., Riveiros, N. V., and Hodell, D. A.: Evolution of South
Atlantic density and chemical stratification across the last deglaciation, P.
Natl. Acad. Sci. USA, 113, 514–519, 10.1073/pnas.1511252113, 2016.
Robinson, L. F., Adkins, J. F., Keigwin, L. D., Southon, J., Fernandez, D. P.,
and Scheirer, D. S.: Radiocarbon variability in the western North Atlantic
during the last deglaciation, Science, 310, 1469–1473, 2005.Roth, R. and Joos, F.: Model limits on the role of volcanic carbon emissions in
regulating glacial-interglacial CO2 variations, Earth Planet. Sc. Lett.,
329, 141–149, 10.1016/j.epsl.2012.02.019, 2012.Schmitt, J., Schneider, R., Elsig, J., Leuenberger, D., Lourantou, A., Chappellaz,
J., Köhler, P., Joos, F., Stocker, T. F., Leuenberger, M., and Fischer, H.:
Carbon Isotope Constraints on the Deglacial CO2 Rise from Ice Cores,
Science, 336, 711–714, 10.1126/science.1217161, 2012.Schmittner, A. and Somes, C. J.: Complementary constraints from carbon (13C)
and nitrogen (15N) isotopes on the glacial ocean's soft-tissue biological
pump, Paleoceanography, 31, 669–693, 10.1002/2015pa002905, 2016.Segschneider, J., and Bendtsen, J.: Temperature-dependent remineralization in
a warming ocean increases surface pCO2 through changes in marine ecosystem
composition, Global Biogeochem. Cy., 27, 1214–1225, 10.1002/2013gb004684, 2013.Sigman, D. M. and Boyle, E. A.: Glacial/interglacial variations in atmospheric
carbon dioxide, Nature, 407, 859–869, 10.1038/35038000, 2000.Six, K. D. and Maier-Reimer, E.: Effects of plankton dynamics on seasonal carbon
fluxes in an ocean general circulation model, Global Biogeochem. Cy., 10,
559–583, 10.1029/96gb02561, 1996.Skinner, L. C., Waelbroeck, C., Scrivner, A., Fallon, E., and Stewart, J.:
Radiocarbon evidence for alternating northern and southern sources of ventilation
of the deep Atlantic carbon pool during the last deglaciation, P. Natl. Acad.
Sci. USA, 111, 5480–5484, 10.1073/pnas.1400668111, 2014.Skinner, L. C., Primeau, F., Freeman, E., de la Fuente, M., Goodwin, P. A.,
Gottschalk, J., Huang, E., McCave, I. N., Noble, T. L., and Scrivner, A. E.:
Radiocarbon constraints on the glacial ocean circulation and its impact on
atmospheric CO2, Nat. Commun., 8, 10, 10.1038/ncomms16010, 2017.Spratt, R. M. and Lisiecki, L. E.: A Late Pleistocene sea level stack, Clim.
Past., 12, 1079–1092, 10.5194/cp-12-1079-2016, 2016.Stephens, B. B. and Keeling, R. F.: The influence of Antarctic sea ice on
glacial-interglacial CO2 variations, Nature, 404, 171–174, 10.1038/35004556, 2000.Sugden, D. E., McCulloch, R. D., Bory, A. J. M., and Hein, A. S.: Influence of
Patagonian glaciers on Antarctic dust deposition during the last glacial period,
Nat. Geosci., 2, 281–285, 10.1038/ngeo474, 2009.Swingedouw, D., Mignot, J., Braconnot, P., Mosquet, E., Kageyama, M., and Alkama,
R.: Impact of Freshwater Release in the North Atlantic under Different Climate
Conditions in an OAGCM, J. Climate, 22, 6377–6403, 10.1175/2009jcli3028.1, 2009.Toggweiler, J. R., Russell, J. L., and Carson, S. R.: Midlatitude westerlies,
atmospheric CO2, and climate change during the ice ages, Paleoceanography,
21, PA2005, 10.1029/2005pa001154, 2006.Waelbroeck, C., Labeyrie, L., Michel, E., Duplessy, J. C., McManus, J. F.,
Lambeck, K., Balbon, E., and Labracherie, M.: Sea-level and deep water temperature
changes derived from benthic foraminifera isotopic records, Quaternary Sci. Rev.,
21, 295–305, 10.1016/s0277-3791(01)00101-9, 2002.Wallmann, K., Schneider, B., and Sarnthein, M.: Effects of eustatic sea-level
change, ocean dynamics, and nutrient utilization on atmospheric pCO2 and
seawater composition over the last 130 000 years: a model study, Clim. Past.,
12, 339–375, 10.5194/cp-12-339-2016, 2016.
Watson, A. J., Bakker, D. C. E., Ridgwell, A. J., Boyd, P. W., and Law, C. S.:
Effect of iron supply on Southern Ocean CO2 uptake and implications for
glacial atmospheric CO2, Nature, 407, 730–733, 10.1038/35037561, 2000.Watson, A. J., Vallis, G. K., and Nikurashin, M.: Southern Ocean buoyancy
forcing of ocean ventilation and glacial atmospheric CO2, Nat. Geosci.,
8, 861–864, 10.1038/ngeo2538, 2015.Willeit, M., Ganopolski, A., Calov, R., Robinson, A., and Maslin, M.: The role
of CO2 decline for the onset of Northern Hemisphere glaciation, Quaternary
Sci. Rev., 119, 22–34, 10.1016/j.quascirev.2015.04.015, 2015.Wolff, E. W., Fischer, H., Fundel, F., Ruth, U., Twarloh, B., Littot, G. C.,
Mulvaney, R., Rothlisberger, R., de Angelis, M., Boutron, C. F., Hansson, M.,
Jonsell, U., Hutterli, M. A., Lambert, F., Kaufmann, P., Stauffer, B., Stocker,
T. F., Steffensen, J. P., Bigler, M., Siggaard-Andersen, M. L., Udisti, R.,
Becagli, S., Castellano, E., Severi, M., Wagenbach, D., Barbante, C., Gabrielli,
P., and Gaspari, V.: Southern Ocean sea-ice extent, productivity and iron flux
over the past eight glacial cycles, Nature, 440, 491–496, 10.1038/nature04614, 2006.Wunsch, C.: Last Glacial Maximum and deglacial abyssal seawater oxygen isotopic
ratios, Clim. Past., 12, 1281–1296, 10.5194/cp-12-1281-2016, 2016.Zhang, X., Lohmann, G., Knorr, G., and Purcell, C.: Abrupt glacial climate shifts
controlled by ice sheet changes, Nature, 512, 290–294, 10.1038/nature13592, 2014.