schmitt2012s accepted all

schmitt2012s accepted all
1
1
2
3
Title:
4
Authors:
5
Jochen Schmitt1,3*, Robert Schneider1, Joachim Elsig1, Daiana Leuenberger1, Anna
6
Lourantou2†, Jérôme Chappellaz2, Peter Köhler3, Fortunat Joos1, Thomas F. Stocker1,
7
Markus Leuenberger1 & Hubertus Fischer1,3
Carbon isotope constraints on the deglacial CO2 rise from ice cores
8
9
Affiliations:
10
1
11
Change Research, University of Bern, Switzerland.
12
2
13
l’Environnement (LGGE) UMR 5183, Grenoble, France.
14
3
15
† Present address: Laboratoire d’Océanographie et du Climat (LOCEAN), Institut Pierre
16
Simon Laplace, Université P. et M. Curie (UPMC), Paris, France.
17
*Correspondence to: [email protected]
Climate and Environmental Physics, Physics Institute & Oeschger Centre for Climate
UJF – Grenoble 1 / CNRS, Laboratoire de Glaciologie et Géophysique de
Alfred Wegener Institute for Polar and Marine Research, Bremerhaven, Germany.
18
19
20
This is the author's version of the work. It is posted here by permission of the AAAS for
21
personal use, not for redistribution. The definitive version was published in Science (336
22
(6082), pp. 711-714, 11 May 2012), doi:10.1126/science.1217161.
23
2
24
Abstract:
25
The stable carbon isotope ratio of atmospheric CO2 (!13Catm) is a key parameter to decipher
26
past carbon cycle changes. Here we present !13Catm data for the last 24,000 years derived
27
from three Antarctic ice cores. We conclude that a pronounced 0.3‰ decrease in !13Catm
28
during the early deglaciation can be best explained by upwelling of old, carbon-enriched
29
waters in the Southern Ocean. Later in the deglaciation, regrowth of the terrestrial
30
biosphere, changes in sea surface temperature, and ocean circulation governed the !13Catm
31
evolution. During the Last Glacial Maximum, !13Catm and CO2 were essentially constant,
32
suggesting that the carbon cycle was in dynamic equilibrium and that the net transfer of
33
carbon to the deep ocean had occurred before then.
34
35
One Sentence Summary:
36
High-resolution ice core records of stable carbon isotopes in atmospheric CO2 constrain
37
carbon cycle changes during the Last Glacial Maximum, the early deglacial, and the
38
transition into the Holocene.
39
40
41
42
Main Text:
43
During the last 800,000 years (800 kyr), atmospheric CO2 concentrations have varied in
44
close relation to Antarctic temperatures (1, 2) and the general waxing and waning of
45
continental ice sheets. In particular, CO2 rose from a stable level of 190 parts per million by
3
46
volume (ppmv) during the Last Glacial Maximum to about 280 ppmv in preindustrial times,
47
showing pronounced differences in atmospheric CO2 rates of change in the course of the
48
last glacial/interglacial transition (3). Many processes have been involved in attempts to
49
explain these CO2 variations, but it has become evident that none of these mechanisms
50
alone can account for the 90 ppmv increase in atmospheric CO2. A combination of
51
processes must have been operating (4, 5), with their exact timing being crucial. However,
52
a unique solution to the deglacial carbon cycle changes has not been yet found.
53
54
In this respect, high-resolution and precise !13Catm records from Antarctic ice cores are
55
needed to better constrain the evolution of carbon cycle changes during the last
56
deglaciation. On millennial time scales, !13Catm is primarily influenced by the !13C of
57
dissolved inorganic carbon (DIC) (!13CDIC) and sea surface temperature (SST), which
58
controls the isotopic fractionation during air/sea gas exchange. The continuous rain of
59
isotopically light organic material to the interior of the ocean draws down carbon from the
60
surface layer to intermediate and deep waters, where the organic carbon is remineralized.
61
Consequently, a vertical !13CDIC gradient is established, controlled by the interplay of the
62
ocean circulation with this so-called “biological pump”. The more intense the circulation,
63
the smaller the gradients are for !13CDIC, DIC, oxygen and nutrients. Superimposed on these
64
marine carbon cycle processes are climate-induced changes in terrestrial biosphere carbon
65
storage, which result in a net change in the carbon isotopic composition of the
66
ocean/atmosphere system. On orbital time scales, weathering and sedimentation of CaCO3
67
affect !13CDIC, !13Catm and atmospheric CO2 as well.
68
4
69
Until recently (6), analytical constraints represented the fundamental limitation on the
70
utility of !13Catm ice core records (7, 8). Here we provide evidence (Fig. 1) about possible
71
causes of carbon cycle changes with measurements of !13Catm from two Antarctic ice cores
72
(EPICA (European Project for Ice Coring in Antarctica) Dome C and Talos Dome),
73
performed with three independent methods in two different labs (referred to as Bern
74
sublimation, Bern cracker and Grenoble mill data) (6, 9). One of our records is based on a
75
sublimation method (10) that avoids the effects associated with incomplete gas extraction
76
and thus yields more precise results (see Supporting Online Material (SOM)). A stringent
77
residual analysis of the three data sets shows virtually no offset between the two Bern data
78
sets and only a small systematic offset between the Bern and Grenoble data of 0.16‰,
79
which can be explained by a method-dependent systematic fractionation. After correction
80
of this offset, we combined the three !13Catm records over the last 24 kyr using an error-
81
weighted Monte Carlo bootstrap approach. This method showed that all three data sets are
82
essentially compatible within their analytical uncertainties. To make full use of the
83
resolution and precision of the data, the inclusion of all three data sets is required, although
84
all our conclusions are also supported by the individual records. The final data set consists
85
of 201 individual measurements, each reflecting typically 2 to 4 replicates and with an
86
analytical 1" error between 0.04 and 0.12‰. Since the resulting Monte Carlo Average
87
(MCA) removes most of the analytical uncertainties, it contains less high-frequency
88
variability compared to the raw data. This is in line with the centennial-scale low-pass
89
filtering inherent to the bubble enclosure process at Dome C. Accordingly, the retained
5
90
variability can be regarded as the signal most representative of millennial !13Catm changes
91
(see SOM for details regarding the MCA and its uncertainty).
92
93
Our !13Catm data is in good agreement with previously published lower-resolution records
94
(6, 9). Our record shows a very stable level between 24 and ca. 19 kyr before present (BP,
95
where present is defined as 1950), with an average !13Catm of -6.45‰ (Tab. S1 and S2),
96
similar to the -6.35‰ of the Late Holocene (Fig. 2B). Given the fact that a large set of
97
environmental parameters such as atmospheric CO2, global SST, terrestrial carbon storage,
98
and ocean circulation have varied between the LGM and the Late Holocene, almost
99
identical !13Catm values indicate that opposing effects must have offset each other (11). This
100
becomes clear if we look at three first-order effects on !13Catm: A SST rise of 1 K translates
101
into a 0.1‰ increase in !13Catm, due to temperature-dependent fractionation between
102
atmospheric CO2 and marine DIC species (12). Assuming a global LGM-to-Holocene SST
103
rise of 3 K would result in about 0.3‰ higher !13Catm for the Holocene, provided that SST
104
distribution and CO2 gross flux exchange patterns remained constant. This effect is further
105
augmented by the uptake of isotopically light carbon by the land biosphere and
106
counterbalanced by the smaller vertical gradient in !13CDIC in the Holocene ocean,
107
supported by marine data (13). The fact that both !13Catm and CO2 show little variation from
108
24 to 19 kyr BP points to the carbon cycle being essentially in dynamic equilibrium at that
109
time. As can also be seen in Fig. 2, the climate variations related to Heinrich stadial 2
110
(HS2) and Dansgaard-Oeschger event 2 (DO2) had little effect on the global carbon cycle
111
during this time interval. However, given the opposing trends for reconstructed atmospheric
6
112
#14C (#14Catm) (14, 15) and the expected #14Catm evolution (16) based on variations in 14C
113
production rate (17, 18), the global 14C budget was not balanced (Fig. 2A).
114
115
After a very small increase in !13Catm at the very end of the glacial, a sharp drop in !13Catm
116
starting at 17.5 kyr parallels the onset of increasing atmospheric CO2. Taken at face value,
117
this would point to an early SST rise that preceded the onset of the CO2 increase. When we
118
apply a crude SST correction to our !13Catm data based on a global estimate of SST
119
temperature changes during the transition (see SOM), this !13Catm increase vanishes (Fig.
120
2B). Note, however, that this 0.06‰ excursion is within the uncertainties of our data and
121
that other effects could also lead to this small enrichment in !13Catm. The 0.3‰ drop in
122
!13Catm after the onset of the transition at 17.5 kyr BP is accompanied by a CO2 increase of
123
about 35 ppmv and a 190‰ drop in #14Catm (19), which has been attributed to a release of
124
old carbon from the deep ocean. This coeval drop in !13Catm and #14Catm during the so-
125
called “mystery interval”, 17.5 – 14 kyr BP (19), is arguably the most enigmatic carbon
126
cycle change in the course of the transition and will be discussed in more detail below.
127
128
After the broad !13Catm minimum is reached at about 16 kyr BP, !13Catm increases slightly
129
by 0.1‰ during the pronounced Bølling-Allerød (BA) warming. Other than circulation
130
changes in the Southern Ocean (20), the regrowth of the terrestrial biosphere in the northern
131
hemisphere could contribute to this increase in !13Catm (4). However, since the SST-
132
corrected !13C evolution (Fig. 2B) does not show any increase, a robust process attribution
133
requires precisely dated SST reconstructions and transient carbon cycle modeling.
7
134
135
An almost linear rise by 0.06‰ per kyr follows the second !13Catm minimum at 12.2 kyr
136
BP, leading to maximum values of -6.33‰ at around 6 kyr BP. This rise might be largely
137
explained by the continuing regrowth of the terrestrial biosphere (21), in concert with
138
smaller contributions from SST warming and changes in circulation and export production
139
(9, 22). From this mid-Holocene maximum, !13Catm values decline slightly to reach values
140
of -6.35‰ at 0.5 kyr BP, as previously reported (6).
141
142
As mentioned above, the carbon cycle changes during the mystery interval have been a
143
matter of intense debate (19, 20, 23). Our high-resolution !13Catm record together with other
144
records of carbon cycle changes and insights from models may help to constrain hypotheses
145
put forward to explain the mystery interval. The rise in CO2 and the decline in !13Catm and
146
#14Catm between 17 and 15 kyr BP fit the concept of bringing DIC-rich waters with old
147
carbon into exchange with the atmosphere. Indicative 14C signals of upwelling of old, CO2-
148
enriched deep water were found in Pacific intermediate waters (24), but others (23) ruled
149
out such old water in the northeast Pacific, and evidence for a
150
ocean remains elusive (19, 23, 25). These #14C studies were usually confronted with
151
variable reservoir age between benthic and planktonic foraminifera. A study using deep sea
152
corals now circumvents this problem by applying absolute U-Th dating and shows that the
153
deep glacial Southern Ocean indeed ventilated its 14C-depleted reservoir during the mystery
154
interval (26).
155
14
C-depleted glacial deep
8
156
The constant !13Catm values during the late glacial indicate that the build-up of such an old,
157
DIC-rich reservoir must have occurred before 24 kyr BP. A large number of records mark
158
the start of the deglaciation around 17 kyr BP (Fig. 2). Within the uncertainty in marine and
159
ice core age scales, the CO2 increase, the pronounced #14Catm drop (15), the resumption of
160
vigorous Southern Ocean upwelling as recorded in intense deposition of biogenic opal (20),
161
and the launch of ice-rafted debris layers at the beginning of the Heinrich 1 stadial (27) all
162
occurred simultaneously. Interestingly, our !13Catm record shows its largest deviation of
163
0.3‰, i.e. the entire !13Catm decrease from the LGM to the Preboreal (PB), within the first 2
164
kyr after the start of the deglaciation. Within the same interval, CO2 rose by 30 ppmv from
165
190 ppmv to 220 ppmv, i.e. only 35% of the LGM-PB rise. Together with the trend reversal
166
in !13Catm towards the end of the mystery interval, this indicates that only a fraction of the
167
glacial/interglacial CO2 increase can be explained by an intensification of deep ocean
168
ventilation bringing isotopically depleted and carbon-rich water to the surface of the
169
Southern Ocean. Our new, high-resolution !13Catm data constrain the period of this release
170
of isotopically depleted carbon from the deep ocean to the atmosphere to between 17.4 kyr
171
BP and 15 kyr BP. This interpretation of the proxy records is quantitatively in line with
172
dynamical ocean model results that link deep ocean ventilation, atmospheric CO2, !13Catm,
173
!13CDIC, opal burial, and radiocarbon (28).
174
175
Alternative hypotheses (29, 30) invoking the release of old carbon from permafrost or
176
carbon locked under continental ice sheets are unlikely to explain the carbon cycle changes
177
in the mystery interval because the amount of terrestrial carbon needed to account for the
9
178
14
179
in atmospheric CO2. Moreover, it would lead to an overall decline in !13CDIC, which is not
180
observed in benthic foraminifera in the deep ocean (13, 22). Also, a carbonate dissolution
181
event at the sea floor that would have to accompany such a large terrestrial carbon release
182
into the atmosphere/ocean system is not imprinted in the deglacial marine CaCO3 record
183
(31)
C drop is very large, at 5000 Gt (25), and would conflict with the moderate 30 ppmv rise
184
14
185
Consequently, despite the fact that the search for an extremely
186
mass in marine records has thus far not been successful (23) and might not even essential to
187
explain the #14Catm anomaly (26), the release of carbon from the deep ocean remains the
188
most plausible scenario to explain the early deglacial drop in our new !13Catm record.
189
Furthermore, model results suggest that a !13Catm decrease of 0.3‰ and a CO2 increase of
190
about 30 ppmv can be accommodated by relatively small (about 20‰) and spatially
191
complex changes in deep ocean #14C (28). These changes may remain undetected in the
192
search for the old abyssal water using benthic foraminifera (19, 25). However, they are also
193
too small to explain the reconstructed #14Catm decline in the mystery interval. Based on
194
these considerations, the currently available marine and ice core information cannot be
195
reconciled with the atmospheric radiocarbon record in a straightforward manner. One
196
possibility to resolve this issue is to also reconsider a larger change in
197
between the Holocene and the glacial, and to work towards independent verification of the
198
#14Catm history.
199
C-depleted deep water
14
C production
10
199
References and Notes:
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
D. Lüthi et al., Nature 453, 379 (2008).
H. Fischer et al., Quat. Sci. Rev. 29, 193 (2010).
E. Monnin et al., Science 291, 112 (2001).
P. Köhler, H. Fischer, G. Munhoven, R. E. Zeebe, Global Biogeochem. Cycles 19,
GB4020 (2005).
D. M. Sigman, M. P. Hain, G. H. Haug, Nature 466, 47 (2010).
J. Elsig et al., Nature 461, 507 (2009).
H. J. Smith, H. Fischer, M. Wahlen, D. Mastroianni, B. Deck, Nature 400, 248
(1999).
M. Leuenberger, U. Siegenthaler, C. C. Langway, Nature 357, 488 (1992).
A. Lourantou et al., Global Biogeochem. Cycles 24, GB2015 (2010).
J. Schmitt, R. Schneider, H. Fischer, Atmos. Meas. Tech. 4, 1445 (2011).
P. Köhler, H. Fischer, J. Schmitt, Paleoceanography 25, PA1213 (2010).
J. Zhang, P. D. Quay, D. O. Wilbur, Geochim Cosmochim Ac 59, 107 (1995).
K. I. C. Oliver et al., Clim. Past 6, 645 (2010).
P. J. Reimer et al., Radiocarbon 51, 1111 (2009).
J. Southon, A. L. Noronha, H. Cheng, R. L. Edwards, Y. Wang, Quat. Sci. Rev. 33,
32 (2012).
P. Köhler, R. Muscheler, H. Fischer, Geochem. Geophys. Geosyst. 7, Q11N06
(2006).
C. Laj, C. Kissel, J. Beer, in Timescales of the Paleomagnetic Field. (American
Geophysical Union, 2004), vol. 145, pp. 255-265.
R. Muscheler, R. Beer, P. W. Kubik, H. A. Synal, Quat. Sci. Rev. 24, 1849 (2005).
W. Broecker, S. Barker, Earth Planet. Sci. Lett. 256, 90 (2007).
R. F. Anderson et al., Science 323, 1443 (2009).
P. Ciais et al., Nat. Geosci 5, 74 (2012).
J. M. Yu et al., Science 330, 1084 (2010).
D. C. Lund, A. C. Mix, J. Southon, Nat. Geosci. 4, 771 (2011).
L. Stott, J. Southon, A. Timmermann, A. Koutavas, Paleoceanography 24, PA2223
(2009).
W. Broecker, E. Clark, Geophys. Res. Lett. 37, L13606 (2010).
A. Burke, L. F. Robinson, Science 335, 557 (2012).
E. Bard, F. Rostek, J. L. Turon, S. Gendreau, Science 289, 1321 (2000).
T. Tschumi, F. Joos, M. Gehlen, C. Heinze, Clim. Past 7, 771 (2011).
R. Zech, Y. Huang, M. Zech, R. Tarozo, W. Zech, Clim. Past 7, 501 (2011).
N. Zeng, Clim. Past 3, 135 (2007).
D. A. Hodell, C. D. Charles, F. J. Sierro, Earth Planet. Sci. Lett. 192, 109 (2001).
B. Lemieux-Dudon et al., Quat. Sci. Rev. 29, 8 (2010).
North Greenland Ice Core Project members, Nature 431, 147 (2004).
EPICA Community Members, Nature 444, 195 (2006).
B. Stenni et al., Science 293, 2074 (2001).
J. Elsig, Dissertation, University of Bern (2009).
D. Lüthi et al., Earth Planet. Sci. Lett. 297, 226 (2010).
11
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
38.
39.
40.
41.
42.
43.
44.
45.
46.
258
Acknowledgements:
259
We thank two anonymous reviewers for carefully reviewing the manuscript. Financial
260
support for this study was provided in part by Deutsche Forschungsgemeinschaft (DFG),
261
Helmholtz Gemeinschaft, and Schweizerischer Nationalfonds (SNF). This work is a
262
contribution to the “European Project for Ice Coring in Antarctica” (EPICA), a joint
263
European Science Foundation/European Commission (EC) scientific program, funded by
264
the EC under the Environment and Climate Program and by national contributions from
265
Belgium, Denmark, France, Germany, Italy, The Netherlands, Norway, Sweden,
266
Switzerland, and U.K. The main logistic support at Dome C was provided by the Institut
267
Polaire Français – Paul Emile Victor (IPEV) and PNRA. Ice core material was also used
268
from the Talos Dome Ice Core Project (TALDICE), a joint European program led by Italy
269
and funded by national contributions from Italy, France, Germany, Switzerland and the
270
United Kingdom. The main logistical support at Talos Dome was provided by PNRA. This
271
is
272
http://doi:10.1594/PANGAEA.772713.
47.
48.
49.
G. Dreyfus, Phd Thesis, Princeton University (2008).
G. B. Dreyfus et al., Quat. Sci. Rev. 29, 28 (2009).
R. J. Francey et al., Tellus 51B, 170 (1999).
R. Spahni et al., Geophys. Res. Lett. 30, 25 (2003).
D. Buiron et al., Clim. Past 7, 1 (2011).
I. G. Enting, J. Geophys. Res.-Atmos. 92, 10977 (1987).
J. Schwander, B. Stauffer, A. Sigg, Annals of Glaciology 10, 141 (1988).
A. Indermühle et al., Nature 398, 121 (1999).
R. E. Zeebe, D. A. Wolf-Gladrow, CO2 in seawater: equilibrium, kinetics, isotopes,
Elsevier Oceanography series (2005), pp. 65.
P. M. Grootes, M. Stuiver, J. Geophys. Res.-Oceans 102, 26455 (1997).
J. Jouzel et al., Science 317, 793 (2007).
K. Visser, R. Thunell, L. Stott, Nature 421, 152 (2003).
EPICA
publication
No.
284.
The
data
is
accessible
online
at
12
273
Fig. 1. Ice core reconstructions of atmospheric !13C and CO2 concentration covering the
274
last 24 kyr. (A) !13Catm of atmospheric CO2 measured with three different methods on two
275
different ice core drill sites. Blue circles: Bern cracker data, green squares: Grenoble mill
276
data (9) after offset correction, red circles: Bern sublimation data. Red stars indicate values
277
from the sublimation method but measured on Talos Dome Ice Core (TALDICE). Error
278
bars represent the standard deviation of replicate measurements. The black line is the result
279
of 4000 Monte Carlo simulations representing an error-weighted average of the different
280
!13Catm data sets. The light and dark shaded areas represent the 2" and 1" error envelope
281
around the Monte Carlo Average (see SI). (B) CO2 concentration. Black circles represent
282
earlier measurements on EDC (3), other symbols are the same as in panel A. Note: All ice
283
core records are plotted on a synchronized age scale (32).
284
13
284
Fig. 2. Ice core reconstructions and marine records illustrating the evolution of major
285
components of the Earth climate system over the last 24 kyr. (A) Reconstructed #14Catm
286
from IntCal09 (14) and the
287
modeled (16) #14Catm assuming a constant carbon cycle under pre-industrial conditions but
288
considering temporal changes in 14C production (either based on 10Be (18), upper and lower
289
estimates enveloped in gray lines, or on paleomagnetic field intensity (17), hatched area).
290
(B) Monte Carlo simulations (this study) of the evolution of !13Catm before (red line
291
represents the MCA, 2" and 1" envelopes are in gray) and after SST correction (gray line;
292
see SI) (C) Opal flux in the Southern Ocean as a proxy for local upwelling (20). (D) Record
293
of ice rafted debris (IRD) in the North Atlantic associated with Heinrich stadials (HS1 and
294
HS2) (27). (E) Greenland temperature proxy !18O (33). (F) Reconstructed atmospheric CH4
295
concentration (34) (G) Antarctic temperature proxy !D from the EDC ice core (35). (H)
296
Compilation of reconstructed CO2 shown in Fig. 1B. Green bars indicate intervals with a
297
strong net terrestrial carbon build-up, blue bars indicate intervals where sequestered deep
298
ocean CO2 was released back to the atmosphere. Note: Ice core records are plotted on a
299
synchronized age scale (32), other records are plotted on their individual age scales.
300
230
Th-dated Hulu Cave #14Catm record (15) compared with
Age (kyr BP)
ï6.2
5
10
15
20
25
A
ï6.3
ï6.4
ï6.5
ï6.6
2m envelope
1m envelope
Monte Carlo Average
Grenoble mill (EDC)
Bern cracker (EDC)
Bern sublimation (EDC)
Bern sublimation (TALDICE)
Monnin et al., 2001 (EDC)
ï6.7
ï6.8
ï6.9
280
260
240
220
200
180
B
0
5
10
15
Age (kyr BP)
20
25
CO2 (ppmv)
b13Catm (‰ VPDB)
0
Age (kyr BP)
5
A
10
15
20
IntCal09
25
600
450
300
150
0
ï150
Hulu Cave
b13C atmosphere (‰)
magnetic field based
10
ï6.2
ï6.3
ï6.4
ï6.5
ï6.6
ï6.7
ï6.8
ï6.9
Be based
corrected for surface
ocean temperature
Southern Ocean upwelling
100
50
HS2
D
Bard et al. 2000
0
E
YD
700
B/A
F
600
ï40
Greenland
ï45
500
G
DO2
ï390
400
280
ACR
ï410
255
H
ï430
Antarctica
230
ï450
205
180
0
5
10
15
Age (kyr BP)
20
25
bD (‰)
CH4 (ppmv)
PB
b O (‰)
ï35
18
IRD (#/g)
Anderson et al. 2009
HS1
6
4
2
0
opal flux
(g cmï2 kyrï1)
B this study
C
CO2 (ppmv)
614C (‰)
0
Supporting Online Material for
Carbon isotope constraints on the deglacial CO2 rise from ice cores
Jochen Schmitt, Robert Schneider, Joachim Elsig, Anna Lourantou, Daiana Leuenberger,
Jérôme Chappellaz, Peter Köhler, Fortunat Joos, Thomas Stocker, Markus Leuenberger,
Hubertus Fischer
correspondence to: [email protected]
This PDF file includes:
Materials and Methods
Figs. S1 to S7
Tables S1 to S2
References
1
Materials and Methods
Analysis of G13C on ice core samples
The measurement of G13C on ice core samples follows well-established protocols
which have been applied in other studies (6, 9, 10, 35). In short, all three methods rely on
four main steps. First, enclosed atmospheric air from the ice sample is released using a
dedicated extraction device (sublimation, needle cracker and a ball mill for the respective
data sets: Bern sublimation, Bern cracker and Grenoble mill). In a second step, the CO2
from the released ice core air is separated from the bulk air (N2, O2 and Ar) using
cryogenic trapping. Third, a gas chromatographic column purifies the CO2 sample from
other trace gases such as N2O. Finally, the stable carbon isotopic ratio of the CO2 sample
is measured against a bracketing standard using an isotope ratio mass spectrometer. The
typical measurement reproducibilities of the three methods are 0.05‰‰ - 0.07‰‰ for the
Bern sublimation (6, 10), 0.07‰‰ for the Bern cracker (6), and 0.10‰‰ for the Grenoble
mill data set (9). Note that the method used for the Bern sublimation data set employs a
novel sublimation technique (10). Besides a somewhat better prevision, this technique
offers the unique advantage of a quantitative gas extraction from the ice sample. The two
other methods use conventional mechanic extraction devices (mill or cracker techniques),
which extract only around 80% of the enclosed air in samples consisting of bubbly ice
(usually the upper part, i.e. <600 m depth, of a deep ice core). This may become more
problematic for ice from the bubble clathrate transition zone, where fractionation between
different gases has been observed for incomplete gas extraction (36). Here we use only
deglacial ice from Dome C located in the pure bubble zone and no gas extraction effects
are to be expected for any extraction method used.
Gravitational correction of the G13C data
Since gases and their isotopes become fractionated in the diffusive firn column
due to gravitational fractionation, the measured G13C values must be corrected for
gravitational enrichment to derive the atmospheric signal (G13Catm). To do this, measured
G15N on N2 serves as a proxy for the gravitational enrichment of the stable carbon
isotopes of CO2. This approach can be used because the isotopic composition of N2 was
constant over the covered time interval, and the gravitational fractionation is only
dependent on the mass difference, 'm, of the molecules, i.e. 'm = 1 for both G15N and
G13C. The three data sets (Bern sublimation, Bern cracker, and Grenoble mill) were
corrected using the same measured G15N data set provided by Dreyfus (37) and by
following previously published procedures (6, 9, 35). The typical gravitational
enrichment for the EDC core ranges between 0.42‰‰ for glacial samples and 0.52‰‰ for
Holocene samples. The reproducibility of the G15N measurements is about 0.01‰‰ (37).
As the G15N measurements and G13C measurements are not carried out on exactly
the same ice core sample nor at the same temporal resolution, an interpolation step must
2
first be performed. To prevent all measurement noise of G15N data from being transferred
to the reconstructed G13Catm record, a spline interpolation is used, which is superior to a
point-to-point interpolation for reasons discussed by Elsig (35). A second approach
makes use of an empirical relation between G15N and the water isotopic composition GD,
which serves a proxy for relevant firn processes, i.e. accumulation rate, thickness of the
diffusive column height and firn temperature (38). This G15N-GD relation must be applied
for time periods where G15N measurements are not available, which is the case for the
time interval between 24 kyr and 22 kyr. As the glacial-interglacial shift of G15N for the
EDC ice core is only 0.1‰‰ and the G15N-GD relation explains 88% of the variability (38),
the differences between the two approaches to correct for gravitational fractionation are
negligible compared to the reproducibility of the G13C measurement. Since the
measurement reproducibility of G15N is with 0.01‰‰ much lower than for G13C, the
introduced error is between 0.01 and 0.02‰‰ and mainly due the interpolation procedure
described above.
The ice core measurements and the calculated Monte Carlo Average are available at:
http://doi:10.1594/PANGAEA.772713
Extracting the atmospheric signal
The following procedure describes the stepwise statistical data analysis to combine
the three individual data sets (““residual analysis””) and to derive a robust atmospheric
signal (““Monte Carlo spline process””). These steps make use of a bootstrap procedure to
account for the specific measurement precision among the data sets and to allow for
identification of outliers, i.e. data points which are incompatible with the common
atmospheric signal in the data set.
Residual analysis
In the main text, we combine three independently measured G13Catm data sets of
different precision, accuracy and resolution for the last 24 kyr. A unified G13Catm time
series of the three individual time series has to account for offsets among the records as
well as for the individual measurement precision.
In the following, we quantify the overall differences in accuracy between the
records. As our sublimation data (6) agree well with measurements from the Law Dome
ice core (39), which have an overlap with firn air reconstructions and archived air from
the Cape Grim air archive, we set the sublimation data as the reference (for details see
discussion in Schmitt et al., 2011 (10)). To account for this agreement with the Law
Dome record (39), the overlapping period of the Law Dome record is added to the
combined data set for the Monte Carlo spline routine as additional data point (see Fig.
S4). In the first step, the potential offsets between the Bern sublimation data to the Bern
cracker data and the data from the Bern sublimation to the Grenoble mill are analyzed
and quantified. As the procedure is the same for the Bern cracker and Grenoble mill data
sets, we generally outline an offset calculation for the Bern sublimation to a 'non3
sublimation' data set. In principle, residual G13C values between two datasets are
determined where they overlap. The resulting mean systematic difference is called offset
and represents the difference in the accuracy of the methods. However, the individual ice
samples of the three data sets are not measured at exactly the same depth, hence gas age,
which complicates direct calculation of the G13C residuals. For each data point of a nonsublimation record, we choose the nearest available Bern sublimation data point in time.
The overall mean time difference between compared points is 't = 29 ± 171 years for the
Bern cracker and 't = 34 ± 175 years for the Grenoble mill data set (Fig. S1B and S2B).
This 't is sufficiently small compared to the width of the gas age distribution in EDC and
Talos Dome ice (40, 41). This is also illustrated in Fig. S1B and S2B, which show that
the absolute values of the residuals do not increase with larger 't. Hence, we can safely
assume that the differences in the compared data points arise essentially from differences
in the accuracy and precision in all three methods. The calculated individual 'G13C
residuals are sorted into bins and plotted in a histogram.
Performing a Gaussian fit yields the mean offset and the standard deviation between
sublimation and non-sublimation data set. To account for the relatively small sample
number for the residual distribution we performed a Monte Carlo procedure to determine
the mean and standard deviation of the distribution. In a first Monte Carlo iteration, the
procedure is repeated 100 times, randomly varying the input data within their 1V standard
deviation. From each of the obtained 100 Monte Carlo runs, a Gaussian fit is calculated
providing a mean offset and 1V standard deviation after averaging over the 100 runs.
Figure S3 shows the histogram of the residuals for the Bern cracker (A) and the Grenoble
mill (B) data set. On top of the histogram, a Gaussian distribution is drawn using the
average mean and standard deviation from the fitted distributions of the 100 Monte Carlo
runs.
If the residuals are only due to the stochastic measurement errors in the two methods
compared, then their distribution has to be Gaussian with the width of the distribution
given by the propagated error of both methods. This is perfectly fulfilled for the residuals
between the Bern sublimation and Bern cracker data (Fig. S3A). For the Bern cracker
record, the calculated offset of -0.01‰‰ ± 0.11‰‰ to the sublimation record is very small
and in line with a similar analysis done on the Holocene data set (6). The distribution of
the residuals is Gaussian, and its width can be entirely explained by error propagation of
the individual measurement error of both methods. For the Grenoble mill data the
distribution is reasonably Gaussian but the width is slightly larger than expected from the
propagated measurement errors. An offset of 0.16‰‰ ± 0.17‰‰ between the Grenoble mill
data and the Bern sublimation data is observed and correcting for this offset improves the
combined record. To apply a constant offset correction, the offset has to be constant with
time. Therefore, we performed a second analysis for the case of the 'G13C residuals
varying with time. Fig. S1A and Fig. S2A show these residuals as a function of the age of
the respective data point. The Bern cracker data shows no temporal dependence of the
offset, the Grenoble mill data shows a slight tendency to higher offsets for younger ages
(higher CO2 concentrations), however, this trend is statistically not significant. Due to the
small sample number available for the residual analysis we refrained from correcting this
insignificant trend. Instead we only shifted the Grenoble data set by its mean offset
relative to the Bern sublimation data. The three involved labs in Bern and Grenoble also
4
conducted a round-robin exercise with three cylinders with working standards to identify
possible scale differences among the labs. For the cylinders, we also measured offsets
which range in the observed differences among Bern and Grenoble ice core data. We
could not completely identify the reasons for these small differences. Note that in view of
the complex analytical systems and the differences in the referencing strategy of the three
methods, the observed systematic offsets are to be expected, rather small and
reproducible.
We acknowledge that the posteriori offset procedure introduces a systematic
additional error to the Grenoble data set. However, we did not add this ““offset error”” to
the measurement error since the Monte Carlo procedure would then put even less weight
onto this data set and the calculated spline would be more biased towards the Bern data
sets. Note that the outcome of the Monte Carlo spline procedure described below, is not
significantly affected by the inclusion of the Grenoble mill data and that all of the
conclusions drawn in our paper also hold, if that data set were to be excluded. However,
inclusion of all three data sets is essential to obtain the highest temporal resolution
possible for the G13Catm record.
Calculating Monte Carlo splines of the combined record
In a second step after the offset correction, cubic splines (42) of the combined
G Catm time series were calculated. To exclude the high frequency measurement noise,
splines with a cut-off period of 375 years were selected (see Fig. S4 showing 400 sample
splines). This value was chosen as for the ice cores used in this study, variability on
shorter time periods is strongly dampened in gas reconstructions due to the low-pass
filtering of the primary atmospheric signal by the continuous bubble enclosure process in
the firn column (43). The spline calculation is repeated 4000 times, with the input data
randomly picked from their 1V error range. For replicate samples, the 1V standard
deviation of these specific data points is used. In the case that the 1V standard deviation
for replicate samples is better than the mean reproducibility of the method, the mean
standard deviation of the data set is used. The same applies for points where only one ice
sample was measured. Assigning the mean standard deviation of the data set to those data
points prevents unjustified pinning of the spline evolution at these positions. As can be
seen in Fig. S4 and S5, regions where data points have small error bars tend to pin the
spline evolution (e.g., compare spline evolution in Fig. S4 at position (D) and the
respective distribution in Fig. S5D). The arithmetic average, here called the Monte Carlo
Average (MCA), is calculated from the 4000 splines. Due the variability of the individual
splines, the calculated MCA is smoother than the individual splines (Fig. S4). This
additional smoothing of the MCA is a function of the measurement precision of a data
point and its temporal distance to its neighbors (i.e. temporal resolution). Since neither
the individual records nor the combined data set are equidistant and the measurement
error varies among the data sets and with time, the additional smoothing of the MCA
varies with time (Fig. S4). In other words, the MCA cannot be characterized by a defined
cut-off period. Instead, our procedure removes high frequency components of the splines
at positions where these were not robust enough as measurement error and temporal
resolution did not allow to constrain the individual splines sufficiently. On the other
13
5
hand, for time intervals with small error bars and dense data coverage, the individual
splines are pinned. Here, the resulting distribution is narrow, and the high frequency
components were retained (Fig. S5D). This procedure is superior to approaches
calculating only one spline with a fixed cut-off period, hence constant smoothing
characteristic, as the MCA can account for inhomogeneous measurement error and
variable temporal resolution, a common feature of our data sets. Finally, the dark gray
(light gray) shaded envelopes in Fig. 1 and 2 are obtained, marking the 1V (2V)
confidence interval of the Monte Carlo average. As the distributions of the splines at a
given point in time are not truly Gaussian (see example histograms in Fig. S5), the 1V
and 2V confidence intervals serve as approximations for the true spline variability at each
point in time. Table S3 provided digitally in the supplement lists the MCA and the
1Venvelope on the synchronized age scale (31).
To investigate the robustness of the chosen approach to combine the three data sets, we
also calculated MCAs for subsets of the combined data. Four cases are selected and
shown in Fig. S6. In case A we tested if the outlier detection does actually improve or
alter the MCA (see discussion below). As illustrated in Fig. S6A this is not the case. Case
B is calculated without the TALDICE samples (which have a different gas age
distribution than Dome C samples) and shows that removing this subset does not change
the MCA. Removing the Grenoble subset in case D leads to only minor changes in the
MCA between 12-15 kyr BP. When the MCA is calculated from the sublimation data
only, the spline routine produces more wiggles as the temporal resolution of the
sublimation data alone is not sufficient for this cut-off period. However, also in this
extreme case the general shape of the MCA is still preserved and the MCAs agree within
their 2Venvelopes. This analysis shows that combining the two Bern data sets and the
Grenoble data set is a beneficial approach as it increases the temporal resolution and
provides a more robust result but does not create spurious trends.
Outlier detection with bootstrapping
Atmospheric stable carbon isotope measurements on the EDC ice core over the
Holocene report a significant measurement outlier at 170 m depth, that can neither be
explained by a climatic excursion nor by poor measurement precision, but is likely
caused by problems during the ice core drilling (6). To test for further outliers in the
combined data set, the above described Monte Carlo procedure is repeated, but each time
one data point is excluded. If the calculated 2V confidence interval of the MCA does not
overlap with the 2V error of the excluded data point, this data point is assigned as an
outlier and excluded from the final spline calculation. This test is done for all data points,
resulting in only four measurements excluded from the data set. Note that for this
bootstrap procedure, all individual measurements were used instead of the replicate
mean. In a last step, the bootstrap procedure was repeated using also the replicate means
instead of the individual replicates. No data point was identified as an outlier during this
final step. Since only four individual measurement points were identified as outlier, the
calculated MCA is not dependent on whether the outliers are included or excluded during
the Monte Carlo procedure as the respective MCA and 2V confidence intervals nicely
overlap (Fig. S6A). In summary, if a data point cannot be identified a priory as an artifact
6
due to either measurement problems or during drilling (e.g. the 170 m depth outliers in
the Holocene section) the bootstrap procedure does not improve the MCA. In other
words, the observed scatter in the combined data set can be explained already by the
combined measurement error and errors due to offset corrections.
Correction for SST changes
With changing sea surface temperature (SST), not only the partial pressure of
dissolved CO2 in water (pCO2) but also the isotopic fractionation during the gas exchange
is altered. The G13C of atmospheric CO2 becomes isotopically heavier by roughly 0.1‰‰
per 1 K increase as the fractionation factor decreases (12, 44, 45). In the following, we
attempted to roughly quantify this effect of changing SST on G13Catm over the transition.
The temperature at 19 kyr BP is set as a reference point, as changes in global mean
sea surface temperature at earlier glacial times are small compared to the variations
during the glacial termination. The isotopic ratio of atmospheric CO2 't years after 19
kyr BP is calculated according to:
G13C (19 kyr BP + 't) = G13C (19 kyr BP) + 'G13C),
(1)
'G13C) = D'T, with D = 0.1‰‰/K.
(2)
where
The evolution of the past global mean SST can be approximated from direct marine
proxies, such as alkenones or Ca/Mg ratios. For regions and time periods without marine
proxy records, scaled temperature evolutions of Greenland and Antarctica serve as crude
representations of the surrounding oceans. In the course of this work, we use the
estimations compiled by Köhler et al. (4) that have been used in other modeling studies
(11). In this compilation, the surface ocean is divided into five regions (corresponding to
the surface boxes used in Köhler et al. (4)): the North Atlantic, the Equatorial Atlantic,
the Southern Ocean, the Equatorial Indo-Pacific and the North Pacific. For each box,
temperature proxies are used to account for the change in SST. The stable water isotopic
signal G18O from the GISP2 (46) and GD from the EDC (47) ice cores were rescaled to a
glacial/interglacial amplitude of 4 K and used to quantify the SST evolutions of the
Northern boxes and the Southern Ocean, respectively. Temperature changes in lower
latitude boxes were adapted from foraminifera records (48). The SST approximations for
the five surface boxes are shown in Fig. S7B (equatorial boxes) and Fig. S7C (high
latitude boxes), together with the area-weighted temperature average (gray curve in Fig.
S7B). Note that gas exchange and thus the isotopic fractionation are largely inhibited for
sea-ice covered areas. Therefore, only areas dominated by open water were used for the
calculation of the area-weighted SST average. These areas are calculated internally
within the BICYCLE model (4). Accounting for the change in sea-ice area leads to a
difference of only 0.8 K in the area-weighted SST estimate over the period between 19
kyr and 10 kyr. The total shift from 19 kyr BP to the Holocene in this reconstruction is
2.8 K.
7
This global SST record is used to calculate the SST-corrected evolution of G13Catm
according to equations (1) and (2) (gray line in Fig. S7A) by subtracting the SST-induced
G13C change from the Monte Carlo Average (red line in Fig. S7A). We note that our
temperature approximation introduces further error to the one and two sigma confidence
interval of the SST corrected record. However, a detailed error propagation analysis is
beyond the scope of the overall approximation since the purpose of this correction was to
provide only a rough estimate of the global SST change on our G13Catm record, i.e. to
show that global SST changes likely shift the G13Catm by approximately 0.3‰‰ from the
LGM to the Holocene. Note also that the proxies used for our crude SST reconstruction
are on their individual age scales; therefore, interpretations on the submillennial timescale
are not possible. For illustration purposes, we also assumed a simple linear SST increase
of 2.8 K between 19 kyr and 10 kyr and a constant SST during the Holocene (Fig. S7B).
The resulting G13C anomaly was subtracted from the G13Catm evolution of the MCA, and
this simple linear SST rise correction (Fig. S7A blue and orange line) is compared with
the proxy-based SST correction. As the proxy based mean SST evolution during the
deglaciation is already rather linear with only small deviations during the YD-BA
oscillation, the two SST reconstructions are similar as well. In summary, our SST
correction serves as an illustration of the effect of an overall glacial/interglacial warming,
while small scale features in the corrected G13Catm record on sub-millennial time scale
should not be interpreted.
8
Fig. S1.
Residual plots for the Bern cracker data set. (A) Residuals plotted against the absolute
age of the Bern cracker measurements. The data points are randomly scattered around the
zero line showing no trend with time. However, the scatter shows a tendency to become
larger for glacial ice with lower CO2 concentration. (B) Residuals plotted against 't, i.e.
the temporal difference between a Bern cracker measurement to its nearest neighbor in
the Bern sublimation data. Most 't values lie between +0.2 kyr and -0.2 kyr, with no
trend of larger 'G13C residuals for larger 't differences.
9
Fig. S2
Residual plots for the Grenoble mill data set. (A) Residuals plotted against the absolute
age of the Grenoble mill measurements. The data points are randomly scattered but show
an offset to the zero line. (B) Residuals plotted against 't, i.e. the temporal difference
between a Grenoble mill measurement to its nearest neighbor in the Bern sublimation
data. Most 't values lie between +0.2 kyr and -0.2 kyr, with no trend of larger 'G13C
residuals for larger 't differences.
10
Fig. S3
Histograms showing the offsets among the data sets: The histograms show the residuals
of the measured mean value at a given sample depth for the Bern cracker (A) and the
Grenoble mill (B) data sets to its nearest neighbor in the Bern sublimation data. The
output of a Monte Carlo procedure (see text) was used to determine the mean and
standard deviation of the Gaussian curve (black line).
11
Fig. S4
Individual Monte Carlo splines: Compilation showing the color-coded G13Catm
measurements. Squares indicate published data, circles new measurements, and triangles
denote individual data points which were detected as outliers in the bootstrap routine. The
Monte Carlo Average is plotted in gray and in black a sample of 400 Monte Carlo splines
with a cut-off period of 375 years. Latin letters A to F connected to vertical lines denote
time intervals where the frequency distribution of the 4000 splines is shown in Fig. S5.
Numerals from 1-7 at the bottom axis mark the time intervals shown in Tab. S2. All
G13Catm records from this study are on a synchronized age scale (31); only the reference
data point at 0.42 kyr from the Law Dome record (39) is on its original age scale.
12
Fig. S5
Frequency distributions of the Monte Carlo splines at selected points in time: Histograms
showing the distributions of the 4000 splines at selected points of our G13Catm time series
(positions marked in Fig. S4 with brown Latin letters). Bars in dark gray mark the 1V
interval, light gray the 2V interval, and white bars are outside the 2V interval. The black
(red) vertical line marks the mean (median) of the distribution. Note that some
distributions are not Gaussian but have a broader shape; therefore, mean and standard
deviation do not entirely describe the frequency distributions.
13
Fig. S6
Comparison of Monte Carlo Averages for different data sets included. Each panel shows
the final MCA in black, i.e. the MCA which is used in the main figures Fig. 1 and Fig. 2
of the paper and on which the interpretations are based. The individual MCAs in the subpanels A-D are plotted in colors. The 2V envelope of the final MCA is in gray and the 2V
envelopes of sub-panels A-D are colored and overlapping 2V envelopes are in the
respective mixed tones. A All three data sets but with the five outliers detected in the
bootstrap routine included. B TALDICE samples excluded C Only Bern cracker and
Bern sublimation data (i.e. Grenoble ball mill data excluded) D Bern sublimation data
only. While the calculated MCA is insensitive to the outliers and the TALDICE samples,
removing the Grenoble ball mill data and Bern cracker data has an effect on the MCA. In
case of removing the Grenoble data, the calculated MCA deviates somewhat from the
final MCA between 12 and 14 kyr, where there is the largest disagreement among the
three data set. However, the overall shape is retained. When only the sublimation data is
used, the MCA becomes unstable and pinned at individual data points, since the
resolution is insufficient for the applied cut-off period of 375 years. All exclusion
experiments show that the respective 2V envelopes overlap and thus record the same
robust G13Catm atmospheric features seen in the final MCA.
14
Fig. S7
SST correction of the G13Catm MCA evolution. A proxy-based SST reconstruction is
compared with a linear temperature increase during the deglaciation. (A) Evolution of
G13Catm (red line) and proxy-based SST corrections (gray line) compared with a linear rise
of 2.8 K between 19 kyr and 10 kyr (blue line) followed by constant SST during the
Holocene (orange line). (B) SST evolution of the equatorial boxes (red lines) and the
area-weighted average of all five surface boxes (gray), the latter being used for the proxy
based SST correction above, (C) SST evolution of the three high latitude surface boxes
(blue lines).
15
16
Table S1.
Selected points of the Monte Carlo Average and their respective mean, median, and 1V
standard deviation as used to calculate the time intervals shown in Tab. S2.
interval boundary
mean G13Catm
median G13Catm
V
(kyr BP)
(‰‰)
(‰‰)
(‰‰)
5.7
-6.32
-6.33
±0.03
12.2
-6.69
-6.68
±0.04
13.3
-6.63
-6.63
±0.04
15.4
-6.70
-6.70
±0.04
16.0
-6.69
-6.69
±0.06
17.4
-6.40
-6.40
±0.05
19.3
-6.46
-6.46
±0.03
24.0
-6.40
-6.40
±0.05
17
Table S2.
Time intervals listing the trends in mean G13Catm and median G13Catm between two interval
boundaries as listed in Tab. S1 and marked in Fig. S4. As a conservative estimate of the
confidence of these differences, the last column lists the sum of both 1V standard
deviations of G13Catm at the interval boundaries. Note that interval 1, 2, 3, 5 and 7
represent the differences between local minima and maxima positions on the MCA. In
contrast, intervals 4 and 7 were selected to represent relatively flat plateaus. Intervals 1
and 5 are indicated in bold to indicate that most of the G13Catm variability is concentrated
in these two intervals.
interval boundaries
'mean G13Catm
'median G13Catm
6V
#
(kyr BP)
(‰‰)
(‰‰)
(‰‰)
1
12.2 - 5.7
0.37 Ĺ
0.35 Ĺ
±0.07
2
13.3 - 12.2
0.06 Ļ
0.05 Ļ
±0.08
3
15.4 - 13.3
0.07 Ĺ
0.07 Ĺ
±0.08
4
16.0 - 15.4
0.01 Ļ
0.01 Ļ
±0.10
5
17.4 - 16.0
0.29 Ļ
0.29 Ļ
±0.11
6
19.3 - 17.4
0.06 Ĺ
0.06 Ĺ
±0.08
7
24.0 - 19.3
0.06 Ļ
0.06 Ļ
±0.08
interval
18
References:
1.
D. Lüthi et al., High-resolution carbon dioxide concentration record 650,000800,000 years before present. Nature 453, 379 (2008). doi:10.1038/nature06949
2.
H. Fischer et al., The role of Southern Ocean processes in orbital and millennial
CO2 variations - A synthesis. Quat. Sci. Rev. 29, 193 (2010).
doi:10.1016/j.quascirev.2009.06.007
3.
E. Monnin et al., Atmospheric CO2 concentration over the last termination.
Science 291, 112 (2001).
4.
P. Köhler, H. Fischer, G. Munhoven, R. E. Zeebe, Quantitative interpretation of
atmospheric carbon records over the last glacial termination. Global Biogeochem.
Cycles 19, GB4020 (2005). doi:10.1029/2004GB002345
5.
D. M. Sigman, M. P. Hain, G. H. Haug, The polar ocean and glacial cycles in
atmospheric CO2 concentration. Nature 466, 47 (2010). doi: 10.1038/nature09149
6.
J. Elsig et al., Stable isotope constraints on Holocene carbon cycle changes from
an Antarctic ice core. Nature 461, 507 (2009). doi:10.1038/nature08393
7.
H. J. Smith, H. Fischer, M. Wahlen, D. Mastroianni, B. Deck, Dual modes of the
carbon cycle since the Last Glacial Maximum. Nature 400, 248 (1999). doi:
10.1038/22291
8.
M. Leuenberger, U. Siegenthaler, C. C. Langway, Carbon isotope composition of
atmospheric CO2 during the last ice age from an Antarctic ice core. Nature 357,
488 (1992).
9.
A. Lourantou et al., Constraint of the CO2 rise by new atmospheric carbon
isotopic measurements during the last deglaciation. Global Biogeochem. Cycles
24, GB2015 (2010). doi:10.1029/2009GB003545
10.
J. Schmitt, R. Schneider, H. Fischer, A sublimation technique for high-precision
measurements of G13CO2 and mixing ratios of CO2 and N2O from air trapped in
ice cores. Atmos. Meas. Tech. 4, 1445 (2011). doi:10.5194/amt-4-1445-2011
11.
P. Köhler, H. Fischer, J. Schmitt, Atmospheric GCO2 and its relation to pCO2
and deep ocean GC during the late Pleistocene. Paleoceanography 25, PA1213
(2010). doi:10.1029/2008PA001703
12.
J. Zhang, P. D. Quay, D. O. Wilbur, Carbon isotope fractionation during gaswater exchange and dissolution of CO2. Geochim Cosmochim Ac 59, 107 (1995).
19
13.
K. I. C. Oliver et al., A synthesis of marine sediment core G3C data over the last
150 000 years. Clim. Past 6, 645 (2010). doi:10.5194/cp-6-645-2010
14.
P. J. Reimer et al., Intcal09 and Marine09 Radiocarbon Age Calibration Curves,
0-50,000 Years Cal BP. Radiocarbon 51, 1111 (2009).
15.
J. Southon, A. L. Noronha, H. Cheng, R. L. Edwards, Y. Wang, A high-resolution
record of atmospheric 14C based on Hulu Cave speleothem H82. Quat. Sci. Rev.
33, 32 (2012). doi: 10.1016/j.quascirev.2011.11.022
16.
P. Köhler, R. Muscheler, H. Fischer, A model-based interpretation of lowfrequency changes in the carbon cycle during the last 120,000 years and its
implications for the reconstruction of atmospheric '14C. Geochem. Geophys.
Geosyst. 7, Q11N06 (2006). doi: 10.1029/2005GC001228
17.
C. Laj, C. Kissel, J. Beer, in Timescales of the Paleomagnetic Field. (American
Geophysical Union, 2004), vol. 145, pp. 255-265.
18.
R. Muscheler, R. Beer, P. W. Kubik, H. A. Synal, Geomagnetic field intensity
during the last 60,000 years based on 10Be and 36Cl from the Summit ice cores and
14
C. Quat. Sci. Rev. 24, 1849 (2005). doi: 10.1016/j.quascirev.2005.01.012
19.
W. Broecker, S. Barker, A 190‰‰ drop in atmosphere's '14C during the "Mystery
Interval" (17.5 to 14.5 kyr). Earth Planet. Sci. Lett. 256, 90 (2007). doi:
10.1016/j.epsl.2007.01.015
20.
R. F. Anderson et al., Wind-Driven Upwelling in the Southern Ocean and the
Deglacial Rise in Atmospheric CO2. Science 323, 1443 (2009).
21.
P. Ciais et al., Large inert carbon pool in the terrestrial biosphere during the Last
Glacial Maximum. Nat. Geosci 5, 74 (2012). doi: 10.1038/NGEO1324
22.
J. M. Yu et al., Loss of Carbon from the Deep Sea Since the Last Glacial
Maximum. Science 330, 1084 (2010).
23.
D. C. Lund, A. C. Mix, J. Southon, Increased ventilation age of the deep northeast
Pacific Ocean during the last deglaciation. Nat. Geosci. 4, 771 (2011).
doi:10.1038/ngeo1272
24.
L. Stott, J. Southon, A. Timmermann, A. Koutavas, Radiocarbon age anomaly at
intermediate water depth in the Pacific Ocean during the last deglaciation.
Paleoceanography 24, PA2223 (2009). doi:10.1029/2008PA001690
25.
W. Broecker, E. Clark, Search for a glacial-age 14C-depleted ocean reservoir.
Geophys. Res. Lett. 37, L13606 (2010). doi:10.1029/2010GL043969
20
26.
A. Burke, L. F. Robinson, The Southern Ocean's role in carbon exchange during
the last deglaciation. Science 335, 557 (2012). doi: 10.1126/science.1208163
27.
E. Bard, F. Rostek, J. L. Turon, S. Gendreau, Hydrological impact of Heinrich
events in the subtropical northeast Atlantic. Science 289, 1321 (2000).
28.
T. Tschumi, F. Joos, M. Gehlen, C. Heinze, Deep ocean ventilation, carbon
isotopes, marine sedimentation and the deglacial CO2 rise. Clim. Past 7, 771
(2011). doi: 10.5194/cp-7-771-2011
29.
R. Zech, Y. Huang, M. Zech, R. Tarozo, W. Zech, High carbon sequestration in
Siberian permafrost loess-paleosols during glacials. Clim. Past 7, 501 (2011). doi:
10.5194/cp-7-501-2011
30.
N. Zeng, Quasi-100 ky glacial-interglacial cycles triggered by subglacial burial
carbon release. Clim. Past 3, 135 (2007).
31.
D. A. Hodell, C. D. Charles, F. J. Sierro, Late Pleistocene evolution of the ocean's
carbonate system. Earth Planet. Sci. Lett. 192, 109 (2001). doi: 10.1016/S0012821X(01)00430-7
32.
B. Lemieux-Dudon et al., Consistent dating for Antarctic and Greenland ice
cores. Quat. Sci. Rev. 29, 8 (2010). doi: 10.1016/j.quascirev.2009.11.010
33.
North Greenland Ice Core Project members, High-resolution record of Northern
Hemisphere climate extending into the last interglacial period. Nature 431, 147
(2004). doi: 10.1038/nature02805
34.
EPICA Community Members, One-to-one coupling of glacial climate variability
in Greenland and Antarctica. Nature 444, 195 (2006). doi: 10.1038/nature05301
35.
B. Stenni et al., An oceanic cold reversal during the last deglaciation. Science
293, 2074 (2001). doi: 10.1126/science.1059702
36.
J. Elsig, Dissertation, University of Bern (2009).
37.
D. Lüthi et al., CO2 and O2/N2 variations in and just below the bubble-clathrate
transformation zone of Antarctic ice cores. Earth Planet. Sci. Lett. 297, 226
(2010). doi:10.1016/j.epsl.2010.06.023
38.
G. Dreyfus, Phd Thesis, Princeton University (2008).
39.
G. B. Dreyfus et al., Firn processes and GN: potential for a gas-phase climate
proxy. Quat. Sci. Rev. 29, 28 (2009).
21
40.
R. J. Francey et al., A 1000-year high precision record of G13C in atmospheric
CO2. Tellus 51B, 170 (1999).
41.
R. Spahni et al., The attenuation of fast atmospheric CH4 variations recorded in
polar ice cores. Geophys. Res. Lett. 30, 25 (2003).
42.
D. Buiron et al., TALDICE-1 age scale of the Talos Dome deep ice core, East
Antarctica. Clim. Past 7, 1 (2011).
43.
I. G. Enting, On the Use of Smoothing Splines to Filter CO2 Data. J. Geophys.
Res.-Atmos. 92, 10977 (1987). doi: 10.1029/JD092iD09p10977
44.
J. Schwander, B. Stauffer, A. Sigg, Air mixing in firn and the age of the air at
pore close-off. Annals of Glaciology 10, 141 (1988).
45.
A. Indermühle et al., Holocene carbon-cycle dynamics based on CO2 trapped in
ice at Taylor Dome, Antarctica. Nature 398, 121 (1999).
46.
R. E. Zeebe, D. A. Wolf-Gladrow, CO2 in seawater: equilibrium, kinetics,
isotopes, Elsevier Oceanography series (2005), pp. 65.
47.
P. M. Grootes, M. Stuiver, Oxygen 18/16 variability in Greenland snow and ice
with 103- to 105 -year time resolution. J. Geophys. Res.-Oceans 102, 26455
(1997).
48.
J. Jouzel et al., Orbital and millennial Antarctic climate variability over the past
800,000 years. Science 317, 793 (2007). doi: 10.1126/science.1141038
49.
K. Visser, R. Thunell, L. Stott, Magnitude and timing of temperature change in
the Indo-Pacific warm pool during deglaciation. Nature 421, 152 (2003).
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