koehler2014nc

koehler2014nc
ARTICLE
Received 7 Apr 2014 | Accepted 9 Oct 2014 | Published 20 Nov 2014
DOI: 10.1038/ncomms6520
OPEN
Permafrost thawing as a possible source of abrupt
carbon release at the onset of the Bølling/Allerød
Peter Köhler1, Gregor Knorr1,2 & Edouard Bard3
One of the most abrupt and yet unexplained past rises in atmospheric CO2 (410 p.p.m.v. in
two centuries) occurred in quasi-synchrony with abrupt northern hemispheric warming into
the Bølling/Allerød, B14,600 years ago. Here we use a U/Th-dated record of atmospheric
D14C from Tahiti corals to provide an independent and precise age control for this CO2 rise.
We also use model simulations to show that the release of old (nearly 14C-free) carbon can
explain these changes in CO2 and D14C. The D14C record provides an independent constraint
on the amount of carbon released (B125 Pg C). We suggest, in line with observations of
atmospheric CH4 and terrigenous biomarkers, that thawing permafrost in high northern
latitudes could have been the source of carbon, possibly with contribution from flooding of
the Siberian continental shelf during meltwater pulse 1A. Our findings highlight the potential
of the permafrost carbon reservoir to modulate abrupt climate changes via greenhouse-gas
feedbacks.
1 Alfred-Wegener-Institut Helmholtz-Zentrum für Polar-und Meeresforschung (AWI), P.O. Box 12 01 61, D-27515 Bremerhaven, Germany. 2 School of Earth
and Ocean Sciences, Cardiff University, Cardiff CF10 3AT, UK. 3 CEREGE, Aix Marseille University, CNRS, IRD, College de France, B.P. 80 Technopole de
l’Arbois, 13545 Aix-en-Provence, France. Correspondence and requests for materials should be addressed to P.K. (email: [email protected]).
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
1
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
C
hanges in the global carbon cycle during the last
deglaciation are so far not completely understood.
However, based on the data and model-based interpretation, the emerging picture indicates that the rise in atmospheric
CO2 of B45 p.p.m.v. during the first half of the deglaciation
(B1 p.p.m.v. per century) was probably fuelled by the release of
old, 13C- and 14C-depleted deep ocean carbon1,2. The processes
responsible for CO2 rise have changed dramatically with the
beginning of the Bølling/Allerød (B/A) B14,600 years before
present (B14.6 kyr BP). Here the abrupt CO2 rise recorded in the
EPICA Dome C (EDC) ice core3,4 was six times faster than
before, about 10 p.p.m.v. in 180 years or B6 p.p.m.v. per century
(Fig. 1). Atmospheric CH4 rose by 150 p.p.b.v. between 18.5 and
14.6 kyr BP and then by the same amount again, but within
450
14.6 kyr BP
350
300
250
Tahiti
200
IntCal13
–0.04‰ per year
150
100
270
260
250
240
230
220
210
200
190
180
–33
800
750
700
550
500
450
–34
400
–35
350
–36
300
2
0
–37
–2
–38
–30
–39
–32
–40
–34
–41
–36
–38
–42
–40
–43
–44
20
4
–42
18
16
14
Time (kyr BP)
12
–44
10
–4
∆TANT (K)
600
CH4 (p.p.b.v.)
650
NGRIP !18O (‰)
EDC CO2 (p.p.m.v.)
–0.10‰ per year
WD !18O (‰)
Atm. ∆14C (‰)
400
–6
–8
–10
–12
Figure 1 | Relevant ice core and D14C data during Termination I.
Atmospheric D14C based on Tahiti corals (magenta circles, mean ±1s)7 or
IntCal13 (grey area, ±1s uncertainty band around the mean)9, the latter
including linear trends with ! 0.04% per year or ! 0.10% per year; CO2
from EDC (blue filled circles)1,22,68; CH4 from EDC (blue)22 and Greenland
(red)69; WAIS Divide (WD) d18O (original (blue) and 100 years running
mean (yellow))54; stack of calculated temperature change3 DTANT from the
five East Antarctic ice cores EDC, EPICA DML, Vostok, Dome Fuji and Talos
Dome (original (black) and 100 years running mean (orange)); NGRIP70
d18O. All Greenland records on GICC05 (ref. 23), all EDC records on
AICC2012 chronology4, WD and DTANT on their own independent
chronology.
2
centuries, around the onset of the B/A. The changes in both
greenhouse gases (GHG) imply that a ratio of both changes
DCH4/DCO2 is a factor of five larger around 14.6 kyr BP than
during the previous four millennia. Such a change in the ratio
DCH4/DCO2 might be the first indication that the wetlands
identified5 as the main contributor to the rapid rise in CH4 at the
onset of the B/A might also have contributed to the abrupt rise in
CO2 at that time.
Although this analysis of CH4 and CO2 changes gives some
first ideas on the potential cause of the abrupt CO2 rise around
the onset of the B/A, its ultimate source was so far not identified.
The d13C signature of terrestrial or marine carbon sources are
different and might allow some source detection. However, the
data uncertainty and density of the atmospheric d13CO2 record
did so far not allow such an identification6. A high-resolution
U/Th-dated time series of atmospheric D14C derived from Tahiti
corals7 over that event offers now some new and independent
insights on the exact timing and magnitude of the carbon release
event and brings some suggestions on its potential origin.
Here we show that the synchronous change in atmospheric
D14C and CO2 derived from the Tahiti and EDC data sets at the
onset of the B/A can be explained by the same process and
suggest permafrost thawing being this process. We finally
examine the climate impact of the GHG changes around
14.6 kyr BP using a state-of-the-art-coupled Earth system model.
Special focus of these investigations is the imprint of the GHG
changes on the Antarctic temperature signature and the relevance
of these changes for the interpretation of bipolar climate linkages
during abrupt climate changes8.
Results
Atmospheric D14C and ice core CO2. The new coral-based
atmospheric D14C record from Tahiti7 shows a prominent decline
around 14.6 kyr BP, an anomaly not visible in the IntCal13 D14C
stack9 (Fig. 2). For comparison, we briefly discuss specific details
related to IntCal13 and what other 14C archives record at that
point in time: After 13.9 kyr BP IntCal13 is based on tree rings
with very little variability. For older samples, however, the various
archives differ by more than the measurement errors. A D14C
anomaly similar to the Tahiti data can be seen in speleothems
from Bahamas10 (Fig. 2). The anomaly is not seen in speleothems
from the Hulu Cave11 or in the marine sediments from Cariaco12
(Fig. 2). The Cariaco record bears some problems—therefore, a
part of it has been excluded by the IntCal13 group specifically
during the Heinrich 1 event, that is, just before the B/A9.
Necessary corrections of speleothem 14C data for its dead carbon
fraction (DCF) introduce large uncertainties to atmospheric D14C
based on them9. Furthermore, the DCF is not constant but
depends itself on climate13 making the speleothems an archive
difficult to interpret, especially during rapid climate changes. The
signal might thus potentially be smoothed out in Hulu, as the
DCF acts as a low-pass filter. The best recorder of atmospheric
D14C available up to now might be the terrestrial plant material
derived from Lake Suigetsu14. Here no corrections for the
reservoir effect or for DCF are necessary. The Lake Suigetsu data,
however, are rather scattered over the time interval of interest,
show a steeper decline in D14C than IntCal13, but neither
strongly support IntCal13 nor Tahiti (Fig. 2). Altogether, the
evidences from D14C data are mixed and further data are
necessary for a conclusive interpretation.
The coral-based D14C record from Tahiti is corrected for a
reservoir age15 of constantly 300 14C years7 to be interpreted as
atmospheric D14C. In principle, the reservoir age might change
over time, mainly due to ocean circulation changes. However,
simulations with three different models16–18 suggest that the
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
300
Tahiti:
Data
Linear model
NL model
280
IntCal13:
Mean stack ± 1"
–0.04‰ per year
–0.10‰ per year
Atm. ∆14C (‰)
260
240
220
200
180
160
15.0
Other records:
Bahamas
Hulu
Cariaco at Hulu2
Lake Suigetsu
14.8
14.6
14.4
14.2
14.0
Time (kyr BP)
Figure 2 | Data analysis of atmospheric D14C around 14.6 kyr BP.
Atmospheric D14C based on Tahiti corals7 (magenta circles) and IntCal13
(grey line, ±1s uncertainty band around the mean)9 are analysed for
trends and compared with various other archives (speleothems from
Bahamas10 (blue diamonds), Hulu Cave11 (dark green squares), marine
sediments in Cariaco12 (light green triangles) plotted on revised Hulu2 age
model9, Lake Suigetsu14 (orange open squares)). All individual data points
are plotted with ±1s in both age and D14C. IntCal13 was approximated by a
linear trend with either ! 0.04% per year (black solid line) or ! 0.10%
per year (black dashed line). Tahiti data were analysed for break points with
two different models (see Methods). For the linear model (red lines), a
statistical package was used. For the non-linear (NL) model (cyan lines),
two data points at beginning of the Tahiti data D14C anomaly from the
IntCal13 data and the eight points around the local minimum (black open
circles) were averaged, plotted with ±1s in both age and D14C (bold large
black open circles) and further analysed. The anomaly in the Tahiti D14C
data following the linear model is D(D14C) ¼ ! 54±8% in
D(age) ¼ 207±95 years and following the NL model: D(D14C) ¼
! 58±14% in D(age) ¼ 258±53 years.
reservoir age is relatively stable in the central Pacific around
Tahiti for various ocean circulation changes (see Supplementary
Note 1 for details). We therefore assume that reservoir ages did
not change over the last 15 kyr in the central low-latitude Pacific
and the D14C signal based on Tahiti corals is not based on local
effects but indeed a recorder of atmospheric D14C changes.
We date the start of this D14C decline seen in the Tahiti data
with two different approaches (Fig. 2, methods) with a 1s
uncertainty of less than a century to 14.6 kyr BP and calculate a
D14C decline of B55% within 200 to 250 years. Having already
excluded changes in reservoir age, we are left with either a
modified carbon cycle or reduced 14C production rates as
potential process explaining the Tahiti D14C data. On the basis
of available 10Be, data changes in 14C production rates cannot
convincingly explain the D14C data (Supplementary Fig. 1,
Supplementary Note 1). All our tests therefore indicate that the
Tahiti D14C drop at 14.6 kyr BP is caused by carbon cycle
changes. This is our working hypothesis on which all else is based
on, but note that its failure cannot entirely be ruled out.
Carbon cycle changes responsible for the D14C anomaly would
also leave their imprints on atmospheric CO2. We can therefore
use the absolute U/Th-dated D14C from the Tahiti corals as an
independent time constraint on the atmospheric CO2 rise. This is
a novel new approach to synchronize atmospheric D14C and
atmospheric CO2, because ice cores archive only a smoothed
version of the atmospheric concentrations making an exact dating
of the abrupt change in atmospheric CO2 very difficult6.
Furthermore, firnification and gas enclosure are still not
completely understood19, and the age difference between ice
matrix and embedded gases complicates gas chronologies3. On
the most recent chronology4, AICC2012, CO2 measured in situ in
the EDC ice core rises by 10 p.p.m.v. between 14.81 and 14.68 kyr
BP. This is more than a century faster when compared with
previous chronologies3 (Figs 3a and 4b), but might in detail be
revised even further once the most recent understanding of
firnification is applied20. Atmospheric changes in CO2 need to
have happened even more abruptly than what is recorded in ice
cores6.
Here we use the Tahiti D14C as an independent age constraint
for the start of the carbon cycle changes (14.6 kyr BP). Recently,
others21 have shown that the rise in atmospheric CH4 and in
temperature in Greenland are near synchronous (5±25 years) at
the onset of the B/A warming. From previous GHG records
measured at the EDC ice core22, it is known that CO2 and CH4
also rise synchronously at 14.6 kyr BP. Combining this
information, we have to conclude that the rise in atmospheric
CO2 and CH4 together with the rapid warming of the northern
hemisphere (NH) happened at the same time, and started at
14.6 kyr BP. This is only 35 years later than the suggested age of
the onset of the B/A (14.635±0.186 kyr BP, ±1s) in the annuallayer counted NGRIP ice core23,24, well within the dating
uncertainty of GICC05 (Fig. 5).
Carbon cycle simulations. A release of 125 Pg of C into the
atmosphere within a time window of 50 to 200 years was
proposed before6 to explain the rise of 10 p.p.m.v. in CO2
measured in EDC. The true atmospheric CO2 then shows an
overshoot whose peak amplitude mainly depends on the length of
the assumed release time (Supplementary Fig. 2). Furthermore,
low-resolution CO2 time series from other ice cores
with amplitudes of 17, 15 and 19 p.p.m.v. in Byrd, Taylor and
Siple Dome, respectively25–28 (Fig. 3a), indicate that the true
atmospheric signal had a larger amplitude than what was measured
in situ in EDC6.
In our previous analysis, we also investigated atmospheric
d13CO2, from which the potential source of the released carbon
might have been identified. The new compilation of ice core
d13CO2 data published in the mean time1 offers a new look on the
information contained in that record. This d13CO2 compilation is
now based on data from EDC, EPICA DML and Talos Dome. In
our time window of interest (15–14 kyr BP), however, only data
from EDC were obtained with some new data points adding to
the previous record. This revised d13CO2 record shows a drop by
0.1% around 14.7 kyr BP followed by a subsequent rise by
B0.2% at 14.4 kyr BP (Fig. 3b). Distinguishing terrestrial from
marine carbon sources for our carbon release leads in our bestguess scenarios (see below how that was chosen) to either a drop
in d13CO2 of 0.4% or less than 0.1% in the true atmospheric
signal, respectively, but only to ! 0.15% and less than ! 0.05%
in a time series that would be recorded in EDC (Fig. 3b). Both
marine- and terrestrial-based d13CO2 simulations fall within the
uncertainties of the measurements before 14.6 kyr BP. The small
rise in d13CO2 after 14.4 kyr BP, however, indicates that directly
after the onset of the B/A other processes released less d13Cdepleted carbon to the atmosphere, for example, CO2 outgassing
from warm oceanic surface waters. On the basis of the data
uncertainty of d13CO2 in EDC, it is still impossible to clearly
identify if the released carbon was of marine or terrestrial origin.
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
3
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
(Supplementary Note 2). We assume a depletion in 14C of the
released carbon with respect to the atmosphere (D(D14C))
15.6
15.4
15.2
15.0
14.8
14.6 between ! 50 and ! 1,250%. This range in the 14C anomalies
260
covers carbon sources from the mean terrestrial biosphere
EDC at AICC2012
potentially released by shelf flooding6 ( ! 50%), suggested
EDC at Parrenin2013
signatures of old carbon of Pacific intermediate waters as
255
Taylor Dome at revised age model
measured off Baja California29 ( ! 400%) and Galapagos30
Siple Dome at own age model
Byrd at GICC05
( ! 700%) to a maximum effect of 14C-free carbon
250
( ! 1,250%). The Shelf Flooding Hypothesis is explained in
detail in the next section, and some more details on our
245
assumptions on 14C are found in the methods.
The highest-simulated anomalies in atmospheric D14C are
obtained
for short release times reaching ! 100% for 50 years
240
and the largest D(D14C) of ! 1,250% (Fig. 4c). The amplitudes
drastically decline with longer release time towards less than
235
! 35%. D14C anomalies are significantly smaller if D(D14C) was
! 700% or less (Fig. 4c). Because of the distinct dynamics of the
D14C data, release times shorter than B110 years are at odds with
230
the Tahiti 14C reconstruction. Simulated anomalies in atmospheric carbon records are nearly identical for Atlantic merAtm. CO2 according to ∆14C
225
idional overturning circulation (AMOC) in the strong or weak
Latest rise in CO2
CO2 as seen at EDC
mode (Methods, Fig. 4c, Supplementary Fig. 3). Combining the
due to ∆14C
At EDC shifted by 50 years
information from both the ice core data and our analysis of the
220
14.6
15.0
14.8
14.2
14.0 Tahiti D14C data leads to a range of scenarios with carbon release
14.4
times between B110 and 200 years in which model results and
Time (kyr BP)
data agree (Fig. 4c). The range of possible scenarios fulfilling the
data constraints also includes some with D(D14C) between ! 700
and ! 1,250%, and so we cannot entirely exclude the possibility
–6.4
that the released carbon still contains some 14C. From these
EDC (Lourantou 2010) at AICC2012
EDC (Schmitt 2012) at AICC2012
possible scenarios, we selected the one with the longest release
–6.5
time of 200 years to be our best-guess scenario, because short
release times lead to higher amplitudes in atmospheric CO2,
–6.6
which are not supported by other ice core data. This scenario
pinpoints to a D(D14C) of ! 1,250% resulting in peak amplitudes
–6.7
of ! 42% in atmospheric D14C (Fig. 4c) and of þ 22 p.p.m.v. in
atmospheric CO2 (Fig. 4d). The depletion in 14C necessary for the
–6.8
model output to agree with the data implies that the shelf flooding
hypothesis connected with meltwater pulse 1A (MWP-1A)6,31
–6.9
seems at a first glance to be in disagreement with the Tahiti-based
–7.0
atmospheric D14C reconstructions (Fig. 4c). We discuss details on
a potential contribution connected with MWP-1A further below.
–7.1
The release of deep ocean carbon, although so far not suggested to
play a role during this rapid CO2 rise, might only potentially be
–7.2
responsible here, if water masses are detected, which are even
more depleted in 14C than what is known until now29,30.
Marine source:
Terrestrial source:
–7.3
The Tahiti D14C data show an excursion from the long-term
Atmospheric signal
Atmospheric signal
declining trend of IntCal13 (ref. 9) (Figs 1 and 4a). Depending on
At EDC
At EDC
–7.4
the time window of interest, IntCal13 might be approximated by
14.8
14.6
15.0
14.4
14.2
14.0
a linear fit with a slope of ! 0.04% per year (the whole Mystery
Interval,
19–14 kyr BP) or ! 0.10% per year (15.0–14.3 kyr BP),
Time (kyr BP)
respectively. These long-term changes are probably caused by a
Figure 3 | Ice core and simulated true atmospheric CO2 and d13CO2.
mixture of changes in 14C production rate and the carbon cycle32.
(a) Ice core CO2 data (±1s) from EDC1,22,68 on two different chronologies3,4 While we are able to force our model with changing 14C
AICC2012 and Parrenin2013, Taylor Dome on revised age model26,27, Siple
production rates (Supplementary Fig. 4), all relevant processes in
Dome on own age model (top x axis)27, Byrd on age model GICC05
the carbon cycle are not yet identified. We therefore compare the
(refs 25,28). Simulated true atmospheric CO2 in our best-guess scenario
D14C data with our original simulation results based on constant
14C production rate, but also with some results that are corrected
according to 14C data (black bold line), filtered to a signal that might be
recorded in EDC (blue dashed line), shifted by 50 years to meet the EDC data
for the trend seen in IntCal13. If corrected accordingly, our best(dashed red line). (b) Ice core d13CO2 data (±1s) from EDC1,68, simulated
guess scenario finally meets the amplitude in the Tahiti D14C data
true atmospheric d13CO2 of our best-guess scenario and how it would have
(Fig. 4a,c).
!13CO2 (‰)
CO2 (p.p.m.v.)
SD age (kyr BP)
been recorded in EDC for either terrestrial or marine origin of the released
carbon, implying a d13C signature of ! 22.5 or ! 8.5%, respectively.
Using the same carbon cycle model6, we repeat simulations
of carbon release with special focus on 14C. The model
dynamic with respect to 14C was extensively evaluated
4
Evidence for permafrost thawing. The synchronicity of the NH
warming and the carbon cycle change together with our suggested
hypothesis for the injection of nearly 14C-free carbon into the
atmosphere make permafrost thawing and a subsequent release of
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
0
260
–10
–50‰
–0.04‰ per year
–0.10‰ per year
–30
–400‰
240
–40
–0.04‰ per year
220
–50
No trend
–700‰
–60
200
–70
Agrees with 14C data
180
160
260
255
Sim
Sim-(0.04‰ per year)
a
Sim-(0.10‰ per year)
Simulated atm.CO2
according to ∆14C
EDC data:
AICC2012 age
Parrenin2013 age
c
–1,250‰
Agrees with CO2 data
–80
–90
–100
50
At EDC
40
250
CO2 (p.p.m.v.)
–20
245
30
240
20
235
230
Atm. ∆ (CO2) (p.p.m.v.)
Atm. ∆14C (‰)
–0
ar
er ye
p
.10‰
Atm. ∆(∆ 14C) (‰)
Tahiti
IntCal13
280
10
225
220
15.0
b
14.8
14.6
14.4
14.2
d
14.0 0
Time (kyr BP)
50
100
150
200
250
300
0
350
Length of C release (year)
Figure 4 | Main carbon cycle simulation results. The transient simulation results (left) showing the impact of a carbon release event on true atmospheric
D14C and CO2 obtained with the carbon cycle model BICYCLE for the best-guess scenario are compared with the data. In sensitivity studies (right), the
length of the release event and the radiocarbon signature D(D14C) of the released carbon are constrained by the data. (a) Atmospheric D14C data from
Tahiti corals7 (magenta, mean ±1s in both age and D14C) and IntCal13 (ref. 9) (grey band, mean ±1s) data. Black bold circles denote start and stop (±1s)
of carbon release in the non-linear model of the Tahiti data interpretation. The vertical black dashed line marks the estimated started of carbon release at
14.6 kyr BP based on a combination of different explanations. Best-guess simulation results of atmospheric D14C (blue) superimposed by a linear trend of
either ! 0.04% per year (long dashed line) or ! 0.10% per year (solid line) (short dashed: no trend superimposed). (b) Atmospheric CO2. EDC ice core
CO2 data (mean ±1s) on two different chronologies3,4 AICC2012 and Parrenin2013. Simulated true atmospheric CO2 rise (black bold line), and how the
signal might be recorded in EDC (dashed red line) after filtering for gas enclosure and shifted by 50 years to meet the data. (c) Simulated peak height in
atmospheric D14C (grey areas) as function of length of carbon release and of the D14C depletion. (d) Simulated peak height in atmospheric CO2 (dark blue
area) as function of length of carbon release. In c,d, simulations result with the AMOC in either a weak or a strong mode are combined spanning a range of
results. Magenta square and circle in c,d mark results of our best-guess scenario for D14C and CO2, respectively. We colour coded the areas in the
parameter space where simulation results agree with the EDC CO2 data (d, light blue) and with the interpretation of the Tahiti D14C data (c, black boxes).
The latter are modified for background linear trends already contained in IntCal13 based on other processes.
old soil carbon a prominent candidate to explain the atmospheric
carbon records. The age of carbon stored in permafrost soils
during glacial times is unknown. Throughout the last glacial cycle
Greenland and the whole NH was perturbed by the rapid
warming of Dansgaard/Oeschger (D/O) events33. However,
during the last 80 kyr, only D/O event 12 around 47 kyr BP
reached in a temperature reconstruction for the site of the NGRIP
ice core in Greenland similar high temperatures as the B/A
(Fig. 6b). In this NGRIP, temperature time series D/O event 2
around 23 kyr BP was rather weak and short, but D/O event 3 at
28 kyr BP reached with ! 36 !C nearly the temperature of
! 33 !C of the B/A33 (Fig. 6b). We assume that most of the NH
follow this temporal changes in temperature observed for
Greenland, although with warmer temperatures closer to the
freezing point further south. It might then be that large areas of
the NH were permanently frozen after D/O event 3, thus about
13 kyr before thawing induced by the onset of the B/A. The D14C
of that permafrost carbon would be depleted by ! 900% with
respect to atmospheric D14C during release around 14.6 kyr BP
(Fig. 6a). However, soil carbon might age significantly in high
latitudes before freezing, for example, present day North
American peatlands are up to 17-kyr old34. Such soil ageing
reduces 14C even further. If the precursor material of the
permafrost soil carbon was photosynthetically produced during
D/O event 12 around 47 kyr BP (the next preceding period
comparable in temperature to the B/A, Fig. 6b), it would be
essentially free of 14C and depleted with respect to atmospheric
D14C by nearly ! 1,250% (Fig. 6a). Permafrost thawing would
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
5
600
220
650
500
450
–34
500
Atm. ∆14C (‰)
700
Greenland CH4(p.p.b.v.)
230
Ages aligned 550
–2
–37
–34
–38
–36
–39
–38
145 years
–40
–42
–3
14.6 14.4 14.2
Time (kyr BP)
–1,250‰
0
–4
Ages aligned
14.8
250‰
200
–1
–36
15.0
300
100
0
–35
400
∆TANT (K)
WD !18O (‰)
Age of C released at B/A (kyr)
30
20
10
0
240
600
–44
14.0
Figure 5 | Implication on the timing of abrupt climate change as obtained
in various ice core records from Greenland and Antarctica. Our results
suggest that anomalies in Tahiti D14C and true atmospheric CO2 are caused
by the same process. This information is used here as an independent age
constraint. The onset of the abrupt rise in atmospheric CO2 (black bold, this
study) is thus tied to 14.6 kyr BP. From previous ice core analysis22, it is
known that the rise in CO2 and CH4 (red circles, Greenland composite69)
occur synchronously here. A new study21 on the NEEM ice core tied the
CH4 rise to be near synchronous to Greenland temperature rise. This
synchronicity of the start of the abrupt changes in atmospheric CO2, CH4
and Greenland temperature tied to 14.6 kyr BP led to the age alignments in
CH4 and NGRIP d18O (high24 (black thin line) and low70 (red line)
resolution). We furthermore show some Antarctic climate records on their
own independent chronologies to illustrate the temporal north–south
offsets. WD54 d18O, original (blue) and 100 years running mean (yellow)
and stack3 of temperature change from five ice cores in East Antarctica,
DTANT, original (black) and 100 years running mean (orange).
then contain a depletion in D14C, which is more negative than for
all other suggested processes6,29,30. An alternative scenario based
on the destabilization of gas hydrates, which also contain 14C-free
carbon, can be rejected based on CH4 isotopes35–37.
For the present day, a rise in global mean temperature
by 5 K, which because of polar amplification might represent a
northern high latitude warming of 10 K, was proposed to lead to
the release of more than 130 Pg of soil carbon from permafrost
thawing within 200 years38. Greenland ice core data33 and
simulations39 suggest that temperatures in the B/A rose by 10–
15 K to near preindustrial levels in central Greenland and
throughout most of the NH land areas. A large inert terrestrial
carbon pool consisting of permafrost soils containing 700 Pg
more C at the Last Glacial Maximum (LGM) than at present day
has been proposed40, which needs to release its excess carbon
during deglaciation. The areal extent of continuous permafrost at
LGM (Fig. 7) was calculated from models41 in PMIP3 to
6
40
60
–26
–28 25
–30
–32
–34
–36
–38
–40
–42
–44
–46
–48
–50
–52
–54
120
20
50
40
30
20
Time (kyr BP)
12
8
3 2
10
0
–26
–28
–30
–32
–34
–36
–38
–40
–42
–44
–46
–48
–50
–52
–54
1
B/A
32 kyr
13 kyr
80
50
40
30
20
0
–100
–200
–300
–400
–500
–600
–700
–800
–900
–1,000
–1,100
–1,200
–1,300
10
∆ (∆14C) (‰)
250
14.6 kyr BP
NGRIP temperature (°C)
260
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
NGRIP !18O (‰)
Atm. CO2 (p.p.m.v.)
ARTICLE
0
Time (kyr BP)
Figure 6 | Radiocarbon depletion of soil carbon of different age and high
northern latitude climate change. (a) Atmospheric D14C based on IntCal13
(ref. 9) over the last 50 kyr (blue, left y axis, mean ±1s). Calculated
radiocarbon depletion resulting in D(D14C) (mean ±1s) of soil carbon
released during the B/A as a function of its age (magenta, right y axis,
upper x axis) and of atmospheric D14C during time of production. (b) NGRIP
temperature reconstruction33 from 120 to 10 kyr BP. The time series is
plotted in two different colours because of the break in the x axis scale at
50 kyr BP. Numbers label selected D/O events. Red labelled arrows
highlight the time which past since NGRIP was similar as warm as during
the B/A (32 kyr since D/O event 12) and since the previous significant
warming before the B/A (13 kyr since D/O 3).
26 $ 1012 m ! 2, agrees with reconstructions42, and is twice as
large as for preindustrial times41.
Previously, methane isotopes36 suggested that a rise in boreal
wetland CH4 emissions by þ 32 Tg CH4 per year would explain
the CH4 rise into the B/A. These findings36 have been challenged
by new methane isotope data37, but so far no revised CH4
emissions from boreal wetlands have been calculated for the B/A.
An alternative interpretation5 of the CH4 cycle based on its
interhemispheric gradient suggests that the rise in CH4 by
150 p.p.b.v. at the onset of the B/A was largely driven by the
increase in CH4 emissions from both tropical ( þ 35 Tg CH4 per
year) and boreal ( þ 15 Tg CH4 per year) wetlands. The CH4
change at the onset of the B/A is thus clearly dominated by
tropical wetlands and its conclusive interpretation is beyond the
scope of this study. However, the rise in CH4 emissions from
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
Figure 7 | PMIP3 simulation results on the LGM permafrost extend.
Results41 show a polar projection of the NH from 20 !N northwards, are
based on soil temperature and distinguish land with ice (dark blue),
permafrost (blue), seasonal frozen (light blue) and not frozen (red). Present
day coastlines are sketched in thin black lines. Magenta points mark
potential core sites (Siberian Shelf, Black Sea, Caspian Sea, Sea of Okhotsk)
from which future 14C measurements on terrigenous material might verify
the age of permafrost possible thawed around 14.6 kyr BP (suggested green
areas).
boreal wetlands is nearly identical to the rise in emissions of up to
þ 14 Tg CH4 per year projected from deep permafrost thawing of
the next century43. If this rise in the boreal CH4 flux is integrated
over the 200-year time window of our carbon release scenario, a
total of 3.0 Pg of CH4 (or 2.25 Pg C in the form of CH4) might
have been released. This is B2% of our total estimated carbon
emissions of 125 Pg C, and in line with an expert assessment on
the future vulnerability of permafrost44 estimating that 2–3% of
carbon released by thawing might enter to the atmosphere in the
form of CH4. Although the contribution from boreal wetlands to
the CH4 rises at the onset of the B/A is small, the nearly 14C-free
signature connected with our proposed permafrost thawing might
be tested by 14C measurements on CH4 derived from ice cores45.
So far, we suggested that NH permafrost is the responsible
source of the released carbon. In the following, we hypothesize
which region might have been affected in detail by permafrost
thawing and how this can be tested in future studies. The PMIP3based map on the LGM permafrost extent clearly indicates that
the largest areas with continuous permafrost are found in
northern Siberia (Fig. 7). Thus, evidences of permafrost thawing
connected to the NH warming should be expected in outflow
originated from the southern edge of the LGM permafrost area
(around 40–50!N), which thawed first. A lot of these areas are
drained via the Amur river into the Sea of Okhotsk and into
coastal seas towards the south (Caspian and Black Sea). Indeed, a
combination of terrestrial biomarkers that clearly indicate the
thawing of permafrost was found at the onset of the B/A in a
sediment core drilled in the Black Sea that records the drainage
from the Fennoscandian Ice Sheet46. Here variations in the
normalized concentrations of different long-chain molecules
provide information on changes in the abundance of peatforming plants. Such data are useful indicators for the variation of
permafrost thawing and of wetland extension as well as for fluvial
periglacial soil erosion in its drainage basin. In details, this
study46 provides evidence that the permafrost melting was very
intense only during the initial part of the Bølling corresponding
indeed to the sharp NH warming.
The map (Fig. 7) also shows that a large fraction of the Siberian
shelf in the Arctic Ocean was during the LGM covered by
permafrost. It might thus be possible that MWP-1A, which was
recorded as a rise in sea level at Tahiti31 from about ! 105 to
! 85 m between 14.65 and 14.31 kyr BP might be partially
responsible for the carbon release as initially suggested6. The
flooding of continental shelves was also proposed to contribute
to the CH4 rise during deglaciation and during D/O events47.
In our earlier study6, we proposed that mainly the flooding of the
Sunda Shelf followed by tropical rain forest decay might have
been responsible for the carbon release. This shelf flooding
scenario was here addressed with a D(D14C) of ! 50% for mean
terrestrial carbon, which failed to meet the D14C data. In our
earlier study6, we also discussed that the existing time series of sea
level change suggest that before MWP-1A the shelves were last
flooded around 30 kyr BP, 15 kyr earlier, leaving ample time for
14C in permafrost carbon on the shelf to decay and to produce a
D(D14C) in the released carbon of down to ! 900% (Fig. 6).
Most recent sediment data48 on iceberg discharge in Antarctica
during Termination I found a significant Antarctic contribution
to MWP-1A. Fingerprint analysis49 of different water sources
for MWP-1A indicate that sea level would rise locally by up to
50% above global average on the Siberian Shelf for freshwater
released in Antarctica. When considering the source-depending
overprint49, we calculate, based on the present day bathymetry50,
a maximum areal extent of 0.4 $ 1012 m ! 2 of the Siberian Shelf,
which might have been flooded by MWP-1A. This is the same
order of magnitude as the present day Siberian Yedoma deposit
extent51 from which an organic carbon content of 30–140 Pg C
has been proposed51. Coastal erosion and sub-sea permafrost
release in Arctic Siberia are also observed for modern times52
with a D14C signature of the released organic carbon as low as
! 800%. Modern organic carbon content in Eurasian Arctic53
river runoff have D14C ages of up to 10 kyr. All these modern data
indicate that old carbon in permafrost exists nowadays, and
potentially was more abundant and older during glacial times.
In which region the thawing of permafrost finally happened
might be verified by future 14C measurements on terrigenous
organic material that are retrieved from marine sediments in the
suggested coastal seas. It will then be possible to finally
attribute the size of the released carbon to either a pure
thermodynamically thawing at the southern edge of the
permafrost area or to a contribution from flooding the Siberian
Shelf during MWP-1A.
Discussion
The rapid CO2 rise at the onset of the B/A is contained with
different amplitude in various ice cores (Fig. 3a). However, the
uncertainty in the proposed age distribution of the CO2 in EDC is
still large6 (Supplementary Fig. 2c) and the assumed carbon
release history and the applied carbon cycle model influence the
amplitude of the proposed true atmospheric CO2 rise. Future CO2
measurement from the WD ice core54 might refine some of these
aspects. The WD ice core has an order of magnitude higher
present day accumulation rate than EDC (20 versusB3 g cm ! 2
per year), thus offers temporally higher resolved gas records.
A potential WD CO2 record still needs to be corrected for the
smoothing during firn enclosure, although this effect will be a lot
smaller than for EDC. Only by considering the 20% uncertainty
in the mean exchange time of CO2 before enclosure in EDC
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
7
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
(Supplementary Fig. 2c) would result in a CO2 release and an
amplitude in the true atmospheric CO2 rise, which are also 20%
smaller than in our best-guess scenario, for example, releasing
100 instead of 125 Pg C leading to a true atmospheric CO2
amplitude of 18 instead of 22 p.p.m.v. (Supplementary Fig. 2d).
Results are then still within the range given by the EDC CO2 data
(Supplementary Fig. 2d). Furthermore, we calculate that a 33%
reduction in the proposed carbon released to the atmosphere
(85 instead of 125 Pg C) still fulfils the data constrains given by
the Tahiti D14C record.
Other rapid CO2 rises are detected in EDC at the end of the
Younger Dryas and in Marine Isotope Stage (MIS) 3 in other ice
cores55. Whether they are also caused by permafrost thawing is
not investigated here. However, these other rapid CO2 jumps are
not always connected to a warming of the NH. Furthermore, the
missing of 14C-depleted CH4 during the CO2 rise at the end of the
Younger Dryas around 12 kyr BP45 suggests that other processes
are responsible. Moreover, our suggested old soil carbon release
during permafrost thawing at 14.6 kyr BP, requires that the same
carbon sources were not tapped during other events within
Termination I. Again, 14C measurements on terrigenous material
might clarify how old the carbon released from permafrost was or
if earlier CO2 rises might already have consumed the old, 14Cdepleted carbon.
Our study suggests that for Termination I, abrupt warming in
the NH might lead to massive permafrost thawing, activating a
long-term immobile carbon reservoir. The abruptly released
carbon then amplified the initial warming as a positive feedback.
Our best-guess scenario generates together with the rise in the
other two important GHG CH4 and N2O a radiative forcing6 of
B0.7 W m ! 2. It is important to quantify the feedback of this
GHG forcing on climate to better understand the impact of
the abrupt GHG changes during the last deglaciation. Since the
abrupt GHG changes are contemporaneous with the onset of the
B/A (in the North) and the beginning of the ACR (in the South),
the sequence of the associated bipolar climate linkages are of
particular interest.
Therefore, we have conducted transient simulations with our
best-guess reconstruction of atmospheric GHG changes during
the beginning of the B/A and the ACR, using the Earth System
Model COSMOS16,56 in a coupled atmosphere-ocean
configuration as outlined in detail in the Supplementary Note 3.
To evaluate the global impacts of the GHG changes it is
instructive to analyse Antarctic temperature changes, since
temperature changes obtained from Antarctic ice cores have
been shown to reflect global scale climate changes associated with
CO2 variations particularly well57. On the basis of our climate
model investigations, we find that the abrupt rise in GHG
concentrations provides an important impact on the Antarctic
temperature signature associated with an abrupt AMOC
strengthening at the end of Heinrich Stadial 1 (Supplementary
Fig. 5). This highlights a potential contribution of abrupt GHG
changes on the bipolar climate signature during deglaciation. In
this sense, the abrupt GHG changes would be a factor that would
offset the timing of the temperature maximum leading into the
ACR, compared to the onset of the B/A. As layed out in detail in
the climate feedback section of the Supplementary Note 3, also
smaller GHG spikes bear the potential to have a substantial effect
on the Antarctic temperature response when compared with
impacts caused by AMOC changes.
So far a synthetic record of Greenland temperature changes8
seems to indicate that rapid climate changes in the north might
indeed have been a universal feature of deglaciations during the
last 800 kyr. Hence, similar to the last deglaciation, abrupt
permafrost thawing might have also occurred regularly during
earlier terminations, although further studies are necessary here.
8
Termination II also contains58 an abrupt rise in CO2,
synchronous to a rise in CH4. A massive drop in atmospheric
d13CO2 accompanying this event58 is consistent with the release
of d13C-depleted CO2 that might indicate a terrestrial source.
However, new d13CO2 data59 did not confirm this negative
d13CO2 anomaly and the revised data give no indication on the
source of this CO2 rise. A synchronous change in deuterium
excess60, a proxy for moisture source shifts, has been used to
suggest that abrupt shifts in southern westerlies might be
connected with the CO2 rise61, but a compelling explanation
remains elusive and further testing of permafrost thawing as a
possible alternative interpretation is needed.
In conclusion, we here suggest that the processes responsible
for the abrupt CO2 rise at the onset of the B/A is also the
underlying cause for the drop seen in atmospheric D14C based on
Tahiti corals. This connection offers a U/Th-dated tie point for
the start of the massive release of carbon at 14.6 kyr BP. Using a
carbon cycle box model, and assuming the release of 125 Pg of
nearly 14C-free carbon, we are able to explain observed anomalies
in atmospheric CO2 and D14C. On the basis of the 14C signature
of the released carbon and the synchronicity to the warming of
the NH, we suggest that the thawing of permafrost was this
responsible process. A potential contribution from MWP-1A
flooding the Siberian Shelf, which might have contained a large
amount of permafrost, is also possible. Future 14C measurements
on terrigenous material might further constrain the source region.
Our interpretation not only provides conceptual insights into the
source of the excursions in the atmospheric carbon records
around 14.6 kyr BP, but also offers an alternative to explanations62,63 for the interhemispheric timing of the B/A and the ACR
as found in ice cores from both hemispheres. Taken together, our
findings highlight a potential climate feedback that might be
obtained from abrupt CO2 release during deglaciation. This
analysis furthermore indicates that the proposed carbon cycle
feedback from an anthropogenic driven permafrost thawing in
the near future38,43,44,64 may already have happened in a similar
way in the past.
Methods
Analysis of the D14C data. For analysis of the drop in the atmospheric D14C data
based on Tahiti corals, we used two different approaches (Fig. 2). First, we used a
linear statistical model Breakfit65, which calculates the break points in time series.
Breakfit searches for two linear functions that are joined at the break point. To
determine the break points, the model is fitted to the data applying an ordinary
least-squares method with a brute-force search for the break points. A measure of
the uncertainty of the break points is based on 2,000-block bootstrap simulations,
applying a moving block bootstrap algorithm with a block length of 1. We were
searching for two break points in the time intervals between 16 and 13 kyr BP. The
two subintervals (one for each break point) were ranging from the outer boundary
next to the break point of interest to the other break point. Subintervals were finally
identified after at least two iterative applications to (in kyr BP): break point 1
[15.74, 14.45] and break point 2 [14.67, 13.16]. Breakfit identified the start in the
D14C drop at 14.66±0.07 kyr BP followed by its decline by 54±8% within
207±95 years. Because of the very distinct dynamics of atmospheric D14C
including a rebound after its minimum (that is, after the carbon release to the
atmosphere stopped), we also analysed the data more subjectively with a non-linear
approach. Here we only calculated the mean time and mean D14C right at the start
of the carbon cycle changes around 14.6 kyr BP (two data points) and at its
minimum (eight data points) assuming that D14C followed a non-linear pathway
between both and included a rebound thereafter. The D14C data then starts to
decline at 14.59±0.04 kyr BP and stop after 258±53 years with a maximum drop
of 58±14% followed by a rebound of atmospheric D14C. This non-linear dynamic
is seen in the Tahiti data but also in our carbon cycle simulations (Figs 2 and 4a).
Combining the linear and non-linear approach brings high confidence that the
D14C drop started at around 14.6 kyr BP. All uncertainties are given as 1s.
Possible D14C signature of permafrost carbon. The maximum possible D(D14C)
of carbon released from permafrost thawing is a function of age and of atmospheric
D14C during time of production. From the D14C signature (IntCal13) (ref. 9)
of the precursor material (atmospheric CO2), which varies before the B/A roughly
between 250 and 550% (Fig. 6a), we first subtract the mean D14C value of
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
terrestrial carbon at the LGM in the model ( ! 50%) before a further reduction
in D14C signature is realised by the radioactive decay of 14C (half-life time
of 5,730 years).
Carbon cycle model. We use the carbon cycle box model BICYCLE in transient
mode to simulate changes in atmospheric CO2 and D14C. The model setup is
identical to an earlier study, which already proposed the magnitude of the CO2
overshoot during the B/A6.
We simulate the release of 125 Pg of carbon into the atmosphere with a constant
rate that varies inversely with the time length of the event between 0.42 Pg C per
year (300 years) and 2.5 Pg C per year (50 years) and configured the AMOC in
either its strong or its weak mode. Both AMOC configurations differ in the strength
of the overturning cell in the Atlantic with 16 Sv deep water production in the
North Atlantic in the strong mode and 2 Sv in the weak mode. We repeated our
previous comparison6 of simulated atmospheric d13CO2 to ice core data from the
EDC because new d13CO2 data were published in the mean time1. For this modeldata comparison of d13CO2, we distinguished terrestrial and marine sources of the
released carbon by assuming a d13C signature of ! 22.5 and ! 8.5%, respectively
(Fig. 3b). More details on these assumptions are found in our previous article6.
D(D14C) of the simulated carbon release is depleted with respect to the atmosphere
between ! 50 and ! 1,250%. Initial conditions of 14C production rates influence
simulated D14C over several ten thousand years32. All simulations therefore start at
60 kyr BP. In our standard case, 14C production rates are assumed to be constant
and 15% higher than present day, leading to atmospheric D14C of þ 250% at
14.6 kyr BP in agreement with IntCal13 (ref. 9). Long-term trends in 14C
production rate as suggested by the geomagnetic field data32 only slightly impact
our simulations (Supplementary Fig. 4).
For model evaluation, BICYCLE is (a) compared in its oceanic carbon uptake
dynamic resulting in a model-specific airborne fraction with other models, (b) used
to simulate the Suess effect (years 1820–1950 AD), (c) the bomb 14C peak (years
1950–2000 AD) and (d) applied on CO2 release experiments for preindustrial
background conditions. The model is compared with the results from another
carbon cycle box model66 (Suess effect and for preindustrial conditions) and with
output from the GENIE model67, an Earth system model of intermediate
complexity (preindustrial conditions). All details on this model evaluation are
found in the Supplementary Note 2 including Supplementary Figs 6–8.
Filtering true atmospheric CO2 into signals recorded in EDC. The smoothing
effect of the gas enclosure process in ice cores that transforms a potential true
atmospheric CO2 into a time series comparable to EDC ice core data is performed
with a log-normal probability density function with an assumed mean value or
width E of 400±80 years (mean ±1s)6 (Supplementary Fig. 2c):
f ðxÞ ¼
lnðxÞ ! m 2
1
pffiffiffiffiffi ' e ! 0:5ð s Þ
x ' s ' 2p
ð1Þ
with x (in years) as the time elapsed since the last exchange with the atmosphere.
From the two free parameters, m and s of the equation, we chose for simplicity
s ¼ 1, which leads to E ¼ em ! 0.5. The application of such a filter function for
the transformation of true atmospheric signals into those that might be recorded
in ice cores during rapid climate change was compared with results from firn
densification models and extensively validated with CH4 data from both
hemisphere6.
References
1. Schmitt, J. et al. Carbon isotope constraints on the deglacial CO2 rise from ice
cores. Science 336, 711–714 (2012).
2. Skinner, L. C., Fallon, S., Waelbroeck, C., Michel, E. & Barker, S. Ventilation of
the deep Southern Ocean and deglacial CO2 Rise. Science 328, 1147–1151
(2010).
3. Parrenin, F. et al. Synchronous change in atmospheric CO2 and Antarctic
temperature during the last deglacial warming. Science 339, 1060–1063 (2013).
4. Veres, D. et al. The Antarctic ice core chronology (AICC2012): an optimized
multi-parameter and multi-site dating approach for the last 120 thousand
years. Clim. Past 9, 1733–1748 (2013).
5. Baumgartner, M. et al. High-resolution interpolar difference of atmospheric
methane around the Last Glacial Maximum. Biogeosciences 9, 3961–3977 (2012).
6. Köhler, P., Knorr, G., Buiron, D., Lourantou, A. & Chappellaz, J. Abrupt rise in
atmospheric CO2 at the onset of the Bølling/Allerød: in-situ ice core data versus
true atmospheric signals. Clim. Past 7, 473–486 (2011).
7. Durand, N. et al. Comparison of 14C and U-Th ages in corals from IODP #310
cores offshore Tahiti. Radiocarbon 55, 1947–1974 (2013).
8. Barker, S. et al. 800,000 Years of abrupt climate variability. Science 334,
347–351 (2011).
9. Reimer, P. J. et al. IntCal13 and Marine13 radiocarbon age calibration curves 050,000 years cal BP. Radiocarbon 55, 1869–1887 (2013).
10. Hoffmann, D. L. et al. Towards radiocarbon calibration beyond 28 ka using
speleothems from the Bahamas. Earth Planet. Sci. Lett. 289, 1–10 (2010).
11. Southon, J., Noronha, A. L., Cheng, H., Edwards, R. L. & Wang, Y.
A high-resolution record of atmospheric 14C based on Hulu Cave speleothem
H82. Quaternary Sci. Rev. 33, 32–41 (2012).
12. Hughen, K., Southon, J., Lehman, S., Bertrand, C. & Turnbull, J. Marinederived 14C calibration and activity record for the past 50,000 years updated
from the Cariaco Basin. Quaternary Sci. Rev. 25, 3216–3227 (2006).
13. Noronha, A. L. et al. Assessing influences on speleothem dead carbon
variability over the Holocene: implications for speleothem-based radiocarbon
calibration. Earth Planet. Sci. Lett. 394, 20–29 (2014).
14. Bronk Ramsey, C. et al. A complete terrestrial radiocarbon record for 11.2 to
52.8 kyr B.P. Science 338, 370–374 (2012).
15. Bard, E. Correction of accelerator mass spectrometry 14C ages measured in
planktonic foraminifera: paleoceanographic implications. Paleoceanography 3,
635–645 (1988).
16. Gong, X., Knorr, G., Lohmann, G. & Zhang, X. Dependence of abrupt Atlantic
meridional ocean circulation changes on climate background states. Geophys.
Res. Lett. 40, 3698–3704 (2013).
17. Butzin, M., Prange, M. & Lohmann, G. Readjustment of glacial radiocarbon
chronologies by self-consistent three-dimensional ocean circulation modeling.
Earth Planet. Sci. Lett. 317-318, 177–184 (2012).
18. Singarayer, J. S. et al. An oceanic origin for the increase of atmospheric
radiocarbon during the Younger Dryas. Geophys. Res. Lett. 35, L14707 (2008).
19. Capron, E. et al. Glacial-interglacial dynamics of Antarctic firn columns:
comparison between simulations and ice core air d15N measurements. Clim.
Past 9, 983–999 (2013).
20. Freitag, J., Kipfstuhl, J., Laepple, T. & Wilhelms, F. Impurity-controlled
densification: a new model for stratified polar firn. J. Glaciol. 59, 1163–1169
(2013).
21. Rosen, J. L. et al. An ice core record of near-synchronous global climate changes
at the Bølling transition. Nat. Geosci. 7, 459–463 (2014).
22. Monnin, E. et al. Atmospheric CO2 concentrations over the last glacial
termination. Science 291, 112–114 (2001).
23. Rasmussen, S. O. et al. A new Greenland ice core chronology for the last glacial
termination. J. Geophys. Res. 111, D06102 (2006).
24. Steffensen, J. P. et al. High-resolution Greenland ice core data show abrupt
climate change happens in few years. Science 321, 680–684 (2008).
25. Neftel, A., Oeschger, H., Staffelbach, T. & Stauffer, B. CO2 record in the Byrd
ice core 50,000-5,000 years BP. Nature 331, 609–611 (1988).
26. Smith, H. J., Fischer, H., Wahlen, M., Mastroianni, D. & Deck, B. Dual modes
of the carbon cycle since the Last Glacial Maximum. Nature 400, 248–250
(1999).
27. Ahn, J. et al. A record of atmospheric CO2 during the last 40,000 years from the
Siple Dome, Antarctica ice core. J. Geophys. Res. 109, D13305 (2004).
28. Pedro, J. B., Rasmussen, S. O. & van Ommen, T. D. Tightened constraints on
the time-lag between Antarctic temperature and CO2 during the last
deglaciation. Clim. Past 8, 1213–1221 (2012).
29. Marchitto, T. M., Lehman, S. J., Oritz, J. D., Flückiger, J. & van Geen, A.
Marine radiocarbon evidence for the mechanism of deglacial atmospheric CO2
rise. Science 316, 1456–1459 (2007).
30. Stott, L., Southon, J., Timmermann, A. & Koutavas, A. Radiocarbon age
anomaly at intermediate water depth in the Pacific Ocean during the last
deglaciation. Paleoceanography 24, PA2223 (2009).
31. Deschamps, P. et al. Ice-sheet collapse and sea-level rise at the Bolling warming
14,600 years ago. Nature 483, 559–564 (2012).
32. Köhler, P., Muscheler, R. & Fischer, H. A model-based interpretation of low
frequency changes in the carbon cycle during the last 120 000 years and its
implications for the reconstruction of atmospheric D14C. Geochem. Geophys.
Geosyst. 7, Q11N06 (2006).
33. Kindler, P. et al. Temperature reconstruction from 10 to 120 kyr b2k from the
NGRIP ice core. Clim. Past 10, 887–902 (2014).
34. Gorham, E., Lehman, C., Dyke, A., Clymo, D. & Janssens, J. Long-term carbon
sequestration in North American peatlands. Quaternary Sci. Rev. 58, 77–82
(2012).
35. Sowers, T. Late Quaternary atmospheric CH4 isotope record suggests marine
clathrates are stable. Science 311, 838–840 (2006).
36. Fischer, H. et al. Changing boreal methane sources and constant biomass
burning during the last termination. Nature 452, 864–867 (2008).
37. Möller, L. et al. Independent variations of CH4 emissions and isotopic
composition over the past 160,000 years. Nat. Geosci. 6, 885–890 (2013).
38. Schaphoff, S. et al. Contribution of permafrost soils to the global carbon budget.
Environ. Res. Lett. 8, 014026 (2013).
39. Liu, Z. et al. Transient simulation of last deglaciation with a new mechanism for
Bølling-Allerød warming. Science 325, 310–314 (2009).
40. Ciais, P. et al. Large inert carbon pool in the terrestrial biosphere during the
Last Glacial Maximum. Nat. Geosci. 5, 74–79 (2012).
41. Saito, K. et al. LGM permafrost distribution: how well can the latest PMIP
multi-model ensembles reconstruct? Clim. Past 9, 1697–1714 (2013).
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
9
ARTICLE
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms6520
42. Vandenberghe, J. et al. The Last Permafrost Maximum (LPM) map of the
Northern Hemisphere: permafrost extent and mean annual air temperatures,
25-17 ka BP. Boreas 43, 652–666 (2014).
43. Koven, C. D. et al. Permafrost carbon-climate feedbacks accelerate global
warming. Proc. Natl Acad. Sci. USA 108, 14769–14774 (2011).
44. Schuur, E. et al. Expert assessment of vulnerability of permafrost carbon to
climate change. Clim. Change 119, 359–374 (2013).
45. Petrenko, V. V. et al. 14CH4 measurements in Greenland ice: investigating last
glacial termination CH4 sources. Science 324, 506–508 (2009).
46. Rostek, F. & Bard, E. Hydrological changes in eastern Europe during the last
40,000 years inferred from biomarkers in Black Sea sediments. Quaternary Res.
80, 502–509 (2013).
47. Ridgwell, A., Maslin, M. & Kaplan, J. O. Flooding of the continental
shelves as a contributor to deglacial CH4 rise. J. Quaternary Sci. 27, 800–806
(2012).
48. Weber, M. E. et al. Millennial-scale variability in Antarctic ice-sheet discharge
during the last deglaciation. Nature 510, 134–138 (2014).
49. Clark, P. U., Mitrovica, J. X., Milne, G. A. & Tamisiea, M. E. Sea-level
fingerprinting as a direct test for the source of global meltwater pulse IA.
Science 295, 2438–2441 (2002).
50. Smith, W. H. & Sandwell, D. T. Global sea floor topography from satellite
altimetry and ship depth soundings. Science 277, 1956–1962 (1997).
51. Strauss, J. et al. The deep permafrost carbon pool of the Yedoma region in
Siberia and Alaska. Geophys. Res. Lett. 40, 6165–6170 (2013).
52. Vonk, J. E. et al. Activation of old carbon by erosion of coastal and subsea
permafrost in Arctic Siberia. Nature 489, 137–140 (2012).
53. Feng, X. et al. Differential mobilization of terrestrial carbon pools in Eurasian
Arctic river basins. Proc. Natl Acad. Sci. USA 110, 14168–14173 (2013).
54. WAIS Divide Project Members. Onset of deglacial warming in West Antarctica
driven by local orbital forcing. Nature 500, 440–444 (2013).
55. Ahn, J. & Brook, E. J. Atmospheric CO2 and climate on millennial time scales
during the last glacial period. Science 322, 83–85 (2008).
56. Zhang, X., Lohmann, G., Knorr, G. & Xu, X. Different ocean states and
transient characteristic in Last Glacial Maximum simulations and implications
for deglaciation. Clim. Past 9, 2319–2333 (2013).
57. Barker, S. & Knorr, G. Antarctic climate signature in the Greenland ice core
record. Proc. Natl Acad. Sci. USA 104, 17278–17282 (2007).
58. Lourantou, A., Chappellaz, J., Barnola, J.-M., Masson-Delmotte, V. & Raynaud, D.
Changes in atmospheric CO2 and its carbon isotopic ratio during the penultimate
deglaciation. Quaternary Sci. Rev. 29, 1983–1992 (2010).
59. Schneider, R., Schmitt, J., Köhler, P., Joos, F. & Fischer, H. A reconstrution of
atmospheric carbon dioxide and its stable carbon isotopic composition from
the penultimate glacial maximum to the last glacial inception. Clim. Past 9,
2507–2523 (2013).
60. Masson-Delmotte, V. et al. Abrupt change of Antarctic moisture origin at the
end of Termination II. Proc. Natl Acad. Sci. 107, 12091–12094 (2010).
61. Landais, A. et al. Two-phase change in CO2, Antarctic temperature and global
climate during Termination II. Nat. Geosci. 6, 1062–1065 (2013).
62. Stenni, B. et al. Expression of the bipolar see-saw in Antarctic climate records
during the last deglaciation. Nat. Geosci. 4, 46–49 (2011).
63. Buiron, D. et al. Regional imprints of millennial variability during the MIS 3
period around Antarctica. Quaternary Sci. Rev. 48, 99–112 (2012).
10
64. Khvorostovsky, D. V. et al. Vulnerability of permafrost carbon to global
warming. Part II: sensitivity of permafrost carbon stock to global warming.
Tellus B 60, 265–275 (2008).
65. Mudelsee, M. Break function regression. Eur. Phys. J. Spec. Top. 174, 49–63
(2009).
66. Bard, E., Raisbeck, G. M., Yiou, F. & Jouzel, J. Solar modulation of cosmogenic
nuclide production over the last millennium: comparison between 14C and 10Be
records. Earth Planet. Sci. Lett. 150, 453–462 (1997).
67. Cao, L. et al. The role of ocean transport in the uptake of anthropogenic CO2.
Biogeosciences 6, 375–390 (2009).
68. Lourantou, A. et al. Constraint of the CO2 rise by new atmospheric carbon
isotopic measurements during the last deglaciation. Global Biogeochem. Cycles
24, GB2015 (2010).
69. EPICA-community-members. One-to-one coupling of glacial climate
variability in Greenland and Antarctica. Nature 444, 195–198 (2006).
70. NGRIP Members. High-resolution record of Northern Hemisphere climate
extending into the last interglacial period. Nature 431, 147–151 (2004).
Acknowledgements
We thank H. Fischer for some insights on CH4 in ice cores and T. Opel for details on
permafrost. R. Muscheler calculated GISP2 10Be fluxes and provided all GISP2 10Be data
on the GICC05 age scale. A. Ridgwell provided helpful comments and GENIE model
results used for carbon cycle model evaluation. K. Saito provided PMIP3 result on LGM
permafrost distribution. M. Mudelsee helped with the application of the Breakfit software, V. Helm transformed IDL data into netCDF. G.K. acknowledges helpful discussions with G. Lohmann and S. Barker. G.K. is funded by REKLIM. E.B. is supported by the
European Community (Project Past for Future) and by the Agence Nationale de la
Recherche (Project EQUIPEX ASTER-CEREGE).
Author contributions
All authors designed research; P.K. performed carbon cycle simulations with BICYCLE;
E.B. performed carbon cycle simulations with other box models; G.K. performed climate
simulations; P.K. drafted the manuscript with contributions from all co-authors.
Additional information
Supplementary Information accompanies this paper at http://www.nature.com/
naturecommunications
Competing financial interests: The authors declare no competing financial interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
How to cite this article: Köhler, P. et al. Permafrost thawing as a possible source of
abrupt carbon release at the onset of the Bølling/Allerød. Nat. Commun. 5:5520
doi: 10.1038/ncomms6520 (2014).
This work is licensed under a Creative Commons Attribution 4.0
International License. The images or other third party material in this
article are included in the article’s Creative Commons license, unless indicated otherwise
in the credit line; if the material is not included under the Creative Commons license,
users will need to obtain permission from the license holder to reproduce the material.
To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
NATURE COMMUNICATIONS | 5:5520 | DOI: 10.1038/ncomms6520 | www.nature.com/naturecommunications
& 2014 Macmillan Publishers Limited. All rights reserved.
14.6 kyr BP
Tahiti
IntCal13
o
-1
-0.04 /oo yr
o
-1
-0.10 /oo yr
20
18
16
14
12
Tahiti
IntCal13
o
-1
-0.04 /oo yr
o
-1
-0.10 /oo yr
Time (kyr BP)
14.8
14.6
-1
-2
Be flux
6
-1
3
225
o
275
250
15.0
10
0.0
GISP2
GRIP
(10 atoms cm yr )
-1
0.1
C ( /oo)
0
450
400
350
300
250
200
150
100
10
0.2
14
GISP2
GRIP
10
0.3
200
175
14.4
14.2
atm
20
Be (10 atoms g )
0.0
10
30
6
0.1
10
40
0.4
o
0.2
0.5
C ( /oo)
50
b
14
0.3
55
50
45
40
35
30
25
20
15
10
atm
3
-1
Be (10 atoms g )
10
60
Be flux
0.5
0.4
0.6
-2
a
(10 atoms cm yr )
14.6 kyr BP
0.6
150
14.0
Time (kyr BP)
Supplementary Figure 1: Comparing atmospheric 10 Be and 14 C data to evaluate the potential
impact of variable 14 C production rates on 14 C. (a) Termination I. (b) Focus on 15 to 14 kyr BP. 10 Be
fluxes and concentration of Greenland ice cores GISP21 (bold closed lines, green and orange) and GRIP2
(dashed lines, blue and cyan) on GICC053 age model. Atmospheric 14 C from IntCal134 (grey band of
±1 around the mean) and Tahiti corals5 (magenta, ±1 in both age and 14 C) including a linear trend
with 0.04h yr 1 or 0.10h yr 1 .
1
050 yr
100 yr
150 yr
200 yr
250 yr
300 yr
a
260
250
270
b
260
250
240
240
230
230
atmosphere
@EDC
220
0.0
0.2
0.4
0.6
0.8
1.0
0.0
Simulation Time (kyr)
o
Probability ( /oo)
6
0.4
0.6
4
220
1.0
atmosp.:
125 PgC
100 PgC
@ EDC:
125 PgC
100 PgC
d
2
f(x)=1/(x
0.8
Simulation Time (kyr)
B/A @ EDC
c
5
0.2
(-1/2 ((ln(x) - )/ )) )
sqrt(2 )) e
EB/[email protected]=400 80yr
270
260
250
3
240
2
230
1
Scenario: 200 yr release, AMOC weak
0
0
200
400
600
800 1000
0.0
Time since exchange with atm (yr)
0.2
0.4
0.6
0.8
atmospheric CO2 (ppmv)
atmospheric CO2 (ppmv)
AMOC strong
atmospheric CO2 (ppmv)
AMOC weak
270
220
1.0
Simulation Time (kyr)
Supplementary Figure 2: Simulated changes in atmospheric CO2 using the BICYCLE carbon cycle
model. The model is set-up for around 14.6 kyr BP. Simulations vary in the length of the carbon release
between 50 yr and 300 yr, AMOC is (a) weak or (b) strong. Both the atmospheric signal (bold) and the
smoothed signal (dashed) that would be recorded in EDC are plotted. No age correction during smoothing
is applied. (c) The atmospheric time series are filtered (smoothed) to address the firn air mixing before gas
enclosure in ice cores with a log-normal probability density (PDF) function, that was tested with output
of firn densification models6 . The mean width E of PDF corresponds to the climatic conditions around
14.6 kyr BP at EDC (E = 400 ± 80 years (1 )). The PDF follows Equation 1 shown in the methods (main
text), in which µ was chosen in order to follow E = eµ 0.5 for the given mean width E = 400 years (bold
lines) or 320 and 480 year, respectively (thin lines). (d) Comparing how the true atmospheric CO2 signal in
two scenarios (125 or 100 PgC released in 200 years, AMOC off) might be recorded in EDC including the
1 uncertainty in the log-normal PDF (subfigure c) used for smoothing during gas enclosure.
2
o
C) ( /oo)
14
(
o
C) ( /oo)
14
(
o
C) ( /oo)
14
(
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
-100
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
-100
0
-10
-20
-30
-40
-50
-60
-70
-80
-90
-100
AMOC weak
AMOC strong
o
-0050 /oo
o
-0400 /oo
o
-0700 /oo
o
-1250 /oo
a
Release time: 050 yr
Release time: 200 yr
b
Release time: 050 yr
Release time: 200 yr
c
Release time: 100 yr
Release time: 250 yr
d
Release time: 100 yr
Release time: 250 yr
e
Release time: 150 yr
Release time: 300 yr
f
Release time: 150 yr
Release time: 300 yr
14.6
14.4
14.2
Time (kyr BP)
14.0
14.6
14.4
14.2
14.0
Time (kyr BP)
Supplementary Figure 3: Simulated changes in atmospheric 14 C using the BICYCLE carbon cycle
model. The model is set-up for around 14.6 kyr BP. Simulations vary in the length of the carbon release:
a,b) 50 and 200 years; c,d) 100 and 250 years; e,f) 150 and and 300 years. AMOC is weak (left) or strong
(right), and 14 C depletion of released carbon with respect to atmosphere varying between –50h and –
1250h. 14 C production rate was constant at 1.15⇥ pre-industrial level, which gives after 45 kyr of spin-up
time an atmospheric 14 C of about 250h, comparable with the observations.
3
AMOC weak
0
AMOC strong
(
14
o
C) ( /oo)
-10
-20
-30
-40
-50
a
1.15 PRE
GLOPIS
-60
14.6
14.4
14.2
14.0
Time (kyr BP)
b
14.6
14.4
14.2
14.0
Time (kyr BP)
Supplementary Figure 4: Impact of long-term change in 14 C production rate on simulated atmospheric 14 C. The 14 C production rate were derived from the geomagnetic field (GLOPIS) and applied in
the BICYCLE model as in a previous application7 . Results for our best guess scenario ( 1250h, 125 Pg
carbon released in 200 years) with 45 kyr of spin-up time are shown. AMOC is in (a) weak or (b) strong
mode. Cyan band spans results based on either a minimum or a maximum change in 14 C production rate as
deduced from GLOPIS7 .
4
200
35
30
25
700
20
650
15
600
CH4 Greenland
(GICC05-80yr)
550
500
TANT (K)
40
0.8
450
0.6
400
10
weak
strong
AMOC
PI
LGM
MIS3
0.4
5
AMOC index (Sv)
CO2 atm
(this study)
220
0.0
CH4 (ppbv)
CO2 (ppmv)
240
270
260
250
240
230
220
210
200
190
Flux (Sv)
0.2
260
N2O (ppbv)
N2O (TALDICE-1 age)
Freshwater hosing experiments
0
0.8
0.6
0.4
CO2_ATM - PRE_BA
CO2_ATM
0.2
0.2
0.0
0.0
-0.2
-0.2
-0.4
200 years
-0.6
0
a
200 years
100 200 300 400 500 600
0
Simulation time (years)
b
TANT (K)
GHG forcing experiments
-0.4
-0.6
100 200 300 400 500 600
Simulation time (years)
Supplementary Figure 5: Results of transient climate model simulations using COSMOS. Shown are
the simulated annual mean Antarctic surface temperature change TANT in (a) our GHG forcing experiments and (b) a reanalysis of North Atlantic freshwater hosing experiments8, 9 , which serve as a surrogate
for the simulation of abrupt climate changes. Recently it has been shown that the recovery and amplification of the AMOC at the end of freshwater perturbations is strongly dependent on the climate background
state8 . Therefore we analysed the Antarctic temperature response for three different climate states (PI: preindustrial; LGM: Last Glacial Maximum at 21 kyr BP; MIS3: 32 kyr BP) that cover a wide spectrum of
glacial-interglacial conditions to evaluate the robustness of the Antarctic temperature response. In Panel
(a) we show the GHG N2 O, CO2 and CH4 as used for forcing the model in simulation CO2 ATM and the
simulated anomaly in TANT in scenario CO2 ATM (green dashed) and of CO2 ATM–PRE BA (blue),
both as 100 years running averages with respect to year 100. Vertical lines bracket the rise in atmospheric
CO2 , representing 14.6 and 14.4 kyr BP, respectively. In panel (b) the North Atlantic freshwater hosing is
shown in Sv (1 Sv = 106 m3 s 1 ). The freshwater perturbation with a freshwater flux of 0.2 Sv is added to
the ice-rafted debris belt in the North Atlantic Ocean, around 40 N – 55 N, 45 W – 20 W of the central
Atlantic Ocean. The intensity of the AMOC is presented by the maximum value of the meridional overturning stream function of the upper 200–3000 m and 30 N northward8 . The simulated anomaly in TANT
for the different background climate states are shown as 100 years running averages with respect to year
230 (i.e. the end of the freshwater perturbation). Hence the corresponding key intervals marked in apricot
background colour in panel a) and b) are between 100-300 years and 230-430, respectively.
5
4
3
2
a
0
1800
1900
o
0.9
0.8
d
2000
Time (yr AD)
Time (yr AD)
310
20
300
atmospheric CO2 (ppmv)
30
(CO2) (ppmv)
Bomb
10
290
0
280
b
-10
270
1850
1900
atmospheric
atmospheric CO2 (ppmv)
Suess effect (1820-1950)
320
1950
400
380
360
340
320
300
280
260
e
1900
C) ( /oo)
o
C ( /oo)
14
o
-30
14
data
BICYCLE TB passive
BICYCLE TB active
box model V03b (Bard1997)
box model V12b (Bard1997)
100
1100
data:
global mean atm
1000
90
Eq Central Pacific corals
900
80
atmosphere:
800
BICYCLE TB passive
70
700
BICYCLE TB active
low lat surface ocean:
60
600
BICYCLE TB passive
BICYCLE TB active
50
500
400
40
300
30
200
20
100
10
0
-100
0
1950 1960 1970 1980 1990 2000
f
(
-20
1850
BICYCLE TB passive
BICYCLE TB active
V03b
V12b
1950 1960 1970 1980 1990 2000
atmospheric
atmospheric
14
o
C ( /oo)
c
0
-10
-30
C (1950-2000)
Time (yr AD)
-10
-20
14
Law Dome
Mauna Loa
Time (yr AD)
0
2500
1950
Time (yr AD)
C production rate (-)
1
1.0
airborne fraction f (-)
C
C prod. rate
C production rate (-)
14
14
14
0
5
1.1
relative
100
6
14
200
7
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
3000
100 Pg
BICYCLE TB-
relative
300
8
1.2
C of annual flux ( /oo)
-1
400
0
-100
-200
-300
-400
-500
-600
-700
-800
-900
-1000
2000
fossil fuel
land use change
annual C flux
cummulative C flux
14
9
Carbon flux (PgC yr )
Cumulative carbon flux [PgC]
500
Time (yr AD)
Supplementary Figure 6: Evaluation of 14 C model performance for Suess effect and bomb 14 C peak.
See caption on the following page.
6
Supplementary Figure 6: Evaluation of 14 C model performance for Suess effect and bomb 14 C peak.
(a) Fluxes of fossil fuel emissions10 (magenta), and land use change11 (green, between 1850 AD and 1750
AD linearly extrapolated to zero). Assumption for 14 C signature of total anthropogenic C flux to the atmosphere (red): Land use change extracts woody parts of trees (fractionated by 40h during photosynthesis
from atmosphere, no further ageing), fossil fuels are 14 C-free ( 14 C= 1000h). Relative changes in 14 C
production rate (cyan) with respect to the long-term Holocene12 . Gray box (1820–1950 AD) covers bombfree Suess effect data13 . (b) Simulated changes in atmospheric CO2 against CO2 data from the Law Dome
ice core14 (circles). BICYCLE model (red) with terrestrial biosphere (TB) either passive (constant, solid
lines) or active (changes as function of CO2 , dashed lines). A second box model in two different versions
(blue) for comparison15 . (c) Simulated changes in atmospheric 14 C against reconstructions (circles with
error bars (1 )13 ). BICYCLE started from 10 kyr BP, the 12-box model started at 0 AD, both used the same
relative 14 C production rate changes12 . (d) Airborne fraction f over time for the injection of 100 PgC in
year 2001 (background CO2 of 389 ppmv) in BICYCLE with terrestrial biosphere passive (TB-). No fossils
fuel emission after year 2000. f is calculated from the difference to simulation without 100 PgC injection
for the next 1000 years. Dotted lines mark f after 100 years. (e) Simulated atmospheric CO2 in the bomb
14
C-peak time window (1950–2000 AD) against CO2 data (Law Dome14 , Mauno Loa CO2 annual mean16 ).
(f) Simulated bomb 14 C-peak against reconstructions. Assumed additional bomb-related 14 C production
rate calculated after Naegler17 , but normalised to the mean natural 14 C production rate used here (85% of
Naegler17 ) leading to the input of 1.2 · 106 g 14 C into the atmosphere. Atmospheric 14 C data18 (global mean
(black filled circles) with range measured in high latitude northern hemisphere and southern hemisphere
(grey area) and coral-based reconstructions of central equatorial Pacific surface waters19 (open circles with
error bars denoting the observed range) against simulations with BICYCLE.
7
Suess effect (1820-1950)
BICYCLE TB-: m=1.07, r =0.98
2
BICYCLE TB+: m = 0.70, r =0.98
2
V03b: m = 0.93, r =0.98
2
V12b: m = 0.52, r =0.98
model - CO2 (ppmv)
(CO2) (ppmv)
model -
0
20
n=33
10
a
-10
-10
0
data -
20
340
320
300
c
1000
-20
-30
b
-20
(
-10
14
o
C ( /oo)
n=115
0
2
BICYCLE TB-: m=1.09, r =0.79
2
BICYCLE TB+: m = 1.14, r =0.79
800
600
model - atm
(
model -
n=43
data - CO2 (ppmv)
2
data -
360
(CO2) (ppmv)
BICYCLE TB-: m=1.10, r =0.94
2
BICYCLE TB+: m = 1.00, r =0.94
2
V03b: m = 0.85, r =0.94
2
V12b: m = 0.71, r =0.94
-30
2
280
280 300 320 340 360 380 400
30
14
-10
C (1950-2000)
BICYCLE TB-: m=1.42, r =1.00
2
BICYCLE TB+: m = 1.00, r =1.00
2
V03b: m = 1.10, r =1.00
2
V12b: m = 0.69, r =1.00
380
14
o
0
10
14
400
2
30
C) ( /oo)
Bomb
n=45
400
200
d
0
0
o
C) ( /oo)
200
400
data - atm
600
800 1000
14
o
C ( /oo)
Supplementary Figure 7: Suess effect and bomb 14 C peak. Scatter plots of model-data comparison for
changes in atmospheric CO2 and 14 C for the Suess effect and the bomb 14 C peak. (a) Atmospheric CO2
for the Suess effect. (b) Atmospheric 14 C for the Suess effect. (c) Atmospheric CO2 for the bomb 14 C
peak. (d) Atmospheric 14 C for the bomb 14 C peak. Model results are resampled only for the n years
in which data points are available. The dotted black lines denotes the perfect model-data agreement (slope
m = 1.0). The calculated slopes m and the correlation coefficients r2 of the linear regressions are contained
in the legends.
8
050 yr CO2 injection time
200 yr CO2 injection time
b
30
30
20
20
10
10
0
0
0
200
400
600
800 1000 0
Simulation Time (years)
200
400
600
800 1000
Simulation Time (years)
10
c
d
0
-10
-20
-20
14
(
-30
-30
Scenario:
AMOC on, pre-ind CC
14
pre-ind C production rate
CO2 injection: 125 PgC
14
o
C of injection: -1000 /oo
-40
-50
-60
0
200
400
600
BICYCLE
box model V03b (Bard1997)
box model V12b (Bard1997)
GENIE
800 1000 0
Simulation Time (years)
200
400
600
-40
14
o
-10
o
C) ( /oo)
0
C) ( /oo)
10
(
(pCO2) (ppmv)
a
40
(pCO2) (ppmv)
40
-50
-60
800 1000
Simulation Time (years)
Supplementary Figure 8: Comparing different carbon cycle models. BICYCLE (red), the other box
model15 (blue, two versions), and the Earth System model of intermediate complexity GENIE20 (cyan)
are compared for a pre-industrial CO2 injection scenario (AMOC strong, pre-industrial carbon cycle, preindustrial 14 C production rate) for 125 PgC injection with 14 C signature of 1000h. Injection started in
simulation year 100 for a length of 50 (left) or 200 (right) years. Top: Changes in atmospheric CO2 , bottom:
changes in atmospheric 14 C.
9
Supplementary Note 1
Atmospheric 14 C Data. Atmospheric 14 C deduced from corals might be influenced by temporal
changes in the reservoir ages, or by variable 14 C production rates. Both potential sources of uncertainty
are briefly discussed below:
(a) Reservoir ages
Reservoir ages might vary with time due to changes in (i) ocean circulation or (ii) gas exchange rates21 :
(i) Especially around the end of Heinrich Stadial 1 the resumption of the Atlantic meridional overturning circulation (AMOC) might lead to variable reservoir ages. This leads, for example, for Atlantic
sites in the Cariaco Basin in an ocean carbon cycle model to uncertainties and changes in the reservoir age
of the order of 100 years within a few centuries22 . This time-dependent change in reservoir age leads to a
difference in atmospheric 14 C of less than 5h if compared to results based on a constant reservoir age.
The influence of the AMOC resumption around 14.6 kyr BP on reservoir ages should be smaller in the central Pacific around Tahiti than in the Caribbean around Cariaco. Another simulation study23 focusing on the
impact of AMOC changes during the Younger Dryas also found rather constant reservoir ages throughout
most of the Pacific. Furthermore, wind-driven circulation or ventilation might also impact on the reservoir
age. However, as discussed in the next paragraph in detail we do not find any indication that surface wind
speed around Tahiti was changing, it stayed remarkable stable in all simulation results which were available
to us covering stable climates for modern times and LGM, and abrupt climate changes during MIS 3 and the
last deglaciation (this study and 8, 9 ). Reservoir age changes based on ocean circulation change is therefore
very small and thus negligible.
(ii) Gas exchange rates of CO2 between atmosphere and surface ocean depend mainly on local wind
speed. In existing simulations local surface wind speed around Tahiti (about 150 W, 18 S) are remarkable
stable. In simulations with the COSMOS coupled Earth system model (see further below) surface wind
speed around Tahiti never changes by more than 10% from its pre-industrial values. This is the case in
simulation results for LGM climate9 , for our simulation scenarios PRE BA and CO2 ATM (this study), or
across an AMOC shutdown and resumption8 . We therefore found no supporting evidence for a wind speed
change in Tahiti across an abrupt stadial/interstadial transition comparable to the B/A around 14.6 kyr BP
and therefore assume that the impact of reservoir age change on the interpretation of the coral-based 14 C
data from Tahiti was small and negligible.
A recent study24 used the same 14 C data from Tahiti corals to come to a different interpretation of
the reservoir age. They assumed that the IntCal09 stack provides the true atmospheric 14 C signal, which
implied that all differences in Tahiti 14 C to IntCal09 has to be explained by a variable reservoir age R.
Consequently, they found R to vary up to 400 years at the onset of the B/A. This is the opposite to our
interpretation. Here, we rely on the Tahiti corals to be the most reliable recorder of 14 C — because of its
small measurement uncertainty and precise U/Th dating. One main concern about the work by Kubota et
al.24 is the usage of the old IntCal09 curve rather than the new updated version4 IntCal13. Indeed, the main
difference between the two IntCal curves occurs around the Heinrich 1 event with significant improvements
included in the IntCal13 curve. This can be briefly illustrated with a simple example of an age within the
Heinrich 1 event: 16,500 calendar years BP corresponds to 13,310 14 C years BP in the IntCal09 curve, but
10
13,682 14 C years BP in the new IntCal13 curve. Consequently, the use of IntCal13 instead of IntCal09
reduces automatically by 372 years the apparent 14 C reservoir age for a coral sample of that calendar age.
This is the magnitude of the change in R that is interpreted erroneously by Kubota et al.24 (cf. their Fig. 4
and related text) as due to CO2 release by the ocean.
(b) Changes in 14 C production rate
Atmospheric 14 C might also vary because of a change in 14 C production rate, which would also be visible in other cosmogenic isotopes, e.g. in the atmospheric concentration of 10 Be which is available from
the GISP2 ice core1 and in high resolution from the GRIP ice core2 . GISP2 10 Be data on the GICC05
chronology3 show a decrease in 10 Be concentration by 50% around 14.65 kyr BP, but a gradual rise in
the 10 Be flux (Supplementary Fig. 1). Depending on the kind of deposition (wet, dry) atmospheric 10 Be
concentration is more related to either 10 Be concentration or 10 Be flux in ice cores. The patterns found in
10
Be data in the GISP2 ice core are reproduced in higher resolution in the data from the GRIP ice core
with some rapid overprints, mainly in the 10 Be flux data. Although during rapid climate changes it is not
clear which deposition is dominant1 , neither of both 10 Be concentration or 10 Be flux shows a pronounced
⇠ 200 250 years long anomaly comparable with the new 14 C data making a 14 C production rate change
a very unlikely explaination (see also25 for a discussion of difficulties linked to wet and dry deposition of
10
Be). Furthermore, upcoming new 10 Be data from the WAIS Divide Ice Core, Antarctica, show no abrupt
changes around 14.6 kyr BP, excluding changes in 14 C production rate as a plausible cause for the Tahiti
14
C anomaly26, 27 .
Supplementary Note 2
Evaluation of the Carbon Cycle Model. For model evaluation BICYCLE is (a) compared in its oceanic
carbon uptake dynamic resulting in a model-specific airborne fraction with other models, (b) used to simulate the Suess effect (years 1820 1950 AD), c) the bomb 14 C peak (years 1950 2000 AD), and (d) applied
on CO2 release experiments for pre-industrial background conditions. The model is compared with results
from another carbon cycle box model15 (Suess effect and for pre-industrial conditions) and with output from
the GENIE model20 , an Earth system model of intermediate complexity (pre-industrial conditions).
(a) Airborne fraction
A recent model inter-comparison28 with 14 models (three comprehensive Earth System Models, seven Earth
System Models of Intermediate Complexity, and four box-type models) calculated the airborne fraction f
of an instantaneous CO2 emission of 100 PgC to a background atmospheric CO2 of 390 ppmv. We here
repeat this scenario28 by adding 100 PgC in year 2000 of our anthropogenic emission scenario (BICYCLE,
terrestrial biosphere passive) having then a CO2 of 389 ppmv. After year 2000 no further anthropogenic
emissions are considered. The airborne fraction f is then calculated from the differences in atmospheric
CO2 for a run without this 100 PgC release (Supplementary Fig. 6d). f declines towards 0.45 after 100
years and to 0.20 after 1000 years, thus slightly larger than the rough estimate of f = 0.17 after 1000 years
given previously6 , in which the experiment and background conditions were different. These results agree
well with the model-mean of the model inter-comparison28 , which obtained f = 0.41 ± 0.13 (2 ) and
f = 0.25 ± 0.09 after 100 and 1000 years, respectively. A smaller airborne fraction (faster oceanic uptake
11
rate) would allow a larger amount of carbon released to the atmosphere to be confined with ice core data.
(b) Suess effect
Further evaluation of the 14 C dynamics in BICYCLE was obtained from simulating the Suess effect13, 29 : the
impact of anthropogenic fossil fuel and land use change emissions10, 11 on atmospheric 14 C (Supplementary Fig. 6a-c). This evaluation was chosen, because the Suess effect contains a well recognised influence
of carbon emissions on atmospheric 14 C, which are of similar magnitude and duration as the carbon release we proposed for the onset of the B/A. We here apply the anthropogenic emission rates and use the
BICYCLE model setup as previously published30, 31 .
The pre-bomb 14 C data reconstructed from well dated tree-rings including the Suess effect13 cover
the years 1820–1950 with atmospheric 14 C decreasing from +3h to 25h (Supplementary Fig. 6c). In
the original study13 the 14 C data continue until the year 1954, however the reconstruction of the 14 C production rate12 used for forcing the model stops at 1950, we therefore restrict our analysis also to results up
to the year 1950. The 14 C data show especially a decline of 23h between 1900 and 1950, furthermore
some multi-decadal feature before 1900. These atmospheric 14 C data contain the combined influence of
the anthropogenic emissions and variations in the 14 C production rate on atmospheric 14 C. The cumulative
anthropogenic carbon emissions during 1820–1950 sum up to 145 Pg of C (Supplementary Fig. 6a). Their
14
C signature depends on the relative share of 14 C-free fossil fuel emissions and land use change. For the
latter we assume a 14 C signature of 40h, being a mean value of pre-industrial terrestrial carbon within
BICYCLE. Anthropogenic emissions were dominated until the beginning of the 20th century by land use
change11 with annual output of fossil fuels being only of secondary importance10, 32 . The assumed mean
14
C of the total anthropogenic emissions before 1850 was therefore only 40 to 100h, but decreased
to 600h around 1950 and 750h around 2000 (Supplementary Fig. 6a). We use the recent calculation
of 14 C production rates based on changes in the geomagnetic field strength and the solar activity12 . These
relative 14 C production rates changes (calculated with respect to the long-term Holocene mean) varies in
the two century from 1750 to 1950 between 82% and 112% (Supplementary Fig. 6a).
Most important features of the observed changes in atmospheric CO2 rise and 14 C decline are met
by our simulations. We focus on simulated changes, not absolute values, therefore all model simulations
shown in Supplementary Figs. 6b,c are shown as anomalies to year 1820 (right y-axes). Absolute values of
simulations results differed from reconstructions by less than 15 ppmv (CO2 ) and less than 5h ( 14 C).
Atmospheric CO2 (Supplementary Fig. 6b) measured in the Law Dome ice core rises by 25 ppmv
between 1820 and 1950. Results with BICYCLE find a 22 ppmv rise in the model version with active
terrestrial biosphere for the same time period, implying that photosynthesis on land is via the fertilisation
effect a function of atmospheric CO2 . If terrestrial carbon content is kept constant (passive terrestrial
biosphere) the simulated rise in atmospheric CO2 between 1820 and 1950 is 35 ppmv, 10 ppmv higher than
in the ice core data.
Changes in atmospheric 14 C (Supplementary Fig. 6c) include both the rather steep decline after the
year 1900, but also important multi-decadal features of the 14 C data consisting of a local maximum in
14
C around 1900, a declining trend before 1840 and rather stable values in-between. Those features are
caused by corresponding anomalies in the 14 C production rates (Supplementary Fig. 6a). In the BICYCLE
12
simulations the steep declines in the first two decades and after year 1900 are met well, in the intermediate period of rather stable 14 C, the model produces an offset with respect to the data of 3 5h. The
difference in the two model applications with either active or passive terrestrial carbon cycle have only
minor influence of ⇠ 2h on the simulated changes in atmospheric 14 C. The model version with active
terrestrial biosphere, which also compares best with changes in atmospheric CO2 , produces a decrease in
atmospheric 14 C of 28h between 1820 and 1950 for the Suess effect (including natural variations in
the 14 C production rates) which is in very good agreement with the data.
The overall model-data misfit for the time window 1820 1950 is estimated from linear regression
analysis of simulated changes against observated changes (Supplementary Fig. 7a,b). Here, only simulation
results are picked for the n years in which observational data are available. An optimal model would have in
such a scatter plot all points on the diagonal with slope m = 1.0. We here find for the simulated atmospheric
CO2 compared to the n = 33 data points of Law Dome regression coefficients m of the slopes of 1.07 and
0.70 BICYCLE with passive and active terrestrial biosphere, respectively. For changes in atmospheric 14 C
(n = 115) the regression coefficients m of the slope are 1.10 and 1.00, listed again for the same simulation
scenarios. These linear regression analyses led furthermore to r2 0.94 for all model-data comparsions.
(c) Bomb 14 C peak
After the year 1950 the natural 14 C signals is overprinted by anthropogenic 14 C production during nuclear
bomb testing. The so-called bomb 14 C peak will be used as an additional model evalution. The magnitude
of the 14 C anomaly was by about a factor of 20 larger and faster than our carbon cycle anomaly during
the onset of the B/A, but the data are still of use for analysing model dynamics.
Anthropogenic carbon emission in the setup for the Suess effect were already extended until year
2000 (Supplementary Fig. 6a). To investigate the bomb 14 C peak we need as additional forcing the bombinduced artificial 14 C production. The background 14 C production rate varied before 1950 by 20% around
the mean Holocene values (of here 440 mol yr 1 ). From results of a 14 C budget closure17 we calculate that
the relative 14 C production rate increased in selected years (cyan points in Supplementary Fig. 6f). In the
peak years 1960 1963 this increase was by a factor of 30 84. The cumulative additional 14 C production
leads to the injection of 1.2 · 106 g 14 C into the atmosphere after 1950. This is 15% smaller than intially17
suggested, because our natural background 14 C production rate is only 85% of that chosen by Naegler17 .
The anthropogenic carbon emissions lead in the Law Dome14 and Mauno Loa16 data to a rise in CO2
of 60 ppmv, from ⇠310 ppmv in year 1950 to nearly 370 ppmv fifty years later (Supplementary Fig. 6e).
The simulated CO2 rise in BICYCLE (terrestrial biosphere active) is identical (slope m = 1.0 in datamodel scatter plot (Supplementary Fig. 7c)), but offset by 15 ppmv. With passive terrestrial biosphere
atmospheric CO2 in BICYCLE rises faster leading to an amplitude of 85 ppmv, or a slope m = 1.42 in the
data-model scatter plot.
The bomb 14 C peak was observed in the atmosphere at various station around the globe, and shows a
distinct north-south gradient before the year 1970. The calculation of a global mean 14 C peak is therefore
not straightforward. We here show a previously18 calculated mean, but also the range of the data between
high northern latitude and southern hemisphere (Supplementary Fig. 6f). The global mean in atmospheric
14
C peaks in the data in the mid 1960s at 700 ± 200h, and declines towards +100h in year 2000
13
thereafter. Additional information can be gained from the rise in 14 C in corals, recording surface ocean
signals. From a coral-data compilation19 we show (Supplementary Fig. 6f) the range of results obtained
in central equatorial Pacific surface waters, because these waters are less perturbed by upwelling or ocean
gyres, and can be used for comparison with our simple carbon cycle box model. These coral data record
the bomb 14 C peak by a rise from 50h in year 1950 towards +100h from year 1970 onwards.
BICYCLE simulates the bomb 14 C peak by an atmospheric 14 C of +1000h in year 1963, and a
decay towards 0h in year 2000. Results differ only slightly for terrestrial biosphere either active or passive.
The model-data scatter plot (Supplementary Fig. 7d) shows a slope m of 1.09 and 1.14, and r2 = 0.79 for
both BICYCLE realisations. Simulated surface equatorial Pacific 14 C nicely shows the rise from 1950
to 1970 by +150h similar to the coral data, but simulations decline thereafter faster than in the data. The
decay of the 14 C peak in atmosphere and surface ocean that is faster in the model than in the data indicates
that the vertical mixing between surface and deep ocean in the model operates faster than in nature. This is
a phenomenom well known for box models (in detail, the mixing in the high latitude is faster in box models
than for GCMs, see the calculation of the Harvarton-Bear index33, 34 , from which we quantified that this
effect is less pronounced in BICYCLE than in other box models34 ). The effect is the more pronounced the
larger the gradient in the tracers between surface and deep ocean is. The bomb 14 C peak is therefore only
of limited usage for the evalution of the model dynamics of BICYCLE, when we want to apply the model
to the B/A event, a problem with more than an order of magnitude smaller changes in the 14 C cycle.
(d) Carbon cycle model comparison
To obtain further knowledge on how the carbon cycle model performs and on how simulations depend on
the specific chosen model setup and parametrization we repeat the Suess effect experiments with another
carbon cycle box model15 (Supplementary Fig. 6). Furthermore, CO2 release experiments, similar to the
release proposed by permafrost thawing around 14.6 kyr BP, but with pre-industrial background conditions
are compared for both carbon cycle models, but also with output from the more complex GENIE model
(Supplementary Fig. 8). This simplified setup is used as a sensitivity test under generic and comparable
conditions.
The other box models have been used previously to study the influence of 14 C production changes15
and thermohaline circulation changes35 . Two geometries have been used with a 12 box model and a simpler
3 box model (V12b & V3b), which have been evaluated by calculating harmonic responses, in amplitude
and phase, to sinusoidal changes of the 14 C production (previous studies15, 25 show a comparison with other
models including the Bern2D model). V12b is a hybrid of PANDORA for the ocean36 and the terrestrial part
of the carbon cycle model of Siegenthaler37 . V12b and V3b were updated to accommodate transient changes
in the amount of 12 C in all reservoirs (assuming a constant Revelle buffer factor for ocean-atmosphere CO2
exchanges). The airborne fraction after a thousand years for the 125 PgC release experiments is 0.18 and
0.11 for V3b and V12b, respectively. V3b is thus similar to the BICYCLE model, while version V12b has
a carbon uptake faster than BICYCLE or most other models28 . Sensitivity tests indicate that this problem is
partly linked to the short residence time of carbon in the terrestrial biosphere (see below).
For the Suess effect, the time series forcing V12b and V3b (anthropogenic emissions, 14 C signature
of emissions, natural changes in 14 C production) were identical to those forcing BICYCLE (Supplementary
Fig. 6a). For the time interval 1820 to 1950 AD, the simulated atmospheric CO2 rise is 17 and 30 ppmv for
14
V12b and V3b, respectively (Supplementary Fig. 6b). For the same 1820-1950 period, the atmospheric 14 C
changes are 21 to 24h for V12b and V3b, respectively (Supplementary Fig. 6c). The carbon loss from
the atmosphere being larger in V12b than in V3b, it is logical to observe that the CO2 rise and atmospheric
14
C depletion are smaller in V12b than in V3b. Increasing the residence time of carbon in the terrestrial
biosphere in V12b by a factor five (from 60 to 300 yr) leads to a total CO2 increase by 24 ppmv and a 14 C
depletion by 23h over the 1820 1950 period.
Overall, Supplementary Figs. 6b and 6c show that for the Suess effect V12b and V3b compare well
with BICYCLE results and with the CO2 and 14 C observations. The slopes m in the model-data scatter
plots (Supplementary Fig. 7a,b) are 0.93 (CO2 V03b), 0.52 (CO2 V12b), 0.85 ( 14 C V03b) and 0.71 ( 14 C
V12b). In particular, V12b and V3b reproduce quite well the natural atmospheric 14 C variations before
1900 AD, mainly caused by geomagnetic field and solar activity. Some systematic differences are nonetheless observed: V12b underestimates by 10 ppmv the observed CO2 rise and by 5h the 14 C depletion in
1950. Surprisingly, V3b matches quite well both observed datasets. As mentioned above, V12b performs
better with a longer carbon residence time for its terrestrial biosphere. This effect is reminiscent to what is
observed for BICYCLE: CO2 observations are bracketed by simulations performed with a passive and active terrestrial biosphere, showing a difference of about 12 ppmv in 1950. BICYCLE slightly overestimate
atmospheric 14 C changes linked to natural forcing.
The anthropogenic CO2 release until year 2000 was also simulated with these box models, but not
the bomb 14 C peak. First tests have shown that the post-bomb 14 C peak decay in atmospheric 14 C in
those models is even faster than in BICYCLE, which again is very likely explained by vertical mixing
parametrizations between surface and deep ocean. V03b and V12b increased their simulated atmospheric
CO2 between year 1950 and 2000 by about 70 and 40 ppmv, respectively (Supplementary Figs. 6e), leading
to slopes m in scatter plot of 1.10 and 0.69 (Supplementary Fig. 7c).
We also performed a model comparison for a generic 125 PgC release over 50 or 200 years with preindustrial background conditions. The 14 C signature of the released carbon was chosen to be 1000h, i.e.
14
C-free carbon, equivalent to fossil fuel emissions. BICYCLE (with passive terrestrial biosphere) simulates
CO2 maxima of 37 and 25 ppmv, and 14 C minima of 55 and 23h for the 50 and 200 years release
times, respectively (Supplementary Fig. 8). As expected from the Suess effect comparison, V3b produces
CO2 and 14 C changes almost as large as BICYCLE while V12b simulates anomalies that are a factor
of two smaller. Again, using a longer biospheric residence time in V12b reduces partly the discrepancy
(e.g. 23 ppmv and 38h instead of 18 ppmv and 34h for the 50 yr release).
For further evaluation these atmospheric carbon cycle perturbations under pre-industrial background
conditions were also performed with GENIE, an Earth system model of intermediate complexity, in experiments without climate feedbacks to make them comparable to the box-models. The model configuration of
GENIE for this experiment was as described earlier20 with the addition of a 14 C source in the atmosphere
that balances the steady-state 14 C decay in ocean and atmosphere. In the experiment with 50 years carbon
release time the peak amplitudes of (pCO2 ) and ( 14 C) in GENIE are +37 ppmv and 54h, respectively, so nearly identical to the results in BICYCLE (Supplementary Fig. 8a,c). For the experiment with
200 years release time the amplitudes in GENIE are with +27 ppmv ( (pCO2 )) and 26h ( ( 14 C))
slightly larger than in BICYCLE (Supplementary Fig. 8b,d). The decline of the atmospheric carbon anomalies in GENIE are more in agreement with the V03b box model, and slightly slower than in BICYCLE. This
agrees with our findings that the airborne fraction f in BICYCLE is on a 100 years time scale comparable
15
with other models, but on a 1000 years time scale slightly smaller than the multi-model mean28 .
In summary, the comparison of the performances of BICYCLE, and the other models is a useful
exercise of model evaluation, giving confidence in the BICYCLE simulations of older carbon cycle perturbations of similar amplitude and timing. We therefore conclude that our carbon cycle model is capable
of simulating dynamics in the carbon and 14 C cycle of magnitude and duration of the Suess effect. For
BICYCLE with passive terrestrial biosphere, the configuration used in the B/A experiments, the relative
uncertainties in both CO2 and 14 C are 7%, and 10%, respectively. The model thus seems to be a suitable
tool for the investigation of the 14 C anomaly of similar characteristics around 14.6 kyr BP.
Supplementary Note 3
This Supplementary Note describes details of the climate simulations and analysis, which represents the
basis for the discussion section that highlights the potential importance of abrupt GHG changes to offset
the timing of the temperature maximum leading into the ACR, compared to the onset of the B/A.
(a) Data-based motivation
Ice core data and simulations show that the temperature signals connected with the bipolar seesaw are not
uniform across Antarctica. Regional differences of millennial climate variability were found in Antarctic
ice cores of different sectors during the last deglaciation38, 39 and MIS 340 . These earlier simulation results
are not directly applicable for the climate transition around 14.6 kyr BP, because here we have to understand
dynamics during an AMOC resumption.
The sequence of bipolar climate changes linking the onset of the B/A and the Antarctic Cold Reversal
(ACR) is of particular interest for our study. However, rapid changes in CH4 are used as age markers to synchronise ice cores from both hemispheres41 leaving only the independently annual layer-counted chronologies of the WD ice core in the south and of NGRIP in the north for further investigations on the timing of
the B/A and the ACR. Note, that temperature change42 based on a stack from five East Antarctic ice cores
(EDC, EPICA DML, Vostok, Dome Fuji, Talos Dome) shows the start of the ACR at approximately the
same time as WD (Fig. 5, main text), although the underlying age model is not entirely independent from
GICC05. If we analyse water isotopes from the two ice cores WD and NGRIP in detail they reveal a time
delayed onset of ⇠180 years in the ACR38 relative to the beginning of the B/A3, 43 , which is reduced to 145
years, once the Greenland temperature rise was aligned to 14.6 kyr BP. The given maximum counting uncertainties at the onset of the B/A are about 186 years for Greenland cores and 4% or 584 years for the ACR
in WD. Note, that these uncertainty estimates already include and capture the suggestion that the GICC05
chronology might be at least 65 years too old around 12 kyr BP44, 45 .
(b) Model setup
To test the importance of the abrupt carbon release and associated CO2 changes for the temperature response
in Antarctica we utilise the comprehensive Earth system model COSMOS in a coupled atmosphere-ocean
configuration, consisting of ECHAM546 and MPI-OM47 without any flux corrections48 . The atmosphere
16
model ECHAM5 was used at T31 resolution (⇠ 3.75 ) with 19 vertical levels. The ocean model MPI-OM
was run at an average resolution of ⇠ 3 with 40 vertical layers. These model components have been
applied for a wide range of glacial, interglacial and Neogene applications8, 49, 50 . Furthermore, COSMOS
has been evaluated with proxy data in an investigation of the LGM9 .
Based on the LGM ocean state and glacial boundary conditions we have performed two experiments
in order to investigate the impact of abrupt CO2 changes on the Antarctic temperature response. In experiment PRE BA we have simulated 1200 model years to mimic a climate state that is characterised by
GHG concentrations prior to the onset of the B/A. The respective concentrations are prescribed by constant
values of CO2 (228 ppmv), N2 O (233 ppbv) and CH4 (498 ppbv) to represent conditions at ⇠14.9 kyr BP
(reference run). Additionally, we have used simulation PRE BA as a basis to perform a transient simulation
(CO2 ATM) with varying GHG concentrations after 600 years of model integration in PRE BA.
In the transient experiment CO2 ATM we apply our suggested true atmospheric CO2 spike of our best
guess emission scenario (200 year long release of 0.625 Pg C yr 1 (in total 125 PgC) into the atmosphere
resulting in a peak amplitude in atmospheric CO2 change of 22 ppmv) together with measured N2 O 51 and
Greenland CH4 time series, the latter dated according to our understanding (Fig. 5 of main text). All GHG
records as taken here are plotted in Supplementary Fig. 5. Using these GHG conditions, CO2 ATM has
been run for 600 years. Hence, the two simulations can be used to deduce the impact of GHG changes
at the inter-hemispheric temperature signature. Such a GHG forcing scenario seems to be reasonable as
suggested by published (Fig. 3a of main text), CO2 measurements, though absolute values in CO2 might
differ from the ones prescribed in our simulation CO2 ATM.
(c) Climate feedback interpretation
The application of the GHG forcing (Supplementary Fig. 5a) leads to a rise of the annual mean Antarctic
surface temperature TANT and the respective timing of the temperature evolution shows similarities with
the temporal signature of the CO2 forcing. The maximum warming of 0.6 K occurs about 250 years after
the onset of the CO2 rise (Supplementary Fig. 5a). The timing and estimated imprint of this temperature
anomaly would generate about +0.5h changes in 18 O if measured in Antarctic ice cores (assuming a slope
in the water isotopic thermometer of 0.8h per K warming52 ). Based on the inter-hemispheric timing of the
ACR in WD (but also in the TANT stack from East Antarctic ice cores) we can derive that the apparent
delay of the maximum isotopic value leading into the ACR with respect to the onset of the B/A in NGRIP
can alternatively also be understood as a further increase of temperature and thus of 18 O in WD (Fig. 5 of
main text). The detected increase of ⇠0.7h in WD suggests that the bulk of the observed signal might be
explained by the simulated Antarctic temperature response to the abrupt increases in the GHG. The exact
contribution will depend on the timing and the strength of the CO2 pulse.
In this context the amplification of the AMOC at the end of Heinrich stadial 1 and associated temperature changes in Antarctica are of interest here. We thus compare our GHG forcing experiments with results
on AMOC strengthening as obtained in previous model simulations using COSMOS8, 9 (Supplementary
Fig. 5). Note, that most simulation studies focus on AMOC weakening40, 53 . Furthermore recently, it has
been shown that the recovery and amplification of the AMOC in abrupt climate change simulations applying
North Atlantic freshwater perturbations is strongly dependent on the climate background state8 . Therefore,
to evaluate the robustness of the Antarctic temperature response ( TANT ) we analysed AMOC strenght17
ening for three different climate states that cover a wide spectrum of glacial-interglacial conditions (PI:
pre-industrial; LGM: Last Glacial Maximum at 21 kyr BP; MIS3: 32 kyr BP). In these AMOC freshwater
hosing experiments, explained in detail elsewhere8, 9 , but briefly summarised in the caption to Supplementary Fig. 5, AMOC is weakened for about 150 years by the freshwater perturbation. Subsequently, the
end of the freshwater input leads to AMOC resumption within 80–200 years. The magnitude and speed of
the AMOC resumption depends on the climate background conditions (Supplementary Fig. 5b), as shown
previously8 .
A comparison between the simulated influence of the GHG forcing and the freshwater hosing experiments highlights the relative importance of the GHG induced Antarctic temperature changes (lower panels
in Supplementary Figs. 5a,b). As discussed above the GHG forcing might lead to a rise in TANT of up to
0.6 K after 250 years, depending on the magnitude and timing of the CO2 pulse. The freshwater hosing experiments are characterised by transient changes in TANT of less than 0.3 K, starting from the time when
freshwater input was shut off. In detail, TANT depends on the background climate state. For example
in the scenario with LGM background conditions TANT was already rising prior to the freshwater shutoff due to transient changes. Furthermore, a comparison of the transient GHG scenario and the transient
AMOC strengthening experiments shows that the TANT response in the GHG simulation (Supplementary
Fig. 5a) is stronger than the simulated impacts by AMOC changes (Supplementary Fig. 5b) already for
GHG changes at the time when about half of the the full 22 ppmv increase in CO2 is reached. Hence, also
smaller GHG spikes bear the potential to have a substantial impact on the Antarctic temperature response
when compared to effects caused by AMOC changes.
In summary, our model investigations suggest that the abrupt GHG changes are more important for the
Antarctic temperature signature than changes associated with an abrupt AMOC strenghening. This highlights the potential of abrupt GHG changes to significantly modulate the Antarctic temperature signature
during abrupt climate changes at the end of the last ice age.
18
Supplementary References
1. Finkel, R. C. & Nishiizumi, K. Beryllium 10 concentrations in the Greenland Ice Sheet Project 2 ice
core from 3–40 ka. Journal of Geophysical Research 102, 26699–26706 (1997).
2. Adolphi, F. et al. Persistent link between solar activity and Greenland climate during the Last Glacial
Maximum. Nature Geoscience 7, 662–666 (2014).
3. Rasmussen, S. O. et al. A new Greenland ice core chronology for the last glacial termination. Journal
of Geophysical Research 111, D06102 (2006).
4. Reimer, P. J. et al. IntCal13 and Marine13 Radiocarbon Age Calibration Curves 0–50,000 Years cal
BP. Radiocarbon 55, 1869–1887 (2013).
5. Durand, N. et al. Comparison of 14 C and U-Th ages in corals from IODP #310 cores offshore Tahiti.
Radiocarbon 55, 1947–1974 (2013).
6. Köhler, P., Knorr, G., Buiron, D., Lourantou, A. & Chappellaz, J. Abrupt rise in atmospheric CO2 at
the onset of the Bølling/Allerød: in-situ ice core data versus true atmospheric signals. Climate of the
Past 7, 473–486 (2011).
7. Köhler, P., Muscheler, R. & Fischer, H. A model-based interpretation of low frequency changes in the
carbon cycle during the last 120 000 years and its implications for the reconstruction of atmospheric
14
C. Geochemistry, Geophysics, Geosystems 7, Q11N06 (2006).
8. Gong, X., Knorr, G., Lohmann, G. & Zhang, X. Dependence of abrupt Atlantic meridional ocean
circulation changes on climate background states. Geophysical Research Letters 40, 3698–3704 (2013).
9. Zhang, X., Lohmann, G., Knorr, G. & Xu, X. Different ocean states and transient characteristic in Last
Glacial Maximum simulations and implications for deglaciation. Climate of the Past 9, 2319–2333
(2013).
10. Marland, G., Boden, T. & Andres, R. J. Global, Regional, and National CO2 Emissions. In Trends:
A Compendium of Data on Global Change (Carbon Dioxide Information Analysis Center, Oak Ridge
National Laboratory, U.S. Department of Energy,, Oak Ridge, Tenn., USA, 2005).
11. Houghton, R. A. Revised estimates of the annual net flux of carbon to the atmosphere from changes in
land use and land management 1850-2000. Tellus 55B, 378–390 (2003).
12. Roth, R. & Joos, F. A reconstruction of radiocarbon production and total solar irradiance from the
Holocene 14 C and CO2 records: implications of data and model uncertainties. Climate of the Past 9,
1879–1909 (2013).
13. Stuiver, M. & Quay, P. D. Atmospheric 14 C changes resulting from fossil fuel CO2 release and cosmic
ray flux variability. Earth and Planetary Science Letters 53, 349–362 (1981).
14. Etheridge, D. M. et al. Natural and anthropogenic changes in atmospheric CO2 over the last 1000 years
from air in Antarctic ice and firn. Journal of Geophysical Research D101, 4115–4128 (1996).
15. Bard, E., Raisbeck, G. M., Yiou, F. & Jouzel, J. Solar modulation of cosmogenic nuclide production
over the last millennium: comparison between 14 C and 10 Be records. Earth and Planetary Science
Letters 150, 453–462 (1997).
19
16. Keeling, R. F., Piper, S., Bollenbacher, A. & Walker, J. Atmospheric CO2 records from sites in the
SIO air sampling network. In Trends: A Compendium of Data on Global Change. (Carbon Dioxide
Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy, Oak Ridge,
Tenn., U.S.A., 2009). doi: 10.3334/CDIAC/atg.035.
17. Naegler, T. & Levin, I. Closing the global radiocarbon budget 1945-2005. Journal of Geophysical
Research 111, D12311 (2006).
18. Hua, Q. & Barbetti, M. Review of tropospheric bomb C-14 data for carbon cycle modeling and age
calibration purposes. Radiocarbon 46, 1273–1298 (2004).
19. Grottoli, A. G. & Eakin, C. M. A review of modern coral
Reviews 81, 67 – 91 (2007).
18
O and
14
C proxy records. Earth-Science
20. Cao, L. et al. The role of ocean transport in the uptake of anthropogenic CO2 . Biogeosciences 6,
375–390 (2009).
21. Bard, E. Correction of accelerator mass spectrometry 14 C ages measured in planktonic foraminifera:
Paleoceanographic implications. Paleoceanography 3, 635–645 (1988).
22. Butzin, M., Prange, M. & Lohmann, G. Readjustment of glacial radiocarbon chronologies by selfconsistent three-dimensional ocean circulation modeling. Earth and Planetary Science Letters 317 318, 177 – 184 (2012).
23. Singarayer, J. S. et al. An oceanic origin for the increase of atmospheric radiocarbon during the Younger
Dryas. Geophysical Research Letters 35, L14707 (2008).
24. Kubota, K., Yokoyama, Y., Ishikawa, T., Obrochta, S. & Suzuki, A. Larger CO2 source at the equatorial
Pacific during the last deglaciation. Scientific Reports 4, 5261 (2014).
25. Delaygue, G. & Bard, E. An Antarctic view of Beryllium-10 and solar activity for the past millennium.
Climate Dynamics 36, 2201–2218 (2011).
26. Woodruff, T., Welten, K., Caffee, M. & Nishiizumi, K. Cosmogenic 10 Be in WAIS Divide from 6–19
kyr BP. In WAIS Divide Ice Core Science Meeting (24–25 Sep 2013, La Jolla, CA, 2013).
27. Woodruff, T., Welten, K., Caffee, M. & Nishiizumi, K. Cosmogenic 36 Cl and 10 Be in the WAIS Divide
ice core from 6–21 kyr BP. In AGU Fall Meeting, Abstract C13A-0665 (9–13 Dec 2013, San Francisco,
CA, 2013).
28. Joos, F. et al. Carbon dioxide and climate impulse response functions for the computation of greenhouse
gas metrics: a multi-model analysis. Atmospheric Chemistry and Physics 13, 2793–2825 (2013).
29. Suess, H. E. Radiocarbon Concentration in Modern Wood. Science 122, 415–417 (1955).
30. Köhler, P., Fischer, H., Schmitt, J. & Munhoven, G. On the application and interpretation of Keeling plots in paleo climatic research — Deciphering 13 C of atmospheric CO2 measured in ice cores.
Biogeosciences 3, 539–556 (2006).
31. Köhler, P., Hartmann, J. & Wolf-Gladrow, D. A. Geoengineering potential of artificially enhanced
silicate weathering of olivine. Proceedings of the National Academy of Science 107, 20228–20233
(2010).
20
32. Andres, R. J. et al. A synthesis of carbon dioxide emissions from fossil-fuel combustion. Biogeosciences 9, 1845–1871 (2012).
33. Broecker, W. et al. How strong is the Harvardton-Bear constraint? Global Biogeochemical Cycles 13,
817–820 (1999).
34. Köhler, P., Fischer, H., Munhoven, G. & Zeebe, R. E. Quantitative interpretation of atmospheric carbon
records over the last glacial termination. Global Biogeochemical Cycles 19, GB4020 (2005).
35. Bard, E. et al. The North Atlantic atmosphere-sea surface 14 C gradient during the Younger Dryas
climatic event. Earth and Planetary Science Letters 126, 275 – 287 (1994).
36. Broecker, W. S. & Peng, T.-H. Carbon Cycle 1985: Glacial to interglacial changes in the operation of
the global carbon cycle. Radiocarbon 28, 309–327 (1986).
37. Siegenthaler, U., Heimann, M. & Oeschger, H.
cycle. Radiocarbon 22, 177–191 (1980).
14
C variations caused by changes in the global carbon
38. WAIS Divide Project Members. Onset of deglacial warming in West Antarctica driven by local orbital
forcing. Nature 500, 440–444 (2013).
39. Stenni, B. et al. Expression of the bipolar see-saw in Antarctic climate records during the last deglaciation. Nature Geoscience 4, 46–49 (2011).
40. Buiron, D. et al. Regional imprints of millennial variability during the MIS 3 period around Antarctica.
Quaternary Science Reviews 48, 99 – 112 (2012).
41. Veres, D. et al. The Antarctic ice core chronology (AICC2012): an optimized multi-parameter and
multi-site dating approach for the last 120 thousand years. Climate of the Past 9, 1733–1748 (2013).
42. Parrenin, F. et al. Synchronous change in atmospheric CO2 and Antarctic temperature during the last
deglacial warming. Science 339, 1060 – 1063 (2013).
43. Steffensen, J. P. et al. High-resolution Greenland ice core data show abrupt climate change happens in
few years. Science 321, 680–684 (2008).
44. Muscheler, R. et al. Tree rings and ice cores reveal
Dryas. Nature Geoscience 1, 263 – 267 (2008).
14
C calibration uncertainties during the Younger
45. Lohne, O. S., Mangerud, J. & Birks, H. H. Precise 14 C ages of the Vedde and Saksunarvatn ashes and
the Younger Dryas boundaries from western Norway and their comparison with the Greenland Ice Core
(GICC05) chronology. Journal of Quaternary Science 28, 490–500 (2013).
46. Roeckner, E. et al. The atmospheric general circulation model ECHAM5, part I: Model description,
vol. 349 of Reports (Max Planck Institute for Meteorology, Hamburg, Germany, 2003).
47. Marsland, S. J., Haak, H., Jungclaus, J. H., Latif, M. & Röske, F. The Max Planck Institute global
ocean/sea ice model with orthogonal curvilinear coordinates. Ocean Modelling 5, 91–127 (2003).
48. Jungclaus, J. et al. Ocean circulation and tropical variability in the coupled model ECHAM5/MPI-OM.
Journal of Climate 19, 3952–3972 (2006).
21
49. Knorr, G., Butzin, M., Micheels, A. & Lohmann, G. A warm Miocene climate at low atmospheric CO2
levels. Geophysical Research Letters 38, L20701 (2011).
50. Zhang, X., Lohmann, G., Knorr, G. & Purcell, C. Abrupt glacial climate shifts controlled by ice sheet
changes. Nature 512, 290–294 (2014).
51. Schilt, A. et al. Atmospheric nitrous oxide during the last 140,000 years. Earth and Planetary Science
Letters 300, 33 – 43 (2010).
52. Masson-Delmotte, V. et al. A Review of Antarctic Surface Snow Isotopic Composition: Observations,
Atmospheric Circulation, and Isotopic Modeling;. Journal of Climate 21, 3359–3387 (2008).
53. Kageyama, M. et al. Climatic impacts of fresh water hosing under Last Glacial Maximum conditions:
a multi-model study. Climate of the Past 9, 935–953 (2013).
22
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement